Merge tag '1.6.0-rc1' into master-1.6.0

Release target Nov 11, 2018

Conflicts:
	configure

.gitignore

@@ -1,3 +1,4 @@
+*~
 *.o
 *.elf
 *.bin

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "ihk"]
+	path = ihk
+	url = https://github.com/RIKEN-SysSoft/ihk.git

arch/arm64/kernel/cpu.c

@@ -30,6 +30,7 @@
 #include <debug-monitors.h>
 #include <sysreg.h>
 #include <cpufeature.h>
+#include <debug.h>
 #ifdef POSTK_DEBUG_ARCH_DEP_65
 #include <hwcap.h>
 #endif /* POSTK_DEBUG_ARCH_DEP_65 */
@@ -39,16 +40,10 @@
 #include "postk_print_sysreg.c"
 
 #ifdef DEBUG_PRINT_CPU
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf kprintf
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf kprintf
 #endif
 
-#define BUG_ON(condition) do { if (condition) { kprintf("PANIC: %s: %s(line:%d)\n",\
-				__FILE__, __FUNCTION__, __LINE__); panic(""); } } while(0)
-
 struct cpuinfo_arm64 cpuinfo_data[NR_CPUS];	/* index is logical cpuid */
 static unsigned int per_cpu_timer_val[NR_CPUS] = { 0 };
 
@@ -1283,7 +1278,6 @@ int ihk_mc_interrupt_cpu(int cpu, int vector)
 	return 0;
 }
 
-#ifdef POSTK_DEBUG_ARCH_DEP_22
 /*
  * @ref.impl linux-linaro/arch/arm64/kernel/process.c::tls_thread_switch()
  */
@@ -1309,14 +1303,13 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 	extern void perf_start(struct mc_perf_event *event);
 	extern void perf_reset(struct mc_perf_event *event);
 	struct thread *last;
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 	struct mcs_rwlock_node_irqsave lock;
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 
 	/* Set up new TLS.. */
 	dkprintf("[%d] arch_switch_context: tlsblock_base: 0x%lX\n", 
 		 ihk_mc_get_processor_id(), next->tlsblock_base);
 
+#ifdef ENABLE_PERF
 	/* Performance monitoring inherit */
 	if(next->proc->monitoring_event) {
 		if(next->proc->perf_status == PP_RESET)
@@ -1326,10 +1319,10 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 			perf_start(next->proc->monitoring_event);
 		}
 	}
+#endif /*ENABLE_PERF*/
 	if (likely(prev)) {
 		tls_thread_switch(prev, next);
 
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 		mcs_rwlock_writer_lock(&prev->proc->update_lock, &lock);
 		if (prev->proc->status & (PS_DELAY_STOPPED | PS_DELAY_TRACED)) {
 			switch (prev->proc->status) {
@@ -1343,11 +1336,12 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 				break;
 			}
 			mcs_rwlock_writer_unlock(&prev->proc->update_lock, &lock);
+
+			/* Wake up the parent who tried wait4 and sleeping */
 			waitq_wakeup(&prev->proc->parent->waitpid_q);
 		} else {
 			mcs_rwlock_writer_unlock(&prev->proc->update_lock, &lock);
 		}
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 
 		last = ihk_mc_switch_context(&prev->ctx, &next->ctx, prev);
 	}
@@ -1357,7 +1351,6 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 
 	return last;
 }
-#endif /* POSTK_DEBUG_ARCH_DEP_22 */
 
 /*@
   @ requires \valid(thread);
@@ -1439,8 +1432,7 @@ void copy_fp_regs(struct thread *from, struct thread *to)
 	}
 }
 
-void
+void clear_fp_regs(void)
-clear_fp_regs(struct thread *thread)
 {
 	if (likely(elf_hwcap & (HWCAP_FP | HWCAP_ASIMD))) {
 #ifdef CONFIG_ARM64_SVE
@@ -1477,7 +1469,7 @@ restore_fp_regs(struct thread *thread)
 	if (likely(elf_hwcap & (HWCAP_FP | HWCAP_ASIMD))) {
 		if (!thread->fp_regs) {
 			// only clear fpregs.
-			clear_fp_regs(thread);
+			clear_fp_regs();
 			return;
 		}
 		thread_fpsimd_load(thread);

arch/arm64/kernel/fpsimd.c

@@ -9,20 +9,16 @@
 #include <prctl.h>
 #include <cpufeature.h>
 #include <kmalloc.h>
+#include <debug.h>
+#include <process.h>
 
 //#define DEBUG_PRINT_FPSIMD
 
 #ifdef DEBUG_PRINT_FPSIMD
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf kprintf
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf kprintf
 #endif
 
-#define BUG_ON(condition) do { if (condition) { kprintf("PANIC: %s: %s(line:%d)\n",\
-				__FILE__, __FUNCTION__, __LINE__); panic(""); } } while(0)
-
 #ifdef CONFIG_ARM64_SVE
 
 /* Maximum supported vector length across all CPUs (initially poisoned) */
@@ -73,9 +69,6 @@ static int get_nr_threads(struct process *proc)
 	return nr_threads;
 }
 
-extern void save_fp_regs(struct thread *thread);
-extern void clear_fp_regs(struct thread *thread);
-extern void restore_fp_regs(struct thread *thread);
 /* @ref.impl arch/arm64/kernel/fpsimd.c::sve_set_vector_length */
 int sve_set_vector_length(struct thread *thread,
 			unsigned long vl, unsigned long flags)
@@ -129,7 +122,7 @@ int sve_set_vector_length(struct thread *thread,
 			/* for self at prctl syscall */
 			if (thread == cpu_local_var(current)) {
 				save_fp_regs(thread);
-				clear_fp_regs(thread);
+				clear_fp_regs();
 				thread_sve_to_fpsimd(thread, &fp_regs);
 				sve_free(thread);
 

arch/arm64/kernel/gencore.c

@@ -7,6 +7,7 @@
 #include <process.h>
 #include <string.h>
 #include <elfcore.h>
+#include <debug.h>
 
 #define	align32(x) ((((x) + 3) / 4) * 4)
 #define	alignpage(x) ((((x) + (PAGE_SIZE) - 1) / (PAGE_SIZE)) * (PAGE_SIZE))
@@ -14,11 +15,8 @@
 //#define DEBUG_PRINT_GENCORE
 
 #ifdef DEBUG_PRINT_GENCORE
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 /*

arch/arm64/kernel/include/arch-lock.h

@@ -6,6 +6,8 @@
 
 #include <ihk/cpu.h>
 #include <ihk/atomic.h>
+#include "affinity.h"
+#include <lwk/compiler.h>
 
 //#define DEBUG_SPINLOCK
 //#define DEBUG_MCS_RWLOCK
@@ -152,6 +154,8 @@ typedef struct mcs_lock_node {
 	unsigned long irqsave;
 } __attribute__((aligned(64))) mcs_lock_node_t;
 
+typedef mcs_lock_node_t mcs_lock_t;
+
 static void mcs_lock_init(struct mcs_lock_node *node)
 {
 	node->locked = 0;
@@ -602,4 +606,16 @@ __mcs_rwlock_reader_unlock(struct mcs_rwlock_lock *lock, struct mcs_rwlock_node_
 #endif
 }
 
+static inline int irqflags_can_interrupt(unsigned long flags)
+{
+#ifdef CONFIG_HAS_NMI
+#warning irqflags_can_interrupt needs testing/fixing on such a target
+	return flags > ICC_PMR_EL1_MASKED;
+#else
+	// PSTATE.DAIF I bit clear means interrupt is possible
+	return !(flags & (1 << 7));
+#endif
+}
+
+
 #endif /* !__HEADER_ARM64_COMMON_ARCH_LOCK_H */

arch/arm64/kernel/include/arch-perfctr.h

@@ -35,38 +35,4 @@ void arm64_disable_pmu(void);
 int armv8pmu_init(struct arm_pmu* cpu_pmu);
 
 /* TODO[PMU]: 共通部に定義があっても良い。今後の動向を見てここの定義を削除する */
-/*
- * Generalized hardware cache events:
- *
- *       { L1-D, L1-I, LLC, ITLB, DTLB, BPU, NODE } x
- *       { read, write, prefetch } x
- *       { accesses, misses }
- */
-enum perf_hw_cache_id {
-	PERF_COUNT_HW_CACHE_L1D			= 0,
-	PERF_COUNT_HW_CACHE_L1I			= 1,
-	PERF_COUNT_HW_CACHE_LL			= 2,
-	PERF_COUNT_HW_CACHE_DTLB		= 3,
-	PERF_COUNT_HW_CACHE_ITLB		= 4,
-	PERF_COUNT_HW_CACHE_BPU			= 5,
-	PERF_COUNT_HW_CACHE_NODE		= 6,
-
-	PERF_COUNT_HW_CACHE_MAX,		/* non-ABI */
-};
-
-enum perf_hw_cache_op_id {
-	PERF_COUNT_HW_CACHE_OP_READ		= 0,
-	PERF_COUNT_HW_CACHE_OP_WRITE		= 1,
-	PERF_COUNT_HW_CACHE_OP_PREFETCH		= 2,
-
-	PERF_COUNT_HW_CACHE_OP_MAX,		/* non-ABI */
-};
-
-enum perf_hw_cache_op_result_id {
-	PERF_COUNT_HW_CACHE_RESULT_ACCESS	= 0,
-	PERF_COUNT_HW_CACHE_RESULT_MISS		= 1,
-
-	PERF_COUNT_HW_CACHE_RESULT_MAX,		/* non-ABI */
-};
-
 #endif

arch/arm64/kernel/include/signal.h

@@ -9,6 +9,11 @@
 #define _NSIG_BPW	64
 #define _NSIG_WORDS (_NSIG / _NSIG_BPW)
 
+static inline int valid_signal(unsigned long sig)
+{
+	return sig <= _NSIG ? 1 : 0;
+}
+
 typedef unsigned long int __sigset_t;
 #define __sigmask(sig)  (((__sigset_t) 1) << ((sig) - 1))
 

arch/arm64/kernel/include/syscall_list.h

@@ -114,14 +114,18 @@ SYSCALL_HANDLED(236, get_mempolicy)
 SYSCALL_HANDLED(237, set_mempolicy)
 SYSCALL_HANDLED(238, migrate_pages)
 SYSCALL_HANDLED(239, move_pages)
+#ifdef PERF_ENABLE
 SYSCALL_HANDLED(241, perf_event_open)
+#endif // PERF_ENABLE
 SYSCALL_HANDLED(260, wait4)
 SYSCALL_HANDLED(270, process_vm_readv)
 SYSCALL_HANDLED(271, process_vm_writev)
+#ifdef PERF_ENABLE
 SYSCALL_HANDLED(601, pmc_init)
 SYSCALL_HANDLED(602, pmc_start)
 SYSCALL_HANDLED(603, pmc_stop)
 SYSCALL_HANDLED(604, pmc_reset)
+#endif // PERF_ENABLE
 SYSCALL_HANDLED(700, get_cpu_id)
 #ifdef PROFILE_ENABLE
 SYSCALL_HANDLED(__NR_profile, profile)

arch/arm64/kernel/irq-gic-v2.c

@@ -7,15 +7,13 @@
 #include <arch/cpu.h>
 #include <memory.h>
 #include <syscall.h>
+#include <debug.h>
 
 // #define DEBUG_GICV2
 
 #ifdef DEBUG_GICV2
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 void *dist_base;

arch/arm64/kernel/irq-gic-v3.c

@@ -7,17 +7,15 @@
 #include <cputype.h>
 #include <process.h>
 #include <syscall.h>
+#include <debug.h>
 
 //#define DEBUG_GICV3
 
 #define USE_CAVIUM_THUNDER_X
 
 #ifdef DEBUG_GICV3
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 #ifdef USE_CAVIUM_THUNDER_X

arch/arm64/kernel/memory.c

@@ -14,9 +14,7 @@
 #include <context.h>
 #include <kmalloc.h>
 #include <vdso.h>
-
+#include <debug.h>
-#define	dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 
 #define NOT_IMPLEMENTED()  do { kprintf("%s is not implemented\n", __func__); while(1);} while(0)
 
@@ -2924,17 +2922,12 @@ int read_process_vm(struct process_vm *vm, void *kdst, const void *usrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, to: %p, pa: %p,"
 				"cpsize: %d\n", __FUNCTION__, to, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_ACTIVE);
 			memcpy(to, va, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);
@@ -3007,17 +3000,12 @@ int write_process_vm(struct process_vm *vm, void *udst, const void *ksrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, from: %p,"
 				"pa: %p, cpsize: %d\n", __FUNCTION__, from, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_WRITABLE|PTATTR_ACTIVE);
 			memcpy(va, from, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);
@@ -3078,17 +3066,12 @@ int patch_process_vm(struct process_vm *vm, void *udst, const void *ksrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, from: %p,"
 				"pa: %p, cpsize: %d\n", __FUNCTION__, from, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_WRITABLE|PTATTR_ACTIVE);
 			memcpy(va, from, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);

arch/arm64/kernel/perfctr.c

@@ -93,21 +93,50 @@ int ihk_mc_perfctr_init(int counter, uint64_t config, int mode)
 	return ret;
 }
 
-int ihk_mc_perfctr_start(int counter)
+int ihk_mc_perfctr_start(unsigned long counter_mask)
 {
-	int ret;
+	int ret = 0;
-	ret = cpu_pmu.enable_counter(counter);
+	int counter;
-	return ret;
+	unsigned long counter_bit;
+
+	for (counter = 0, counter_bit = 1;
+	     counter_bit < counter_mask;
+	     counter++, counter_bit <<= 1) {
+		if (!(counter_mask & counter_bit))
+			continue;
+
+		ret = cpu_pmu.enable_counter(counter_mask);
+		if (ret < 0)
+			break;
+	}
+
+	return ret < 0 ? ret : 0;
 }
 
-int ihk_mc_perfctr_stop(int counter)
+int ihk_mc_perfctr_stop(unsigned long counter_mask)
 {
-	cpu_pmu.disable_counter(counter);
+	int ret = 0;
+	int counter;
+	unsigned long counter_bit;
 
-	// ihk_mc_perfctr_startが呼ばれるときには、
+	for (counter = 0, counter_bit = 1;
-	// init系関数が呼ばれるのでdisableにする。
+	     counter_bit < counter_mask;
-	cpu_pmu.disable_intens(counter);
+	     counter++, counter_bit <<= 1) {
-	return 0;
+		if (!(counter_mask & counter_bit))
+			continue;
+
+		ret = cpu_pmu.disable_counter(counter);
+		if (ret < 0)
+			break;
+
+		// ihk_mc_perfctr_startが呼ばれるときには、
+		// init系関数が呼ばれるのでdisableにする。
+		ret = cpu_pmu.disable_intens(counter);
+		if (ret < 0)
+			break;
+	}
+
+	return ret < 0 ? ret : 0;
 }
 
 int ihk_mc_perfctr_reset(int counter)

arch/arm64/kernel/perfctr_armv8pmu.c

@@ -4,16 +4,14 @@
 #include <ihk/perfctr.h>
 #include <errno.h>
 #include <ihk/debug.h>
+#include <debug.h>
 
 #define BIT(nr) (1UL << (nr))
 
 //#define DEBUG_PRINT_PMU
 #ifdef DEBUG_PRINT_PMU
-#define	dkprintf(...) kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...) kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...) kprintf(__VA_ARGS__)
 #endif
 
 

arch/arm64/kernel/psci.c

@@ -21,15 +21,13 @@
 #include <ihk/debug.h>
 #include <compiler.h>
 #include <lwk/compiler.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_PSCI
 
 #ifdef DEBUG_PRINT_PSCI
-#define dkprintf(...) kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf(...) kprintf(__VA_ARGS__)
 #endif
 
 #define PSCI_POWER_STATE_TYPE_POWER_DOWN	1

arch/arm64/kernel/ptrace.c

@@ -11,22 +11,17 @@
 #include <hwcap.h>
 #include <string.h>
 #include <thread_info.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_SC
 
 #ifdef DEBUG_PRINT_SC
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
 #endif
 
 #define NOT_IMPLEMENTED()  do { kprintf("%s is not implemented\n", __func__); while(1);} while(0)
 
-#define BUG_ON(condition) do { if (condition) { kprintf("PANIC: %s: %s(line:%d)\n",\
-				__FILE__, __FUNCTION__, __LINE__); panic(""); } } while(0)
-
 extern void save_debugreg(unsigned long *debugreg);
 extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern int interrupt_from_user(void *);
@@ -959,11 +954,7 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	}
 	thread->exit_status = sig;
 	/* Transition thread state */
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 	proc->status = PS_DELAY_TRACED;
-#else /* POSTK_DEBUG_TEMP_FIX_41 */
-	proc->status = PS_TRACED;
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 	thread->status = PS_TRACED;
 	proc->ptrace &= ~PT_TRACE_SYSCALL;
 	if (sig == SIGSTOP || sig == SIGTSTP ||
@@ -982,10 +973,6 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	info._sifields._sigchld.si_pid = thread->tid;
 	info._sifields._sigchld.si_status = thread->exit_status;
 	do_kill(cpu_local_var(current), parent_pid, -1, SIGCHLD, &info, 0);
-#ifndef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
-	/* Wake parent (if sleeping in wait4()) */
-	waitq_wakeup(&proc->parent->waitpid_q);
-#endif /* !POSTK_DEBUG_TEMP_FIX_41 */
 
 	dkprintf("ptrace_report_signal,sleeping\n");
 	/* Sleep */

arch/arm64/kernel/syscall.c

@@ -14,6 +14,8 @@
 #include <prctl.h>
 #include <limits.h>
 #include <syscall.h>
+#include <uio.h>
+#include <debug.h>
 
 extern void ptrace_report_signal(struct thread *thread, int sig);
 extern void clear_single_step(struct thread *thread);
@@ -27,18 +29,12 @@ static void __check_signal(unsigned long rc, void *regs, int num, int irq_disabl
 //#define DEBUG_PRINT_SC
 
 #ifdef DEBUG_PRINT_SC
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
 #endif
 
 #define NOT_IMPLEMENTED()  do { kprintf("%s is not implemented\n", __func__); while(1);} while(0)
 
-#define BUG_ON(condition) do { if (condition) { kprintf("PANIC: %s: %s(line:%d)\n",\
-				__FILE__, __FUNCTION__, __LINE__); panic(""); } } while(0)
-
 uintptr_t debug_constants[] = {
 	sizeof(struct cpu_local_var),
 	offsetof(struct cpu_local_var, current),
@@ -59,7 +55,7 @@ static int cpuid_head = 1;
 
 extern int num_processors;
 
-int obtain_clone_cpuid(cpu_set_t *cpu_set) {
+int obtain_clone_cpuid(cpu_set_t *cpu_set, int use_last) {
 	int min_queue_len = -1;
 	int i, min_cpu = -1;
 
@@ -1177,19 +1173,10 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				/* Reap and set new signal_flags */
 				proc->signal_flags = SIGNAL_STOP_STOPPED;
 
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 				proc->status = PS_DELAY_STOPPED;
-#else /* POSTK_DEBUG_TEMP_FIX_41 */
-				proc->status = PS_STOPPED;
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 				thread->status = PS_STOPPED;
 				mcs_rwlock_writer_unlock(&proc->update_lock, &lock);	
 
-#ifndef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
-				/* Wake up the parent who tried wait4 and sleeping */
-				waitq_wakeup(&proc->parent->waitpid_q);
-#endif /* !POSTK_DEBUG_TEMP_FIX_41 */
-
 				dkprintf("do_signal(): pid: %d, tid: %d SIGSTOP, sleeping\n", 
 					proc->pid, thread->tid);
 				/* Sleep */
@@ -1206,19 +1193,10 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			/* Update thread state in fork tree */
 			mcs_rwlock_writer_lock(&proc->update_lock, &lock);	
 			thread->exit_status = SIGTRAP;
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 			proc->status = PS_DELAY_TRACED;
-#else /* POSTK_DEBUG_TEMP_FIX_41 */
-			proc->status = PS_TRACED;
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 			thread->status = PS_TRACED;
 			mcs_rwlock_writer_unlock(&proc->update_lock, &lock);	
 
-#ifndef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
-			/* Wake up the parent who tried wait4 and sleeping */
-			waitq_wakeup(&thread->proc->parent->waitpid_q);
-#endif /* !POSTK_DEBUG_TEMP_FIX_41 */
-
 			/* Sleep */
 			dkprintf("do_signal,SIGTRAP,sleeping\n");
 
@@ -1594,7 +1572,7 @@ done:
 		return 0;
 	}
 
-	if (tthread->thread_offloaded) {
+	if (tthread->uti_state == UTI_STATE_RUNNING_IN_LINUX) {
 		interrupt_syscall(tthread, sig);
 		release_thread(tthread);
 		return 0;
@@ -1729,7 +1707,7 @@ SYSCALL_DECLARE(mmap)
 		| MAP_NONBLOCK		// 0x10000
 		;
 
-	const intptr_t addr0 = ihk_mc_syscall_arg0(ctx);
+	const uintptr_t addr0 = ihk_mc_syscall_arg0(ctx);
 	const size_t len0 = ihk_mc_syscall_arg1(ctx);
 	const int prot = ihk_mc_syscall_arg2(ctx);
 	const int flags0 = ihk_mc_syscall_arg3(ctx);
@@ -1738,7 +1716,7 @@ SYSCALL_DECLARE(mmap)
 	struct thread *thread = cpu_local_var(current);
 	struct vm_regions *region = &thread->vm->region;
 	int error;
-	intptr_t addr = 0;
+	uintptr_t addr = 0;
 	size_t len;
 	int flags = flags0;
 	size_t pgsize;

arch/arm64/kernel/vdso.c

@@ -14,15 +14,13 @@
 #include <ihk/debug.h>
 #include <ikc/queue.h>
 #include <vdso.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_VDSO
 
 #ifdef DEBUG_PRINT_VDSO
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 #ifdef POSTK_DEBUG_ARCH_DEP_52

arch/arm64/kernel/vdso/gettimeofday.c

@@ -1,5 +1,7 @@
 /* gettimeofday.c COPYRIGHT FUJITSU LIMITED 2016 */
 
+#include <affinity.h>
+#include <arch-memory.h>
 #include <time.h>
 #include <syscall.h>
 #include <registers.h>

arch/x86_64/elfboot/raw.lds

@@ -9,29 +9,29 @@ PHDRS
 SECTIONS
 {
 	. = SIZEOF_HEADERS;
-    . = ALIGN(4096);
+	. = ALIGN(4096);
 	.text :	{
-	      *(.text)
+		*(.text)
 	} :text
 	.data : {
-	      *(.data)
+		*(.data)
-	      *(.data.*)
+		*(.data.*)
 	} :data
 	.rodata : {
-	      *(.rodata .rodata.*)
+		*(.rodata .rodata.*)
 	} :data
 
 	. = ALIGN(8);
 	.bss : {
-	     _bss_start = .;
+		_bss_start = .;
-	     *(.bss .bss.*)
+		*(.bss .bss.*)
-	     _bss_end = .;
+		_bss_end = .;
-         . = ALIGN(4096);
+		. = ALIGN(4096);
-         _stack_end = .;
+		_stack_end = .;
-    } :data
+	} :data
 
 	/DISCARD/ : {
-	     *(.eh_frame)
+		*(.eh_frame)
-         *(.note.gnu.build-id)
+		*(.note.gnu.build-id)
 	}
 }

arch/x86_64/kernel/cpu.c

@@ -31,6 +31,7 @@
 #include <prctl.h>
 #include <page.h>
 #include <kmalloc.h>
+#include <debug.h>
 
 #define LAPIC_ID            0x020
 #define LAPIC_TIMER         0x320
@@ -69,11 +70,8 @@
 //#define DEBUG_PRINT_CPU
 
 #ifdef DEBUG_PRINT_CPU
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf kprintf
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf kprintf
 #endif
 
 static void *lapic_vp;
@@ -96,6 +94,8 @@ int gettime_local_support = 0;
 extern int ihk_mc_pt_print_pte(struct page_table *pt, void *virt);
 extern int kprintf(const char *format, ...);
 extern int interrupt_from_user(void *);
+extern void perf_start(struct mc_perf_event *event);
+extern void perf_reset(struct mc_perf_event *event);
 
 static struct idt_entry{
 	uint32_t desc[4];
@@ -847,9 +847,6 @@ void setup_x86_ap(void (*next_func)(void))
 }
 
 void arch_show_interrupt_context(const void *reg);
-void set_signal(int sig, void *regs, struct siginfo *info);
-void check_signal(unsigned long, void *, int);
-void check_sig_pending();
 extern void tlb_flush_handler(int vector);
 
 void __show_stack(uintptr_t *sp) {
@@ -877,7 +874,7 @@ void interrupt_exit(struct x86_user_context *regs)
 		cpu_enable_interrupt();
 		check_sig_pending();
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 	else {
 		check_sig_pending();
@@ -1010,6 +1007,12 @@ void handle_interrupt(int vector, struct x86_user_context *regs)
 	set_cputime(interrupt_from_user(regs)? 0: 1);
 
 	--v->in_interrupt;
+
+	/* for migration by IPI */
+	if (v->flags & CPU_FLAG_NEED_MIGRATE) {
+		schedule();
+		check_signal(0, regs, 0);
+	}
 }
 
 void gpe_handler(struct x86_user_context *regs)
@@ -1644,12 +1647,10 @@ int ihk_mc_interrupt_cpu(int cpu, int vector)
 	return 0;
 }
 
-#ifdef POSTK_DEBUG_ARCH_DEP_22
-extern void perf_start(struct mc_perf_event *event);
-extern void perf_reset(struct mc_perf_event *event);
 struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 {
 	struct thread *last;
+	struct mcs_rwlock_node_irqsave lock;
 
 	dkprintf("[%d] schedule: tlsblock_base: 0x%lX\n",
 	         ihk_mc_get_processor_id(), next->tlsblock_base);
@@ -1668,7 +1669,7 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 	}
 
 #ifdef PROFILE_ENABLE
-	if (prev->profile && prev->profile_start_ts != 0) {
+	if (prev && prev->profile && prev->profile_start_ts != 0) {
 		prev->profile_elapsed_ts +=
 			(rdtsc() - prev->profile_start_ts);
 		prev->profile_start_ts = 0;
@@ -1680,6 +1681,28 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 #endif
 
 	if (prev) {
+		mcs_rwlock_writer_lock(&prev->proc->update_lock, &lock);
+		if (prev->proc->status & (PS_DELAY_STOPPED | PS_DELAY_TRACED)) {
+			switch (prev->proc->status) {
+			case PS_DELAY_STOPPED:
+				prev->proc->status = PS_STOPPED;
+				break;
+			case PS_DELAY_TRACED:
+				prev->proc->status = PS_TRACED;
+				break;
+			default:
+				break;
+			}
+			mcs_rwlock_writer_unlock(&prev->proc->update_lock,
+						&lock);
+
+			/* Wake up the parent who tried wait4 and sleeping */
+			waitq_wakeup(&prev->proc->parent->waitpid_q);
+		} else {
+			mcs_rwlock_writer_unlock(&prev->proc->update_lock,
+						&lock);
+		}
+
 		last = ihk_mc_switch_context(&prev->ctx, &next->ctx, prev);
 	}
 	else {
@@ -1687,7 +1710,6 @@ struct thread *arch_switch_context(struct thread *prev, struct thread *next)
 	}
 	return last;
 }
-#endif
 
 /*@
   @ requires \valid(thread);
@@ -1762,14 +1784,6 @@ void copy_fp_regs(struct thread *from, struct thread *to)
 	}
 }
 
-#ifdef POSTK_DEBUG_TEMP_FIX_19
-void
-clear_fp_regs(struct thread *thread)
-{
-	return;
-}
-#endif /* POSTK_DEBUG_TEMP_FIX_19 */
-
 /*@
   @ requires \valid(thread);
   @ assigns thread->fp_regs;
@@ -1777,8 +1791,11 @@ clear_fp_regs(struct thread *thread)
 void
 restore_fp_regs(struct thread *thread)
 {
-	if (!thread->fp_regs)
+	if (!thread->fp_regs) {
+		// only clear fpregs.
+		clear_fp_regs();
 		return;
+	}
 
 	if (xsave_available) {
 		unsigned int low, high;
@@ -1797,6 +1814,13 @@ restore_fp_regs(struct thread *thread)
 	//release_fp_regs(thread);
 }
 
+void clear_fp_regs(void)
+{
+	struct cpu_local_var *v = get_this_cpu_local_var();
+
+	restore_fp_regs(&v->idle);
+}
+
 ihk_mc_user_context_t *lookup_user_context(struct thread *thread)
 {
 	ihk_mc_user_context_t *uctx = thread->uctx;

arch/x86_64/kernel/gencore.c

@@ -6,6 +6,7 @@
 #include <process.h>
 #include <string.h>
 #include <elfcore.h>
+#include <debug.h>
 
 #define	align32(x) ((((x) + 3) / 4) * 4)
 #define	alignpage(x) ((((x) + (PAGE_SIZE) - 1) / (PAGE_SIZE)) * (PAGE_SIZE))
@@ -13,13 +14,16 @@
 //#define DEBUG_PRINT_GENCORE
 
 #ifdef DEBUG_PRINT_GENCORE
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
+/* Exclude reserved (mckernel's internal use), device file,
+ * hole created by mprotect
+ */
+#define GENCORE_RANGE_IS_INACCESSIBLE(range) \
+	    ((range->flag & (VR_RESERVED | VR_MEMTYPE_UC | VR_DONTDUMP)))
+
 /*
  * Generate a core file image, which consists of many chunks.
  * Returns an allocated table, an etnry of which is a pair of the address 
@@ -309,12 +313,10 @@ int gencore(struct thread *thread, void *regs,
 
 		dkprintf("start:%lx end:%lx flag:%lx objoff:%lx\n", 
 			 range->start, range->end, range->flag, range->objoff);
-		/* We omit reserved areas because they are only for
+
-		   mckernel's internal use. */		   
+		if (GENCORE_RANGE_IS_INACCESSIBLE(range)) {
-		if (range->flag & VR_RESERVED)
-			continue;
-		if (range->flag & VR_DONTDUMP)
 			continue;
+		}
 		/* We need a chunk for each page for a demand paging area.
 		   This can be optimized for spacial complexity but we would
 		   lose simplicity instead. */
@@ -403,8 +405,9 @@ int gencore(struct thread *thread, void *regs,
 		unsigned long flag = range->flag;
 		unsigned long size = range->end - range->start;
 
-		if (range->flag & VR_RESERVED)
+		if (GENCORE_RANGE_IS_INACCESSIBLE(range)) {
 			continue;
+		}
 
 		ph[i].p_type = PT_LOAD;
 		ph[i].p_flags = ((flag & VR_PROT_READ) ? PF_R : 0)
@@ -446,8 +449,9 @@ int gencore(struct thread *thread, void *regs,
 
 		unsigned long phys;
 
-		if (range->flag & VR_RESERVED)
+		if (GENCORE_RANGE_IS_INACCESSIBLE(range)) {
 			continue;
+		}
 		if (range->flag & VR_DEMAND_PAGING) {
 			/* Just an ad hoc kluge. */
 			unsigned long p, start, phys;

arch/x86_64/kernel/include/arch-futex.h

@@ -64,12 +64,13 @@ static inline int futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval,
 	return oldval;
 }
 
-static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
+static inline int futex_atomic_op_inuser(int encoded_op,
+					 int __user *uaddr)
 {
 	int op = (encoded_op >> 28) & 7;
 	int cmp = (encoded_op >> 24) & 15;
-	int oparg = (encoded_op << 8) >> 20;
+	int oparg = (encoded_op & 0x00fff000) >> 12;
-	int cmparg = (encoded_op << 20) >> 20;
+	int cmparg = encoded_op & 0xfff;
 	int oldval = 0, ret, tem;
 
 	if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28))

arch/x86_64/kernel/include/arch-lock.h

@@ -6,6 +6,7 @@
 
 #include <ihk/cpu.h>
 #include <ihk/atomic.h>
+#include <lwk/compiler.h>
 
 //#define DEBUG_SPINLOCK
 //#define DEBUG_MCS_RWLOCK
@@ -14,7 +15,17 @@
 int __kprintf(const char *format, ...);
 #endif
 
-typedef int ihk_spinlock_t;
+typedef unsigned short __ticket_t;
+typedef unsigned int __ticketpair_t;
+
+typedef struct ihk_spinlock {
+	union {
+		__ticketpair_t head_tail;
+		struct __raw_tickets {
+			__ticket_t head, tail;
+		} tickets;
+	};
+} ihk_spinlock_t;
 
 extern void preempt_enable(void);
 extern void preempt_disable(void);
@@ -23,9 +34,61 @@ extern void preempt_disable(void);
 
 static void ihk_mc_spinlock_init(ihk_spinlock_t *lock)
 {
-	*lock = 0;
+	lock->head_tail = 0;
+}
+#define SPIN_LOCK_UNLOCKED { .head_tail = 0 }
+
+
+#ifdef DEBUG_SPINLOCK
+#define ihk_mc_spinlock_trylock_noirq(l) { int rc;						\
+__kprintf("[%d] call ihk_mc_spinlock_trylock_noirq %p %s:%d\n", ihk_mc_get_processor_id(), (l), __FILE__, __LINE__); \
+rc = __ihk_mc_spinlock_trylock_noirq(l); \
+ __kprintf("[%d] ret ihk_mc_spinlock_trylock_noirq\n", ihk_mc_get_processor_id()); rc; \
+}
+#else
+#define ihk_mc_spinlock_trylock_noirq __ihk_mc_spinlock_trylock_noirq
+#endif
+
+static int __ihk_mc_spinlock_trylock_noirq(ihk_spinlock_t *lock)
+{
+	ihk_spinlock_t cur = { .head_tail = lock->head_tail };
+	ihk_spinlock_t next = { .tickets.head = cur.tickets.head, .tickets.tail = cur.tickets.tail + 2 };
+	int success;
+
+	if (cur.tickets.head != cur.tickets.tail) {
+		return 0;
+	}
+
+	preempt_disable();
+
+	/* Use the same increment amount as other functions! */
+	success = __sync_bool_compare_and_swap((__ticketpair_t*)lock, cur.head_tail, next.head_tail);
+
+	if (!success) {
+		preempt_enable();
+	}
+	return success;
+}
+
+#ifdef DEBUG_SPINLOCK
+#define ihk_mc_spinlock_trylock(l, result) ({ unsigned long rc;		\
+__kprintf("[%d] call ihk_mc_spinlock_trylock %p %s:%d\n", ihk_mc_get_processor_id(), (l), __FILE__, __LINE__); \
+ rc = __ihk_mc_spinlock_trylock(l, result);									\
+__kprintf("[%d] ret ihk_mc_spinlock_trylock\n", ihk_mc_get_processor_id()); rc;\
+})
+#else
+#define ihk_mc_spinlock_trylock __ihk_mc_spinlock_trylock
+#endif
+static unsigned long __ihk_mc_spinlock_trylock(ihk_spinlock_t *lock, int *result)
+{
+	unsigned long flags;
+	
+	flags = cpu_disable_interrupt_save();
+
+	*result = __ihk_mc_spinlock_trylock_noirq(lock);
+
+	return flags;
 }
-#define SPIN_LOCK_UNLOCKED 0
 
 #ifdef DEBUG_SPINLOCK
 #define ihk_mc_spinlock_lock_noirq(l) { \
@@ -39,40 +102,24 @@ __kprintf("[%d] ret ihk_mc_spinlock_lock_noirq\n", ihk_mc_get_processor_id()); \
 
 static void __ihk_mc_spinlock_lock_noirq(ihk_spinlock_t *lock)
 {
-	int inc = 0x00010000;
+	register struct __raw_tickets inc = { .tail = 0x0002 };
-	int tmp;
-
-#if 0
-	asm volatile("lock ; xaddl %0, %1\n"
-	             "movzwl %w0, %2\n\t"
-	             "shrl $16, %0\n\t"
-	             "1:\t"
-	             "cmpl %0, %2\n\t"
-	             "je 2f\n\t"
-	             "rep ; nop\n\t"
-	             "movzwl %1, %2\n\t"
-	             "jmp 1b\n"
-	             "2:"
-	             : "+Q" (inc), "+m" (*lock), "=r" (tmp) : : "memory", "cc");
-#endif
 
 	preempt_disable();
 
-	asm volatile("lock; xaddl %0, %1\n"
+	asm volatile ("lock xaddl %0, %1\n"
-			"movzwl %w0, %2\n\t"
+			: "+r" (inc), "+m" (*(lock)) : : "memory", "cc");
-			"shrl $16, %0\n\t"
-			"1:\t"
-			"cmpl %0, %2\n\t"
-			"je 2f\n\t"
-			"rep ; nop\n\t"
-			"movzwl %1, %2\n\t"
-			/* don't need lfence here, because loads are in-order */
-			"jmp 1b\n"
-			"2:"
-			: "+r" (inc), "+m" (*lock), "=&r" (tmp)
-			:
-			: "memory", "cc");
 
+	if (inc.head == inc.tail)
+		goto out;
+
+	for (;;) {
+		if (*((volatile __ticket_t *)&lock->tickets.head) == inc.tail)
+			goto out;
+		cpu_pause();
+	}
+
+out:
+	barrier();	/* make sure nothing creeps before the lock is taken */
 }
 
 #ifdef DEBUG_SPINLOCK
@@ -106,7 +153,10 @@ __kprintf("[%d] ret ihk_mc_spinlock_unlock_noirq\n", ihk_mc_get_processor_id());
 #endif
 static void __ihk_mc_spinlock_unlock_noirq(ihk_spinlock_t *lock)
 {
-	asm volatile ("lock incw %0" : "+m"(*lock) : : "memory", "cc");
+	__ticket_t inc = 0x0002;
+
+	asm volatile ("lock addw %1, %0\n"
+			: "+m" (lock->tickets.head) : "ri" (inc) : "memory", "cc");
 
 	preempt_enable();
 }
@@ -602,4 +652,9 @@ __mcs_rwlock_reader_unlock(struct mcs_rwlock_lock *lock, struct mcs_rwlock_node_
 #endif
 }
 
+static inline int irqflags_can_interrupt(unsigned long flags)
+{
+	return !!(flags & 0x200);
+}
+
 #endif

arch/x86_64/kernel/include/arch-memory.h

@@ -42,16 +42,34 @@
 
 #define USER_END           0x0000800000000000UL
 #define TASK_UNMAPPED_BASE 0x00002AAAAAA00000UL
+
+/*
+ * Canonical negative addresses (i.e., the smallest kernel virtual address)
+ * on x86 64 bit mode (in its most restricted 48 bit format) starts from
+ * 0xffff800000000000, but Linux starts mapping physical memory at 0xffff880000000000.
+ * The 0x80000000000 long gap (8TBs, i.e., 16 PGD level entries in the page tables)
+ * is used for Xen hyervisor (see arch/x86/include/asm/page.h) and that is
+ * what we utilize for McKernel.
+ * This gives us the benefit of being able to use Linux kernel virtual
+ * addresses identically as in Linux.
+ *
+ * NOTE: update these also in eclair.c when modified!
+ */
 #define MAP_ST_START       0xffff800000000000UL
-#define MAP_VMAP_START     0xfffff00000000000UL
+#define MAP_VMAP_START     0xffff850000000000UL
-#define MAP_FIXED_START    0xffffffff70000000UL
+#define MAP_FIXED_START    0xffff860000000000UL
-#define MAP_KERNEL_START   0xffffffff80000000UL
+#define LINUX_PAGE_OFFSET  0xffff880000000000UL
+/*
+ * MAP_KERNEL_START is 8MB below MODULES_END in Linux.
+ * Placing the LWK image in the virtual address space at the end of
+ * the Linux modules section enables us to map the LWK TEXT in Linux
+ * as well, so that Linux can also call into LWK text.
+ */
+#define MAP_KERNEL_START   0xFFFFFFFFFE800000UL
 #define STACK_TOP(region)  ((region)->user_end)
 
 #define MAP_VMAP_SIZE      0x0000000100000000UL
 
-#define KERNEL_PHYS_OFFSET MAP_ST_START
-
 #define PTL4_SHIFT         39
 #define PTL4_SIZE          (1UL << PTL4_SHIFT)
 #define PTL3_SHIFT         30

arch/x86_64/kernel/include/syscall_list.h

@@ -114,7 +114,7 @@ SYSCALL_HANDLED(160, setrlimit)
 SYSCALL_HANDLED(164, settimeofday)
 SYSCALL_HANDLED(186, gettid)
 SYSCALL_HANDLED(200, tkill)
-SYSCALL_DELEGATED(201, time)
+SYSCALL_HANDLED(201, time)
 SYSCALL_HANDLED(202, futex)
 SYSCALL_HANDLED(203, sched_setaffinity)
 SYSCALL_HANDLED(204, sched_getaffinity)
@@ -161,6 +161,7 @@ SYSCALL_HANDLED(__NR_profile, profile)
 SYSCALL_HANDLED(730, util_migrate_inter_kernel)
 SYSCALL_HANDLED(731, util_indicate_clone)
 SYSCALL_HANDLED(732, get_system)
+SYSCALL_HANDLED(733, util_register_desc)
 
 /* McKernel Specific */
 SYSCALL_HANDLED(801, swapout)

arch/x86_64/kernel/local.c

@@ -107,9 +107,17 @@ void init_boot_processor_local(void)
   @ ensures \result == %gs;
   @ assigns \nothing;
   */
+extern int num_processors;
 int ihk_mc_get_processor_id(void)
 {
 	int id;
+	void *gs;
+
+	gs = (void *)rdmsr(MSR_GS_BASE);
+	if (gs < (void *)locals ||
+			gs > ((void *)locals + LOCALS_SPAN * num_processors)) {
+		return -1;
+	}
 
 	asm volatile("movl %%gs:0, %0" : "=r"(id));
 

arch/x86_64/kernel/memory.c

@@ -25,15 +25,13 @@
 #include <cls.h>
 #include <kmalloc.h>
 #include <rusage_private.h>
+#include <debug.h>
 
 //#define DEBUG
 
 #ifdef DEBUG
-#define	dkprintf(...)	do { kprintf(__VA_ARGS__); } while (0)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	do { kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { } while (0)
-#define ekprintf(...) do { kprintf(__VA_ARGS__); } while (0)
 #endif
 
 static char *last_page;
@@ -41,6 +39,8 @@ extern char _head[], _end[];
 
 extern unsigned long x86_kernel_phys_base;
 
+int safe_kernel_map = 0;
+
 /* Arch specific early allocation routine */
 void *early_alloc_pages(int nr_pages)
 {
@@ -109,6 +109,7 @@ struct page_table {
 };
 
 static struct page_table *init_pt;
+static int init_pt_loaded = 0;
 static ihk_spinlock_t init_pt_lock;
 
 static int use_1gb_page = 0;
@@ -167,30 +168,6 @@ static unsigned long setup_l3(struct page_table *pt,
 	return virt_to_phys(pt);
 }
 
-static void init_normal_area(struct page_table *pt)
-{
-	unsigned long map_start, map_end, phys, pt_phys;
-	int ident_index, virt_index;
-
-	map_start = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0);
-	map_end = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0);
-
-	kprintf("map_start = %lx, map_end = %lx\n", map_start, map_end);
-	ident_index = map_start >> PTL4_SHIFT;
-	virt_index = (MAP_ST_START >> PTL4_SHIFT) & (PT_ENTRIES - 1);
-
-	memset(pt, 0, sizeof(struct page_table));
-
-	for (phys = (map_start & ~(PTL4_SIZE - 1)); phys < map_end;
-	     phys += PTL4_SIZE) {
-		pt_phys = setup_l3(ihk_mc_alloc_pages(1, IHK_MC_AP_CRITICAL), phys,
-		                   map_start, map_end);
-
-		pt->entry[ident_index++] = pt_phys | PFL4_PDIR_ATTR;
-		pt->entry[virt_index++] = pt_phys | PFL4_PDIR_ATTR;
-	}
-}
-
 static struct page_table *__alloc_new_pt(ihk_mc_ap_flag ap_flag)
 {
 	struct page_table *newpt = ihk_mc_alloc_pages(1, ap_flag);
@@ -258,6 +235,11 @@ static unsigned long attr_to_l1attr(enum ihk_mc_pt_attribute attr)
 	}
 }
 
+#define PTLX_SHIFT(index) PTL ## index ## _SHIFT
+
+#define GET_VIRT_INDEX(virt, index, dest) \
+	dest = ((virt) >> PTLX_SHIFT(index)) & (PT_ENTRIES - 1)
+
 #define GET_VIRT_INDICES(virt, l4i, l3i, l2i, l1i) \
 	l4i = ((virt) >> PTL4_SHIFT) & (PT_ENTRIES - 1); \
 	l3i = ((virt) >> PTL3_SHIFT) & (PT_ENTRIES - 1); \
@@ -1518,12 +1500,12 @@ static int clear_range_l1(void *args0, pte_t *ptep, uint64_t base,
 	if (page) {
 		dkprintf("%s: page=%p,is_in_memobj=%d,(old & PFL1_DIRTY)=%lx,memobj=%p,args->memobj->flags=%x\n", __FUNCTION__, page, page_is_in_memobj(page), (old & PFL1_DIRTY), args->memobj, args->memobj ? args->memobj->flags : -1);
 	}
-	if (page && page_is_in_memobj(page) && (old & PFL1_DIRTY) && (args->memobj) &&
+	if (page && page_is_in_memobj(page) && pte_is_dirty(&old, PTL1_SIZE) &&
-			!(args->memobj->flags & MF_ZEROFILL)) {
+			args->memobj && !(args->memobj->flags & MF_ZEROFILL)) {
 		memobj_flush_page(args->memobj, phys, PTL1_SIZE);
 	}
 
-	if (!(old & PFL1_FILEOFF)) {
+	if (!pte_is_fileoff(&old, PTL1_SIZE)) {
 		if(args->free_physical) {
 			if (!page) {
 				/* Anonymous || !XPMEM attach */
@@ -1585,11 +1567,11 @@ static int clear_range_l2(void *args0, pte_t *ptep, uint64_t base,
 			page = phys_to_page(phys);
 		}
 
-		if (page && page_is_in_memobj(page) && (old & PFL2_DIRTY)) {
+		if (page && page_is_in_memobj(page) && pte_is_dirty(&old, PTL2_SIZE)) {
 			memobj_flush_page(args->memobj, phys, PTL2_SIZE);
 		}
 
-		if (!(old & PFL2_FILEOFF)) {
+		if (!pte_is_fileoff(&old, PTL2_SIZE)) {
 			if(args->free_physical) {
 				if (!page) {
 					/* Anonymous || !XPMEM attach */
@@ -1666,13 +1648,13 @@ static int clear_range_l3(void *args0, pte_t *ptep, uint64_t base,
 			page = phys_to_page(phys);
 		}
 
-		if (page && page_is_in_memobj(page) && (old & PFL3_DIRTY)) {
+		if (page && page_is_in_memobj(page) && pte_is_dirty(&old, PTL3_SIZE)) {
 			memobj_flush_page(args->memobj, phys, PTL3_SIZE);
 		}
 
 		dkprintf("%s: phys=%ld, pte_get_phys(&old),PTL3_SIZE\n", __FUNCTION__, pte_get_phys(&old));
 
-		if (!(old & PFL3_FILEOFF)) {
+		if (!pte_is_fileoff(&old, PTL3_SIZE)) {
 			if(args->free_physical) {
 				if (!page) {
 					/* Anonymous || !XPMEM attach */
@@ -2540,6 +2522,82 @@ static void init_fixed_area(struct page_table *pt)
 	return;
 }
 
+static void init_normal_area(struct page_table *pt)
+{
+	unsigned long map_start, map_end, phys;
+	void *virt;
+
+	map_start = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0);
+	map_end = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0);
+	virt = (void *)MAP_ST_START + map_start;
+
+	kprintf("map_start = %lx, map_end = %lx, virt %lx\n",
+		map_start, map_end, virt);
+
+	for (phys = map_start; phys < map_end; phys += LARGE_PAGE_SIZE) {
+		if (set_pt_large_page(pt, virt, phys, PTATTR_WRITABLE) != 0) {
+			kprintf("%s: error setting mapping for 0x%lx\n",
+					__func__, virt);
+		}
+		virt += LARGE_PAGE_SIZE;
+	}
+}
+
+static void init_linux_kernel_mapping(struct page_table *pt)
+{
+	unsigned long map_start, map_end, phys;
+	void *virt;
+	int nr_memory_chunks, chunk_id, numa_id;
+
+	/* In case of safe_kernel_map option (safe_kernel_map == 1),
+	 * processing to prevent destruction of the memory area on Linux side
+	 * is executed */
+	if (safe_kernel_map == 0) {
+		kprintf("Straight-map entire physical memory\n");
+
+		/* Map 2 TB for now */
+		map_start = 0;
+		map_end = 0x20000000000;
+
+		virt = (void *)LINUX_PAGE_OFFSET;
+
+		kprintf("Linux kernel virtual: 0x%lx - 0x%lx -> 0x%lx - 0x%lx\n",
+			LINUX_PAGE_OFFSET, LINUX_PAGE_OFFSET + map_end, 0, map_end);
+
+		for (phys = map_start; phys < map_end; phys += LARGE_PAGE_SIZE) {
+			if (set_pt_large_page(pt, virt, phys, PTATTR_WRITABLE) != 0) {
+				kprintf("%s: error setting mapping for 0x%lx\n", __FUNCTION__, virt);
+			}
+			virt += LARGE_PAGE_SIZE;
+		}
+	} else {
+		kprintf("Straight-map physical memory areas allocated to McKernel\n");
+
+		nr_memory_chunks = ihk_mc_get_nr_memory_chunks();
+		if (nr_memory_chunks == 0) {
+			kprintf("%s: ERROR: No memory chunk available.\n", __FUNCTION__);
+			return;
+		}
+
+		for (chunk_id = 0; chunk_id < nr_memory_chunks; chunk_id++) {
+			if (ihk_mc_get_memory_chunk(chunk_id, &map_start, &map_end, &numa_id)) {
+				kprintf("%s: ERROR: Memory chunk id (%d) out of range.\n", __FUNCTION__, chunk_id);
+				continue;
+			}
+
+			dkprintf("Linux kernel virtual: 0x%lx - 0x%lx -> 0x%lx - 0x%lx\n",
+					 LINUX_PAGE_OFFSET + map_start, LINUX_PAGE_OFFSET + map_end, map_start, map_end);
+
+			virt = (void *)(LINUX_PAGE_OFFSET + map_start);
+			for (phys = map_start; phys < map_end; phys += LARGE_PAGE_SIZE, virt += LARGE_PAGE_SIZE) {
+				if (set_pt_large_page(pt, virt, phys, PTATTR_WRITABLE) != 0) {
+					kprintf("%s: set_pt_large_page() failed for 0x%lx\n", __FUNCTION__, virt);
+				}
+			}
+		}
+	}
+}
+
 void init_text_area(struct page_table *pt)
 {
 	unsigned long __end, phys, virt;
@@ -2624,17 +2682,19 @@ void init_page_table(void)
 	init_pt = ihk_mc_alloc_pages(1, IHK_MC_AP_CRITICAL);
 	ihk_mc_spinlock_init(&init_pt_lock);
 	
-	memset(init_pt, 0, sizeof(PAGE_SIZE));
+	memset(init_pt, 0, sizeof(*init_pt));
 
 	/* Normal memory area */
 	init_normal_area(init_pt);
+	init_linux_kernel_mapping(init_pt);
 	init_fixed_area(init_pt);
 	init_low_area(init_pt);
 	init_text_area(init_pt);
 	init_vsyscall_area(init_pt);
 
 	load_page_table(init_pt);
-	kprintf("Page table is now at %p\n", init_pt);
+	init_pt_loaded = 1;
+	kprintf("Page table is now at 0x%lx\n", init_pt);
 }
 
 extern void __reserve_arch_pages(unsigned long, unsigned long,
@@ -2664,15 +2724,31 @@ unsigned long virt_to_phys(void *v)
 	unsigned long va = (unsigned long)v;
 
 	if (va >= MAP_KERNEL_START) {
+		dkprintf("%s: MAP_KERNEL_START <= 0x%lx <= LINUX_PAGE_OFFSET\n",
+				__FUNCTION__, va);
 		return va - MAP_KERNEL_START + x86_kernel_phys_base;
-	} else {
+	}
+	else if (va >= LINUX_PAGE_OFFSET) {
+		return va - LINUX_PAGE_OFFSET;
+	}
+	else if (va >= MAP_FIXED_START) {
+		return va - MAP_FIXED_START;
+	}
+	else {
+		dkprintf("%s: MAP_ST_START <= 0x%lx <= MAP_FIXED_START\n",
+				__FUNCTION__, va);
 		return va - MAP_ST_START;
 	}
 }
 
 void *phys_to_virt(unsigned long p)
 {
-	return (void *)(p + MAP_ST_START);
+	/* Before loading our own PT use straight mapping */
+	if (!init_pt_loaded) {
+		return (void *)(p + MAP_ST_START);
+	}
+
+	return (void *)(p + LINUX_PAGE_OFFSET);
 }
 
 int copy_from_user(void *dst, const void *src, size_t siz)
@@ -2840,17 +2916,12 @@ int read_process_vm(struct process_vm *vm, void *kdst, const void *usrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, to: %p, pa: %p,"
 				"cpsize: %d\n", __FUNCTION__, to, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_ACTIVE);
 			memcpy(to, va, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);
@@ -2924,17 +2995,12 @@ int write_process_vm(struct process_vm *vm, void *udst, const void *ksrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, from: %p,"
 				"pa: %p, cpsize: %d\n", __FUNCTION__, from, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_ACTIVE);
 			memcpy(va, from, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);
@@ -2995,17 +3061,12 @@ int patch_process_vm(struct process_vm *vm, void *udst, const void *ksrc, size_t
 			return error;
 		}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* NUMA support(memory area determination) */
+		if (!is_mckernel_memory(pa, pa + cpsize)) {
-		if (!is_mckernel_memory(pa)) {
-#else
-		if (pa < ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0) ||
-			pa >= ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0)) {
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 			dkprintf("%s: pa is outside of LWK memory, from: %p,"
 				"pa: %p, cpsize: %d\n", __FUNCTION__, from, pa, cpsize);
 			va = ihk_mc_map_virtual(pa, 1, PTATTR_ACTIVE);
 			memcpy(va, from, cpsize);
-			ihk_mc_unmap_virtual(va, 1, 1);
+			ihk_mc_unmap_virtual(va, 1);
 		}
 		else {
 			va = phys_to_virt(pa);

arch/x86_64/kernel/mikc.c

@@ -30,7 +30,7 @@ int ihk_mc_ikc_init_first_local(struct ihk_ikc_channel_desc *channel,
 
 	memset(channel, 0, sizeof(struct ihk_ikc_channel_desc));
 
-	mikc_queue_pages = ((2 * num_processors * MASTER_IKCQ_PKTSIZE)
+	mikc_queue_pages = ((4 * num_processors * MASTER_IKCQ_PKTSIZE)
 			+ (PAGE_SIZE - 1)) / PAGE_SIZE;
 
 	/* Place both sides in this side */

arch/x86_64/kernel/perfctr.c

@@ -16,20 +16,16 @@
 #include <registers.h>
 #include <mc_perf_event.h>
 #include <config.h>
+#include <debug.h>
 
 extern unsigned int *x86_march_perfmap;
 extern int running_on_kvm(void);
-#ifdef POSTK_DEBUG_TEMP_FIX_31
 int ihk_mc_perfctr_fixed_init(int counter, int mode);
-#endif/*POSTK_DEBUG_TEMP_FIX_31*/
 
 //#define PERFCTR_DEBUG
 #ifdef PERFCTR_DEBUG
-#define	dkprintf(...)	do { kprintf(__VA_ARGS__); } while (0)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	do { kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define	dkprintf(...)	do { } while (0)
-#define	ekprintf(...)	do { kprintf(__VA_ARGS__); } while (0)
 #endif
 
 #define X86_CR4_PCE     0x00000100
@@ -43,11 +39,11 @@ int ihk_mc_perfctr_fixed_init(int counter, int mode);
 		}                                                               \
     } while(0)
 
-int perf_counters_discovered = 0;
+int perf_counters_discovered;
-int X86_IA32_NUM_PERF_COUNTERS = 0;
+int NUM_PERF_COUNTERS;
-unsigned long X86_IA32_PERF_COUNTERS_MASK = 0;
+unsigned long PERF_COUNTERS_MASK;
-int X86_IA32_NUM_FIXED_PERF_COUNTERS = 0;
+int NUM_FIXED_PERF_COUNTERS;
-unsigned long X86_IA32_FIXED_PERF_COUNTERS_MASK = 0;
+unsigned long FIXED_PERF_COUNTERS_MASK;
 
 void x86_init_perfctr(void)
 {
@@ -78,17 +74,17 @@ void x86_init_perfctr(void)
 		op = 0x0a;
 		asm volatile("cpuid" : "=a"(eax),"=b"(ebx),"=c"(ecx),"=d"(edx):"a"(op));
 
-		X86_IA32_NUM_PERF_COUNTERS = ((eax & 0xFF00) >> 8);
+		NUM_PERF_COUNTERS = ((eax & 0xFF00) >> 8);
-		X86_IA32_PERF_COUNTERS_MASK = (1 << X86_IA32_NUM_PERF_COUNTERS) - 1;
+		PERF_COUNTERS_MASK = (1 << NUM_PERF_COUNTERS) - 1;
 		
-		X86_IA32_NUM_FIXED_PERF_COUNTERS = (edx & 0x0F);
+		NUM_FIXED_PERF_COUNTERS = (edx & 0x0F);
-		X86_IA32_FIXED_PERF_COUNTERS_MASK = 
+		FIXED_PERF_COUNTERS_MASK =
-			((1UL << X86_IA32_NUM_FIXED_PERF_COUNTERS) - 1) <<
+			((1UL << NUM_FIXED_PERF_COUNTERS) - 1) <<
-			X86_IA32_BASE_FIXED_PERF_COUNTERS;
+			BASE_FIXED_PERF_COUNTERS;
 	
 		perf_counters_discovered = 1;
-		kprintf("X86_IA32_NUM_PERF_COUNTERS: %d, X86_IA32_NUM_FIXED_PERF_COUNTERS: %d\n",
+		kprintf("NUM_PERF_COUNTERS: %d, NUM_FIXED_PERF_COUNTERS: %d\n",
-				X86_IA32_NUM_PERF_COUNTERS, X86_IA32_NUM_FIXED_PERF_COUNTERS);
+				NUM_PERF_COUNTERS, NUM_FIXED_PERF_COUNTERS);
 	}
 
 	/* Clear Fixed Counter Control */
@@ -97,20 +93,20 @@ void x86_init_perfctr(void)
 	wrmsr(MSR_PERF_FIXED_CTRL, value);
 
 	/* Clear Generic Counter Control */
-	for(i = 0; i < X86_IA32_NUM_PERF_COUNTERS; i++) {
+	for (i = 0; i < NUM_PERF_COUNTERS; i++) {
 		wrmsr(MSR_IA32_PERFEVTSEL0 + i, 0);
 	}
 
 	/* Enable PMC Control */
 	value = rdmsr(MSR_PERF_GLOBAL_CTRL);
-	value |= X86_IA32_PERF_COUNTERS_MASK;
+	value |= PERF_COUNTERS_MASK;
-	value |= X86_IA32_FIXED_PERF_COUNTERS_MASK;
+	value |= FIXED_PERF_COUNTERS_MASK;
 	wrmsr(MSR_PERF_GLOBAL_CTRL, value);
 }
 
 static int set_perfctr_x86_direct(int counter, int mode, unsigned int value)
 {
-	if (counter < 0 || counter >= X86_IA32_NUM_PERF_COUNTERS) {
+	if (counter < 0 || counter >= NUM_PERF_COUNTERS) {
 		return -EINVAL;
 	}
 
@@ -149,13 +145,14 @@ static int set_pmc_x86_direct(int counter, long val)
 	val &= 0x000000ffffffffff; // 40bit Mask
 
 	cnt_bit = 1UL << counter;
-	if ( cnt_bit & X86_IA32_PERF_COUNTERS_MASK ) {
+	if (cnt_bit & PERF_COUNTERS_MASK) {
 		// set generic pmc
 		wrmsr(MSR_IA32_PMC0 + counter, val);
 	}
-	else if ( cnt_bit & X86_IA32_FIXED_PERF_COUNTERS_MASK ) {
+	else if (cnt_bit & FIXED_PERF_COUNTERS_MASK) {
 		// set fixed pmc
-		wrmsr(MSR_IA32_FIXED_CTR0 + counter - X86_IA32_BASE_FIXED_PERF_COUNTERS, val);
+		wrmsr(MSR_IA32_FIXED_CTR0 +
+			counter - BASE_FIXED_PERF_COUNTERS, val);
 	}
 	else {
 		return -EINVAL;
@@ -175,10 +172,10 @@ static int set_fixed_counter(int counter, int mode)
 {
 	unsigned long value = 0;
 	unsigned int  ctr_mask = 0xf;
-	int counter_idx = counter - X86_IA32_BASE_FIXED_PERF_COUNTERS ;
+	int counter_idx = counter - BASE_FIXED_PERF_COUNTERS;
 	unsigned int  set_val = 0;
 
-	if (counter_idx < 0 || counter_idx >= X86_IA32_NUM_FIXED_PERF_COUNTERS) {
+	if (counter_idx < 0 || counter_idx >= NUM_FIXED_PERF_COUNTERS) {
 		return -EINVAL;
 	}
 
@@ -208,14 +205,13 @@ int ihk_mc_perfctr_init_raw(int counter, uint64_t config, int mode)
 int ihk_mc_perfctr_init_raw(int counter, unsigned int code, int mode)
 #endif /*POSTK_DEBUG_TEMP_FIX_29*/
 {
-#ifdef POSTK_DEBUG_TEMP_FIX_31
 	// PAPI_REF_CYC counted by fixed counter
-	if (counter >= X86_IA32_BASE_FIXED_PERF_COUNTERS) {
+	if (counter >= BASE_FIXED_PERF_COUNTERS &&
+		counter < BASE_FIXED_PERF_COUNTERS + NUM_FIXED_PERF_COUNTERS) {
 		return ihk_mc_perfctr_fixed_init(counter, mode);
 	}
-#endif /*POSTK_DEBUG_TEMP_FIX_31*/
 
-	if (counter < 0 || counter >= X86_IA32_NUM_PERF_COUNTERS) {
+	if (counter < 0 || counter >= NUM_PERF_COUNTERS) {
 		return -EINVAL;
 	}
 
@@ -248,7 +244,7 @@ int ihk_mc_perfctr_init(int counter, enum ihk_perfctr_type type, int mode)
 	}
 #endif /*POSTK_DEBUG_TEMP_FIX_29*/
 
-	if (counter < 0 || counter >= X86_IA32_NUM_PERF_COUNTERS) {
+	if (counter < 0 || counter >= NUM_PERF_COUNTERS) {
 		return -EINVAL;
 	}
 	if (type < 0 || type >= PERFCTR_MAX_TYPE) {
@@ -300,18 +296,11 @@ int ihk_mc_perfctr_set_extra(struct mc_perf_event *event)
 extern void x86_march_perfctr_start(unsigned long counter_mask);
 #endif
 
-#ifdef POSTK_DEBUG_TEMP_FIX_30
-int ihk_mc_perfctr_start(int counter)
-#else
 int ihk_mc_perfctr_start(unsigned long counter_mask)
-#endif /*POSTK_DEBUG_TEMP_FIX_30*/
 {
 	int ret = 0;
 	unsigned long value = 0;
-	unsigned long mask = X86_IA32_PERF_COUNTERS_MASK | X86_IA32_FIXED_PERF_COUNTERS_MASK;
+	unsigned long mask = PERF_COUNTERS_MASK | FIXED_PERF_COUNTERS_MASK;
-#ifdef POSTK_DEBUG_TEMP_FIX_30
-	unsigned long counter_mask = 1UL << counter;
-#endif /*POSTK_DEBUG_TEMP_FIX_30*/
 
 	PERFCTR_CHKANDJUMP(counter_mask & ~mask, "counter_mask out of range", -EINVAL);
 
@@ -328,18 +317,11 @@ int ihk_mc_perfctr_start(unsigned long counter_mask)
 	goto fn_exit;
 }
 
-#ifdef POSTK_DEBUG_TEMP_FIX_30
-int ihk_mc_perfctr_stop(int counter)
-#else
 int ihk_mc_perfctr_stop(unsigned long counter_mask)
-#endif/*POSTK_DEBUG_TEMP_FIX_30*/
 {
 	int ret = 0;
 	unsigned long value;
-	unsigned long mask = X86_IA32_PERF_COUNTERS_MASK | X86_IA32_FIXED_PERF_COUNTERS_MASK;
+	unsigned long mask = PERF_COUNTERS_MASK | FIXED_PERF_COUNTERS_MASK;
-#ifdef POSTK_DEBUG_TEMP_FIX_30
-	unsigned long counter_mask = 1UL << counter;
-#endif/*POSTK_DEBUG_TEMP_FIX_30*/
 
 	PERFCTR_CHKANDJUMP(counter_mask & ~mask, "counter_mask out of range", -EINVAL);
 
@@ -376,10 +358,10 @@ int ihk_mc_perfctr_fixed_init(int counter, int mode)
 {
 	unsigned long value = 0;
 	unsigned int  ctr_mask = 0xf;
-	int counter_idx = counter - X86_IA32_BASE_FIXED_PERF_COUNTERS ;
+	int counter_idx = counter - BASE_FIXED_PERF_COUNTERS;
 	unsigned int  set_val = 0;
 
-	if (counter_idx < 0 || counter_idx >= X86_IA32_NUM_FIXED_PERF_COUNTERS) {
+	if (counter_idx < 0 || counter_idx >= NUM_FIXED_PERF_COUNTERS) {
 		return -EINVAL;
 	}
 
@@ -420,7 +402,7 @@ int ihk_mc_perfctr_read_mask(unsigned long counter_mask, unsigned long *value)
 {
 	int i, j;
 
-	for (i = 0, j = 0; i < X86_IA32_NUM_PERF_COUNTERS && counter_mask;
+	for (i = 0, j = 0; i < NUM_PERF_COUNTERS && counter_mask;
 	     i++, counter_mask >>= 1) {
 		if (counter_mask & 1) {
 			value[j++] = rdpmc(i);
@@ -440,13 +422,14 @@ unsigned long ihk_mc_perfctr_read(int counter)
 
 	cnt_bit = 1UL << counter;
 
-	if ( cnt_bit & X86_IA32_PERF_COUNTERS_MASK ) {
+	if (cnt_bit & PERF_COUNTERS_MASK) {
 		// read generic pmc
 		retval = rdpmc(counter);
 	}
-	else if ( cnt_bit & X86_IA32_FIXED_PERF_COUNTERS_MASK ) {
+	else if (cnt_bit & FIXED_PERF_COUNTERS_MASK) {
 		// read fixed pmc
-		retval = rdpmc((1 << 30) + (counter - X86_IA32_BASE_FIXED_PERF_COUNTERS));
+		retval = rdpmc((1 << 30) +
+			(counter - BASE_FIXED_PERF_COUNTERS));
 	}
 	else {
 		retval = -EINVAL;
@@ -468,12 +451,12 @@ unsigned long ihk_mc_perfctr_read_msr(int counter)
 
 	cnt_bit = 1UL << counter;
 
-	if ( cnt_bit & X86_IA32_PERF_COUNTERS_MASK ) {
+	if (cnt_bit & PERF_COUNTERS_MASK) {
 		// read generic pmc
 		idx = MSR_IA32_PMC0 + counter;
 		retval = (unsigned long) rdmsr(idx);
 	}
-	else if ( cnt_bit & X86_IA32_FIXED_PERF_COUNTERS_MASK ) {
+	else if (cnt_bit & FIXED_PERF_COUNTERS_MASK) {
 		// read fixed pmc
 		idx = MSR_IA32_FIXED_CTR0 + counter;
 		retval = (unsigned long) rdmsr(idx);
@@ -506,8 +489,8 @@ int ihk_mc_perfctr_alloc_counter(unsigned int *type, unsigned long *config, unsi
 	}
 
 	// find avail generic counter
-       	for(i = 0; i < X86_IA32_NUM_PERF_COUNTERS; i++) {
+	for (i = 0; i < NUM_PERF_COUNTERS; i++) {
-		if(!(pmc_status & (1 << i))) {
+		if (!(pmc_status & (1 << i))) {
 			ret = i;
 			break;
 		}

arch/x86_64/kernel/syscall.c

@@ -31,12 +31,11 @@
 #include <page.h>
 #include <limits.h>
 #include <syscall.h>
+#include <debug.h>
 
 void terminate_mcexec(int, int);
 extern long do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact);
 long syscall(int num, ihk_mc_user_context_t *ctx);
-void set_signal(int sig, void *regs0, siginfo_t *info);
-void check_signal(unsigned long rc, void *regs0, int num);
 extern unsigned long do_fork(int, unsigned long, unsigned long, unsigned long,
 	unsigned long, unsigned long, unsigned long);
 extern int get_xsave_size();
@@ -45,11 +44,8 @@ extern uint64_t get_xsave_mask();
 //#define DEBUG_PRINT_SC
 
 #ifdef DEBUG_PRINT_SC
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf(...) kprintf(__VA_ARGS__)
 #endif
 
 uintptr_t debug_constants[] = {
@@ -92,33 +88,45 @@ static ptrdiff_t vdso_offset;
 
 extern int num_processors;
 
-int obtain_clone_cpuid(cpu_set_t *cpu_set) {
+int obtain_clone_cpuid(cpu_set_t *cpu_set, int use_last) {
 	int min_queue_len = -1;
-	int cpu, min_cpu = -1;
+	int cpu, min_cpu = -1, uti_cpu = -1;
+	unsigned long irqstate;
 	
+	irqstate = ihk_mc_spinlock_lock(&runq_reservation_lock);
 	/* Find the first allowed core with the shortest run queue */
 	for (cpu = 0; cpu < num_processors; ++cpu) {
 		struct cpu_local_var *v;
-		unsigned long irqstate;
 
 		if (!CPU_ISSET(cpu, cpu_set)) continue;
 
 		v = get_cpu_local_var(cpu);
-		irqstate = ihk_mc_spinlock_lock(&v->runq_lock);
+		ihk_mc_spinlock_lock_noirq(&v->runq_lock);
-		if (min_queue_len == -1 || v->runq_len < min_queue_len) {
+		dkprintf("%s: cpu=%d,runq_len=%d,runq_reserved=%d\n", __FUNCTION__, cpu, v->runq_len, v->runq_reserved);
-			min_queue_len = v->runq_len;
+		if (min_queue_len == -1 || v->runq_len + v->runq_reserved < min_queue_len) {
+			min_queue_len = v->runq_len + v->runq_reserved;
 			min_cpu = cpu;
 		}
-		ihk_mc_spinlock_unlock(&v->runq_lock, irqstate);
 
+		/* Record the last tie CPU */
+		if (min_cpu != cpu && v->runq_len + v->runq_reserved == min_queue_len) {
+			uti_cpu = cpu;
+		}
+		dkprintf("%s: cpu=%d,runq_len=%d,runq_reserved=%d,min_cpu=%d,uti_cpu=%d\n", __FUNCTION__, cpu, v->runq_len, v->runq_reserved, min_cpu, uti_cpu);
+		ihk_mc_spinlock_unlock_noirq(&v->runq_lock);
+#if 0
 		if (min_queue_len == 0)
 			break;
+#endif
 	}
 
+	min_cpu = use_last ? uti_cpu : min_cpu;
 	if (min_cpu != -1) {
 		if (get_cpu_local_var(min_cpu)->status != CPU_STATUS_RESERVED)
 			get_cpu_local_var(min_cpu)->status = CPU_STATUS_RESERVED;
+		__sync_fetch_and_add(&get_cpu_local_var(min_cpu)->runq_reserved, 1);
 	}
+	ihk_mc_spinlock_unlock(&runq_reservation_lock, irqstate);
 
     return min_cpu;
 }
@@ -251,7 +259,7 @@ SYSCALL_DECLARE(rt_sigreturn)
 		info.si_code = TRAP_TRACE;
 		set_signal(SIGTRAP, regs, &info);
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 
 	if(ksigsp.fpregs && xsavesize){
@@ -276,7 +284,6 @@ SYSCALL_DECLARE(rt_sigreturn)
 }
 
 extern struct cpu_local_var *clv;
-extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern void interrupt_syscall(struct thread *, int sig);
 extern void terminate(int, int);
 extern int num_processors;
@@ -530,23 +537,32 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	dkprintf("ptrace_report_signal, tid=%d, pid=%d\n", thread->tid, thread->proc->pid);
 
 	mcs_rwlock_writer_lock(&proc->update_lock, &lock);	
-	if(!(proc->ptrace & PT_TRACED)){
+	if (!(thread->ptrace & PT_TRACED)) {
 		mcs_rwlock_writer_unlock(&proc->update_lock, &lock);
 		return;
 	}
-	thread->exit_status = sig;
+
 	/* Transition thread state */
-	proc->status = PS_TRACED;
+	thread->exit_status = sig;
 	thread->status = PS_TRACED;
-	proc->ptrace &= ~PT_TRACE_SYSCALL;
+	thread->ptrace &= ~PT_TRACE_SYSCALL;
-	if (sig == SIGSTOP || sig == SIGTSTP ||
-			sig == SIGTTIN || sig == SIGTTOU) {
-		proc->signal_flags |= SIGNAL_STOP_STOPPED;
-	} else {
-		proc->signal_flags &= ~SIGNAL_STOP_STOPPED;
-	}
-	parent_pid = proc->parent->pid;
 	save_debugreg(thread->ptrace_debugreg);
+	if (sig == SIGSTOP || sig == SIGTSTP ||
+	    sig == SIGTTIN || sig == SIGTTOU) {
+		thread->signal_flags |= SIGNAL_STOP_STOPPED;
+	}
+	else {
+		thread->signal_flags &= ~SIGNAL_STOP_STOPPED;
+	}
+
+	if (thread == proc->main_thread) {
+		proc->status = PS_DELAY_TRACED;
+		parent_pid = proc->parent->pid;
+	}
+	else {
+		parent_pid = thread->report_proc->pid;
+		waitq_wakeup(&thread->report_proc->waitpid_q);
+	}
 	mcs_rwlock_writer_unlock(&proc->update_lock, &lock);
 
 	memset(&info, '\0', sizeof info);
@@ -555,8 +571,6 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	info._sifields._sigchld.si_pid = thread->tid;
 	info._sifields._sigchld.si_status = thread->exit_status;
 	do_kill(cpu_local_var(current), parent_pid, -1, SIGCHLD, &info, 0);
-	/* Wake parent (if sleeping in wait4()) */
-	waitq_wakeup(&proc->parent->waitpid_q);
 
 	dkprintf("ptrace_report_signal,sleeping\n");
 	/* Sleep */
@@ -569,9 +583,8 @@ ptrace_arch_prctl(int pid, long code, long addr)
 {
 	long rc = -EIO;
 	struct thread *child;
-	struct mcs_rwlock_node_irqsave lock;
 
-	child = find_thread(pid, pid, &lock);
+	child = find_thread(pid, pid);
 	if (!child)
 		return -ESRCH;
 	if (child->proc->status & (PS_TRACED | PS_STOPPED)) {
@@ -613,7 +626,7 @@ ptrace_arch_prctl(int pid, long code, long addr)
 			break;
 		}
 	}
-	thread_unlock(child, &lock);
+	thread_unlock(child);
 
 	return rc;
 }
@@ -635,11 +648,13 @@ arch_ptrace(long request, int pid, long addr, long data)
 static int
 isrestart(int num, unsigned long rc, int sig, int restart)
 {
-	if(sig == SIGKILL || sig == SIGSTOP)
+	if (sig == SIGKILL || sig == SIGSTOP)
 		return 0;
-	if(num == 0 || rc != -EINTR)
+	if (num < 0 || rc != -EINTR)
 		return 0;
-	switch(num){
+	if (sig == SIGCHLD)
+		return 1;
+	switch (num) {
 	    case __NR_pause:
 	    case __NR_rt_sigsuspend:
 	    case __NR_rt_sigtimedwait:
@@ -660,14 +675,12 @@ isrestart(int num, unsigned long rc, int sig, int restart)
 	    case __NR_io_getevents:
 		return 0;
 	}
-	if(sig == SIGCHLD)
+	if (restart)
-		return 1;
-	if(restart)
 		return 1;
 	return 0;
 }
 
-void
+int
 do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pending *pending, int num)
 {
 	struct x86_user_context *regs = regs0;
@@ -679,14 +692,15 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 	int	ptraceflag = 0;
 	struct mcs_rwlock_node_irqsave lock;
 	struct mcs_rwlock_node_irqsave mcs_rw_node;
+	int restart = 0;
 
 	for(w = pending->sigmask.__val[0], sig = 0; w; sig++, w >>= 1);
 	dkprintf("do_signal(): tid=%d, pid=%d, sig=%d\n", thread->tid, proc->pid, sig);
 	orgsig = sig;
 
-	if((proc->ptrace & PT_TRACED) &&
+	if ((thread->ptrace & PT_TRACED) &&
-	   pending->ptracecont == 0 &&
+	    pending->ptracecont == 0 &&
-	   sig != SIGKILL) {
+	    sig != SIGKILL) {
 		ptraceflag = 1;
 		sig = SIGSTOP;
 	}
@@ -707,7 +721,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 	if(k->sa.sa_handler == SIG_IGN){
 		kfree(pending);
 		mcs_rwlock_writer_unlock(&thread->sigcommon->lock, &mcs_rw_node);
-		return;
+		goto out;
 	}
 	else if(k->sa.sa_handler){
 		unsigned long *usp; /* user stack */
@@ -757,9 +771,8 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 		memcpy(&ksigsp.sigstack, &thread->sigstack, sizeof(stack_t));
 		ksigsp.sigrc = rc;
 		ksigsp.num = num;
-		ksigsp.restart = isrestart(num, rc, sig, k->sa.sa_flags & SA_RESTART);
+		restart = isrestart(num, rc, sig, k->sa.sa_flags & SA_RESTART);
-		if(num != 0 && rc == -EINTR && sig == SIGCHLD)
+		ksigsp.restart = restart;
-			ksigsp.restart = 1;
 		if(xsavesize){
 			uint64_t xsave_mask = get_xsave_mask();
 			unsigned int low = (unsigned int)xsave_mask;
@@ -772,7 +785,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				kfree(_kfpregs);
 				kprintf("do_signal,no space available\n");
 				terminate(0, sig);
-				return;
+				goto out;
 			}
 			kfpregs = (void *)((((unsigned long)_kfpregs) + 63) & ~63);
 			memset(kfpregs, '\0', xsavesize);
@@ -782,7 +795,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				kfree(_kfpregs);
 				kprintf("do_signal,write_process_vm failed\n");
 				terminate(0, sig);
-				return;
+				goto out;
 			}
 			ksigsp.fpregs = (void *)fpregs;
 			kfree(_kfpregs);
@@ -794,7 +807,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			mcs_rwlock_writer_unlock(&thread->sigcommon->lock, &mcs_rw_node);
 			kprintf("do_signal,write_process_vm failed\n");
 			terminate(0, sig);
-			return;
+			goto out;
 		}
 
 		usp = (unsigned long *)sigsp;
@@ -824,12 +837,13 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			info.si_code = TRAP_TRACE;
 			set_signal(SIGTRAP, regs, &info);
 			check_need_resched();
-			check_signal(0, regs, 0);
+			check_signal(0, regs, -1);
 		}
 	}
 	else {
 		int	coredumped = 0;
 		siginfo_t info;
+		int ptc = pending->ptracecont;
 
 		if(ptraceflag){
 			if(thread->ptrace_recvsig)
@@ -856,25 +870,37 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 				info.si_code = CLD_STOPPED;
 				info._sifields._sigchld.si_pid = thread->proc->pid;
 				info._sifields._sigchld.si_status = (sig << 8) | 0x7f;
-				do_kill(cpu_local_var(current), thread->proc->parent->pid, -1, SIGCHLD, &info, 0);
+				if (ptc == 2 &&
-				dkprintf("do_signal,SIGSTOP,changing state\n");
+				    thread != thread->proc->main_thread) {
+					thread->signal_flags =
+							    SIGNAL_STOP_STOPPED;
+					thread->status = PS_STOPPED;
+					thread->exit_status = SIGSTOP;
+					do_kill(thread,
+						thread->report_proc->pid, -1,
+						SIGCHLD, &info, 0);
+					waitq_wakeup(
+					       &thread->report_proc->waitpid_q);
+				}
+				else {
+					/* Update thread state in fork tree */
+					mcs_rwlock_writer_lock(
+						     &proc->update_lock, &lock);
+					proc->group_exit_status = SIGSTOP;
 
-				/* Update thread state in fork tree */
+					/* Reap and set new signal_flags */
-				mcs_rwlock_writer_lock(&proc->update_lock, &lock);	
+					proc->main_thread->signal_flags =
-				proc->group_exit_status = SIGSTOP;
+							    SIGNAL_STOP_STOPPED;
 
-				/* Reap and set new signal_flags */
+					proc->status = PS_DELAY_STOPPED;
-				proc->signal_flags = SIGNAL_STOP_STOPPED;
+					thread->status = PS_STOPPED;
+					mcs_rwlock_writer_unlock(
+						     &proc->update_lock, &lock);
 
-				proc->status = PS_STOPPED;
+					do_kill(thread,
-				thread->status = PS_STOPPED;
+						thread->proc->parent->pid, -1,
-				mcs_rwlock_writer_unlock(&proc->update_lock, &lock);	
+						SIGCHLD, &info, 0);
-
+				}
-				/* Wake up the parent who tried wait4 and sleeping */
-				waitq_wakeup(&proc->parent->waitpid_q);
-
-				dkprintf("do_signal(): pid: %d, tid: %d SIGSTOP, sleeping\n", 
-					proc->pid, thread->tid);
 				/* Sleep */
 				schedule();
 				dkprintf("SIGSTOP(): woken up\n");
@@ -882,19 +908,28 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			break;
 		case SIGTRAP:
 			dkprintf("do_signal,SIGTRAP\n");
-			if(!(proc->ptrace & PT_TRACED)) {
+			if (!(thread->ptrace & PT_TRACED)) {
 				goto core;
 			}
 
 			/* Update thread state in fork tree */
-			mcs_rwlock_writer_lock(&proc->update_lock, &lock);	
 			thread->exit_status = SIGTRAP;
-			proc->status = PS_TRACED;
 			thread->status = PS_TRACED;
-			mcs_rwlock_writer_unlock(&proc->update_lock, &lock);	
+			if (thread == proc->main_thread) {
-
+				mcs_rwlock_writer_lock(&proc->update_lock,
-			/* Wake up the parent who tried wait4 and sleeping */
+						       &lock);
-			waitq_wakeup(&thread->proc->parent->waitpid_q);
+				proc->group_exit_status = SIGTRAP;
+				proc->status = PS_DELAY_TRACED;
+				mcs_rwlock_writer_unlock(&proc->update_lock,
+							 &lock);
+				do_kill(thread, thread->proc->parent->pid, -1,
+					SIGCHLD, &info, 0);
+			}
+			else {
+				do_kill(thread, thread->report_proc->pid, -1,
+					SIGCHLD, &info, 0);
+				waitq_wakeup(&thread->report_proc->waitpid_q);
+			}
 
 			/* Sleep */
 			dkprintf("do_signal,SIGTRAP,sleeping\n");
@@ -909,7 +944,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			info._sifields._sigchld.si_pid = proc->pid;
 			info._sifields._sigchld.si_status = 0x0000ffff;
 			do_kill(cpu_local_var(current), proc->parent->pid, -1, SIGCHLD, &info, 0);
-			proc->signal_flags = SIGNAL_STOP_CONTINUED;
+			proc->main_thread->signal_flags = SIGNAL_STOP_CONTINUED;
 			proc->status = PS_RUNNING;
 			dkprintf("do_signal,SIGCONT,do nothing\n");
 			break;
@@ -938,6 +973,8 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			break;
 		}
 	}
+out:
+	return restart;
 }
 
 static struct sig_pending *
@@ -957,10 +994,12 @@ getsigpending(struct thread *thread, int delflag){
 	lock = &thread->sigcommon->lock;
 	head = &thread->sigcommon->sigpending;
 	for(;;) {
-		if (delflag)
+		if (delflag) {
 			mcs_rwlock_writer_lock(lock, &mcs_rw_node);
-		else 
+		}
+		else {
 			mcs_rwlock_reader_lock(lock, &mcs_rw_node);
+		}
 
 		list_for_each_entry_safe(pending, next, head, list){
 			for(x = pending->sigmask.__val[0], sig = 0; x; sig++, x >>= 1);
@@ -973,19 +1012,23 @@ getsigpending(struct thread *thread, int delflag){
 					if(delflag) 
 						list_del(&pending->list);
 
-					if (delflag)
+					if (delflag) {
 						mcs_rwlock_writer_unlock(lock, &mcs_rw_node);
-					else 
+					}
+					else {
 						mcs_rwlock_reader_unlock(lock, &mcs_rw_node);
+					}
 					return pending;
 				}
 			}
 		}
 
-		if (delflag)
+		if (delflag) {
 			mcs_rwlock_writer_unlock(lock, &mcs_rw_node);
-		else 
+		}
+		else {
 			mcs_rwlock_reader_unlock(lock, &mcs_rw_node);
+		}
 
 		if(lock == &thread->sigpendinglock)
 			return NULL;
@@ -1000,6 +1043,11 @@ getsigpending(struct thread *thread, int delflag){
 struct sig_pending *
 hassigpending(struct thread *thread)
 {
+	if (list_empty(&thread->sigpending) &&
+	    list_empty(&thread->sigcommon->sigpending)) {
+		return NULL;
+	}
+
 	return getsigpending(thread, 0);
 }
 
@@ -1017,6 +1065,12 @@ void save_syscall_return_value(int num, unsigned long rc)
 	return;
 }
 
+/** \brief check arrived signals and processing
+ *
+ * @param rc    return value of syscall
+ * @param regs0 context
+ * @param num   syscall number (-1: Not called on exiting system call)
+ */
 void
 check_signal(unsigned long rc, void *regs0, int num)
 {
@@ -1050,6 +1104,11 @@ check_signal(unsigned long rc, void *regs0, int num)
 		goto out;
 	}
 
+	if (list_empty(&thread->sigpending) &&
+	    list_empty(&thread->sigcommon->sigpending)) {
+		goto out;
+	}
+
 	for(;;){
 		pending = getsigpending(thread, 1);
 		if(!pending) {
@@ -1057,7 +1116,9 @@ check_signal(unsigned long rc, void *regs0, int num)
 			goto out;
 		}
 
-		do_signal(rc, regs, thread, pending, num);
+		if (do_signal(rc, regs, thread, pending, num)) {
+			num = -1;
+		}
 	}
 
 out:
@@ -1137,7 +1198,7 @@ check_sig_pending_thread(struct thread *thread)
 }
 
 void
-check_sig_pending()
+check_sig_pending(void)
 {
 	struct thread *thread;
 	struct cpu_local_var *v;
@@ -1158,7 +1219,7 @@ repeat:
 			continue;
 		}
 
-		if (thread->proc->exit_status & 0x0000000100000000L) {
+		if (thread->proc->group_exit_status & 0x0000000100000000L) {
 			continue;
 		}
 
@@ -1367,7 +1428,8 @@ done:
 		return 0;
 	}
 
-	if (tthread->thread_offloaded) {
+	/* Forward signal to Linux by interrupt_syscall mechanism */
+	if (tthread->uti_state == UTI_STATE_RUNNING_IN_LINUX) {
 		if (!tthread->proc->nohost) {
 			interrupt_syscall(tthread, sig);
 		}
@@ -1384,10 +1446,10 @@ done:
 	   in check_signal */
 	rc = 0;
 	k = tthread->sigcommon->action + sig - 1;
-	if((sig != SIGKILL && (tproc->ptrace & PT_TRACED)) ||
+	if ((sig != SIGKILL && (tthread->ptrace & PT_TRACED)) ||
-			(k->sa.sa_handler != (void *)1 &&
+	    (k->sa.sa_handler != (void *)1 &&
-			 (k->sa.sa_handler != NULL ||
+	     (k->sa.sa_handler != NULL ||
-			  (sig != SIGCHLD && sig != SIGURG)))){
+	      (sig != SIGCHLD && sig != SIGURG)))) {
 		struct sig_pending *pending = NULL;
 		if (sig < 33) { // SIGRTMIN - SIGRTMAX
 			list_for_each_entry(pending, head, list){
@@ -1471,7 +1533,7 @@ set_signal(int sig, void *regs0, siginfo_t *info)
 
 SYSCALL_DECLARE(mmap)
 {
-	const int supported_flags = 0
+	const unsigned int supported_flags = 0
 		| MAP_SHARED		// 01
 		| MAP_PRIVATE		// 02
 		| MAP_FIXED		// 10
@@ -1479,7 +1541,7 @@ SYSCALL_DECLARE(mmap)
 		| MAP_LOCKED		// 2000
 		| MAP_POPULATE		// 8000
 		| MAP_HUGETLB		// 00040000
-		| (0x3F << MAP_HUGE_SHIFT) // FC000000
+		| (0x3FU << MAP_HUGE_SHIFT) // FC000000
 		;
 	const int ignored_flags = 0
 #ifdef	USE_NOCACHE_MMAP
@@ -1498,7 +1560,7 @@ SYSCALL_DECLARE(mmap)
 		| MAP_NONBLOCK		// 00010000
 		;
 
-	const intptr_t addr0 = ihk_mc_syscall_arg0(ctx);
+	const uintptr_t addr0 = ihk_mc_syscall_arg0(ctx);
 	const size_t len0 = ihk_mc_syscall_arg1(ctx);
 	const int prot = ihk_mc_syscall_arg2(ctx);
 	const int flags0 = ihk_mc_syscall_arg3(ctx);
@@ -1507,7 +1569,7 @@ SYSCALL_DECLARE(mmap)
 	struct thread *thread = cpu_local_var(current);
 	struct vm_regions *region = &thread->vm->region;
 	int error;
-	intptr_t addr = 0;
+	uintptr_t addr = 0;
 	size_t len;
 	int flags = flags0;
 	size_t pgsize;
@@ -1699,6 +1761,11 @@ SYSCALL_DECLARE(arch_prctl)
 	                     ihk_mc_syscall_arg1(ctx));
 }
 
+SYSCALL_DECLARE(time)
+{
+	return time();
+}
+
 static int vdso_get_vdso_info(void)
 {
 	int error;
@@ -2081,7 +2148,7 @@ int do_process_vm_read_writev(int pid,
 
 	range = lookup_process_memory_range(lthread->vm, 
 			(uintptr_t)local_iov, 
-			(uintptr_t)(local_iov + liovcnt * sizeof(struct iovec)));
+			(uintptr_t)(local_iov + liovcnt));
 
 	if (!range) {
 		ret = -EFAULT; 
@@ -2090,7 +2157,7 @@ int do_process_vm_read_writev(int pid,
 
 	range = lookup_process_memory_range(lthread->vm, 
 			(uintptr_t)remote_iov, 
-			(uintptr_t)(remote_iov + riovcnt * sizeof(struct iovec)));
+			(uintptr_t)(remote_iov + riovcnt));
 
 	if (!range) {
 		ret = -EFAULT; 
@@ -2366,8 +2433,6 @@ int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
 			case 0:
 				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
 						sizeof(void *) * count);
-				memcpy(mpsr->status, mpsr->user_status,
-						sizeof(int) * count);
 				memcpy(mpsr->nodes, mpsr->user_nodes,
 						sizeof(int) * count);
 				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
@@ -2387,41 +2452,38 @@ int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
 			case 0:
 				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
 						sizeof(void *) * count);
-				memcpy(mpsr->status, mpsr->user_status,
-						sizeof(int) * count);
-			case 1:
 				memcpy(mpsr->nodes, mpsr->user_nodes,
 						sizeof(int) * count);
+				mpsr->nodes_ready = 1;
+				break;
+			case 1:
 				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
 				memset(mpsr->status, 0, sizeof(int) * count);
 				memset(mpsr->nr_pages, 0, sizeof(int) * count);
 				memset(mpsr->dst_phys, 0,
 						sizeof(unsigned long) * count);
-				mpsr->nodes_ready = 1;
 				break;
 
 			default:
 				break;
 		}
 	}
-	else if (nr_cpus >= 4 && nr_cpus < 8) {
+	else if (nr_cpus >= 4 && nr_cpus < 7) {
 		switch (cpu_index) {
 			case 0:
 				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
 						sizeof(void *) * count);
 				break;
 			case 1:
-				memcpy(mpsr->status, mpsr->user_status,
-						sizeof(int) * count);
-				break;
-			case 2:
 				memcpy(mpsr->nodes, mpsr->user_nodes,
 						sizeof(int) * count);
 				mpsr->nodes_ready = 1;
 				break;
-			case 3:
+			case 2:
 				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
 				memset(mpsr->status, 0, sizeof(int) * count);
+				break;
+			case 3:
 				memset(mpsr->nr_pages, 0, sizeof(int) * count);
 				memset(mpsr->dst_phys, 0,
 						sizeof(unsigned long) * count);
@@ -2431,7 +2493,7 @@ int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
 				break;
 		}
 	}
-	else if (nr_cpus >= 8) {
+	else {
 		switch (cpu_index) {
 			case 0:
 				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
@@ -2443,28 +2505,23 @@ int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
 						sizeof(void *) * (count / 2));
 				break;
 			case 2:
-				memcpy(mpsr->status, mpsr->user_status,
-						sizeof(int) * count);
-				break;
-			case 3:
 				memcpy(mpsr->nodes, mpsr->user_nodes,
 						sizeof(int) * count);
 				mpsr->nodes_ready = 1;
 				break;
-			case 4:
+			case 3:
 				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
 				break;
-			case 5:
+			case 4:
 				memset(mpsr->status, 0, sizeof(int) * count);
 				break;
-			case 6:
+			case 5:
 				memset(mpsr->nr_pages, 0, sizeof(int) * count);
 				break;
-			case 7:
+			case 6:
 				memset(mpsr->dst_phys, 0,
 						sizeof(unsigned long) * count);
 				break;
-
 			default:
 				break;
 		}
@@ -2672,11 +2729,19 @@ out:
 
 time_t time(void) {
 	struct syscall_request sreq IHK_DMA_ALIGN;
-	struct thread *thread = cpu_local_var(current);
+	struct timespec ats;
-	time_t ret;
+	time_t ret = 0;
-	sreq.number = __NR_time;
+
-	sreq.args[0] = (uintptr_t)NULL;
+	if (gettime_local_support) {
-	ret = (time_t)do_syscall(&sreq, ihk_mc_get_processor_id(), thread->proc->pid);
+		calculate_time_from_tsc(&ats);
+		ret = ats.tv_sec;
+	}
+	else {
+		sreq.number = __NR_time;
+		sreq.args[0] = (uintptr_t)NULL;
+		ret = (time_t)do_syscall(&sreq, ihk_mc_get_processor_id());
+	}
+
 	return ret;
 }
 

arch/x86_64/kernel/vsyscall.c

@@ -31,51 +31,6 @@ struct tod_data_s tod_data
 	.version =	IHK_ATOMIC64_INIT(0),
 };
 
-static inline void cpu_pause_for_vsyscall(void)
-{
-	asm volatile ("pause" ::: "memory");
-	return;
-} /* cpu_pause_for_vsyscall() */
-
-static inline void calculate_time_from_tsc(struct timespec *ts)
-{
-	long ver;
-	unsigned long current_tsc;
-	__time_t sec_delta;
-	long ns_delta;
-
-	for (;;) {
-		while ((ver = ihk_atomic64_read(&tod_data.version)) & 1) {
-			/* settimeofday() is in progress */
-			cpu_pause_for_vsyscall();
-		}
-		rmb();
-		*ts = tod_data.origin;
-		rmb();
-		if (ver == ihk_atomic64_read(&tod_data.version)) {
-			break;
-		}
-
-		/* settimeofday() has intervened */
-		cpu_pause_for_vsyscall();
-	}
-
-	current_tsc = rdtsc();
-	sec_delta = current_tsc / tod_data.clocks_per_sec;
-	ns_delta = NS_PER_SEC * (current_tsc % tod_data.clocks_per_sec)
-		/ tod_data.clocks_per_sec;
-	/* calc. of ns_delta overflows if clocks_per_sec exceeds 18.44 GHz */
-
-	ts->tv_sec += sec_delta;
-	ts->tv_nsec += ns_delta;
-	if (ts->tv_nsec >= NS_PER_SEC) {
-		ts->tv_nsec -= NS_PER_SEC;
-		++ts->tv_sec;
-	}
-
-	return;
-} /* calculate_time_from_tsc() */
-
 int vsyscall_gettimeofday(struct timeval *tv, void *tz)
 {
 	int error;

arch/x86_64/tools/mcreboot-smp-x86.sh.in

@@ -45,11 +45,12 @@ fi
 
 turbo=""
 ihk_irq=""
+safe_kernel_map=""
 umask_old=`umask`
 idle_halt=""
 allow_oversubscribe=""
 
-while getopts :tk:c:m:o:f:r:q:i:d:e:hO OPT
+while getopts stk:c:m:o:f:r:q:i:d:e:hO OPT
 do
 	case ${OPT} in
 	f)	facility=${OPTARG}
@@ -62,6 +63,8 @@ do
 		;;
 	m) mem=${OPTARG}
 		;;
+	s) safe_kernel_map="safe_kernel_map"
+		;;
 	r) ikc_map=${OPTARG}
 		;;
 	q) ihk_irq=${OPTARG}
@@ -78,8 +81,8 @@ do
 		;;
 	O) allow_oversubscribe="allow_oversubscribe"
 		;;
-	*)  echo "invalid option -${OPT}" >&2
+	\?) exit 1
-		exit 1
+		;;
 	esac
 done
 
@@ -232,7 +235,7 @@ if [ "${ENABLE_MCOVERLAYFS}" == "yes" ]; then
 			enable_mcoverlay="yes"
 		fi
 	else
-		if [ ${linux_version_code} -eq 199168 -a ${rhel_release} -ge 327 -a ${rhel_release} -le 693 ]; then
+		if [ ${linux_version_code} -eq 199168 -a ${rhel_release} -ge 327 -a ${rhel_release} -le 862 ]; then
 			enable_mcoverlay="yes"
 		fi
 		if [ ${linux_version_code} -ge 262144 -a ${linux_version_code} -lt 262400 ]; then
@@ -446,7 +449,7 @@ if ! ${SBINDIR}/ihkosctl 0 load ${KERNDIR}/mckernel.img; then
 fi
 
 # Set kernel arguments
-if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo $idle_halt dump_level=${DUMP_LEVEL} $extra_kopts $allow_oversubscribe"; then
+if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo $safe_kernel_map $idle_halt dump_level=${DUMP_LEVEL} $extra_kopts $allow_oversubscribe"; then
 	echo "error: setting kernel arguments" >&2
 	error_exit "os_created"
 fi

config.h.in

@@ -54,48 +54,6 @@
 /* Define to 1 if you have the <unistd.h> header file. */
 #undef HAVE_UNISTD_H
 
-/* Define to address of kernel symbol __vvar_page, or 0 if exported */
-#undef MCCTRL_KSYM___vvar_page
-
-/* Define to address of kernel symbol hpet_address, or 0 if exported */
-#undef MCCTRL_KSYM_hpet_address
-
-/* Define to address of kernel symbol hv_clock, or 0 if exported */
-#undef MCCTRL_KSYM_hv_clock
-
-/* Define to address of kernel symbol sys_mount, or 0 if exported */
-#undef MCCTRL_KSYM_sys_mount
-
-/* Define to address of kernel symbol sys_readlink, or 0 if exported */
-#undef MCCTRL_KSYM_sys_readlink
-
-/* Define to address of kernel symbol sys_umount, or 0 if exported */
-#undef MCCTRL_KSYM_sys_umount
-
-/* Define to address of kernel symbol sys_unshare, or 0 if exported */
-#undef MCCTRL_KSYM_sys_unshare
-
-/* Define to address of kernel symbol vdso_end, or 0 if exported */
-#undef MCCTRL_KSYM_vdso_end
-
-/* Define to address of kernel symbol vdso_image_64, or 0 if exported */
-#undef MCCTRL_KSYM_vdso_image_64
-
-/* Define to address of kernel symbol vdso_pages, or 0 if exported */
-#undef MCCTRL_KSYM_vdso_pages
-
-/* Define to address of kernel symbol vdso_spec, or 0 if exported */
-#undef MCCTRL_KSYM_vdso_spec
-
-/* Define to address of kernel symbol vdso_start, or 0 if exported */
-#undef MCCTRL_KSYM_vdso_start
-
-/* Define to address of kernel symbol walk_page_range, or 0 if exported */
-#undef MCCTRL_KSYM_walk_page_range
-
-/* Define to address of kernel symbol zap_page_range, or 0 if exported */
-#undef MCCTRL_KSYM_zap_page_range
-
 /* McKernel specific headers */
 #undef MCKERNEL_INCDIR
 
@@ -128,3 +86,6 @@
 
 /* Define to 1 if you have the ANSI C header files. */
 #undef STDC_HEADERS
+
+/* whether or not syscall_intercept library is linked */
+#undef WITH_SYSCALL_INTERCEPT

configure

@@ -628,9 +628,12 @@ IHK_RELEASE_DATE
 DCFA_VERSION
 MCKERNEL_VERSION
 IHK_VERSION
+WITH_SYSCALL_INTERCEPT
 ENABLE_QLMPI
 ENABLE_RUSAGE
 ENABLE_MCOVERLAYFS
+LDFLAGS_SYSCALL_INTERCEPT
+CPPFLAGS_SYSCALL_INTERCEPT
 MANDIR
 KERNDIR
 KMODDIR
@@ -702,6 +705,9 @@ enable_option_checking
 with_mpi
 with_mpi_include
 with_mpi_lib
+with_syscall_intercept
+with_syscall_intercept_include
+with_syscall_intercept_lib
 with_kernelsrc
 with_target
 with_system_map
@@ -1346,6 +1352,15 @@ Optional Packages:
   --with-mpi-include=PATH specify path where mpi include directory can be
                           found
   --with-mpi-lib=PATH     specify path where mpi lib directory can be found
+  --with-syscall_intercept=PATH
+                          specify path where syscall_intercept include
+                          directory and lib directory can be found
+  --with-syscall_intercept-include=PATH
+                          specify path where syscall_intercept include
+                          directory can be found
+  --with-syscall_intercept-lib=PATH
+                          specify path where syscall_intercept lib directory
+                          can be found
   --with-kernelsrc=path   Path to 'kernel src', default is
                           /lib/modules/uname_r/build
   --with-target={attached-mic | builtin-mic | builtin-x86 | smp-x86}
@@ -2082,6 +2097,8 @@ ac_compiler_gnu=$ac_cv_c_compiler_gnu
 
 
 
+
+
 IHK_VERSION=1.5.1
 MCKERNEL_VERSION=1.5.1
 DCFA_VERSION=DCFA_VERSION_m4
@@ -3513,6 +3530,195 @@ fi
 
 
 
+
+# Check whether --with-syscall_intercept was given.
+if test "${with_syscall_intercept+set}" = set; then :
+  withval=$with_syscall_intercept; case "$withval" in #(
+  yes|no|'') :
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: --without-syscall_intercept=PATH expects a valid PATH" >&5
+$as_echo "$as_me: WARNING: --without-syscall_intercept=PATH expects a valid PATH" >&2;}
+                          with_syscall_intercept="" ;; #(
+  *) :
+     ;;
+esac
+else
+  with_syscall_intercept=
+fi
+
+
+# Check whether --with-syscall_intercept-include was given.
+if test "${with_syscall_intercept_include+set}" = set; then :
+  withval=$with_syscall_intercept_include; case "$withval" in #(
+  yes|no|'') :
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: --without-syscall_intercept-include=PATH expects a valid PATH" >&5
+$as_echo "$as_me: WARNING: --without-syscall_intercept-include=PATH expects a valid PATH" >&2;}
+                          with_syscall_intercept_include="" ;; #(
+  *) :
+     ;;
+esac
+fi
+
+
+# Check whether --with-syscall_intercept-lib was given.
+if test "${with_syscall_intercept_lib+set}" = set; then :
+  withval=$with_syscall_intercept_lib; case "$withval" in #(
+  yes|no|'') :
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: --without-syscall_intercept-lib=PATH expects a valid PATH" >&5
+$as_echo "$as_me: WARNING: --without-syscall_intercept-lib=PATH expects a valid PATH" >&2;}
+                          with_syscall_intercept_lib="" ;; #(
+  *) :
+     ;;
+esac
+fi
+
+
+    # The args have been sanitized into empty/non-empty values above.
+    # Now append -I/-L args to CPPFLAGS/LDFLAGS, with more specific options
+    # taking priority
+
+    if test -n "${with_syscall_intercept_include}"; then :
+
+
+	if echo "$CPPFLAGS_SYSCALL_INTERCEPT" | $FGREP -e "\<-I${with_syscall_intercept_include}\>" >/dev/null 2>&1; then :
+  echo "CPPFLAGS_SYSCALL_INTERCEPT(='$CPPFLAGS_SYSCALL_INTERCEPT') contains '-I${with_syscall_intercept_include}', not appending" >&5
+else
+  echo "CPPFLAGS_SYSCALL_INTERCEPT(='$CPPFLAGS_SYSCALL_INTERCEPT') does not contain '-I${with_syscall_intercept_include}', appending" >&5
+		CPPFLAGS_SYSCALL_INTERCEPT="$CPPFLAGS_SYSCALL_INTERCEPT -I${with_syscall_intercept_include}"
+
+fi
+
+else
+  if test -n "${with_syscall_intercept}"; then :
+
+
+	if echo "$CPPFLAGS_SYSCALL_INTERCEPT" | $FGREP -e "\<-I${with_syscall_intercept}/include\>" >/dev/null 2>&1; then :
+  echo "CPPFLAGS_SYSCALL_INTERCEPT(='$CPPFLAGS_SYSCALL_INTERCEPT') contains '-I${with_syscall_intercept}/include', not appending" >&5
+else
+  echo "CPPFLAGS_SYSCALL_INTERCEPT(='$CPPFLAGS_SYSCALL_INTERCEPT') does not contain '-I${with_syscall_intercept}/include', appending" >&5
+		CPPFLAGS_SYSCALL_INTERCEPT="$CPPFLAGS_SYSCALL_INTERCEPT -I${with_syscall_intercept}/include"
+
+fi
+
+fi
+fi
+
+    if test -n "${with_syscall_intercept_lib}"; then :
+
+
+	if echo "$LDFLAGS_SYSCALL_INTERCEPT" | $FGREP -e "\<-L${with_syscall_intercept_lib} -Wl,-rpath,${with_syscall_intercept_lib}\>" >/dev/null 2>&1; then :
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') contains '-L${with_syscall_intercept_lib} -Wl,-rpath,${with_syscall_intercept_lib}', not appending" >&5
+else
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') does not contain '-L${with_syscall_intercept_lib} -Wl,-rpath,${with_syscall_intercept_lib}', appending" >&5
+		LDFLAGS_SYSCALL_INTERCEPT="$LDFLAGS_SYSCALL_INTERCEPT -L${with_syscall_intercept_lib} -Wl,-rpath,${with_syscall_intercept_lib}"
+
+fi
+
+else
+  if test -n "${with_syscall_intercept}"; then :
+
+
+	if echo "$LDFLAGS_SYSCALL_INTERCEPT" | $FGREP -e "\<-L${with_syscall_intercept}/lib -Wl,-rpath,${with_syscall_intercept}/lib\>" >/dev/null 2>&1; then :
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') contains '-L${with_syscall_intercept}/lib -Wl,-rpath,${with_syscall_intercept}/lib', not appending" >&5
+else
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') does not contain '-L${with_syscall_intercept}/lib -Wl,-rpath,${with_syscall_intercept}/lib', appending" >&5
+		LDFLAGS_SYSCALL_INTERCEPT="$LDFLAGS_SYSCALL_INTERCEPT -L${with_syscall_intercept}/lib -Wl,-rpath,${with_syscall_intercept}/lib"
+
+fi
+
+                  if test -d "${with_syscall_intercept}/lib64"; then :
+
+
+	if echo "$LDFLAGS_SYSCALL_INTERCEPT" | $FGREP -e "\<-L${with_syscall_intercept}/lib64 -Wl,-rpath,${with_syscall_intercept}/lib64\>" >/dev/null 2>&1; then :
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') contains '-L${with_syscall_intercept}/lib64 -Wl,-rpath,${with_syscall_intercept}/lib64', not appending" >&5
+else
+  echo "LDFLAGS_SYSCALL_INTERCEPT(='$LDFLAGS_SYSCALL_INTERCEPT') does not contain '-L${with_syscall_intercept}/lib64 -Wl,-rpath,${with_syscall_intercept}/lib64', appending" >&5
+		LDFLAGS_SYSCALL_INTERCEPT="$LDFLAGS_SYSCALL_INTERCEPT -L${with_syscall_intercept}/lib64 -Wl,-rpath,${with_syscall_intercept}/lib64"
+
+fi
+
+fi
+
+fi
+
+fi
+
+    if test -n "${with_syscall_intercept}" || test -n "${with_syscall_intercept_include}" || test -n "${with_syscall_intercept_lib}"; then :
+  WITH_SYSCALL_INTERCEPT=yes
+else
+  WITH_SYSCALL_INTERCEPT=no
+fi
+
+
+if test "x$WITH_SYSCALL_INTERCEPT" == "xno" ; then
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for syscall_no_intercept in -lsyscall_intercept" >&5
+$as_echo_n "checking for syscall_no_intercept in -lsyscall_intercept... " >&6; }
+if ${ac_cv_lib_syscall_intercept_syscall_no_intercept+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  ac_check_lib_save_LIBS=$LIBS
+LIBS="-lsyscall_intercept -lcapstone -ldl $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* Override any GCC internal prototype to avoid an error.
+   Use char because int might match the return type of a GCC
+   builtin and then its argument prototype would still apply.  */
+#ifdef __cplusplus
+extern "C"
+#endif
+char syscall_no_intercept ();
+int
+main ()
+{
+return syscall_no_intercept ();
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  ac_cv_lib_syscall_intercept_syscall_no_intercept=yes
+else
+  ac_cv_lib_syscall_intercept_syscall_no_intercept=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_syscall_intercept_syscall_no_intercept" >&5
+$as_echo "$ac_cv_lib_syscall_intercept_syscall_no_intercept" >&6; }
+if test "x$ac_cv_lib_syscall_intercept_syscall_no_intercept" = xyes; then :
+  syscall_intercept_lib_found=yes
+else
+  syscall_intercept_lib_found=no
+fi
+
+	if test "x$syscall_intercept_lib_found" != "xyes"; then :
+  { $as_echo "$as_me:${as_lineno-$LINENO}: libsyscall_intercept.so not found" >&5
+$as_echo "$as_me: libsyscall_intercept.so not found" >&6;}
+fi
+
+	ac_fn_c_check_header_mongrel "$LINENO" "libsyscall_intercept_hook_point.h" "ac_cv_header_libsyscall_intercept_hook_point_h" "$ac_includes_default"
+if test "x$ac_cv_header_libsyscall_intercept_hook_point_h" = xyes; then :
+  syscall_intercept_header_found=yes
+else
+  syscall_intercept_header_found=no
+fi
+
+
+	if test "x$syscall_intercept_header_found" != "xyes"; then :
+  { $as_echo "$as_me:${as_lineno-$LINENO}: libsyscall_intercept_hook_point.h not found" >&5
+$as_echo "$as_me: libsyscall_intercept_hook_point.h not found" >&6;}
+fi
+
+	if test "x$syscall_intercept_lib_found" == "xyes" && test "x$syscall_intercept_header_found" == "xyes"; then :
+  WITH_SYSCALL_INTERCEPT=yes
+else
+  WITH_SYSCALL_INTERCEPT=no
+fi
+fi
+
+
+
 # Check whether --with-kernelsrc was given.
 if test "${with_kernelsrc+set}" = set; then :
   withval=$with_kernelsrc; WITH_KERNELSRC=$withval
@@ -4396,399 +4602,6 @@ KDIR="$WITH_KERNELSRC"
 UNAME_R="$WITH_UNAME_R"
 TARGET="$WITH_TARGET"
 
-MCCTRL_LINUX_SYMTAB=""
-case "X$WITH_SYSTEM_MAP" in
-    Xyes | Xno | X)
-	MCCTRL_LINUX_SYMTAB=""
-	;;
-    *)
-	MCCTRL_LINUX_SYMTAB="$WITH_SYSTEM_MAP"
-	;;
-esac
-
-{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for System.map" >&5
-$as_echo_n "checking for System.map... " >&6; }
-if test -r "$MCCTRL_LINUX_SYMTAB"; then
-	MCCTRL_LINUX_SYMTAB="$MCCTRL_LINUX_SYMTAB"
-elif test -r "/boot/System.map-`uname -r`"; then
-	MCCTRL_LINUX_SYMTAB="/boot/System.map-`uname -r`"
-elif test -r "$KDIR/System.map"; then
-	MCCTRL_LINUX_SYMTAB="$KDIR/System.map"
-fi
-
-if test "$MCCTRL_LINUX_SYMTAB" == ""; then
-	as_fn_error $? "could not find" "$LINENO" 5
-fi
-
-if test -z "`eval cat $MCCTRL_LINUX_SYMTAB`"; then
-	as_fn_error $? "could not read System.map file, no read permission?" "$LINENO" 5
-fi
-{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $MCCTRL_LINUX_SYMTAB" >&5
-$as_echo "$MCCTRL_LINUX_SYMTAB" >&6; }
-
-MCCTRL_LINUX_SYMTAB_CMD="cat $MCCTRL_LINUX_SYMTAB"
-
-# MCCTRL_FIND_KSYM(SYMBOL)
-# ------------------------------------------------------
-# Search System.map for address of the given symbol and
-# do one of three things in config.h:
-# If not found, leave MCCTRL_KSYM_foo undefined
-# If found to be exported, "#define MCCTRL_KSYM_foo 0"
-# If found not to be exported, "#define MCCTRL_KSYM_foo 0x<value>"
-
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol sys_mount" >&5
-$as_echo_n "checking System.map for symbol sys_mount... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " sys_mount\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_sys_mount\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_sys_mount $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol sys_umount" >&5
-$as_echo_n "checking System.map for symbol sys_umount... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " sys_umount\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_sys_umount\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_sys_umount $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol sys_unshare" >&5
-$as_echo_n "checking System.map for symbol sys_unshare... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " sys_unshare\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_sys_unshare\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_sys_unshare $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol zap_page_range" >&5
-$as_echo_n "checking System.map for symbol zap_page_range... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " zap_page_range\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_zap_page_range\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_zap_page_range $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol vdso_image_64" >&5
-$as_echo_n "checking System.map for symbol vdso_image_64... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " vdso_image_64\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_vdso_image_64\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_vdso_image_64 $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol vdso_start" >&5
-$as_echo_n "checking System.map for symbol vdso_start... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " vdso_start\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_vdso_start\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_vdso_start $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol vdso_end" >&5
-$as_echo_n "checking System.map for symbol vdso_end... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " vdso_end\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_vdso_end\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_vdso_end $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol vdso_pages" >&5
-$as_echo_n "checking System.map for symbol vdso_pages... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " vdso_pages\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_vdso_pages\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_vdso_pages $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol __vvar_page" >&5
-$as_echo_n "checking System.map for symbol __vvar_page... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __vvar_page\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab___vvar_page\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM___vvar_page $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol hpet_address" >&5
-$as_echo_n "checking System.map for symbol hpet_address... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " hpet_address\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_hpet_address\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_hpet_address $mcctrl_addr
-_ACEOF
-
-  fi
-
-# POSTK_DEBUG_ARCH_DEP_50, add:find kernel symbol.
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol vdso_spec" >&5
-$as_echo_n "checking System.map for symbol vdso_spec... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " vdso_spec\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_vdso_spec\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_vdso_spec $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol hv_clock" >&5
-$as_echo_n "checking System.map for symbol hv_clock... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " hv_clock\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_hv_clock\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_hv_clock $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol sys_readlink" >&5
-$as_echo_n "checking System.map for symbol sys_readlink... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " sys_readlink\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_sys_readlink\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_sys_readlink $mcctrl_addr
-_ACEOF
-
-  fi
-
-
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking System.map for symbol walk_page_range" >&5
-$as_echo_n "checking System.map for symbol walk_page_range... " >&6; }
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " walk_page_range\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: not found" >&5
-$as_echo "not found" >&6; }
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_walk_page_range\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: result: $mcctrl_result" >&5
-$as_echo "$mcctrl_result" >&6; }
-
-cat >>confdefs.h <<_ACEOF
-#define MCCTRL_KSYM_walk_page_range $mcctrl_addr
-_ACEOF
-
-  fi
-
-
 case $ENABLE_MEMDUMP in
 	yes|no|auto)
 		;;
@@ -4986,6 +4799,17 @@ else
 $as_echo "$as_me: perf is disabled" >&6;}
 fi
 
+if test "x$WITH_SYSCALL_INTERCEPT" = "xyes" ; then
+
+$as_echo "#define WITH_SYSCALL_INTERCEPT 1" >>confdefs.h
+
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: syscall_intercept library is linked" >&5
+$as_echo "$as_me: syscall_intercept library is linked" >&6;}
+else
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: syscall_intercept library isn't linked" >&5
+$as_echo "$as_me: syscall_intercept library isn't linked" >&6;}
+fi
+
 if test "x$MCKERNEL_INCDIR" != "x" ; then
 
 cat >>confdefs.h <<_ACEOF
@@ -5052,6 +4876,9 @@ fi
 
 
 
+
+
+
 
 
 
@@ -5060,9 +4887,14 @@ ac_config_headers="$ac_config_headers config.h"
 
 # POSTK_DEBUG_ARCH_DEP_37
 # AC_CONFIG_FILES arch dependfiles separate
-ac_config_files="$ac_config_files Makefile executer/user/Makefile executer/user/mcexec.1:executer/user/mcexec.1in executer/user/vmcore2mckdump executer/user/arch/$ARCH/Makefile executer/user/arch/x86_64/Makefile executer/kernel/mcctrl/Makefile executer/kernel/mcctrl/arch/$ARCH/Makefile executer/kernel/mcoverlayfs/Makefile executer/kernel/mcoverlayfs/linux-3.10.0-327.36.1.el7/Makefile executer/kernel/mcoverlayfs/linux-4.0.9/Makefile executer/kernel/mcoverlayfs/linux-4.6.7/Makefile executer/include/qlmpilib.h kernel/Makefile kernel/Makefile.build kernel/include/swapfmt.h arch/x86_64/tools/mcreboot-attached-mic.sh arch/x86_64/tools/mcshutdown-attached-mic.sh arch/x86_64/tools/mcreboot-builtin-x86.sh arch/x86_64/tools/mcreboot-smp-x86.sh arch/x86_64/tools/mcstop+release-smp-x86.sh arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh arch/x86_64/tools/mcoverlay-create-smp-x86.sh arch/x86_64/tools/eclair-dump-backtrace.exp arch/x86_64/tools/mcshutdown-builtin-x86.sh arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in arch/x86_64/tools/irqbalance_mck.service arch/x86_64/tools/irqbalance_mck.in tools/mcstat/Makefile"
+ac_config_files="$ac_config_files Makefile executer/user/Makefile executer/user/mcexec.1:executer/user/mcexec.1in executer/user/vmcore2mckdump executer/user/arch/$ARCH/Makefile executer/user/arch/x86_64/Makefile executer/kernel/mcctrl/Makefile executer/kernel/mcctrl/arch/$ARCH/Makefile executer/kernel/mcoverlayfs/Makefile executer/kernel/mcoverlayfs/linux-3.10.0-327.36.1.el7/Makefile executer/kernel/mcoverlayfs/linux-4.0.9/Makefile executer/kernel/mcoverlayfs/linux-4.6.7/Makefile executer/include/qlmpilib.h kernel/Makefile kernel/Makefile.build kernel/include/swapfmt.h arch/x86_64/tools/mcreboot-attached-mic.sh arch/x86_64/tools/mcshutdown-attached-mic.sh arch/x86_64/tools/mcreboot-builtin-x86.sh arch/x86_64/tools/mcreboot-smp-x86.sh arch/x86_64/tools/mcstop+release-smp-x86.sh arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh arch/x86_64/tools/mcoverlay-create-smp-x86.sh arch/x86_64/tools/eclair-dump-backtrace.exp arch/x86_64/tools/mcshutdown-builtin-x86.sh arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in arch/x86_64/tools/irqbalance_mck.service arch/x86_64/tools/irqbalance_mck.in tools/mcstat/mcstat.1:tools/mcstat/mcstat.1in tools/mcstat/Makefile"
 
 
+if test -e "${ABS_SRCDIR}/test"; then
+ac_config_files="$ac_config_files mck_test_config.sample:test/mck_test_config.sample.in"
+
+fi
+
 if test "$TARGET" = "smp-x86"; then
 ac_config_files="$ac_config_files arch/x86_64/kernel/Makefile.arch"
 
@@ -5797,7 +5629,9 @@ do
     "arch/x86_64/tools/mcreboot.1") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in" ;;
     "arch/x86_64/tools/irqbalance_mck.service") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/irqbalance_mck.service" ;;
     "arch/x86_64/tools/irqbalance_mck.in") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/irqbalance_mck.in" ;;
+    "tools/mcstat/mcstat.1") CONFIG_FILES="$CONFIG_FILES tools/mcstat/mcstat.1:tools/mcstat/mcstat.1in" ;;
     "tools/mcstat/Makefile") CONFIG_FILES="$CONFIG_FILES tools/mcstat/Makefile" ;;
+    "mck_test_config.sample") CONFIG_FILES="$CONFIG_FILES mck_test_config.sample:test/mck_test_config.sample.in" ;;
     "arch/x86_64/kernel/Makefile.arch") CONFIG_FILES="$CONFIG_FILES arch/x86_64/kernel/Makefile.arch" ;;
     "kernel/config/config.smp-arm64") CONFIG_FILES="$CONFIG_FILES kernel/config/config.smp-arm64" ;;
     "arch/arm64/kernel/vdso/Makefile") CONFIG_FILES="$CONFIG_FILES arch/arm64/kernel/vdso/Makefile" ;;

configure.ac

@@ -77,6 +77,58 @@ AC_DEFUN([PAC_SET_HEADER_LIB_PATH],[
           ])
 ])
 
+AC_DEFUN([PAC_SET_HEADER_LIB_PATH_SYSCALL_INTERCEPT],[
+    AC_ARG_WITH([$1],
+                [AC_HELP_STRING([--with-$1=PATH],
+                                [specify path where $1 include directory and lib directory can be found])],
+
+                [AS_CASE(["$withval"],
+                         [yes|no|''],
+                         [AC_MSG_WARN([--with[out]-$1=PATH expects a valid PATH])
+                          with_$1=""])],
+                [with_$1=$2])
+    AC_ARG_WITH([$1-include],
+                [AC_HELP_STRING([--with-$1-include=PATH],
+                                [specify path where $1 include directory can be found])],
+                [AS_CASE(["$withval"],
+                         [yes|no|''],
+                         [AC_MSG_WARN([--with[out]-$1-include=PATH expects a valid PATH])
+                          with_$1_include=""])],
+                [])
+    AC_ARG_WITH([$1-lib],
+                [AC_HELP_STRING([--with-$1-lib=PATH],
+                                [specify path where $1 lib directory can be found])],
+                [AS_CASE(["$withval"],
+                         [yes|no|''],
+                         [AC_MSG_WARN([--with[out]-$1-lib=PATH expects a valid PATH])
+                          with_$1_lib=""])],
+                [])
+
+    # The args have been sanitized into empty/non-empty values above.
+    # Now append -I/-L args to CPPFLAGS/LDFLAGS, with more specific options
+    # taking priority
+
+    AS_IF([test -n "${with_$1_include}"],
+          [PAC_APPEND_FLAG([-I${with_$1_include}],[CPPFLAGS_SYSCALL_INTERCEPT])],
+          [AS_IF([test -n "${with_$1}"],
+                 [PAC_APPEND_FLAG([-I${with_$1}/include],[CPPFLAGS_SYSCALL_INTERCEPT])])])
+
+    AS_IF([test -n "${with_$1_lib}"],
+          [PAC_APPEND_FLAG([-L${with_$1_lib} -Wl,-rpath,${with_$1_lib}],[LDFLAGS_SYSCALL_INTERCEPT])],
+          [AS_IF([test -n "${with_$1}"],
+                 dnl is adding lib64 by default really the right thing to do?  What if
+                 dnl we are on a 32-bit host that happens to have both lib dirs available?
+                 [PAC_APPEND_FLAG([-L${with_$1}/lib -Wl,-rpath,${with_$1}/lib],[LDFLAGS_SYSCALL_INTERCEPT])
+                  AS_IF([test -d "${with_$1}/lib64"],
+		        [PAC_APPEND_FLAG([-L${with_$1}/lib64 -Wl,-rpath,${with_$1}/lib64],[LDFLAGS_SYSCALL_INTERCEPT])])
+                 ])
+          ])
+
+    AS_IF([test -n "${with_$1}" || test -n "${with_$1_include}" || test -n "${with_$1_lib}"],
+	      [WITH_SYSCALL_INTERCEPT=yes],
+	      [WITH_SYSCALL_INTERCEPT=no])
+])
+
 IHK_VERSION=IHK_VERSION_m4
 MCKERNEL_VERSION=MCKERNEL_VERSION_m4
 DCFA_VERSION=DCFA_VERSION_m4
@@ -95,6 +147,23 @@ AS_IF([test "x$numa_lib_found" != "xyes"],
 
 PAC_SET_HEADER_LIB_PATH([mpi])
 
+PAC_SET_HEADER_LIB_PATH_SYSCALL_INTERCEPT([syscall_intercept])
+
+if test "x$WITH_SYSCALL_INTERCEPT" == "xno" ; then
+	AC_CHECK_LIB([syscall_intercept],[syscall_no_intercept],[syscall_intercept_lib_found=yes],[syscall_intercept_lib_found=no],[-lcapstone -ldl])
+	AS_IF([test "x$syscall_intercept_lib_found" != "xyes"],
+		[AC_MSG_NOTICE([libsyscall_intercept.so not found])])
+
+	AC_CHECK_HEADER([libsyscall_intercept_hook_point.h],[syscall_intercept_header_found=yes],[syscall_intercept_header_found=no])
+	AS_IF([test "x$syscall_intercept_header_found" != "xyes"],
+		[AC_MSG_NOTICE([libsyscall_intercept_hook_point.h not found])])
+
+	AS_IF([test "x$syscall_intercept_lib_found" == "xyes" && test "x$syscall_intercept_header_found" == "xyes"],
+		[WITH_SYSCALL_INTERCEPT=yes],
+		[WITH_SYSCALL_INTERCEPT=no])
+fi
+
+
 AC_ARG_WITH([kernelsrc],
   AC_HELP_STRING(
     [--with-kernelsrc=path],[Path to 'kernel src', default is /lib/modules/uname_r/build]),
@@ -339,78 +408,6 @@ KDIR="$WITH_KERNELSRC"
 UNAME_R="$WITH_UNAME_R"
 TARGET="$WITH_TARGET"
 
-MCCTRL_LINUX_SYMTAB=""
-case "X$WITH_SYSTEM_MAP" in
-    Xyes | Xno | X)
-	MCCTRL_LINUX_SYMTAB=""
-	;;
-    *)
-	MCCTRL_LINUX_SYMTAB="$WITH_SYSTEM_MAP"
-	;;
-esac
-
-AC_MSG_CHECKING([[for System.map]])
-if test -r "$MCCTRL_LINUX_SYMTAB"; then
-	MCCTRL_LINUX_SYMTAB="$MCCTRL_LINUX_SYMTAB"
-elif test -r "/boot/System.map-`uname -r`"; then
-	MCCTRL_LINUX_SYMTAB="/boot/System.map-`uname -r`"
-elif test -r "$KDIR/System.map"; then
-	MCCTRL_LINUX_SYMTAB="$KDIR/System.map"
-fi
-
-if test "$MCCTRL_LINUX_SYMTAB" == ""; then
-	AC_MSG_ERROR([could not find])
-fi
-
-if test -z "`eval cat $MCCTRL_LINUX_SYMTAB`"; then
-	AC_MSG_ERROR([could not read System.map file, no read permission?])
-fi
-AC_MSG_RESULT([$MCCTRL_LINUX_SYMTAB])
-
-MCCTRL_LINUX_SYMTAB_CMD="cat $MCCTRL_LINUX_SYMTAB"
-
-# MCCTRL_FIND_KSYM(SYMBOL)
-# ------------------------------------------------------
-# Search System.map for address of the given symbol and
-# do one of three things in config.h:
-# If not found, leave MCCTRL_KSYM_foo undefined
-# If found to be exported, "#define MCCTRL_KSYM_foo 0"
-# If found not to be exported, "#define MCCTRL_KSYM_foo 0x<value>"
-AC_DEFUN([MCCTRL_FIND_KSYM],[
-  AC_MSG_CHECKING([[System.map for symbol $1]])
-  mcctrl_addr=`eval $MCCTRL_LINUX_SYMTAB_CMD | grep " $1\$" | cut -d\  -f1`
-  if test -z $mcctrl_addr; then
-    AC_MSG_RESULT([not found])
-  else
-    mcctrl_result=$mcctrl_addr
-    mcctrl_addr="0x$mcctrl_addr"
-    m4_ifval([$2],[],[
-      if `eval $MCCTRL_LINUX_SYMTAB_CMD | grep " __ksymtab_$1\$" >/dev/null`; then
-        mcctrl_result="exported"
-        mcctrl_addr="0"
-      fi
-    ])
-    AC_MSG_RESULT([$mcctrl_result])
-    AC_DEFINE_UNQUOTED(MCCTRL_KSYM_[]$1,$mcctrl_addr,[Define to address of kernel symbol $1, or 0 if exported])
-  fi
-])
-
-MCCTRL_FIND_KSYM([sys_mount])
-MCCTRL_FIND_KSYM([sys_umount])
-MCCTRL_FIND_KSYM([sys_unshare])
-MCCTRL_FIND_KSYM([zap_page_range])
-MCCTRL_FIND_KSYM([vdso_image_64])
-MCCTRL_FIND_KSYM([vdso_start])
-MCCTRL_FIND_KSYM([vdso_end])
-MCCTRL_FIND_KSYM([vdso_pages])
-MCCTRL_FIND_KSYM([__vvar_page])
-MCCTRL_FIND_KSYM([hpet_address])
-# POSTK_DEBUG_ARCH_DEP_50, add:find kernel symbol.
-MCCTRL_FIND_KSYM([vdso_spec])
-MCCTRL_FIND_KSYM([hv_clock])
-MCCTRL_FIND_KSYM([sys_readlink])
-MCCTRL_FIND_KSYM([walk_page_range])
-
 case $ENABLE_MEMDUMP in
 	yes|no|auto)
 		;;
@@ -489,6 +486,13 @@ else
 	AC_MSG_NOTICE([perf is disabled])
 fi
 
+if test "x$WITH_SYSCALL_INTERCEPT" = "xyes" ; then
+	AC_DEFINE([WITH_SYSCALL_INTERCEPT],[1],[whether or not syscall_intercept library is linked])
+	AC_MSG_NOTICE([syscall_intercept library is linked])
+else
+	AC_MSG_NOTICE([syscall_intercept library isn't linked])
+fi
+
 if test "x$MCKERNEL_INCDIR" != "x" ; then
    AC_DEFINE_UNQUOTED(MCKERNEL_INCDIR,"$MCKERNEL_INCDIR",[McKernel specific headers])
 fi
@@ -526,9 +530,12 @@ AC_SUBST(KMODDIR)
 AC_SUBST(KERNDIR)
 AC_SUBST(MANDIR)
 AC_SUBST(CFLAGS)
+AC_SUBST(CPPFLAGS_SYSCALL_INTERCEPT)
+AC_SUBST(LDFLAGS_SYSCALL_INTERCEPT)
 AC_SUBST(ENABLE_MCOVERLAYFS)
 AC_SUBST(ENABLE_RUSAGE)
 AC_SUBST(ENABLE_QLMPI)
+AC_SUBST(WITH_SYSCALL_INTERCEPT)
 
 AC_SUBST(IHK_VERSION)
 AC_SUBST(MCKERNEL_VERSION)
@@ -570,9 +577,16 @@ AC_CONFIG_FILES([
 	arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in
 	arch/x86_64/tools/irqbalance_mck.service
 	arch/x86_64/tools/irqbalance_mck.in
+	tools/mcstat/mcstat.1:tools/mcstat/mcstat.1in
 	tools/mcstat/Makefile
 ])
 
+if test -e "${ABS_SRCDIR}/test"; then
+AC_CONFIG_FILES([
+mck_test_config.sample:test/mck_test_config.sample.in
+])
+fi
+
 if test "$TARGET" = "smp-x86"; then
 AC_CONFIG_FILES([
 	arch/x86_64/kernel/Makefile.arch

executer/include/uprotocol.h

@@ -55,13 +55,14 @@
 #define MCEXEC_UP_SYS_UMOUNT     0x30a02915
 #define MCEXEC_UP_SYS_UNSHARE    0x30a02916
 
-#define MCEXEC_UP_UTIL_THREAD1   0x30a02920
+#define MCEXEC_UP_UTI_GET_CTX    0x30a02920
-#define MCEXEC_UP_UTIL_THREAD2   0x30a02921
+#define MCEXEC_UP_UTI_SAVE_FS    0x30a02921
 #define MCEXEC_UP_SIG_THREAD     0x30a02922
 #define MCEXEC_UP_SYSCALL_THREAD 0x30a02924
 #define MCEXEC_UP_TERMINATE_THREAD 0x30a02925
 #define MCEXEC_UP_GET_NUM_POOL_THREADS  0x30a02926
 #define MCEXEC_UP_UTI_ATTR       0x30a02927
+#define MCEXEC_UP_RELEASE_USER_SPACE 0x30a02928
 
 #define MCEXEC_UP_DEBUG_LOG     0x40000000
 
@@ -91,6 +92,7 @@ struct program_image_section {
 
 struct get_cpu_set_arg {
 	int nr_processes;
+	int *process_rank;
 	void *cpu_set;
 	size_t cpu_set_size;	// Size in bytes
 	int *target_core;
@@ -140,8 +142,10 @@ struct program_load_desc {
 	unsigned long heap_extension;
 	long stack_premap;
 	unsigned long mpol_bind_mask;
+	int uti_thread_rank; /* N-th clone() spawns a thread on Linux CPU */
+	int uti_use_last_cpu; /* Work-around not to share CPU with OpenMP thread */
 	int nr_processes;
-	char shell_path[SHELL_PATH_MAX_LEN];
+	int process_rank;
 	__cpu_set_unit cpu_set[PLD_CPU_SET_SIZE];
 	int profile;
 	struct program_image_section sections[0];
@@ -242,6 +246,28 @@ struct sys_unshare_desc {
 	unsigned long unshare_flags;
 };
 
+struct release_user_space_desc {
+	unsigned long user_start;
+	unsigned long user_end;
+};
+
+struct terminate_thread_desc {
+	int pid;
+	int tid;
+
+	long code; 
+	/* 32------32  31--16  15--------8  7----0
+       exit_group          exit-status  signal */
+
+	unsigned long tsk; /* struct task_struct * */
+};
+
+struct rpgtable_desc {
+	uintptr_t rpgtable;
+	uintptr_t start;
+	uintptr_t len;
+};
+
 enum perf_ctrl_type {
 	PERF_CTRL_SET,
 	PERF_CTRL_GET,
@@ -251,6 +277,7 @@ enum perf_ctrl_type {
 
 struct perf_ctrl_desc {
 	enum perf_ctrl_type ctrl_type;
+	int err;
 	union {
 		/* for SET, GET */
 		struct {
@@ -290,6 +317,10 @@ struct perf_ctrl_desc {
 #define UTI_FLAG_HIGH_PRIORITY (1ULL<<12)
 #define UTI_FLAG_NON_COOPERATIVE (1ULL<<13)
 
+#define UTI_FLAG_PREFER_LWK (1ULL << 14)
+#define UTI_FLAG_PREFER_FWK (1ULL << 15)
+#define UTI_FLAG_FABRIC_INTR_AFFINITY (1ULL << 16)
+
 /* Linux default value is used */
 #define UTI_MAX_NUMA_DOMAINS (1024)
 
@@ -308,6 +339,30 @@ struct kuti_attr {
 
 struct uti_attr_desc {
 	unsigned long phys_attr;
+	char *uti_cpu_set_str; /* UTI_CPU_SET environmental variable */
+	size_t uti_cpu_set_len;
+};
+
+struct uti_ctx {
+	union {
+		char ctx[4096]; /* TODO: Get the size from config.h */
+		struct {
+			int uti_refill_tid;
+		};
+	};
+}; 
+
+struct uti_get_ctx_desc {
+	unsigned long rp_rctx; /* Remote physical address of remote context */
+	void *rctx; /* Remote context */
+	void *lctx; /* Local context */
+	int uti_refill_tid;
+	unsigned long key; /* OUT: struct task_struct* of mcexec thread, used to search struct host_thread */
+};
+
+struct uti_save_fs_desc {
+	void *rctx; /* Remote context */
+	void *lctx; /* Local context */
 };
 
 #endif

executer/include/uti.h

@@ -0,0 +1,31 @@
+#ifndef UTI_H_INCLUDED
+#define UTI_H_INCLUDED
+
+struct syscall_struct {
+	int number;
+	unsigned long args[6];
+	unsigned long ret;
+	unsigned long uti_clv; /* copy of a clv in McKernel */
+};
+
+#define UTI_SZ_SYSCALL_STACK 16
+
+/* Variables accessed by mcexec.c and syscall_intercept.c */
+struct uti_desc {
+	char lctx[4096]; /* TODO: Get the size from config.h */
+	char rctx[4096]; /* TODO: Get the size from config.h */
+	int mck_tid; /* TODO: Move this out for multiple migrated-to-Linux threads */
+	unsigned long key; /* struct task_struct* of mcexec thread, used to search struct host_thread */
+	int pid, tid; /* Used as the id of tracee when issuing MCEXEC_UP_TERMINATE_THREAD */
+	unsigned long uti_clv; /* copy of McKernel clv */
+
+	int fd; /* /dev/mcosX */
+	struct syscall_struct syscall_stack[UTI_SZ_SYSCALL_STACK]; /* stack of system call arguments and return values */
+	int syscall_stack_top; /* stack-pointer of syscall arguments list */
+	long syscalls[512], syscalls2[512]; /* Syscall profile counters */
+	int start_syscall_intercept; /* Used to sync between mcexec.c and syscall_intercept.c */
+};
+
+
+#endif
+

executer/kernel/mcctrl/arch/arm64/archdeps.c

@@ -1,6 +1,7 @@
 /* archdeps.c COPYRIGHT FUJITSU LIMITED 2016 */
 #include <linux/version.h>
 #include <linux/mm_types.h>
+#include <linux/kallsyms.h>
 #include <asm/vdso.h>
 #include "../../../config.h"
 #include "../../mcctrl.h"
@@ -17,29 +18,31 @@
 
 #define D(fmt, ...) printk("%s(%d) " fmt, __func__, __LINE__, ##__VA_ARGS__)
 
-#ifdef MCCTRL_KSYM_vdso_start
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 0, 0)
-# if MCCTRL_KSYM_vdso_start
+void *vdso_start;
-void *vdso_start = (void *)MCCTRL_KSYM_vdso_start;
+void *vdso_end;
-# endif
+static struct vm_special_mapping (*vdso_spec)[2];
-#else
-# error missing address of vdso_start.
 #endif
 
-#ifdef MCCTRL_KSYM_vdso_end
+int arch_symbols_init(void)
-# if MCCTRL_KSYM_vdso_end
+{
-void *vdso_end = (void *)MCCTRL_KSYM_vdso_end;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 0, 0)
-# endif
+	vdso_start = (void *) kallsyms_lookup_name("vdso_start");
-#else
+	if (WARN_ON(!vdso_start))
-# error missing address of vdso_end.
+		return -EFAULT;
+
+	vdso_end = (void *) kallsyms_lookup_name("vdso_end");
+	if (WARN_ON(!vdso_end))
+		return -EFAULT;
+
+	vdso_spec = (void *) kallsyms_lookup_name("vdso_spec");
+	if (WARN_ON(!vdso_spec))
+		return -EFAULT;
 #endif
 
-#ifdef MCCTRL_KSYM_vdso_spec
+	return 0;
-# if MCCTRL_KSYM_vdso_spec
+}
-static struct vm_special_mapping (*vdso_spec)[2] = (void*)MCCTRL_KSYM_vdso_spec;
+
-# endif
-#else
-# error missing address of vdso_spec.
-#endif
 
 #ifdef POSTK_DEBUG_ARCH_DEP_52
 #define VDSO_MAXPAGES 1

executer/kernel/mcctrl/arch/x86_64/archdeps.c

@@ -1,5 +1,6 @@
 /* archdeps.c COPYRIGHT FUJITSU LIMITED 2016 */
 #include <linux/version.h>
+#include <linux/kallsyms.h>
 #include "../../../config.h"
 #include "../../mcctrl.h"
 
@@ -13,57 +14,46 @@
 #endif
 #endif /* POSTK_DEBUG_ARCH_DEP_83 */
 
-#ifdef MCCTRL_KSYM_vdso_image_64
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
-#if MCCTRL_KSYM_vdso_image_64
+static struct vdso_image *vdso_image_64;
-struct vdso_image *vdso_image = (void *)MCCTRL_KSYM_vdso_image_64;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
+static void *vdso_start;
+static void *vdso_end;
+static struct page **vdso_pages;
 #endif
+static void *__vvar_page;
+static long *hpet_address;
+static void **hv_clock;
+
+int arch_symbols_init(void)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
+	vdso_image_64 = (void *) kallsyms_lookup_name("vdso_image_64");
+	if (WARN_ON(!vdso_image_64))
+		return -EFAULT;
+#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
+	vdso_start = (void *) kallsyms_lookup_name("vdso_start");
+	if (WARN_ON(!vdso_start))
+		return -EFAULT;
+
+	vdso_end = (void *) kallsyms_lookup_name("vdso_end");
+	if (WARN_ON(!vdso_end))
+		return -EFAULT;
+
+	vdso_pages = (void *) kallsyms_lookup_name("vdso_pages");
+	if (WARN_ON(!vdso_pages))
+		return -EFAULT;
 #endif
 
-#ifdef MCCTRL_KSYM_vdso_start
+	__vvar_page = (void *) kallsyms_lookup_name("__vvar_page");
-#if MCCTRL_KSYM_vdso_start
+	if (WARN_ON(!__vvar_page))
-void *vdso_start = (void *)MCCTRL_KSYM_vdso_start;
+		return -EFAULT;
-#endif
-#endif
 
-#ifdef MCCTRL_KSYM_vdso_end
+	hpet_address = (void *) kallsyms_lookup_name("hpet_address");
-#if MCCTRL_KSYM_vdso_end
+	hv_clock = (void *) kallsyms_lookup_name("hv_clock");
-void *vdso_end = (void *)MCCTRL_KSYM_vdso_end;
+	return 0;
-#endif
+}
-#endif
 
-#ifdef MCCTRL_KSYM_vdso_pages
-#if MCCTRL_KSYM_vdso_pages
-struct page **vdso_pages = (void *)MCCTRL_KSYM_vdso_pages;
-#endif
-#endif
-
-#ifdef MCCTRL_KSYM___vvar_page
-#if MCCTRL_KSYM___vvar_page
-void *__vvar_page = (void *)MCCTRL_KSYM___vvar_page;
-#endif
-#endif
-
-long *hpet_addressp
-#ifdef MCCTRL_KSYM_hpet_address
-#if MCCTRL_KSYM_hpet_address
-	= (void *)MCCTRL_KSYM_hpet_address;
-#else
-	= &hpet_address;
-#endif
-#else
-	= NULL;
-#endif
-
-void **hv_clockp
-#ifdef MCCTRL_KSYM_hv_clock
-#if MCCTRL_KSYM_hv_clock
-	= (void *)MCCTRL_KSYM_hv_clock;
-#else
-	= &hv_clock;
-#endif
-#else
-	= NULL;
-#endif
 
 #ifdef POSTK_DEBUG_ARCH_DEP_52
 #define VDSO_MAXPAGES 2
@@ -138,7 +128,7 @@ void get_vdso_info(ihk_os_t os, long vdso_rpa)
 
 	/* VDSO pages */
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,16,0)
-	size = vdso_image->size;
+	size = vdso_image_64->size;
 	vdso->vdso_npages = size >> PAGE_SHIFT;
 
 	if (vdso->vdso_npages > VDSO_MAXPAGES) {
@@ -148,7 +138,7 @@ void get_vdso_info(ihk_os_t os, long vdso_rpa)
 
 	for (i = 0; i < vdso->vdso_npages; ++i) {
 		vdso->vdso_physlist[i] = virt_to_phys(
-				vdso_image->data + (i * PAGE_SIZE));
+				vdso_image_64->data + (i * PAGE_SIZE));
 	}
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 	size = vdso_end - vdso_start;
@@ -185,36 +175,36 @@ void get_vdso_info(ihk_os_t os, long vdso_rpa)
 #endif
 
 	/* HPET page */
-	if (hpet_addressp && *hpet_addressp) {
+	if (hpet_address && *hpet_address) {
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,5,0)
 		vdso->hpet_is_global = 0;
 		vdso->hpet_virt = (void *)(-2 * PAGE_SIZE);
-		vdso->hpet_phys = *hpet_addressp;
+		vdso->hpet_phys = *hpet_address;
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0)
 		vdso->hpet_is_global = 0;
 		vdso->hpet_virt = (void *)(-1 * PAGE_SIZE);
-		vdso->hpet_phys = *hpet_addressp;
+		vdso->hpet_phys = *hpet_address;
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(3,16,0)
 		vdso->hpet_is_global = 0;
 		vdso->hpet_virt = (void *)((vdso->vdso_npages + 1) * PAGE_SIZE);
-		vdso->hpet_phys = *hpet_addressp;
+		vdso->hpet_phys = *hpet_address;
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 		vdso->hpet_is_global = 1;
 		vdso->hpet_virt = (void *)fix_to_virt(VSYSCALL_HPET);
-		vdso->hpet_phys = *hpet_addressp;
+		vdso->hpet_phys = *hpet_address;
 #endif
 	}
 
 	/* struct pvlock_vcpu_time_info table */
-	if (hv_clockp && *hv_clockp) {
+	if (hv_clock && *hv_clock) {
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,5,0)
 		vdso->pvti_is_global = 0;
 		vdso->pvti_virt = (void *)(-1 * PAGE_SIZE);
-		vdso->pvti_phys = virt_to_phys(*hv_clockp);
+		vdso->pvti_phys = virt_to_phys(*hv_clock);
 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(3,8,0)
 		vdso->pvti_is_global = 1;
 		vdso->pvti_virt = (void *)fix_to_virt(PVCLOCK_FIXMAP_BEGIN);
-		vdso->pvti_phys = virt_to_phys(*hv_clockp);
+		vdso->pvti_phys = virt_to_phys(*hv_clock);
 #endif
 	}
 
@@ -289,6 +279,14 @@ get_fs_ctx(void *ctx)
 	return tctx->fs;
 }
 
+unsigned long
+get_rsp_ctx(void *ctx)
+{
+	struct trans_uctx *tctx = ctx;
+
+	return tctx->rsp;
+}
+
 #ifdef POSTK_DEBUG_ARCH_DEP_83 /* arch depend translate_rva_to_rpa() move */
 int translate_rva_to_rpa(ihk_os_t os, unsigned long rpt, unsigned long rva,
 		unsigned long *rpap, unsigned long *pgsizep)

executer/kernel/mcctrl/binfmt_mcexec.c

@@ -125,7 +125,6 @@ static int load_elf(struct linux_binprm *bprm
 		for(i = 0, st = 0; mode != 2;){
 			if(st == 0){
 				off = p & ~PAGE_MASK;
-#ifdef POSTK_DEBUG_ARCH_DEP_41 /* HOST-Linux version switch add */
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)
 				rc = get_user_pages_remote(current, bprm->mm,
 				                        bprm->p, 1, FOLL_FORCE, &page, NULL, NULL);
@@ -141,17 +140,6 @@ static int load_elf(struct linux_binprm *bprm
 				                        bprm->p, 1, 0, 1,
 				                        &page, NULL);
 #endif
-#else /* POSTK_DEBUG_ARCH_DEP_41 */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0)
-				rc = get_user_pages_remote(current, bprm->mm,
-				                        bprm->p, 1, 0, 1,
-				                        &page, NULL);
-#else
-				rc = get_user_pages(current, bprm->mm,
-				                        bprm->p, 1, 0, 1,
-				                        &page, NULL);
-#endif
-#endif /* POSTK_DEBUG_ARCH_DEP_41 */
 				if(rc <= 0) {
 					kfree(pbuf);
 					return -EFAULT;

executer/kernel/mcctrl/driver.c

@@ -28,6 +28,7 @@
 #include <linux/slab.h>
 #include <linux/device.h>
 #include <linux/delay.h>
+#include <linux/kallsyms.h>
 #include "mcctrl.h"
 #include <ihk/ihk_host_user.h>
 
@@ -43,8 +44,6 @@ extern void mcctrl_syscall_init(void);
 extern void procfs_init(int);
 extern void procfs_exit(int);
 
-extern void rus_page_hash_init(void);
-extern void rus_page_hash_put_pages(void);
 extern void uti_attr_finalize(void);
 extern void binfmt_mcexec_init(void);
 extern void binfmt_mcexec_exit(void);
@@ -84,13 +83,14 @@ static struct ihk_os_user_call_handler mcctrl_uchs[] = {
 	{ .request = MCEXEC_UP_SYS_MOUNT, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_SYS_UMOUNT, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_SYS_UNSHARE, .func = mcctrl_ioctl },
-	{ .request = MCEXEC_UP_UTIL_THREAD1, .func = mcctrl_ioctl },
+	{ .request = MCEXEC_UP_UTI_GET_CTX, .func = mcctrl_ioctl },
-	{ .request = MCEXEC_UP_UTIL_THREAD2, .func = mcctrl_ioctl },
+	{ .request = MCEXEC_UP_UTI_SAVE_FS, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_SIG_THREAD, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_SYSCALL_THREAD, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_TERMINATE_THREAD, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_GET_NUM_POOL_THREADS, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_UTI_ATTR, .func = mcctrl_ioctl },
+	{ .request = MCEXEC_UP_RELEASE_USER_SPACE, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_DEBUG_LOG, .func = mcctrl_ioctl },
 	{ .request = IHK_OS_AUX_PERF_NUM, .func = mcctrl_ioctl },
 	{ .request = IHK_OS_AUX_PERF_SET, .func = mcctrl_ioctl },
@@ -178,6 +178,7 @@ int mcctrl_os_shutdown_notifier(int os_index)
 			mdelay(200);
 		}
 
+		pager_cleanup();
 		sysfsm_cleanup(os[os_index]);
 		free_topology_info(os[os_index]);
 		ihk_os_unregister_user_call_handlers(os[os_index], mcctrl_uc + os_index);
@@ -185,9 +186,6 @@ int mcctrl_os_shutdown_notifier(int os_index)
 		destroy_ikc_channels(os[os_index]);
 		procfs_exit(os_index);
 	}
-#ifdef POSTK_DEBUG_TEMP_FIX_35 /* in shutdown phase, rus_page_hash_put_pages() call added. */
-	rus_page_hash_put_pages();
-#endif /* POSTK_DEBUG_TEMP_FIX_35 */
 
 	os[os_index] = NULL;
 
@@ -214,6 +212,68 @@ static struct ihk_os_notifier mcctrl_os_notifier = {
 	.ops = &mcctrl_os_notifier_ops,
 };
 
+
+
+int (*mcctrl_sys_mount)(char *dev_name, char *dir_name, char *type,
+			unsigned long flags, void *data);
+int (*mcctrl_sys_umount)(char *dir_name, int flags);
+int (*mcctrl_sys_unshare)(unsigned long unshare_flags);
+long (*mcctrl_sched_setaffinity)(pid_t pid, const struct cpumask *in_mask);
+int (*mcctrl_sched_setscheduler_nocheck)(struct task_struct *p, int policy,
+					 const struct sched_param *param);
+
+ssize_t (*mcctrl_sys_readlink)(const char *path, char *buf,
+			       size_t bufsiz);
+void (*mcctrl_zap_page_range)(struct vm_area_struct *vma,
+			      unsigned long start,
+			      unsigned long size,
+			      struct zap_details *details);
+
+struct inode_operations *mcctrl_hugetlbfs_inode_operations;
+
+
+static int symbols_init(void)
+{
+	mcctrl_sys_mount = (void *) kallsyms_lookup_name("sys_mount");
+	if (WARN_ON(!mcctrl_sys_mount))
+		return -EFAULT;
+
+	mcctrl_sys_umount = (void *) kallsyms_lookup_name("sys_umount");
+	if (WARN_ON(!mcctrl_sys_umount))
+		return -EFAULT;
+
+	mcctrl_sys_unshare = (void *) kallsyms_lookup_name("sys_unshare");
+	if (WARN_ON(!mcctrl_sys_unshare))
+		return -EFAULT;
+
+	mcctrl_sched_setaffinity =
+		(void *) kallsyms_lookup_name("sched_setaffinity");
+	if (WARN_ON(!mcctrl_sched_setaffinity))
+		return -EFAULT;
+
+	mcctrl_sched_setscheduler_nocheck =
+		(void *) kallsyms_lookup_name("sched_setscheduler_nocheck");
+	if (WARN_ON(!mcctrl_sched_setscheduler_nocheck))
+		return -EFAULT;
+
+	mcctrl_sys_readlink =
+		(void *) kallsyms_lookup_name("sys_readlink");
+	if (WARN_ON(!mcctrl_sys_readlink))
+		return -EFAULT;
+
+	mcctrl_zap_page_range =
+		(void *) kallsyms_lookup_name("zap_page_range");
+	if (WARN_ON(!mcctrl_zap_page_range))
+		return -EFAULT;
+
+	mcctrl_hugetlbfs_inode_operations =
+		(void *) kallsyms_lookup_name("hugetlbfs_inode_operations");
+	if (WARN_ON(!mcctrl_hugetlbfs_inode_operations))
+		return -EFAULT;
+
+	return arch_symbols_init();
+}
+
 static int __init mcctrl_init(void)
 {
 	int ret = 0;
@@ -227,10 +287,11 @@ static int __init mcctrl_init(void)
 		os[i] = NULL;
 	}
 
-	rus_page_hash_init();
-
 	binfmt_mcexec_init();
 
+	if ((ret = symbols_init()))
+		goto error;
+
 	if ((ret = ihk_host_register_os_notifier(&mcctrl_os_notifier)) != 0) {
 		printk("mcctrl: error: registering OS notifier\n");
 		goto error;
@@ -241,7 +302,6 @@ static int __init mcctrl_init(void)
 
 error:
 	binfmt_mcexec_exit();
-	rus_page_hash_put_pages();
 
 	return ret;
 }
@@ -253,7 +313,6 @@ static void __exit mcctrl_exit(void)
 	}
 
 	binfmt_mcexec_exit();
-	rus_page_hash_put_pages();
 	uti_attr_finalize();
 
 	printk("mcctrl: unregistered.\n");

executer/kernel/mcctrl/ikc.c

@@ -52,6 +52,8 @@
 static void mcctrl_ikc_init(ihk_os_t os, int cpu, unsigned long rphys, struct ihk_ikc_channel_desc *c);
 int mcexec_syscall(struct mcctrl_usrdata *ud, struct ikc_scd_packet *packet);
 void sig_done(unsigned long arg, int err);
+void mcctrl_perf_ack(ihk_os_t os, struct ikc_scd_packet *packet);
+void mcctrl_futex_wake(struct ikc_scd_packet *pisp);
 void mcctrl_os_read_write_cpu_response(ihk_os_t os,
 		struct ikc_scd_packet *pisp);
 void mcctrl_eventfd(ihk_os_t os, struct ikc_scd_packet *pisp);
@@ -154,7 +156,7 @@ int mcctrl_ikc_send_wait(ihk_os_t os, int cpu, struct ikc_scd_packet *pisp,
 		spin_lock_irqsave(&usrdata->wakeup_descs_lock, flags);
 		list_add(&desc->chain, &usrdata->wakeup_descs_list);
 		spin_unlock_irqrestore(&usrdata->wakeup_descs_lock, flags);
-		if (free_addrs_count)
+		if (do_frees)
 			*do_frees = 0;
 		return ret < 0 ? ret : -ETIME;
 	}
@@ -182,6 +184,7 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 	case SCD_MSG_PREPARE_PROCESS_ACKED:
 	case SCD_MSG_PERF_ACK:
 	case SCD_MSG_SEND_SIGNAL_ACK:
+	case SCD_MSG_PROCFS_ANSWER:
 		mcctrl_wakeup_cb(__os, pisp);
 		break;
 
@@ -189,11 +192,6 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		mcexec_syscall(usrdata, pisp);
 		break;
 
-	case SCD_MSG_PROCFS_ANSWER:
-		procfs_answer(usrdata, pisp->pid);
-		break;
-
-
 	case SCD_MSG_SYSFS_REQ_CREATE:
 	case SCD_MSG_SYSFS_REQ_MKDIR:
 	case SCD_MSG_SYSFS_REQ_SYMLINK:
@@ -209,7 +207,8 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 
 	case SCD_MSG_PROCFS_TID_CREATE:
 	case SCD_MSG_PROCFS_TID_DELETE:
-		procfsm_packet_handler(__os, pisp->msg, pisp->pid, pisp->arg);
+		procfsm_packet_handler(__os, pisp->msg, pisp->pid, pisp->arg,
+				       pisp->resp_pa);
 		break;
 
 	case SCD_MSG_GET_VDSO_INFO:
@@ -225,6 +224,10 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		mcctrl_eventfd(__os, pisp);
 		break;
 
+	case SCD_MSG_FUTEX_WAKE:
+		mcctrl_futex_wake(pisp);
+		break;
+
 	default:
 		printk(KERN_ERR "mcctrl:syscall_packet_handler:"
 				"unknown message (%d.%d.%d.%d.%d.%#lx)\n",

executer/kernel/mcctrl/mcctrl.h

@@ -67,6 +67,7 @@
 #define	SCD_MSG_PROCFS_DELETE		0x11
 #define	SCD_MSG_PROCFS_REQUEST		0x12
 #define	SCD_MSG_PROCFS_ANSWER		0x13
+#define	SCD_MSG_PROCFS_RELEASE		0x15
 
 #define SCD_MSG_DEBUG_LOG		0x20
 
@@ -101,23 +102,18 @@
 #define SCD_MSG_CPU_RW_REG              0x52
 #define SCD_MSG_CPU_RW_REG_RESP         0x53
 
+#define SCD_MSG_FUTEX_WAKE              0x60
+
 #define DMA_PIN_SHIFT                   21
 
 #define DO_USER_MODE
 
 #define	__NR_coredump			999
 
-#ifdef POSTK_DEBUG_TEMP_FIX_61 /* Core table size and lseek return value to loff_t */
 struct coretable {
 	loff_t len;
 	unsigned long addr;
 };
-#else /* POSTK_DEBUG_TEMP_FIX_61 */
-struct coretable {
-	int len;
-	unsigned long addr;
-};
-#endif /* POSTK_DEBUG_TEMP_FIX_61 */
 
 enum mcctrl_os_cpu_operation {
 	MCCTRL_OS_CPU_READ_REGISTER,
@@ -125,6 +121,12 @@ enum mcctrl_os_cpu_operation {
 	MCCTRL_OS_CPU_MAX_OP
 };
 
+/* Used to wake-up a Linux thread futex_wait()-ing */
+struct uti_futex_resp {
+	int done;
+	wait_queue_head_t wq;
+};
+
 struct ikc_scd_packet {
 	int msg;
 	int err;
@@ -147,7 +149,7 @@ struct ikc_scd_packet {
 			long sysfs_arg3;
 		};
 
-		/* SCD_MSG_SCHEDULE_THREAD */
+		/* SCD_MSG_WAKE_UP_SYSCALL_THREAD */
 		struct {
 			int ttid;
 		};
@@ -163,6 +165,12 @@ struct ikc_scd_packet {
 		struct {
 			int eventfd_type;
 		};
+
+		/* SCD_MSG_FUTEX_WAKE */
+		struct {
+			void *resp;
+			int *spin_sleep; /* 1: waiting in linux_wait_event() 0: woken up by someone else */
+		} futex;
 	};
 	char padding[8];
 };
@@ -213,9 +221,12 @@ struct mcctrl_channel {
 };
 
 struct mcctrl_per_thread_data {
+	struct mcctrl_per_proc_data *ppd;
 	struct list_head hash;
 	struct task_struct *task;
 	void *data;
+	int tid; /* debug */
+	atomic_t refcount;
 };
 
 #define MCCTRL_PER_THREAD_DATA_HASH_SHIFT 8
@@ -315,6 +326,7 @@ struct mcctrl_part_exec {
 	struct mutex lock;	
 	int nr_processes;
 	int nr_processes_left;
+	int process_rank;
 	cpumask_t cpus_used;
 	struct list_head pli_list;
 };
@@ -400,10 +412,30 @@ int mcctrl_ikc_send_wait(ihk_os_t os, int cpu, struct ikc_scd_packet *pisp,
 
 ihk_os_t osnum_to_os(int n);
 
+/* look up symbols, plus arch-specific ones */
+extern int (*mcctrl_sys_mount)(char *dev_name, char *dir_name, char *type,
+			       unsigned long flags, void *data);
+extern int (*mcctrl_sys_umount)(char *dir_name, int flags);
+extern int (*mcctrl_sys_unshare)(unsigned long unshare_flags);
+extern long (*mcctrl_sched_setaffinity)(pid_t pid,
+					const struct cpumask *in_mask);
+extern int (*mcctrl_sched_setscheduler_nocheck)(struct task_struct *p,
+						int policy,
+						const struct sched_param *param);
+extern ssize_t (*mcctrl_sys_readlink)(const char *path, char *buf,
+				      size_t bufsiz);
+extern void (*mcctrl_zap_page_range)(struct vm_area_struct *vma,
+				     unsigned long start,
+				     unsigned long size,
+				     struct zap_details *details);
+extern struct inode_operations *mcctrl_hugetlbfs_inode_operations;
+
 /* syscall.c */
 void pager_add_process(void);
 void pager_remove_process(struct mcctrl_per_proc_data *ppd);
+void pager_cleanup(void);
 
+int __do_in_kernel_irq_syscall(ihk_os_t os, struct ikc_scd_packet *packet);
 int __do_in_kernel_syscall(ihk_os_t os, struct ikc_scd_packet *packet);
 int mcctrl_add_per_proc_data(struct mcctrl_usrdata *ud, int pid, 
 	struct mcctrl_per_proc_data *ppd);
@@ -412,20 +444,18 @@ struct mcctrl_per_proc_data *mcctrl_get_per_proc_data(
 		struct mcctrl_usrdata *ud, int pid);
 void mcctrl_put_per_proc_data(struct mcctrl_per_proc_data *ppd);
 
-int mcctrl_add_per_thread_data(struct mcctrl_per_proc_data* ppd,
+int mcctrl_add_per_thread_data(struct mcctrl_per_proc_data *ppd, void *data);
-	struct task_struct *task, void *data);
+void mcctrl_put_per_thread_data_unsafe(struct mcctrl_per_thread_data *ptd);
-int mcctrl_delete_per_thread_data(struct mcctrl_per_proc_data* ppd,
+void mcctrl_put_per_thread_data(struct mcctrl_per_thread_data* ptd);
-	struct task_struct *task);
 #ifdef POSTK_DEBUG_ARCH_DEP_56 /* Strange how to use inline declaration fix. */
-static inline struct mcctrl_per_thread_data *mcctrl_get_per_thread_data(
+inline struct mcctrl_per_thread_data *mcctrl_get_per_thread_data(struct mcctrl_per_proc_data *ppd, struct task_struct *task)
-	struct mcctrl_per_proc_data *ppd, struct task_struct *task)
 {
 	struct mcctrl_per_thread_data *ptd_iter, *ptd = NULL;
 	int hash = (((uint64_t)task >> 4) & MCCTRL_PER_THREAD_DATA_HASH_MASK);
 	unsigned long flags;
 
-	/* Check if data for this thread exists and return it */
+	/* Check if data for this thread exists */
-	read_lock_irqsave(&ppd->per_thread_data_hash_lock[hash], flags);
+	write_lock_irqsave(&ppd->per_thread_data_hash_lock[hash], flags);
 
 	list_for_each_entry(ptd_iter, &ppd->per_thread_data_hash[hash], hash) {
 		if (ptd_iter->task == task) {
@@ -434,16 +464,27 @@ static inline struct mcctrl_per_thread_data *mcctrl_get_per_thread_data(
 		}
 	}
 
-	read_unlock_irqrestore(&ppd->per_thread_data_hash_lock[hash], flags);
+	if (ptd) {
-	return ptd ? ptd->data : NULL;
+		if (atomic_read(&ptd->refcount) <= 0) {
+			printk("%s: ERROR: use-after-free detected (%d)", __FUNCTION__, atomic_read(&ptd->refcount));
+			ptd = NULL;
+			goto out;
+		}
+		atomic_inc(&ptd->refcount);
+	}
+
+ out:
+	write_unlock_irqrestore(&ppd->per_thread_data_hash_lock[hash], flags);
+	return ptd;
 }
 #else /* POSTK_DEBUG_ARCH_DEP_56 */
-inline struct mcctrl_per_thread_data *mcctrl_get_per_thread_data(
+inline struct mcctrl_per_thread_data *mcctrl_get_per_thread_data(struct mcctrl_per_proc_data *ppd, struct task_struct *task);
-	struct mcctrl_per_proc_data *ppd, struct task_struct *task);
 #endif /* POSTK_DEBUG_ARCH_DEP_56 */
+int mcctrl_clear_pte_range(uintptr_t start, uintptr_t len);
 
 void __return_syscall(ihk_os_t os, struct ikc_scd_packet *packet, 
 		long ret, int stid);
+int clear_pte_range(uintptr_t start, uintptr_t len);
 
 int mcctrl_os_alive(void);
 
@@ -455,7 +496,6 @@ struct procfs_read {
 	int count;		/* bytes to read (request) */
 	int eof;		/* if eof is detected, 1 otherwise 0. (answer)*/
 	int ret;		/* read bytes (answer) */
-	int status;		/* non-zero if done (answer) */
 	int newcpu;		/* migrated new cpu (answer) */
 	int readwrite;		/* 0:read, 1:write */
 	char fname[PROCFS_NAME_MAX];	/* procfs filename (request) */
@@ -468,7 +508,8 @@ struct procfs_file {
 };
 
 void procfs_answer(struct mcctrl_usrdata *ud, int pid);
-int procfsm_packet_handler(void *os, int msg, int pid, unsigned long arg);
+int procfsm_packet_handler(void *os, int msg, int pid, unsigned long arg,
+			   unsigned long resp_pa);
 void add_tid_entry(int osnum, int pid, int tid);
 void add_pid_entry(int osnum, int pid);
 void delete_tid_entry(int osnum, int pid, int tid);
@@ -504,7 +545,9 @@ struct vdso {
 
 int reserve_user_space(struct mcctrl_usrdata *usrdata, unsigned long *startp,
 		unsigned long *endp);
+int release_user_space(uintptr_t start, uintptr_t len);
 void get_vdso_info(ihk_os_t os, long vdso_pa);
+int arch_symbols_init(void);
 
 struct get_cpu_mapping_req {
 	int busy;		/* INOUT: */

executer/kernel/mcctrl/procfs.c

@@ -103,33 +103,6 @@ getpath(struct procfs_list_entry *e, char *buf, int bufsize)
 	}
 }
 
-/**
- * \brief Process SCD_MSG_PROCFS_ANSWER message.
- *
- * \param ud mcctrl_usrdata pointer
- * \param pid PID of the requesting process
- */
-void procfs_answer(struct mcctrl_usrdata *ud, int pid)
-{
-	struct mcctrl_per_proc_data *ppd = NULL;
-
-	if (pid > 0) {
-		ppd = mcctrl_get_per_proc_data(ud, pid);
-
-		if (unlikely(!ppd)) {
-			kprintf("%s: ERROR: no per-process structure for PID %d\n",
-					__FUNCTION__, pid);
-			return;
-		}
-	}
-
-	wake_up_all(pid > 0 ? &ppd->wq_procfs : &ud->wq_procfs);
-
-	if (pid > 0) {
-		mcctrl_put_per_proc_data(ppd);
-	}
-}
-
 static struct procfs_list_entry *
 find_procfs_entry(struct procfs_list_entry *parent, const char *name)
 {
@@ -321,6 +294,8 @@ get_base_entry(int osnum)
 	if(!e){
 		e = add_procfs_entry(NULL, name, S_IFDIR | 0555,
 		                     uid, gid, NULL);
+		if (!e)
+			return NULL;
 		e->osnum = osnum;
 	}
 	return e;
@@ -456,6 +431,8 @@ proc_exe_link(int osnum, int pid, const char *path)
 
 		e = add_procfs_entry(parent, "exe", S_IFLNK | 0777, uid, gid,
 		                     path);
+		if (!e)
+			goto out;
 		e->data = kmalloc(strlen(path) + 1, GFP_KERNEL);
 		strcpy(e->data, path);
 		task = find_procfs_entry(parent, "task");
@@ -464,6 +441,7 @@ proc_exe_link(int osnum, int pid, const char *path)
 			                 uid, gid, path);
 		}
 	}
+out:
 	up(&procfs_file_list_lock);
 }
 
@@ -509,7 +487,6 @@ procfs_exit(int osnum)
  * This function conforms to the 2) way of fs/proc/generic.c
  * from linux-2.6.39.4.
  */
-#ifdef POSTK_DEBUG_TEMP_FIX_43 /* Fixed an issue that failed pread / pwrite of size larger than 4MB */
 static ssize_t __mckernel_procfs_read_write(
 		struct file *file,
 		char __user *buf, size_t nbytes,
@@ -520,7 +497,7 @@ static ssize_t __mckernel_procfs_read_write(
 	int order = 0;
 	volatile struct procfs_read *r = NULL;
 	struct ikc_scd_packet isp;
-	int ret, osnum, pid, retw;
+	int ret, osnum, pid;
 	unsigned long pbuf;
 	size_t count = nbytes;
 	size_t copy_size = 0;
@@ -615,11 +592,11 @@ static ssize_t __mckernel_procfs_read_write(
 
 	while (count > 0) {
 		int this_len = min_t(ssize_t, count, copy_size);
+		int do_free;
 
 		r->pbuf = pbuf;
 		r->eof = 0;
 		r->ret = -EIO; /* default */
-		r->status = 0;
 		r->offset = offset;
 		r->count = this_len;
 		r->readwrite = read_write;
@@ -629,50 +606,26 @@ static ssize_t __mckernel_procfs_read_write(
 		isp.arg = virt_to_phys(r);
 		isp.pid = pid;
 
-		ret = mcctrl_ikc_send(osnum_to_os(e->osnum),
+		ret = mcctrl_ikc_send_wait(osnum_to_os(e->osnum),
-				(pid > 0) ? ppd->ikc_target_cpu : 0, &isp);
+					   (pid > 0) ? ppd->ikc_target_cpu : 0,
+					   &isp, HZ, NULL, &do_free, 1, r);
+
+		if (!do_free && ret >= 0) {
+			ret = -EIO;
+		}
 
 		if (ret < 0) {
-			goto out; /* error */
+			if (ret == -ETIME) {
-		}
+				pr_info("%s: error: timeout (1 sec)\n",
-
+				       __func__);
-		/* Wait for a reply. */
+			}
-		ret = -EIO; /* default exit code */
+			else if (ret == -ERESTARTSYS) {
-		dprintk("%s: waiting for reply\n", __FUNCTION__);
+				ret = -ERESTART;
-
+			}
-retry_wait:
+			if (!do_free)
-		/* Wait for the status field of the procfs_read structure,
+				r = NULL;
-		 * wait on per-process or OS specific data depending on
-		 * who the request is for.
-		 */
-		if (pid > 0) {
-			retw = wait_event_interruptible_timeout(ppd->wq_procfs,
-					r->status != 0, HZ);
-		}
-		else {
-			retw = wait_event_interruptible_timeout(udp->wq_procfs,
-					r->status != 0, HZ);
-		}
-
-		/* Timeout? */
-		if (retw == 0 && r->status == 0) {
-			printk("%s: error: timeout (1 sec)\n", __FUNCTION__);
 			goto out;
 		}
-		/* Interrupted? */
-		else if (retw == -ERESTARTSYS) {
-			ret = -ERESTART;
-			goto out;
-		}
-		/* Were we woken up by a reply to another procfs request? */
-		else if (r->status == 0) {
-			/* TODO: r->status is not set atomically, we could be woken
-			 * up with status == 0 and it could change to 1 while in this
-			 * code, we could potentially miss the wake_up()... 
-			 */
-			printk("%s: stale wake-up, retrying\n", __FUNCTION__);
-			goto retry_wait;
-		}
 
 		/* Wake up and check the result. */
 		dprintk("%s: woke up. ret: %d, eof: %d\n",
@@ -717,193 +670,6 @@ out:
 
 	return ret;
 }
-#else /* POSTK_DEBUG_TEMP_FIX_43 */
-static ssize_t __mckernel_procfs_read_write(
-		struct file *file,
-		char __user *buf, size_t nbytes,
-		loff_t *ppos, int read_write)
-{
-	struct inode * inode = file->f_inode;
-	char *kern_buffer = NULL;
-	int order = 0;
-	volatile struct procfs_read *r = NULL;
-	struct ikc_scd_packet isp;
-	int ret, osnum, pid, retw;
-	unsigned long pbuf;
-	unsigned long count = nbytes;
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3,10,0)
-	struct proc_dir_entry *dp = PDE(inode);
-	struct procfs_list_entry *e = dp->data;
-#else
-	struct procfs_list_entry *e = PDE_DATA(inode);
-#endif
-	loff_t offset = *ppos;
-	char pathbuf[PROCFS_NAME_MAX];
-	char *path, *p;
-	ihk_os_t os = NULL;
-	struct mcctrl_usrdata *udp = NULL;
-	struct mcctrl_per_proc_data *ppd = NULL;
-
-	if (count <= 0 || offset < 0) {
-		return 0;
-	}
-
-	path = getpath(e, pathbuf, PROCFS_NAME_MAX);
-	dprintk("%s: invoked for %s, offset: %lu, count: %lu\n",
-			__FUNCTION__, path,
-			(unsigned long)offset, count);
-
-	/* Verify OS number */
-	ret = sscanf(path, "mcos%d/", &osnum);
-	if (ret != 1) {
-		printk("%s: error: couldn't determine OS number\n", __FUNCTION__);
-		return -EINVAL;
-	}
-
-	if (osnum != e->osnum) {
-		printk("%s: error: OS numbers don't match\n", __FUNCTION__);
-		return -EINVAL;
-	}
-
-	/* Is this request for a specific process? */
-	p = strchr(path, '/') + 1;
-	ret = sscanf(p, "%d/", &pid);
-	if (ret != 1) {
-		pid = -1;
-	}
-
-	os = osnum_to_os(osnum);
-	if (!os) {
-		printk("%s: error: no IHK OS data found for OS %d\n",
-				__FUNCTION__, osnum);
-		return -EINVAL;
-	}
-
-	udp = ihk_host_os_get_usrdata(os);
-	if (!udp) {
-		printk("%s: error: no MCCTRL data found for OS %d\n",
-				__FUNCTION__, osnum);
-		return -EINVAL;
-	}
-
-	if (pid > 0) {
-		ppd = mcctrl_get_per_proc_data(udp, pid);
-
-		if (unlikely(!ppd)) {
-			printk("%s: error: no per-process structure for PID %d",
-					__FUNCTION__, pid);
-			return -EINVAL;
-		}
-	}
-
-	while ((1 << order) < count) ++order;
-	if (order > 12) {
-		order -= 12;
-	}
-	else {
-		order = 1;
-	}
-
-	/* NOTE: we need physically contigous memory to pass through IKC */
-	kern_buffer = (char *)__get_free_pages(GFP_KERNEL, order);
-	if (!kern_buffer) {
-		printk("%s: ERROR: allocating kernel buffer\n", __FUNCTION__);
-		ret = -ENOMEM;
-		goto out;
-	}
-
-	pbuf = virt_to_phys(kern_buffer);
-
-	r = kmalloc(sizeof(struct procfs_read), GFP_KERNEL);
-	if (r == NULL) {
-		ret = -ENOMEM;
-		goto out;
-	}
-	r->pbuf = pbuf;
-	r->eof = 0;
-	r->ret = -EIO; /* default */
-	r->status = 0;
-	r->offset = offset;
-	r->count = count;
-	r->readwrite = read_write;
-	strncpy((char *)r->fname, path, PROCFS_NAME_MAX);
-	isp.msg = SCD_MSG_PROCFS_REQUEST;
-	isp.ref = 0;
-	isp.arg = virt_to_phys(r);
-	isp.pid = pid;
-
-	ret = mcctrl_ikc_send(osnum_to_os(e->osnum),
-			(pid > 0) ? ppd->ikc_target_cpu : 0, &isp);
-
-	if (ret < 0) {
-		goto out; /* error */
-	}
-
-	/* Wait for a reply. */
-	ret = -EIO; /* default exit code */
-	dprintk("%s: waiting for reply\n", __FUNCTION__);
-
-retry_wait:
-	/* Wait for the status field of the procfs_read structure,
-	 * wait on per-process or OS specific data depending on
-	 * who the request is for.
-	 */
-	if (pid > 0) {
-		retw = wait_event_interruptible_timeout(ppd->wq_procfs,
-				r->status != 0, 5 * HZ);
-	}
-	else {
-		retw = wait_event_interruptible_timeout(udp->wq_procfs,
-				r->status != 0, 5 * HZ);
-	}
-
-	/* Timeout? */
-	if (retw == 0 && r->status == 0) {
-		printk("%s: error: timeout (1 sec)\n", __FUNCTION__);
-		goto out;
-	}
-	/* Interrupted? */
-	else if (retw == -ERESTARTSYS) {
-		ret = -ERESTART;
-		goto out;
-	}
-	/* Were we woken up by a reply to another procfs request? */
-	else if (r->status == 0) {
-		/* TODO: r->status is not set atomically, we could be woken
-		 * up with status == 0 and it could change to 1 while in this
-		 * code, we could potentially miss the wake_up()... 
-		 */
-		printk("%s: stale wake-up, retrying\n", __FUNCTION__);
-		goto retry_wait;
-	}
-
-	/* Wake up and check the result. */
-	dprintk("%s: woke up. ret: %d, eof: %d\n",
-			__FUNCTION__, r->ret, r->eof);
-
-	if (r->ret > 0) {
-		if (read_write == 0) {
-			if (copy_to_user(buf, kern_buffer, r->ret)) {
-				printk("%s: ERROR: copy_to_user failed.\n", __FUNCTION__);
-				ret = -EFAULT;
-				goto out;
-			}
-		}
-		*ppos += r->ret;
-	}
-	ret = r->ret;
-
-out:
-	if (ppd)
-		mcctrl_put_per_proc_data(ppd);
-	if (kern_buffer)
-		free_pages((uintptr_t)kern_buffer, order);
-	if (r)
-		kfree((void *)r);
-
-	return ret;
-}
-#endif /* POSTK_DEBUG_TEMP_FIX_43 */
 
 static ssize_t mckernel_procfs_read(struct file *file,
 		char __user *buf, size_t nbytes, loff_t *ppos)
@@ -939,33 +705,48 @@ struct procfs_work {
 	int msg;
 	int pid;
 	unsigned long arg;
+	unsigned long resp_pa;
 	struct work_struct work;
 };
 
 static void procfsm_work_main(struct work_struct *work0)
 {
 	struct procfs_work *work = container_of(work0, struct procfs_work, work);
+	unsigned long phys;
+	int *done;
 
 	switch (work->msg) {
-		case SCD_MSG_PROCFS_TID_CREATE:
+	case SCD_MSG_PROCFS_TID_CREATE:
-			add_tid_entry(ihk_host_os_get_index(work->os), work->pid, work->arg);
+		add_tid_entry(ihk_host_os_get_index(work->os),
-			break;
+				work->pid, work->arg);
+		phys = ihk_device_map_memory(ihk_os_to_dev(work->os),
+					     work->resp_pa, sizeof(int));
+		done = ihk_device_map_virtual(ihk_os_to_dev(work->os),
+					      phys, sizeof(int), NULL, 0);
+		*done = 1;
+		ihk_device_unmap_virtual(ihk_os_to_dev(work->os),
+						 done, sizeof(int));
+		ihk_device_unmap_memory(ihk_os_to_dev(work->os),
+					phys, sizeof(int));
+		break;
 
-		case SCD_MSG_PROCFS_TID_DELETE:
+	case SCD_MSG_PROCFS_TID_DELETE:
-			delete_tid_entry(ihk_host_os_get_index(work->os), work->pid, work->arg);
+		delete_tid_entry(ihk_host_os_get_index(work->os),
-			break;
+				 work->pid, work->arg);
+		break;
 
-		default:
+	default:
-			printk("%s: unknown work: msg: %d, pid: %d, arg: %lu)\n",
+		pr_warn("%s: unknown work: msg: %d, pid: %d, arg: %lu)\n",
-					__FUNCTION__, work->msg, work->pid, work->arg);
+			__func__, work->msg, work->pid, work->arg);
-			break;
+		break;
 	}
 
 	kfree(work);
 	return;
 }
 
-int procfsm_packet_handler(void *os, int msg, int pid, unsigned long arg)
+int procfsm_packet_handler(void *os, int msg, int pid, unsigned long arg,
+			   unsigned long resp_pa)
 {
 	struct procfs_work *work = NULL;
 
@@ -979,6 +760,7 @@ int procfsm_packet_handler(void *os, int msg, int pid, unsigned long arg)
 	work->msg = msg;
 	work->pid = pid;
 	work->arg = arg;
+	work->resp_pa = resp_pa;
 	INIT_WORK(&work->work, &procfsm_work_main);
 
 	schedule_work(&work->work);
@@ -997,6 +779,303 @@ static const struct file_operations mckernel_forward = {
 	.write		= mckernel_procfs_write,
 };
 
+#define PA_NULL (-1L)
+
+struct mckernel_procfs_buffer_info {
+	unsigned long top_pa;
+	unsigned long cur_pa;
+	ihk_os_t os;
+	int pid;
+	char path[0];
+};
+
+struct mckernel_procfs_buffer {
+	unsigned long next_pa;
+	unsigned long pos;
+	unsigned long size;
+	char buf[0];
+};
+
+static int mckernel_procfs_buff_open(struct inode *inode, struct file *file)
+{
+	struct mckernel_procfs_buffer_info *info;
+	int pid;
+	int ret;
+	char *path;
+	char *path_buf;
+	char *p;
+	ihk_os_t os;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 10, 0)
+	struct proc_dir_entry *dp = PDE(inode);
+	struct procfs_list_entry *e = dp->data;
+#else
+	struct procfs_list_entry *e = PDE_DATA(inode);
+#endif
+
+	os = osnum_to_os(e->osnum);
+	if (!os) {
+		return -EINVAL;
+	}
+	path_buf = kmalloc(PROCFS_NAME_MAX, GFP_KERNEL);
+	if (!path_buf) {
+		return -ENOMEM;
+	}
+	path = getpath(e, path_buf, PROCFS_NAME_MAX);
+	p = strchr(path, '/') + 1;
+	ret = sscanf(p, "%d/", &pid);
+	if (ret != 1) {
+		pid = -1;
+	}
+
+	info = kmalloc(sizeof(struct mckernel_procfs_buffer_info) +
+		       strlen(path) + 1, GFP_KERNEL);
+	if (!info) {
+		kfree(path_buf);
+		return -ENOMEM;
+	}
+	info->top_pa = PA_NULL;
+	info->cur_pa = PA_NULL;
+	info->os = os;
+	info->pid = pid;
+	strcpy(info->path, path);
+	file->private_data = info;
+
+	kfree(path_buf);
+	return 0;
+}
+
+static int mckernel_procfs_buff_release(struct inode *inode, struct file *file)
+{
+	struct mckernel_procfs_buffer_info *info = file->private_data;
+	int rc = 0;
+
+	if (!info) {
+		return -EIO;
+	}
+
+	file->private_data = NULL;
+	if (info->top_pa != PA_NULL) {
+		int ret;
+		struct procfs_read *r = NULL;
+		struct ikc_scd_packet isp;
+		int do_free;
+
+		r = kmalloc(sizeof(struct procfs_read), GFP_KERNEL);
+		if (r == NULL) {
+			rc = -ENOMEM;
+			goto out;
+		}
+		memset(r, '\0', sizeof(struct procfs_read));
+		r->pbuf = info->top_pa;
+		r->ret = -EIO; /* default */
+		r->fname[0] = '\0';
+		isp.msg = SCD_MSG_PROCFS_RELEASE;
+		isp.ref = 0;
+		isp.arg = virt_to_phys(r);
+		isp.pid = 0;
+
+		rc = -EIO;
+		ret = mcctrl_ikc_send_wait(info->os, 0,
+					   &isp, 5 * HZ, NULL, &do_free, 1, r);
+
+		if (!do_free && ret >= 0) {
+			ret = -EIO;
+		}
+
+		if (ret < 0) {
+			rc = ret;
+			if (ret == -ETIME) {
+				pr_info("%s: error: timeout (1 sec)\n",
+				       __func__);
+			}
+			else if (ret == -ERESTARTSYS) {
+				rc = -ERESTART;
+			}
+			if (!do_free)
+				r = NULL;
+			goto out;
+		}
+
+		if (r->ret < 0) {
+			rc = r->ret;
+			goto out;
+		}
+		rc = 0;
+out:
+		if (r)
+			kfree((void *)r);
+	}
+	kfree(info);
+	return rc;
+}
+
+static ssize_t mckernel_procfs_buff_read(struct file *file, char __user *ubuf,
+					 size_t nbytes, loff_t *ppos)
+{
+	struct mckernel_procfs_buffer_info *info = file->private_data;
+	unsigned long phys;
+	struct mckernel_procfs_buffer *buf;
+	int pos = *ppos;
+	ssize_t l = 0;
+	int done = 0;
+	ihk_os_t os;
+
+	if (nbytes <= 0 || *ppos < 0) {
+		return 0;
+	}
+
+	if (!info) {
+		return -EIO;
+	}
+
+	os = info->os;
+	if (info->top_pa == PA_NULL) {
+		int ret;
+		int pid = info->pid;
+		struct procfs_read *r = NULL;
+		struct ikc_scd_packet isp;
+		struct mcctrl_usrdata *udp = NULL;
+		struct mcctrl_per_proc_data *ppd = NULL;
+		int do_free;
+
+		udp = ihk_host_os_get_usrdata(os);
+		if (!udp) {
+			pr_err("%s: no MCCTRL data found for OS\n",
+					__func__);
+			return -EINVAL;
+		}
+
+		if (pid > 0) {
+			ppd = mcctrl_get_per_proc_data(udp, pid);
+
+			if (unlikely(!ppd)) {
+				pr_err("%s: no per-process structure for PID %d",
+						__func__, pid);
+				return -EINVAL;
+			}
+		}
+
+		r = kmalloc(sizeof(struct procfs_read), GFP_KERNEL);
+		if (r == NULL) {
+			l = -ENOMEM;
+			done = 1;
+			goto out;
+		}
+		memset(r, '\0', sizeof(struct procfs_read));
+		r->pbuf = PA_NULL;
+		r->ret = -EIO; /* default */
+		strncpy((char *)r->fname, info->path, PROCFS_NAME_MAX);
+		isp.msg = SCD_MSG_PROCFS_REQUEST;
+		isp.ref = 0;
+		isp.arg = virt_to_phys(r);
+		isp.pid = pid;
+
+		l = -EIO;
+		done = 1;
+		ret = mcctrl_ikc_send_wait(os,
+					   (pid > 0) ? ppd->ikc_target_cpu : 0,
+					   &isp, 5 * HZ, NULL, &do_free, 1, r);
+
+		if (!do_free && ret >= 0) {
+			ret = -EIO;
+		}
+
+		if (ret < 0) {
+			l = ret;
+			if (ret == -ETIME) {
+				pr_info("%s: error: timeout (1 sec)\n",
+				       __func__);
+			}
+			else if (ret == -ERESTARTSYS) {
+				l = -ERESTART;
+			}
+			if (!do_free)
+				r = NULL;
+			goto out;
+		}
+
+		if (r->ret < 0) {
+			l = r->ret;
+			goto out;
+		}
+
+		done = 0;
+		l = 0;
+		info->top_pa = info->cur_pa = r->pbuf;
+
+out:
+		if (ppd)
+			mcctrl_put_per_proc_data(ppd);
+		if (r)
+			kfree((void *)r);
+	}
+
+	if (info->cur_pa == PA_NULL) {
+		info->cur_pa = info->top_pa;
+	}
+
+	while (!done && info->cur_pa != PA_NULL) {
+		long bpos;
+		long bsize;
+
+		phys = ihk_device_map_memory(ihk_os_to_dev(os), info->cur_pa,
+					     PAGE_SIZE);
+#ifdef CONFIG_MIC
+		buf = ioremap_wc(phys, PAGE_SIZE);
+#else
+		buf = ihk_device_map_virtual(ihk_os_to_dev(os), phys,
+					     PAGE_SIZE, NULL, 0);
+#endif
+
+		if (pos < buf->pos) {
+			info->cur_pa = info->top_pa;
+			goto rep;
+		}
+
+		if (pos >= buf->pos + buf->size) {
+			info->cur_pa = buf->next_pa;
+			goto rep;
+		}
+
+		bpos = pos - buf->pos;
+		bsize = (buf->pos + buf->size) - pos;
+		if (bsize > (nbytes - l)) {
+			bsize = nbytes - l;
+		}
+		if (copy_to_user(ubuf, buf->buf + bpos, bsize)) {
+			done = 1;
+			pos = *ppos;
+			l = -EFAULT;
+		}
+		else {
+			ubuf += bsize;
+			pos += bsize;
+			l += bsize;
+			if (l == nbytes) {
+				done = 1;
+			}
+		}
+rep:
+#ifdef CONFIG_MIC
+		iounmap(buf);
+#else
+		ihk_device_unmap_virtual(ihk_os_to_dev(os), buf, PAGE_SIZE);
+#endif
+		ihk_device_unmap_memory(ihk_os_to_dev(os), phys, PAGE_SIZE);
+	};
+
+	*ppos = pos;
+	return l;
+}
+
+static const struct file_operations mckernel_buff_io = {
+	.llseek		= mckernel_procfs_lseek,
+	.read		= mckernel_procfs_buff_read,
+	.write		= NULL,
+	.open		= mckernel_procfs_buff_open,
+	.release	= mckernel_procfs_buff_release,
+};
+
 static const struct procfs_entry tid_entry_stuff[] = {
 //	PROC_REG("auxv",       S_IRUSR, NULL),
 //	PROC_REG("clear_refs", S_IWUSR, NULL),
@@ -1006,10 +1085,10 @@ static const struct procfs_entry tid_entry_stuff[] = {
 //	PROC_LNK("exe",        mckernel_readlink),
 //	PROC_REG("limits",     S_IRUSR|S_IWUSR, NULL),
 //	PROC_REG("maps",       S_IRUGO, NULL),
-	PROC_REG("mem",        S_IRUSR|S_IWUSR, NULL),
+	PROC_REG("mem",        0600, NULL),
 //	PROC_REG("pagemap",    S_IRUGO, NULL),
 //	PROC_REG("smaps",      S_IRUGO, NULL),
-	PROC_REG("stat",       S_IRUGO, NULL),
+	PROC_REG("stat",       0444, &mckernel_buff_io),
 //	PROC_REG("statm",      S_IRUGO, NULL),
 //	PROC_REG("status",     S_IRUGO, NULL),
 //	PROC_REG("syscall",    S_IRUGO, NULL),
@@ -1018,26 +1097,26 @@ static const struct procfs_entry tid_entry_stuff[] = {
 };
 
 static const struct procfs_entry pid_entry_stuff[] = {
-	PROC_REG("auxv",       S_IRUSR, NULL),
+	PROC_REG("auxv",       0400, &mckernel_buff_io),
 	/* Support the case where McKernel process retrieves its job-id under the Fujitsu TCS suite. */
 //	PROC_REG("cgroup",     S_IXUSR, NULL),
 //	PROC_REG("clear_refs", S_IWUSR, NULL),
-	PROC_REG("cmdline",    S_IRUGO, NULL),
+	PROC_REG("cmdline",    0444, &mckernel_buff_io),
-//	PROC_REG("comm",       S_IRUGO|S_IWUSR, NULL),
+	PROC_REG("comm",       0644, &mckernel_buff_io),
 //	PROC_REG("coredump_filter", S_IRUGO|S_IWUSR, NULL),
-	PROC_REG("cpuset",     S_IXUSR, NULL),
+//	PROC_REG("cpuset",     S_IRUGO, NULL),
 //	PROC_REG("environ",    S_IRUSR, NULL),
 //	PROC_LNK("exe",        mckernel_readlink),
 //	PROC_REG("limits",     S_IRUSR|S_IWUSR, NULL),
-	PROC_REG("maps",       S_IRUGO, NULL),
+	PROC_REG("maps",       0444, &mckernel_buff_io),
-	PROC_REG("mem",        S_IRUSR|S_IWUSR, NULL),
+	PROC_REG("mem",        0400, NULL),
-	PROC_REG("pagemap",    S_IRUGO, NULL),
+	PROC_REG("pagemap",    0444, NULL),
-	PROC_REG("smaps",      S_IRUGO, NULL),
+//	PROC_REG("smaps",      S_IRUGO, NULL),
-//	PROC_REG("stat",       S_IRUGO, NULL),
+//	PROC_REG("stat",       0444, &mckernel_buff_io),
 //	PROC_REG("statm",      S_IRUGO, NULL),
-	PROC_REG("status",     S_IRUGO, NULL),
+	PROC_REG("status",     0444, &mckernel_buff_io),
 //	PROC_REG("syscall",    S_IRUGO, NULL),
-	PROC_DIR("task",       S_IRUGO|S_IXUGO),
+	PROC_DIR("task",       0555),
 //	PROC_REG("wchan",      S_IRUGO, NULL),
 	PROC_TERM
 };
@@ -1045,14 +1124,14 @@ static const struct procfs_entry pid_entry_stuff[] = {
 static const struct procfs_entry base_entry_stuff[] = {
 //	PROC_REG("cmdline",    S_IRUGO, NULL),
 #ifdef POSTK_DEBUG_ARCH_DEP_42 /* /proc/cpuinfo support added. */
-	PROC_REG("cpuinfo",    S_IRUGO, NULL),
+	PROC_REG("cpuinfo",    0444, &mckernel_buff_io),
 #else /* POSTK_DEBUG_ARCH_DEP_42 */
 //	PROC_REG("cpuinfo",    S_IRUGO, NULL),
 #endif /* POSTK_DEBUG_ARCH_DEP_42 */
 //	PROC_REG("meminfo",    S_IRUGO, NULL),
 //	PROC_REG("pagetypeinfo",S_IRUGO, NULL),
 //	PROC_REG("softirq",    S_IRUGO, NULL),
-	PROC_REG("stat",       S_IRUGO, NULL),
+	PROC_REG("stat",       0444, &mckernel_buff_io),
 //	PROC_REG("uptime",     S_IRUGO, NULL),
 //	PROC_REG("version",    S_IRUGO, NULL),
 //	PROC_REG("vmallocinfo",S_IRUSR, NULL),

executer/kernel/mcctrl/sysfs_files.c

@@ -790,6 +790,7 @@ out:
 	return error;
 } /* setup_node_files() */
 
+#ifdef SETUP_PCI_FILES
 static int read_file(void *buf, size_t size, char *fmt, va_list ap)
 {
 	int error;
@@ -798,7 +799,6 @@ static int read_file(void *buf, size_t size, char *fmt, va_list ap)
 	int n;
 	struct file *fp = NULL;
 	loff_t off;
-	mm_segment_t ofs;
 	ssize_t ss;
 
 	dprintk("read_file(%p,%ld,%s,%p)\n", buf, size, fmt, ap);
@@ -824,13 +824,14 @@ static int read_file(void *buf, size_t size, char *fmt, va_list ap)
 	}
 
 	off = 0;
-	ofs = get_fs();
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 14, 0)
-	set_fs(KERNEL_DS);
+	ss = kernel_read(fp, buf, size, &off);
-	ss = vfs_read(fp, buf, size, &off);
+#else
-	set_fs(ofs);
+	ss = kernel_read(fp, off, buf, size);
+#endif
 	if (ss < 0) {
 		error = ss;
-		eprintk("mcctrl:read_file:vfs_read failed. %d\n", error);
+		eprintk("mcctrl:read_file:kernel_read failed. %d\n", error);
 		goto out;
 	}
 	if (ss >= size) {
@@ -892,16 +893,6 @@ out:
 	return error;
 } /* read_long() */
 
-#ifdef MCCTRL_KSYM_sys_readlink
-static ssize_t (*mcctrl_sys_readlink)(const char *path, char *buf,
-		size_t bufsiz)
-#if MCCTRL_KSYM_sys_readlink
-	= (void *)MCCTRL_KSYM_sys_readlink;
-#else
-	= &sys_readlink;
-#endif
-#endif
-
 static int read_link(char *buf, size_t bufsize, char *fmt, ...)
 {
 	int error;
@@ -951,30 +942,14 @@ out:
 	return error;
 } /* read_link() */
 
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 static int setup_one_pci(struct mcctrl_usrdata *udp, const char *name)
 {
-#else /* POSTK_DEBUG_TEMP_FIX_22 */
-static int setup_one_pci(void *arg0, const char *name, int namlen,
-		loff_t offset, u64 ino, unsigned d_type)
-{
-	struct mcctrl_usrdata *udp = arg0;
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 	int error;
 	char *buf = NULL;
 	long node;
 	struct sysfsm_bitmap_param param;
 
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 	dprintk("setup_one_pci(%p,%s)\n", udp, name);
-#else /* POSTK_DEBUG_TEMP_FIX_22 */
-	dprintk("setup_one_pci(%p,%s,%d,%#lx,%#lx,%d)\n",
-			arg0, name, namlen, (long)offset, (long)ino, d_type);
-	if (namlen != 12) {
-		error = 0;
-		goto out;
-	}
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 
 	buf = (void *)__get_free_pages(GFP_KERNEL, 0);
 	if (!buf) {
@@ -1026,26 +1001,39 @@ static int setup_one_pci(void *arg0, const char *name, int namlen,
 	error = 0;
 out:
 	free_pages((long)buf, 0);
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 	dprintk("setup_one_pci(%p,%s): %d\n", udp, name, error);
-#else /* POSTK_DEBUG_TEMP_FIX_22 */
-	dprintk("setup_one_pci(%p,%s,%d,%#lx,%#lx,%d): %d\n",
-			arg0, name, namlen, (long)offset, (long)ino, d_type,
-			error);
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 	return error;
 } /* setup_one_pci() */
 
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 LIST_HEAD(pci_file_name_list);
 struct pci_file_name {
 	char *name;
 	struct list_head chain;
 };
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0) || \
+	(defined(RHEL_RELEASE_CODE) && RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 5))
+struct mcctrl_filler_args {
+	struct dir_context ctx;
+	void *buf;
+};
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 19, 0)
+static int pci_file_name_gen(struct dir_context *ctx, const char *name,
+		int namlen, loff_t offset, u64 ino, unsigned int d_type)
+#else
+static int pci_file_name_gen(void *ctx, const char *name,
+		int namlen, loff_t offset, u64 ino, unsigned int d_type)
+#endif
+{
+	struct mcctrl_filler_args *args
+		= container_of(ctx, struct mcctrl_filler_args, ctx);
+	void *buf = args->buf;
+#else
 static int pci_file_name_gen(void *buf, const char *name, int namlen,
 		loff_t offset, u64 ino, unsigned d_type)
 {
+#endif
 	struct pci_file_name *p;
 	int error = -1;
 
@@ -1083,56 +1071,31 @@ out:
 			buf, name, namlen, (long)offset, (long)ino, d_type, error);
 	return error;
 }
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,11,0)
+static inline int mcctrl_vfs_readdir(struct file *file, filldir_t filler,
-typedef int (*mcctrl_filldir_t)(void *buf, const char *name, int namlen,
+				     void *buf)
-		loff_t offset, u64 ino, unsigned d_type);
-
-struct mcctrl_filler_args {
-	struct dir_context ctx;
-	mcctrl_filldir_t filler;
-	void *buf;
-};
-
-static int mcctrl_filler(struct dir_context *ctx, const char *name,
-		int namlen, loff_t offset, u64 ino, unsigned d_type)
-{
-	struct mcctrl_filler_args *args
-		= container_of(ctx, struct mcctrl_filler_args, ctx);
-
-	return (*args->filler)(args->buf, name, namlen, offset, ino, d_type);
-} /* mcctrl_filler() */
-
-static inline int mcctrl_vfs_readdir(struct file *file,
-		mcctrl_filldir_t filler, void *buf)
 {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0) || \
+	(defined(RHEL_RELEASE_CODE) && RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 5))
 	struct mcctrl_filler_args args = {
-		.ctx.actor = &mcctrl_filler,
+		.ctx.actor = filler,
-		.filler = (void *)filler,
 		.buf = buf,
 	};
 
 	return iterate_dir(file, &args.ctx);
-} /* mcctrl_vfs_readdir() */
 #else
-static inline int mcctrl_vfs_readdir(struct file *file, filldir_t filler,
-		void *buf)
-{
 	return vfs_readdir(file, filler, buf);
-} /* mcctrl_vfs_readdir() */
 #endif
+} /* mcctrl_vfs_readdir() */
 
 static int setup_pci_files(struct mcctrl_usrdata *udp)
 {
 	int error;
 	int er;
 	struct file *fp = NULL;
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 	int ret = 0;
 	struct pci_file_name *cur;
 	struct pci_file_name *next;
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 
 	dprintk("setup_pci_files(%p)\n", udp);
 	fp = filp_open("/sys/bus/pci/devices", O_DIRECTORY, 0);
@@ -1142,18 +1105,13 @@ static int setup_pci_files(struct mcctrl_usrdata *udp)
 		goto out;
 	}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 	error = mcctrl_vfs_readdir(fp, &pci_file_name_gen, udp);
-#else /* POSTK_DEBUG_TEMP_FIX_22 */
-	error = mcctrl_vfs_readdir(fp, &setup_one_pci, udp);
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 	if (error) {
 		eprintk("mcctrl:setup_pci_files:"
 				"mcctrl_vfs_readdir failed. %d\n", error);
 		goto out;
 	}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_22 /* iterate_dir() deadlock */
 	list_for_each_entry_safe(cur, next, &pci_file_name_list, chain) {
 		if (!ret) {
 			ret = setup_one_pci(udp, cur->name);
@@ -1162,7 +1120,6 @@ static int setup_pci_files(struct mcctrl_usrdata *udp)
 		kfree(cur->name);
 		kfree(cur);
 	}
-#endif /* POSTK_DEBUG_TEMP_FIX_22 */
 
 	error = 0;
 out:
@@ -1176,6 +1133,7 @@ out:
 	dprintk("setup_pci_files(%p): %d\n", udp, error);
 	return error;
 } /* setup_pci_files() */
+#endif // SETUP_PCI_FILES
 
 void setup_sysfs_files(ihk_os_t os)
 {
@@ -1215,7 +1173,9 @@ void setup_sysfs_files(ihk_os_t os)
 	setup_cpus_sysfs_files(udp);
 	setup_node_files(udp);
 	setup_cpus_sysfs_files_node_link(udp);
-	//setup_pci_files(udp);
+#ifdef SETUP_PCI_FILES
+	setup_pci_files(udp);
+#endif
 
 	/* Indicate sysfs files setup completion for boot script */
 	error = sysfsm_mkdirf(os, NULL, "/sys/setup_complete");

executer/kernel/mcoverlayfs/Makefile.in

@@ -21,7 +21,7 @@ endif
 endif
 ifeq ($(BUILD_MODULE_TMP),rhel)
 ifeq ($(BUILD_MODULE),none)
-BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -eq 199168 -a ${RHEL_RELEASE} -ge 327 -a ${RHEL_RELEASE} -le 693 ]; then echo "linux-3.10.0-327.36.1.el7"; else echo "none"; fi)
+BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -eq 199168 -a ${RHEL_RELEASE} -ge 327 -a ${RHEL_RELEASE} -le 862 ]; then echo "linux-3.10.0-327.36.1.el7"; else echo "none"; fi)
 endif
 ifeq ($(BUILD_MODULE),none)
 BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -ge 262144 -a ${LINUX_VERSION_CODE} -lt 262400 ]; then echo "linux-4.0.9"; else echo "none"; fi)

executer/kernel/mcoverlayfs/linux-3.10.0-327.36.1.el7/readdir.c

@@ -15,6 +15,7 @@
 #include <linux/rbtree.h>
 #include <linux/security.h>
 #include <linux/cred.h>
+#include <linux/version.h>
 #include "overlayfs.h"
 
 struct ovl_cache_entry {
@@ -34,10 +35,18 @@ struct ovl_dir_cache {
 	struct list_head entries;
 };
 
+/* vfs_readdir vs. iterate_dir compat */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0) || \
+	(defined(RHEL_RELEASE_CODE) && RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 5))
+#define USE_ITERATE_DIR 1
+#endif
+
+#ifndef USE_ITERATE_DIR
 struct dir_context {
 	const filldir_t actor;
 	//loff_t pos;
 };
+#endif
 
 struct ovl_readdir_data {
 	struct dir_context ctx;
@@ -256,7 +265,11 @@ static inline int ovl_dir_read(struct path *realpath,
 	do {
 		rdd->count = 0;
 		rdd->err = 0;
+#ifdef USE_ITERATE_DIR
+		err = iterate_dir(realfile, &rdd->ctx);
+#else
 		err = vfs_readdir(realfile, rdd->ctx.actor, rdd);
+#endif
 		if (err >= 0)
 			err = rdd->err;
 	} while (!err && rdd->count);
@@ -365,6 +378,22 @@ static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 	return cache;
 }
 
+#ifdef USE_ITERATE_DIR
+struct iterate_wrapper {
+	struct dir_context ctx;
+	filldir_t actor;
+	void *buf;
+};
+
+static int ovl_wrap_readdir(void *ctx, const char *name, int namelen,
+			    loff_t offset, u64 ino, unsigned int d_type)
+{
+	struct iterate_wrapper *w = ctx;
+
+	return w->actor(w->buf, name, namelen, offset, ino, d_type);
+}
+#endif
+
 static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
 {
 	struct ovl_dir_file *od = file->private_data;
@@ -376,7 +405,16 @@ static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
 		ovl_dir_reset(file);
 
 	if (od->is_real) {
+#ifdef USE_ITERATE_DIR
+		struct iterate_wrapper w = {
+			.ctx.actor = ovl_wrap_readdir,
+			.actor = filler,
+			.buf = buf,
+		};
+		res = iterate_dir(od->realfile, &w.ctx);
+#else
 		res = vfs_readdir(od->realfile, filler, buf);
+#endif
 		file->f_pos = od->realfile->f_pos;
 
 		return res;

executer/user/Makefile.in

@@ -13,6 +13,8 @@ KDIR ?= @KDIR@
 ARCH=@ARCH@
 CFLAGS=-Wall -O -I. -I$(VPATH)/arch/${ARCH} -I${IHKDIR} -I@abs_builddir@/../../../ihk/linux/include
 LDFLAGS=@LDFLAGS@
+CPPFLAGS_SYSCALL_INTERCEPT=@CPPFLAGS_SYSCALL_INTERCEPT@
+LDFLAGS_SYSCALL_INTERCEPT=@LDFLAGS_SYSCALL_INTERCEPT@
 RPATH=$(shell echo $(LDFLAGS)|awk '{for(i=1;i<=NF;i++){if($$i~/^-L/){w=$$i;sub(/^-L/,"-Wl,-rpath,",w);print w}}}')
 VPATH=@abs_srcdir@
 TARGET=mcexec libsched_yield ldump2mcdump.so
@@ -21,12 +23,17 @@ LIBS=@LIBS@
 IHKDIR ?= $(VPATH)/../../../ihk/linux/include/
 MCEXEC_LIBS=-lmcexec -lrt -lnuma -pthread -L@abs_builddir@/../../../ihk/linux/user -lihk -Wl,-rpath,$(MCKERNEL_LIBDIR)
 ENABLE_QLMPI=@ENABLE_QLMPI@
+WITH_SYSCALL_INTERCEPT=@WITH_SYSCALL_INTERCEPT@
 
 ifeq ($(ENABLE_QLMPI),yes)
 	MCEXEC_LIBS += -lmpi
 	TARGET+= libqlmpi.so ql_server ql_mpiexec_start ql_mpiexec_finalize ql_talker libqlfort.so
 endif
 
+ifeq ($(WITH_SYSCALL_INTERCEPT),yes)
+	TARGET += syscall_intercept.so
+endif
+
 ifeq ($(ARCH), arm64)
 CFLAGS += $(foreach i, $(shell seq 1 100), $(addprefix -DPOSTK_DEBUG_ARCH_DEP_, $(i)))
 CFLAGS += $(foreach i, $(shell seq 1 100), $(addprefix -DPOSTK_DEBUG_TEMP_FIX_, $(i)))
@@ -40,10 +47,10 @@ mcexec: mcexec.c libmcexec.a
 # POSTK_DEBUG_ARCH_DEP_34, eclair arch depend separate.
 ifeq ($(ARCH), arm64)
 eclair: eclair.c arch/$(ARCH)/arch-eclair.c
-	$(CC) -I.. -I. -I./arch/$(ARCH)/include -I$(VPATH)/.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include $(CFLAGS) -o $@ $^ $(LIBS)
+	$(CC) -I.. -I. -I./arch/$(ARCH)/include -I$(VPATH)/.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include $(CFLAGS) -o $@ $^ $(LIBS) -ldl -lz
 else
-eclair: eclair.c
+eclair: eclair.c arch/$(ARCH)/arch-eclair.c
-	$(CC) $(CFLAGS) -I${IHKDIR} -o $@ $^ $(LIBS)
+	$(CC) -I.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include $(CFLAGS) -o $@ $^ $(LIBS)
 endif
 
 ldump2mcdump.so: ldump2mcdump.c
@@ -52,6 +59,12 @@ ldump2mcdump.so: ldump2mcdump.c
 libsched_yield: libsched_yield.c
 	$(CC) -shared -fPIC -Wl,-soname,sched_yield.so.1 -o libsched_yield.so.1.0.0 $^ -lc -ldl
 
+syscall_intercept.so: syscall_intercept.c libsyscall_intercept_arch.a
+	$(CC) $(CPPFLAGS_SYSCALL_INTERCEPT) -g -O2 $(LDFLAGS_SYSCALL_INTERCEPT) -lsyscall_intercept -fpic -shared -L. -lsyscall_intercept_arch $^ -o $@ 
+
+libsyscall_intercept_arch.a::
+	+(cd arch/${ARCH}; $(MAKE))
+
 libmcexec.a::
 	+(cd arch/${ARCH}; $(MAKE))
 
@@ -99,6 +112,9 @@ ifeq ($(ENABLE_QLMPI),yes)
 	install -m 755 ql_mpiexec_start $(BINDIR)
 	install -m 755 ql_mpiexec_finalize $(BINDIR)
 	install -m 755 ql_talker $(SBINDIR)
+endif
+ifeq ($(WITH_SYSCALL_INTERCEPT),yes)
+	install -m 755 syscall_intercept.so $(MCKERNEL_LIBDIR)
 endif
 	@uncomment_if_ENABLE_MEMDUMP@install -m 755 eclair $(BINDIR)
 	@uncomment_if_ENABLE_MEMDUMP@install -m 755 vmcore2mckdump $(BINDIR)

executer/user/arch/x86_64/Makefile.in

@@ -4,7 +4,7 @@ BINDIR=@BINDIR@
 KDIR ?= @KDIR@
 CFLAGS=-Wall -O -I.
 VPATH=@abs_srcdir@
-TARGET=../../libmcexec.a
+TARGET=../../libmcexec.a ../../libsyscall_intercept_arch.a
 LIBS=@LIBS@
 
 all: $(TARGET)
@@ -18,6 +18,12 @@ archdep.o: archdep.S
 arch_syscall.o: arch_syscall.c
 	$(CC) -c -I${KDIR} $(CFLAGS) $(EXTRA_CFLAGS) -fPIE -pie -pthread $<
 
+../../libsyscall_intercept_arch.a: archdep_c.o
+	$(AR) cr ../../libsyscall_intercept_arch.a archdep_c.o
+
+archdep_c.o: archdep_c.c
+	$(CC) -c -I${KDIR} $(CFLAGS) $(EXTRA_CFLAGS) -fPIE -pie -pthread $<
+
 clean:
 	$(RM) $(TARGET) *.o
 

executer/user/arch/x86_64/arch-eclair.c

@@ -42,7 +42,7 @@ int print_kregs(char *rbp, size_t rbp_size, const struct arch_kregs *kregs)
 	}
 
 	for (i = 0; i < sizeof(regs_1)/sizeof(regs_1[0]); i++) {	/* rsi, rdi, rbp, rsp */
-		ret = print_bin(rbp, rbp_size, (void *)regs_1[i], sizeof(regs_1[0]));
+		ret = print_bin(rbp, rbp_size, regs_1 + i, sizeof(regs_1[0]));
 		if (ret < 0) {
 			return ret;
 		}
@@ -62,7 +62,7 @@ int print_kregs(char *rbp, size_t rbp_size, const struct arch_kregs *kregs)
 	}
 
 	for (i = 0; i < sizeof(regs_2)/sizeof(regs_2[0]); i++) {	/* r12-r15 */
-		ret = print_bin(rbp, rbp_size, (void *)regs_2[i], sizeof(regs_2[0]));
+		ret = print_bin(rbp, rbp_size, regs_2 + i, sizeof(regs_2[0]));
 		if (ret < 0) {
 			return ret;
 		}

executer/user/arch/x86_64/arch_args.h

@@ -67,6 +67,12 @@ get_syscall_arg6(syscall_args *args)
 	return args->r9;
 }
 
+static inline unsigned long
+get_syscall_rip(syscall_args *args)
+{
+	return args->rip;
+}
+
 static inline void
 set_syscall_number(syscall_args *args, unsigned long value)
 {

executer/user/arch/x86_64/arch_syscall.c

@@ -48,7 +48,7 @@ archdep_syscall(struct syscall_wait_desc *w, long *ret)
 			if (*ret >= PATH_MAX) {
 				*ret = -ENAMETOOLONG;
 			}
-			if (ret < 0) {
+			if (*ret < 0) {
 				return 0;
 			}
 			__dprintf("open: %s\n", pathbuf);

executer/user/arch/x86_64/archdep.S

@@ -1,15 +1,22 @@
 /*
-arg: rdi, rsi, rdx, rcx, r8, r9
+Calling convention:
-ret: rax
+  arg: rdi, rsi, rdx, rcx, r8, r9
+  ret: rax
 
-rax syscall number
+  rdi: fd
-syscall: (rax:num) rdi rsi rdx r10 r8 r9 (rcx:ret addr)
+  rsi: cmd
-fd, cmd, param
+  rdx: param
-rdi: fd
+  rcx: save area
-rsi: cmd
+  r8: new thread context
-rdx: param
+
-rcx: save area
+Syscam call convention:
-r8: new thread context
+  syscall number: rax
+  arg: rdi, rsi, rdx, r10, r8, r9
+  return addr: rcx
+
+  rdi:	 fd
+  rsi:	cmd
+  rdx:	param
 */
 
 .global switch_ctx
@@ -91,6 +98,7 @@ switch_ctx:
 
 1:
 	mov	$0xffffffffffffffff,%eax
+	retq
 2:
 	pushq	%rax
 	movq	$158,%rax /* arch_prctl */
@@ -146,4 +154,3 @@ compare_and_swap_int:
 	lock
 	cmpxchgl	%edx,0(%rdi)
 	retq
-

executer/user/arch/x86_64/archdep_c.c

@@ -0,0 +1,52 @@
+/*
+function call convention
+rdi, rsi, rdx, rcx, r8, r9: 	IN	arguments
+rax:				OUT	return value
+
+syscall convention:
+rax:				IN	syscall number
+rdi, rsi, rdx, r10, r8, r9:	IN	arguments
+rax:				OUT	return value
+rcx, r11:			CLOBBER
+*/
+long uti_syscall6(long syscall_number, long arg0, long arg1, long arg2, long arg3, long arg4, long arg5)
+{
+	long ret;
+	asm volatile ("movq %[arg3],%%r10; movq %[arg4],%%r8; movq %[arg5],%%r9; syscall"
+				  : "=a" (ret)
+				  : "a" (syscall_number),
+					"D" (arg0), "S" (arg1), "d" (arg2),
+					[arg3] "g" (arg3), [arg4] "g" (arg4), [arg5] "g" (arg5)
+				  : "rcx", "r11", "r10", "r8", "r9", "memory");
+	return ret;
+}
+
+long uti_syscall3(long syscall_number, long arg0, long arg1, long arg2)
+{
+	long ret;
+	asm volatile ("syscall"
+				  : "=a" (ret)
+				  : "a" (syscall_number), "D" (arg0), "S" (arg1), "d" (arg2)
+				  : "rcx", "r11", "memory");
+	return ret;
+}
+
+long uti_syscall1(long syscall_number, long arg0)
+{
+	long ret;
+	asm volatile ("syscall"
+				  : "=a" (ret)
+				  : "a" (syscall_number), "D" (arg0)
+				  : "rcx", "r11", "memory");
+	return ret;
+}
+
+long uti_syscall0(long syscall_number)
+{
+	long ret;
+	asm volatile ("syscall"
+				  : "=a" (ret)
+				  : "a" (syscall_number)
+				  : "rcx", "r11", "memory");
+	return ret;
+}

executer/user/arch/x86_64/include/arch-eclair.h

@@ -2,8 +2,18 @@
 #ifndef HEADER_USER_X86_ECLAIR_H
 #define HEADER_USER_X86_ECLAIR_H
 
-#define MAP_KERNEL	0xFFFFFFFF80000000
+#ifndef POSTK_DEBUG_ARCH_DEP_34
-#define MAP_ST		0xFFFF800000000000
+#define MAP_ST_START       0xffff800000000000UL
+#define MAP_VMAP_START     0xffff850000000000UL
+#define MAP_FIXED_START    0xffff860000000000UL
+#define LINUX_PAGE_OFFSET  0xffff880000000000UL
+#define MAP_KERNEL_START   0xFFFFFFFFFE800000UL
+#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
+
+/* TODO: these should be updated when McKernel changes */
+#define MCKERNEL_ELF_START "0xFFFFFFFFFE801000"
+#define MCKERNEL_ELF_LEN   "0x0000000000100000"
+
 
 #define ARCH_CLV_SPAN	"x86_cpu_local_variables_span"
 

executer/user/archdep.h

@@ -1,4 +1,6 @@
-extern int switch_ctx(int fd, unsigned long cmd, void **param, void *lctx, void *rctx);
+#include "../include/uprotocol.h"
+
+extern int switch_ctx(int fd, unsigned long cmd, struct uti_save_fs_desc *desc, void *lctx, void *rctx);
 extern unsigned long compare_and_swap(unsigned long *addr, unsigned long old, unsigned long new);
 extern unsigned int compare_and_swap_int(unsigned int *addr, unsigned int old, unsigned int new);
 extern int archdep_syscall(struct syscall_wait_desc *w, long *ret);

executer/user/archdep_uti.h

@@ -0,0 +1,5 @@
+extern long uti_syscall6(long syscall_number, long arg0, long arg1, long arg2, long arg3, long arg4, long arg5);
+extern long uti_syscall3(long syscall_number, long arg0, long arg1, long arg2);
+extern long uti_syscall1(long syscall_number, long arg0);
+extern long uti_syscall0(long syscall_number);
+ 

executer/user/eclair.c

@@ -8,9 +8,7 @@
  * 	Copyright (C) 2015  RIKEN AICS
  */
 
-#ifdef POSTK_DEBUG_ARCH_DEP_33
 #include "../config.h"
-#endif	/* POSTK_DEBUG_ARCH_DEP_33 */
 #include <bfd.h>
 #include <fcntl.h>
 #include <inttypes.h>
@@ -22,10 +20,8 @@
 #include <arpa/inet.h>
 #include <sys/ioctl.h>
 #include <ihk/ihk_host_user.h>
-#ifdef POSTK_DEBUG_ARCH_DEP_34
 #include <eclair.h>
 #include <arch-eclair.h>
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 
 #define CPU_TID_BASE 1000000
 
@@ -85,11 +81,7 @@ static struct thread_info *curr_thread = NULL;
 static uintptr_t ihk_mc_switch_context = -1;
 #endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 
-#ifdef POSTK_DEBUG_ARCH_DEP_34
 uintptr_t lookup_symbol(char *name) {
-#else	/* POSTK_DEBUG_ARCH_DEP_34 */
-static uintptr_t lookup_symbol(char *name) {
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 	int i;
 
 	for (i = 0; i < nsyms; ++i) {
@@ -101,22 +93,22 @@ static uintptr_t lookup_symbol(char *name) {
 	return NOSYMBOL;
 } /* lookup_symbol() */
 
+#define NOPHYS ((uintptr_t)-1)
 
 static uintptr_t virt_to_phys(uintptr_t va) {
-#ifndef POSTK_DEBUG_ARCH_DEP_34
+	if (va >= MAP_KERNEL_START) {
-#define MAP_KERNEL 0xFFFFFFFF80000000
+		return va - MAP_KERNEL_START + kernel_base;
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
-	if (va >= MAP_KERNEL) {
-		return (va - MAP_KERNEL + kernel_base);
 	}
-#ifndef POSTK_DEBUG_ARCH_DEP_34
+	else if (va >= LINUX_PAGE_OFFSET) {
-#define MAP_ST 0xFFFF800000000000
+		return va - LINUX_PAGE_OFFSET;
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
-	if (va >= MAP_ST) {
-		return (va - MAP_ST);
 	}
-	if (0) printf("virt_to_phys(%lx): -1\n", va);
+	else if (va >= MAP_FIXED_START) {
-#define NOPHYS ((uintptr_t)-1)
+		return va - MAP_FIXED_START;
+	}
+	else if (va >= MAP_ST_START) {
+		return va - MAP_ST_START;
+	}
+
 	return NOPHYS;
 } /* virt_to_phys() */
 
@@ -673,11 +665,7 @@ static int setup_dump(char *fname) {
 	return 0;
 } /* setup_dump() */
 
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 static ssize_t print_hex(char *buf, size_t buf_size, char *str) {
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-static ssize_t print_hex(char *buf, char *str) {
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 
 	char *p;
 	char *q;
@@ -702,11 +690,7 @@ static ssize_t print_hex(char *buf, char *str) {
 	return (q - buf);
 } /* print_hex() */
 
-#if defined(POSTK_DEBUG_ARCH_DEP_34) && defined(POSTK_DEBUG_ARCH_DEP_38)
 ssize_t print_bin(char *buf, size_t buf_size, void *data, size_t size) {
-#else	/* POSTK_DEBUG_ARCH_DEP_34 && POSTK_DEBUG_ARCH_DEP_38*/
-static ssize_t print_bin(char *buf, void *data, size_t size) {
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 && POSTK_DEBUG_ARCH_DEP_38*/
 	uint8_t *p;
 	char *q;
 	int i;
@@ -733,13 +717,8 @@ static ssize_t print_bin(char *buf, void *data, size_t size) {
 	return (q - buf);
 } /* print_bin() */
 
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 static void command(const char *cmd, char *res, size_t res_size) {
 	const char *p;
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-static void command(char *cmd, char *res) {
-	char *p;
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 	char *rbp;
 
 	p = cmd;
@@ -801,11 +780,7 @@ static void command(char *cmd, char *res) {
 #endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 			rbp += sprintf(rbp, "l");
 			if (0)
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 			rbp += print_hex(rbp, res_size, str);
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-			rbp += print_hex(rbp, str);
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 			rbp += sprintf(rbp, "%s", str);
 		}
 		else if (!strcmp(p, "D")) {
@@ -814,20 +789,9 @@ static void command(char *cmd, char *res) {
 		}
 		else if (!strcmp(p, "g")) {
 			if (curr_thread->cpu < 0) {
-#ifndef POSTK_DEBUG_ARCH_DEP_34
-				struct x86_kregs {
-					uintptr_t rsp, rbp, rbx, rsi;
-					uintptr_t rdi, r12, r13, r14;
-					uintptr_t r15, rflags, rsp0;
-				};
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 
 				int error;
-#ifdef POSTK_DEBUG_ARCH_DEP_34
 				struct arch_kregs kregs;
-#else	/* POSTK_DEBUG_ARCH_DEP_34 */
-				struct x86_kregs kregs;
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 
 				error = read_mem(curr_thread->process+K(CTX_OFFSET),
 						&kregs, sizeof(kregs));
@@ -836,36 +800,7 @@ static void command(char *cmd, char *res) {
 					break;
 				}
 
-#ifdef POSTK_DEBUG_ARCH_DEP_34
 				print_kregs(rbp, res_size, &kregs);
-#else	/* POSTK_DEBUG_ARCH_DEP_34 */
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* rax */
-				rbp += print_bin(rbp, &kregs.rbx, sizeof(uint64_t));
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* rcx */
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* rdx */
-				rbp += print_bin(rbp, &kregs.rsi, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.rdi, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.rbp, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.rsp, sizeof(uint64_t));
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* r8 */
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* r9 */
-
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* r10 */
-				rbp += sprintf(rbp, "xxxxxxxxxxxxxxxx");	/* r11 */
-				rbp += print_bin(rbp, &kregs.r12, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.r13, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.r14, sizeof(uint64_t));
-				rbp += print_bin(rbp, &kregs.r15, sizeof(uint64_t));
-				rbp += print_bin(rbp, &ihk_mc_switch_context,
-						sizeof(uint64_t));		/* rip */
-				rbp += print_bin(rbp, &kregs.rflags, sizeof(uint32_t));
-				rbp += sprintf(rbp, "xxxxxxxx");		/* cs */
-				rbp += sprintf(rbp, "xxxxxxxx");		/* ss */
-				rbp += sprintf(rbp, "xxxxxxxx");		/* ds */
-				rbp += sprintf(rbp, "xxxxxxxx");		/* es */
-				rbp += sprintf(rbp, "xxxxxxxx");		/* fs */
-				rbp += sprintf(rbp, "xxxxxxxx");		/* gs */
-#endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 			}
 			else {
 				int error;
@@ -943,11 +878,7 @@ static void command(char *cmd, char *res) {
 #endif	/* POSTK_DEBUG_ARCH_DEP_34 */
 			rbp += sprintf(rbp, "l");
 			if (0)
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 			rbp += print_hex(rbp, res_size, str);
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-			rbp += print_hex(rbp, str);
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 			rbp += sprintf(rbp, "%s", str);
 		}
 		else if (!strncmp(p, "T", 1)) {
@@ -1039,11 +970,7 @@ static void command(char *cmd, char *res) {
 			else {
 				q += sprintf(q, "status=%#x", ti->status);
 			}
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 			rbp += print_hex(rbp, res_size, buf);
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-			rbp += print_hex(rbp, buf);
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 		}
 	} while (0);
 
@@ -1272,11 +1199,7 @@ int main(int argc, char *argv[]) {
 			}
 			mode = 0;
 			fputc('+', ofp);
-#ifdef POSTK_DEBUG_ARCH_DEP_38
 			command(lbuf, rbuf, sizeof(rbuf));
-#else	/* POSTK_DEBUG_ARCH_DEP_38 */
-			command(lbuf, rbuf);
-#endif	/* POSTK_DEBUG_ARCH_DEP_38 */
 			sum = 0;
 			for (p = rbuf; *p != '\0'; ++p) {
 				sum += *p;

executer/user/eclair.h

@@ -3,11 +3,7 @@
 #ifndef HEADER_USER_COMMON_ECLAIR_H
 #define HEADER_USER_COMMON_ECLAIR_H
 
-#ifdef POSTK_DEBUG_ARCH_DEP_76 /* header path fix */
 #include "../config.h"
-#else /* POSTK_DEBUG_ARCH_DEP_76 */
-#include <config.h>
-#endif /* POSTK_DEBUG_ARCH_DEP_76 */
 #include <stdio.h>
 #include <inttypes.h>
 #include <arch-eclair.h>

executer/user/libsched_yield.c

@@ -11,7 +11,9 @@
 
 typedef int (*int_void_fn)(void);
 
+#if 0
 static int_void_fn orig_sched_yield = 0;
+#endif
 
 int sched_yield(void)
 {

executer/user/syscall_intercept.c

@@ -0,0 +1,139 @@
+#include <libsyscall_intercept_hook_point.h>
+#include <errno.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <syscall.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include "../include/uprotocol.h"
+#include "../include/uti.h"
+#include "./archdep_uti.h"
+
+static struct uti_desc uti_desc;
+
+#define DEBUG_UTI
+
+static int
+hook(long syscall_number,
+	 long arg0, long arg1,
+	 long arg2, long arg3,
+	 long arg4, long arg5,
+	 long *result)
+{
+	//return 1; /* debug */
+	int tid = uti_syscall0(__NR_gettid);
+	struct terminate_thread_desc term_desc;
+	unsigned long code;
+	int stack_top;
+		
+	if (!uti_desc.start_syscall_intercept) {
+		return 1; /* System call isn't taken over */
+	}
+	if (tid != uti_desc.mck_tid) {
+		if (uti_desc.syscalls2 && syscall_number >= 0 && syscall_number < 512) {
+			uti_desc.syscalls2[syscall_number]++;
+		}
+		return 1;
+	}
+#ifdef DEBUG_UTI
+	if (uti_desc.syscalls && syscall_number >= 0 && syscall_number < 512) {
+		uti_desc.syscalls[syscall_number]++;
+	}
+#endif
+
+	switch (syscall_number) {
+	case __NR_gettid:
+		*result = uti_desc.mck_tid;
+		return 0;
+	case __NR_futex:
+	case __NR_brk:
+	case __NR_mmap:
+	case __NR_munmap:
+	case __NR_mprotect:
+	case __NR_mremap:
+		/* Overflow check */
+		if (uti_desc.syscall_stack_top == -1) {
+			*result = -ENOMEM;
+			return 0;
+		}
+	
+		/* Sanity check */
+		if (uti_desc.syscall_stack_top < 0 || uti_desc.syscall_stack_top >= UTI_SZ_SYSCALL_STACK) {
+			*result = -EINVAL;
+			return 0;
+		}
+
+		/* Store the return value in the stack to prevent it from getting corrupted 
+		   when an interrupt happens just after ioctl() and before copying the return
+		   value to *result */
+		stack_top = __sync_fetch_and_sub(&uti_desc.syscall_stack_top, 1);
+		
+		uti_desc.syscall_stack[stack_top].number = syscall_number;
+		uti_desc.syscall_stack[stack_top].args[0] = arg0;
+		uti_desc.syscall_stack[stack_top].args[1] = arg1;
+		uti_desc.syscall_stack[stack_top].args[2] = arg2;
+		uti_desc.syscall_stack[stack_top].args[3] = arg3;
+		uti_desc.syscall_stack[stack_top].args[4] = arg4;
+		uti_desc.syscall_stack[stack_top].args[5] = arg5;
+		uti_desc.syscall_stack[stack_top].uti_clv = uti_desc.uti_clv;
+		uti_desc.syscall_stack[stack_top].ret = -EINVAL;
+
+		uti_syscall3(__NR_ioctl, uti_desc.fd, MCEXEC_UP_SYSCALL_THREAD, (long)(uti_desc.syscall_stack + stack_top));
+		*result = uti_desc.syscall_stack[stack_top].ret;
+		
+		/* push syscall_struct list */
+		__sync_fetch_and_add(&uti_desc.syscall_stack_top, 1);
+
+		return 0; /* System call is taken over */
+	case __NR_exit_group:
+		code = 0x100000000;
+		goto make_remote_thread_exit;
+	case __NR_exit:
+		code = 0;
+	make_remote_thread_exit:
+		/* Make migrated-to-Linux thread on the McKernel side call do_exit() or terminate() */
+		term_desc.pid = uti_desc.pid;
+		term_desc.tid = uti_desc.tid; /* tid of mcexec */
+		term_desc.code = code | ((arg0 & 255) << 8);
+		term_desc.tsk = uti_desc.key;
+
+		uti_syscall3(__NR_ioctl, uti_desc.fd, MCEXEC_UP_TERMINATE_THREAD, (long)&term_desc);
+		return 1;
+	case __NR_clone:
+	case __NR_fork:
+	case __NR_vfork:
+	case __NR_execve:
+		*result = -ENOSYS;
+		return 0;
+#if 0 /* debug */
+	case __NR_set_robust_list:
+		*result = -ENOSYS;
+		return 0;
+#endif
+	case 888:
+        *result = (long)&uti_desc;
+		return 0;
+	default:
+		return 1;
+	}
+
+	return 0;
+}
+
+static __attribute__((constructor)) void
+init(void)
+{
+	/* Set up the callback function */
+	intercept_hook_point = hook;
+
+	/* Initialize uti_desc */
+	uti_desc.syscall_stack_top = UTI_SZ_SYSCALL_STACK - 1;
+
+	/* Pass address of uti_desc to McKernel */
+	uti_syscall1(733, (unsigned long)&uti_desc);
+}
+
+static __attribute__((destructor)) void
+dtor(void)
+{
+}

kernel/Makefile.build.in

@@ -6,7 +6,7 @@ IHKDIR=$(IHKBASE)/$(TARGETDIR)
 OBJS = init.o mem.o debug.o mikc.o listeners.o ap.o syscall.o cls.o host.o
 OBJS += process.o copy.o waitq.o futex.o timer.o plist.o fileobj.o shmobj.o
 OBJS += zeroobj.o procfs.o devobj.o sysfs.o xpmem.o profile.o freeze.o
-OBJS += rbtree.o
+OBJS += rbtree.o hugefileobj.o
 OBJS += pager.o
 # POSTK_DEBUG_ARCH_DEP_18 coredump arch separation.
 DEPSRCS=$(wildcard $(SRC)/*.c)
@@ -19,7 +19,7 @@ endif
 
 CFLAGS += -I$(SRC)/include -I@abs_builddir@/../ -I@abs_builddir@/include -D__KERNEL__  -g -fno-omit-frame-pointer -fno-inline -fno-inline-small-functions
 ifneq ($(ARCH), arm64)
-CFLAGS += -mcmodel=large -mno-red-zone
+CFLAGS += -mcmodel=large -mno-red-zone -mno-sse
 endif
 LDFLAGS += -e arch_start
 IHKOBJ = ihk/ihk.o

kernel/ap.c

@@ -29,15 +29,13 @@
 #include <time.h>
 #include <syscall.h>
 #include <rusage_private.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_AP
 
 #ifdef DEBUG_PRINT_AP
-#define dkprintf(...) do { kprintf(__VA_ARGS__); } while (0)
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) do { kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { } while (0)
-#define ekprintf(...) do { kprintf(__VA_ARGS__); } while (0)
 #endif
 
 int num_processors = 1;
@@ -209,8 +207,10 @@ store_fake_cpu_info(struct sysfs_ops *ops0, void *instance, void *buf,
 
 static struct fake_cpu_info_ops show_fci_online = {
 	.member = ONLINE,
-	.ops.show = &show_fake_cpu_info,
+	.ops = {
-	.ops.store = &store_fake_cpu_info,
+		.show = &show_fake_cpu_info,
+		.store = &store_fake_cpu_info,
+	},
 };
 
 void

kernel/config/attached-mic.lds

@@ -1,24 +1,28 @@
 PHDRS 
 {
 	text PT_LOAD FLAGS(5);
-    data PT_LOAD FLAGS(7);
+	data PT_LOAD FLAGS(7);
 }
 SECTIONS
 {
 	. = 0xffffffff80001000;
-    _head = .;
+	_head = .;
 
-    .text : {
+	.text : {
-        *(.text);
+		*(.text);
-    } : text
+	} : text
 
-    . = ALIGN(4096);
+	. = ALIGN(4096);
 	.data : {
-	      *(.data)
+		*(.data)
-	      *(.data.*)
+		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
-	      *(.rodata .rodata.*)
+		*(.rodata .rodata.*)
 	} :data
 
 	.vsyscall : ALIGN(0x1000) {
@@ -37,14 +41,14 @@ SECTIONS
 		. = ALIGN(4096);
 	} : data = 0xf4
 
-    .bss : {
+	.bss : {
-          *(.bss .bss.*)
+		*(.bss .bss.*)
-    }
+	}
-    . = ALIGN(4096);
+	. = ALIGN(4096);
-    _end = .;
+	_end = .;
 
 	/DISCARD/ : {
-	     *(.eh_frame)
+		*(.eh_frame)
-         *(.note.gnu.build-id)
+		*(.note.gnu.build-id)
 	}
 }

kernel/config/builtin-mic.lds

@@ -1,24 +1,28 @@
 PHDRS 
 {
 	text PT_LOAD FLAGS(5);
-    data PT_LOAD FLAGS(7);
+	data PT_LOAD FLAGS(7);
 }
 SECTIONS
 {
 	. = 0xffffffff80001000;
-    _head = .;
+	_head = .;
 
-    .text : {
+	.text : {
-        *(.text);
+		*(.text);
-    } : text
+	} : text
 
-    . = ALIGN(4096);
+	. = ALIGN(4096);
 	.data : {
-	      *(.data)
+		*(.data)
-	      *(.data.*)
+		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
-	      *(.rodata .rodata.*)
+		*(.rodata .rodata.*)
 	} :data
 
 	.vsyscall : ALIGN(0x1000) {
@@ -37,14 +41,14 @@ SECTIONS
 		. = ALIGN(4096);
 	} : data = 0xf4
 
-    .bss : {
+	.bss : {
-          *(.bss .bss.*)
+		*(.bss .bss.*)
-    }
+	}
-    . = ALIGN(4096);
+	. = ALIGN(4096);
-    _end = .;
+	_end = .;
 
 	/DISCARD/ : {
-	     *(.eh_frame)
+		*(.eh_frame)
-         *(.note.gnu.build-id)
+		*(.note.gnu.build-id)
 	}
 }

kernel/config/builtin-x86.lds

@@ -1,24 +1,28 @@
 PHDRS 
 {
 	text PT_LOAD FLAGS(5);
-    data PT_LOAD FLAGS(7);
+	data PT_LOAD FLAGS(7);
 }
 SECTIONS
 {
 	. = 0xffffffff80001000;
-    _head = .;
+	_head = .;
 
-    .text : {
+	.text : {
-        *(.text);
+		*(.text);
-    } : text
+	} : text
 
-    . = ALIGN(4096);
+	. = ALIGN(4096);
 	.data : {
-	      *(.data)
+		*(.data)
-	      *(.data.*)
+		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
-	      *(.rodata .rodata.*)
+		*(.rodata .rodata.*)
 	} :data
 
 	.vsyscall : ALIGN(0x1000) {
@@ -37,10 +41,10 @@ SECTIONS
 		. = ALIGN(4096);
 	} : data = 0xf4
 
-    .bss : {
+	.bss : {
-          *(.bss .bss.*)
+		*(.bss .bss.*)
-    }
+	}
-    . = ALIGN(4096);
+	. = ALIGN(4096);
-    _end = .;
+	_end = .;
 
 }

kernel/config/smp-arm64_type1.lds

@@ -16,6 +16,10 @@ SECTIONS
 	.data : {
 		*(.data)
 		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
 		*(.rodata .rodata.*)

kernel/config/smp-arm64_type2.lds

@@ -16,6 +16,10 @@ SECTIONS
 	.data : {
 		*(.data)
 		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
 		*(.rodata .rodata.*)

kernel/config/smp-arm64_type3.lds

@@ -16,6 +16,10 @@ SECTIONS
 	.data : {
 		*(.data)
 		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
 		*(.rodata .rodata.*)

kernel/config/smp-arm64_type4.lds

@@ -16,6 +16,10 @@ SECTIONS
 	.data : {
 		*(.data)
 		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
 		*(.rodata .rodata.*)

kernel/config/smp-x86.lds

@@ -1,24 +1,28 @@
 PHDRS 
 {
 	text PT_LOAD FLAGS(5);
-    data PT_LOAD FLAGS(7);
+	data PT_LOAD FLAGS(7);
 }
 SECTIONS
 {
-	. = 0xffffffff80001000;
+	. = 0xFFFFFFFFFE801000;
-    _head = .;
+	_head = .;
 
-    .text : {
+	.text : {
-        *(.text);
+		*(.text);
-    } : text
+	} : text
 
-    . = ALIGN(4096);
+	. = ALIGN(4096);
 	.data : {
-	      *(.data)
+		*(.data)
-	      *(.data.*)
+		*(.data.*)
+		. = ALIGN(8);
+		__start___verbose = .;
+		*(__verbose);
+		__stop___verbose = .;
 	} :data
 	.rodata : {
-	      *(.rodata .rodata.*)
+		*(.rodata .rodata.*)
 	} :data
 
 	.vsyscall : ALIGN(0x1000) {
@@ -37,9 +41,9 @@ SECTIONS
 		. = ALIGN(4096);
 	} : data = 0xf4
 
-    .bss : {
+	.bss : {
-          *(.bss .bss.*)
+		*(.bss .bss.*)
-    }
+	}
-    . = ALIGN(4096);
+	. = ALIGN(4096);
-    _end = .;
+	_end = .;
 }

kernel/debug.c

@@ -18,6 +18,9 @@
 #include <ihk/lock.h>
 #include <ihk/monitor.h>
 #include <errno.h>
+#include <sysfs.h>
+#include <debug.h>
+#include <limits.h>
 
 struct ihk_kmsg_buf *kmsg_buf;
 
@@ -84,7 +87,8 @@ void kputs(char *buf)
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
 	kprintf_unlock(flags_outer);
 
-	if (DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
+	if (irqflags_can_interrupt(flags_outer) &&
+			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);
 	}
@@ -123,8 +127,8 @@ int __kprintf(const char *format, ...)
 	}
 
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
-
+	if (irqflags_can_interrupt(flags_inner) &&
-	if (DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
+			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);
 	}
@@ -165,7 +169,8 @@ int kprintf(const char *format, ...)
 	debug_spin_unlock_irqrestore(&kmsg_buf->lock, flags_inner);
 	kprintf_unlock(flags_outer);
 
-	if (DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
+	if (irqflags_can_interrupt(flags_outer) &&
+			DEBUG_KMSG_USED > IHK_KMSG_HIGH_WATER_MARK) {
 		eventfd(IHK_OS_EVENTFD_TYPE_KMSG);
 		ihk_mc_delay_us(IHK_KMSG_NOTIFY_DELAY);
 	}
@@ -178,3 +183,147 @@ void kmsg_init()
 {
 	ihk_mc_spinlock_init(&kmsg_lock);
 }
+
+extern struct ddebug __start___verbose[];
+extern struct ddebug __stop___verbose[];
+
+static ssize_t dynamic_debug_sysfs_show(struct sysfs_ops *ops,
+		void *instance, void *buf, size_t size)
+{
+	struct ddebug *dbg;
+	ssize_t n = 0;
+
+	n = snprintf(buf, size, "# filename:lineno function flags format\n");
+
+	for (dbg = __start___verbose; dbg < __stop___verbose; dbg++) {
+		n += snprintf(buf + n, size - n, "%s:%d %s =%s\n",
+				dbg->file, dbg->line, dbg->func,
+				dbg->flags ? "p" : "_");
+
+		if (n >= size)
+			break;
+	}
+
+	return n;
+}
+
+static ssize_t dynamic_debug_sysfs_store(struct sysfs_ops *ops,
+		void *instance, void *buf, size_t size)
+{
+	char *cur = buf;
+	char *file = NULL, *func = NULL;
+	long int line_start = 0, line_end = INT_MAX;
+	int set_flag = -1;
+	struct ddebug *dbg;
+
+
+	// assume line was new-line terminated and squash last newline
+	cur[size-1] = '\0';
+
+	/* basic line parsing, combinaisons of:
+	 *   file <file>
+	 *   func <func>
+	 *   line <line|line-line|line-|-line>
+	 *   and must end with [+-=][p_] (set/clear print flag)
+	 */
+again:
+	while (cur && cur < ((char *)buf) + size && *cur) {
+		dkprintf("looking at %.*s, size left %d\n",
+			size - (cur - (char *)buf), cur,
+			(char *)buf - cur + size);
+
+		if (strncmp(cur, "func ", 5) == 0) {
+			cur += 5;
+			func = cur;
+		} else if (strncmp(cur, "file ", 5) == 0) {
+			cur += 5;
+			file = cur;
+		} else if (strncmp(cur, "line ", 5) == 0) {
+			cur += 5;
+			if (*cur != '-') {
+				line_start = strtol(cur, &cur, 0);
+			}
+			if (*cur != '-') {
+				line_end = line_start;
+			} else {
+				cur++;
+				if (*cur == ' ' || *cur == '\0') {
+					line_end = INT_MAX;
+				} else {
+					line_end = strtol(cur, &cur, 0);
+				}
+			}
+		} else if (strchr("+-=", *cur)) {
+			switch ((*cur) + 256 * (*(cur+1))) {
+			case '+' + 256*'p':
+			case '=' + 256*'p':
+				set_flag = DDEBUG_PRINT;
+				break;
+			case '-' + 256*'p':
+			case '=' + 256*'_':
+				set_flag = DDEBUG_NONE;
+				break;
+			default:
+				kprintf("invalid flag: %.*s\n",
+					size - (cur - (char *)buf), cur);
+				return -EINVAL;
+			}
+			/* XXX check 3rd char is end of input or \n or ; */
+			cur += 3;
+			break;
+
+		} else {
+			kprintf("dynamic debug control: unrecognized keyword: %.*s\n",
+				size - (cur - (char *)buf), cur);
+			return -EINVAL;
+		}
+		cur = strpbrk(cur, " \n");
+		if (cur) {
+			*cur = '\0';
+			cur++;
+		}
+	}
+	dkprintf("func %s, file %s, lines %d-%d, flag %x\n",
+		func, file, line_start, line_end, set_flag);
+
+	if (set_flag < 0) {
+		kprintf("dynamic debug control: no flag set?\n");
+		return -EINVAL;
+	}
+	if (!func && !file) {
+		kprintf("at least file or func should be set\n");
+		return -EINVAL;
+	}
+
+	for (dbg = __start___verbose; dbg < __stop___verbose; dbg++) {
+		/* TODO: handle wildcards */
+		if ((!func || strcmp(func, dbg->func) == 0) &&
+		    (!file || strcmp(file, dbg->file) == 0) &&
+		    dbg->line >= line_start &&
+		    dbg->line <= line_end) {
+			dbg->flags = set_flag;
+		}
+	}
+
+	if (cur && cur < ((char *)buf) + size && *cur)
+		goto again;
+
+	return size;
+}
+
+static struct sysfs_ops dynamic_debug_sysfs_ops = {
+	.show = &dynamic_debug_sysfs_show,
+	.store = &dynamic_debug_sysfs_store,
+};
+
+void dynamic_debug_sysfs_setup(void)
+{
+	int error;
+
+	error = sysfs_createf(&dynamic_debug_sysfs_ops, NULL, 0644,
+			      "/sys/kernel/debug/dynamic_debug/control");
+	if (error) {
+		kprintf("%s: ERROR: creating dynamic_debug/control sysfs file",
+			__func__);
+	}
+}

kernel/devobj.c

@@ -36,15 +36,13 @@
 #include <syscall.h>
 #include <process.h>
 #include <rusage_private.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_DEVOBJ
 
 #ifdef DEBUG_PRINT_DEVOBJ
-#define	dkprintf(...) kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...) kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...) kprintf(__VA_ARGS__)
 #endif
 
 
@@ -54,16 +52,15 @@ struct devobj {
 	uintptr_t	handle;
 	off_t		pfn_pgoff;
 	uintptr_t *	pfn_table;
+	ihk_spinlock_t  pfn_table_lock;
 	size_t		npages;
 };
 
-static memobj_release_func_t devobj_release;
+static memobj_free_func_t devobj_free;
-static memobj_ref_func_t devobj_ref;
 static memobj_get_page_func_t devobj_get_page;
 
 static struct memobj_ops devobj_ops = {
-	.release =	&devobj_release,
+	.free =		&devobj_free,
-	.ref =		&devobj_ref,
 	.get_page =	&devobj_get_page,
 };
 
@@ -88,12 +85,9 @@ int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxp
 	int error;
 	struct devobj *obj  = NULL;
 	const size_t npages = (len + PAGE_SIZE - 1) / PAGE_SIZE;
-#ifdef POSTK_DEBUG_TEMP_FIX_36
 	const size_t uintptr_per_page = (PAGE_SIZE / sizeof(uintptr_t));
-	const size_t pfn_npages = (npages + uintptr_per_page - 1) / uintptr_per_page;
+	const size_t pfn_npages =
-#else
+		(npages + uintptr_per_page - 1) / uintptr_per_page;
-	const size_t pfn_npages = (npages / (PAGE_SIZE / sizeof(uintptr_t))) + 1;
-#endif /*POSTK_DEBUG_TEMP_FIX_36*/
 
 	dkprintf("%s: fd: %d, len: %lu, off: %lu \n", __FUNCTION__, fd, len, off);
 
@@ -122,6 +116,8 @@ int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxp
 	ihk_mc_syscall_arg4(&ctx) = virt_to_phys(&result);
 	ihk_mc_syscall_arg5(&ctx) = prot | populate_flags;
 
+	memset(&result, 0, sizeof(result));
+
 	error = syscall_generic_forwarding(__NR_mmap, &ctx);
 	if (error) {
 		kprintf("%s: error: fd: %d, len: %lu, off: %lu map failed.\n", 
@@ -135,6 +131,7 @@ int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxp
 	obj->memobj.ops = &devobj_ops;
 	obj->memobj.flags = MF_HAS_PAGER | MF_DEV_FILE;
 	obj->memobj.size = len;
+	ihk_atomic_set(&obj->memobj.refcnt, 1);
 	obj->handle = result.handle;
 
 	dkprintf("%s: path=%s\n", __FUNCTION__, result.path);
@@ -148,10 +145,9 @@ int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxp
 		strncpy(obj->memobj.path, result.path, PATH_MAX);
 	}
 
-	obj->ref = 1;
+	obj->pfn_pgoff = off >> PAGE_SHIFT;
-	obj->pfn_pgoff = off / PAGE_SIZE;
 	obj->npages = npages;
-	ihk_mc_spinlock_init(&obj->memobj.lock);
+	ihk_mc_spinlock_init(&obj->pfn_table_lock);
 
 	error = 0;
 	*objp = to_memobj(obj);
@@ -170,81 +166,50 @@ out:
 	return error;
 }
 
-static void devobj_ref(struct memobj *memobj)
+static void devobj_free(struct memobj *memobj)
 {
 	struct devobj *obj = to_devobj(memobj);
-
-	dkprintf("devobj_ref(%p %lx):\n", obj, obj->handle);
-	memobj_lock(&obj->memobj);
-	++obj->ref;
-	memobj_unlock(&obj->memobj);
-	return;
-}
-
-static void devobj_release(struct memobj *memobj)
-{
-	struct devobj *obj = to_devobj(memobj);
-	struct devobj *free_obj = NULL;
 	uintptr_t handle;
-#ifndef POSTK_DEBUG_TEMP_FIX_36
+	const size_t uintptr_per_page = (PAGE_SIZE / sizeof(uintptr_t));
 	const size_t pfn_npages =
-		(obj->npages / (PAGE_SIZE / sizeof(uintptr_t))) + 1;
+		(obj->npages + uintptr_per_page - 1) / uintptr_per_page;
-#endif /*!POSTK_DEBUG_TEMP_FIX_36*/
+	int error;
+	ihk_mc_user_context_t ctx;
 
-	dkprintf("devobj_release(%p %lx)\n", obj, obj->handle);
+	dkprintf("%s(%p %lx)\n", __func__, obj, obj->handle);
 
-	memobj_lock(&obj->memobj);
-	--obj->ref;
-	if (obj->ref <= 0) {
-		free_obj = obj;
-	}
 	handle = obj->handle;
-	memobj_unlock(&obj->memobj);
 
-	if (free_obj) {
+	ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_UNMAP;
-		if (!(free_obj->memobj.flags & MF_HOST_RELEASED)) {
+	ihk_mc_syscall_arg1(&ctx) = handle;
-			int error;
+	ihk_mc_syscall_arg2(&ctx) = 1;
-			ihk_mc_user_context_t ctx;
 
-			ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_UNMAP;
+	error = syscall_generic_forwarding(__NR_mmap, &ctx);
-			ihk_mc_syscall_arg1(&ctx) = handle;
+	if (error) {
-			ihk_mc_syscall_arg2(&ctx) = 1;
+		kprintf("%s(%p %lx): release failed. %d\n",
-
+			__func__, obj, handle, error);
-			error = syscall_generic_forwarding(__NR_mmap, &ctx);
+		/* through */
-			if (error) {
-				kprintf("devobj_release(%p %lx):"
-						"release failed. %d\n",
-						free_obj, handle, error);
-				/* through */
-			}
-		}
-
-		if (obj->pfn_table) {
-			// Don't call memory_stat_rss_sub() because devobj related pages don't reside in main memory
-#ifdef POSTK_DEBUG_TEMP_FIX_36
-			const size_t uintptr_per_page = (PAGE_SIZE / sizeof(uintptr_t));
-			const size_t pfn_npages = (obj->npages + uintptr_per_page - 1) / uintptr_per_page;
-			ihk_mc_free_pages(obj->pfn_table, pfn_npages);
-#else
-			ihk_mc_free_pages(obj->pfn_table, pfn_npages);
-#endif /*POSTK_DEBUG_TEMP_FIX_36*/
-		}
-
-		if (to_memobj(free_obj)->path) {
-			kfree(to_memobj(free_obj)->path);
-		}
-
-		kfree(free_obj);
 	}
 
-	dkprintf("devobj_release(%p %lx):free %p\n",
+	if (obj->pfn_table) {
-			obj, handle, free_obj);
+		// Don't call memory_stat_rss_sub() because devobj related
+		// pages don't reside in main memory
+		ihk_mc_free_pages(obj->pfn_table, pfn_npages);
+	}
+
+	if (to_memobj(obj)->path) {
+		kfree(to_memobj(obj)->path);
+	}
+
+	kfree(obj);
+
+	dkprintf("%s(%p %lx):free\n", __func__, obj, handle);
 	return;
 }
 
 static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintptr_t *physp, unsigned long *flag, uintptr_t virt_addr)
 {
-	const off_t pgoff = off / PAGE_SIZE;
+	const off_t pgoff = off >> PAGE_SHIFT;
 	struct devobj *obj = to_devobj(memobj);
 	int error;
 	uintptr_t pfn;
@@ -262,17 +227,14 @@ static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintpt
 	ix = pgoff - obj->pfn_pgoff;
 	dkprintf("ix: %ld\n", ix);
 
-	memobj_lock(&obj->memobj);
-	pfn = obj->pfn_table[ix];
 #ifdef PROFILE_ENABLE
 	profile_event_add(PROFILE_page_fault_dev_file, PAGE_SIZE);
 #endif // PROFILE_ENABLE
+	pfn = obj->pfn_table[ix];
 	if (!(pfn & PFN_VALID)) {
-		memobj_unlock(&obj->memobj);
-
 		ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_PFN;
 		ihk_mc_syscall_arg1(&ctx) = obj->handle;
-		ihk_mc_syscall_arg2(&ctx) = pgoff << PAGE_SHIFT;
+		ihk_mc_syscall_arg2(&ctx) = off & ~(PAGE_SIZE - 1);
 		ihk_mc_syscall_arg3(&ctx) = virt_to_phys(&pfn);
 
 		error = syscall_generic_forwarding(__NR_mmap, &ctx);
@@ -303,11 +265,9 @@ static int devobj_get_page(struct memobj *memobj, off_t off, int p2align, uintpt
 			dkprintf("devobj_get_page(%p %lx,%lx,%d):PFN_PRESENT after %#lx\n", memobj, obj->handle, off, p2align, pfn);
 		}
 
-		memobj_lock(&obj->memobj);
 		obj->pfn_table[ix] = pfn;
 		// Don't call memory_stat_rss_add() because devobj related pages don't reside in main memory
 	}
-	memobj_unlock(&obj->memobj);
 
 	if (!(pfn & PFN_PRESENT)) {
 		kprintf("devobj_get_page(%p %lx,%lx,%d):not present. %lx\n", memobj, obj->handle, off, p2align, pfn);

kernel/fileobj.c

@@ -27,15 +27,13 @@
 #include <string.h>
 #include <syscall.h>
 #include <rusage_private.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_FILEOBJ
 
 #ifdef DEBUG_PRINT_FILEOBJ
-#define	dkprintf(...)	do { if (1) kprintf(__VA_ARGS__); } while (0)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define	dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 mcs_lock_t fileobj_list_lock;
@@ -47,24 +45,21 @@ static LIST_HEAD(fileobj_list);
 
 struct fileobj {
 	struct memobj memobj;		/* must be first */
-	long sref;
+	uint64_t sref;
-	long cref;
 	uintptr_t handle;
 	struct list_head list;
 	struct list_head page_hash[FILEOBJ_PAGE_HASH_SIZE];
 	mcs_lock_t page_hash_locks[FILEOBJ_PAGE_HASH_SIZE];
 };
 
-static memobj_release_func_t fileobj_release;
+static memobj_free_func_t fileobj_free;
-static memobj_ref_func_t fileobj_ref;
 static memobj_get_page_func_t fileobj_get_page;
 static memobj_flush_page_func_t fileobj_flush_page;
 static memobj_invalidate_page_func_t fileobj_invalidate_page;
 static memobj_lookup_page_func_t fileobj_lookup_page;
 
 static struct memobj_ops fileobj_ops = {
-	.release =	&fileobj_release,
+	.free =	&fileobj_free,
-	.ref =		&fileobj_ref,
 	.get_page =	&fileobj_get_page,
 	.copy_page =	NULL,
 	.flush_page =	&fileobj_flush_page,
@@ -170,22 +165,22 @@ static void obj_list_remove(struct fileobj *obj)
 /* return NULL or locked fileobj */
 static struct fileobj *obj_list_lookup(uintptr_t handle)
 {
-	struct fileobj *obj;
 	struct fileobj *p;
 
-	obj = NULL;
 	list_for_each_entry(p, &fileobj_list, list) {
 		if (p->handle == handle) {
-			memobj_lock(&p->memobj);
+			/* for the interval between last put and fileobj_free
-			if (p->cref > 0) {
+			 * taking list_lock
-				obj = p;
+			 */
-				break;
+			if (memobj_ref(&p->memobj) <= 1) {
+				ihk_atomic_dec(&p->memobj.refcnt);
+				continue;
 			}
-			memobj_unlock(&p->memobj);
+			return p;
 		}
 	}
 
-	return obj;
+	return NULL;
 }
 
 /***********************************************************************
@@ -200,13 +195,7 @@ int fileobj_create(int fd, struct memobj **objp, int *maxprotp, uintptr_t virt_a
 	struct fileobj *obj;
 	struct mcs_lock_node node;
 
-	dkprintf("fileobj_create(%d)\n", fd);
+	dkprintf("%s(%d)\n", __func__, fd);
-	newobj = kmalloc(sizeof(*newobj), IHK_MC_AP_NOWAIT);
-	if (!newobj) {
-		error = -ENOMEM;
-		kprintf("fileobj_create(%d):kmalloc failed. %d\n", fd, error);
-		goto out;
-	}
 
 	ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_CREATE;
 	ihk_mc_syscall_arg1(&ctx) = fd;
@@ -214,20 +203,41 @@ int fileobj_create(int fd, struct memobj **objp, int *maxprotp, uintptr_t virt_a
 	memset(&result, 0, sizeof(result));
 
 	error = syscall_generic_forwarding(__NR_mmap, &ctx);
+
 	if (error) {
-		dkprintf("fileobj_create(%d):create failed. %d\n", fd, error);
+		/* -ESRCH doesn't mean an error but requesting a fall
+		 * back to treat the file as a device file
+		 */
+		if (error != -ESRCH) {
+			kprintf("%s(%d):create failed. %d\n",
+				__func__, fd, error);
+		}
 		goto out;
 	}
 
+	if (result.flags & MF_HUGETLBFS) {
+		return hugefileobj_pre_create(&result, objp, maxprotp);
+	}
+
+	mcs_lock_lock(&fileobj_list_lock, &node);
+	obj = obj_list_lookup(result.handle);
+	if (obj)
+		goto found;
+	mcs_lock_unlock(&fileobj_list_lock, &node);
+
+	// not found: alloc new object and lookup again
+	newobj = kmalloc(sizeof(*newobj), IHK_MC_AP_NOWAIT);
+	if (!newobj) {
+		error = -ENOMEM;
+		kprintf("%s(%d):kmalloc failed. %d\n", __func__, fd, error);
+		goto out;
+	}
 	memset(newobj, 0, sizeof(*newobj));
 	newobj->memobj.ops = &fileobj_ops;
 	newobj->memobj.flags = MF_HAS_PAGER | MF_REG_FILE;
 	newobj->handle = result.handle;
 
-	newobj->sref = 1;
-	newobj->cref = 1;
 	fileobj_page_hash_init(newobj);
-	ihk_mc_spinlock_init(&newobj->memobj.lock);
 
 	mcs_lock_lock_noirq(&fileobj_list_lock, &node);
 	obj = obj_list_lookup(result.handle);
@@ -237,6 +247,8 @@ int fileobj_create(int fd, struct memobj **objp, int *maxprotp, uintptr_t virt_a
 		to_memobj(obj)->size = result.size;
 		to_memobj(obj)->flags |= result.flags;
 		to_memobj(obj)->status = MEMOBJ_READY;
+		ihk_atomic_set(&to_memobj(obj)->refcnt, 1);
+		obj->sref = 1;
 		if (to_memobj(obj)->flags & MF_PREFETCH) {
 			to_memobj(obj)->status = MEMOBJ_TO_BE_PREFETCHED;
 		}
@@ -305,20 +317,17 @@ error_cleanup:
 		}
 
 		newobj = NULL;
-		dkprintf("%s: new obj 0x%lx cref: %d, %s\n",
+		dkprintf("%s: new obj 0x%lx %s\n",
 			__FUNCTION__,
 			obj,
-			obj->cref,
 			to_memobj(obj)->flags & MF_ZEROFILL ? "zerofill" : "");
 	}
 	else {
-		++obj->sref;
+found:
-		++obj->cref;
+		obj->sref++;
-		memobj_unlock(&obj->memobj);	/* locked by obj_list_lookup() */
+		dkprintf("%s: existing obj 0x%lx, %s\n",
-		dkprintf("%s: existing obj 0x%lx cref: %d, %s\n",
 			__FUNCTION__,
 			obj,
-			obj->cref,
 			to_memobj(obj)->flags & MF_ZEROFILL ? "zerofill" : "");
 	}
 
@@ -332,152 +341,111 @@ out:
 	if (newobj) {
 		kfree(newobj);
 	}
-	dkprintf("fileobj_create(%d):%d %p %x\n", fd, error, *objp, *maxprotp);
+	dkprintf("%s(%d):%d %p %x\n", __func__, fd, error, *objp, *maxprotp);
 	return error;
 }
 
-static void fileobj_ref(struct memobj *memobj)
+static void fileobj_free(struct memobj *memobj)
 {
 	struct fileobj *obj = to_fileobj(memobj);
-
-	dkprintf("fileobj_ref(%p %lx):\n", obj, obj->handle);
-	memobj_lock(&obj->memobj);
-	++obj->cref;
-	memobj_unlock(&obj->memobj);
-	return;
-}
-
-static void fileobj_release(struct memobj *memobj)
-{
-	struct fileobj *obj = to_fileobj(memobj);
-	long free_sref = 0;
-	uintptr_t free_handle;
-	struct fileobj *free_obj = NULL;
 	struct mcs_lock_node node;
+	int error;
+	ihk_mc_user_context_t ctx;
 
-	dkprintf("fileobj_release(%p %lx)\n", obj, obj->handle);
 
-	memobj_lock(&obj->memobj);
+	dkprintf("%s: free obj 0x%lx, %s\n", __func__,
-	--obj->cref;
+		 obj, to_memobj(obj)->flags & MF_ZEROFILL ? "zerofill" : "");
-	free_sref = obj->sref - 1;	/* surplus sref */
-	if (obj->cref <= 0) {
-		free_sref = obj->sref;
-		free_obj = obj;
-	}
-	obj->sref -= free_sref;
-	free_handle = obj->handle;
-	memobj_unlock(&obj->memobj);
-	if (obj->memobj.flags & MF_HOST_RELEASED) {
-		free_sref = 0; // don't call syscall_generic_forwarding
-	}
 
-	if (free_obj) {
+	mcs_lock_lock_noirq(&fileobj_list_lock, &node);
-		dkprintf("%s: release obj 0x%lx cref: %d, free_obj: 0x%lx, %s\n",
+	obj_list_remove(obj);
-				__FUNCTION__,
+	mcs_lock_unlock_noirq(&fileobj_list_lock, &node);
-				obj,
-				obj->cref,
-				free_obj,
-				to_memobj(obj)->flags & MF_ZEROFILL ? "zerofill" : "");
-		mcs_lock_lock_noirq(&fileobj_list_lock, &node);
-		/* zap page_list */
-		for (;;) {
-			struct page *page;
-			void *page_va;
-			uintptr_t phys;
 
-			page = fileobj_page_hash_first(obj);
+	/* zap page_list */
-			if (!page) {
+	for (;;) {
-				break;
+		struct page *page;
-			}
+		void *page_va;
-			__fileobj_page_hash_remove(page);
+		uintptr_t phys;
-			phys = page_to_phys(page);
-			page_va = phys_to_virt(phys);
 
-			/* Count must be one because set to one on the first get_page() invoking fileobj_do_pageio and
+		page = fileobj_page_hash_first(obj);
-			   incremented by the second get_page() reaping the pageio and decremented by clear_range().
+		if (!page) {
+			break;
+		}
+		__fileobj_page_hash_remove(page);
+		phys = page_to_phys(page);
+		page_va = phys_to_virt(phys);
+		/* Count must be one because set to one on the first
+		 * get_page() invoking fileobj_do_pageio and incremented by
+		 * the second get_page() reaping the pageio and decremented
+		 * by clear_range().
+		 */
+		if (ihk_atomic_read(&page->count) != 1) {
+			kprintf("%s: WARNING: page count is %d for phys 0x%lx is invalid, flags: 0x%lx\n",
+				__func__, ihk_atomic_read(&page->count),
+				page->phys, to_memobj(obj)->flags);
+		}
+		else if (page_unmap(page)) {
+			ihk_mc_free_pages_user(page_va, 1);
+			/* Track change in page->count for !MF_PREMAP pages.
+			 * It is decremented here or in clear_range()
 			 */
-			if (ihk_atomic_read(&page->count) != 1) {
+			dkprintf("%lx-,%s: calling memory_stat_rss_sub(),phys=%lx,size=%ld,pgsize=%ld\n",
-				kprintf("%s: WARNING: page count is %d for phys 0x%lx is invalid, flags: 0x%lx\n",
+				 phys, __func__, phys, PAGE_SIZE, PAGE_SIZE);
-						__FUNCTION__,
+			rusage_memory_stat_mapped_file_sub(PAGE_SIZE,
-					ihk_atomic_read(&page->count),
+							   PAGE_SIZE);
-					page->phys,
-					to_memobj(free_obj)->flags);
-			}
-			else if (page_unmap(page)) {
-				ihk_mc_free_pages_user(page_va, 1);
-				/* Track change in page->count for !MF_PREMAP pages. It is decremented here or in clear_range() */
-				dkprintf("%lx-,%s: calling memory_stat_rss_sub(),phys=%lx,size=%ld,pgsize=%ld\n", phys, __FUNCTION__, phys, PAGE_SIZE, PAGE_SIZE);
-				rusage_memory_stat_mapped_file_sub(PAGE_SIZE, PAGE_SIZE); 
-			}
-#if 0
-			count = ihk_atomic_sub_return(1, &page->count);
-
-			if (!((page->mode == PM_WILL_PAGEIO)
-					|| (page->mode == PM_DONE_PAGEIO)
-					|| (page->mode == PM_PAGEIO_EOF)
-					|| (page->mode == PM_PAGEIO_ERROR)
-					|| ((page->mode == PM_MAPPED)
-						&& (count <= 0)))) {
-				kprintf("fileobj_release(%p %lx): "
-					       "mode %x, count %d, off %lx\n",
-					       obj, obj->handle, page->mode,
-					       count, page->offset);
-				panic("fileobj_release");
-			}
-
-			page->mode = PM_NONE;
-#endif
-		}
-
-		/* Pre-mapped? */
-		if (to_memobj(free_obj)->flags & MF_PREMAP) {
-			int i;
-			for (i = 0; i < to_memobj(free_obj)->nr_pages; ++i) {
-				if (to_memobj(free_obj)->pages[i]) {
-					dkprintf("%s: pages[i]=%p\n", __FUNCTION__, i, to_memobj(free_obj)->pages[i]);
-					// Track change in fileobj->pages[] for MF_PREMAP pages
-					// Note that page_unmap() isn't called for MF_PREMAP in 
-					// free_process_memory_range() --> ihk_mc_pt_free_range()
-					dkprintf("%lx-,%s: memory_stat_rss_sub,phys=%lx,size=%ld,pgsize=%ld\n",
-							virt_to_phys(to_memobj(free_obj)->pages[i]), __FUNCTION__, virt_to_phys(to_memobj(free_obj)->pages[i]), PAGE_SIZE, PAGE_SIZE);
-					rusage_memory_stat_mapped_file_sub(PAGE_SIZE, PAGE_SIZE);
-					ihk_mc_free_pages_user(to_memobj(free_obj)->pages[i], 1);
-				}
-			}
-
-			kfree(to_memobj(free_obj)->pages);
-		}
-
-		if (to_memobj(free_obj)->path) {
-			dkprintf("%s: %s\n", __FUNCTION__, to_memobj(free_obj)->path);
-			kfree(to_memobj(free_obj)->path);
-		}
-
-		obj_list_remove(free_obj);
-		mcs_lock_unlock_noirq(&fileobj_list_lock, &node);
-		kfree(free_obj);
-	}
-
-	if (free_sref) {
-		int error;
-		ihk_mc_user_context_t ctx;
-
-		ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_RELEASE;
-		ihk_mc_syscall_arg1(&ctx) = free_handle;
-		ihk_mc_syscall_arg2(&ctx) = free_sref;
-
-		error = syscall_generic_forwarding(__NR_mmap, &ctx);
-		if (error) {
-			kprintf("fileobj_release(%p %lx):"
-					"release %ld failed. %d\n",
-					obj, free_handle, free_sref, error);
-			/* through */
 		}
 	}
 
-	dkprintf("fileobj_release(%p %lx):free %ld %p\n",
+	/* Pre-mapped? */
-			obj, free_handle, free_sref, free_obj);
+	if (to_memobj(obj)->flags & MF_PREMAP) {
+		int i;
+
+		for (i = 0; i < to_memobj(obj)->nr_pages; ++i) {
+			if (to_memobj(obj)->pages[i]) {
+				dkprintf("%s: pages[i]=%p\n", __func__, i,
+					 to_memobj(obj)->pages[i]);
+				// Track change in fileobj->pages[] for MF_PREMAP pages
+				// Note that page_unmap() isn't called for MF_PREMAP in
+				// free_process_memory_range() --> ihk_mc_pt_free_range()
+				dkprintf("%lx-,%s: memory_stat_rss_sub,phys=%lx,size=%ld,pgsize=%ld\n",
+					 virt_to_phys(to_memobj(obj)->pages[i]),
+					 __func__,
+					 virt_to_phys(to_memobj(obj)->pages[i]),
+					 PAGE_SIZE, PAGE_SIZE);
+				rusage_memory_stat_mapped_file_sub(PAGE_SIZE,
+								   PAGE_SIZE);
+				ihk_mc_free_pages_user(to_memobj(obj)->pages[i],
+						       1);
+			}
+		}
+
+		kfree(to_memobj(obj)->pages);
+	}
+
+	if (to_memobj(obj)->path) {
+		dkprintf("%s: %s\n", __func__, to_memobj(obj)->path);
+		kfree(to_memobj(obj)->path);
+	}
+
+	/* linux side
+	 * sref is necessary because handle is used as key, so there could
+	 * be a new mckernel pager with the same handle being created as
+	 * this one is being destroyed
+	 */
+	ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_RELEASE;
+	ihk_mc_syscall_arg1(&ctx) = obj->handle;
+	ihk_mc_syscall_arg2(&ctx) = obj->sref;
+
+	error = syscall_generic_forwarding(__NR_mmap, &ctx);
+	if (error) {
+		kprintf("%s(%p %lx): free failed. %d\n", __func__,
+			obj, obj->handle, error);
+		/* through */
+	}
+
+	dkprintf("%s(%p %lx):free\n", __func__, obj, obj->handle);
+	kfree(obj);
 	return;
+
 }
 
 struct pageio_args {
@@ -570,7 +538,7 @@ static void fileobj_do_pageio(void *args0)
 out:
 	mcs_lock_unlock_noirq(&obj->page_hash_locks[hash],
 			&mcs_node);
-	fileobj_release(&obj->memobj);		/* got fileobj_get_page() */
+	memobj_unref(&obj->memobj);		/* got fileobj_get_page() */
 	kfree(args0);
 	dkprintf("fileobj_do_pageio(%p,%lx,%lx):\n", obj, off, pgsize);
 	return;
@@ -656,7 +624,9 @@ static int fileobj_get_page(struct memobj *memobj, off_t off,
 			npages = 1 << p2align;
 
 			virt = ihk_mc_alloc_pages_user(npages, (IHK_MC_AP_NOWAIT |
-					(to_memobj(obj)->flags & MF_ZEROFILL) ? IHK_MC_AP_USER : 0), virt_addr);
+					((to_memobj(obj)->flags & MF_ZEROFILL) ?
+						IHK_MC_AP_USER : 0)),
+					virt_addr);
 			if (!virt) {
 				error = -ENOMEM;
 				kprintf("fileobj_get_page(%p,%lx,%x,%x,%p):"
@@ -681,9 +651,7 @@ static int fileobj_get_page(struct memobj *memobj, off_t off,
 			page->mode = PM_WILL_PAGEIO;
 		}
 
-		memobj_lock(&obj->memobj);
+		memobj_ref(&obj->memobj);
-		++obj->cref;	/* for fileobj_do_pageio() */
-		memobj_unlock(&obj->memobj);
 
 		args->fileobj = obj;
 		args->objoff = off;
@@ -744,10 +712,6 @@ static int fileobj_flush_page(struct memobj *memobj, uintptr_t phys,
 		return 0;
 	}
 
-	if (memobj->flags & MF_HOST_RELEASED) {
-		return 0;
-	}
-
 	page = phys_to_page(phys);
 	if (!page) {
 		kprintf("%s: warning: tried to flush non-existing page for phys addr: 0x%lx\n", 
@@ -755,8 +719,6 @@ static int fileobj_flush_page(struct memobj *memobj, uintptr_t phys,
 		return 0;
 	}
 
-	memobj_unlock(&obj->memobj);
-
 	ihk_mc_syscall_arg0(&ctx) = PAGER_REQ_WRITE;
 	ihk_mc_syscall_arg1(&ctx) = obj->handle;
 	ihk_mc_syscall_arg2(&ctx) = page->offset;
@@ -771,7 +733,6 @@ static int fileobj_flush_page(struct memobj *memobj, uintptr_t phys,
 		/* through */
 	}
 
-	memobj_lock(&obj->memobj);
 	return 0;
 }
 

kernel/futex.c

@@ -70,15 +70,22 @@
 #include <cls.h>
 #include <kmsg.h>
 #include <timer.h>
+#include <debug.h>
+#include <syscall.h>
 
 //#define DEBUG_PRINT_FUTEX
 
 #ifdef DEBUG_PRINT_FUTEX
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
+#define DDEBUG_DEFAULT DDEBUG_PRINT
+#define uti_dkprintf(...) do { ((clv_override && linux_printk) ? (*linux_printk) : kprintf)(__VA_ARGS__); } while (0)
 #else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
+#define uti_dkprintf(...) do { } while (0)
 #endif
+#define uti_kprintf(...) do { ((clv_override && linux_printk) ? (*linux_printk) : kprintf)(__VA_ARGS__); } while (0)
 
+
+unsigned long ihk_mc_get_ns_per_tsc(void);
 int futex_cmpxchg_enabled;
 
 /**
@@ -108,6 +115,9 @@ struct futex_q {
 	union futex_key key;
 	union futex_key *requeue_pi_key;
 	uint32_t bitset;
+
+	/* Used to wake-up a thread running on a Linux CPU */
+	void *uti_futex_resp; 
 };
 
 /*
@@ -180,11 +190,12 @@ static void drop_futex_key_refs(union futex_key *key)
  * lock_page() might sleep, the caller should not hold a spinlock.
  */
 static int
-get_futex_key(uint32_t *uaddr, int fshared, union futex_key *key)
+get_futex_key(uint32_t *uaddr, int fshared, union futex_key *key, struct cpu_local_var *clv_override)
 {
 	unsigned long address = (unsigned long)uaddr;
 	unsigned long phys;
-	struct process_vm *mm = cpu_local_var(current)->vm;
+	struct thread *thread = cpu_local_var_with_override(current, clv_override);
+	struct process_vm *mm = thread->vm;
 
 	/*
 	 * The futex address must be "naturally" aligned.
@@ -250,7 +261,7 @@ static int cmpxchg_futex_value_locked(uint32_t __user *uaddr, uint32_t uval, uin
  * The hash bucket lock must be held when this is called.
  * Afterwards, the futex_q must not be accessed.
  */
-static void wake_futex(struct futex_q *q)
+static void wake_futex(struct futex_q *q, struct cpu_local_var *clv_override)
 {
 	struct thread *p = q->task;
 
@@ -272,8 +283,31 @@ static void wake_futex(struct futex_q *q)
 	barrier();
 	q->lock_ptr = NULL;
 
-	dkprintf("wake_futex(): waking up tid %d\n", p->tid);
+
-	sched_wakeup_thread(p, PS_NORMAL);
+	if (q->uti_futex_resp) { 
+		int rc;
+		uti_dkprintf("wake_futex(): waking up migrated-to-Linux thread (tid %d),uti_futex_resp=%p\n", p->tid, q->uti_futex_resp);
+		/* TODO: Add the case when a Linux thread waking up another Linux thread */
+		if (clv_override) {
+			uti_dkprintf("%s: ERROR: A Linux thread is waking up migrated-to-Linux thread\n", __FUNCTION__);
+		}
+		if (p->spin_sleep == 0) {
+			uti_dkprintf("%s: INFO: woken up by someone else\n", __FUNCTION__);
+		}
+
+		struct ikc_scd_packet pckt;
+		struct ihk_ikc_channel_desc *resp_channel = cpu_local_var_with_override(ikc2linux, clv_override);
+		pckt.msg = SCD_MSG_FUTEX_WAKE;
+		pckt.futex.resp = q->uti_futex_resp;
+		pckt.futex.spin_sleep = &p->spin_sleep;
+		rc = ihk_ikc_send(resp_channel, &pckt, 0);
+		if (rc) {
+			uti_dkprintf("%s: ERROR: ihk_ikc_send returned %d, resp_channel=%p\n", __FUNCTION__, rc, resp_channel);
+		}
+	} else {
+		uti_dkprintf("wake_futex(): waking up McKernel thread (tid %d)\n", p->tid);
+		sched_wakeup_thread(p, PS_NORMAL);
+	}
 }
 
 /*
@@ -303,7 +337,7 @@ double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
 /*
  * Wake up waiters matching bitset queued on this futex (uaddr).
  */
-static int futex_wake(uint32_t *uaddr, int fshared, int nr_wake, uint32_t bitset)
+static int futex_wake(uint32_t *uaddr, int fshared, int nr_wake, uint32_t bitset, struct cpu_local_var *clv_override)
 {
 	struct futex_hash_bucket *hb;
 	struct futex_q *this, *next;
@@ -314,7 +348,7 @@ static int futex_wake(uint32_t *uaddr, int fshared, int nr_wake, uint32_t bitset
 	if (!bitset)
 		return -EINVAL;
 
-	ret = get_futex_key(uaddr, fshared, &key);
+	ret = get_futex_key(uaddr, fshared, &key, clv_override);
 	if ((ret != 0))
 		goto out;
 
@@ -330,7 +364,7 @@ static int futex_wake(uint32_t *uaddr, int fshared, int nr_wake, uint32_t bitset
 			if (!(this->bitset & bitset))
 				continue;
 
-			wake_futex(this);
+			wake_futex(this, clv_override);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -348,7 +382,8 @@ out:
  */
 static int
 futex_wake_op(uint32_t *uaddr1, int fshared, uint32_t *uaddr2,
-	      int nr_wake, int nr_wake2, int op)
+			  int nr_wake, int nr_wake2, int op,
+			  struct cpu_local_var *clv_override)
 {
 	union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
 	struct futex_hash_bucket *hb1, *hb2;
@@ -357,10 +392,10 @@ futex_wake_op(uint32_t *uaddr1, int fshared, uint32_t *uaddr2,
 	int ret, op_ret;
 
 retry:
-	ret = get_futex_key(uaddr1, fshared, &key1);
+	ret = get_futex_key(uaddr1, fshared, &key1, clv_override);
 	if ((ret != 0))
 		goto out;
-	ret = get_futex_key(uaddr2, fshared, &key2);
+	ret = get_futex_key(uaddr2, fshared, &key2, clv_override);
 	if ((ret != 0))
 		goto out_put_key1;
 
@@ -394,7 +429,7 @@ retry_private:
 
 	plist_for_each_entry_safe(this, next, head, list) {
 		if (match_futex (&this->key, &key1)) {
-			wake_futex(this);
+			wake_futex(this, clv_override);
 			if (++ret >= nr_wake)
 				break;
 		}
@@ -406,7 +441,7 @@ retry_private:
 		op_ret = 0;
 		plist_for_each_entry_safe(this, next, head, list) {
 			if (match_futex (&this->key, &key2)) {
-				wake_futex(this);
+				wake_futex(this, clv_override);
 				if (++op_ret >= nr_wake2)
 					break;
 			}
@@ -469,7 +504,7 @@ void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
  */
 static int futex_requeue(uint32_t *uaddr1, int fshared, uint32_t *uaddr2,
 			 int nr_wake, int nr_requeue, uint32_t *cmpval,
-			 int requeue_pi)
+						 int requeue_pi, struct cpu_local_var *clv_override)
 {
 	union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
 	int drop_count = 0, task_count = 0, ret;
@@ -477,10 +512,10 @@ static int futex_requeue(uint32_t *uaddr1, int fshared, uint32_t *uaddr2,
 	struct plist_head *head1;
 	struct futex_q *this, *next;
 
-	ret = get_futex_key(uaddr1, fshared, &key1);
+	ret = get_futex_key(uaddr1, fshared, &key1, clv_override);
 	if ((ret != 0))
 		goto out;
-	ret = get_futex_key(uaddr2, fshared, &key2);
+	ret = get_futex_key(uaddr2, fshared, &key2, clv_override);
 	if ((ret != 0))
 		goto out_put_key1;
 
@@ -515,7 +550,7 @@ static int futex_requeue(uint32_t *uaddr1, int fshared, uint32_t *uaddr2,
 		 */
 		/* RIKEN: no requeue_pi at this moment */
 		if (++task_count <= nr_wake) {
-			wake_futex(this);
+			wake_futex(this, clv_override);
 			continue;
 		}
 
@@ -574,7 +609,7 @@ queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
  * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
  * an example).
  */
-static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
+static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb, struct cpu_local_var *clv_override)
 {
 	int prio;
 
@@ -595,7 +630,7 @@ static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
 	q->list.plist.spinlock = &hb->lock;
 #endif
 	plist_add(&q->list, &hb->chain);
-	q->task = cpu_local_var(current);
+	q->task = cpu_local_var_with_override(current, clv_override);
 	ihk_mc_spinlock_unlock_noirq(&hb->lock);
 }
 
@@ -658,46 +693,64 @@ retry:
 /* RIKEN: this function has been rewritten so that it returns the remaining
  * time in case we are waken.
  */
-static uint64_t futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
+static int64_t futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
-				uint64_t timeout)
+				   uint64_t timeout, struct cpu_local_var *clv_override)
 {
-	uint64_t time_remain = 0;
+	int64_t time_remain = 0;
 	unsigned long irqstate;
-	struct thread *thread = cpu_local_var(current);
+	struct thread *thread = cpu_local_var_with_override(current, clv_override);
 	/*
 	 * The task state is guaranteed to be set before another task can
 	 * wake it. 
 	 * queue_me() calls spin_unlock() upon completion, serializing
 	 * access to the hash list and forcing a memory barrier.
 	 */
-	xchg4(&(cpu_local_var(current)->status), PS_INTERRUPTIBLE);
+	xchg4(&(thread->status), PS_INTERRUPTIBLE);
 
-	/* Indicate spin sleep */
+	/* Indicate spin sleep. Note that schedule_timeout() with
-	if (!idle_halt) {
+	 * idle_halt should use spin sleep because sleep with timeout
+	 * is not implemented.
+	 */
+	if (!idle_halt || timeout) {
 		irqstate = ihk_mc_spinlock_lock(&thread->spin_sleep_lock);
 		thread->spin_sleep = 1;
 		ihk_mc_spinlock_unlock(&thread->spin_sleep_lock, irqstate);
 	}
 
-	queue_me(q, hb);
+	queue_me(q, hb, clv_override);
 	
 	if (!plist_node_empty(&q->list)) {
+		if (clv_override) {
+			uti_dkprintf("%s: tid: %d is trying to sleep\n", __FUNCTION__, thread->tid);
+			/* Note that the unit of timeout is nsec */
+			time_remain = (*linux_wait_event)(q->uti_futex_resp, timeout);
+			
+			/* Note that time_remain == 0 indicates contidion evaluated to false after the timeout elapsed */
+			if (time_remain < 0) {
+				if (time_remain == -ERESTARTSYS) { /* Interrupted by signal */
+					uti_dkprintf("%s: DEBUG: wait_event returned -ERESTARTSYS\n", __FUNCTION__);
+				} else {
+					uti_kprintf("%s: ERROR: wait_event returned %d\n", __FUNCTION__, time_remain);
+				}
+			}
+			uti_dkprintf("%s: tid: %d woken up\n", __FUNCTION__, thread->tid);
+		} else {
 		
 		if (timeout) {
-			dkprintf("futex_wait_queue_me(): tid: %d schedule_timeout()\n", cpu_local_var(current)->tid);
+			dkprintf("futex_wait_queue_me(): tid: %d schedule_timeout()\n", thread->tid);
 			time_remain = schedule_timeout(timeout);
 		}
 		else {
-			dkprintf("futex_wait_queue_me(): tid: %d schedule()\n", cpu_local_var(current)->tid);
+			dkprintf("futex_wait_queue_me(): tid: %d schedule()\n", thread->tid);
 			spin_sleep_or_schedule();
 			time_remain = 0;
 		}
-		
+		dkprintf("futex_wait_queue_me(): tid: %d woken up\n", thread->tid);
-		dkprintf("futex_wait_queue_me(): tid: %d woken up\n", cpu_local_var(current)->tid);
+		}
 	}
 	
 	/* This does not need to be serialized */
-	cpu_local_var(current)->status = PS_RUNNING;
+	thread->status = PS_RUNNING;
 	thread->spin_sleep = 0;
 	
 	return time_remain;
@@ -721,7 +774,8 @@ static uint64_t futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q
  * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked
  */
 static int futex_wait_setup(uint32_t __user *uaddr, uint32_t val, int fshared,
-			   struct futex_q *q, struct futex_hash_bucket **hb)
+			    struct futex_q *q, struct futex_hash_bucket **hb,
+			    struct cpu_local_var *clv_override)
 {
 	uint32_t uval;
 	int ret;
@@ -744,7 +798,7 @@ static int futex_wait_setup(uint32_t __user *uaddr, uint32_t val, int fshared,
 	 * rare, but normal.
 	 */
 	q->key = FUTEX_KEY_INIT;
-	ret = get_futex_key(uaddr, fshared, &q->key);
+	ret = get_futex_key(uaddr, fshared, &q->key, clv_override);
 	if (ret != 0)
 		return ret;
 
@@ -768,49 +822,59 @@ static int futex_wait_setup(uint32_t __user *uaddr, uint32_t val, int fshared,
 }
 
 static int futex_wait(uint32_t __user *uaddr, int fshared,
-		      uint32_t val, uint64_t timeout, uint32_t bitset, int clockrt)
+		      uint32_t val, uint64_t timeout, uint32_t bitset, int clockrt,
+		      struct cpu_local_var *clv_override)
 {
 	struct futex_hash_bucket *hb;
 	struct futex_q q;
-	uint64_t time_remain;
+	int64_t time_remain;
 	int ret;
 
 	if (!bitset)
 		return -EINVAL;
 
 #ifdef PROFILE_ENABLE
-	if (cpu_local_var(current)->profile &&
+	if (cpu_local_var_with_override(current, clv_override)->profile &&
-			cpu_local_var(current)->profile_start_ts) {
+		cpu_local_var_with_override(current, clv_override)->profile_start_ts) {
-		cpu_local_var(current)->profile_elapsed_ts +=
+		cpu_local_var_with_override(current, clv_override)->profile_elapsed_ts +=
-			(rdtsc() - cpu_local_var(current)->profile_start_ts);
+			(rdtsc() - cpu_local_var_with_override(current, clv_override)->profile_start_ts);
-		cpu_local_var(current)->profile_start_ts = 0;
+		cpu_local_var_with_override(current, clv_override)->profile_start_ts = 0;
 	}
 #endif
 
 	q.bitset = bitset;
 	q.requeue_pi_key = NULL;
+	q.uti_futex_resp = cpu_local_var_with_override(uti_futex_resp, clv_override);
 
 retry:
 	/* Prepare to wait on uaddr. */
-	ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
+	ret = futex_wait_setup(uaddr, val, fshared, &q, &hb, clv_override);
-	if (ret)
+	if (ret) {
+		uti_dkprintf("%s: tid=%d futex_wait_setup returns zero, no need to sleep\n", __FUNCTION__, cpu_local_var_with_override(current, clv_override)->tid);
 		goto out;
+	}
 
 	/* queue_me and wait for wakeup, timeout, or a signal. */
-	time_remain = futex_wait_queue_me(hb, &q, timeout);
+	time_remain = futex_wait_queue_me(hb, &q, timeout, clv_override);
 
 	/* If we were woken (and unqueued), we succeeded, whatever. */
 	ret = 0;
-	if (!unqueue_me(&q))
+	if (!unqueue_me(&q)) {
+		uti_dkprintf("%s: tid=%d unqueued\n", __FUNCTION__, cpu_local_var_with_override(current, clv_override)->tid);
 		goto out_put_key;
+	}
 	ret = -ETIMEDOUT;
 
 	/* RIKEN: timer expired case (indicated by !time_remain) */
-	if (timeout && !time_remain)
+	if (timeout && !time_remain) {
+		uti_dkprintf("%s: tid=%d timer expired\n", __FUNCTION__, cpu_local_var_with_override(current, clv_override)->tid);
 		goto out_put_key;
+	}
 
-	if (hassigpending(cpu_local_var(current))) {
+	/* RIKEN: futex_wait_queue_me() returns -ERESTARTSYS when waiting on Linux CPU and woken up by signal */
+	if (hassigpending(cpu_local_var_with_override(current, clv_override)) || time_remain == -ERESTARTSYS) {
 		ret = -EINTR;
+		uti_dkprintf("%s: tid=%d woken up by signal\n", __FUNCTION__, cpu_local_var_with_override(current, clv_override)->tid);
 		goto out_put_key;
 	}
 
@@ -822,19 +886,22 @@ out_put_key:
 	put_futex_key(fshared, &q.key);
 out:
 #ifdef PROFILE_ENABLE
-	if (cpu_local_var(current)->profile) {
+	if (cpu_local_var_with_override(current, clv_override)->profile) {
-		cpu_local_var(current)->profile_start_ts = rdtsc();
+		cpu_local_var_with_override(current, clv_override)->profile_start_ts = rdtsc();
 	}
 #endif
 	return ret;
 }
 
 int futex(uint32_t *uaddr, int op, uint32_t val, uint64_t timeout,
-		uint32_t *uaddr2, uint32_t val2, uint32_t val3, int fshared)
+	  uint32_t *uaddr2, uint32_t val2, uint32_t val3, int fshared,
+	  struct cpu_local_var *clv_override)
 {
 	int clockrt, ret = -ENOSYS;
 	int cmd = op & FUTEX_CMD_MASK;
 
+	uti_dkprintf("%s: uaddr=%p, op=%x, val=%x, timeout=%ld, uaddr2=%p, val2=%x, val3=%x, fshared=%d, clv=%p\n", __FUNCTION__, uaddr, op, val, timeout, uaddr2, val2, val3, fshared, clv_override);
+
 	clockrt = op & FUTEX_CLOCK_REALTIME;
 	if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
 		return -ENOSYS;
@@ -843,21 +910,21 @@ int futex(uint32_t *uaddr, int op, uint32_t val, uint64_t timeout,
 	case FUTEX_WAIT:
 		val3 = FUTEX_BITSET_MATCH_ANY;
 	case FUTEX_WAIT_BITSET:
-		ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt);
+		ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt, clv_override);
 		break;
 	case FUTEX_WAKE:
 		val3 = FUTEX_BITSET_MATCH_ANY;
 	case FUTEX_WAKE_BITSET:
-		ret = futex_wake(uaddr, fshared, val, val3);
+		ret = futex_wake(uaddr, fshared, val, val3, clv_override);
 		break;
 	case FUTEX_REQUEUE:
-		ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0);
+		ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0, clv_override);
 		break;
 	case FUTEX_CMP_REQUEUE:
-		ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, 0);
+		ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, 0, clv_override);
 		break;
 	case FUTEX_WAKE_OP:
-		ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
+		ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3, clv_override);
 		break;
 	/* RIKEN: these calls are not supported for now.	
 	case FUTEX_LOCK_PI:

kernel/host.c

@@ -34,13 +34,13 @@
 #include <sysfs.h>
 #include <ihk/perfctr.h>
 #include <rusage_private.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_HOST
 
 #ifdef DEBUG_PRINT_HOST
-#define dkprintf kprintf
+#undef DDEBUG_DEFAULT
-#else
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
 #endif
 
 /* Linux channel table, indexec by Linux CPU id */
@@ -78,7 +78,6 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 	unsigned long args_envs_p, args_envs_rp;
 	unsigned long s, e, up;
 	char **argv;
-	char **a;
 	int i, n, argc, envc, args_envs_npages;
 	char **env;
 	int range_npages;
@@ -306,7 +305,7 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 
 	/* Only unmap remote address if it wasn't specified as an argument */
 	if (!args) {
-		ihk_mc_unmap_virtual(args_envs_r, args_envs_npages, 0);
+		ihk_mc_unmap_virtual(args_envs_r, args_envs_npages);
 		ihk_mc_unmap_memory(NULL, args_envs_rp, p->args_len);
 	}
 	flush_tlb();
@@ -341,7 +340,7 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 
 	/* Only map remote address if it wasn't specified as an argument */
 	if (!envs) {
-		ihk_mc_unmap_virtual(args_envs_r, args_envs_npages, 0);
+		ihk_mc_unmap_virtual(args_envs_r, args_envs_npages);
 		ihk_mc_unmap_memory(NULL, args_envs_rp, p->envs_len);
 	}
 	flush_tlb();
@@ -357,12 +356,13 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 		proc->saved_cmdline_len = 0;
 	}
 
-	proc->saved_cmdline = kmalloc(p->args_len, IHK_MC_AP_NOWAIT);
+	proc->saved_cmdline_len = p->args_len - ((argc + 2) * sizeof(char **));
+	proc->saved_cmdline = kmalloc(proc->saved_cmdline_len,
+				      IHK_MC_AP_NOWAIT);
 	if (!proc->saved_cmdline) {
 		goto err;
 	}
 
-	proc->saved_cmdline_len = p->args_len - ((argc + 2) * sizeof(char **));
 	memcpy(proc->saved_cmdline,
 			(char *)args_envs + ((argc + 2) * sizeof(char **)),
 			proc->saved_cmdline_len);
@@ -370,21 +370,18 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 			__FUNCTION__,
 			proc->saved_cmdline);
 
-	for (a = argv; *a; a++) {
+	for (i = 0; i < argc; i++) {
-		*a = (char *)addr + (unsigned long)*a; // Process' address space!
+		// Process' address space!
+		argv[i] = (char *)addr + (unsigned long)argv[i];
 	}
 
 	envc = *((long *)(args_envs + p->args_len));
 	dkprintf("envc: %d\n", envc);
 
 	env = (char **)(args_envs + p->args_len + sizeof(long));
-	while (*env) {
+	for (i = 0; i < envc; i++) {
-		char **_env = env;
+		env[i] = addr + p->args_len + env[i];
-		//dkprintf("%s\n", args_envs + p->args_len + (unsigned long)*env);
-		*env = (char *)addr + p->args_len + (unsigned long)*env;
-		env = ++_env;
 	}
-	env = (char **)(args_envs + p->args_len + sizeof(long));
 
 	dkprintf("env OK\n");
 
@@ -449,7 +446,7 @@ static int process_msg_prepare_process(unsigned long rphys)
 	if((pn = kmalloc(sizeof(struct program_load_desc) 
 					+ sizeof(struct program_image_section) * n,
 					IHK_MC_AP_NOWAIT)) == NULL){
-		ihk_mc_unmap_virtual(p, npages, 0);
+		ihk_mc_unmap_virtual(p, npages);
 		ihk_mc_unmap_memory(NULL, phys, sz);
 		return -ENOMEM;
 	}
@@ -460,7 +457,7 @@ static int process_msg_prepare_process(unsigned long rphys)
 					(unsigned long *)&p->cpu_set,
 					sizeof(p->cpu_set))) == NULL) {
 		kfree(pn);
-		ihk_mc_unmap_virtual(p, npages, 1);
+		ihk_mc_unmap_virtual(p, npages);
 		ihk_mc_unmap_memory(NULL, phys, sz);
 		return -ENOMEM;
 	}
@@ -482,6 +479,7 @@ static int process_msg_prepare_process(unsigned long rphys)
 	proc->mpol_flags = pn->mpol_flags;
 	proc->mpol_threshold = pn->mpol_threshold;
 	proc->nr_processes = pn->nr_processes;
+	proc->process_rank = pn->process_rank;
 	proc->heap_extension = pn->heap_extension;
 
 	/* Update NUMA binding policy if requested */
@@ -504,6 +502,9 @@ static int process_msg_prepare_process(unsigned long rphys)
 		vm->numa_mem_policy = MPOL_BIND;
 	}
 
+	proc->uti_thread_rank = pn->uti_thread_rank;
+	proc->uti_use_last_cpu = pn->uti_use_last_cpu;
+
 #ifdef PROFILE_ENABLE
 	proc->profile = pn->profile;
 	thread->profile = pn->profile;
@@ -542,14 +543,14 @@ static int process_msg_prepare_process(unsigned long rphys)
 
 	kfree(pn);
 
-	ihk_mc_unmap_virtual(p, npages, 1);
+	ihk_mc_unmap_virtual(p, npages);
 	ihk_mc_unmap_memory(NULL, phys, sz);
 	flush_tlb();
 
 	return 0;
 err:
 	kfree(pn);
-	ihk_mc_unmap_virtual(p, npages, 1);
+	ihk_mc_unmap_virtual(p, npages);
 	ihk_mc_unmap_memory(NULL, phys, sz);
 	destroy_thread(thread);
 	return -ENOMEM;
@@ -562,7 +563,6 @@ static void syscall_channel_send(struct ihk_ikc_channel_desc *c,
 }
 
 extern unsigned long do_kill(struct thread *, int, int, int, struct siginfo *, int ptracecont);
-extern void process_procfs_request(struct ikc_scd_packet *rpacket);
 extern void terminate_host(int pid);
 extern void debug_log(long);
 
@@ -573,7 +573,6 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 	struct ikc_scd_packet pckt;
 	struct ihk_ikc_channel_desc *resp_channel = cpu_local_var(ikc2linux);
 	int rc;
-	struct mcs_rwlock_node_irqsave lock;
 	struct thread *thread;
 	struct process *proc;
 	struct mcctrl_signal {
@@ -610,7 +609,7 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 	case SCD_MSG_SCHEDULE_PROCESS:
 		thread = (struct thread *)packet->arg;
 
-		cpuid = obtain_clone_cpuid(&thread->cpu_set);
+		cpuid = obtain_clone_cpuid(&thread->cpu_set, 0);
 		if (cpuid == -1) {
 			kprintf("No CPU available\n");
 			ret = -1;
@@ -634,14 +633,14 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 	 * the waiting thread
 	 */
 	case SCD_MSG_WAKE_UP_SYSCALL_THREAD:
-		thread = find_thread(0, packet->ttid, &lock);
+		thread = find_thread(0, packet->ttid);
 		if (!thread) {
 			kprintf("%s: WARNING: no thread for SCD reply? TID: %d\n",
 				__FUNCTION__, packet->ttid);
 			ret = -EINVAL;
 			break;
 		}
-		thread_unlock(thread, &lock);
+		thread_unlock(thread);
 
 		dkprintf("%s: SCD_MSG_WAKE_UP_SYSCALL_THREAD: waking up tid %d\n",
 			__FUNCTION__, packet->ttid);
@@ -653,7 +652,7 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		pp = ihk_mc_map_memory(NULL, packet->arg, sizeof(struct mcctrl_signal));
 		sp = (struct mcctrl_signal *)ihk_mc_map_virtual(pp, 1, PTATTR_WRITABLE | PTATTR_ACTIVE);
 		memcpy(&info, sp, sizeof(struct mcctrl_signal));
-		ihk_mc_unmap_virtual(sp, 1, 0);
+		ihk_mc_unmap_virtual(sp, 1);
 		ihk_mc_unmap_memory(NULL, pp, sizeof(struct mcctrl_signal));
 		pckt.msg = SCD_MSG_SEND_SIGNAL_ACK;
 		pckt.err = 0;
@@ -668,7 +667,14 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		break;
 
 	case SCD_MSG_PROCFS_REQUEST:
-		process_procfs_request(packet);
+	case SCD_MSG_PROCFS_RELEASE:
+		pckt.msg = SCD_MSG_PROCFS_ANSWER;
+		pckt.ref = packet->ref;
+		pckt.arg = packet->arg;
+		pckt.err = process_procfs_request(packet);
+		pckt.reply = packet->reply;
+		pckt.pid = packet->pid;
+		syscall_channel_send(resp_channel, &pckt);
 		ret = 0;
 		break;
 
@@ -705,17 +711,26 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 			if (!pcd->exclude_user) {
 				mode |= PERFCTR_USER_MODE;
 			}
-			ihk_mc_perfctr_init_raw(pcd->target_cntr, pcd->config, mode);
+
-			ihk_mc_perfctr_stop(1 << pcd->target_cntr);
+			ret = ihk_mc_perfctr_init_raw(pcd->target_cntr, pcd->config, mode);
-			ihk_mc_perfctr_reset(pcd->target_cntr);
+			if (ret != 0) {
+				break;
+			}
+
+			ret = ihk_mc_perfctr_stop(1 << pcd->target_cntr);
+			if (ret != 0) {
+				break;
+			}
+
+			ret = ihk_mc_perfctr_reset(pcd->target_cntr);
 			break;
 
 		case PERF_CTRL_ENABLE:
-			ihk_mc_perfctr_start(pcd->target_cntr_mask);
+			ret = ihk_mc_perfctr_start(pcd->target_cntr_mask);
 			break;
 			
 		case PERF_CTRL_DISABLE:
-			ihk_mc_perfctr_stop(pcd->target_cntr_mask);
+			ret = ihk_mc_perfctr_stop(pcd->target_cntr_mask);
 			break;
 
 		case PERF_CTRL_GET:
@@ -726,16 +741,15 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 			kprintf("%s: SCD_MSG_PERF_CTRL unexpected ctrl_type\n", __FUNCTION__);
 		}
 
-		ihk_mc_unmap_virtual(pcd, 1, 0);
+		ihk_mc_unmap_virtual(pcd, 1);
 		ihk_mc_unmap_memory(NULL, pp, sizeof(struct perf_ctrl_desc));
 
 		pckt.msg = SCD_MSG_PERF_ACK;
-		pckt.err = 0;
+		pckt.err = ret;
 		pckt.arg = packet->arg;
 		pckt.reply = packet->reply;
 		ihk_ikc_send(resp_channel, &pckt, 0);
 
-		ret = 0;
 		break;
 
 	case SCD_MSG_CPU_RW_REG:

kernel/hugefileobj.c

@@ -0,0 +1,303 @@
+#include <memobj.h>
+#include <ihk/mm.h>
+#include <kmsg.h>
+#include <kmalloc.h>
+#include <string.h>
+#include <debug.h>
+
+#if DEBUG_HUGEFILEOBJ
+#undef DDEBUG_DEFAULT
+#define DDEBUG_DEFAULT DDEBUG_PRINT
+#endif
+
+struct hugefilechunk {
+	struct list_head list;
+	off_t pgoff;
+	int npages;
+	void *mem;
+};
+
+struct hugefileobj {
+	struct memobj memobj;
+	size_t pgsize;
+	uintptr_t handle;
+	unsigned int pgshift;
+	struct list_head chunk_list;
+	ihk_spinlock_t chunk_lock;
+	struct list_head obj_list;
+};
+
+static ihk_spinlock_t hugefileobj_list_lock;
+static LIST_HEAD(hugefileobj_list);
+
+static struct hugefileobj *to_hugefileobj(struct memobj *memobj)
+{
+	return (struct hugefileobj *)memobj;
+}
+
+static struct memobj *to_memobj(struct hugefileobj *obj)
+{
+	return &obj->memobj;
+}
+
+static struct hugefileobj *hugefileobj_lookup(uintptr_t handle)
+{
+	struct hugefileobj *p;
+
+	list_for_each_entry(p, &hugefileobj_list, obj_list) {
+		if (p->handle == handle) {
+			/* for the interval between last put and fileobj_free
+			 * taking list_lock
+			 */
+			if (memobj_ref(&p->memobj) <= 1) {
+				ihk_atomic_dec(&p->memobj.refcnt);
+				continue;
+			}
+			return p;
+		}
+	}
+
+	return NULL;
+}
+
+static int hugefileobj_get_page(struct memobj *memobj, off_t off,
+				int p2align, uintptr_t *physp,
+				unsigned long *pflag, uintptr_t virt_addr)
+{
+	struct hugefileobj *obj = to_hugefileobj(memobj);
+	struct hugefilechunk *chunk;
+	off_t pgoff;
+
+	if (p2align != obj->pgshift - PTL1_SHIFT) {
+		kprintf("%s: p2align %d but expected %d\n",
+			__func__, p2align, obj->pgshift - PTL1_SHIFT);
+		return -ENOMEM;
+	}
+
+	pgoff = off >> obj->pgshift;
+	ihk_mc_spinlock_lock_noirq(&obj->chunk_lock);
+	list_for_each_entry(chunk, &obj->chunk_list, list) {
+		if (pgoff >= chunk->pgoff + chunk->npages)
+			continue;
+		if (pgoff >= chunk->pgoff)
+			break;
+		kprintf("%s: no segment found for pgoff %lx (obj %p)\n",
+			__func__, pgoff, obj);
+		chunk = NULL;
+		break;
+	}
+	ihk_mc_spinlock_unlock_noirq(&obj->chunk_lock);
+	if (!chunk)
+		return -EIO;
+
+	*physp = virt_to_phys(chunk->mem + (off - chunk->pgoff * PAGE_SIZE));
+
+	return 0;
+}
+
+static void hugefileobj_free(struct memobj *memobj)
+{
+	struct hugefileobj *obj = to_hugefileobj(memobj);
+	struct hugefilechunk *chunk, *next;
+
+	dkprintf("Destroying hugefileobj %p\n", memobj);
+
+	ihk_mc_spinlock_lock_noirq(&hugefileobj_list_lock);
+	list_del(&obj->obj_list);
+	ihk_mc_spinlock_unlock_noirq(&hugefileobj_list_lock);
+
+	kfree(memobj->path);
+	/* don't bother with chunk_lock, memobj refcounting makes this safe */
+	list_for_each_entry_safe(chunk, next, &obj->chunk_list, list) {
+		ihk_mc_free_pages_user(chunk->mem, chunk->npages);
+		kfree(chunk);
+	}
+	kfree(memobj);
+}
+
+struct memobj_ops hugefileobj_ops = {
+	.free = hugefileobj_free,
+	.get_page = hugefileobj_get_page,
+
+};
+
+void hugefileobj_cleanup(void)
+{
+	struct hugefileobj *obj;
+	int refcnt;
+
+	while (true) {
+		ihk_mc_spinlock_lock_noirq(&hugefileobj_list_lock);
+		if (list_empty(&hugefileobj_list)) {
+			ihk_mc_spinlock_unlock_noirq(&hugefileobj_list_lock);
+			break;
+		}
+		obj = list_first_entry(&hugefileobj_list, struct hugefileobj,
+				       obj_list);
+		ihk_mc_spinlock_unlock_noirq(&hugefileobj_list_lock);
+
+		if ((refcnt = memobj_unref(to_memobj(obj))) != 0) {
+			kprintf("%s: obj %p had refcnt %ld > 1, destroying anyway\n",
+				__func__, obj, refcnt + 1);
+			hugefileobj_free(to_memobj(obj));
+		}
+	}
+}
+
+int hugefileobj_pre_create(struct pager_create_result *result,
+			   struct memobj **objp, int *maxprotp)
+{
+	struct hugefileobj *obj;
+
+	ihk_mc_spinlock_lock_noirq(&hugefileobj_list_lock);
+	obj = hugefileobj_lookup(result->handle);
+	if (obj)
+		goto out_unlock;
+
+	obj = kmalloc(sizeof(*obj), IHK_MC_AP_NOWAIT);
+	if (!obj)
+		return -ENOMEM;
+
+	obj->handle = result->handle;
+	obj->pgsize = result->size;
+	obj->pgshift = 0;
+	INIT_LIST_HEAD(&obj->chunk_list);
+	ihk_mc_spinlock_init(&obj->chunk_lock);
+	obj->memobj.flags = result->flags;
+	obj->memobj.status = MEMOBJ_TO_BE_PREFETCHED;
+	obj->memobj.ops = &hugefileobj_ops;
+	/* keep mapping around when process is gone */
+	ihk_atomic_set(&obj->memobj.refcnt, 2);
+	if (result->path[0]) {
+		obj->memobj.path = kmalloc(PATH_MAX, IHK_MC_AP_NOWAIT);
+		if (!obj->memobj.path) {
+			kfree(obj);
+			return -ENOMEM;
+		}
+		strncpy(obj->memobj.path, result->path, PATH_MAX);
+	}
+
+	list_add(&obj->obj_list, &hugefileobj_list);
+out_unlock:
+	ihk_mc_spinlock_unlock_noirq(&hugefileobj_list_lock);
+
+	*maxprotp = result->maxprot;
+	*objp = to_memobj(obj);
+
+	return 0;
+}
+
+int hugefileobj_create(struct memobj *memobj, size_t len, off_t off,
+		       int *pgshiftp, uintptr_t virt_addr)
+{
+	struct hugefileobj *obj = to_hugefileobj(memobj);
+	struct hugefilechunk *chunk = NULL, *old_chunk = NULL;
+	int p2align;
+	unsigned int pgshift;
+	int npages, npages_left;
+	void *v;
+	off_t pgoff, next_pgoff;
+	int error;
+
+	error = arch_get_smaller_page_size(NULL, obj->pgsize + 1, NULL,
+					   &p2align);
+	if (error)
+		return error;
+	pgshift = p2align + PTL1_SHIFT;
+	if (1 << pgshift != obj->pgsize) {
+		dkprintf("invalid hugefileobj pagesize: %d\n",
+			obj->pgsize);
+		return -EINVAL;
+	}
+
+	if (len & ((1 << pgshift) - 1)) {
+		dkprintf("invalid hugetlbfs mmap size %d (pagesize %d)\n",
+			len, 1 << pgshift);
+		obj->pgshift = 0;
+		return -EINVAL;
+	}
+	if (off & ((1 << pgshift) - 1)) {
+		dkprintf("invalid hugetlbfs mmap offset %d (pagesize %d)\n",
+			off, 1 << pgshift);
+		obj->pgshift = 0;
+		return -EINVAL;
+	}
+
+
+	ihk_mc_spinlock_lock_noirq(&obj->chunk_lock);
+	if (obj->pgshift && obj->pgshift != pgshift) {
+		kprintf("pgshift changed between two calls on same inode?! had %d now %d\n",
+			obj->pgshift, pgshift);
+		ihk_mc_spinlock_unlock_noirq(&obj->chunk_lock);
+		return -EINVAL;
+	}
+	obj->pgshift = pgshift;
+
+	/* Prealloc upfront, we need to fail here if not enough memory. */
+	if (!list_empty(&obj->chunk_list))
+		old_chunk = list_first_entry(&obj->chunk_list,
+					     struct hugefilechunk, list);
+	pgoff = off >> PAGE_SHIFT;
+	npages_left = len >> PAGE_SHIFT;
+	npages = npages_left;
+	while (npages_left) {
+		while (old_chunk &&
+				pgoff >= old_chunk->pgoff + old_chunk->npages) {
+			if (list_is_last(&old_chunk->list, &obj->chunk_list)) {
+				old_chunk = NULL;
+				break;
+			}
+			old_chunk = list_entry(old_chunk->list.next,
+					       struct hugefilechunk, list);
+		}
+		if (old_chunk) {
+			next_pgoff = old_chunk->pgoff + old_chunk->npages;
+			if (pgoff >= old_chunk->pgoff && pgoff < next_pgoff) {
+				npages_left -= next_pgoff - pgoff;
+				pgoff = next_pgoff;
+				continue;
+			}
+		}
+		if (!chunk) {
+			chunk = kmalloc(sizeof(*chunk), IHK_MC_AP_NOWAIT);
+		}
+		if (!chunk) {
+			kprintf("could not allocate hugefileobj chunk\n");
+			return -ENOMEM;
+		}
+		if (npages > npages_left)
+			npages = npages_left;
+		v = ihk_mc_alloc_aligned_pages_user(npages, p2align,
+				IHK_MC_AP_NOWAIT | IHK_MC_AP_USER, virt_addr);
+		if (!v) {
+			if (npages == 1) {
+				dkprintf("could not allocate more pages wth pgshift %d\n",
+					 pgshift);
+				kfree(chunk);
+				/* caller will cleanup the rest */
+				return -ENOMEM;
+			}
+			/* exponential backoff, try less aggressive? */
+			npages /= 2;
+			continue;
+		}
+		memset(v, 0, npages * PAGE_SIZE);
+		chunk->npages = npages;
+		chunk->mem = v;
+		chunk->pgoff = pgoff;
+		/* ordered list: insert before next (bigger) element */
+		if (old_chunk)
+			list_add(&chunk->list, old_chunk->list.prev);
+		else
+			list_add(&chunk->list, obj->chunk_list.prev);
+		pgoff += npages;
+		npages_left -= npages;
+	}
+	obj->memobj.size = len;
+
+	ihk_mc_spinlock_unlock_noirq(&obj->chunk_lock);
+
+	*pgshiftp = pgshift;
+
+	return 0;
+}

kernel/include/cls.h

@@ -21,7 +21,7 @@
 
 struct kmalloc_header {
 	unsigned int front_magic;
-	unsigned int cpu_id;
+	int cpu_id;
 	struct list_head list;
 	int size; /* The size of this chunk without the header */
 	unsigned int end_magic;
@@ -74,6 +74,7 @@ struct cpu_local_var {
 	struct thread *current;
 	struct list_head runq;
 	size_t runq_len;
+	size_t runq_reserved; /* Number of threads which are about to be added to runq */
 
 	struct ihk_ikc_channel_desc *ikc2linux;
 
@@ -99,6 +100,9 @@ struct cpu_local_var {
 	struct list_head smp_func_req_list;
 
 	struct process_vm *on_fork_vm;
+
+	/* UTI */
+	void *uti_futex_resp;
 } __attribute__((aligned(64)));
 
 
@@ -110,4 +114,6 @@ static struct cpu_local_var *get_this_cpu_local_var(void)
 
 #define cpu_local_var(name) get_this_cpu_local_var()->name
 
+#define cpu_local_var_with_override(name, clv_override) (clv_override ? clv_override->name : get_this_cpu_local_var()->name)
+
 #endif

kernel/include/debug.h

@@ -0,0 +1,54 @@
+#ifndef DEBUG_H
+#define DEBUG_H
+
+#include "lwk/compiler.h"
+
+void panic(const char *);
+
+/* when someone has a lot of time, add attribute __printf(1, 2) to kprintf */
+int kprintf(const char *format, ...);
+
+struct ddebug {
+	const char *file;
+	const char *func;
+	const char *fmt;
+	unsigned int line:24;
+	unsigned int flags:8;
+} __aligned(8);
+
+#define DDEBUG_NONE  0x0
+#define DDEBUG_PRINT 0x1
+
+#define DDEBUG_DEFAULT DDEBUG_NONE
+
+#define DDEBUG_SYMBOL()                                  \
+	static struct ddebug __aligned(8)                \
+	__attribute__((section("__verbose"))) ddebug = { \
+		.file = __FILE__,                        \
+		.func = __func__,                        \
+		.line = __LINE__,                        \
+		.flags = DDEBUG_DEFAULT,                 \
+	}
+
+#define DDEBUG_TEST ddebug.flags
+
+
+
+#define dkprintf(fmt, args...)        \
+do {                                  \
+	DDEBUG_SYMBOL();              \
+	if (DDEBUG_TEST)              \
+		kprintf(fmt, ##args); \
+} while (0)
+#define ekprintf(fmt, args...) kprintf(fmt, ##args)
+
+#define BUG_ON(condition) do {                         \
+	if (condition) {                               \
+		kprintf("PANIC: %s: %s(line:%d)\n",    \
+			__FILE__, __func__, __LINE__); \
+		panic("");                             \
+	}                                              \
+} while (0)
+#define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)]))
+
+#endif

kernel/include/futex.h

@@ -63,7 +63,7 @@
 #define FUTEX_OP_ANDN		3	/* *(int *)UADDR2 &= ~OPARG; */
 #define FUTEX_OP_XOR		4	/* *(int *)UADDR2 ^= OPARG; */
 
-#define FUTEX_OP_OPARG_SHIFT	8	/* Use (1 << OPARG) instead of OPARG.  */
+#define FUTEX_OP_OPARG_SHIFT	8U	/* Use (1 << OPARG) instead of OPARG. */
 
 #define FUTEX_OP_CMP_EQ		0	/* if (oldval == CMPARG) wake */
 #define FUTEX_OP_CMP_NE		1	/* if (oldval != CMPARG) wake */
@@ -150,6 +150,7 @@ union futex_key {
 
 extern int futex_init(void);
 
+struct cpu_local_var;
 extern int
 futex(
 	uint32_t __user *		uaddr,
@@ -159,7 +160,8 @@ futex(
 	uint32_t __user *		uaddr2,
 	uint32_t				val2,
 	uint32_t				val3,
-	int                     fshared
+	int                     fshared,
+	struct cpu_local_var *clv_override
 );
 
 

kernel/include/init.h

@@ -33,6 +33,7 @@ extern void cpu_sysfs_setup(void);
 extern void numa_sysfs_setup(void);
 extern void rusage_sysfs_setup(void);
 extern void status_sysfs_setup(void);
+extern void dynamic_debug_sysfs_setup(void);
 
 extern char *find_command_line(char *name);
 

kernel/include/kmalloc.h

@@ -13,11 +13,9 @@
 #ifndef __HEADER_KMALLOC_H
 #define __HEADER_KMALLOC_H
 
-#include <ihk/mm.h>
+#include "ihk/mm.h"
-#include <cls.h>
+#include "cls.h"
-
+#include "debug.h"
-void panic(const char *);
-int kprintf(const char *format, ...);
 
 #define kmalloc(size, flag) ({\
 void *r = _kmalloc(size, flag, __FILE__, __LINE__);\

kernel/include/lwk/compiler-gcc.h

@@ -12,11 +12,8 @@
 /* Optimization barrier */
 
 /* The "volatile" is due to gcc bugs */
-/* XXX: barrier is also defined in lib/include/ihk/cpu.h,
+#define barrier() __asm__ __volatile__("": : :"memory")
- * it would be cleaner to restore this here at some point, but we have
+
- * quite a few C files not including either this or kernel's compiler.h
- * #define barrier() __asm__ __volatile__("": : :"memory")
- */
 /*
  * This version is i.e. to prevent dead stores elimination on @ptr
  * where gcc and llvm may behave differently when otherwise using

kernel/include/lwk/compiler.h

@@ -3,6 +3,8 @@
 
 #ifndef __ASSEMBLY__
 
+#include <types.h>
+
 #ifdef __CHECKER__
 # define __user		__attribute__((noderef, address_space(1)))
 # define __kernel	__attribute__((address_space(0)))
@@ -175,11 +177,6 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 # define unlikely(x)	__builtin_expect(!!(x), 0)
 #endif
 
-/* Optimization barrier */
-#ifndef barrier
-# define barrier() __memory_barrier()
-#endif
-
 #ifndef barrier_data
 # define barrier_data(ptr) barrier()
 #endif
@@ -490,4 +487,62 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 	(_________p1); \
 })
 
+extern void *memcpy(void *dest, const void *src, size_t n);
+
+static __always_inline void __read_once_size(const volatile void *p, void *res, int size)
+{
+	switch (size) {
+	case 1: *(unsigned char *)res = *(volatile unsigned char *)p; break;
+	case 2: *(unsigned short *)res = *(volatile unsigned short *)p; break;
+	case 4: *(unsigned int *)res = *(volatile unsigned int *)p; break;
+	case 8: *(unsigned long long *)res = *(volatile unsigned long long *)p; break;
+	default:
+		barrier();
+		memcpy((void *)res, (const void *)p, size);
+		barrier();
+	}
+}
+
+static __always_inline void __write_once_size(volatile void *p, void *res, int size)
+{
+	switch (size) {
+	case 1: *(volatile unsigned char *)p = *(unsigned char *)res; break;
+	case 2: *(volatile unsigned short *)p = *(unsigned short *)res; break;
+	case 4: *(volatile unsigned int *)p = *(unsigned int *)res; break;
+	case 8: *(volatile unsigned long long *)p = *(unsigned long long *)res; break;
+	default:
+		barrier();
+		memcpy((void *)p, (const void *)res, size);
+		barrier();
+	}
+}
+
+/*
+ * Prevent the compiler from merging or refetching reads or writes. The
+ * compiler is also forbidden from reordering successive instances of
+ * READ_ONCE, WRITE_ONCE and ACCESS_ONCE (see below), but only when the
+ * compiler is aware of some particular ordering.  One way to make the
+ * compiler aware of ordering is to put the two invocations of READ_ONCE,
+ * WRITE_ONCE or ACCESS_ONCE() in different C statements.
+ *
+ * In contrast to ACCESS_ONCE these two macros will also work on aggregate
+ * data types like structs or unions. If the size of the accessed data
+ * type exceeds the word size of the machine (e.g., 32 bits or 64 bits)
+ * READ_ONCE() and WRITE_ONCE()  will fall back to memcpy and print a
+ * compile-time warning.
+ *
+ * Their two major use cases are: (1) Mediating communication between
+ * process-level code and irq/NMI handlers, all running on the same CPU,
+ * and (2) Ensuring that the compiler does not  fold, spindle, or otherwise
+ * mutilate accesses that either do not require ordering or that interact
+ * with an explicit memory barrier or atomic instruction that provides the
+ * required ordering.
+ */
+
+#define READ_ONCE(x) \
+	({ union { typeof(x) __val; char __c[1]; } __u; __read_once_size(&(x), __u.__c, sizeof(x)); __u.__val; })
+
+#define WRITE_ONCE(x, val) \
+	({ typeof(x) __val = (val); __write_once_size(&(x), &__val, sizeof(__val)); __val; })
+
 #endif /* __LWK_COMPILER_H */

kernel/include/lwk/futex.h

@@ -25,7 +25,7 @@
 #define FUTEX_OP_ANDN		3	/* *(int *)UADDR2 &= ~OPARG; */
 #define FUTEX_OP_XOR		4	/* *(int *)UADDR2 ^= OPARG; */
 
-#define FUTEX_OP_OPARG_SHIFT	8	/* Use (1 << OPARG) instead of OPARG.  */
+#define FUTEX_OP_OPARG_SHIFT	8U	/* Use (1 << OPARG) instead of OPARG. */
 
 #define FUTEX_OP_CMP_EQ		0	/* if (oldval == CMPARG) wake */
 #define FUTEX_OP_CMP_NE		1	/* if (oldval != CMPARG) wake */

kernel/include/memobj.h

@@ -19,6 +19,7 @@
 #include <ihk/lock.h>
 #include <errno.h>
 #include <list.h>
+#include <pager.h>
 
 #ifdef POSTK_DEBUG_ARCH_DEP_18 /* coredump arch separation. */
 #else /* POSTK_DEBUG_ARCH_DEP_18 */
@@ -44,8 +45,7 @@ enum {
 	MF_XPMEM   = 0x10000, /* To identify XPMEM attachment pages for rusage accounting */
 	MF_ZEROOBJ = 0x20000, /* To identify pages of anonymous, on-demand paging ranges for rusage accounting */
 	MF_SHM =     0x40000,
-	MF_HOST_RELEASED = 0x80000000,
+	MF_HUGETLBFS = 0x100000,
-	MF_END
 };
 
 #define MEMOBJ_READY              0
@@ -56,7 +56,7 @@ struct memobj {
 	uint32_t flags;
 	uint32_t status;
 	size_t size;
-	ihk_spinlock_t lock;
+	ihk_atomic_t refcnt;
 
 	/* For pre-mapped memobjects */
 	void **pages;
@@ -64,8 +64,7 @@ struct memobj {
 	char *path;
 };
 
-typedef void memobj_release_func_t(struct memobj *obj);
+typedef void memobj_free_func_t(struct memobj *obj);
-typedef void memobj_ref_func_t(struct memobj *obj);
 typedef int memobj_get_page_func_t(struct memobj *obj, off_t off, int p2align, uintptr_t *physp, unsigned long *flag, uintptr_t virt_addr);
 typedef uintptr_t memobj_copy_page_func_t(struct memobj *obj, uintptr_t orgphys, int p2align);
 typedef int memobj_flush_page_func_t(struct memobj *obj, uintptr_t phys, size_t pgsize);
@@ -73,27 +72,28 @@ typedef int memobj_invalidate_page_func_t(struct memobj *obj, uintptr_t phys, si
 typedef int memobj_lookup_page_func_t(struct memobj *obj, off_t off, int p2align, uintptr_t *physp, unsigned long *flag);
 
 struct memobj_ops {
-	memobj_release_func_t *		release;
+	memobj_free_func_t *free;
-	memobj_ref_func_t *		ref;
+	memobj_get_page_func_t *get_page;
-	memobj_get_page_func_t *	get_page;
+	memobj_copy_page_func_t *copy_page;
-	memobj_copy_page_func_t *	copy_page;
+	memobj_flush_page_func_t *flush_page;
-	memobj_flush_page_func_t *	flush_page;
+	memobj_invalidate_page_func_t *invalidate_page;
-	memobj_invalidate_page_func_t *	invalidate_page;
+	memobj_lookup_page_func_t *lookup_page;
-	memobj_lookup_page_func_t *	lookup_page;
 };
 
-static inline void memobj_release(struct memobj *obj)
+static inline int memobj_ref(struct memobj *obj)
 {
-	if (obj->ops->release) {
+	return ihk_atomic_inc_return(&obj->refcnt);
-		(*obj->ops->release)(obj);
-	}
 }
 
-static inline void memobj_ref(struct memobj *obj)
+static inline int memobj_unref(struct memobj *obj)
 {
-	if (obj->ops->ref) {
+	int cnt;
-		(*obj->ops->ref)(obj);
+
+	if ((cnt = ihk_atomic_dec_return(&obj->refcnt)) == 0) {
+		(*obj->ops->free)(obj);
 	}
+
+	return cnt;
 }
 
 static inline int memobj_get_page(struct memobj *obj, off_t off,
@@ -140,16 +140,6 @@ static inline int memobj_lookup_page(struct memobj *obj, off_t off,
 	return -ENXIO;
 }
 
-static inline void memobj_lock(struct memobj *obj)
-{
-	ihk_mc_spinlock_lock_noirq(&obj->lock);
-}
-
-static inline void memobj_unlock(struct memobj *obj)
-{
-	ihk_mc_spinlock_unlock_noirq(&obj->lock);
-}
-
 static inline int memobj_has_pager(struct memobj *obj)
 {
 	return !!(obj->flags & MF_HAS_PAGER);
@@ -166,5 +156,10 @@ int shmobj_create(struct shmid_ds *ds, struct memobj **objp);
 int zeroobj_create(struct memobj **objp);
 int devobj_create(int fd, size_t len, off_t off, struct memobj **objp, int *maxprotp,
 	int prot, int populate_flags);
+int hugefileobj_pre_create(struct pager_create_result *result,
+			   struct memobj **objp, int *maxprotp);
+int hugefileobj_create(struct memobj *obj, size_t len, off_t off,
+		       int *pgshiftp, uintptr_t virt_addr);
+void hugefileobj_cleanup(void);
 
 #endif /* HEADER_MEMOBJ_H */

kernel/include/process.h

@@ -70,10 +70,8 @@
 #define PS_TRACED            0x40 /* Set to "not running" by a ptrace related event */
 #define PS_STOPPING          0x80
 #define PS_TRACING           0x100
-#ifdef POSTK_DEBUG_TEMP_FIX_41 /* early to wait4() wakeup for ptrace, fix. */
 #define PS_DELAY_STOPPED     0x200
 #define PS_DELAY_TRACED      0x400
-#endif /* POSTK_DEBUG_TEMP_FIX_41 */
 
 #define PS_NORMAL	(PS_INTERRUPTIBLE | PS_UNINTERRUPTIBLE)
 
@@ -244,6 +242,11 @@ enum mpol_rebind_step {
 #define SPAWN_TO_REMOTE 1
 #define SPAWNING_TO_REMOTE 1001
 
+#define UTI_STATE_DEAD 0
+#define UTI_STATE_PROLOGUE 1
+#define UTI_STATE_RUNNING_IN_LINUX 2
+#define UTI_STATE_EPILOGUE 3
+
 #include <waitq.h>
 #include <futex.h>
 
@@ -277,6 +280,7 @@ extern struct list_head	resource_set_list;
 extern mcs_rwlock_lock_t	resource_set_lock;
 extern int idle_halt;
 extern int allow_oversubscribe;
+extern ihk_spinlock_t runq_reservation_lock; /* mutex for cpuid reservation (clv->runq_reserved) */
 
 struct process_hash {
 	struct list_head	list[HASH_SIZE];
@@ -460,6 +464,14 @@ struct process {
 
 	// threads and children
 	struct list_head threads_list;
+	struct list_head report_threads_list;
+
+	/*
+	 * main_thread is used to refer to thread information using process ID.
+	 * 1) signal related state in signal_flags
+	 * 2) status of trace
+	 */
+	struct thread *main_thread;
 	mcs_rwlock_lock_t threads_lock; // lock for threads_list
 	/* TID set of proxy process */
 	struct mcexec_tid *tids;
@@ -488,7 +500,6 @@ struct process {
 			// V       +----   |
 			// PS_STOPPED -----+
 			// (PS_TRACED)
-	unsigned long exit_status; // only for zombie
 
 	/* Store exit_status for a group of threads when stopped by SIGSTOP.
 	   exit_status can't be used because values of exit_status of threads
@@ -520,22 +531,6 @@ struct process {
 	long saved_cmdline_len;
 	cpu_set_t cpu_set;
 
-	/* Store ptrace flags.
-	 * The lower 8 bits are PTRACE_O_xxx of the PTRACE_SETOPTIONS request.
-	 * Other bits are for inner use of the McKernel.
-	 */
-	int ptrace;
-
-	/* Store ptrace event message.
-	 * PTRACE_O_xxx will store event message here.
-	 * PTRACE_GETEVENTMSG will get from here.
-	 */
-	unsigned long ptrace_eventmsg;
-
-	/* Store event related to signal. For example, 
-	   it represents that the proceess has been resumed by SIGCONT. */
-	int signal_flags;
-
 	/* Store signal sent to parent when the process terminates. */
 	int termsig;
 
@@ -557,6 +552,9 @@ struct process {
 	size_t mpol_threshold;
 	unsigned long heap_extension;
 	unsigned long mpol_bind_mask;
+	int uti_thread_rank; /* Spawn on Linux CPU when clone_count reaches this */
+	int uti_use_last_cpu; /* Work-around not to share CPU with OpenMP thread */
+	int clone_count;
 
 	// perf_event
 	int perf_status;
@@ -572,6 +570,7 @@ struct process {
 	unsigned long profile_elapsed_ts;
 #endif // PROFILE_ENABLE
 	int nr_processes; /* For partitioned execution */
+	int process_rank; /* Rank in partition */
 };
 
 /*
@@ -602,7 +601,7 @@ struct thread {
 	// thread info
 	int cpu_id;
 	int tid;
-	int status;	// PS_RUNNING -> PS_EXITED
+	int status;	// PS_RUNNING -> PS_EXITED (-> ZOMBIE / ptrace)
 			// |       ^       ^
 			// |       |       |
 			// V       |       |
@@ -612,6 +611,14 @@ struct thread {
 			// PS_UNINTERRUPTIBLE
 	int exit_status;
 
+	/*
+	 * Store event related to signal. For example,
+	 * it represents that the proceess has been resumed by SIGCONT.
+	 */
+	int signal_flags;
+
+	int termsig;
+
 	// process vm
 	struct process_vm *vm;
 
@@ -631,6 +638,22 @@ struct thread {
 	ihk_spinlock_t spin_sleep_lock;
 	int spin_sleep;
 
+	// for ptrace
+	struct process *report_proc;
+	struct list_head report_siblings_list; // lock process
+
+	/* Store ptrace flags.
+	 * The lower 8 bits are PTRACE_O_xxx of the PTRACE_SETOPTIONS request.
+	 * Other bits are for inner use of the McKernel.
+	 */
+	int ptrace;
+
+	/* Store ptrace event message.
+	 * PTRACE_O_xxx will store event message here.
+	 * PTRACE_GETEVENTMSG will get from here.
+	 */
+	unsigned long ptrace_eventmsg;
+
 	ihk_atomic_t refcount;
 
 	int	*clear_child_tid;
@@ -687,10 +710,11 @@ struct thread {
 	/* Syscall offload wait queue head */
 	struct waitq scd_wq;
 
-	int thread_offloaded;
+	int uti_state;
 	int mod_clone;
 	struct uti_attr *mod_clone_arg;
 	int parent_cpuid;
+	int uti_refill_tid;
 
 	// for performance counter
 	unsigned long pmc_alloc_map;
@@ -716,6 +740,8 @@ struct process_vm {
     // 2. addition of process page table (allocate_pages, update_process_page_table)
     // note that physical memory allocator (ihk_mc_alloc_pages, ihk_pagealloc_alloc)
     // is protected by its own lock (see ihk/manycore/generic/page_alloc.c)
+	unsigned long is_memory_range_lock_taken;
+	/* #986: Fix deadlock between do_page_fault_process_vm() and set_host_vma() */
 
 	ihk_atomic_t refcount;
 	int exiting;
@@ -819,14 +845,32 @@ void cpu_clear_and_set(int c_cpu, int s_cpu,
 
 void release_cpuid(int cpuid);
 
-struct thread *find_thread(int pid, int tid, struct mcs_rwlock_node_irqsave *lock);
+struct thread *find_thread(int pid, int tid);
-void thread_unlock(struct thread *thread, struct mcs_rwlock_node_irqsave *lock);
+void thread_unlock(struct thread *thread);
 struct process *find_process(int pid, struct mcs_rwlock_node_irqsave *lock);
 void process_unlock(struct process *proc, struct mcs_rwlock_node_irqsave *lock);
 void chain_process(struct process *);
 void chain_thread(struct thread *);
 void proc_init(void);
-void set_timer(void);
+void set_timer(int runq_locked);
 struct sig_pending *hassigpending(struct thread *thread);
+extern int do_signal(unsigned long rc, void *regs0, struct thread *thread,
+		     struct sig_pending *pending, int num);
+extern void check_signal(unsigned long rc, void *regs0, int num);
+extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig,
+			     struct siginfo *info, int ptracecont);
+extern void set_signal(int sig, void *regs, struct siginfo *info);
+extern void check_sig_pending(void);
+void clear_single_step(struct thread *thread);
+
+void release_fp_regs(struct thread *proc);
+void save_fp_regs(struct thread *proc);
+void copy_fp_regs(struct thread *from, struct thread *to);
+void restore_fp_regs(struct thread *proc);
+void clear_fp_regs(void);
+
+#define VERIFY_READ 0
+#define VERIFY_WRITE 1
+int access_ok(struct process_vm *vm, int type, uintptr_t addr, size_t len);
 
 #endif

kernel/include/rusage_private.h

@@ -10,6 +10,7 @@
 #include <rusage.h>
 #include <ihk/ihk_monitor.h>
 #include <arch_rusage.h>
+#include <debug.h>
 
 #ifdef ENABLE_RUSAGE
 
@@ -55,7 +56,7 @@ rusage_rss_add(unsigned long size)
 	}
 
 	vm->currss += size;
-	if (vm->currss > vm->proc->maxrss) {
+	if (vm->proc && vm->currss > vm->proc->maxrss) {
 		vm->proc->maxrss = vm->currss;
 	}
 }
@@ -118,8 +119,9 @@ static inline int rusage_memory_stat_add(struct vm_range *range, uintptr_t phys,
 	struct page *page = phys_to_page(phys);
 
 	/* Is It file map and cow page? */
-	if ((range->memobj->flags & (MF_DEV_FILE | MF_REG_FILE)) &&
+	if ((range->memobj->flags & (MF_DEV_FILE | MF_REG_FILE |
-		!page) {
+				     MF_HUGETLBFS)) &&
+	    !page) {
 		//kprintf("%s: cow,phys=%lx\n", __FUNCTION__, phys);
 		memory_stat_rss_add(size, pgsize);
 		return 1;

kernel/include/shm.h

@@ -57,6 +57,7 @@ struct shmobj {
 	struct shmlock_user *	user;
 	struct shmid_ds		ds;
 	struct list_head	page_list;
+	ihk_spinlock_t		page_list_lock;
 	struct list_head	chain;		/* shmobj_list */
 };
 
@@ -104,7 +105,6 @@ static inline void shmlock_users_unlock(void)
 void shmobj_list_lock(void);
 void shmobj_list_unlock(void);
 int shmobj_create_indexed(struct shmid_ds *ds, struct shmobj **objp);
-void shmobj_destroy(struct shmobj *obj);
 void shmlock_user_free(struct shmlock_user *user);
 int shmlock_user_get(uid_t ruid, struct shmlock_user **userp);
 

kernel/include/syscall.h

@@ -49,6 +49,7 @@
 #define	SCD_MSG_PROCFS_DELETE		0x11
 #define	SCD_MSG_PROCFS_REQUEST		0x12
 #define	SCD_MSG_PROCFS_ANSWER		0x13
+#define	SCD_MSG_PROCFS_RELEASE		0x15
 
 #define	SCD_MSG_DEBUG_LOG		0x20
 
@@ -82,6 +83,8 @@
 #define SCD_MSG_CPU_RW_REG              0x52
 #define SCD_MSG_CPU_RW_REG_RESP         0x53
 
+#define SCD_MSG_FUTEX_WAKE              0x60
+
 /* Cloning flags.  */
 # define CSIGNAL       0x000000ff /* Signal mask to be sent at exit.  */
 # define CLONE_VM      0x00000100 /* Set if VM shared between processes.  */
@@ -197,8 +200,10 @@ struct program_load_desc {
 	unsigned long heap_extension;
 	long stack_premap;
 	unsigned long mpol_bind_mask;
+	int uti_thread_rank; /* N-th clone() spawns a thread on Linux CPU */
+	int uti_use_last_cpu; /* Work-around not to share CPU with OpenMP thread */
 	int nr_processes;
-	char shell_path[SHELL_PATH_MAX_LEN];
+	int process_rank;
 	__cpu_set_unit cpu_set[PLD_CPU_SET_SIZE];
 	int profile;
 	struct program_image_section sections[0];
@@ -258,7 +263,7 @@ struct ikc_scd_packet {
 			long sysfs_arg3;
 		};
 
-		/* SCD_MSG_SCHEDULE_THREAD */
+		/* SCD_MSG_WAKE_UP_SYSCALL_THREAD */
 		struct {
 			int ttid;
 		};
@@ -274,6 +279,12 @@ struct ikc_scd_packet {
 		struct {
 			int eventfd_type;
 		};
+
+		/* SCD_MSG_FUTEX_WAKE */
+		struct {
+			void *resp;
+			int *spin_sleep; /* 1: waiting in linux_wait_event() 0: woken up by someone else */
+		} futex;
 	};
 	char padding[8];
 };
@@ -336,10 +347,10 @@ struct syscall_post {
 	SYSCALL_ARG_##a2(2); SYSCALL_ARG_##a3(3); \
 	SYSCALL_ARG_##a4(4); SYSCALL_ARG_##a5(5);
 
-#define SYSCALL_FOOTER return do_syscall(&request, ihk_mc_get_processor_id(), 0)
+#define SYSCALL_FOOTER return do_syscall(&request, ihk_mc_get_processor_id())
 
-extern long do_syscall(struct syscall_request *req, int cpu, int pid);
+extern long do_syscall(struct syscall_request *req, int cpu);
-int obtain_clone_cpuid(cpu_set_t *cpu_set);
+int obtain_clone_cpuid(cpu_set_t *cpu_set, int use_last);
 extern long syscall_generic_forwarding(int n, ihk_mc_user_context_t *ctx);
 
 #define DECLARATOR(number,name)		__NR_##name = number,
@@ -353,17 +364,10 @@ enum {
 #undef	SYSCALL_DELEGATED
 
 #define	__NR_coredump 999	/* pseudo syscall for coredump */
-#ifdef POSTK_DEBUG_TEMP_FIX_61 /* Core table size and lseek return value to loff_t */
 struct coretable {		/* table entry for a core chunk */
 	off_t len;		/* length of the chunk */
 	unsigned long addr;	/* physical addr of the chunk */
 };
-#else /* POSTK_DEBUG_TEMP_FIX_61 */
-struct coretable {		/* table entry for a core chunk */
-	int len;		/* length of the chunk */
-	unsigned long addr;	/* physical addr of the chunk */
-};
-#endif /* POSTK_DEBUG_TEMP_FIX_61 */
 
 #ifdef POSTK_DEBUG_TEMP_FIX_1
 void create_proc_procfs_files(int pid, int tid, int cpuid);
@@ -383,7 +387,6 @@ struct procfs_read {
 	int count;		/* bytes to read (request) */
 	int eof;		/* if eof is detected, 1 otherwise 0. (answer)*/
 	int ret;		/* read bytes (answer) */
-	int status;		/* non-zero if done (answer) */
 	int newcpu;		/* migrated new cpu (answer) */
 	int readwrite;		/* 0:read, 1:write */
 	char fname[PROCFS_NAME_MAX];	/* procfs filename (request) */
@@ -395,6 +398,8 @@ struct procfs_file {
 	char fname[PROCFS_NAME_MAX];	/* procfs filename (request) */
 };
 
+int process_procfs_request(struct ikc_scd_packet *rpacket);
+
 #define RUSAGE_SELF 0
 #define RUSAGE_CHILDREN -1
 #define RUSAGE_THREAD 1
@@ -459,8 +464,8 @@ static inline unsigned long timespec_to_jiffy(const struct timespec *ats)
 
 void reset_cputime(void);
 void set_cputime(int mode);
-int do_munmap(void *addr, size_t len);
+int do_munmap(void *addr, size_t len, int holding_memory_range_lock);
-intptr_t do_mmap(intptr_t addr0, size_t len0, int prot, int flags, int fd,
+intptr_t do_mmap(uintptr_t addr0, size_t len0, int prot, int flags, int fd,
 		off_t off0);
 void clear_host_pte(uintptr_t addr, size_t len);
 typedef int32_t key_t;
@@ -471,7 +476,16 @@ int arch_setup_vdso(void);
 int arch_cpu_read_write_register(struct ihk_os_cpu_register *desc,
 		enum mcctrl_os_cpu_operation op);
 struct vm_range_numa_policy *vm_range_policy_search(struct process_vm *vm, uintptr_t addr);
+void calculate_time_from_tsc(struct timespec *ts);
 time_t time(void);
+long do_futex(int n, unsigned long arg0, unsigned long arg1,
+			  unsigned long arg2, unsigned long arg3,
+			  unsigned long arg4, unsigned long arg5,
+			  unsigned long _uti_clv,
+			  void *uti_futex_resp,
+			  void *_linux_wait_event,
+			  void *_linux_printk,
+			  void *_linux_clock_gettime);
 
 #ifndef POSTK_DEBUG_ARCH_DEP_52
 #define VDSO_MAXPAGES 2
@@ -519,6 +533,7 @@ enum perf_ctrl_type {
 
 struct perf_ctrl_desc {
 	enum perf_ctrl_type ctrl_type;
+	int err;
 	union {
 		/* for SET, GET */
 		struct {
@@ -569,6 +584,15 @@ typedef struct uti_attr {
 	uint64_t flags; /* Representing location and behavior hints by bitmap */
 } uti_attr_t;
 
+struct uti_ctx {
+	union {
+		char ctx[4096];
+		struct {
+			int uti_refill_tid;
+		};
+	};
+}; 
+
 struct move_pages_smp_req {
 	unsigned long count;
 	const void **user_virt_addr;
@@ -589,4 +613,9 @@ struct move_pages_smp_req {
 #define PROCESS_VM_READ		0
 #define PROCESS_VM_WRITE	1
 
+/* uti: function pointers pointing to Linux codes */
+extern long (*linux_wait_event)(void *_resp, unsigned long nsec_timeout);
+extern int (*linux_printk)(const char *fmt, ...);
+extern int (*linux_clock_gettime)(clockid_t clk_id, struct timespec *tp);
+
 #endif

kernel/include/time.h

@@ -25,6 +25,8 @@
 #define CLOCK_PROCESS_CPUTIME_ID	2
 #define CLOCK_THREAD_CPUTIME_ID		3
 
+typedef int clockid_t;
+
 typedef long int __time_t;
 
 /* POSIX.1b structure for a time value.  This is like a `struct timeval' but

kernel/include/xpmem_private.h

@@ -26,20 +26,17 @@
 
 #include <mc_xpmem.h>
 #include <xpmem.h>
+#include <debug.h>
 
 #define XPMEM_CURRENT_VERSION           0x00026003
 
 //#define DEBUG_PRINT_XPMEM
 
 #ifdef DEBUG_PRINT_XPMEM
-#define dkprintf(...) kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#define XPMEM_DEBUG(format, a...) kprintf("[%d] %s: "format"\n", cpu_local_var(current)->proc->rgid, __func__, ##a)
-#else
-#define dkprintf(...) do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define ekprintf(...) kprintf(__VA_ARGS__)
-#define XPMEM_DEBUG(format, a...) do { if (0) kprintf("\n"); } while (0)
 #endif
+#define XPMEM_DEBUG(format, a...) dkprintf("[%d] %s: "format"\n", cpu_local_var(current)->proc->rgid, __func__, ##a)
 
 //#define USE_DBUG_ON
 

kernel/init.c

@@ -32,6 +32,7 @@
 #include <syscall.h>
 #include <sysfs.h>
 #include <ihk/monitor.h>
+#include <debug.h>
 
 //#define IOCTL_FUNC_EXTENSION
 #ifdef IOCTL_FUNC_EXTENSION
@@ -41,11 +42,8 @@
 //#define DEBUG_PRINT_INIT
 
 #ifdef DEBUG_PRINT_INIT
-#define dkprintf(...) do { kprintf(__VA_ARGS__); } while (0)
+#undef DDEBUG_DEFAULT
-#define ekprintf(...) do { kprintf(__VA_ARGS__); } while (0)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...) do { } while (0)
-#define ekprintf(...) do { kprintf(__VA_ARGS__); } while (0)
 #endif
 
 #define DUMP_LEVEL_USER_UNUSED_EXCLUDE 24
@@ -61,6 +59,13 @@ static void handler_init(void)
 	ihk_mc_set_syscall_handler(syscall);
 }
 
+
+/* Symbols with name conflict with the linux kernel
+ * Give the possibility to load all symbols at the same time
+ */
+int *mck_num_processors = &num_processors;
+
+
 unsigned long data[1024] __attribute__((aligned(64)));
 
 #ifdef USE_DMA
@@ -125,6 +130,8 @@ char *find_command_line(char *name)
 	return strstr(cmdline, name);
 }
 
+extern int safe_kernel_map;
+
 static void parse_kargs(void)
 {
 	char *ptr;
@@ -145,6 +152,11 @@ static void parse_kargs(void)
 	}
 	ihk_mc_set_dump_level(dump_level);
 
+	ptr = find_command_line("safe_kernel_map");
+	if (ptr) {
+		safe_kernel_map = 1;
+	}
+
 	/* idle_halt option */
 	ptr = find_command_line("idle_halt");
 	if (ptr) {
@@ -246,6 +258,11 @@ static void nmi_init()
 	ihk_set_nmi_mode_addr(phys);
 }
 
+static void uti_init()
+{
+	ihk_set_mckernel_do_futex((unsigned long)do_futex);
+}
+
 static void rest_init(void)
 {
 	handler_init();
@@ -261,6 +278,7 @@ static void rest_init(void)
 #endif /* !POSTK_DEBUG_TEMP_FIX_73 */
 	cpu_local_var_init();
 	nmi_init();
+	uti_init();
 	time_init();
 	kmalloc_init();
 
@@ -331,6 +349,7 @@ static void populate_sysfs(void)
 {
 	cpu_sysfs_setup();
 	numa_sysfs_setup();
+	dynamic_debug_sysfs_setup();
 	//setup_remote_snooping_samples();
 } /* populate_sysfs() */
 

kernel/listeners.c

@@ -19,15 +19,13 @@
 #include <ihk/ikc.h>
 #include <ikc/master.h>
 #include <arch/cpu.h>
+#include <debug.h>
 
 //#define DEBUG_LISTENERS
 
 #ifdef DEBUG_LISTENERS
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 void testmem(void *v, unsigned long size)
@@ -71,7 +69,7 @@ static int test_packet_handler(struct ihk_ikc_channel_desc *c,
 		
 		testmem(v, 4 * 1024 * 1024);
 
-		ihk_mc_unmap_virtual(v, 4 * 1024, 1);
+		ihk_mc_unmap_virtual(v, 4 * 1024);
 		ihk_mc_unmap_memory(NULL, pp, 4 * 1024 * 1024);
 	} else if (packet->msg == 0x11110012) {
 		p.msg = 0x11110013;

kernel/mem.c

@@ -44,15 +44,13 @@
 #include <process.h>
 #include <limits.h>
 #include <sysfs.h>
+#include <debug.h>
 
 //#define DEBUG_PRINT_MEM
 
 #ifdef DEBUG_PRINT_MEM
-#define	dkprintf(...)	kprintf(__VA_ARGS__)
+#undef DDEBUG_DEFAULT
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
+#define DDEBUG_DEFAULT DDEBUG_PRINT
-#else
-#define dkprintf(...)	do { if (0) kprintf(__VA_ARGS__); } while (0)
-#define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
 static unsigned long pa_start, pa_end;
@@ -547,7 +545,7 @@ static void *mckernel_allocate_aligned_pages_node(int npages, int p2align,
 		ihk_mc_ap_flag flag, int pref_node, int is_user, uintptr_t virt_addr)
 {
 	unsigned long pa = 0;
-	int i, node;
+	int i = 0, node;
 #ifndef IHK_RBTREE_ALLOCATOR
 	struct ihk_page_allocator_desc *pa_allocator;
 #endif
@@ -962,8 +960,6 @@ static struct ihk_mc_interrupt_handler query_free_mem_handler = {
 	.priv = NULL,
 };
 
-void set_signal(int sig, void *regs, struct siginfo *info);
-void check_signal(unsigned long, void *, int);
 int gencore(struct thread *, void *, struct coretable **, int *);
 void freecore(struct coretable **);
 
@@ -981,11 +977,9 @@ void coredump(struct thread *thread, void *regs)
 	struct coretable *coretable;
 	int chunks;
 
-#ifdef POSTK_DEBUG_ARCH_DEP_67 /* use limit corefile size. (temporarily fix.) */
 	if (thread->proc->rlimit[MCK_RLIMIT_CORE].rlim_cur == 0) {
 		return;
 	}
-#endif /* POSTK_DEBUG_ARCH_DEP_67 */
 
 #ifndef POSTK_DEBUG_ARCH_DEP_18
 	ret = gencore(thread, regs, &coretable, &chunks);
@@ -997,7 +991,7 @@ void coredump(struct thread *thread, void *regs)
 	request.args[0] = chunks;
 	request.args[1] = virt_to_phys(coretable);
 	/* no data for now */
-	ret = do_syscall(&request, thread->cpu_id, thread->proc->pid);
+	ret = do_syscall(&request, thread->cpu_id);
 	if (ret == 0) {
 		kprintf("dumped core.\n");
 	} else {
@@ -1223,7 +1217,7 @@ out:
 	if(interrupt_from_user(regs)){
 		cpu_enable_interrupt();
 		check_need_resched();
-		check_signal(0, regs, 0);
+		check_signal(0, regs, -1);
 	}
 	set_cputime(interrupt_from_user(regs)? 0: 1);
 #ifdef PROFILE_ENABLE
@@ -1671,7 +1665,7 @@ void *ihk_mc_map_virtual(unsigned long phys, int npages,
 	return (char *)p + offset;
 }
 
-void ihk_mc_unmap_virtual(void *va, int npages, int free_physical)
+void ihk_mc_unmap_virtual(void *va, int npages)
 {
 	unsigned long i;
 
@@ -1681,13 +1675,7 @@ void ihk_mc_unmap_virtual(void *va, int npages, int free_physical)
 		flush_tlb_single((unsigned long)(va + (i << PAGE_SHIFT)));
 	}
 
-#ifdef POSTK_DEBUG_TEMP_FIX_51 /* ihk_mc_unmap_virtual() free_physical disabled */
 	ihk_pagealloc_free(vmap_allocator, (unsigned long)va, npages);
-#else /* POSTK_DEBUG_TEMP_FIX_51 */
-	if (free_physical) {
-		ihk_pagealloc_free(vmap_allocator, (unsigned long)va, npages);
-	}
-#endif /* POSTK_DEBUG_TEMP_FIX_51 */
 }
 
 #ifdef ATTACHED_MIC
@@ -2304,76 +2292,37 @@ void ___kmalloc_print_free_list(struct list_head *list)
 	kprintf_unlock(irqflags);
 }
 
-#ifdef POSTK_DEBUG_ARCH_DEP_27
-int search_free_space(struct thread *thread, size_t len, intptr_t hint,
-		      int pgshift, intptr_t *addrp)
-{
-	struct vm_regions *region = &thread->vm->region;
-	intptr_t addr;
-	int error;
-	struct vm_range *range;
-	size_t pgsize = (size_t)1 << pgshift;
-
-	dkprintf("search_free_space(%lx,%lx,%d,%p)\n", len, hint, pgshift, addrp);
-
-	addr = hint;
-	for (;;) {
-		addr = (addr + pgsize - 1) & ~(pgsize - 1);
-		if ((region->user_end <= addr)
-				|| ((region->user_end - len) < addr)) {
-			ekprintf("search_free_space(%lx,%lx,%p):"
-					"no space. %lx %lx\n",
-					len, hint, addrp, addr,
-					region->user_end);
-			error = -ENOMEM;
-			goto out;
-		}
-
-		range = lookup_process_memory_range(thread->vm, addr, addr+len);
-		if (range == NULL) {
-			break;
-		}
-		addr = range->end;
-	}
-
-	error = 0;
-	*addrp = addr;
-
-out:
-	dkprintf("search_free_space(%lx,%lx,%d,%p): %d %lx\n",
-			len, hint, pgshift, addrp, error, addr);
-	return error;
-}
-#endif	/* POSTK_DEBUG_ARCH_DEP_27 */
-
-#ifdef POSTK_DEBUG_TEMP_FIX_52 /* supports NUMA for memory area determination */
 #ifdef IHK_RBTREE_ALLOCATOR
-int is_mckernel_memory(unsigned long phys)
+int is_mckernel_memory(unsigned long start, unsigned long end)
 {
 	int i;
 
 	for (i = 0; i < ihk_mc_get_nr_memory_chunks(); ++i) {
-		unsigned long start, end;
+		unsigned long chunk_start, chunk_end;
 		int numa_id;
 
-		ihk_mc_get_memory_chunk(i, &start, &end, &numa_id);
+		ihk_mc_get_memory_chunk(i, &chunk_start, &chunk_end, &numa_id);
-		if (start <= phys && phys < end) {
+		if ((chunk_start <= start && start < chunk_end) &&
+		    (chunk_start <= end && end < chunk_end)) {
 			return 1;
 		}
 	}
 	return 0;
 }
 #else /* IHK_RBTREE_ALLOCATOR */
-int is_mckernel_memory(unsigned long phys)
+int is_mckernel_memory(unsigned long start, unsigned long end)
 {
 	int i;
 
 	for (i = 0; i < ihk_mc_get_nr_numa_nodes(); ++i) {
 		struct ihk_page_allocator_desc *pa_allocator;
+		unsigned long area_start = pa_allocator->start;
+		unsigned long area_end = pa_allocator->end;
 
 		list_for_each_entry(pa_allocator,
 				    &memory_nodes[i].allocators, list) {
-			if (pa_allocator->start <= phys && phys < pa_allocator->end) {
+			if ((area_start <= start && start < area_end) &&
+			    (area_start <= end && end < area_end)) {
 				return 1;
 			}
 		}
@@ -2381,7 +2330,6 @@ int is_mckernel_memory(unsigned long phys)
 	return 0;
 }
 #endif /* IHK_RBTREE_ALLOCATOR */
-#endif /* POSTK_DEBUG_TEMP_FIX_52 */
 
 void ihk_mc_query_mem_areas(void){