Merge branch 'master' of postpeta.pccluster.org:mckernel

arch/x86/kernel/cpu.c

@@ -82,6 +82,7 @@ void x86_init_perfctr(void);
 int gettime_local_support = 0;
 
 extern int kprintf(const char *format, ...);
+extern int interrupt_from_user(void *);
 
 static struct idt_entry{
 	uint32_t desc[4];
@@ -786,6 +787,8 @@ void handle_interrupt(int vector, struct x86_user_context *regs)
 	lapic_ack();
 	++v->in_interrupt;
 
+	set_cputime(interrupt_from_user(regs)? 1: 2);
+
 	dkprintf("CPU[%d] got interrupt, vector: %d, RIP: 0x%lX\n", 
 	         ihk_mc_get_processor_id(), vector, regs->gpr.rip);
 
@@ -862,12 +865,14 @@ void handle_interrupt(int vector, struct x86_user_context *regs)
 
 	check_signal(0, regs, 0);
 	check_need_resched();
+	set_cputime(0);
 
 	--v->in_interrupt;
 }
 
 void gpe_handler(struct x86_user_context *regs)
 {
+	set_cputime(interrupt_from_user(regs)? 1: 2);
 	kprintf("General protection fault (err: %lx, %lx:%lx)\n",
 	        regs->gpr.error, regs->gpr.cs, regs->gpr.rip);
 	arch_show_interrupt_context(regs);
@@ -877,6 +882,7 @@ void gpe_handler(struct x86_user_context *regs)
 	set_signal(SIGSEGV, regs, NULL);
 	check_signal(0, regs, 0);
 	check_need_resched();
+	set_cputime(0);
 	// panic("GPF");
 }
 
@@ -886,6 +892,7 @@ void debug_handler(struct x86_user_context *regs)
 	int si_code = 0;
 	struct siginfo info;
 
+	set_cputime(interrupt_from_user(regs)? 1: 2);
 #ifdef DEBUG_PRINT_CPU
 	kprintf("debug exception (err: %lx, %lx:%lx)\n",
 	        regs->gpr.error, regs->gpr.cs, regs->gpr.rip);
@@ -905,12 +912,14 @@ void debug_handler(struct x86_user_context *regs)
 	set_signal(SIGTRAP, regs, &info);
 	check_signal(0, regs, 0);
 	check_need_resched();
+	set_cputime(0);
 }
 
 void int3_handler(struct x86_user_context *regs)
 {
 	struct siginfo info;
 
+	set_cputime(interrupt_from_user(regs)? 1: 2);
 #ifdef DEBUG_PRINT_CPU
 	kprintf("int3 exception (err: %lx, %lx:%lx)\n",
 	        regs->gpr.error, regs->gpr.cs, regs->gpr.rip);
@@ -922,6 +931,7 @@ void int3_handler(struct x86_user_context *regs)
 	set_signal(SIGTRAP, regs, &info);
 	check_signal(0, regs, 0);
 	check_need_resched();
+	set_cputime(0);
 }
 
 

arch/x86/kernel/include/syscall_list.h

@@ -20,7 +20,7 @@
  *     syscall_name[] only, no handler exists.
  */
 
-SYSCALL_DELEGATED(0, read)
+SYSCALL_HANDLED(0, read)
 SYSCALL_DELEGATED(1, write)
 SYSCALL_DELEGATED(2, open)
 SYSCALL_HANDLED(3, close)
@@ -35,7 +35,7 @@ SYSCALL_HANDLED(12, brk)
 SYSCALL_HANDLED(13, rt_sigaction)
 SYSCALL_HANDLED(14, rt_sigprocmask)
 SYSCALL_HANDLED(15, rt_sigreturn)
-SYSCALL_DELEGATED(16, ioctl)
+SYSCALL_HANDLED(16, ioctl)
 SYSCALL_DELEGATED(17, pread64)
 SYSCALL_DELEGATED(18, pwrite64)
 SYSCALL_DELEGATED(20, writev)
@@ -51,6 +51,8 @@ SYSCALL_HANDLED(30, shmat)
 SYSCALL_HANDLED(31, shmctl)
 SYSCALL_HANDLED(34, pause)
 SYSCALL_HANDLED(35, nanosleep)
+SYSCALL_HANDLED(36, getitimer)
+SYSCALL_HANDLED(38, setitimer)
 SYSCALL_HANDLED(39, getpid)
 SYSCALL_HANDLED(56, clone)
 SYSCALL_DELEGATED(57, fork)
@@ -69,6 +71,7 @@ SYSCALL_DELEGATED(79, getcwd)
 SYSCALL_DELEGATED(89, readlink)
 SYSCALL_HANDLED(96, gettimeofday)
 SYSCALL_HANDLED(97, getrlimit)
+SYSCALL_HANDLED(100, times)
 SYSCALL_HANDLED(101, ptrace)
 SYSCALL_HANDLED(102, getuid)
 SYSCALL_HANDLED(104, getgid)
@@ -107,6 +110,7 @@ SYSCALL_HANDLED(158, arch_prctl)
 SYSCALL_HANDLED(160, setrlimit)
 SYSCALL_HANDLED(164, settimeofday)
 SYSCALL_HANDLED(186, gettid)
+SYSCALL_HANDLED(200, tkill)
 SYSCALL_DELEGATED(201, time)
 SYSCALL_HANDLED(202, futex)
 SYSCALL_HANDLED(203, sched_setaffinity)
@@ -133,6 +137,7 @@ SYSCALL_HANDLED(279, move_pages)
 SYSCALL_DELEGATED(281, epoll_pwait)
 SYSCALL_HANDLED(282, signalfd)
 SYSCALL_HANDLED(289, signalfd4)
+SYSCALL_HANDLED(298, perf_event_open)
 #ifdef DCFA_KMOD
 SYSCALL_HANDLED(303, mod_call)
 #endif

arch/x86/kernel/interrupt.S

@@ -209,7 +209,9 @@ enter_user_mode:
 		callq	release_runq_lock
 		movq	$0,  %rdi
 		movq	%rsp, %rsi
-		call check_signal
+		call	check_signal
+		movq	$0,  %rdi
+		call	set_cputime
 		POP_ALL_REGS
 		addq	$8, %rsp
 		iretq

arch/x86/kernel/syscall.c

@@ -32,6 +32,8 @@ int write_process_vm(struct process_vm *vm, void *dst, const void *src, size_t s
 long do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact);
 long syscall(int num, ihk_mc_user_context_t *ctx);
 extern void save_fp_regs(struct thread *proc);
+void set_signal(int sig, void *regs0, siginfo_t *info);
+void check_signal(unsigned long rc, void *regs0, int num);
 
 //#define DEBUG_PRINT_SC
 
@@ -178,6 +180,17 @@ SYSCALL_DECLARE(rt_sigreturn)
 	if(sigsp->restart){
 		return syscall(sigsp->num, (ihk_mc_user_context_t *)regs);
 	}
+	if(regs->gpr.rflags & RFLAGS_TF){
+		struct siginfo info;
+
+		regs->gpr.rax = sigsp->sigrc;
+		memset(&info, '\0', sizeof info);
+		regs->gpr.rflags &= ~RFLAGS_TF;
+		info.si_code = TRAP_TRACE;
+		set_signal(SIGTRAP, regs, &info);
+		check_signal(0, regs, 0);
+		check_need_resched();
+	}
 	return sigsp->sigrc;
 }
 
@@ -595,6 +608,16 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			thread->sigmask.__val[0] |= pending->sigmask.__val[0];
 		kfree(pending);
 		ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+		if(regs->gpr.rflags & RFLAGS_TF){
+			struct siginfo info;
+
+			memset(&info, '\0', sizeof info);
+			regs->gpr.rflags &= ~RFLAGS_TF;
+			info.si_code = TRAP_TRACE;
+			set_signal(SIGTRAP, regs, &info);
+			check_signal(0, regs, 0);
+			check_need_resched();
+		}
 	}
 	else {
 		int	coredumped = 0;
@@ -756,6 +779,14 @@ hassigpending(struct thread *thread)
 	return getsigpending(thread, 0);
 }
 
+int
+interrupt_from_user(void *regs0)
+{
+	struct x86_user_context *regs = regs0;
+
+	return !(regs->gpr.rsp & 0x8000000000000000);
+}
+
 void
 check_signal(unsigned long rc, void *regs0, int num)
 {
@@ -785,7 +816,7 @@ check_signal(unsigned long rc, void *regs0, int num)
 		return;
 	}
 
-	if(regs != NULL && (regs->gpr.rsp & 0x8000000000000000)) {
+	if(regs != NULL && !interrupt_from_user(regs)) {
 		return;
 	}
 

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

@@ -80,6 +80,7 @@ fi
 # Remove mcoverlay if loaded
 if [ "$enable_mcoverlay" != "" ]; then
 	if [ "`lsmod | grep mcoverlay`" != "" ]; then
+		if [ "`cat /proc/mounts | grep /tmp/mcos/mcos0_sys`" != "" ]; then umount -l /tmp/mcos/mcos0_sys; fi
 		if [ "`cat /proc/mounts | grep /tmp/mcos/mcos0_proc`" != "" ]; then umount -l /tmp/mcos/mcos0_proc; fi
 		if [ "`cat /proc/mounts | grep /tmp/mcos`" != "" ]; then umount -l /tmp/mcos; fi
 		if [ -e /tmp/mcos ]; then rm -rf /tmp/mcos; fi
@@ -159,6 +160,25 @@ if [ "$enable_mcoverlay" != "" ]; then
 	if [ ! -e /tmp/mcos/mcos0_proc_work ]; then mkdir -p /tmp/mcos/mcos0_proc_work; fi
 	if ! mount -t mcoverlay mcoverlay -o lowerdir=/proc/mcos0:/proc,upperdir=/tmp/mcos/mcos0_proc_upper,workdir=/tmp/mcos/mcos0_proc_work,nocopyupw,nofscheck /tmp/mcos/mcos0_proc; then echo "error: mount /tmp/mcos/mcos0_proc"; exit; fi
 	mount --make-rprivate /proc
+	while [ ! -e /sys/devices/virtual/mcos/mcos0/sys ]
+	do
+		sleep 1
+	done
+	if [ ! -e /tmp/mcos/mcos0_sys ]; then mkdir -p /tmp/mcos/mcos0_sys; fi
+	if [ ! -e /tmp/mcos/mcos0_sys_upper ]; then mkdir -p /tmp/mcos/mcos0_sys_upper; fi
+	if [ ! -e /tmp/mcos/mcos0_sys_work ]; then mkdir -p /tmp/mcos/mcos0_sys_work; fi
+	if ! mount -t mcoverlay mcoverlay -o lowerdir=/sys/devices/virtual/mcos/mcos0/sys:/sys,upperdir=/tmp/mcos/mcos0_sys_upper,workdir=/tmp/mcos/mcos0_sys_work,nocopyupw,nofscheck /tmp/mcos/mcos0_sys; then echo "error: mount /tmp/mcos/mcos0_sys"; exit; fi
+	mount --make-rprivate /sys
+	for cpuid in `find /sys/devices/system/cpu/* -maxdepth 0 -name "cpu[0123456789]*" -printf "%f "`; do
+		if [ ! -e "/sys/devices/virtual/mcos/mcos0/sys/devices/system/cpu/$cpuid" ]; then
+			rm -rf /tmp/mcos/mcos0_sys/devices/system/cpu/$cpuid
+		fi
+	done
+	for cpuid in `find /sys/bus/cpu/devices/* -maxdepth 0 -name "cpu[0123456789]*" -printf "%f "`; do
+		if [ ! -e "/sys/devices/virtual/mcos/mcos0/sys/bus/cpu/devices/$cpuid" ]; then
+			rm -rf /tmp/mcos/mcos0_sys/bus/cpu/devices/$cpuid
+		fi
+	done
 fi
 if [ ${LOGMODE} -ne 0 ]
 then

configure

@@ -649,6 +649,7 @@ ac_user_opts='
 enable_option_checking
 with_kernelsrc
 with_target
+with_system_map
 enable_dcfa
 '
       ac_precious_vars='build_alias
@@ -1277,6 +1278,8 @@ Optional Packages:
                           /lib/modules/uname_r/build
   --with-target={attached-mic | builtin-mic | builtin-x86 | smp-x86}
                           target, default is attached-mic
+  --with-system_map=path  Path to 'System.map file', default is
+                          /boot/System.map-uname_r
 
 Some influential environment variables:
   CC          C compiler command
@@ -1784,6 +1787,15 @@ else
 fi
 
 
+
+# Check whether --with-system_map was given.
+if test "${with_system_map+set}" = set; then :
+  withval=$with_system_map; WITH_SYSTEM_MAP=$withval
+else
+  WITH_SYSTEM_MAP=yes
+fi
+
+
 # Check whether --enable-dcfa was given.
 if test "${enable_dcfa+set}" = set; then :
   enableval=$enable_dcfa;
@@ -3798,6 +3810,96 @@ esac
 KDIR="$WITH_KERNELSRC"
 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
+$as_echo_n "checking for System.map... " >&6; }
+if test -f "$MCCTRL_LINUX_SYMTAB"; then
+	MCCTRL_LINUX_SYMTAB="$MCCTRL_LINUX_SYMTAB"
+elif test -f "/boot/System.map-`uname -r`"; then
+	MCCTRL_LINUX_SYMTAB="/boot/System.map-`uname -r`"
+elif test -f "$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_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
 
 
 
@@ -3816,6 +3918,10 @@ TARGET="$WITH_TARGET"
 
 
 
+
+
+ac_config_headers="$ac_config_headers executer/kernel/mcctrl/config.h"
+
 ac_config_files="$ac_config_files Makefile executer/user/Makefile executer/kernel/mcctrl/Makefile executer/kernel/mcoverlayfs/Makefile kernel/Makefile kernel/Makefile.build arch/x86/tools/mcreboot-attached-mic.sh arch/x86/tools/mcshutdown-attached-mic.sh arch/x86/tools/mcreboot-builtin-x86.sh arch/x86/tools/mcreboot-smp-x86.sh arch/x86/tools/mcstop+release-smp-x86.sh arch/x86/tools/mcshutdown-builtin-x86.sh arch/x86/tools/mcreboot.1:arch/x86/tools/mcreboot.1in"
 
 
@@ -3915,43 +4021,7 @@ test "x$prefix" = xNONE && prefix=$ac_default_prefix
 # Let make expand exec_prefix.
 test "x$exec_prefix" = xNONE && exec_prefix='${prefix}'
 
-# Transform confdefs.h into DEFS.
+DEFS=-DHAVE_CONFIG_H
-# Protect against shell expansion while executing Makefile rules.
-# Protect against Makefile macro expansion.
-#
-# If the first sed substitution is executed (which looks for macros that
-# take arguments), then branch to the quote section.  Otherwise,
-# look for a macro that doesn't take arguments.
-ac_script='
-:mline
-/\\$/{
- N
- s,\\\n,,
- b mline
-}
-t clear
-:clear
-s/^[	 ]*#[	 ]*define[	 ][	 ]*\([^	 (][^	 (]*([^)]*)\)[	 ]*\(.*\)/-D\1=\2/g
-t quote
-s/^[	 ]*#[	 ]*define[	 ][	 ]*\([^	 ][^	 ]*\)[	 ]*\(.*\)/-D\1=\2/g
-t quote
-b any
-:quote
-s/[	 `~#$^&*(){}\\|;'\''"<>?]/\\&/g
-s/\[/\\&/g
-s/\]/\\&/g
-s/\$/$$/g
-H
-:any
-${
-	g
-	s/^\n//
-	s/\n/ /g
-	p
-}
-'
-DEFS=`sed -n "$ac_script" confdefs.h`
-
 
 ac_libobjs=
 ac_ltlibobjs=
@@ -4385,11 +4455,15 @@ case $ac_config_files in *"
 "*) set x $ac_config_files; shift; ac_config_files=$*;;
 esac
 
+case $ac_config_headers in *"
+"*) set x $ac_config_headers; shift; ac_config_headers=$*;;
+esac
 
 
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 # Files that config.status was made for.
 config_files="$ac_config_files"
+config_headers="$ac_config_headers"
 
 _ACEOF
 
@@ -4410,10 +4484,15 @@ Usage: $0 [OPTION]... [TAG]...
       --recheck    update $as_me by reconfiguring in the same conditions
       --file=FILE[:TEMPLATE]
                    instantiate the configuration file FILE
+      --header=FILE[:TEMPLATE]
+                   instantiate the configuration header FILE
 
 Configuration files:
 $config_files
 
+Configuration headers:
+$config_headers
+
 Report bugs to the package provider."
 
 _ACEOF
@@ -4474,7 +4553,18 @@ do
     esac
     as_fn_append CONFIG_FILES " '$ac_optarg'"
     ac_need_defaults=false;;
-  --he | --h |  --help | --hel | -h )
+  --header | --heade | --head | --hea )
+    $ac_shift
+    case $ac_optarg in
+    *\'*) ac_optarg=`$as_echo "$ac_optarg" | sed "s/'/'\\\\\\\\''/g"` ;;
+    esac
+    as_fn_append CONFIG_HEADERS " '$ac_optarg'"
+    ac_need_defaults=false;;
+  --he | --h)
+    # Conflict between --help and --header
+    as_fn_error $? "ambiguous option: \`$1'
+Try \`$0 --help' for more information.";;
+  --help | --hel | -h )
     $as_echo "$ac_cs_usage"; exit ;;
   -q | -quiet | --quiet | --quie | --qui | --qu | --q \
   | -silent | --silent | --silen | --sile | --sil | --si | --s)
@@ -4530,6 +4620,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 for ac_config_target in $ac_config_targets
 do
   case $ac_config_target in
+    "executer/kernel/mcctrl/config.h") CONFIG_HEADERS="$CONFIG_HEADERS executer/kernel/mcctrl/config.h" ;;
     "Makefile") CONFIG_FILES="$CONFIG_FILES Makefile" ;;
     "executer/user/Makefile") CONFIG_FILES="$CONFIG_FILES executer/user/Makefile" ;;
     "executer/kernel/mcctrl/Makefile") CONFIG_FILES="$CONFIG_FILES executer/kernel/mcctrl/Makefile" ;;
@@ -4556,6 +4647,7 @@ done
 # bizarre bug on SunOS 4.1.3.
 if $ac_need_defaults; then
   test "${CONFIG_FILES+set}" = set || CONFIG_FILES=$config_files
+  test "${CONFIG_HEADERS+set}" = set || CONFIG_HEADERS=$config_headers
 fi
 
 # Have a temporary directory for convenience.  Make it in the build tree
@@ -4743,8 +4835,116 @@ fi
 cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 fi # test -n "$CONFIG_FILES"
 
+# Set up the scripts for CONFIG_HEADERS section.
+# No need to generate them if there are no CONFIG_HEADERS.
+# This happens for instance with `./config.status Makefile'.
+if test -n "$CONFIG_HEADERS"; then
+cat >"$ac_tmp/defines.awk" <<\_ACAWK ||
+BEGIN {
+_ACEOF
 
-eval set X "  :F $CONFIG_FILES      "
+# Transform confdefs.h into an awk script `defines.awk', embedded as
+# here-document in config.status, that substitutes the proper values into
+# config.h.in to produce config.h.
+
+# Create a delimiter string that does not exist in confdefs.h, to ease
+# handling of long lines.
+ac_delim='%!_!# '
+for ac_last_try in false false :; do
+  ac_tt=`sed -n "/$ac_delim/p" confdefs.h`
+  if test -z "$ac_tt"; then
+    break
+  elif $ac_last_try; then
+    as_fn_error $? "could not make $CONFIG_HEADERS" "$LINENO" 5
+  else
+    ac_delim="$ac_delim!$ac_delim _$ac_delim!! "
+  fi
+done
+
+# For the awk script, D is an array of macro values keyed by name,
+# likewise P contains macro parameters if any.  Preserve backslash
+# newline sequences.
+
+ac_word_re=[_$as_cr_Letters][_$as_cr_alnum]*
+sed -n '
+s/.\{148\}/&'"$ac_delim"'/g
+t rset
+:rset
+s/^[	 ]*#[	 ]*define[	 ][	 ]*/ /
+t def
+d
+:def
+s/\\$//
+t bsnl
+s/["\\]/\\&/g
+s/^ \('"$ac_word_re"'\)\(([^()]*)\)[	 ]*\(.*\)/P["\1"]="\2"\
+D["\1"]=" \3"/p
+s/^ \('"$ac_word_re"'\)[	 ]*\(.*\)/D["\1"]=" \2"/p
+d
+:bsnl
+s/["\\]/\\&/g
+s/^ \('"$ac_word_re"'\)\(([^()]*)\)[	 ]*\(.*\)/P["\1"]="\2"\
+D["\1"]=" \3\\\\\\n"\\/p
+t cont
+s/^ \('"$ac_word_re"'\)[	 ]*\(.*\)/D["\1"]=" \2\\\\\\n"\\/p
+t cont
+d
+:cont
+n
+s/.\{148\}/&'"$ac_delim"'/g
+t clear
+:clear
+s/\\$//
+t bsnlc
+s/["\\]/\\&/g; s/^/"/; s/$/"/p
+d
+:bsnlc
+s/["\\]/\\&/g; s/^/"/; s/$/\\\\\\n"\\/p
+b cont
+' <confdefs.h | sed '
+s/'"$ac_delim"'/"\\\
+"/g' >>$CONFIG_STATUS || ac_write_fail=1
+
+cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
+  for (key in D) D_is_set[key] = 1
+  FS = ""
+}
+/^[\t ]*#[\t ]*(define|undef)[\t ]+$ac_word_re([\t (]|\$)/ {
+  line = \$ 0
+  split(line, arg, " ")
+  if (arg[1] == "#") {
+    defundef = arg[2]
+    mac1 = arg[3]
+  } else {
+    defundef = substr(arg[1], 2)
+    mac1 = arg[2]
+  }
+  split(mac1, mac2, "(") #)
+  macro = mac2[1]
+  prefix = substr(line, 1, index(line, defundef) - 1)
+  if (D_is_set[macro]) {
+    # Preserve the white space surrounding the "#".
+    print prefix "define", macro P[macro] D[macro]
+    next
+  } else {
+    # Replace #undef with comments.  This is necessary, for example,
+    # in the case of _POSIX_SOURCE, which is predefined and required
+    # on some systems where configure will not decide to define it.
+    if (defundef == "undef") {
+      print "/*", prefix defundef, macro, "*/"
+      next
+    }
+  }
+}
+{ print }
+_ACAWK
+_ACEOF
+cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
+  as_fn_error $? "could not setup config headers machinery" "$LINENO" 5
+fi # test -n "$CONFIG_HEADERS"
+
+
+eval set X "  :F $CONFIG_FILES  :H $CONFIG_HEADERS    "
 shift
 for ac_tag
 do
@@ -4952,7 +5152,30 @@ which seems to be undefined.  Please make sure it is defined" >&2;}
   esac \
   || as_fn_error $? "could not create $ac_file" "$LINENO" 5
  ;;
-
+  :H)
+  #
+  # CONFIG_HEADER
+  #
+  if test x"$ac_file" != x-; then
+    {
+      $as_echo "/* $configure_input  */" \
+      && eval '$AWK -f "$ac_tmp/defines.awk"' "$ac_file_inputs"
+    } >"$ac_tmp/config.h" \
+      || as_fn_error $? "could not create $ac_file" "$LINENO" 5
+    if diff "$ac_file" "$ac_tmp/config.h" >/dev/null 2>&1; then
+      { $as_echo "$as_me:${as_lineno-$LINENO}: $ac_file is unchanged" >&5
+$as_echo "$as_me: $ac_file is unchanged" >&6;}
+    else
+      rm -f "$ac_file"
+      mv "$ac_tmp/config.h" "$ac_file" \
+	|| as_fn_error $? "could not create $ac_file" "$LINENO" 5
+    fi
+  else
+    $as_echo "/* $configure_input  */" \
+      && eval '$AWK -f "$ac_tmp/defines.awk"' "$ac_file_inputs" \
+      || as_fn_error $? "could not create -" "$LINENO" 5
+  fi
+ ;;
 
 
   esac

configure.ac

@@ -27,6 +27,11 @@ AC_ARG_WITH([target],
     [--with-target={attached-mic | builtin-mic | builtin-x86 | smp-x86}],[target, default is attached-mic]),
   [WITH_TARGET=$withval],[WITH_TARGET=yes])
 
+AC_ARG_WITH([system_map],
+  AS_HELP_STRING(
+    [--with-system_map=path],[Path to 'System.map file', default is /boot/System.map-uname_r]),
+  [WITH_SYSTEM_MAP=$withval],[WITH_SYSTEM_MAP=yes])
+
 AC_ARG_ENABLE([dcfa],
   [AS_HELP_STRING(
   	[--enable-dcfa],[Enable DCFA modules])],[],[enable_dcfa=no])
@@ -139,6 +144,65 @@ esac
 KDIR="$WITH_KERNELSRC"
 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 -f "$MCCTRL_LINUX_SYMTAB"; then
+	MCCTRL_LINUX_SYMTAB="$MCCTRL_LINUX_SYMTAB"
+elif test -f "/boot/System.map-`uname -r`"; then
+	MCCTRL_LINUX_SYMTAB="/boot/System.map-`uname -r`"
+elif test -f "$KDIR/System.map"; then
+	MCCTRL_LINUX_SYMTAB="$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_unshare])
+
 AC_SUBST(CC)
 AC_SUBST(XCC)
 AC_SUBST(ARCH)
@@ -157,6 +221,7 @@ AC_SUBST(IHK_RELEASE_DATE)
 AC_SUBST(MCKERNEL_RELEASE_DATE)
 AC_SUBST(DCFA_RESEASE_DATE)
 
+AC_CONFIG_HEADERS([executer/kernel/mcctrl/config.h])
 AC_CONFIG_FILES([
 	Makefile
 	executer/user/Makefile

executer/include/uprotocol.h

@@ -48,6 +48,9 @@
 #define MCEXEC_UP_OPEN_EXEC      0x30a02912
 #define MCEXEC_UP_CLOSE_EXEC     0x30a02913
 
+#define MCEXEC_UP_SYS_MOUNT      0x30a02914
+#define MCEXEC_UP_SYS_UNSHARE    0x30a02915
+
 #define MCEXEC_UP_DEBUG_LOG     0x40000000
 
 #define MCEXEC_UP_TRANSFER_TO_REMOTE	0
@@ -167,4 +170,16 @@ struct newprocess_desc {
 	int pid;
 };
 
+struct sys_mount_desc {
+	char *dev_name;
+	char *dir_name;
+	char *type;
+	unsigned long flags;
+	void *data;
+};
+
+struct sys_unshare_desc {
+	unsigned long unshare_flags;
+};
+
 #endif

executer/kernel/mcctrl/Makefile.in

@@ -7,7 +7,7 @@ IHK_BASE=$(src)/../../../../ihk
 
 obj-m += mcctrl.o
 
-ccflags-y := -I$(IHK_BASE)/linux/include -I$(IHK_BASE)/ikc/include -I$(IHK_BASE)/include -I$(src)/../../include -mcmodel=kernel -mno-red-zone -DMCEXEC_PATH=\"$(BINDIR)/mcexec\"
+ccflags-y := -I$(IHK_BASE)/linux/include -I$(IHK_BASE)/ikc/include -I$(IHK_BASE)/include -I$(src)/../../include -mcmodel=kernel -mno-red-zone -DMCEXEC_PATH=\"$(BINDIR)/mcexec\" -I@abs_builddir@
 
 mcctrl-y := driver.o control.o ikc.o syscall.o procfs.o binfmt_mcexec.o
 mcctrl-y += sysfs.o sysfs_files.o

executer/kernel/mcctrl/config.h.in

@@ -0,0 +1,25 @@
+/* executer/kernel/mcctrl/config.h.in.  Generated from configure.ac by autoheader.  */
+
+/* Define to address of kernel symbol sys_mount, or 0 if exported */
+#undef MCCTRL_KSYM_sys_mount
+
+/* Define to address of kernel symbol sys_unshare, or 0 if exported */
+#undef MCCTRL_KSYM_sys_unshare
+
+/* Define to the address where bug reports for this package should be sent. */
+#undef PACKAGE_BUGREPORT
+
+/* Define to the full name of this package. */
+#undef PACKAGE_NAME
+
+/* Define to the full name and version of this package. */
+#undef PACKAGE_STRING
+
+/* Define to the one symbol short name of this package. */
+#undef PACKAGE_TARNAME
+
+/* Define to the home page for this package. */
+#undef PACKAGE_URL
+
+/* Define to the version of this package. */
+#undef PACKAGE_VERSION

executer/kernel/mcctrl/control.c

@@ -36,6 +36,7 @@
 #include <asm/delay.h>
 #include <asm/msr.h>
 #include <asm/io.h>
+#include "config.h"
 #include "mcctrl.h"
 
 //#define DEBUG
@@ -46,12 +47,31 @@
 #define dprintk(...)
 #endif
 
+#ifdef MCCTRL_KSYM_sys_unshare
+#if MCCTRL_KSYM_sys_unshare
+typedef int (*int_star_fn_ulong_t)(unsigned long);
+int (*mcctrl_sys_unshare)(unsigned long unshare_flags) =
+        (int_star_fn_ulong_t)
+        MCCTRL_KSYM_sys_unshare;
+#else // exported
+int (*mcctrl_sys_unshare)(unsigned long unshare_flags) = NULL;
+#endif
+#endif
+
+#ifdef MCCTRL_KSYM_sys_mount
+#if MCCTRL_KSYM_sys_mount
+typedef int (*int_star_fn_char_char_char_ulong_void_t)(char *, char *, char *, unsigned long, void *);
+int (*mcctrl_sys_mount)(char *dev_name,char *dir_name, char *type, unsigned long flags, void *data) =
+        (int_star_fn_char_char_char_ulong_void_t)
+        MCCTRL_KSYM_sys_mount;
+#else // exported
+int (*mcctrl_sys_mount)(char *dev_name,char *dir_name, char *type, unsigned long flags, void *data) = NULL;
+#endif
+#endif
+
 //static DECLARE_WAIT_QUEUE_HEAD(wq_prepare);
 //extern struct mcctrl_channel *channels;
 int mcctrl_ikc_set_recv_cpu(ihk_os_t os, int cpu);
-extern int procfs_create_entry(void *os, int ref, int osnum, int pid, char *name,
-		int mode, void *opaque);
-extern void procfs_delete_entry(void *os, int osnum, char *fname);
 
 static long mcexec_prepare_image(ihk_os_t os,
                                  struct program_load_desc * __user udesc)
@@ -105,10 +125,10 @@ static long mcexec_prepare_image(ihk_os_t os,
 
 	pdesc->args = (void*)virt_to_phys(args);
 	printk("args: 0x%lX\n", (unsigned long)pdesc->args);
-	printk("argc: %d\n", *(int*)args);
+	printk("argc: %ld\n", *(long *)args);
 	pdesc->envs = (void*)virt_to_phys(envs);
 	printk("envs: 0x%lX\n", (unsigned long)pdesc->envs);
-	printk("envc: %d\n", *(int*)envs);
+	printk("envc: %ld\n", *(long *)envs);
 
 	isp.msg = SCD_MSG_PREPARE_PROCESS;
 	isp.ref = pdesc->cpu;
@@ -267,12 +287,15 @@ static void release_handler(ihk_os_t os, void *param)
 {
 	struct handlerinfo *info = param;
 	struct ikc_scd_packet isp;
+	int os_ind = ihk_host_os_get_index(os);
 
 	memset(&isp, '\0', sizeof isp);
 	isp.msg = SCD_MSG_CLEANUP_PROCESS;
 	isp.pid = info->pid;
 
 	mcctrl_ikc_send(os, 0, &isp);
+	if(os_ind >= 0)
+		delete_pid_entry(os_ind, info->pid);
 	kfree(param);
 }
 
@@ -838,7 +861,7 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 	struct mckernel_exec_file *mcef_iter;
 	int retval;
 	int os_ind = ihk_host_os_get_index(os);
-	char *proc_name, *pathbuf, *fullpath;
+	char *pathbuf, *fullpath;
 
 	if (os_ind < 0) {
 		return EINVAL;
@@ -849,12 +872,6 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 		return ENOMEM;
 	}
 
-	proc_name = kmalloc(PATH_MAX, GFP_TEMPORARY);
-	if (!proc_name) {
-		retval = ENOMEM;
-		goto out_error_free_path;
-	}
-
 	file = open_exec(filename);
 	retval = PTR_ERR(file);
 	if (IS_ERR(file)) {
@@ -873,8 +890,6 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 		goto out_put_file;
 	}
 
-	snprintf(proc_name, 1024, "mcos%d/%d/exe", os_ind, current->tgid);
-
 	spin_lock_irq(&mckernel_exec_file_lock);
 	/* Find previous file (if exists) and drop it */
 	list_for_each_entry(mcef_iter, &mckernel_exec_files, list) {
@@ -883,9 +898,6 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 			fput(mcef_iter->fp);
 			list_del(&mcef_iter->list);
 			kfree(mcef_iter);
-			/* Drop old /proc/self/exe */
-			procfs_delete_entry(os, os_ind, proc_name);
-			dprintk("%d open_exec dropped previous executable \n", (int)current->tgid);
 			break;
 		}
 	}
@@ -897,15 +909,12 @@ int mcexec_open_exec(ihk_os_t os, char * __user filename)
 	list_add_tail(&mcef->list, &mckernel_exec_files);
 
 	/* Create /proc/self/exe entry */
-	if (procfs_create_entry(os, 0, os_ind, current->tgid, proc_name, 
+	add_pid_entry(os_ind, current->tgid);
-				S_IFLNK, fullpath) != 0) {
+	proc_exe_link(os_ind, current->tgid, fullpath);
-		printk("ERROR: could not create a procfs entry for %s.\n", proc_name);
-	}
 	spin_unlock(&mckernel_exec_file_lock);
 
 	dprintk("%d open_exec and holding file: %s\n", (int)current->tgid, filename);
 
-	kfree(proc_name);
 	kfree(pathbuf);
 
 	return 0;
@@ -914,9 +923,6 @@ out_put_file:
 	fput(file);
 
 out_error_free:
-	kfree(proc_name);
-
-out_error_free_path:
 	kfree(pathbuf);
 	return -retval;
 }
@@ -927,7 +933,6 @@ int mcexec_close_exec(ihk_os_t os)
 	struct mckernel_exec_file *mcef = NULL;
 	int found = 0;
 	int os_ind = ihk_host_os_get_index(os);	
-	char proc_name[1024];
 
 	if (os_ind < 0) {
 		return EINVAL;
@@ -946,14 +951,6 @@ int mcexec_close_exec(ihk_os_t os)
 		}
 	}
 
-	/* Remove /proc/self/exe and /proc/self directory 
-	 * TODO: instead of removing directory explicitly, detect in procfs_delete_entry() 
-	 * when a directory becomes empty and remove it automatically */
-	snprintf(proc_name, 1024, "mcos%d/%d/exe", os_ind, current->tgid);
-	procfs_delete_entry(os, os_ind, proc_name);
-	snprintf(proc_name, 1024, "mcos%d/%d", os_ind, current->tgid);
-	procfs_delete_entry(os, os_ind, proc_name);
-
 	spin_unlock(&mckernel_exec_file_lock);
 
 	return (found ? 0 : EINVAL);
@@ -1011,6 +1008,67 @@ long mcexec_strncpy_from_user(ihk_os_t os, struct strncpy_from_user_desc * __use
 	return 0;
 }
 
+long mcexec_sys_mount(struct sys_mount_desc *__user arg)
+{
+	struct sys_mount_desc desc;
+	struct cred *promoted;
+	const struct cred *original;
+	int ret;
+
+	if (copy_from_user(&desc, arg, sizeof(desc))) {
+		return -EFAULT;
+	}
+
+	promoted = prepare_creds();
+	if (!promoted) {
+		return -ENOMEM;
+	}
+	cap_raise(promoted->cap_effective, CAP_SYS_ADMIN);
+	original = override_creds(promoted);
+
+#if MCCTRL_KSYM_sys_mount
+	ret = mcctrl_sys_mount(desc.dev_name, desc.dir_name, desc.type,
+		desc.flags, desc.data);
+#else
+	ret = -EFAULT;
+#endif
+
+	revert_creds(original);
+	put_cred(promoted);
+
+	return ret;
+}
+
+long mcexec_sys_unshare(struct sys_unshare_desc *__user arg)
+{
+	struct sys_unshare_desc desc;
+	struct cred *promoted;
+	const struct cred *original;
+	int ret;
+
+	if (copy_from_user(&desc, arg, sizeof(desc))) {
+		return -EFAULT;
+	}
+
+	promoted = prepare_creds();
+	if (!promoted) {
+		return -ENOMEM;
+	}
+	cap_raise(promoted->cap_effective, CAP_SYS_ADMIN);
+	original = override_creds(promoted);
+
+#if MCCTRL_KSYM_sys_unshare
+	ret = mcctrl_sys_unshare(desc.unshare_flags);
+#else
+	ret = -EFAULT;
+#endif
+
+	revert_creds(original);
+	put_cred(promoted);
+
+	return ret;
+}
+
 long __mcctrl_control(ihk_os_t os, unsigned int req, unsigned long arg,
                       struct file *file)
 {
@@ -1065,6 +1123,12 @@ long __mcctrl_control(ihk_os_t os, unsigned int req, unsigned long arg,
 	case MCEXEC_UP_GET_CREDV:
 		return mcexec_getcredv((int *)arg);
 
+	case MCEXEC_UP_SYS_MOUNT:
+		return mcexec_sys_mount((struct sys_mount_desc *)arg);
+
+	case MCEXEC_UP_SYS_UNSHARE:
+		return mcexec_sys_unshare((struct sys_unshare_desc *)arg);
+
 	case MCEXEC_UP_DEBUG_LOG:
 		return mcexec_debug_log(os, arg);
 	}

executer/kernel/mcctrl/driver.c

@@ -68,6 +68,8 @@ static struct ihk_os_user_call_handler mcctrl_uchs[] = {
 	{ .request = MCEXEC_UP_CLOSE_EXEC, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_GET_CRED, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_GET_CREDV, .func = mcctrl_ioctl },
+	{ .request = MCEXEC_UP_SYS_MOUNT, .func = mcctrl_ioctl },
+	{ .request = MCEXEC_UP_SYS_UNSHARE, .func = mcctrl_ioctl },
 	{ .request = MCEXEC_UP_DEBUG_LOG, .func = mcctrl_ioctl },
 };
 
@@ -80,6 +82,12 @@ static struct ihk_os_user_call mcctrl_uc[OS_MAX_MINOR];
 
 static ihk_os_t os[OS_MAX_MINOR];
 
+ihk_os_t
+osnum_to_os(int n)
+{
+	return os[n];
+}
+
 static int __init mcctrl_init(void)
 {
 	int	i;

executer/kernel/mcctrl/ikc.c

@@ -41,9 +41,6 @@
 void mcexec_prepare_ack(ihk_os_t os, unsigned long arg, int err);
 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_channel *c, int pid, unsigned long arg);
-void procfs_create(void *__os, int ref, int osnum, int pid, unsigned long arg);
-void procfs_delete(void *__os, int osnum, unsigned long arg);
-void procfs_answer(unsigned long arg, int err);
 void sig_done(unsigned long arg, int err);
 
 static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
@@ -69,14 +66,6 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		mcexec_syscall(usrdata->channels + pisp->ref, pisp->pid, pisp->arg);
 		break;
 
-	case SCD_MSG_PROCFS_CREATE:
-		procfs_create(__os, pisp->ref, pisp->osnum, pisp->pid, pisp->arg);
-		break;
-
-	case SCD_MSG_PROCFS_DELETE:
-		procfs_delete(__os, pisp->osnum, pisp->arg);
-		break;
-
 	case SCD_MSG_PROCFS_ANSWER:
 		procfs_answer(pisp->arg, pisp->err);
 		break;
@@ -98,6 +87,14 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 				pisp->sysfs_arg1, pisp->sysfs_arg2);
 		break;
 
+	case SCD_MSG_PROCFS_TID_CREATE:
+		add_tid_entry(ihk_host_os_get_index(__os), pisp->pid, pisp->arg);
+		break;
+
+	case SCD_MSG_PROCFS_TID_DELETE:
+		delete_tid_entry(ihk_host_os_get_index(__os), pisp->pid, pisp->arg);
+		break;
+
 	default:
 		printk(KERN_ERR "mcctrl:syscall_packet_handler:"
 				"unknown message (%d.%d.%d.%d.%d.%#lx)\n",

executer/kernel/mcctrl/mcctrl.h

@@ -81,6 +81,8 @@
 #define SCD_MSG_SYSFS_RESP_SETUP        0x41
 /* #define SCD_MSG_SYSFS_REQ_CLEANUP    0x42 */
 /* #define SCD_MSG_SYSFS_RESP_CLEANUP   0x43 */
+#define SCD_MSG_PROCFS_TID_CREATE	0x44
+#define SCD_MSG_PROCFS_TID_DELETE	0x45
 
 #define DMA_PIN_SHIFT                   21
 
@@ -226,6 +228,8 @@ int mcctrl_ikc_is_valid_thread(ihk_os_t os, int cpu);
 int reserve_user_space(struct mcctrl_usrdata *usrdata, unsigned long *startp,
 		unsigned long *endp);
 
+ihk_os_t osnum_to_os(int n);
+
 /* syscall.c */
 int init_peer_channel_registry(struct mcctrl_usrdata *ud);
 void destroy_peer_channel_registry(struct mcctrl_usrdata *ud);
@@ -244,6 +248,7 @@ struct procfs_read {
 	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) */
 };
 
@@ -253,6 +258,15 @@ struct procfs_file {
 	char fname[PROCFS_NAME_MAX];	/* procfs filename (request) */
 };
 
+void procfs_answer(unsigned int arg, int err);
+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);
+void delete_pid_entry(int osnum, int pid);
+void proc_exe_link(int osnum, int pid, const char *path);
+void procfs_init(int osnum);
+void procfs_exit(int osnum);
+
 /* sysfs_files.c */
 void setup_sysfs_files(ihk_os_t os);
 

executer/user/Makefile.in

@@ -1,5 +1,6 @@
 CC=@CC@
 BINDIR=@BINDIR@
+KDIR ?= @KDIR@
 CFLAGS=-Wall -O -fPIE -pie
 VPATH=@abs_srcdir@
 TARGET=mcexec
@@ -7,7 +8,7 @@ TARGET=mcexec
 all: $(TARGET)
 
 mcexec: mcexec.c
-	$(CC) $(CFLAGS) $(EXTRA_CFLAGS) -pthread -o $@ $^ $(EXTRA_OBJS)
+	$(CC) -I${KDIR} $(CFLAGS) $(EXTRA_CFLAGS) -pthread -o $@ $^ $(EXTRA_OBJS)
 
 clean:
 	$(RM) $(TARGET) *.o
@@ -16,5 +17,5 @@ clean:
 
 install:
 	mkdir -p -m 755 $(BINDIR)
-	install -o root -m 4755 mcexec $(BINDIR)
+	install -m 755 mcexec $(BINDIR)
 

executer/user/mcexec.c

@@ -60,7 +60,7 @@
 #include <signal.h>
 #include <sys/signalfd.h>
 #include <sys/mount.h>
-#include <linux/version.h>
+#include <include/generated/uapi/linux/version.h>
 #include "../include/uprotocol.h"
 
 //#define DEBUG
@@ -185,7 +185,8 @@ struct program_load_desc *load_elf(FILE *fp, char **interp_pathp)
 	
 	desc = malloc(sizeof(struct program_load_desc)
 	              + sizeof(struct program_image_section) * nhdrs);
-	memset(desc, '\0', sizeof(struct program_load_desc));
+	memset(desc, '\0', sizeof(struct program_load_desc)
+	                   + sizeof(struct program_image_section) * nhdrs);
 	desc->shell_path[0] = '\0';
 	fseek(fp, hdr.e_phoff, SEEK_SET);
 	j = 0;
@@ -673,6 +674,7 @@ void transfer_image(int fd, struct program_load_desc *desc)
 		          desc->sections[i].offset, flen);
 
 		while (s < e) {
+			memset(&pt, '\0', sizeof pt);
 			pt.rphys = rpa;
 			pt.userp = dma_buf;
 			pt.size = PAGE_SIZE;
@@ -776,7 +778,7 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	}
 
 	/* Count full length */
-	full_len = sizeof(int) + sizeof(char *); // Counter and terminating NULL
+	full_len = sizeof(long) + sizeof(char *); // Counter and terminating NULL
 	if (first) {
 		full_len += sizeof(char *) + strlen(first) + 1; 
 	}
@@ -786,6 +788,8 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 		full_len += sizeof(char *) + strlen(strings[string_i]) + 1; 
 	}
 
+	full_len = (full_len + sizeof(long) - 1) & ~(sizeof(long) - 1);
+
 	_flat = (char *)malloc(full_len);
 	if (!_flat) {
 		return 0;
@@ -794,14 +798,14 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	memset(_flat, 0, full_len);
 
 	/* Number of strings */
-	*((int*)_flat) = nr_strings + (first ? 1 : 0);
+	*((long *)_flat) = nr_strings + (first ? 1 : 0);
 	
 	// Actual offset
-	flat_offset = sizeof(int) + sizeof(char *) * (nr_strings + 1 + 
+	flat_offset = sizeof(long) + sizeof(char *) * (nr_strings + 1 + 
 			(first ? 1 : 0)); 
 
 	if (first) {
-		*((char **)(_flat + sizeof(int))) = (void *)flat_offset;
+		*((char **)(_flat + sizeof(long))) = (void *)flat_offset;
 		memcpy(_flat + flat_offset, first, strlen(first) + 1);
 		flat_offset += strlen(first) + 1;
 	}
@@ -809,7 +813,7 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	for (string_i = 0; string_i < nr_strings; ++string_i) {
 		
 		/* Fabricate the string */
-		*((char **)(_flat + sizeof(int) + (string_i + (first ? 1 : 0)) 
+		*((char **)(_flat + sizeof(long) + (string_i + (first ? 1 : 0)) 
 					* sizeof(char *))) = (void *)flat_offset;
 		memcpy(_flat + flat_offset, strings[string_i], strlen(strings[string_i]) + 1);
 		flat_offset += strlen(strings[string_i]) + 1;
@@ -892,6 +896,7 @@ sendsig(int sig, siginfo_t *siginfo, void *context)
 		remote_tid = -1;
 	}
 
+	memset(&sigdesc, '\0', sizeof sigdesc);
 	sigdesc.cpu = cpu;
 	sigdesc.pid = (int)pid;
 	sigdesc.tid = remote_tid;
@@ -918,6 +923,7 @@ act_signalfd4(struct syscall_wait_desc *w)
 	switch(mode){
 	    case 0: /* new signalfd */
 		sfd = malloc(sizeof(struct sigfd));
+		memset(sfd, '\0', sizeof(struct sigfd));
 		tmp = w->sr.args[1];
 		flags = 0;
 		if(tmp & SFD_NONBLOCK)
@@ -1301,39 +1307,6 @@ int main(int argc, char **argv)
 		++optind;
 	}
 
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,0,0)
-	__dprintf("mcoverlay enable\n");
-	char mcos_procdir[PATH_MAX];
-
-	error = isunshare();
-	if (error == 0) {
-		if (unshare(CLONE_NEWNS)) {
-			fprintf(stderr, "Error: Failed to unshare. (%s)\n", 
-				strerror(errno));
-			return 1;
-		}
-
-		sprintf(mcos_procdir, "/tmp/mcos/mcos%d_proc", mcosid);
-		if (mount(mcos_procdir, "/proc", NULL, MS_BIND, NULL)) {
-			fprintf(stderr, "Error: Failed to mount. (%s)\n", 
-				strerror(errno));
-			return 1;
-		}
-	} else if (error == -1) {
-		return 1;
-	}
-#else
-	__dprintf("mcoverlay disable\n");
-#endif
-
-	__dprintf("before seteuid(): uid=%d, euid=%d\n", getuid(), geteuid());
-	if (seteuid(getuid())) {
-		fprintf(stderr, "Error: Failed to seteuid. (%s)\n", 
-			strerror(errno));
-		return 1;
-	}
-	__dprintf("after seteuid():  uid=%d, euid=%d\n", getuid(), geteuid());
-
 	sprintf(dev, "/dev/mcos%d", mcosid);
 
 	/* No more arguments? */
@@ -1355,6 +1328,58 @@ int main(int argc, char **argv)
 		return 1;
 	}
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,0,0)
+	__dprintf("mcoverlay enable\n");
+	char mcos_procdir[PATH_MAX];
+	char mcos_sysdir[PATH_MAX];
+
+	error = isunshare();
+	if (error == 0) {
+		struct sys_unshare_desc unshare_desc;
+		struct sys_mount_desc mount_desc;
+
+		memset(&unshare_desc, '\0', sizeof unshare_desc);
+		memset(&mount_desc, '\0', sizeof mount_desc);
+		unshare_desc.unshare_flags = CLONE_NEWNS;
+		if (ioctl(fd, MCEXEC_UP_SYS_UNSHARE, 
+			(unsigned long)&unshare_desc) != 0) {
+			fprintf(stderr, "Error: Failed to unshare. (%s)\n", 
+				strerror(errno));
+			return 1;
+		}
+
+		sprintf(mcos_procdir, "/tmp/mcos/mcos%d_proc", mcosid);
+		mount_desc.dev_name = mcos_procdir;
+		mount_desc.dir_name = "/proc";
+		mount_desc.type = NULL;
+		mount_desc.flags = MS_BIND;
+		mount_desc.data = NULL;
+		if (ioctl(fd, MCEXEC_UP_SYS_MOUNT, 
+			(unsigned long)&mount_desc) != 0) {
+			fprintf(stderr, "Error: Failed to mount /proc. (%s)\n", 
+				strerror(errno));
+			return 1;
+		}
+
+		sprintf(mcos_sysdir, "/tmp/mcos/mcos%d_sys", mcosid);
+		mount_desc.dev_name = mcos_sysdir;
+		mount_desc.dir_name = "/sys";
+		mount_desc.type = NULL;
+		mount_desc.flags = MS_BIND;
+		mount_desc.data = NULL;
+		if (ioctl(fd, MCEXEC_UP_SYS_MOUNT, 
+			(unsigned long)&mount_desc) != 0) {
+			fprintf(stderr, "Error: Failed to mount /sys. (%s)\n", 
+				strerror(errno));
+			return 1;
+		}
+	} else if (error == -1) {
+		return 1;
+	}
+#else
+	__dprintf("mcoverlay disable\n");
+#endif
+
 	if (lookup_exec_path(argv[optind], path, sizeof(path)) != 0) {
 		fprintf(stderr, "error: finding file: %s\n", argv[optind]);
 		return 1;
@@ -1522,6 +1547,7 @@ void do_syscall_return(int fd, int cpu,
 {
 	struct syscall_ret_desc desc;
 
+	memset(&desc, '\0', sizeof desc);
 	desc.cpu = cpu;
 	desc.ret = ret;
 	desc.src = src;
@@ -1538,6 +1564,7 @@ void do_syscall_load(int fd, int cpu, unsigned long dest, unsigned long src,
 {
 	struct syscall_load_desc desc;
 
+	memset(&desc, '\0', sizeof desc);
 	desc.cpu = cpu;
 	desc.src = src;
 	desc.dest = dest;
@@ -1587,6 +1614,7 @@ static long do_strncpy_from_user(int fd, void *dest, void *src, unsigned long n)
 	struct strncpy_from_user_desc desc;
 	int ret;
 
+	memset(&desc, '\0', sizeof desc);
 	desc.dest = dest;
 	desc.src = src;
 	desc.n = n;
@@ -1718,6 +1746,7 @@ int main_loop(int fd, int cpu, pthread_mutex_t *lock)
 	char pathbuf[PATH_MAX];
 	char tmpbuf[PATH_MAX];
 
+	memset(&w, '\0', sizeof w);
 	w.cpu = cpu;
 	w.pid = getpid();
 
@@ -2140,6 +2169,7 @@ return_execve1:
 						fprintf(stderr, "execve(): error allocating desc\n");
 						goto return_execve2;
 					}
+					memset(desc, '\0', w.sr.args[2]);
 
 					/* Copy descriptor from co-kernel side */
 					trans.userp = (void*)desc;
@@ -2182,6 +2212,11 @@ return_execve2:
 			do_syscall_return(fd, cpu, ret, 0, 0, 0, 0);
 			break;
 
+		case __NR_perf_event_open:
+			ret = open("/dev/null", O_RDONLY);
+			do_syscall_return(fd, cpu, ret, 0, 0, 0, 0);
+			break;
+
 		case __NR_rt_sigaction:
 			act_sigaction(&w);
 			do_syscall_return(fd, cpu, 0, 0, 0, 0, 0);

kernel/host.c

@@ -242,7 +242,7 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 		p->args_len = args_len;
 	}
 
-	dkprintf("args copy, nr: %d\n", *((int*)args_envs_r));
+	dkprintf("args copy, nr: %d\n", *((long *)args_envs_r));
 
 	memcpy_long(args_envs, args_envs_r, p->args_len + sizeof(long) - 1);
 
@@ -277,7 +277,7 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 		p->envs_len = envs_len;
 	}
 
-	dkprintf("envs copy, nr: %d\n", *((int*)args_envs_r));
+	dkprintf("envs copy, nr: %d\n", *((long *)args_envs_r));
 
 	memcpy_long(args_envs + p->args_len, args_envs_r, p->envs_len + sizeof(long) - 1);
 
@@ -289,10 +289,10 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 	flush_tlb();
 
 	// Update variables
-	argc = *((int*)(args_envs));
+	argc = *((long *)(args_envs));
 	dkprintf("argc: %d\n", argc);
 
-	argv = (char **)(args_envs + (sizeof(int)));
+	argv = (char **)(args_envs + (sizeof(long)));
 	if(proc->saved_cmdline){
 		kfree(proc->saved_cmdline);
 		proc->saved_cmdline_len = 0;
@@ -309,17 +309,17 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 		*a = (char *)addr + (unsigned long)*a; // Process' address space!
 	}
 
-	envc = *((int*)(args_envs + p->args_len));
+	envc = *((long *)(args_envs + p->args_len));
 	dkprintf("envc: %d\n", envc);
 
-	env = (char **)(args_envs + p->args_len + sizeof(int));
+	env = (char **)(args_envs + p->args_len + sizeof(long));
 	while (*env) {
 		char **_env = env;
 		//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(int));
+	env = (char **)(args_envs + p->args_len + sizeof(long));
 
 	dkprintf("env OK\n");
 
@@ -449,7 +449,7 @@ static void process_msg_init(struct ikc_scd_init_param *pcp, struct syscall_para
 
 static void process_msg_init_acked(struct ihk_ikc_channel_desc *c, unsigned long pphys)
 {
-	struct ikc_scd_init_param *param = (void *)pphys;
+	struct ikc_scd_init_param *param = phys_to_virt(pphys);
 	struct syscall_params *lparam;
 	enum ihk_mc_pt_attribute attr;
 

kernel/include/process.h

@@ -312,18 +312,22 @@ struct vm_regions {
 
 struct process_vm;
 
-struct sigfd {
+struct mckfd {
-	struct sigfd *next;
+	struct mckfd *next;
 	int fd;
-	__sigset_t mask;
+	long data;
+	void *opt;
+	long (*read_cb)(struct mckfd *, ihk_mc_user_context_t *);
+	int (*ioctl_cb)(struct mckfd *, ihk_mc_user_context_t *);
+	int (*close_cb)(struct mckfd *, ihk_mc_user_context_t *);
 };
+
 #define SFD_CLOEXEC 02000000
 #define SFD_NONBLOCK 04000
 
 struct sig_common {
 	ihk_spinlock_t	lock;
 	ihk_atomic_t use;
-	struct sigfd *sigfd;
 	struct k_sigaction action[_NSIG];
 	struct list_head sigpending;
 };
@@ -343,7 +347,7 @@ typedef void pgio_func_t(void *arg);
  * special "init" process */
 struct process {
 	struct list_head hash_list;
-	mcs_rwlock_lock_t update_lock; // lock for parent, status, ...?
+	mcs_rwlock_lock_t update_lock; // lock for parent, status, cpu time...
 
 	// process vm
 	struct process_vm *vm;
@@ -425,6 +429,16 @@ struct process {
 	/* Store signal sent to parent when the process terminates. */
 	int termsig;
 
+	ihk_spinlock_t mckfd_lock;
+	struct mckfd *mckfd;
+
+	// cpu time (summary)
+	struct timespec stime;
+	struct timespec utime;
+
+	// cpu time (children)
+	struct timespec stime_children;
+	struct timespec utime_children;
 };
 
 void hold_thread(struct thread *ftn);
@@ -512,6 +526,20 @@ struct thread {
 	unsigned long *ptrace_debugreg;	/* debug registers for ptrace */
 	struct sig_pending *ptrace_recvsig;
 	struct sig_pending *ptrace_sendsig;
+
+	// cpu time
+	struct timespec stime;
+	struct timespec utime;
+	struct timespec btime;
+	int times_update;
+	int in_kernel;
+
+	// interval timers
+	int itimer_enabled;
+	struct itimerval itimer_virtual;
+	struct itimerval itimer_prof;
+	struct timespec itimer_virtual_value;
+	struct timespec itimer_prof_value;
 };
 
 struct process_vm {
@@ -615,5 +643,6 @@ 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 set_timer();
 
 #endif

kernel/include/syscall.h

@@ -66,6 +66,8 @@
 #define SCD_MSG_SYSFS_RESP_SETUP        0x41
 /* #define SCD_MSG_SYSFS_REQ_CLEANUP    0x42 */
 /* #define SCD_MSG_SYSFS_RESP_CLEANUP   0x43 */
+#define SCD_MSG_PROCFS_TID_CREATE	0x44
+#define SCD_MSG_PROCFS_TID_DELETE	0x45
 
 #define ARCH_SET_GS 0x1001
 #define ARCH_SET_FS 0x1002
@@ -314,6 +316,7 @@ struct procfs_read {
 	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) */
 };
 
@@ -334,4 +337,7 @@ struct tod_data_s {
 };
 extern struct tod_data_s tod_data;	/* residing in arch-dependent file */
 
+void reset_cputime();
+void set_cputime(int mode);
+
 #endif

kernel/include/time.h

@@ -53,5 +53,72 @@ struct timezone
     int tz_dsttime;		/* Nonzero if DST is ever in effect.  */
   };
 
+#define ITIMER_REAL	0
+#define ITIMER_VIRTUAL	1
+#define ITIMER_PROF	2
+
+struct itimerval {
+	struct timeval it_interval;
+	struct timeval it_value;
+};
+
+static inline void
+ts_add(struct timespec *ats, const struct timespec *bts)
+{
+	ats->tv_sec += bts->tv_sec;
+	ats->tv_nsec += bts->tv_nsec;
+	while(ats->tv_nsec >= 1000000000){
+		ats->tv_sec++;
+		ats->tv_nsec -= 1000000000;
+	}
+}
+
+static inline void
+ts_sub(struct timespec *ats, const struct timespec *bts)
+{
+	ats->tv_sec -= bts->tv_sec;
+	ats->tv_nsec -= bts->tv_nsec;
+	while(ats->tv_nsec < 0){
+		ats->tv_sec--;
+		ats->tv_nsec += 1000000000;
+	}
+}
+
+static inline void
+tv_add(struct timeval *ats, const struct timeval *bts)
+{
+	ats->tv_sec += bts->tv_sec;
+	ats->tv_usec += bts->tv_usec;
+	while(ats->tv_usec >= 1000000){
+		ats->tv_sec++;
+		ats->tv_usec -= 1000000;
+	}
+}
+
+static inline void
+tv_sub(struct timeval *ats, const struct timeval *bts)
+{
+	ats->tv_sec -= bts->tv_sec;
+	ats->tv_usec -= bts->tv_usec;
+	while(ats->tv_usec < 0){
+		ats->tv_sec--;
+		ats->tv_usec += 1000000;
+	}
+}
+
+static inline void
+tv_to_ts(struct timespec *ats, const struct timeval *bts)
+{
+	ats->tv_sec = bts->tv_sec;
+	ats->tv_nsec = bts->tv_usec * 1000;
+}
+
+static inline void
+ts_to_tv(struct timeval *ats, const struct timespec *bts)
+{
+	ats->tv_sec = bts->tv_sec;
+	ats->tv_usec = bts->tv_nsec / 1000;
+}
+
 #endif // __TIME_H
 

kernel/init.c

@@ -283,7 +283,6 @@ static void post_init(void)
 	zero_tsc();
 	ap_start();
 
-	create_os_procfs_files();
 	sysfs_init();
 	populate_sysfs();
 }

kernel/mem.c

@@ -53,6 +53,7 @@ static unsigned long pa_start, pa_end;
 static struct page *pa_pages;
 
 extern int ihk_mc_pt_print_pte(struct page_table *pt, void *virt);
+extern int interrupt_from_user(void *);
 
 struct tlb_flush_entry tlb_flush_vector[IHK_TLB_FLUSH_IRQ_VECTOR_SIZE];
 
@@ -369,6 +370,7 @@ static void page_fault_handler(void *fault_addr, uint64_t reason, void *regs)
 	struct thread *thread = cpu_local_var(current);
 	int error;
 
+	set_cputime(interrupt_from_user(regs)? 1: 2);
 	dkprintf("[%d]page_fault_handler(%p,%lx,%p)\n",
 			ihk_mc_get_processor_id(), fault_addr, reason, regs);
 
@@ -427,6 +429,7 @@ out:
 			ihk_mc_get_processor_id(), fault_addr, reason,
 			regs, error);
 	check_need_resched();
+	set_cputime(0);
 	return;
 }
 

kernel/process.c

@@ -63,6 +63,8 @@ extern int num_processors;
 extern ihk_spinlock_t cpuid_head_lock;
 int ptrace_detach(int pid, int data);
 extern unsigned long do_kill(struct thread *, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
+extern void procfs_create_thread(struct thread *);
+extern void procfs_delete_thread(struct thread *);
 
 struct list_head resource_set_list;
 mcs_rwlock_lock_t    resource_set_lock;
@@ -96,6 +98,7 @@ init_process(struct process *proc, struct process *parent)
 	INIT_LIST_HEAD(&proc->ptraced_children_list);
 	mcs_rwlock_init(&proc->threads_lock);
 	mcs_rwlock_init(&proc->children_lock);
+	ihk_mc_spinlock_init(&proc->mckfd_lock);
 	waitq_init(&proc->waitpid_q);
 	ihk_atomic_set(&proc->refcount, 2);
 }
@@ -2158,13 +2161,20 @@ void destroy_thread(struct thread *thread)
 void release_thread(struct thread *thread)
 {
 	struct process_vm *vm;
+	struct mcs_rwlock_node lock;
 
 	if (!ihk_atomic_dec_and_test(&thread->refcount)) {
 		return;
 	}
 
+	mcs_rwlock_writer_lock_noirq(&thread->proc->update_lock, &lock);
+	ts_add(&thread->proc->stime, &thread->stime);
+	ts_add(&thread->proc->utime, &thread->utime);
+	mcs_rwlock_writer_unlock_noirq(&thread->proc->update_lock, &lock);
+
 	vm = thread->vm;
 
+	procfs_delete_thread(thread);
 	destroy_thread(thread);
 
 	release_process_vm(vm);
@@ -2468,6 +2478,26 @@ ack:
 	ihk_mc_spinlock_unlock(&cur_v->migq_lock, irqstate);
 }
 
+void
+set_timer()
+{
+	struct cpu_local_var *v = get_this_cpu_local_var();
+
+	/* Toggle timesharing if CPU core is oversubscribed */
+	if (v->runq_len > 1 || v->current->itimer_enabled) {
+		if (!cpu_local_var(timer_enabled)) {
+			lapic_timer_enable(10000000);
+			cpu_local_var(timer_enabled) = 1;
+		}
+	}
+	else {
+		if (cpu_local_var(timer_enabled)) {
+			lapic_timer_disable();
+			cpu_local_var(timer_enabled) = 0;
+		}
+	}
+}
+
 void schedule(void)
 {
 	struct cpu_local_var *v;
@@ -2506,20 +2536,6 @@ redo:
 			list_add_tail(&prev->sched_list, &(v->runq));
 			++v->runq_len;
 		}
-
-		/* Toggle timesharing if CPU core is oversubscribed */
-		if (v->runq_len > 1) {
-			if (!cpu_local_var(timer_enabled)) {
-				lapic_timer_enable(10000000);
-				cpu_local_var(timer_enabled) = 1;
-			}
-		}
-		else {
-			if (cpu_local_var(timer_enabled)) {
-				lapic_timer_disable();
-				cpu_local_var(timer_enabled) = 0;
-			}
-		}
 	}
 
 	if (v->flags & CPU_FLAG_NEED_MIGRATE) {
@@ -2543,8 +2559,11 @@ redo:
 	if (prev != next) {
 		switch_ctx = 1;
 		v->current = next;
+		reset_cputime();
 	}
 
+	set_timer();
+
 	if (switch_ctx) {
 		dkprintf("schedule: %d => %d \n",
 		        prev ? prev->tid : 0, next ? next->tid : 0);
@@ -2752,7 +2771,7 @@ void runq_add_thread(struct thread *thread, int cpu_id)
 	__runq_add_thread(thread, cpu_id);
 	ihk_mc_spinlock_unlock(&(v->runq_lock), irqstate);
 
-	create_proc_procfs_files(thread->proc->pid, cpu_id);
+	procfs_create_thread(thread);
 
 	/* Kick scheduler */
 	if (cpu_id != ihk_mc_get_processor_id())

kernel/procfs.c

@@ -38,212 +38,37 @@ extern int sscanf(const char * buf, const char * fmt, ...);
 
 extern int osnum;
 
-void create_proc_procfs_files(int pid, int cpuid);
-void delete_proc_procfs_files(int pid);
-void create_os_procfs_files(void);
-void delete_os_procfs_files(void);
-
-static void create_proc_procfs_file(int pid, char *fname, int mode, int cpuid);
-static void delete_proc_procfs_file(int pid, char *fname);
-static void operate_proc_procfs_file(int pid, char *fname, int msg, int mode, int cpuid);
-
 int copy_from_user(void *dst, const void *src, size_t siz);
 int copy_to_user(void *dst, const void *src, size_t siz);
 
-/**
+static void
- * \brief Create all procfs files for process.
+procfs_thread_ctl(struct thread *thread, int msg)
- *
- * \param pid pid of the process
- * \param cpuid cpuid of the process
- */
-
-void create_proc_procfs_files(int pid, int cpuid)
-{
-	char fname[PROCFS_NAME_MAX];
-
-	dprintf("create procfs files:\n");
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/auxv", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0400, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/cmdline", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0444, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/mem", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0400, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/maps", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0444, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/pagemap", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0444, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/status", osnum, pid);
-	create_proc_procfs_file(pid, fname, 0444, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task/%d/mem", osnum, pid, pid);
-	create_proc_procfs_file(pid, fname, 0400, cpuid);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task/%d/stat", osnum, pid, pid);
-	create_proc_procfs_file(pid, fname, 0444, cpuid);
-
-	dprintf("create procfs files: done\n");
-}
-
-/**
- * \brief Create a procfs file for process.
- *
- * \param pid pid of the process
- * \param fname file name of the procfs file
- * \param mode file mode
- * \param cpuid cpuid of the process
- */
-
-static void create_proc_procfs_file(int pid, char *fname, int mode, int cpuid)
-{
-	dprintf("create procfs file: %s, mode: %o, cpuid: %d\n", fname, mode, cpuid);
-	operate_proc_procfs_file(pid, fname, SCD_MSG_PROCFS_CREATE, mode, cpuid);
-}
-
-/**
- * \brief Delete all procfs files for process.
- *
- * \param pid pid of the process
- */
-
-void delete_proc_procfs_files(int pid)
-{
-	char fname[PROCFS_NAME_MAX];
-
-	dprintf("delete procfs files for pid %d.\n", pid);
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task/%d/mem", osnum, pid, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task/%d/stat", osnum, pid, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task/%d", osnum, pid, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/task", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/mem", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/maps", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/status", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/pagemap", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/cmdline", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	snprintf(fname, PROCFS_NAME_MAX, "mcos%d/%d/auxv", osnum, pid);
-	delete_proc_procfs_file(pid, fname);
-
-	/* NOTE: Directory is removed on the host when mcexec drops the executable */
-	dprintf("delete procfs files for pid %d: done\n", pid);
-}
-
-/**
- * \brief Delete a procfs file for process.
- *
- * \param pid pid of the process
- * \param fname file name of the procfs file
- */
-
-static void delete_proc_procfs_file(int pid, char *fname)
-{
-	dprintf("delete procfs file: %s\n", fname);
-	operate_proc_procfs_file(pid, fname, SCD_MSG_PROCFS_DELETE, 0, 0);
-	dprintf("delete procfs file: %s done\n", fname);
-}
-
-/**
- * \brief create a procfs file for this operating system
- * \param fname relative path name from "host:/proc".
- * \param mode permissions of the file to be created
- *
- * Though operate_proc_procfs_file() is intended to create a process
- * specific file, it is reused to create a OS specific file by
- * specifying -1 as the pid parameter.
- */
-static void create_os_procfs_file(char *fname, int mode)
-{
-	const pid_t pid = -1;
-	const int msg = SCD_MSG_PROCFS_CREATE;
-	const int cpuid = ihk_mc_get_processor_id();	/* i.e. BSP */
-
-	operate_proc_procfs_file(pid, fname, msg, mode, cpuid);
-	return;
-}
-
-/**
- * \brief create all procfs files for this operating system
- */
-void create_os_procfs_files(void)
-{
-	char *fname = NULL;
-	size_t n;
-
-	fname = kmalloc(PROCFS_NAME_MAX, IHK_MC_AP_CRITICAL);
-
-	n = snprintf(fname, PROCFS_NAME_MAX, "mcos%d/stat", osnum);
-	if (n >= PROCFS_NAME_MAX) panic("/proc/stat");
-	create_os_procfs_file(fname, 0444);
-
-	return;
-}
-
-/**
- * \brief Create/delete a procfs file for process.
- *
- * \param pid pid of the process
- * \param fname file name of the procfs file
- * \param msg message (create/delete) 
- * \param mode file mode
- * \param cpuid cpuid of the process
- */
-
-static void operate_proc_procfs_file(int pid, char *fname, int msg, int mode, int cpuid)
 {
 	struct ihk_ikc_channel_desc *syscall_channel;
-	struct ikc_scd_packet pckt;
+	struct ikc_scd_packet packet;
-	struct procfs_file *f;
-	int ret;
 
 	syscall_channel = cpu_local_var(syscall_channel);
+	memset(&packet, '\0', sizeof packet);
+	packet.arg = thread->tid;
+	packet.msg = msg;
+	packet.osnum = osnum;
+	packet.ref = thread->cpu_id;
+	packet.pid = thread->proc->pid;
+	packet.err = 0;
 
-	f = kmalloc(sizeof(struct procfs_file), IHK_MC_AP_NOWAIT);
+	ihk_ikc_send(syscall_channel, &packet, 0);
-	if (!f) {
+}
-		kprintf("ERROR: not enough memory for dealing procfs file %s!",
-			fname);
-		return;
-	}
-	f->status = 0;
-	f->mode = mode;
-	strncpy(f->fname, fname, PROCFS_NAME_MAX);
-	pckt.arg = virt_to_phys(f);
-	pckt.msg = msg;
-	pckt.osnum = osnum;
-	pckt.ref = cpuid;
-	pckt.pid = pid;
-	pckt.err = 0;
 
-	ret = ihk_ikc_send(syscall_channel, &pckt, 0);
+void
-	if (ret < 0) {
+procfs_create_thread(struct thread *thread)
-		kprintf("ERROR: sending IKC msg, ret: %d\n", ret);
+{
-	}
+	procfs_thread_ctl(thread, SCD_MSG_PROCFS_TID_CREATE);
+}
 
-	while (f->status != 1) {
+void
-		cpu_pause();
+procfs_delete_thread(struct thread *thread)
-	}
+{
-	kfree(f);
+	procfs_thread_ctl(thread, SCD_MSG_PROCFS_TID_DELETE);
 }
 
 /**
@@ -251,12 +76,13 @@ static void operate_proc_procfs_file(int pid, char *fname, int msg, int mode, in
  *
  * \param rarg returned argument
  */
-
+void
-void process_procfs_request(unsigned long rarg)
+process_procfs_request(unsigned long rarg)
 {
 	unsigned long parg, pbuf;
-        struct thread *thread = cpu_local_var(current);
+        struct thread *thread = NULL;
-	struct process *proc = thread->proc;
+	struct process *proc = NULL;
+	struct process_vm *vm = NULL;
 	struct procfs_read *r;
 	struct ikc_scd_packet packet;
 	int rosnum, ret, pid, tid, ans = -EIO, eof = 0;
@@ -266,7 +92,7 @@ void process_procfs_request(unsigned long rarg)
 	unsigned long offset;
 	int count;
 	int npages;
-	int is_current = 1;	/* is 'proc' same as 'current'? */
+	int readwrite = 0;
 
 	dprintf("process_procfs_request: invoked.\n");
 
@@ -296,6 +122,7 @@ void process_procfs_request(unsigned long rarg)
 		goto bufunavail;
 	}
 
+	readwrite = r->readwrite;
 	count = r->count;
 	offset = r->offset;
 	dprintf("fname: %s, offset: %lx, count:%d.\n", r->fname, r->offset, r->count);
@@ -334,32 +161,47 @@ void process_procfs_request(unsigned long rarg)
 	 */
 	ret = sscanf(p, "%d/", &pid);
 	if (ret == 1) {
-		if (pid != cpu_local_var(current)->proc->pid) {
+		struct mcs_rwlock_node tlock;
-			/* We are not located in the proper cpu for some reason. */
+		int tids;
+		struct thread *thread1 = NULL;
 
-			dprintf("mismatched pid. We are %d, but requested pid is %d.\n",
+		proc = find_process(pid, &lock);
-				pid, cpu_local_var(current)->pid);
+		if(proc == NULL){
-			tid = pid;	/* main thread */
+			kprintf("process_procfs_request: no such pid %d\n", pid);
-			thread = find_thread(pid, tid, &lock);
+			goto end;
-			if (!thread) {
-				dprintf("We cannot find the proper cpu for requested pid.\n");
-				goto end;
-			}
-			else if (thread->cpu_id != ihk_mc_get_processor_id()) {
-				/* The target process has gone by migration. */
-				r->newcpu = thread->cpu_id;
-				dprintf("expected cpu id is %d.\n", thread->cpu_id);
-				thread_unlock(thread, &lock);
-				ans = 0;
-				goto end;
-			}
-			else {
-				thread_unlock(thread, &lock);
-				/* 'proc' is not 'current' */
-				is_current = 0;
-			}
-			proc = thread->proc;
 		}
+		p = strchr(p, '/') + 1;
+		if((tids = sscanf(p, "task/%d/", &tid)) == 1){
+			p = strchr(p, '/') + 1;
+			p = strchr(p, '/') + 1;
+		}
+		else
+			tid = pid;
+
+		mcs_rwlock_reader_lock_noirq(&proc->threads_lock, &tlock);
+		list_for_each_entry(thread, &proc->threads_list, siblings_list){
+			if(thread->tid == tid)
+				break;
+			if(!thread1)
+				thread1 = thread;
+		}
+		if(thread == NULL){
+			kprintf("process_procfs_request: no such tid %d-%d\n", pid, tid);
+			if(tids){
+				process_unlock(proc, &lock);
+				mcs_rwlock_reader_unlock_noirq(&proc->threads_lock, &tlock);
+				goto end;
+			}
+			thread = thread1;
+		}
+		if(thread)
+			hold_thread(thread);
+		mcs_rwlock_reader_unlock_noirq(&proc->threads_lock, &tlock);
+		hold_process(proc);
+		vm = proc->vm;
+		if(vm)
+			hold_process_vm(vm);
+		process_unlock(proc, &lock);
 	}
 	else if (!strcmp(p, "stat")) {	/* "/proc/stat" */
 		extern int num_processors;	/* kernel/ap.c */
@@ -390,10 +232,9 @@ void process_procfs_request(unsigned long rarg)
 		goto end;
 	}
 	else {
+		kprintf("unsupported procfs entry: %s\n", p);
 		goto end;
 	}
-	dprintf("matched PID: %d.\n", pid);
-	p = strchr(p, '/') + 1;
 
 	/* 
 	 * mcos%d/PID/mem
@@ -402,74 +243,55 @@ void process_procfs_request(unsigned long rarg)
 	 * of the process. The count is the length of the area.
 	 */
 	if (strcmp(p, "mem") == 0) {
-		struct vm_range *range;
+		uint64_t reason = PF_POPULATE | PF_WRITE | PF_USER;
-		struct process_vm *vm = proc->vm;
+		unsigned long offset = r->offset;
+		unsigned long left = r->count;
+		int ret;
+		struct page_table *pt = vm->address_space->page_table;
 
-		if (!is_current) {
+		ans = 0;
-			uint64_t reason = PF_POPULATE | PF_WRITE | PF_USER;
+		if(left == 0)
-			unsigned long offset = r->offset;
+			goto end;
-			unsigned long left = r->count;
-			int ret;
 
-			ans = 0;
+#if 0
-			if(left == 0)
+		if(!(proc->ptrace & PT_TRACED) ||
-				goto end;
+		   !(proc->status & (PS_STOPPED | PS_TRACED))){
-
+			ans = -EIO;
-			while(left){
+			goto end;
-				unsigned long pa;
-				char *va;
-				int pos = offset & (PAGE_SIZE - 1);
-				int size = PAGE_SIZE - pos;
-
-				if(size > left)
-					size = left;
-				ret = page_fault_process_vm(proc->vm,
-				                (void *)offset, reason);
-				if(ret){
-					if(ans == 0)
-						ans = -EIO;
-					goto end;
-				}
-				ret = ihk_mc_pt_virt_to_phys(vm->address_space->page_table,
-				                (void *)offset, &pa);
-				if(ret){
-					if(ans == 0)
-						ans = -EIO;
-					goto end;
-				}
-				va = phys_to_virt(pa);
-				memcpy(buf + ans, va, size);
-				offset += size;
-				left -= size;
-				ans += size;
-			}
 		}
-		else{
+#endif
-			unsigned long offset = r->offset;
+
-			unsigned long left = r->count;
+		if(readwrite == 0)
-			unsigned long pos;
+			reason = PF_POPULATE | PF_USER;
-			unsigned long l;
+
-			ans = 0;
+		while(left){
-			list_for_each_entry(range, &vm->vm_range_list, list) {
+			unsigned long pa;
-				dprintf("range: %lx - %lx\n", range->start, range->end);
+			char *va;
-				while (left &&
+			int pos = offset & (PAGE_SIZE - 1);
-				       (range->start <= offset) && 
+			int size = PAGE_SIZE - pos;
-				       (offset < range->end)) {
+
-					pos = offset & (PAGE_SIZE - 1);
+			if(size > left)
-					l = PAGE_SIZE - pos;
+				size = left;
-					if(l > left)
+			ret = page_fault_process_vm(vm, (void *)offset, reason);
-						l = left;
+			if(ret){
-					if(copy_from_user(buf, (void *)offset, l)){
+				if(ans == 0)
-						if(ans == 0)
+					ans = -EIO;
-							ans = -EIO;
+				goto end;
-						goto end;
-					}
-					buf += l;
-					ans += l;
-					offset += l;
-					left -= l;
-				}
 			}
+			ret = ihk_mc_pt_virt_to_phys(pt, (void *)offset, &pa);
+			if(ret){
+				if(ans == 0)
+					ans = -EIO;
+				goto end;
+			}
+			va = phys_to_virt(pa);
+			if(readwrite)
+				memcpy(va, buf + ans, size);
+			else
+				memcpy(buf + ans, va, size);
+			offset += size;
+			left -= size;
+			ans += size;
 		}
 		goto end;
 	}
@@ -479,7 +301,6 @@ void process_procfs_request(unsigned long rarg)
 	 */
 	if (strcmp(p, "maps") == 0) {
 		struct vm_range *range;
-		struct process_vm *vm = proc->vm;
 		int left = r->count - 1; /* extra 1 for terminating NULL */
 		int written = 0;
 		char *_buf = buf;
@@ -537,7 +358,6 @@ void process_procfs_request(unsigned long rarg)
 	 * mcos%d/PID/pagemap
 	 */
 	if (strcmp(p, "pagemap") == 0) {
-		struct process_vm *vm = proc->vm;
 		uint64_t *_buf = (uint64_t *)buf;
 		uint64_t start, end;
 		
@@ -620,7 +440,7 @@ void process_procfs_request(unsigned long rarg)
 	 * mcos%d/PID/auxv
 	 */
 	if (strcmp(p, "auxv") == 0) {
-		unsigned int limit = AUXV_LEN * sizeof(int);
+		unsigned int limit = AUXV_LEN * sizeof(unsigned long);
 		unsigned int len = r->count;
 		if (r->offset < limit) {
 			if (limit < r->offset + r->count) {
@@ -673,112 +493,71 @@ void process_procfs_request(unsigned long rarg)
 	 * The offset is treated as the beginning of the virtual address area
 	 * of the process. The count is the length of the area.
 	 */
-	tid = pid;
-	ret = sscanf(p, "task/%d/", &tid);
-	if (ret == 1) {
-		p = strchr(p, '/') + 1;
-		p = strchr(p, '/') + 1;
 
-		if (!strcmp(p, "mem")){
+	if (!strcmp(p, "stat")) {
-			struct vm_range *range;
+		char tmp[1024];
-			struct process_vm *vm = proc->vm;
+		int len;
 
-			if (!is_current) {
+		/*
-				goto end;
+		 * pid (comm) state ppid
+		 * pgrp session tty_nr tpgid
+		 * flags minflt cminflt majflt
+		 * cmajflt utime stime cutime
+		 * cstime priority nice num_threads
+		 * itrealvalue starttime vsize rss
+		 * rsslim startcode endcode startstack
+		 * kstkesp kstkeip signal blocked
+		 * sigignore sigcatch wchan nswap
+		 * cnswap exit_signal processor rt_priority
+		 * policy delayacct_blkio_ticks guest_time cguest_time
+		 */
+		ans = sprintf(tmp,
+		    "%d (%s) %c %d "	      // pid...
+		    "%d %d %d %d "	      // pgrp...
+		    "%u %lu %lu %lu "	      // flags...
+		    "%lu %lu %lu %ld "	      // cmajflt...
+		    "%ld %ld %ld %ld "	      // cstime...
+		    "%ld %llu %lu %ld "	      // itrealvalue...
+		    "%lu %lu %lu %lu "	      // rsslim...
+		    "%lu %lu %lu %lu "	      // kstkesp...
+		    "%lu %lu %lu %lu "	      // sigignore...
+		    "%lu %d %d %u "	      // cnswap...
+		    "%u %llu %lu %ld\n",      // policy...
+		    0, "exe", 'R', 0,	      // pid...
+		    0, 0, 0, 0,      	      // pgrp...
+		    0, 0L, 0L, 0L,	      // flags...
+		    0L, 0L, 0L, 0L,	      // cmajflt...
+		    0L, 0L, 0L, 0L,	      // cstime...
+		    0L, 0LL, 0L, 0L,	      // itrealvalue...
+		    0L, 0L, 0L, 0L,	      // rsslim...
+		    0L, 0L, 0L, 0L,	      // kstkesp...
+		    0L, 0L, 0L, 0L,	      // sigignore...
+		    0L, 0, thread->cpu_id, 0, // cnswap...
+		    0, 0LL, 0L, 0L	      // policy...
+		);
+		dprintf("tmp=%s\n", tmp);
+
+		len = strlen(tmp);
+		if (r->offset < len) {
+			if (r->offset + r->count < len) {
+				ans = r->count;
+			} else {
+				eof = 1;
+				ans = len;
 			}
-			if (pid != tid) {
+			strncpy(buf, tmp + r->offset, ans);
-				/* We are not multithreaded yet. */
+		} else if (r->offset == len) {
-				goto end;
+			ans = 0;
-			} 
+			eof = 1;
-			list_for_each_entry(range, &vm->vm_range_list, list) {
-				dprintf("range: %lx - %lx\n", range->start, range->end);
-				if ((range->start <= r->offset) && 
-				    (r->offset < range->end)) {
-					unsigned int len = r->count;
-					if (range->end < r->offset + r->count) {
-						len = range->end - r->offset;
-					}
-					memcpy((void *)buf, (void *)range->start, len);
-					ans = len;
-					break;
-				}
-			}
-			goto end;
 		}
-
-		if (!strcmp(p, "stat")) {
-			char tmp[1024];
-			int len;
-
-			if ((thread = find_thread(pid, tid, &lock))){
-				dprintf("thread found! pid=%d tid=%d\n", pid, tid);
-				/*
-				 * pid (comm) state ppid
-				 * pgrp session tty_nr tpgid
-				 * flags minflt cminflt majflt
-				 * cmajflt utime stime cutime
-				 * cstime priority nice num_threads
-				 * itrealvalue starttime vsize rss
-				 * rsslim startcode endcode startstack
-				 * kstkesp kstkeip signal blocked
-				 * sigignore sigcatch wchan nswap
-				 * cnswap exit_signal processor rt_priority
-				 * policy delayacct_blkio_ticks guest_time cguest_time
-				 */
-				ans = sprintf(tmp,
-				    "%d (%s) %c %d "	      // pid...
-				    "%d %d %d %d "	      // pgrp...
-				    "%u %lu %lu %lu "	      // flags...
-				    "%lu %lu %lu %ld "	      // cmajflt...
-				    "%ld %ld %ld %ld "	      // cstime...
-				    "%ld %llu %lu %ld "	      // itrealvalue...
-				    "%lu %lu %lu %lu "	      // rsslim...
-				    "%lu %lu %lu %lu "	      // kstkesp...
-				    "%lu %lu %lu %lu "	      // sigignore...
-				    "%lu %d %d %u "	      // cnswap...
-				    "%u %llu %lu %ld\n",      // policy...
-				    0, "exe", 'R', 0,	      // pid...
-				    0, 0, 0, 0,      	      // pgrp...
-				    0, 0L, 0L, 0L,	      // flags...
-				    0L, 0L, 0L, 0L,	      // cmajflt...
-				    0L, 0L, 0L, 0L,	      // cstime...
-				    0L, 0LL, 0L, 0L,	      // itrealvalue...
-				    0L, 0L, 0L, 0L,	      // rsslim...
-				    0L, 0L, 0L, 0L,	      // kstkesp...
-				    0L, 0L, 0L, 0L,	      // sigignore...
-				    0L, 0, thread->cpu_id, 0, // cnswap...
-				    0, 0LL, 0L, 0L	      // policy...
-				);
-				thread_unlock(thread, &lock);
-				dprintf("tmp=%s\n", tmp);
-
-				len = strlen(tmp);
-				if (r->offset < len) {
-					if (r->offset + r->count < len) {
-						ans = r->count;
-					} else {
-						eof = 1;
-						ans = len;
-					}
-					strncpy(buf, tmp + r->offset, ans);
-				} else if (r->offset == len) {
-					ans = 0;
-					eof = 1;
-				}
-				goto end;
-			}
-			else{
-				dprintf("no thread found pid=%d tid=%d\n", pid, tid);
-			}
-		}
-		dprintf("could not find a matching entry for task/%d/%s.\n", tid, p); 
 		goto end;
 	}
 
-	/* 
+	if(thread)
-	 * Processing for pattern "mcos%d/PID/xxx" files should be here.
+		kprintf("unsupported procfs entry: %d/task/%d/%s\n", pid, tid, p);
-	*/
+	else
-	dprintf("could not find a matching entry for %s.\n", p); 
+		kprintf("unsupported procfs entry: %d/%s\n", pid, p);
+
 end:
 	ihk_mc_unmap_virtual(buf, npages, 0);
 	dprintf("ret: %d, eof: %d\n", ans, eof);
@@ -799,6 +578,12 @@ dataunavail:
 	if (ret < 0) {
 		kprintf("ERROR: sending IKC msg, ret: %d\n", ret);
 	}
+	if(proc)
+		release_process(proc);
+	if(thread)
+		release_thread(thread);
+	if(vm)
+		release_process_vm(vm);
 	return;
 
 }

kernel/syscall.c

@@ -113,6 +113,7 @@ extern unsigned long ihk_mc_get_ns_per_tsc(void);
 extern int ptrace_detach(int pid, int data);
 extern void debug_log(unsigned long);
 extern void free_all_process_memory_range(struct process_vm *vm);
+extern struct cpu_local_var *clv;
 
 int prepare_process_ranges_args_envs(struct thread *thread, 
 		struct program_load_desc *pn,
@@ -401,6 +402,12 @@ do_wait(int pid, int *status, int options, void *rusage)
 				ret = wait_zombie(thread, child, status, options);
 				mcs_rwlock_writer_unlock_noirq(&thread->proc->children_lock, &lock);
 				if(!(options & WNOWAIT)){
+					mcs_rwlock_writer_lock_noirq(&proc->update_lock, &lock);
+					ts_add(&proc->stime_children, &child->stime);
+					ts_add(&proc->utime_children, &child->utime);
+					ts_add(&proc->stime_children, &child->stime_children);
+					ts_add(&proc->utime_children, &child->utime_children);
+					mcs_rwlock_writer_unlock_noirq(&proc->update_lock, &lock);
 					release_process(child);
 				}
 				goto out_found;
@@ -632,8 +639,6 @@ terminate(int rc, int sig)
 	list_add_tail(&mythread->siblings_list, &proc->threads_list);
 	mcs_rwlock_writer_unlock(&proc->threads_lock, &lock);
 
-	delete_proc_procfs_files(proc->pid);
-
 	vm = proc->vm;
 	free_all_process_memory_range(vm);
 
@@ -2036,6 +2041,49 @@ SYSCALL_DECLARE(set_tid_address)
 	return cpu_local_var(current)->proc->pid;
 }
 
+static unsigned long
+timespec_to_jiffy(const struct timespec *ats)
+{
+	return ats->tv_sec * 100 + ats->tv_nsec / 10000000;
+}
+
+SYSCALL_DECLARE(times)
+{
+	struct tms {
+		unsigned long tms_utime;
+		unsigned long tms_stime;
+		unsigned long tms_cutime;
+		unsigned long tms_cstime;
+	};
+	struct tms mytms;
+	struct tms *buf = (struct tms *)ihk_mc_syscall_arg0(ctx);
+	struct thread *thread = cpu_local_var(current);
+	struct process *proc = thread->proc;
+	struct timespec ats;
+
+	mytms.tms_utime = timespec_to_jiffy(&thread->utime);
+	mytms.tms_stime = timespec_to_jiffy(&thread->stime);
+	ats.tv_sec = proc->utime.tv_sec;
+	ats.tv_nsec = proc->utime.tv_nsec;
+	ts_add(&ats, &proc->utime_children);
+	mytms.tms_cutime = timespec_to_jiffy(&ats);
+	ats.tv_sec = proc->stime.tv_sec;
+	ats.tv_nsec = proc->stime.tv_nsec;
+	ts_add(&ats, &proc->stime_children);
+	mytms.tms_cstime = timespec_to_jiffy(&ats);
+	if(copy_to_user(buf, &mytms, sizeof mytms))
+		return -EFAULT;
+	if(gettime_local_support){
+		calculate_time_from_tsc(&ats);
+	}
+	else{
+		ats.tv_sec = 0;
+		ats.tv_nsec = 0;
+	}
+
+	return timespec_to_jiffy(&ats);
+}
+
 SYSCALL_DECLARE(kill)
 {
 	int pid = ihk_mc_syscall_arg0(ctx);
@@ -2077,6 +2125,24 @@ SYSCALL_DECLARE(tgkill)
 	return do_kill(thread, tgid, tid, sig, &info, 0);
 }
 
+SYSCALL_DECLARE(tkill)
+{
+	int tid = ihk_mc_syscall_arg0(ctx);
+	int sig = ihk_mc_syscall_arg1(ctx);
+	struct thread *thread = cpu_local_var(current);
+	struct siginfo info;
+
+	memset(&info, '\0', sizeof info);
+	info.si_signo = sig;
+	info.si_code = SI_TKILL;
+	info._sifields._kill.si_pid = thread->proc->pid;
+
+	if(tid <= 0)
+		return -EINVAL;
+
+	return do_kill(thread, -1, tid, sig, &info, 0);
+}
+
 int *
 getcred(int *_buf)
 {
@@ -2359,34 +2425,85 @@ do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
 	return 0;
 }
 
+SYSCALL_DECLARE(read)
+{
+	int fd = ihk_mc_syscall_arg0(ctx);
+	long rc;
+	struct thread *thread = cpu_local_var(current);
+	struct process *proc = thread->proc;
+	struct mckfd *fdp;
+	long irqstate;
+
+	irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
+	for(fdp = proc->mckfd; fdp; fdp = fdp->next)
+		if(fdp->fd == fd)
+			break;
+	ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
+
+	if(fdp && fdp->read_cb){
+kprintf("read: found system fd %d\n", fd);
+		rc = fdp->read_cb(fdp, ctx);
+	}
+	else{
+		rc = syscall_generic_forwarding(__NR_read, ctx);
+	}
+	return rc;
+}
+
+SYSCALL_DECLARE(ioctl)
+{
+	int fd = ihk_mc_syscall_arg0(ctx);
+	long rc;
+	struct thread *thread = cpu_local_var(current);
+	struct process *proc = thread->proc;
+	struct mckfd *fdp;
+	long irqstate;
+
+	irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
+	for(fdp = proc->mckfd; fdp; fdp = fdp->next)
+		if(fdp->fd == fd)
+			break;
+	ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
+
+	if(fdp && fdp->ioctl_cb){
+kprintf("ioctl: found system fd %d\n", fd);
+		rc = fdp->ioctl_cb(fdp, ctx);
+	}
+	else{
+		rc = syscall_generic_forwarding(__NR_ioctl, ctx);
+	}
+	return rc;
+}
+
 SYSCALL_DECLARE(close)
 {
 	int fd = ihk_mc_syscall_arg0(ctx);
-	int rc;
+	long rc;
 	struct thread *thread = cpu_local_var(current);
-	struct sigfd *sfd;
+	struct process *proc = thread->proc;
-	struct sigfd *sb;
+	struct mckfd *fdp;
-	long	irqstate;
+	struct mckfd *fdq;
+	long irqstate;
 
-	irqstate = ihk_mc_spinlock_lock(&thread->sigcommon->lock);
+	irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
-	for(sfd = thread->sigcommon->sigfd, sb = NULL; sfd; sb = sfd, sfd = sfd->next)
+	for(fdp = proc->mckfd, fdq = NULL; fdp; fdq = fdp, fdp = fdp->next)
-		if(sfd->fd == fd)
+		if(fdp->fd == fd)
 			break;
-	if(sfd){
+
-		struct syscall_request request IHK_DMA_ALIGN;
+	if(fdp){
-		if(sb)
+kprintf("close: found system fd %d pid=%d\n", fd, proc->pid);
-			sb->next = sfd->next;
+		if(fdq)
+			fdq->next = fdp->next;
 		else
-			thread->sigcommon->sigfd = sfd->next;
+			proc->mckfd = fdp->next;
-		ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+		ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
-		request.number = __NR_signalfd4;
+		if(fdp->close_cb)
-		request.args[0] = 1;
+			fdp->close_cb(fdp, ctx);
-		request.args[1] = sfd->fd;
+		kfree(fdp);
-		kfree(sfd);
+		rc = syscall_generic_forwarding(__NR_close, ctx);
-		rc = do_syscall(&request, ihk_mc_get_processor_id(), 0);
 	}
 	else{
-		ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+		ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
 		rc = syscall_generic_forwarding(__NR_close, ctx);
 	}
 	return rc;
@@ -2487,7 +2604,8 @@ SYSCALL_DECLARE(signalfd4)
 {
 	int fd = ihk_mc_syscall_arg0(ctx);
 	struct thread *thread = cpu_local_var(current);
-	struct sigfd *sfd;
+	struct process *proc = thread->proc;
+	struct mckfd *sfd;
 	long    irqstate;
 	sigset_t *maskp = (sigset_t *)ihk_mc_syscall_arg1(ctx);;
 	__sigset_t mask;
@@ -2501,10 +2619,9 @@ SYSCALL_DECLARE(signalfd4)
 	if(flags & ~(SFD_NONBLOCK | SFD_CLOEXEC))
 		return -EINVAL;
 
-	irqstate = ihk_mc_spinlock_lock(&thread->sigcommon->lock);
 	if(fd == -1){
 		struct syscall_request request IHK_DMA_ALIGN;
-		ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+
 		request.number = __NR_signalfd4;
 		request.args[0] = 0;
 		request.args[1] = flags;
@@ -2512,25 +2629,73 @@ SYSCALL_DECLARE(signalfd4)
 		if(fd < 0){
 			return fd;
 		}
-		sfd = kmalloc(sizeof(struct sigfd), IHK_MC_AP_NOWAIT);
+		sfd = kmalloc(sizeof(struct mckfd), IHK_MC_AP_NOWAIT);
 		if(!sfd)
 			return -ENOMEM;
 		sfd->fd = fd;
-		irqstate = ihk_mc_spinlock_lock(&thread->sigcommon->lock);
+		irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
-		sfd->next = thread->sigcommon->sigfd;
+		sfd->next = proc->mckfd;
-		thread->sigcommon->sigfd = sfd;
+		proc->mckfd = sfd;
 	}
 	else{
-		for(sfd = thread->sigcommon->sigfd; sfd; sfd = sfd->next)
+		irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
+		for(sfd = proc->mckfd; sfd; sfd = sfd->next)
 			if(sfd->fd == fd)
 				break;
 		if(!sfd){
-			ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+			ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
 			return -EINVAL;
 		}
 	}
-	memcpy(&sfd->mask, &mask, sizeof mask);
+	memcpy(&sfd->data, &mask, sizeof mask);
-	ihk_mc_spinlock_unlock(&thread->sigcommon->lock, irqstate);
+	ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
+	return sfd->fd;
+}
+
+static long
+perf_event_read(struct mckfd *sfd, ihk_mc_user_context_t *ctx)
+{
+	return 0;
+}
+
+static int
+perf_event_ioctl(struct mckfd *sfd, ihk_mc_user_context_t *ctx)
+{
+	return 0;
+}
+
+static int
+perf_event_close(struct mckfd *sfd, ihk_mc_user_context_t *ctx)
+{
+	return 0;
+}
+
+SYSCALL_DECLARE(perf_event_open)
+{
+	struct syscall_request request IHK_DMA_ALIGN;
+	struct thread *thread = cpu_local_var(current);
+	struct process *proc = thread->proc;
+	struct mckfd *sfd;
+	int fd;
+	long irqstate;
+
+	request.number = __NR_perf_event_open;
+	request.args[0] = 0;
+	fd = do_syscall(&request, ihk_mc_get_processor_id(), 0);
+	if(fd < 0){
+		return fd;
+	}
+	sfd = kmalloc(sizeof(struct mckfd), IHK_MC_AP_NOWAIT);
+	if(!sfd)
+		return -ENOMEM;
+	sfd->fd = fd;
+	sfd->read_cb = perf_event_read;
+	sfd->ioctl_cb = perf_event_ioctl;
+	sfd->close_cb = perf_event_close;
+	irqstate = ihk_mc_spinlock_lock(&proc->mckfd_lock);
+	sfd->next = proc->mckfd;
+	proc->mckfd = sfd;
+	ihk_mc_spinlock_unlock(&proc->mckfd_lock, irqstate);
 	return sfd->fd;
 }
 
@@ -5073,6 +5238,125 @@ static void calculate_time_from_tsc(struct timespec *ts)
 	return;
 }
 
+SYSCALL_DECLARE(setitimer)
+{
+	int which = (int)ihk_mc_syscall_arg0(ctx);
+	struct itimerval *new = (struct itimerval *)ihk_mc_syscall_arg1(ctx);
+	struct itimerval *old = (struct itimerval *)ihk_mc_syscall_arg2(ctx);
+	struct syscall_request request IHK_DMA_ALIGN;
+	struct thread *thread = cpu_local_var(current);
+	int timer_start = 1;
+	struct itimerval wkval;
+	struct timeval tv;
+
+	if(which != ITIMER_REAL &&
+	   which != ITIMER_VIRTUAL &&
+	   which != ITIMER_PROF)
+		return -EINVAL;
+
+	if(which == ITIMER_REAL){
+		request.number = __NR_setitimer;
+		request.args[0] = ihk_mc_syscall_arg0(ctx);
+		request.args[1] = ihk_mc_syscall_arg1(ctx);
+		request.args[2] = ihk_mc_syscall_arg2(ctx);
+
+		return do_syscall(&request, ihk_mc_get_processor_id(), 0);
+	}
+	else if(which == ITIMER_VIRTUAL){
+		if(old){
+			memcpy(&wkval, &thread->itimer_virtual, sizeof wkval);
+			if(wkval.it_value.tv_sec != 0 ||
+			   wkval.it_value.tv_usec != 0){
+				ts_to_tv(&tv, &thread->itimer_virtual_value);
+				tv_sub(&wkval.it_value, &tv);
+			}
+			if(copy_to_user(old, &wkval, sizeof wkval))
+				return -EFAULT;
+		}
+		if(!new){
+			return 0;
+		}
+		if(copy_from_user(&thread->itimer_virtual, new, sizeof(struct itimerval)))
+		thread->itimer_virtual_value.tv_sec = 0;
+		thread->itimer_virtual_value.tv_nsec = 0;
+		if(thread->itimer_virtual.it_value.tv_sec == 0 &&
+		   thread->itimer_virtual.it_value.tv_usec == 0)
+			timer_start = 0;
+	}
+	else if(which == ITIMER_PROF){
+		if(old){
+			memcpy(&wkval, &thread->itimer_prof, sizeof wkval);
+			if(wkval.it_value.tv_sec != 0 ||
+			   wkval.it_value.tv_usec != 0){
+				ts_to_tv(&tv, &thread->itimer_prof_value);
+				tv_sub(&wkval.it_value, &tv);
+			}
+			if(copy_to_user(old, &wkval, sizeof wkval))
+				return -EFAULT;
+		}
+		if(!new){
+			return 0;
+		}
+		if(copy_from_user(&thread->itimer_prof, new, sizeof(struct itimerval)))
+		thread->itimer_prof_value.tv_sec = 0;
+		thread->itimer_prof_value.tv_nsec = 0;
+		if(thread->itimer_prof.it_value.tv_sec == 0 &&
+		   thread->itimer_prof.it_value.tv_usec == 0)
+			timer_start = 0;
+	}
+	thread->itimer_enabled = timer_start;
+	set_timer();
+	return 0;
+}
+
+SYSCALL_DECLARE(getitimer)
+{
+	int which = (int)ihk_mc_syscall_arg0(ctx);
+	struct itimerval *old = (struct itimerval *)ihk_mc_syscall_arg1(ctx);
+	struct syscall_request request IHK_DMA_ALIGN;
+	struct thread *thread = cpu_local_var(current);
+	struct itimerval wkval;
+	struct timeval tv;
+
+	if(which != ITIMER_REAL &&
+	   which != ITIMER_VIRTUAL &&
+	   which != ITIMER_PROF)
+		return -EINVAL;
+
+	if(which == ITIMER_REAL){
+		request.number = __NR_getitimer;
+		request.args[0] = ihk_mc_syscall_arg0(ctx);
+		request.args[1] = ihk_mc_syscall_arg1(ctx);
+
+		return do_syscall(&request, ihk_mc_get_processor_id(), 0);
+	}
+	else if(which == ITIMER_VIRTUAL){
+		if(old){
+			memcpy(&wkval, &thread->itimer_virtual, sizeof wkval);
+			if(wkval.it_value.tv_sec != 0 ||
+			   wkval.it_value.tv_usec != 0){
+				ts_to_tv(&tv, &thread->itimer_virtual_value);
+				tv_sub(&wkval.it_value, &tv);
+			}
+			if(copy_to_user(old, &wkval, sizeof wkval))
+				return -EFAULT;
+		}
+	}
+	else if(which == ITIMER_PROF){
+		if(old){
+			memcpy(&wkval, &thread->itimer_prof, sizeof wkval);
+			if(wkval.it_value.tv_sec != 0 ||
+			   wkval.it_value.tv_usec != 0){
+				ts_to_tv(&tv, &thread->itimer_prof_value);
+				tv_sub(&wkval.it_value, &tv);
+			}
+			if(copy_to_user(old, &wkval, sizeof wkval))
+				return -EFAULT;
+		}
+	}
+	return 0;
+}
+
 SYSCALL_DECLARE(clock_gettime)
 {
 	/* TODO: handle clock_id */
@@ -5096,6 +5380,41 @@ SYSCALL_DECLARE(clock_gettime)
 		dkprintf("clock_gettime(): %d\n", error);
 		return error;
 	}
+	else if(clock_id == CLOCK_PROCESS_CPUTIME_ID){
+		struct thread *thread = cpu_local_var(current);
+		struct process *proc = thread->proc;
+		struct thread *child;
+		struct mcs_rwlock_node lock;
+
+		mcs_rwlock_reader_lock_noirq(&proc->children_lock, &lock);
+		list_for_each_entry(child, &proc->threads_list, siblings_list){
+			if(child != thread &&
+			   child->status == PS_RUNNING &&
+			   !child->in_kernel){
+				child->times_update = 0;
+				ihk_mc_interrupt_cpu(get_x86_cpu_local_variable(child->cpu_id)->apic_id, 0xd1);
+			}
+		}
+		ats.tv_sec = proc->utime.tv_sec;
+		ats.tv_nsec = proc->utime.tv_nsec;
+		ts_add(&ats, &proc->stime);
+		list_for_each_entry(child, &proc->threads_list, siblings_list){
+			while(!child->times_update)
+				cpu_pause();
+			ts_add(&ats, &child->utime);
+			ts_add(&ats, &child->stime);
+		}
+		mcs_rwlock_reader_unlock_noirq(&proc->children_lock, &lock);
+		return copy_to_user(ts, &ats, sizeof ats);
+	}
+	else if(clock_id == CLOCK_THREAD_CPUTIME_ID){
+		struct thread *thread = cpu_local_var(current);
+
+		ats.tv_sec = thread->utime.tv_sec;
+		ats.tv_nsec = thread->utime.tv_nsec;
+		ts_add(&ats, &thread->stime);
+		return copy_to_user(ts, &ats, sizeof ats);
+	}
 
 	/* Otherwise offload */
 	request.number = __NR_clock_gettime;
@@ -6489,13 +6808,136 @@ SYSCALL_DECLARE(pmc_reset)
     return ihk_mc_perfctr_reset(counter);
 }
 
+void
+reset_cputime()
+{
+	struct thread *thread;
+
+	if(clv == NULL)
+		return;
+
+	if(!(thread = cpu_local_var(current)))
+		return;
+
+	thread->btime.tv_sec = 0;
+	thread->btime.tv_nsec = 0;
+}
+
+/**
+ * mode == 0: kernel -> user
+ * mode == 1: user -> kernel
+ * mode == 2: kernel -> kernel
+ */
+void
+set_cputime(int mode)
+{
+	struct thread *thread;
+	struct timespec ats;
+	struct cpu_local_var *v;
+
+	if(clv == NULL)
+		return;
+
+	v = get_this_cpu_local_var();
+	if(!(thread = v->current))
+		return;
+
+	if(!gettime_local_support){
+		thread->times_update = 1;
+		return;
+	}
+
+	calculate_time_from_tsc(&ats);
+	if(thread->btime.tv_sec != 0 && thread->btime.tv_nsec != 0){
+		struct timespec dts;
+
+		dts.tv_sec = ats.tv_sec;
+		dts.tv_nsec = ats.tv_nsec;
+		ts_sub(&dts, &thread->btime);
+		if(mode == 1){
+			ts_add(&thread->utime, &dts);
+			ts_add(&thread->itimer_virtual_value, &dts);
+			ts_add(&thread->itimer_prof_value, &dts);
+		}
+		else{
+			ts_add(&thread->stime, &dts);
+			ts_add(&thread->itimer_prof_value, &dts);
+		}
+	}
+
+	if(mode == 2){
+		thread->btime.tv_sec = 0;
+		thread->btime.tv_nsec = 0;
+	}
+	else{
+		thread->btime.tv_sec = ats.tv_sec;
+		thread->btime.tv_nsec = ats.tv_nsec;
+	}
+	thread->times_update = 1;
+	thread->in_kernel = mode;
+
+	if(thread->itimer_enabled){
+		struct timeval tv;
+		int ev = 0;
+
+		if(thread->itimer_virtual.it_value.tv_sec != 0 ||
+		   thread->itimer_virtual.it_value.tv_usec){
+			ts_to_tv(&tv, &thread->itimer_virtual_value);
+			tv_sub(&tv, &thread->itimer_virtual.it_value);
+			if(tv.tv_sec > 0 ||
+			   (tv.tv_sec == 0 &&
+			    tv.tv_usec > 0)){
+				thread->itimer_virtual_value.tv_sec = 0;
+				thread->itimer_virtual_value.tv_nsec = 0;
+				thread->itimer_virtual.it_value.tv_sec =
+				    thread->itimer_virtual.it_interval.tv_sec;
+				thread->itimer_virtual.it_value.tv_usec =
+				    thread->itimer_virtual.it_interval.tv_usec;
+				do_kill(thread, thread->proc->pid, thread->tid,
+				        SIGVTALRM, NULL, 0);
+				ev = 1;
+			}
+		}
+
+		if(thread->itimer_prof.it_value.tv_sec != 0 ||
+		   thread->itimer_prof.it_value.tv_usec){
+			ts_to_tv(&tv, &thread->itimer_prof_value);
+			tv_sub(&tv, &thread->itimer_prof.it_value);
+			if(tv.tv_sec > 0 ||
+			   (tv.tv_sec == 0 &&
+			    tv.tv_usec > 0)){
+				thread->itimer_prof_value.tv_sec = 0;
+				thread->itimer_prof_value.tv_nsec = 0;
+				thread->itimer_prof.it_value.tv_sec =
+				    thread->itimer_prof.it_interval.tv_sec;
+				thread->itimer_prof.it_value.tv_usec =
+				    thread->itimer_prof.it_interval.tv_usec;
+				do_kill(thread, thread->proc->pid, thread->tid,
+				        SIGPROF, NULL, 0);
+				ev = 1;
+			}
+		}
+		if(ev){
+			if(thread->itimer_virtual.it_value.tv_sec == 0 &&
+			   thread->itimer_virtual.it_value.tv_usec == 0 &&
+			   thread->itimer_prof.it_value.tv_sec == 0 &&
+			   thread->itimer_prof.it_value.tv_usec == 0){
+				thread->itimer_enabled = 0;
+				set_timer();
+			}
+		}
+	}
+}
+
 long syscall(int num, ihk_mc_user_context_t *ctx)
 {
 	long l;
 
+	set_cputime(1);
 	if(cpu_local_var(current)->proc->status == PS_EXITED &&
 	   (num != __NR_exit && num != __NR_exit_group)){
 		check_signal(-EINVAL, NULL, 0);
+		set_cputime(0);
 		return -EINVAL;
 	}
 
@@ -6554,5 +6996,6 @@ long syscall(int num, ihk_mc_user_context_t *ctx)
 		ptrace_syscall_exit(cpu_local_var(current));
 	}
 
+	set_cputime(0);
 	return l;
 }

lib/string.c

@@ -223,7 +223,7 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	}
 
 	/* Count full length */
-	full_len = sizeof(int) + sizeof(char *); // Counter and terminating NULL
+	full_len = sizeof(long) + sizeof(char *); // Counter and terminating NULL
 	if (first) {
 		full_len += sizeof(char *) + strlen(first) + 1; 
 	}
@@ -233,6 +233,8 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 		full_len += sizeof(char *) + strlen(strings[string_i]) + 1; 
 	}
 
+	full_len = (full_len + sizeof(long) - 1) & ~(sizeof(long) - 1);
+
 	_flat = (char *)kmalloc(full_len, IHK_MC_AP_NOWAIT);
 	if (!_flat) {
 		return 0;
@@ -241,14 +243,14 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	memset(_flat, 0, full_len);
 
 	/* Number of strings */
-	*((int*)_flat) = nr_strings + (first ? 1 : 0);
+	*((long *)_flat) = nr_strings + (first ? 1 : 0);
 	
 	// Actual offset
-	flat_offset = sizeof(int) + sizeof(char *) * (nr_strings + 1 + 
+	flat_offset = sizeof(long) + sizeof(char *) * (nr_strings + 1 + 
 			(first ? 1 : 0)); 
 
 	if (first) {
-		*((char **)(_flat + sizeof(int))) = (void *)flat_offset;
+		*((char **)(_flat + sizeof(long))) = (void *)flat_offset;
 		memcpy(_flat + flat_offset, first, strlen(first) + 1);
 		flat_offset += strlen(first) + 1;
 	}
@@ -256,7 +258,7 @@ int flatten_strings(int nr_strings, char *first, char **strings, char **flat)
 	for (string_i = 0; string_i < nr_strings; ++string_i) {
 		
 		/* Fabricate the string */
-		*((char **)(_flat + sizeof(int) + (string_i + (first ? 1 : 0)) 
+		*((char **)(_flat + sizeof(long) + (string_i + (first ? 1 : 0)) 
 					* sizeof(char *))) = (void *)flat_offset;
 		memcpy(_flat + flat_offset, strings[string_i], strlen(strings[string_i]) + 1);
 		flat_offset += strlen(strings[string_i]) + 1;