configure.ac: Update version number to 1.5.1-knl+hfi

Change-Id: Icbd08c9c5f65b22d007ec479a34acd20062e0e90

.gitignore

@@ -14,3 +14,4 @@ elfboot/elfboot_test
 linux/executer/mcexec
 linux/mod_test*
 linux/target
+kernel/script/dwarf-extract-struct

Makefile.in

@@ -1,16 +1,25 @@
 TARGET = @TARGET@
 SBINDIR = @SBINDIR@
+BINDIR = @BINDIR@
 INCDIR =  @INCDIR@
 ETCDIR = @ETCDIR@
 MANDIR = @MANDIR@
 
-all::
-	@(cd executer/kernel/mcctrl; make modules)
-	@(cd executer/kernel/mcoverlayfs; make modules)
-	@(cd executer/user; make)
-	@case "$(TARGET)" in \
+all: executer-mcctrl executer-mcoverlayfs executer-user mckernel mck-tools
+
+executer-mcctrl:
+	+@(cd executer/kernel/mcctrl; $(MAKE) modules)
+
+executer-mcoverlayfs:
+	+@(cd executer/kernel/mcoverlayfs; $(MAKE) modules)
+
+executer-user:
+	+@(cd executer/user; $(MAKE))
+
+mckernel:
+	+@case "$(TARGET)" in \
 	    attached-mic | builtin-x86 | builtin-mic | smp-x86 | smp-arm64) \
-		(cd kernel; make) \
+		(cd kernel; $(MAKE)) \
 		;; \
 	    *) \
 		echo "unknown target $(TARGET)" >&2 \
@@ -18,13 +27,16 @@ all::
 		;; \
 	esac
 
-install::
-	@(cd executer/kernel/mcctrl; make install)
-	@(cd executer/kernel/mcoverlayfs; make install)
-	@(cd executer/user; make install)
+mck-tools:
+	+@(cd tools/mcstat; $(MAKE))
+
+install:
+	@(cd executer/kernel/mcctrl; $(MAKE) install)
+	@(cd executer/kernel/mcoverlayfs; $(MAKE) install)
+	@(cd executer/user; $(MAKE) install)
 	@case "$(TARGET)" in \
 	    attached-mic | builtin-x86 | builtin-mic | smp-x86 | smp-arm64) \
-		(cd kernel; make install) \
+		(cd kernel; $(MAKE) install) \
 		;; \
 	    *) \
 		echo "unknown target $(TARGET)" >&2 \
@@ -32,51 +44,41 @@ install::
 		;; \
 	esac
 	@case "$(TARGET)" in \
-	    attached-mic) \
-		mkdir -p -m 755 $(SBINDIR); \
-		install -m 755 arch/x86/tools/mcreboot-attached-mic.sh $(SBINDIR)/mcreboot; \
-		install -m 755 arch/x86/tools/mcshutdown-attached-mic.sh $(SBINDIR)/mcshutdown; \
-		mkdir -p -m 755 $(MANDIR)/man1; \
-		install -m 644 arch/x86/tools/mcreboot.1 $(MANDIR)/man1/mcreboot.1; \
-		;; \
-	    builtin-x86) \
-		mkdir -p -m 755 $(SBINDIR); \
-		install -m 755 arch/x86/tools/mcreboot-builtin-x86.sh $(SBINDIR)/mcreboot; \
-		install -m 755 arch/x86/tools/mcshutdown-builtin-x86.sh $(SBINDIR)/mcshutdown; \
-		mkdir -p -m 755 $(MANDIR)/man1; \
-		install -m 644 arch/x86/tools/mcreboot.1 $(MANDIR)/man1/mcreboot.1; \
-		;; \
 		smp-x86 | smp-arm64) \
 		mkdir -p -m 755 $(SBINDIR); \
-		install -m 755 arch/x86/tools/mcreboot-smp-x86.sh $(SBINDIR)/mcreboot.sh; \
-		install -m 755 arch/x86/tools/mcstop+release-smp-x86.sh $(SBINDIR)/mcstop+release.sh; \
-		install -m 600 arch/x86/tools/mcoverlay-destroy-smp-x86.sh $(SBINDIR)/mcoverlay-destroy.sh; \
-		install -m 600 arch/x86/tools/mcoverlay-create-smp-x86.sh $(SBINDIR)/mcoverlay-create.sh; \
-		install -m 755 arch/x86/tools/eclair-dump-backtrace.exp $(SBINDIR)/eclair-dump-backtrace.exp;\
+		install -m 755 arch/x86_64/tools/mcreboot-smp-x86.sh $(SBINDIR)/mcreboot.sh; \
+		install -m 755 arch/x86_64/tools/mcstop+release-smp-x86.sh $(SBINDIR)/mcstop+release.sh; \
+		install -m 755 arch/x86_64/tools/mpimcexec $(BINDIR)/mpimcexec; \
+		install -m 755 arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh $(SBINDIR)/mcoverlay-destroy.sh; \
+		install -m 755 arch/x86_64/tools/mcoverlay-create-smp-x86.sh $(SBINDIR)/mcoverlay-create.sh; \
+		install -m 755 arch/x86_64/tools/eclair-dump-backtrace.exp $(SBINDIR)/eclair-dump-backtrace.exp;\
 		mkdir -p -m 755 $(ETCDIR); \
-		install -m 644 arch/x86/tools/irqbalance_mck.service $(ETCDIR)/irqbalance_mck.service; \
-		install -m 644 arch/x86/tools/irqbalance_mck.in $(ETCDIR)/irqbalance_mck.in; \
+		install -m 644 arch/x86_64/tools/irqbalance_mck.service $(ETCDIR)/irqbalance_mck.service; \
+		install -m 644 arch/x86_64/tools/irqbalance_mck.in $(ETCDIR)/irqbalance_mck.in; \
 		mkdir -p -m 755 $(INCDIR); \
 		install -m 644 kernel/include/swapfmt.h $(INCDIR); \
 		mkdir -p -m 755 $(MANDIR)/man1; \
-		install -m 644 arch/x86/tools/mcreboot.1 $(MANDIR)/man1/mcreboot.1; \
+		install -m 644 arch/x86_64/tools/mcreboot.1 $(MANDIR)/man1/mcreboot.1; \
+		install -m 644 arch/x86_64/tools/mpimcexec.1 $(MANDIR)/man1/mpimcexec.1; \
 		;; \
 	    *) \
 		echo "unknown target $(TARGET)" >&2 \
 		exit 1 \
 		;; \
 	esac
+	@(cd tools/mcstat/; $(MAKE) install)
 
-clean::
-	@(cd executer/kernel/mcctrl; make clean)
-	@(cd executer/kernel/mcoverlayfs; make clean)
-	@(cd executer/user; make clean)
+clean:
+	@(cd executer/kernel/mcctrl; $(MAKE) clean)
+	@(cd executer/kernel/mcoverlayfs; $(MAKE) clean)
+	@(cd executer/user; $(MAKE) clean)
 	@case "$(TARGET)" in \
 	    attached-mic | builtin-x86 | builtin-mic | smp-x86 | smp-arm64) \
-		(cd kernel; make clean) \
+		(cd kernel; $(MAKE) clean) \
 		;; \
 	    *) \
 		echo "unknown target $(TARGET)" >&2 \
 		exit 1 \
 		;; \
 	esac
+	@(cd tools/mcstat; $(MAKE) clean)

arch/arm64/kernel/cpu.c

@@ -590,6 +590,8 @@ static void show_context_stack(struct pt_regs *regs)
 		return;
 	}
 
+	ihk_mc_debug_show_interrupt_context(regs);
+
 	sp = (uintptr_t)regs + sizeof(*regs);
 	stack_top = ALIGN_UP(sp, (uintptr_t)KERNEL_STACK_SIZE);
 	max_loop = (stack_top - sp) / min_stack_frame_size;
@@ -1170,8 +1172,6 @@ void arch_clone_thread(struct thread *othread, unsigned long pc,
 	asm("mrs %0, tpidr_el0" : "=r" (tls));
 	othread->tlsblock_base = nthread->tlsblock_base = tls;
 
-	/* copy fp_regs values from parent. */
-	save_fp_regs(othread);
 	if ((othread->fp_regs != NULL) && (check_and_allocate_fp_regs(nthread) == 0)) {
 		memcpy(nthread->fp_regs, othread->fp_regs, sizeof(fp_regs_struct));
 	}
@@ -1205,6 +1205,10 @@ void ihk_mc_delay_us(int us)
 	arch_delay(us);
 }
 
+void arch_print_stack()
+{
+}
+
 void arch_show_interrupt_context(const void *reg)
 {
 	const struct pt_regs *regs = (struct pt_regs *)reg;
@@ -1428,6 +1432,13 @@ save_fp_regs(struct thread *thread)
 	}
 }
 
+void copy_fp_regs(struct thread *from, struct thread *to)
+{
+	if ((from->fp_regs != NULL) && (check_and_allocate_fp_regs(to) == 0)) {
+		memcpy(to->fp_regs, from->fp_regs, sizeof(fp_regs_struct));
+	}
+}
+
 void
 clear_fp_regs(struct thread *thread)
 {
@@ -1499,7 +1510,6 @@ unhandled_page_fault(struct thread *thread, void *fault_addr, void *regs)
 	const uintptr_t address = (uintptr_t)fault_addr;
 	struct process_vm *vm = thread->vm;
 	struct vm_range *range;
-	char found;
 	unsigned long irqflags;
 	unsigned long error = 0;
 
@@ -1513,17 +1523,12 @@ unhandled_page_fault(struct thread *thread, void *fault_addr, void *regs)
 			(error & PF_RSVD ? "was" : "wasn't"),
 			(error & PF_INSTR ? "was" : "wasn't"));
 
-	found = 0;
-	list_for_each_entry(range, &vm->vm_range_list, list) {
-		if (range->start <= address && range->end > address) {
-			found = 1;
-			__kprintf("address is in range, flag: 0x%lx\n",
-					range->flag);
-			ihk_mc_pt_print_pte(vm->address_space->page_table, (void*)address);
-			break;
-		}
-	}
-	if (!found) {
+	range = lookup_process_memory_range(vm, address, address+1);
+	if (range) {
+		__kprintf("address is in range, flag: 0x%lx\n",
+				range->flag);
+		ihk_mc_pt_print_pte(vm->address_space->page_table, (void*)address);
+	} else {
 		__kprintf("address is out of range! \n");
 	}
 

arch/arm64/kernel/gencore.c

@@ -221,7 +221,7 @@ int gencore(struct thread *thread, void *regs,
 	Elf64_Ehdr eh;
 	Elf64_Phdr *ph = NULL;
 	void *note = NULL;
-	struct vm_range *range;
+	struct vm_range *range, *next;
 	struct process_vm *vm = thread->vm;
 	int segs = 1;	/* the first one is for NOTE */
 	int notesize, phsize, alignednotesize;
@@ -235,7 +235,10 @@ int gencore(struct thread *thread, void *regs,
 		return -1;
 	}
 
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		dkprintf("start:%lx end:%lx flag:%lx objoff:%lx\n", 
 			 range->start, range->end, range->flag, range->objoff);
 		/* We omit reserved areas because they are only for
@@ -323,7 +326,10 @@ int gencore(struct thread *thread, void *regs,
 
 	/* program header for each memory chunk */
 	i = 1;
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		unsigned long flag = range->flag;
 		unsigned long size = range->end - range->start;
 
@@ -364,7 +370,10 @@ int gencore(struct thread *thread, void *regs,
 	dkprintf("coretable[2]: %lx@%lx(%lx)\n", ct[2].len, ct[2].addr, note);
 
 	i = 3;	/* memory segments */
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		unsigned long phys;
 
 		if (range->flag & VR_RESERVED)

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

@@ -134,4 +134,12 @@ futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval, int newval)
 	return ret;
 }
 
+static inline int get_futex_value_locked(uint32_t *dest, uint32_t *from)
+{
+
+	*dest = *(volatile uint32_t *)from;
+
+	return 0;
+}
+
 #endif /* !__HEADER_ARM64_COMMON_ARCH_FUTEX_H */

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

@@ -215,7 +215,7 @@ static const unsigned int  PTL1_ENTRIES = __PTL1_ENTRIES;
 #define PAGE_P2ALIGN        0
 #define page_offset(addr)   __page_offset(addr, PAGE_SIZE)
 #define page_align(addr)    __page_align(addr, PAGE_SIZE)
-#define page_align_up(addr) __page_align_up((addr, PAGE_SIZE)
+#define page_align_up(addr) __page_align_up(addr, PAGE_SIZE)
 
 /*
  * large page
@@ -263,6 +263,8 @@ static const unsigned int  PTL1_ENTRIES = __PTL1_ENTRIES;
 
 #define PTE_FILEOFF		PTE_SPECIAL
 
+#define PT_ENTRIES		(PAGE_SIZE >> 3)
+
 #ifndef __ASSEMBLY__
 
 #include <ihk/types.h>

arch/arm64/kernel/include/arch/cpu.h

@@ -25,6 +25,8 @@
 #define smp_rmb()	dmb(ishld)
 #define smp_wmb()	dmb(ishst)
 
+#define arch_barrier()	smp_mb()
+
 #define smp_store_release(p, v)						\
 do {									\
 	compiletime_assert_atomic_type(*p);				\

arch/arm64/kernel/include/arch_rusage.h

@@ -0,0 +1,34 @@
+#ifndef ARCH_RUSAGE_H_INCLUDED
+#define ARCH_RUSAGE_H_INCLUDED
+
+#include <arch-memory.h>
+
+//#define DEBUG_RUSAGE
+
+extern struct rusage_global *rusage;
+
+#define IHK_OS_PGSIZE_4KB  0
+#define IHK_OS_PGSIZE_16KB 1
+#define IHK_OS_PGSIZE_64KB 2
+
+static inline int rusage_pgsize_to_pgtype(size_t pgsize)
+{
+	int ret = IHK_OS_PGSIZE_4KB;
+	switch (pgsize) {
+	case __PTL1_SIZE:
+		ret = IHK_OS_PGSIZE_4KB;
+		break;
+	case __PTL2_SIZE:
+		ret = IHK_OS_PGSIZE_16KB;
+		break;
+	case __PTL3_SIZE:
+		ret = IHK_OS_PGSIZE_64KB;
+		break;
+	default:
+		kprintf("%s: Error: Unknown pgsize=%ld\n", __FUNCTION__, pgsize);
+		break;
+	}
+	return ret;
+}
+
+#endif /* !defined(ARCH_RUSAGE_H_INCLUDED) */

arch/arm64/kernel/include/assembler.h

@@ -6,6 +6,8 @@
 
 #if defined(CONFIG_HAS_NMI)
 #include <arm-gic-v3.h>
+#else /* defined(CONFIG_HAS_NMI) */
+#include <sysreg.h>
 #endif /* defined(CONFIG_HAS_NMI) */
 
 #if defined(CONFIG_HAS_NMI)

arch/arm64/kernel/include/cputype.h

@@ -35,6 +35,8 @@
 #define MIDR_IMPLEMENTOR(midr)	\
 	(((midr) & MIDR_IMPLEMENTOR_MASK) >> MIDR_IMPLEMENTOR_SHIFT)
 
+#define ARM_CPU_IMP_CAVIUM	0x43
+
 #ifndef __ASSEMBLY__
 
 static unsigned int read_cpuid_id(void)

arch/arm64/kernel/include/ihk/context.h

@@ -73,6 +73,7 @@ static inline void pt_regs_write_reg(struct pt_regs *regs, int r,
 #define ihk_mc_syscall_arg5(uc) (uc)->regs[5]
 
 #define ihk_mc_syscall_ret(uc)  (uc)->regs[0]
+#define ihk_mc_syscall_number(uc)  (uc)->regs[8]
 
 #define ihk_mc_syscall_pc(uc)   (uc)->pc
 #define ihk_mc_syscall_sp(uc)   (uc)->sp

arch/arm64/kernel/include/syscall_list.h

@@ -144,5 +144,3 @@ SYSCALL_HANDLED(1045, signalfd)
 SYSCALL_DELEGATED(1049, stat)
 SYSCALL_DELEGATED(1060, getpgrp)
 SYSCALL_DELEGATED(1062, time)
-SYSCALL_HANDLED(1071, vfork)
-SYSCALL_DELEGATED(1079, fork)

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

@@ -10,6 +10,8 @@
 
 //#define DEBUG_GICV3
 
+#define USE_CAVIUM_THUNDER_X
+
 #ifdef DEBUG_GICV3
 #define	dkprintf(...)	kprintf(__VA_ARGS__)
 #define	ekprintf(...)	kprintf(__VA_ARGS__)
@@ -18,6 +20,10 @@
 #define	ekprintf(...)	kprintf(__VA_ARGS__)
 #endif
 
+#ifdef USE_CAVIUM_THUNDER_X
+static char is_cavium_thunderx = 0;
+#endif
+
 void *dist_base;
 void *rdist_base[NR_CPUS];
 
@@ -108,8 +114,8 @@ static uint64_t gic_read_iar_cavium_thunderx(void)
 	asm volatile("nop;nop;nop;nop;");
 	asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
 	asm volatile("nop;nop;nop;nop;");
-	mb();
 #endif /* CONFIG_HAS_NMI */
+	mb();
 
 	return irqstat;
 }
@@ -118,7 +124,7 @@ static uint64_t gic_read_iar_cavium_thunderx(void)
 static uint64_t gic_read_iar(void)
 {
 #ifdef USE_CAVIUM_THUNDER_X
-	if (static_key_false(&is_cavium_thunderx))
+	if (is_cavium_thunderx)
 		return gic_read_iar_cavium_thunderx();
 	else
 #endif
@@ -266,6 +272,7 @@ void arm64_issue_ipi_gicv3(uint32_t cpuid, uint32_t vector)
 {
 	dkprintf("Send irq#%d to cpuid=%d\n", vector, cpuid);
 
+	barrier();
 	if(vector < 16){
 		// send SGI
 		arm64_raise_sgi_gicv3(cpuid, vector);
@@ -304,7 +311,9 @@ void gic_dist_init_gicv3(unsigned long dist_base_pa, unsigned long size)
 
 #ifdef USE_CAVIUM_THUNDER_X
 	/* Cavium ThunderX erratum 23154 */
-	gicv3_check_capabilities();
+	if (MIDR_IMPLEMENTOR(read_cpuid_id()) == ARM_CPU_IMP_CAVIUM) {
+		is_cavium_thunderx = 1;
+	}
 #endif
 }
 
@@ -399,6 +408,7 @@ void gic_enable_gicv3(void)
 	/* Set specific IPI to NMI */
 	writeb_relaxed(GICD_INT_NMI_PRI, rd_sgi_base + GIC_DIST_PRI + INTRID_CPU_STOP);
 	writeb_relaxed(GICD_INT_NMI_PRI, rd_sgi_base + GIC_DIST_PRI + INTRID_MEMDUMP);
+	writeb_relaxed(GICD_INT_NMI_PRI, rd_sgi_base + GIC_DIST_PRI + INTRID_STACK_TRACE);
 
 	/* sync wait */
 	gic_do_wait_for_rwp(rbase);

arch/arm64/kernel/local.c

@@ -20,10 +20,11 @@ size_t arm64_cpu_local_variables_span = LOCALS_SPAN;	/* for debugger */
 void init_processors_local(int max_id)
 {
 	int i = 0;
+	const int sz = (max_id + 1) * KERNEL_STACK_SIZE;
 	union arm64_cpu_local_variables *tmp;
 
 	/* allocate one more for alignment */
-	locals = ihk_mc_alloc_pages((max_id + 1) * (KERNEL_STACK_SIZE / PAGE_SIZE), IHK_MC_AP_CRITICAL);
+	locals = ihk_mc_alloc_pages(((sz + PAGE_SIZE - 1) / PAGE_SIZE), IHK_MC_AP_CRITICAL);
 	locals = (union arm64_cpu_local_variables *)ALIGN_UP((unsigned long)locals, KERNEL_STACK_SIZE);
 
 	/* clear struct process, struct process_vm, struct thread_info area */

arch/arm64/kernel/memory.c

@@ -1760,6 +1760,12 @@ int visit_pte_range(page_table_t pt, void *start0, void *end0, int pgshift,
 	return initial_lookup.walk(tt, 0, start, end, initial_lookup.callback, &args);
 }
 
+int visit_pte_range_safe(page_table_t pt, void *start0, void *end0, int pgshift,
+                        enum visit_pte_flag flags, pte_visitor_t *funcp, void *arg)
+{
+	return 0;
+}
+
 struct clear_range_args {
 	int free_physical;
 	struct memobj *memobj;
@@ -1801,7 +1807,6 @@ static int clear_range_l1(void *args0, pte_t *ptep, uint64_t base,
 			ihk_mc_free_pages_user(phys_to_virt(phys), npages);
 			dkprintf("%s: freeing regular page at 0x%lx\n", __FUNCTION__, base);
 		}
-		args->vm->currss -= PTL1_SIZE;
 	}
 	
 	return 0;
@@ -1881,7 +1886,6 @@ static int clear_range_middle(void *args0, pte_t *ptep, uint64_t base,
 				ihk_mc_free_pages_user(phys_to_virt(phys), npages);
 				dkprintf("%s(level=%d): freeing large page at 0x%lx\n", __FUNCTION__, level, base);
 			}
-			args->vm->currss -= tbl.pgsize;
 		}
 
 		return 0;

arch/arm64/kernel/ptrace.c

@@ -965,7 +965,7 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	proc->status = PS_TRACED;
 #endif /* POSTK_DEBUG_TEMP_FIX_41 */
 	thread->status = PS_TRACED;
-	proc->ptrace &= ~PT_TRACE_SYSCALL_MASK;
+	proc->ptrace &= ~PT_TRACE_SYSCALL;
 	if (sig == SIGSTOP || sig == SIGTSTP ||
 			sig == SIGTTIN || sig == SIGTTOU) {
 		proc->signal_flags |= SIGNAL_STOP_STOPPED;

arch/arm64/kernel/syscall.c

@@ -12,6 +12,8 @@
 #include <lwk/compiler.h>
 #include <hwcap.h>
 #include <prctl.h>
+#include <limits.h>
+#include <syscall.h>
 
 extern void ptrace_report_signal(struct thread *thread, int sig);
 extern void clear_single_step(struct thread *thread);
@@ -1321,6 +1323,17 @@ interrupt_from_user(void *regs0)
 	return((regs->pstate & PSR_MODE_MASK) == PSR_MODE_EL0t);
 }
 
+void save_syscall_return_value(int num, unsigned long rc)
+{
+	/*
+	 * Save syscall return value.
+	 */
+	if (cpu_local_var(current) && cpu_local_var(current)->uctx &&
+			num != __NR_rt_sigsuspend) {
+		ihk_mc_syscall_arg0(cpu_local_var(current)->uctx) = rc;
+	}
+}
+
 void
 check_signal(unsigned long rc, void *regs0, int num)
 {
@@ -1345,16 +1358,6 @@ __check_signal(unsigned long rc, void *regs0, int num, int irq_disabled)
 		return;
 	thread = cpu_local_var(current);
 
-	/**
-	 * If check_signal is called from syscall(), 
-	 * then save syscall return value.
-	 */
-	if((regs == NULL)&&(num != __NR_rt_sigsuspend)){ /* It's call from syscall! */
-	 	// Get user context through current thread
-	 	// and update syscall return.
-	 	ihk_mc_syscall_arg0(thread->uctx) = rc;
-	}
-
 	if(thread == NULL || thread->proc->pid == 0){
 		struct thread *t;
 		irqstate = ihk_mc_spinlock_lock(&(cpu_local_var(runq_lock)));
@@ -1866,4 +1869,645 @@ save_uctx(void *uctx, struct pt_regs *regs)
 	/* TODO: skeleton for UTI */
 }
 
+int do_process_vm_read_writev(int pid, 
+		const struct iovec *local_iov,
+		unsigned long liovcnt,
+		const struct iovec *remote_iov,
+		unsigned long riovcnt,
+		unsigned long flags,
+		int op)
+{
+	int ret = -EINVAL;	
+	int li, ri;
+	int pli, pri;
+	off_t loff, roff;
+	size_t llen = 0, rlen = 0;
+	size_t copied = 0;
+	size_t to_copy;
+	struct thread *lthread = cpu_local_var(current);
+	struct process *rproc;
+	struct process *lproc = lthread->proc;
+	struct process_vm *rvm = NULL;
+	unsigned long rphys;
+	unsigned long rpage_left;
+	unsigned long psize;
+	void *rva;
+	struct vm_range *range;
+	struct mcs_rwlock_node_irqsave lock;
+	struct mcs_rwlock_node update_lock;
+
+	/* Sanity checks */
+	if (flags) {
+		return -EINVAL;
+	}
+	
+	if (liovcnt > IOV_MAX || riovcnt > IOV_MAX) {
+		return -EINVAL;
+	}
+
+	/* Check if parameters are okay */
+	ihk_mc_spinlock_lock_noirq(&lthread->vm->memory_range_lock);
+
+	range = lookup_process_memory_range(lthread->vm, 
+			(uintptr_t)local_iov, 
+			(uintptr_t)(local_iov + liovcnt * sizeof(struct iovec)));
+
+	if (!range) {
+		ret = -EFAULT; 
+		goto arg_out;
+	}
+
+	range = lookup_process_memory_range(lthread->vm, 
+			(uintptr_t)remote_iov, 
+			(uintptr_t)(remote_iov + riovcnt * sizeof(struct iovec)));
+
+	if (!range) {
+		ret = -EFAULT; 
+		goto arg_out;
+	}
+
+	ret = 0;
+arg_out:
+	ihk_mc_spinlock_unlock_noirq(&lthread->vm->memory_range_lock);
+
+	if (ret != 0) {
+		goto out;
+	}
+
+	for (li = 0; li < liovcnt; ++li) {
+		llen += local_iov[li].iov_len;
+		dkprintf("local_iov[%d].iov_base: 0x%lx, len: %lu\n",
+			li, local_iov[li].iov_base, local_iov[li].iov_len);
+	}
+
+	for (ri = 0; ri < riovcnt; ++ri) {
+		rlen += remote_iov[ri].iov_len;
+		dkprintf("remote_iov[%d].iov_base: 0x%lx, len: %lu\n",
+			ri, remote_iov[ri].iov_base, remote_iov[ri].iov_len);
+	}
+
+	if (llen != rlen) {
+		return -EINVAL;
+	}
+	
+	/* Find remote process */
+	rproc = find_process(pid, &lock);
+	if (!rproc) {
+		ret = -ESRCH;
+		goto out;
+	}
+
+	mcs_rwlock_reader_lock_noirq(&rproc->update_lock, &update_lock);
+	if(rproc->status == PS_EXITED ||
+	   rproc->status == PS_ZOMBIE){
+		mcs_rwlock_reader_unlock_noirq(&rproc->update_lock, &update_lock);
+		process_unlock(rproc, &lock);
+		ret = -ESRCH;
+		goto out;
+	}
+	rvm = rproc->vm;
+	hold_process_vm(rvm);
+	mcs_rwlock_reader_unlock_noirq(&rproc->update_lock, &update_lock);
+	process_unlock(rproc, &lock);
+
+	if (lproc->euid != 0 &&
+	    (lproc->ruid != rproc->ruid ||
+	     lproc->ruid != rproc->euid ||
+	     lproc->ruid != rproc->suid ||
+	     lproc->rgid != rproc->rgid ||
+	     lproc->rgid != rproc->egid ||
+	     lproc->rgid != rproc->sgid)) {
+		ret = -EPERM;
+		goto out;
+	}
+
+	dkprintf("pid %d found, doing %s: liovcnt: %d, riovcnt: %d \n", pid,
+		(op == PROCESS_VM_READ) ? "PROCESS_VM_READ" : "PROCESS_VM_WRITE",
+		liovcnt, riovcnt);
+
+	pli = pri = -1; /* Previous indeces in iovecs */
+	li = ri = 0; /* Current indeces in iovecs */
+	loff = roff = 0; /* Offsets in current iovec */
+
+	/* Now iterate and do the copy */
+	while (copied < llen) {
+		int faulted = 0;
+
+		/* New local vector? */
+		if (pli != li) {
+			struct vm_range *range;
+
+			ihk_mc_spinlock_lock_noirq(&lthread->vm->memory_range_lock);
+
+			/* Is base valid? */
+			range = lookup_process_memory_range(lthread->vm,
+					(uintptr_t)local_iov[li].iov_base,
+					(uintptr_t)(local_iov[li].iov_base + 1));
+
+			if (!range) {
+				ret = -EFAULT;
+				goto pli_out;
+			}
+
+			/* Is range valid? */
+			range = lookup_process_memory_range(lthread->vm,
+					(uintptr_t)local_iov[li].iov_base,
+					(uintptr_t)(local_iov[li].iov_base + local_iov[li].iov_len));
+
+			if (range == NULL) {
+				ret = -EINVAL;
+				goto pli_out;
+			}
+
+			if (!(range->flag & ((op == PROCESS_VM_READ) ?
+				VR_PROT_WRITE : VR_PROT_READ))) {
+				ret = -EFAULT;
+				goto pli_out;
+			}
+
+			ret = 0;
+pli_out:
+			ihk_mc_spinlock_unlock_noirq(&lthread->vm->memory_range_lock);
+
+			if (ret != 0) {
+				goto out;
+			}
+
+			pli = li;
+		}
+
+		/* New remote vector? */
+		if (pri != ri) {
+			struct vm_range *range;
+
+			ihk_mc_spinlock_lock_noirq(&rvm->memory_range_lock);
+
+			/* Is base valid? */
+			range = lookup_process_memory_range(rvm,
+					(uintptr_t)remote_iov[li].iov_base,
+					(uintptr_t)(remote_iov[li].iov_base + 1));
+
+			if (range == NULL) {
+				ret = -EFAULT;
+				goto pri_out;
+			}
+
+			/* Is range valid? */
+			range = lookup_process_memory_range(rvm,
+					(uintptr_t)remote_iov[li].iov_base,
+					(uintptr_t)(remote_iov[li].iov_base + remote_iov[li].iov_len));
+
+			if (range == NULL) {
+				ret = -EINVAL;
+				goto pri_out;
+			}
+
+			if (!(range->flag & ((op == PROCESS_VM_READ) ?
+				VR_PROT_READ : VR_PROT_WRITE))) {
+				ret = -EFAULT;
+				goto pri_out;
+			}
+
+			ret = 0;
+pri_out:
+			ihk_mc_spinlock_unlock_noirq(&rvm->memory_range_lock);
+
+			if (ret != 0) {
+				goto out;
+			}
+
+			pri = ri;
+		}
+
+		/* Figure out how much we can copy at most in this iteration */
+		to_copy = (local_iov[li].iov_len - loff);	
+		if ((remote_iov[ri].iov_len - roff) < to_copy) {
+			to_copy = remote_iov[ri].iov_len - roff;
+		}
+
+retry_lookup:
+		/* TODO: remember page and do this only if necessary */
+		ret = ihk_mc_pt_virt_to_phys_size(rvm->address_space->page_table,
+				remote_iov[ri].iov_base + roff, &rphys, &psize);
+
+		if (ret) {
+			uint64_t reason = PF_POPULATE | PF_WRITE | PF_USER;
+			void *addr;
+
+			if (faulted) {
+				ret = -EFAULT;
+				goto out;
+			}
+
+			/* Fault in pages */
+			for (addr = (void *)
+					(((unsigned long)remote_iov[ri].iov_base + roff)
+					& PAGE_MASK);
+					addr < (remote_iov[ri].iov_base + roff + to_copy);
+					addr += PAGE_SIZE) {
+
+				ret = page_fault_process_vm(rvm, addr, reason);
+				if (ret) {
+					ret = -EFAULT;
+					goto out;
+				}
+			}
+
+			faulted = 1;
+			goto retry_lookup;
+		}
+
+		rpage_left = ((((unsigned long)remote_iov[ri].iov_base + roff +
+			psize) & ~(psize - 1)) -
+			((unsigned long)remote_iov[ri].iov_base + roff));
+		if (rpage_left < to_copy) {
+			to_copy = rpage_left;
+		}
+
+		rva = phys_to_virt(rphys);
+
+		fast_memcpy(
+				(op == PROCESS_VM_READ) ? local_iov[li].iov_base + loff : rva,
+				(op == PROCESS_VM_READ) ? rva : local_iov[li].iov_base + loff,
+				to_copy);
+
+		copied += to_copy;
+		dkprintf("local_iov[%d]: 0x%lx %s remote_iov[%d]: 0x%lx, %lu copied, psize: %lu, rpage_left: %lu\n",
+			li, local_iov[li].iov_base + loff, 
+			(op == PROCESS_VM_READ) ? "<-" : "->", 
+			ri, remote_iov[ri].iov_base + roff, to_copy,
+			psize, rpage_left);
+
+		loff += to_copy;
+		roff += to_copy;
+
+		if (loff == local_iov[li].iov_len) {
+			li++;
+			loff = 0;
+		}
+
+		if (roff == remote_iov[ri].iov_len) {
+			ri++;
+			roff = 0;
+		}
+	}
+
+	release_process_vm(rvm);
+
+	return copied;
+
+out:
+	if(rvm)
+		release_process_vm(rvm);
+	return ret;
+}
+
+int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
+{
+	int i, i_s, i_e, phase = 1;
+	struct move_pages_smp_req *mpsr =
+		(struct move_pages_smp_req *)arg;
+	struct process_vm *vm = mpsr->proc->vm;
+	int count = mpsr->count;
+	struct page_table *save_pt;
+	extern struct page_table *get_init_page_table(void);
+
+	i_s = (count / nr_cpus) * cpu_index;
+	i_e = i_s + (count / nr_cpus);
+	if (cpu_index == (nr_cpus - 1)) {
+		i_e = count;
+	}
+
+	/* Load target process' PT so that we can access user-space */
+	save_pt = cpu_local_var(current) == &cpu_local_var(idle) ?
+		get_init_page_table() :
+		cpu_local_var(current)->vm->address_space->page_table;
+
+	if (save_pt != vm->address_space->page_table) {
+		ihk_mc_load_page_table(vm->address_space->page_table);
+	}
+	else {
+		save_pt = NULL;
+	}
+
+	if (nr_cpus == 1) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				mpsr->nodes_ready = 1;
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus > 1 && nr_cpus < 4) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+			case 1:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				mpsr->nodes_ready = 1;
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus >= 4 && nr_cpus < 8) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				break;
+			case 1:
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				break;
+			case 2:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				mpsr->nodes_ready = 1;
+				break;
+			case 3:
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus >= 8) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * (count / 2));
+				break;
+			case 1:
+				memcpy(mpsr->virt_addr + (count / 2),
+						mpsr->user_virt_addr + (count / 2),
+						sizeof(void *) * (count / 2));
+				break;
+			case 2:
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				break;
+			case 3:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				mpsr->nodes_ready = 1;
+				break;
+			case 4:
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				break;
+			case 5:
+				memset(mpsr->status, 0, sizeof(int) * count);
+				break;
+			case 6:
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				break;
+			case 7:
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				break;
+
+			default:
+				break;
+		}
+	}
+
+	while (!(volatile int)mpsr->nodes_ready) {
+		cpu_pause();
+	}
+
+	/* NUMA verification in parallel */
+	for (i = i_s; i < i_e; i++) {
+		if (mpsr->nodes[i] < 0 ||
+				mpsr->nodes[i] >= ihk_mc_get_nr_numa_nodes() ||
+				!test_bit(mpsr->nodes[i],
+					mpsr->proc->vm->numa_mask)) {
+			mpsr->phase_ret = -EINVAL;
+			break;
+		}
+	}
+
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	/* PTE lookup in parallel */
+	for (i = i_s; i < i_e; i++) {
+		void *phys;
+		size_t pgsize;
+		int p2align;
+		/*
+		 * XXX: No page structures for anonymous mappings.
+		 * Look up physical addresses by scanning page tables.
+		 */
+		mpsr->ptep[i] = ihk_mc_pt_lookup_pte(vm->address_space->page_table,
+				(void *)mpsr->virt_addr[i], 0, &phys, &pgsize, &p2align);
+
+		/* PTE valid? */
+		if (!mpsr->ptep[i] || !pte_is_present(mpsr->ptep[i])) {
+			mpsr->status[i] = -ENOENT;
+			mpsr->ptep[i] = NULL;
+			continue;
+		}
+
+		/* PTE is file? */
+		if (pte_is_fileoff(mpsr->ptep[i], PAGE_SIZE)) {
+			mpsr->status[i] = -EINVAL;
+			mpsr->ptep[i] = NULL;
+			continue;
+		}
+
+		dkprintf("%s: virt 0x%lx:%lu requested to be moved to node %d\n",
+			__FUNCTION__, mpsr->virt_addr[i], pgsize, mpsr->nodes[i]);
+
+		/* Large page? */
+		if (pgsize > PAGE_SIZE) {
+			int nr_sub_pages = (pgsize / PAGE_SIZE);
+			int j;
+
+			if (i + nr_sub_pages > count) {
+				kprintf("%s: ERROR: page at index %d exceeds the region\n",
+						__FUNCTION__, i);
+				mpsr->status[i] = -EINVAL;
+				break;
+			}
+
+			/* Is it contiguous across nr_sub_pages and all
+			 * requested to be moved to the same target node? */
+			for (j = 0; j < nr_sub_pages; ++j) {
+				if (mpsr->virt_addr[i + j] !=
+				(mpsr->virt_addr[i] + (j * PAGE_SIZE)) ||
+						mpsr->nodes[i] != mpsr->nodes[i + j]) {
+					kprintf("%s: ERROR: virt address or node at index %d"
+							" is inconsistent\n",
+							__FUNCTION__, i + j);
+					mpsr->phase_ret = -EINVAL;
+					goto pte_out;
+				}
+			}
+
+			mpsr->nr_pages[i] = nr_sub_pages;
+			i += (nr_sub_pages - 1);
+		}
+		else {
+			mpsr->nr_pages[i] = 1;
+		}
+	}
+
+pte_out:
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	if (cpu_index == 0) {
+		/* Allocate new pages on target NUMA nodes */
+		for (i = 0; i < count; i++) {
+			int pgalign = 0;
+			int j;
+			void *dst;
+
+			if (!mpsr->ptep[i] || mpsr->status[i] < 0 || !mpsr->nr_pages[i])
+				continue;
+
+			/* TODO: store pgalign info in an array as well? */
+			if (mpsr->nr_pages[i] > 1) {
+				if (mpsr->nr_pages[i] * PAGE_SIZE == PTL2_SIZE)
+					pgalign = PTL2_SHIFT - PTL1_SHIFT;
+			}
+
+			dst = ihk_mc_alloc_aligned_pages_node(mpsr->nr_pages[i],
+					pgalign, IHK_MC_AP_USER, mpsr->nodes[i]);
+
+			if (!dst) {
+				mpsr->status[i] = -ENOMEM;
+				continue;
+			}
+
+			for (j = i; j < (i + mpsr->nr_pages[i]); ++j) {
+				mpsr->status[j] = mpsr->nodes[i];
+			}
+
+			mpsr->dst_phys[i] = virt_to_phys(dst);
+
+			dkprintf("%s: virt 0x%lx:%lu to node %d, pgalign: %d,"
+					" allocated phys: 0x%lx\n",
+					__FUNCTION__, mpsr->virt_addr[i],
+					mpsr->nr_pages[i] * PAGE_SIZE,
+					mpsr->nodes[i], pgalign, mpsr->dst_phys[i]);
+		}
+	}
+
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	/* Copy, PTE update, memfree in parallel */
+	for (i = i_s; i < i_e; ++i) {
+		if (!mpsr->dst_phys[i])
+			continue;
+
+		fast_memcpy(phys_to_virt(mpsr->dst_phys[i]),
+				phys_to_virt(pte_get_phys(mpsr->ptep[i])),
+				mpsr->nr_pages[i] * PAGE_SIZE);
+
+		ihk_mc_free_pages(
+				phys_to_virt(pte_get_phys(mpsr->ptep[i])),
+				mpsr->nr_pages[i]);
+
+		pte_update_phys(mpsr->ptep[i], mpsr->dst_phys[i]);
+
+		dkprintf("%s: virt 0x%lx:%lu copied and remapped to phys: 0x%lu\n",
+				__FUNCTION__, mpsr->virt_addr[i],
+				mpsr->nr_pages[i] * PAGE_SIZE,
+				mpsr->dst_phys[i]);
+	}
+
+	/* XXX: do a separate SMP call with only CPUs running threads
+	 * of this process? */
+	if (cpu_local_var(current)->proc == mpsr->proc) {
+		/* Invalidate all TLBs */
+		for (i = 0; i < mpsr->count; i++) {
+			if (!mpsr->dst_phys[i])
+				continue;
+
+			flush_tlb_single((unsigned long)mpsr->virt_addr[i]);
+		}
+	}
+
+out:
+	if (save_pt) {
+		ihk_mc_load_page_table(save_pt);
+	}
+
+	return mpsr->phase_ret;
+}
+
+time_t time(void) {
+	struct timespec ats;
+
+	if (gettime_local_support) {
+		calculate_time_from_tsc(&ats);
+		return ats.tv_sec;
+	}
+
+	return (time_t)0;
+}
+
 /*** End of File ***/

arch/x86/kernel/Makefile.arch

@@ -1,7 +0,0 @@
-IHK_OBJS += cpu.o interrupt.o memory.o trampoline.o local.o context.o
-IHK_OBJS += perfctr.o syscall.o vsyscall.o
-# POSTK_DEBUG_ARCH_DEP_18 coredump arch separation.
-# IHK_OBJS added coredump.o
-ifeq ($(ARCH), arm64)
-IHK_OBJS += coredump.o
-endif

arch/x86_64/kernel/Makefile.arch.in

@@ -0,0 +1,2 @@
+IHK_OBJS += cpu.o interrupt.o memory.o trampoline.o local.o context.o
+IHK_OBJS += perfctr.o syscall.o vsyscall.o

arch/x86_64/kernel/cpu.c

@@ -849,6 +849,7 @@ void setup_x86_ap(void (*next_func)(void))
 void arch_show_interrupt_context(const void *reg);
 void set_signal(int sig, void *regs, struct siginfo *info);
 void check_signal(unsigned long, void *, int);
+void check_sig_pending();
 extern void tlb_flush_handler(int vector);
 
 void __show_stack(uintptr_t *sp) {
@@ -870,6 +871,19 @@ void show_context_stack(uintptr_t *rbp) {
 	return;
 }
 
+void interrupt_exit(struct x86_user_context *regs)
+{
+	if (interrupt_from_user(regs)) {
+		cpu_enable_interrupt();
+		check_sig_pending();
+		check_need_resched();
+		check_signal(0, regs, 0);
+	}
+	else {
+		check_sig_pending();
+	}
+}
+
 void handle_interrupt(int vector, struct x86_user_context *regs)
 {
 	struct ihk_mc_interrupt_handler *h;
@@ -992,12 +1006,8 @@ void handle_interrupt(int vector, struct x86_user_context *regs)
 		}
 	}
 
-	if(interrupt_from_user(regs)){
-		cpu_enable_interrupt();
-		check_signal(0, regs, 0);
-		check_need_resched();
-	}
-	set_cputime(0);
+	interrupt_exit(regs);
+	set_cputime(interrupt_from_user(regs)? 0: 1);
 
 	--v->in_interrupt;
 }
@@ -1012,13 +1022,9 @@ void gpe_handler(struct x86_user_context *regs)
 		panic("gpe_handler");
 	}
 	set_signal(SIGSEGV, regs, NULL);
-	if(interrupt_from_user(regs)){
-		cpu_enable_interrupt();
-		check_signal(0, regs, 0);
-		check_need_resched();
-	}
-	set_cputime(0);
-	// panic("GPF");
+	interrupt_exit(regs);
+	set_cputime(interrupt_from_user(regs)? 0: 1);
+	panic("GPF");
 }
 
 void debug_handler(struct x86_user_context *regs)
@@ -1045,12 +1051,8 @@ void debug_handler(struct x86_user_context *regs)
 	memset(&info, '\0', sizeof info);
 	info.si_code = si_code;
 	set_signal(SIGTRAP, regs, &info);
-	if(interrupt_from_user(regs)){
-		cpu_enable_interrupt();
-		check_signal(0, regs, 0);
-		check_need_resched();
-	}
-	set_cputime(0);
+	interrupt_exit(regs);
+	set_cputime(interrupt_from_user(regs)? 0: 1);
 }
 
 void int3_handler(struct x86_user_context *regs)
@@ -1067,12 +1069,8 @@ void int3_handler(struct x86_user_context *regs)
 	memset(&info, '\0', sizeof info);
 	info.si_code = TRAP_BRKPT;
 	set_signal(SIGTRAP, regs, &info);
-	if(interrupt_from_user(regs)){
-		cpu_enable_interrupt();
-		check_signal(0, regs, 0);
-		check_need_resched();
-	}
-	set_cputime(0);
+	interrupt_exit(regs);
+	set_cputime(interrupt_from_user(regs)? 0: 1);
 }
 
 void
@@ -1081,7 +1079,6 @@ unhandled_page_fault(struct thread *thread, void *fault_addr, void *regs)
 	const uintptr_t address = (uintptr_t)fault_addr;
 	struct process_vm *vm = thread->vm;
 	struct vm_range *range;
-	char found;
 	unsigned long irqflags;
 	unsigned long error = ((struct x86_user_context *)regs)->gpr.error;
 
@@ -1095,17 +1092,12 @@ unhandled_page_fault(struct thread *thread, void *fault_addr, void *regs)
 			(error & PF_RSVD ? "was" : "wasn't"),
 			(error & PF_INSTR ? "was" : "wasn't"));
 
-	found = 0;
-	list_for_each_entry(range, &vm->vm_range_list, list) {
-		if (range->start <= address && range->end > address) {
-			found = 1;
-			__kprintf("address is in range, flag: 0x%lx\n",
-					range->flag);
-			ihk_mc_pt_print_pte(vm->address_space->page_table, (void*)address);
-			break;
-		}
-	}
-	if (!found) {
+	range = lookup_process_memory_range(vm, address, address+1);
+	if (range) {
+		__kprintf("address is in range, flag: 0x%lx\n",
+				range->flag);
+		ihk_mc_pt_print_pte(vm->address_space->page_table, (void*)address);
+	} else {
 		__kprintf("address is out of range! \n");
 	}
 
@@ -1233,6 +1225,13 @@ void cpu_pause(void)
 	asm volatile("pause" ::: "memory");
 }
 
+/* From: kernel-xppsl_1.5.2/arch/x86/include/asm/processor.h */
+/* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
+void cpu_relax(void)
+{
+	asm volatile("rep; nop" ::: "memory");
+}
+
 /*@
   @ assigns \nothing;
   @ ensures \interrupt_disabled > 0;
@@ -1473,29 +1472,91 @@ void ihk_mc_delay_us(int us)
 	arch_delay(us);
 }
 
-#define EXTENDED_ARCH_SHOW_CONTEXT
-#ifdef EXTENDED_ARCH_SHOW_CONTEXT
 void arch_show_extended_context(void)
 {
-	unsigned long cr0, cr4, msr, xcr0;
+	unsigned long cr0, cr4, msr, xcr0 = 0;
 
 	/*  Read and print CRs, MSR_EFER, XCR0  */
 	asm volatile("movq %%cr0, %0" : "=r"(cr0));
 	asm volatile("movq %%cr4, %0" : "=r"(cr4));
 	msr = rdmsr(MSR_EFER);
-	xcr0 = xgetbv(0);
-
+	if (xsave_available) {
+		xcr0 = xgetbv(0);
+	}
 	__kprintf("\n             CR0              CR4\n");
 	__kprintf("%016lX %016lX\n", cr0, cr4);
 
 	__kprintf("             MSR_EFER\n");
 	__kprintf("%016lX\n", msr);
 
-	__kprintf("             XCR0\n");
-	__kprintf("%016lX\n", xcr0);
-
+	if (xsave_available) {
+		__kprintf("             XCR0\n");
+		__kprintf("%016lX\n", xcr0);
+	}
+}
+
+struct stack {
+	struct stack *rbp;
+	unsigned long eip;
+};
+
+/* KPRINTF_LOCAL_BUF_LEN is 1024, useless to go further */
+#define STACK_BUF_LEN (1024-sizeof("[  0]: "))
+static void __print_stack(struct stack *rbp, unsigned long first) {
+	char buf[STACK_BUF_LEN];
+	size_t len;
+
+	/* Build string in buffer to output a single line */
+	len = snprintf(buf, STACK_BUF_LEN,
+		       "addr2line -e smp-x86/kernel/mckernel.img -fpia");
+
+	if (first)
+		len += snprintf(buf + len, STACK_BUF_LEN - len,
+				" %#16lx", first);
+
+	while ((unsigned long)rbp > 0xffff880000000000 &&
+			STACK_BUF_LEN - len > sizeof(" 0x0123456789abcdef")) {
+		len += snprintf(buf + len, STACK_BUF_LEN - len,
+				" %#16lx", rbp->eip);
+		rbp = rbp->rbp;
+	}
+	__kprintf("%s\n", buf);
+}
+
+void arch_print_pre_interrupt_stack(const struct x86_basic_regs *regs) {
+	struct stack *rbp;
+
+	/* only for kernel stack */
+	if (regs->error & PF_USER)
+		return;
+
+	__kprintf("Pre-interrupt stack trace:\n");
+
+	/* interrupt stack heuristics:
+	 * - the first entry looks like it is always garbage, so skip.
+	 * (that is done by taking regs->rsp instead of &regs->rsp)
+	 * - that still looks sometimes wrong. For now, if it is not
+	 * within 64k of itself, look for the next entry that matches.
+	 */
+
+	rbp = (struct stack*)regs->rsp;
+
+	while ((uintptr_t)rbp > (uintptr_t)rbp->rbp
+			|| (uintptr_t)rbp + 0x10000 < (uintptr_t)rbp->rbp)
+		rbp = (struct stack *)(((uintptr_t *)rbp) + 1);
+
+	__print_stack(rbp, regs->rip);
+}
+
+void arch_print_stack() {
+	struct stack *rbp;
+
+	__kprintf("Approximative stack trace:\n");
+
+	asm("mov %%rbp, %0" : "=r"(rbp) );
+
+	__print_stack(rbp, 0);
 }
-#endif
 
 /*@
   @ requires \valid(reg);
@@ -1526,9 +1587,11 @@ void arch_show_interrupt_context(const void *reg)
 	__kprintf("%16lx %16lx %16lx %16lx\n",
 	        regs->cs, regs->ss, regs->rflags, regs->error);
 
-#ifdef EXTENDED_ARCH_SHOW_CONTEXT
-        arch_show_extended_context();
-#endif	
+kprintf_unlock(irqflags);
+return;
+	arch_show_extended_context();
+
+	arch_print_pre_interrupt_stack(regs);
 
 	kprintf_unlock(irqflags);
 }
@@ -1651,13 +1714,11 @@ release_fp_regs(struct thread *thread)
 	thread->fp_regs = NULL;
 }
 
-/*@
-  @ requires \valid(thread);
-  @*/
-void
-save_fp_regs(struct thread *thread)
+static int
+check_and_allocate_fp_regs(struct thread *thread)
 {
-	int	pages;
+	int pages;
+	int result = 0;
 
 	if (!thread->fp_regs) {
 		pages = (xsave_size + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
@@ -1666,12 +1727,26 @@ save_fp_regs(struct thread *thread)
 
 		if (!thread->fp_regs) {
 			kprintf("error: allocating fp_regs pages\n");
-			return;
+			result = 1;
+			goto out;
 		}
 
-		memset(thread->fp_regs, 0, sizeof(fp_regs_struct));
 		memset(thread->fp_regs, 0, pages * PAGE_SIZE);
 	}
+out:
+	return result;
+}
+
+/*@
+  @ requires \valid(thread);
+  @*/
+void
+save_fp_regs(struct thread *thread)
+{
+	if (check_and_allocate_fp_regs(thread) != 0) {
+		// alloc error
+		return;
+	}
 
 	if (xsave_available) {
 		unsigned int low, high;
@@ -1687,6 +1762,13 @@ save_fp_regs(struct thread *thread)
 	}
 }
 
+void copy_fp_regs(struct thread *from, struct thread *to)
+{
+	if ((from->fp_regs != NULL) && (check_and_allocate_fp_regs(to) == 0)) {
+		memcpy(to->fp_regs, from->fp_regs, sizeof(fp_regs_struct));
+	}
+}
+
 #ifdef POSTK_DEBUG_TEMP_FIX_19
 void
 clear_fp_regs(struct thread *thread)

arch/x86_64/kernel/gencore.c

@@ -289,7 +289,7 @@ int gencore(struct thread *thread, void *regs,
 	Elf64_Ehdr eh;
 	Elf64_Phdr *ph = NULL;
 	void *note = NULL;
-	struct vm_range *range;
+	struct vm_range *range, *next;
 	struct process_vm *vm = thread->vm;
 	int segs = 1;	/* the first one is for NOTE */
 	int notesize, phsize, alignednotesize;
@@ -303,13 +303,18 @@ int gencore(struct thread *thread, void *regs,
 		return -1;
 	}
 
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		dkprintf("start:%lx end:%lx flag:%lx objoff:%lx\n", 
 			 range->start, range->end, range->flag, range->objoff);
 		/* We omit reserved areas because they are only for
 		   mckernel's internal use. */		   
 		if (range->flag & VR_RESERVED)
 			continue;
+		if (range->flag & VR_DONTDUMP)
+			continue;
 		/* We need a chunk for each page for a demand paging area.
 		   This can be optimized for spacial complexity but we would
 		   lose simplicity instead. */
@@ -391,7 +396,10 @@ int gencore(struct thread *thread, void *regs,
 
 	/* program header for each memory chunk */
 	i = 1;
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		unsigned long flag = range->flag;
 		unsigned long size = range->end - range->start;
 
@@ -432,7 +440,10 @@ int gencore(struct thread *thread, void *regs,
 	dkprintf("coretable[2]: %lx@%lx(%lx)\n", ct[2].len, ct[2].addr, note);
 
 	i = 3;	/* memory segments */
-	list_for_each_entry(range, &vm->vm_range_list, list) {
+	next = lookup_process_memory_range(vm, 0, -1);
+	while ((range = next)) {
+		next = next_process_memory_range(vm, range);
+
 		unsigned long phys;
 
 		if (range->flag & VR_RESERVED)

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

@@ -64,7 +64,6 @@ static inline int futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval,
 	return oldval;
 }
 
-#ifdef POSTK_DEBUG_ARCH_DEP_8 /* arch depend hide */
 static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
 {
 	int op = (encoded_op >> 28) & 7;
@@ -128,6 +127,13 @@ static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
 	}
 	return ret;
 }
-#endif	/* !POSTK_DEBUG_ARCH_DEP_8 */
+
+static inline int get_futex_value_locked(uint32_t *dest, uint32_t *from)
+{
+
+	*dest = *(volatile uint32_t *)from;
+
+	return 0;
+}
 
 #endif

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

@@ -14,7 +14,17 @@
 int __kprintf(const char *format, ...);
 #endif
 
-typedef int ihk_spinlock_t;
+typedef unsigned short __ticket_t;
+typedef unsigned int __ticketpair_t;
+
+typedef struct ihk_spinlock {
+	union {
+		__ticketpair_t head_tail;
+		struct __raw_tickets {
+			__ticket_t head, tail;
+		} tickets;
+	};
+} ihk_spinlock_t;
 
 extern void preempt_enable(void);
 extern void preempt_disable(void);
@@ -23,9 +33,9 @@ extern void preempt_disable(void);
 
 static void ihk_mc_spinlock_init(ihk_spinlock_t *lock)
 {
-	*lock = 0;
+	lock->head_tail = 0;
 }
-#define SPIN_LOCK_UNLOCKED 0
+#define SPIN_LOCK_UNLOCKED { .head_tail = 0 }
 
 #ifdef DEBUG_SPINLOCK
 #define ihk_mc_spinlock_lock_noirq(l) { \
@@ -39,40 +49,24 @@ __kprintf("[%d] ret ihk_mc_spinlock_lock_noirq\n", ihk_mc_get_processor_id()); \
 
 static void __ihk_mc_spinlock_lock_noirq(ihk_spinlock_t *lock)
 {
-	int inc = 0x00010000;
-	int tmp;
-
-#if 0
-	asm volatile("lock ; xaddl %0, %1\n"
-	             "movzwl %w0, %2\n\t"
-	             "shrl $16, %0\n\t"
-	             "1:\t"
-	             "cmpl %0, %2\n\t"
-	             "je 2f\n\t"
-	             "rep ; nop\n\t"
-	             "movzwl %1, %2\n\t"
-	             "jmp 1b\n"
-	             "2:"
-	             : "+Q" (inc), "+m" (*lock), "=r" (tmp) : : "memory", "cc");
-#endif
+	register struct __raw_tickets inc = { .tail = 0x0002 };
 
 	preempt_disable();
 
-	asm volatile("lock; xaddl %0, %1\n"
-			"movzwl %w0, %2\n\t"
-			"shrl $16, %0\n\t"
-			"1:\t"
-			"cmpl %0, %2\n\t"
-			"je 2f\n\t"
-			"rep ; nop\n\t"
-			"movzwl %1, %2\n\t"
-			/* don't need lfence here, because loads are in-order */
-			"jmp 1b\n"
-			"2:"
-			: "+r" (inc), "+m" (*lock), "=&r" (tmp)
-			:
-			: "memory", "cc");
+	asm volatile ("lock xaddl %0, %1\n"
+			: "+r" (inc), "+m" (*(lock)) : : "memory", "cc");
 
+	if (inc.head == inc.tail)
+		goto out;
+
+	for (;;) {
+		if (*((volatile __ticket_t *)&lock->tickets.head) == inc.tail)
+			goto out;
+		cpu_pause();
+	}
+
+out:
+	barrier();	/* make sure nothing creeps before the lock is taken */
 }
 
 #ifdef DEBUG_SPINLOCK
@@ -106,8 +100,11 @@ __kprintf("[%d] ret ihk_mc_spinlock_unlock_noirq\n", ihk_mc_get_processor_id());
 #endif
 static void __ihk_mc_spinlock_unlock_noirq(ihk_spinlock_t *lock)
 {
-	asm volatile ("lock incw %0" : "+m"(*lock) : : "memory", "cc");
-	
+	__ticket_t inc = 0x0002;
+
+	asm volatile ("lock addw %1, %0\n"
+			: "+m" (lock->tickets.head) : "ri" (inc) : "memory", "cc");
+
 	preempt_enable();
 }
 
@@ -134,6 +131,8 @@ typedef struct mcs_lock_node {
 	unsigned long irqsave;
 } __attribute__((aligned(64))) mcs_lock_node_t;
 
+typedef mcs_lock_node_t mcs_lock_t;
+
 static void mcs_lock_init(struct mcs_lock_node *node)
 {
 	node->locked = 0;

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

@@ -40,18 +40,42 @@
 #define LARGE_PAGE_MASK    (~((unsigned long)LARGE_PAGE_SIZE - 1))
 #define LARGE_PAGE_P2ALIGN (LARGE_PAGE_SHIFT - PAGE_SHIFT)
 
+#define GB_PAGE_SHIFT   30
+#define GB_PAGE_SIZE    (1UL << GB_PAGE_SHIFT)
+#define GB_PAGE_MASK    (~((unsigned long)GB_PAGE_SIZE - 1))
+#define GB_PAGE_P2ALIGN (GB_PAGE_SHIFT - PAGE_SHIFT)
+
+
 #define USER_END           0x0000800000000000UL
 #define TASK_UNMAPPED_BASE 0x00002AAAAAA00000UL
+
+/*
+ * Canonical negative addresses (i.e., the smallest kernel virtual address)
+ * on x86 64 bit mode (in its most restricted 48 bit format) starts from
+ * 0xffff800000000000, but Linux starts mapping physical memory at 0xffff880000000000.
+ * The 0x80000000000 long gap (8TBs, i.e., 16 PGD level entries in the page tables)
+ * is used for Xen hyervisor (see arch/x86/include/asm/page.h) and that is
+ * what we utilize for McKernel.
+ * This gives us the benefit of being able to use Linux kernel virtual
+ * addresses identically as in Linux.
+ *
+ * NOTE: update these also in eclair.c when modified!
+ */
 #define MAP_ST_START       0xffff800000000000UL
-#define MAP_VMAP_START     0xfffff00000000000UL
-#define MAP_FIXED_START    0xffffffff70000000UL
-#define MAP_KERNEL_START   0xffffffff80000000UL
+#define MAP_VMAP_START     0xffff850000000000UL
+#define MAP_FIXED_START    0xffff860000000000UL
+#define LINUX_PAGE_OFFSET  0xffff880000000000UL
+/*
+ * MAP_KERNEL_START is 8MB below MODULES_END in Linux.
+ * Placing the LWK image in the virtual address space at the end of
+ * the Linux modules section enables us to map the LWK TEXT in Linux
+ * as well, so that Linux can also call into LWK text.
+ */
+#define MAP_KERNEL_START   0xFFFFFFFFFE800000UL
 #define STACK_TOP(region)  ((region)->user_end)
 
 #define MAP_VMAP_SIZE      0x0000000100000000UL
 
-#define KERNEL_PHYS_OFFSET MAP_ST_START
-
 #define PTL4_SHIFT         39
 #define PTL4_SIZE          (1UL << PTL4_SHIFT)
 #define PTL3_SHIFT         30

arch/x86_64/kernel/include/arch/cpu.h

@@ -13,16 +13,16 @@
 #ifndef ARCH_CPU_H
 #define ARCH_CPU_H
 
-#include <ihk/cpu.h>
+#define arch_barrier()	asm volatile("" : : : "memory")
 
 static inline void rmb(void)
 {
-	barrier();
+	arch_barrier();
 }
 
 static inline void wmb(void)
 {
-	barrier();
+	arch_barrier();
 }
 
 static unsigned long read_tsc(void)

arch/x86_64/kernel/include/ihk/atomic.h

@@ -133,7 +133,7 @@ static inline void ihk_atomic64_inc(ihk_atomic64_t *v)
  * Note 2: xchg has side effect, so that attribute volatile is necessary,
  *	  but generally the primitive is invalid, *ptr is output argument. --ANK
  */
-#define __xg(x) ((volatile long *)(x))
+#define __xg(x) ((volatile typeof(x))(x))
 
 #define xchg4(ptr, x)						\
 ({									\

arch/x86_64/kernel/include/ihk/context.h

@@ -49,6 +49,7 @@ typedef struct x86_user_context ihk_mc_user_context_t;
 #define ihk_mc_syscall_arg5(uc) (uc)->gpr.r9
 
 #define ihk_mc_syscall_ret(uc)  (uc)->gpr.rax
+#define ihk_mc_syscall_number(uc)  (uc)->gpr.orig_rax
 
 #define ihk_mc_syscall_pc(uc)   (uc)->gpr.rip
 #define ihk_mc_syscall_sp(uc)   (uc)->gpr.rsp

arch/x86_64/kernel/include/registers.h

@@ -189,9 +189,30 @@ struct tss64 {
 } __attribute__((packed));
 
 struct x86_basic_regs {
-	unsigned long r15, r14, r13, r12, rbp, rbx, r11, r10;
-	unsigned long r9, r8, rax, rcx, rdx, rsi, rdi, error;
-	unsigned long rip, cs, rflags, rsp, ss;
+	unsigned long r15;
+	unsigned long r14;
+	unsigned long r13;
+	unsigned long r12;
+	unsigned long rbp;
+	unsigned long rbx;
+	unsigned long r11;
+	unsigned long r10;
+	unsigned long r9;
+	unsigned long r8;
+	unsigned long rax;
+	unsigned long rcx;
+	unsigned long rdx;
+	unsigned long rsi;
+	unsigned long rdi;
+	union {
+		unsigned long orig_rax; /* syscall */
+		unsigned long error;    /* interrupts */
+	};
+	unsigned long rip;
+	unsigned long cs;
+	unsigned long rflags;
+	unsigned long rsp;
+	unsigned long ss;
 };
 
 struct x86_sregs {

arch/x86_64/kernel/include/signal.h

@@ -18,6 +18,11 @@
 #define _NSIG_BPW	64
 #define _NSIG_WORDS (_NSIG / _NSIG_BPW)
 
+static inline int valid_signal(unsigned long sig)
+{
+	return sig <= _NSIG ? 1 : 0;
+}
+
 typedef unsigned long int __sigset_t;
 #define __sigmask(sig)  (((__sigset_t) 1) << ((sig) - 1))
 

arch/x86_64/kernel/include/syscall_list.h

@@ -39,7 +39,7 @@ SYSCALL_HANDLED(15, rt_sigreturn)
 SYSCALL_HANDLED(16, ioctl)
 SYSCALL_DELEGATED(17, pread64)
 SYSCALL_DELEGATED(18, pwrite64)
-SYSCALL_DELEGATED(20, writev)
+SYSCALL_HANDLED(20, writev)
 SYSCALL_DELEGATED(21, access)
 SYSCALL_DELEGATED(23, select)
 SYSCALL_HANDLED(24, sched_yield)
@@ -56,7 +56,7 @@ SYSCALL_HANDLED(36, getitimer)
 SYSCALL_HANDLED(38, setitimer)
 SYSCALL_HANDLED(39, getpid)
 SYSCALL_HANDLED(56, clone)
-SYSCALL_DELEGATED(57, fork)
+SYSCALL_HANDLED(57, fork)
 SYSCALL_HANDLED(58, vfork)
 SYSCALL_HANDLED(59, execve)
 SYSCALL_HANDLED(60, exit)

arch/x86_64/kernel/interrupt.S

@@ -145,6 +145,8 @@ nmi:
 		movq	%rsp,%gs:PANIC_REGS+0x08
 
 		movl	nmi_mode(%rip),%eax
+		cmp $3,%rax
+		je  4f
 		cmp	$1,%rax
 		je	1f
 		cmp	$2,%rax
@@ -199,9 +201,9 @@ nmi:
 		movl	%eax,%gs:PANIC_REGS+0xA0
 		movq	$1,%gs:PANICED
 		call	ihk_mc_query_mem_areas
-1:
+4:
 		hlt
-		jmp	1b
+		jmp	4b
 
 .globl x86_syscall
 x86_syscall:

arch/x86_64/kernel/local.c

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

arch/x86_64/kernel/memory.c

@@ -41,6 +41,8 @@ extern char _head[], _end[];
 
 extern unsigned long x86_kernel_phys_base;
 
+int safe_kernel_map = 0;
+
 /* Arch specific early allocation routine */
 void *early_alloc_pages(int nr_pages)
 {
@@ -109,6 +111,7 @@ struct page_table {
 };
 
 static struct page_table *init_pt;
+static int init_pt_loaded = 0;
 static ihk_spinlock_t init_pt_lock;
 
 static int use_1gb_page = 0;
@@ -172,19 +175,23 @@ static void init_normal_area(struct page_table *pt)
 	unsigned long map_start, map_end, phys, pt_phys;
 	int ident_index, virt_index;
 
-	map_start = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_START, 0);
+	/*
+	 * This has to start from 0x00, see load_file() in IHK-SMP.
+	 * For security reasons, we could skip holes in the LWK
+	 * assigned physical memory, but Linux mappings already map
+	 * those anyway.
+	 */
+	map_start = 0;
 	map_end = ihk_mc_get_memory_address(IHK_MC_GMA_MAP_END, 0);
 
-	kprintf("map_start = %lx, map_end = %lx\n", map_start, map_end);
 	ident_index = map_start >> PTL4_SHIFT;
 	virt_index = (MAP_ST_START >> PTL4_SHIFT) & (PT_ENTRIES - 1);
 
 	memset(pt, 0, sizeof(struct page_table));
 
-	for (phys = (map_start & ~(PTL4_SIZE - 1)); phys < map_end;
-	     phys += PTL4_SIZE) {
-		pt_phys = setup_l3(ihk_mc_alloc_pages(1, IHK_MC_AP_CRITICAL), phys,
-		                   map_start, map_end);
+	for (phys = map_start; phys < map_end; phys += PTL4_SIZE) {
+		pt_phys = setup_l3(ihk_mc_alloc_pages(1, IHK_MC_AP_CRITICAL),
+				phys, map_start, map_end);
 
 		pt->entry[ident_index++] = pt_phys | PFL4_PDIR_ATTR;
 		pt->entry[virt_index++] = pt_phys | PFL4_PDIR_ATTR;
@@ -493,7 +500,7 @@ uint64_t ihk_mc_pt_virt_to_pagemap(struct page_table *pt, unsigned long virt)
 
 	error = ihk_mc_pt_virt_to_phys(pt, (void *)virt, &phys);
 	if (error) {
-		return 0;
+		return PM_PSHIFT(PAGE_SHIFT);
 	}
 
 	pagemap = PM_PFRAME(phys >> PAGE_SHIFT);
@@ -724,6 +731,26 @@ static void destroy_page_table(int level, struct page_table *pt)
 	return;
 }
 
+void ihk_mc_pt_destroy_pgd_subtree(struct page_table *pt, void *virt)
+{
+	int l4idx, l3idx, l2idx, l1idx;
+	unsigned long v = (unsigned long)virt;
+	struct page_table *lower;
+
+	GET_VIRT_INDICES(v, l4idx, l3idx, l2idx, l1idx);
+
+	if (!(pt->entry[l4idx] & PF_PRESENT))
+		return;
+
+	lower = (struct page_table *)
+		phys_to_virt(pt->entry[l4idx] & PT_PHYSMASK);
+	destroy_page_table(3, lower);
+
+	pt->entry[l4idx] = 0;
+	dkprintf("%s: virt: 0x%lx, l4idx: %d subtree destroyed\n",
+		__FUNCTION__, virt, l4idx);
+}
+
 void ihk_mc_pt_destroy(struct page_table *pt)
 {
 	const int level = 4;	/* PML4 */
@@ -1542,7 +1569,6 @@ static int clear_range_l1(void *args0, pte_t *ptep, uint64_t base,
 				dkprintf("%lx-,%s: calling memory_stat_rss_sub(),phys=%lx,size=%ld,pgsize=%ld\n", pte_get_phys(&old), __FUNCTION__, pte_get_phys(&old), PTL1_SIZE, PTL1_SIZE);
 				rusage_memory_stat_sub(args->memobj, PTL1_SIZE, PTL1_SIZE); 
 			}
-			args->vm->currss -= PTL1_SIZE;
 		}  else {
 			dkprintf("%s: !calling memory_stat_rss_sub(),virt=%lx,phys=%lx\n", __FUNCTION__, base, pte_get_phys(&old));
 		}
@@ -1611,7 +1637,6 @@ static int clear_range_l2(void *args0, pte_t *ptep, uint64_t base,
 					dkprintf("%lx-,%s: calling memory_stat_rss_sub(),phys=%lx,size=%ld,pgsize=%ld\n", pte_get_phys(&old), __FUNCTION__, pte_get_phys(&old), PTL2_SIZE, PTL2_SIZE);
 					rusage_memory_stat_sub(args->memobj, PTL2_SIZE, PTL2_SIZE); 
 				}
-				args->vm->currss -= PTL2_SIZE;
 			}
 		}
 
@@ -1693,7 +1718,6 @@ static int clear_range_l3(void *args0, pte_t *ptep, uint64_t base,
 					dkprintf("%lx-,%s: calling memory_stat_rss_sub(),phys=%lx,size=%ld,pgsize=%ld\n", pte_get_phys(&old), __FUNCTION__, pte_get_phys(&old), PTL3_SIZE, PTL3_SIZE);
 					rusage_memory_stat_sub(args->memobj, PTL3_SIZE, PTL3_SIZE); 
 				}
-				args->vm->currss -= PTL3_SIZE;
 			}
 		}
 
@@ -1963,6 +1987,28 @@ out:
 	return ptep;
 }
 
+pte_t *ihk_mc_pt_lookup_fault_pte(struct process_vm *vm, void *virt,
+		int pgshift, void **basep, size_t *sizep, int *p2alignp)
+{
+	int faulted = 0;
+	pte_t *ptep;
+
+retry:
+	ptep = ihk_mc_pt_lookup_pte(vm->address_space->page_table,
+			virt, pgshift, basep, sizep, p2alignp);
+	if (!faulted && (!ptep || !pte_is_present(ptep))) {
+		page_fault_process_vm(vm, virt, PF_POPULATE | PF_USER);
+		faulted = 1;
+		goto retry;
+	}
+
+	if (faulted && ptep && pte_is_present(ptep)) {
+		kprintf("%s: successfully faulted 0x%lx\n", __FUNCTION__, virt);
+	}
+
+	return ptep;
+}
+
 pte_t *ihk_mc_pt_lookup_pte(page_table_t pt, void *virt, int pgshift,
 		void **basep, size_t *sizep, int *p2alignp)
 {
@@ -2262,7 +2308,7 @@ out:
 
 int ihk_mc_pt_set_range(page_table_t pt, struct process_vm *vm, void *start, 
 		void *end, uintptr_t phys, enum ihk_mc_pt_attribute attr,
-						int pgshift, struct vm_range *range)
+		int pgshift, struct vm_range *range)
 {
 	int error;
 	struct set_range_args args;
@@ -2606,6 +2652,61 @@ void init_low_area(struct page_table *pt)
 	set_pt_large_page(pt, 0, 0, PTATTR_NO_EXECUTE|PTATTR_WRITABLE);
 }
 
+static void init_linux_kernel_mapping(struct page_table *pt)
+{
+	unsigned long map_start, map_end, phys;
+	void *virt;
+	int nr_memory_chunks, chunk_id, numa_id;
+	
+	/* In case of safe_kernel_map option (safe_kernel_map == 1), 
+	   processing to prevent destruction of the memory area on Linux side 
+	   is executed */
+	if (safe_kernel_map == 0) {
+		kprintf("Straight-map entire physical memory\n");
+
+		/* Map 2 TB for now */
+		map_start = 0;
+		map_end = 0x20000000000;
+
+		virt = (void *)LINUX_PAGE_OFFSET;
+
+		kprintf("Linux kernel virtual: 0x%lx - 0x%lx -> 0x%lx - 0x%lx\n",
+			LINUX_PAGE_OFFSET, LINUX_PAGE_OFFSET + map_end, 0, map_end);
+
+		for (phys = map_start; phys < map_end; phys += LARGE_PAGE_SIZE) {
+			if (set_pt_large_page(pt, virt, phys, PTATTR_WRITABLE) != 0) {
+				kprintf("%s: error setting mapping for 0x%lx\n", __FUNCTION__, virt);
+			}
+			virt += LARGE_PAGE_SIZE;
+		}
+	} else {
+		kprintf("Straight-map physical memory areas allocated to McKernel\n");
+
+		nr_memory_chunks = ihk_mc_get_nr_memory_chunks();
+		if (nr_memory_chunks == 0) {
+			kprintf("%s: ERROR: No memory chunk available.\n", __FUNCTION__);
+			return;
+		}
+
+		for (chunk_id = 0; chunk_id < nr_memory_chunks; chunk_id++) {
+			if (ihk_mc_get_memory_chunk(chunk_id, &map_start, &map_end, &numa_id)) {
+				kprintf("%s: ERROR: Memory chunk id (%d) out of range.\n", __FUNCTION__, chunk_id);
+				continue;
+			}
+			
+			dkprintf("Linux kernel virtual: 0x%lx - 0x%lx -> 0x%lx - 0x%lx\n",
+					 LINUX_PAGE_OFFSET + map_start, LINUX_PAGE_OFFSET + map_end, map_start, map_end);
+
+			virt = (void *)(LINUX_PAGE_OFFSET + map_start);
+			for (phys = map_start; phys < map_end; phys += LARGE_PAGE_SIZE, virt += LARGE_PAGE_SIZE) {
+				if (set_pt_large_page(pt, virt, phys, PTATTR_WRITABLE) != 0) {
+					kprintf("%s: set_pt_large_page() failed for 0x%lx\n", __FUNCTION__, virt);
+				}
+			}
+		}
+	}
+}
+
 static void init_vsyscall_area(struct page_table *pt)
 {
 	extern char vsyscall_page[];
@@ -2631,13 +2732,15 @@ void init_page_table(void)
 
 	/* Normal memory area */
 	init_normal_area(init_pt);
+	init_linux_kernel_mapping(init_pt);
 	init_fixed_area(init_pt);
 	init_low_area(init_pt);
 	init_text_area(init_pt);
 	init_vsyscall_area(init_pt);
 
 	load_page_table(init_pt);
-	kprintf("Page table is now at %p\n", init_pt);
+	init_pt_loaded = 1;
+	kprintf("Page table is now at 0x%lx\n", init_pt);
 }
 
 extern void __reserve_arch_pages(unsigned long, unsigned long,
@@ -2665,17 +2768,33 @@ void ihk_mc_reserve_arch_pages(struct ihk_page_allocator_desc *pa_allocator,
 unsigned long virt_to_phys(void *v)
 {
 	unsigned long va = (unsigned long)v;
-	
+
 	if (va >= MAP_KERNEL_START) {
+		dkprintf("%s: MAP_KERNEL_START <= 0x%lx <= LINUX_PAGE_OFFSET\n",
+				__FUNCTION__, va);
 		return va - MAP_KERNEL_START + x86_kernel_phys_base;
-	} else {
+	}
+	else if (va >= LINUX_PAGE_OFFSET) {
+		return va - LINUX_PAGE_OFFSET;
+	}
+	else if (va >= MAP_FIXED_START) {
+		return va - MAP_FIXED_START;
+	}
+	else {
+		dkprintf("%s: MAP_ST_START <= 0x%lx <= MAP_FIXED_START\n",
+				__FUNCTION__, va);
 		return va - MAP_ST_START;
 	}
 }
 
 void *phys_to_virt(unsigned long p)
 {
-	return (void *)(p + MAP_ST_START);
+	/* Before loading our own PT use straight mapping */
+	if (!init_pt_loaded) {
+		return (void *)(p + MAP_ST_START);
+	}
+
+	return (void *)(p + LINUX_PAGE_OFFSET);
 }
 
 int copy_from_user(void *dst, const void *src, size_t siz)

arch/x86_64/kernel/perfctr.c

@@ -10,9 +10,12 @@
 #include <ihk/perfctr.h>
 #include <march.h>
 #include <errno.h>
+#include <cls.h>
 #include <ihk/debug.h>
+#include <ihk/cpu.h>
 #include <registers.h>
 #include <mc_perf_event.h>
+#include <config.h>
 
 extern unsigned int *x86_march_perfmap;
 extern int running_on_kvm(void);
@@ -57,6 +60,10 @@ void x86_init_perfctr(void)
 	uint64_t ecx;
 	uint64_t edx;
 
+#ifndef ENABLE_PERF
+	return;
+#endif //ENABLE_PERF
+
 	/* Do not do it on KVM */
 	if (running_on_kvm()) return;
 
@@ -93,7 +100,7 @@ void x86_init_perfctr(void)
 	for(i = 0; i < X86_IA32_NUM_PERF_COUNTERS; i++) {
 		wrmsr(MSR_IA32_PERFEVTSEL0 + i, 0);
 	}
-		
+
 	/* Enable PMC Control */
 	value = rdmsr(MSR_PERF_GLOBAL_CTRL);
 	value |= X86_IA32_PERF_COUNTERS_MASK;
@@ -254,6 +261,41 @@ int ihk_mc_perfctr_init(int counter, enum ihk_perfctr_type type, int mode)
 	return set_perfctr_x86_direct(counter, mode, x86_march_perfmap[type]);
 }
 
+int ihk_mc_perfctr_set_extra(struct mc_perf_event *event)
+{
+	struct thread *thread = cpu_local_var(current);
+
+	// allocate extra_reg
+	if (thread->extra_reg_alloc_map & (1UL << event->extra_reg.idx)) {
+		if (event->extra_reg.idx == EXTRA_REG_RSP_0) {
+			event->extra_reg.idx = EXTRA_REG_RSP_1;
+		}
+		else if (event->extra_reg.idx == EXTRA_REG_RSP_1) {
+			event->extra_reg.idx = EXTRA_REG_RSP_0;
+		}
+
+		if (thread->extra_reg_alloc_map & (1UL << event->extra_reg.idx)) {
+			// extra_regs are full
+			return -1;
+		}
+	}
+
+	if (event->extra_reg.idx == EXTRA_REG_RSP_0) {
+		event->hw_config &= ~0xffUL;
+		event->hw_config |= ihk_mc_get_extra_reg_event(EXTRA_REG_RSP_0);
+		event->extra_reg.reg = MSR_OFFCORE_RSP_0;
+	}
+	else if (event->extra_reg.idx == EXTRA_REG_RSP_1) {
+		event->hw_config &= ~0xffUL;
+		event->hw_config |= ihk_mc_get_extra_reg_event(EXTRA_REG_RSP_1);
+		event->extra_reg.reg = MSR_OFFCORE_RSP_1;
+	}
+		
+	thread->extra_reg_alloc_map |= (1UL << event->extra_reg.idx);
+	wrmsr(event->extra_reg.reg, event->extra_reg.config);
+	return 0;
+}
+
 #ifdef HAVE_MARCH_PERFCTR_START
 extern void x86_march_perfctr_start(unsigned long counter_mask);
 #endif

arch/x86_64/kernel/syscall.c

@@ -29,11 +29,12 @@
 #include <prctl.h>
 #include <ihk/ikc.h>
 #include <page.h>
+#include <limits.h>
+#include <syscall.h>
 
-void terminate(int, int);
+void terminate_mcexec(int, int);
 extern long do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact);
 long syscall(int num, ihk_mc_user_context_t *ctx);
-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);
 extern unsigned long do_fork(int, unsigned long, unsigned long, unsigned long,
@@ -142,8 +143,6 @@ SYSCALL_DECLARE(rt_sigaction)
 	struct k_sigaction new_sa, old_sa;
 	int rc;
 
-	if(sig == SIGKILL || sig == SIGSTOP || sig <= 0 || sig > 64)
-		return -EINVAL;
 	if (sigsetsize != sizeof(sigset_t))
 		return -EINVAL;
 
@@ -251,8 +250,8 @@ SYSCALL_DECLARE(rt_sigreturn)
 		regs->gpr.rflags &= ~RFLAGS_TF;
 		info.si_code = TRAP_TRACE;
 		set_signal(SIGTRAP, regs, &info);
-		check_signal(0, regs, 0);
 		check_need_resched();
+		check_signal(0, regs, 0);
 	}
 
 	if(ksigsp.fpregs && xsavesize){
@@ -279,6 +278,7 @@ SYSCALL_DECLARE(rt_sigreturn)
 extern struct cpu_local_var *clv;
 extern unsigned long do_kill(struct thread *thread, int pid, int tid, int sig, struct siginfo *info, int ptracecont);
 extern void interrupt_syscall(struct thread *, int sig);
+extern void terminate(int, int);
 extern int num_processors;
 
 #define RFLAGS_MASK (RFLAGS_CF | RFLAGS_PF | RFLAGS_AF | RFLAGS_ZF | \
@@ -460,7 +460,6 @@ void set_single_step(struct thread *thread)
 
 long ptrace_read_fpregs(struct thread *thread, void *fpregs)
 {
-	save_fp_regs(thread);
 	if (thread->fp_regs == NULL) {
 		return -ENOMEM;
 	}
@@ -470,7 +469,6 @@ long ptrace_read_fpregs(struct thread *thread, void *fpregs)
 
 long ptrace_write_fpregs(struct thread *thread, void *fpregs)
 {
-	save_fp_regs(thread);
 	if (thread->fp_regs == NULL) {
 		return -ENOMEM;
 	}
@@ -540,7 +538,7 @@ void ptrace_report_signal(struct thread *thread, int sig)
 	/* Transition thread state */
 	proc->status = PS_TRACED;
 	thread->status = PS_TRACED;
-	proc->ptrace &= ~PT_TRACE_SYSCALL_MASK;
+	proc->ptrace &= ~PT_TRACE_SYSCALL;
 	if (sig == SIGSTOP || sig == SIGTSTP ||
 			sig == SIGTTIN || sig == SIGTTOU) {
 		proc->signal_flags |= SIGNAL_STOP_STOPPED;
@@ -809,6 +807,11 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 		regs->gpr.rip = (unsigned long)k->sa.sa_handler;
 		regs->gpr.rsp = (unsigned long)usp;
 
+		// check signal handler is ONESHOT
+		if (k->sa.sa_flags & SA_RESETHAND) {
+			k->sa.sa_handler = SIG_DFL; 
+		}
+
 		if(!(k->sa.sa_flags & SA_NODEFER))
 			thread->sigmask.__val[0] |= pending->sigmask.__val[0];
 		kfree(pending);
@@ -820,8 +823,8 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			regs->gpr.rflags &= ~RFLAGS_TF;
 			info.si_code = TRAP_TRACE;
 			set_signal(SIGTRAP, regs, &info);
-			check_signal(0, regs, 0);
 			check_need_resched();
+			check_signal(0, regs, 0);
 		}
 	}
 	else {
@@ -927,6 +930,7 @@ do_signal(unsigned long rc, void *regs0, struct thread *thread, struct sig_pendi
 			break;
 		case SIGCHLD:
 		case SIGURG:
+		case SIGWINCH:
 			break;
 		default:
 			dkprintf("do_signal,default,terminate,sig=%d\n", sig);
@@ -1007,6 +1011,12 @@ interrupt_from_user(void *regs0)
 	return !(regs->gpr.rsp & 0x8000000000000000);
 }
 
+void save_syscall_return_value(int num, unsigned long rc)
+{
+	/* Empty on x86 */
+	return;
+}
+
 void
 check_signal(unsigned long rc, void *regs0, int num)
 {
@@ -1054,6 +1064,110 @@ out:
 	return;
 }
 
+static int
+check_sig_pending_thread(struct thread *thread)
+{
+	int found = 0;
+	struct list_head *head;
+	mcs_rwlock_lock_t *lock;
+	struct mcs_rwlock_node_irqsave mcs_rw_node;
+	struct sig_pending *next;
+	struct sig_pending *pending;
+	__sigset_t w;
+	__sigset_t x;
+	int sig = 0;
+	struct k_sigaction *k;
+	struct cpu_local_var *v;
+
+	v = get_this_cpu_local_var();
+	w = thread->sigmask.__val[0];
+
+	lock = &thread->sigcommon->lock;
+	head = &thread->sigcommon->sigpending;
+	for (;;) {
+		mcs_rwlock_reader_lock(lock, &mcs_rw_node);
+
+		list_for_each_entry_safe(pending, next, head, list){
+			for (x = pending->sigmask.__val[0], sig = 0; x;
+			     sig++, x >>= 1);
+			k = thread->sigcommon->action + sig - 1;
+			if ((sig != SIGCHLD && sig != SIGURG) ||
+			    (k->sa.sa_handler != (void *)1 &&
+			     k->sa.sa_handler != NULL)) {
+				if (!(pending->sigmask.__val[0] & w)) {
+					if (pending->interrupted == 0) {
+						pending->interrupted = 1;
+						found = 1;
+						if (sig != SIGCHLD &&
+						    sig != SIGURG &&
+						    !k->sa.sa_handler) {
+							found = 2;
+							break;
+						}
+					}
+				}
+			}
+		}
+
+		mcs_rwlock_reader_unlock(lock, &mcs_rw_node);
+
+		if (found == 2) {
+			break;
+		}
+
+		if (lock == &thread->sigpendinglock) {
+			break;
+		}
+
+		lock = &thread->sigpendinglock;
+		head = &thread->sigpending;
+	}
+
+	if (found == 2) {
+		ihk_mc_spinlock_unlock(&v->runq_lock, v->runq_irqstate);
+		terminate_mcexec(0, sig);
+		return 1;
+	}
+	else if (found == 1) {
+		ihk_mc_spinlock_unlock(&v->runq_lock, v->runq_irqstate);
+		interrupt_syscall(thread, 0);
+		return 1;
+	}
+	return 0;
+}
+
+void
+check_sig_pending()
+{
+	struct thread *thread;
+	struct cpu_local_var *v;
+
+	if (clv == NULL)
+		return;
+
+	v = get_this_cpu_local_var();
+repeat:
+	v->runq_irqstate = ihk_mc_spinlock_lock(&v->runq_lock);
+	list_for_each_entry(thread, &(v->runq), sched_list) {
+
+		if (thread == NULL || thread == &cpu_local_var(idle)) {
+			continue;
+		}
+
+		if (thread->in_syscall_offload == 0) {
+			continue;
+		}
+
+		if (thread->proc->exit_status & 0x0000000100000000L) {
+			continue;
+		}
+
+		if (check_sig_pending_thread(thread))
+			goto repeat;
+	}
+	ihk_mc_spinlock_unlock(&v->runq_lock, v->runq_irqstate);
+}
+
 unsigned long
 do_kill(struct thread *thread, int pid, int tid, int sig, siginfo_t *info,
         int ptracecont)
@@ -1215,15 +1329,19 @@ done:
 		mcs_rwlock_reader_lock_noirq(&tproc->update_lock, &updatelock);
 		savelock = &tthread->sigpendinglock;
 		head = &tthread->sigpending;
-		if(sig == SIGKILL ||
-		   (tproc->status != PS_EXITED &&
-		    tproc->status != PS_ZOMBIE &&
-		    tthread->status != PS_EXITED)){
-			hold_thread(tthread);
+		mcs_rwlock_reader_lock_noirq(&tproc->threads_lock, &lock);
+		if (tthread->status != PS_EXITED &&
+			(sig == SIGKILL ||
+			 (tproc->status != PS_EXITED && tproc->status != PS_ZOMBIE))) {
+			if ((rc = hold_thread(tthread))) {
+				kprintf("%s: ERROR hold_thread returned %d,tid=%d\n", __FUNCTION__, rc, tthread->tid);
+				tthread = NULL;
+			}
 		}
 		else{
 			tthread = NULL;
 		}
+		mcs_rwlock_reader_unlock_noirq(&tproc->threads_lock, &lock);
 		mcs_rwlock_reader_unlock_noirq(&tproc->update_lock, &updatelock);
 		mcs_rwlock_reader_unlock_noirq(&thash->lock[hash], &lock);
 	}
@@ -1250,7 +1368,9 @@ done:
 	}
 
 	if (tthread->thread_offloaded) {
-		interrupt_syscall(tthread, sig);
+		if (!tthread->proc->nohost) {
+			interrupt_syscall(tthread, sig);
+		}
 		release_thread(tthread);
 		return 0;
 	}
@@ -1285,6 +1405,7 @@ done:
 				rc = -ENOMEM;
 			}
 			else{
+				memset(pending, 0, sizeof(struct sig_pending));
 				pending->sigmask.__val[0] = mask;
 				memcpy(&pending->info, info, sizeof(siginfo_t));
 				pending->ptracecont = ptracecont;
@@ -1308,9 +1429,6 @@ done:
 			ihk_mc_interrupt_cpu(get_x86_cpu_local_variable(tthread->cpu_id)->apic_id, 0xd0);
 		}
 
-		if(!tthread->proc->nohost)
-			interrupt_syscall(tthread, 0);
-
 		if (status != PS_RUNNING) {
 			if(sig == SIGKILL){
 				/* Wake up the target only when stopped by ptrace-reporting */
@@ -1336,15 +1454,19 @@ set_signal(int sig, void *regs0, siginfo_t *info)
 	struct x86_user_context *regs = regs0;
 	struct thread *thread = cpu_local_var(current);
 
-	if(thread == NULL || thread->proc->pid == 0)
+	if (thread == NULL || thread->proc->pid == 0)
 		return;
 
-	if((__sigmask(sig) & thread->sigmask.__val[0]) ||
-	   (regs->gpr.rsp & 0x8000000000000000)){
+        if (!interrupt_from_user(regs)) {
+		ihk_mc_debug_show_interrupt_context(regs);
+                panic("panic: kernel mode signal");
+        }
+
+	if ((__sigmask(sig) & thread->sigmask.__val[0])) {
 		coredump(thread, regs0);
 		terminate(0, sig | 0x80);
 	}
-		do_kill(thread, thread->proc->pid, thread->tid, sig, info, 0);
+	do_kill(thread, thread->proc->pid, thread->tid, sig, info, 0);
 }
 
 SYSCALL_DECLARE(mmap)
@@ -1483,6 +1605,16 @@ SYSCALL_DECLARE(clone)
                    ihk_mc_syscall_sp(ctx));
 }
 
+SYSCALL_DECLARE(fork)
+{
+	return do_fork(SIGCHLD, 0, 0, 0, 0, ihk_mc_syscall_pc(ctx), ihk_mc_syscall_sp(ctx));
+}
+
+SYSCALL_DECLARE(vfork)
+{
+	return do_fork(CLONE_VFORK|SIGCHLD, 0, 0, 0, 0, ihk_mc_syscall_pc(ctx), ihk_mc_syscall_sp(ctx));
+}
+
 SYSCALL_DECLARE(shmget)
 {
 	const key_t key = ihk_mc_syscall_arg0(ctx);
@@ -1907,4 +2039,644 @@ save_uctx(void *uctx, struct x86_user_context *regs)
 	ctx->fregsize = 0;
 }
 
+int do_process_vm_read_writev(int pid, 
+		const struct iovec *local_iov,
+		unsigned long liovcnt,
+		const struct iovec *remote_iov,
+		unsigned long riovcnt,
+		unsigned long flags,
+		int op)
+{
+	int ret = -EINVAL;	
+	int li, ri;
+	int pli, pri;
+	off_t loff, roff;
+	size_t llen = 0, rlen = 0;
+	size_t copied = 0;
+	size_t to_copy;
+	struct thread *lthread = cpu_local_var(current);
+	struct process *rproc;
+	struct process *lproc = lthread->proc;
+	struct process_vm *rvm = NULL;
+	unsigned long rphys;
+	unsigned long rpage_left;
+	unsigned long psize;
+	void *rva;
+	struct vm_range *range;
+	struct mcs_rwlock_node_irqsave lock;
+	struct mcs_rwlock_node update_lock;
+
+	/* Sanity checks */
+	if (flags) {
+		return -EINVAL;
+	}
+	
+	if (liovcnt > IOV_MAX || riovcnt > IOV_MAX) {
+		return -EINVAL;
+	}
+
+	/* Check if parameters are okay */
+	ihk_mc_spinlock_lock_noirq(&lthread->vm->memory_range_lock);
+
+	range = lookup_process_memory_range(lthread->vm, 
+			(uintptr_t)local_iov, 
+			(uintptr_t)(local_iov + liovcnt * sizeof(struct iovec)));
+
+	if (!range) {
+		ret = -EFAULT; 
+		goto arg_out;
+	}
+
+	range = lookup_process_memory_range(lthread->vm, 
+			(uintptr_t)remote_iov, 
+			(uintptr_t)(remote_iov + riovcnt * sizeof(struct iovec)));
+
+	if (!range) {
+		ret = -EFAULT; 
+		goto arg_out;
+	}
+
+	ret = 0;
+arg_out:
+	ihk_mc_spinlock_unlock_noirq(&lthread->vm->memory_range_lock);
+
+	if (ret != 0) {
+		goto out;
+	}
+
+	for (li = 0; li < liovcnt; ++li) {
+		llen += local_iov[li].iov_len;
+		dkprintf("local_iov[%d].iov_base: 0x%lx, len: %lu\n",
+			li, local_iov[li].iov_base, local_iov[li].iov_len);
+	}
+
+	for (ri = 0; ri < riovcnt; ++ri) {
+		rlen += remote_iov[ri].iov_len;
+		dkprintf("remote_iov[%d].iov_base: 0x%lx, len: %lu\n",
+			ri, remote_iov[ri].iov_base, remote_iov[ri].iov_len);
+	}
+
+	if (llen != rlen) {
+		return -EINVAL;
+	}
+	
+	/* Find remote process */
+	rproc = find_process(pid, &lock);
+	if (!rproc) {
+		ret = -ESRCH;
+		goto out;
+	}
+
+	mcs_rwlock_reader_lock_noirq(&rproc->update_lock, &update_lock);
+	if(rproc->status == PS_EXITED ||
+	   rproc->status == PS_ZOMBIE){
+		mcs_rwlock_reader_unlock_noirq(&rproc->update_lock, &update_lock);
+		process_unlock(rproc, &lock);
+		ret = -ESRCH;
+		goto out;
+	}
+	rvm = rproc->vm;
+	hold_process_vm(rvm);
+	mcs_rwlock_reader_unlock_noirq(&rproc->update_lock, &update_lock);
+	process_unlock(rproc, &lock);
+
+	if (lproc->euid != 0 &&
+	    (lproc->ruid != rproc->ruid ||
+	     lproc->ruid != rproc->euid ||
+	     lproc->ruid != rproc->suid ||
+	     lproc->rgid != rproc->rgid ||
+	     lproc->rgid != rproc->egid ||
+	     lproc->rgid != rproc->sgid)) {
+		ret = -EPERM;
+		goto out;
+	}
+
+	dkprintf("pid %d found, doing %s: liovcnt: %d, riovcnt: %d \n", pid,
+		(op == PROCESS_VM_READ) ? "PROCESS_VM_READ" : "PROCESS_VM_WRITE",
+		liovcnt, riovcnt);
+
+	pli = pri = -1; /* Previous indeces in iovecs */
+	li = ri = 0; /* Current indeces in iovecs */
+	loff = roff = 0; /* Offsets in current iovec */
+
+	/* Now iterate and do the copy */
+	while (copied < llen) {
+		int faulted = 0;
+
+		/* New local vector? */
+		if (pli != li) {
+			struct vm_range *range;
+
+			ihk_mc_spinlock_lock_noirq(&lthread->vm->memory_range_lock);
+
+			/* Is base valid? */
+			range = lookup_process_memory_range(lthread->vm,
+					(uintptr_t)local_iov[li].iov_base,
+					(uintptr_t)(local_iov[li].iov_base + 1));
+
+			if (!range) {
+				ret = -EFAULT;
+				goto pli_out;
+			}
+
+			/* Is range valid? */
+			range = lookup_process_memory_range(lthread->vm,
+					(uintptr_t)local_iov[li].iov_base,
+					(uintptr_t)(local_iov[li].iov_base + local_iov[li].iov_len));
+
+			if (range == NULL) {
+				ret = -EINVAL;
+				goto pli_out;
+			}
+
+			if (!(range->flag & ((op == PROCESS_VM_READ) ?
+				VR_PROT_WRITE : VR_PROT_READ))) {
+				ret = -EFAULT;
+				goto pli_out;
+			}
+
+			ret = 0;
+pli_out:
+			ihk_mc_spinlock_unlock_noirq(&lthread->vm->memory_range_lock);
+
+			if (ret != 0) {
+				goto out;
+			}
+
+			pli = li;
+		}
+
+		/* New remote vector? */
+		if (pri != ri) {
+			struct vm_range *range;
+
+			ihk_mc_spinlock_lock_noirq(&rvm->memory_range_lock);
+
+			/* Is base valid? */
+			range = lookup_process_memory_range(rvm,
+					(uintptr_t)remote_iov[li].iov_base,
+					(uintptr_t)(remote_iov[li].iov_base + 1));
+
+			if (range == NULL) {
+				ret = -EFAULT;
+				goto pri_out;
+			}
+
+			/* Is range valid? */
+			range = lookup_process_memory_range(rvm,
+					(uintptr_t)remote_iov[li].iov_base,
+					(uintptr_t)(remote_iov[li].iov_base + remote_iov[li].iov_len));
+
+			if (range == NULL) {
+				ret = -EINVAL;
+				goto pri_out;
+			}
+
+			if (!(range->flag & ((op == PROCESS_VM_READ) ?
+				VR_PROT_READ : VR_PROT_WRITE))) {
+				ret = -EFAULT;
+				goto pri_out;
+			}
+
+			ret = 0;
+pri_out:
+			ihk_mc_spinlock_unlock_noirq(&rvm->memory_range_lock);
+
+			if (ret != 0) {
+				goto out;
+			}
+
+			pri = ri;
+		}
+
+		/* Figure out how much we can copy at most in this iteration */
+		to_copy = (local_iov[li].iov_len - loff);	
+		if ((remote_iov[ri].iov_len - roff) < to_copy) {
+			to_copy = remote_iov[ri].iov_len - roff;
+		}
+
+retry_lookup:
+		/* TODO: remember page and do this only if necessary */
+		ret = ihk_mc_pt_virt_to_phys_size(rvm->address_space->page_table,
+				remote_iov[ri].iov_base + roff, &rphys, &psize);
+
+		if (ret) {
+			uint64_t reason = PF_POPULATE | PF_WRITE | PF_USER;
+			void *addr;
+
+			if (faulted) {
+				ret = -EFAULT;
+				goto out;
+			}
+
+			/* Fault in pages */
+			for (addr = (void *)
+					(((unsigned long)remote_iov[ri].iov_base + roff)
+					& PAGE_MASK);
+					addr < (remote_iov[ri].iov_base + roff + to_copy);
+					addr += PAGE_SIZE) {
+
+				ret = page_fault_process_vm(rvm, addr, reason);
+				if (ret) {
+					ret = -EFAULT;
+					goto out;
+				}
+			}
+
+			faulted = 1;
+			goto retry_lookup;
+		}
+
+		rpage_left = ((((unsigned long)remote_iov[ri].iov_base + roff +
+			psize) & ~(psize - 1)) -
+			((unsigned long)remote_iov[ri].iov_base + roff));
+		if (rpage_left < to_copy) {
+			to_copy = rpage_left;
+		}
+
+		rva = phys_to_virt(rphys);
+
+		fast_memcpy(
+				(op == PROCESS_VM_READ) ? local_iov[li].iov_base + loff : rva,
+				(op == PROCESS_VM_READ) ? rva : local_iov[li].iov_base + loff,
+				to_copy);
+
+		copied += to_copy;
+		dkprintf("local_iov[%d]: 0x%lx %s remote_iov[%d]: 0x%lx, %lu copied, psize: %lu, rpage_left: %lu\n",
+			li, local_iov[li].iov_base + loff, 
+			(op == PROCESS_VM_READ) ? "<-" : "->", 
+			ri, remote_iov[ri].iov_base + roff, to_copy,
+			psize, rpage_left);
+
+		loff += to_copy;
+		roff += to_copy;
+
+		if (loff == local_iov[li].iov_len) {
+			li++;
+			loff = 0;
+		}
+
+		if (roff == remote_iov[ri].iov_len) {
+			ri++;
+			roff = 0;
+		}
+	}
+
+	release_process_vm(rvm);
+
+	return copied;
+
+out:
+	if(rvm)
+		release_process_vm(rvm);
+	return ret;
+}
+
+int move_pages_smp_handler(int cpu_index, int nr_cpus, void *arg)
+{
+	int i, i_s, i_e, phase = 1;
+	struct move_pages_smp_req *mpsr =
+		(struct move_pages_smp_req *)arg;
+	struct process_vm *vm = mpsr->proc->vm;
+	int count = mpsr->count;
+	struct page_table *save_pt;
+	extern struct page_table *get_init_page_table(void);
+
+	i_s = (count / nr_cpus) * cpu_index;
+	i_e = i_s + (count / nr_cpus);
+	if (cpu_index == (nr_cpus - 1)) {
+		i_e = count;
+	}
+
+	/* Load target process' PT so that we can access user-space */
+	save_pt = cpu_local_var(current) == &cpu_local_var(idle) ?
+		get_init_page_table() :
+		cpu_local_var(current)->vm->address_space->page_table;
+
+	if (save_pt != vm->address_space->page_table) {
+		ihk_mc_load_page_table(vm->address_space->page_table);
+	}
+	else {
+		save_pt = NULL;
+	}
+
+	if (nr_cpus == 1) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				mpsr->nodes_ready = 1;
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus > 1 && nr_cpus < 4) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+			case 1:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				mpsr->nodes_ready = 1;
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus >= 4 && nr_cpus < 8) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * count);
+				break;
+			case 1:
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				break;
+			case 2:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				mpsr->nodes_ready = 1;
+				break;
+			case 3:
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				memset(mpsr->status, 0, sizeof(int) * count);
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				break;
+
+			default:
+				break;
+		}
+	}
+	else if (nr_cpus >= 8) {
+		switch (cpu_index) {
+			case 0:
+				memcpy(mpsr->virt_addr, mpsr->user_virt_addr,
+						sizeof(void *) * (count / 2));
+				break;
+			case 1:
+				memcpy(mpsr->virt_addr + (count / 2),
+						mpsr->user_virt_addr + (count / 2),
+						sizeof(void *) * (count / 2));
+				break;
+			case 2:
+				memcpy(mpsr->status, mpsr->user_status,
+						sizeof(int) * count);
+				break;
+			case 3:
+				memcpy(mpsr->nodes, mpsr->user_nodes,
+						sizeof(int) * count);
+				mpsr->nodes_ready = 1;
+				break;
+			case 4:
+				memset(mpsr->ptep, 0, sizeof(pte_t) * count);
+				break;
+			case 5:
+				memset(mpsr->status, 0, sizeof(int) * count);
+				break;
+			case 6:
+				memset(mpsr->nr_pages, 0, sizeof(int) * count);
+				break;
+			case 7:
+				memset(mpsr->dst_phys, 0,
+						sizeof(unsigned long) * count);
+				break;
+
+			default:
+				break;
+		}
+	}
+
+	while (!(volatile int)mpsr->nodes_ready) {
+		cpu_pause();
+	}
+
+	/* NUMA verification in parallel */
+	for (i = i_s; i < i_e; i++) {
+		if (mpsr->nodes[i] < 0 ||
+				mpsr->nodes[i] >= ihk_mc_get_nr_numa_nodes() ||
+				!test_bit(mpsr->nodes[i],
+					mpsr->proc->vm->numa_mask)) {
+			mpsr->phase_ret = -EINVAL;
+			break;
+		}
+	}
+
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	/* PTE lookup in parallel */
+	for (i = i_s; i < i_e; i++) {
+		void *phys;
+		size_t pgsize;
+		int p2align;
+		/*
+		 * XXX: No page structures for anonymous mappings.
+		 * Look up physical addresses by scanning page tables.
+		 */
+		mpsr->ptep[i] = ihk_mc_pt_lookup_pte(vm->address_space->page_table,
+				(void *)mpsr->virt_addr[i], 0, &phys, &pgsize, &p2align);
+
+		/* PTE valid? */
+		if (!mpsr->ptep[i] || !pte_is_present(mpsr->ptep[i])) {
+			mpsr->status[i] = -ENOENT;
+			mpsr->ptep[i] = NULL;
+			continue;
+		}
+
+		/* PTE is file? */
+		if (pte_is_fileoff(mpsr->ptep[i], PAGE_SIZE)) {
+			mpsr->status[i] = -EINVAL;
+			mpsr->ptep[i] = NULL;
+			continue;
+		}
+
+		dkprintf("%s: virt 0x%lx:%lu requested to be moved to node %d\n",
+			__FUNCTION__, mpsr->virt_addr[i], pgsize, mpsr->nodes[i]);
+
+		/* Large page? */
+		if (pgsize > PAGE_SIZE) {
+			int nr_sub_pages = (pgsize / PAGE_SIZE);
+			int j;
+
+			if (i + nr_sub_pages > count) {
+				kprintf("%s: ERROR: page at index %d exceeds the region\n",
+						__FUNCTION__, i);
+				mpsr->status[i] = -EINVAL;
+				break;
+			}
+
+			/* Is it contiguous across nr_sub_pages and all
+			 * requested to be moved to the same target node? */
+			for (j = 0; j < nr_sub_pages; ++j) {
+				if (mpsr->virt_addr[i + j] !=
+				(mpsr->virt_addr[i] + (j * PAGE_SIZE)) ||
+						mpsr->nodes[i] != mpsr->nodes[i + j]) {
+					kprintf("%s: ERROR: virt address or node at index %d"
+							" is inconsistent\n",
+							__FUNCTION__, i + j);
+					mpsr->phase_ret = -EINVAL;
+					goto pte_out;
+				}
+			}
+
+			mpsr->nr_pages[i] = nr_sub_pages;
+			i += (nr_sub_pages - 1);
+		}
+		else {
+			mpsr->nr_pages[i] = 1;
+		}
+	}
+
+pte_out:
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	if (cpu_index == 0) {
+		/* Allocate new pages on target NUMA nodes */
+		for (i = 0; i < count; i++) {
+			int pgalign = 0;
+			int j;
+			void *dst;
+
+			if (!mpsr->ptep[i] || mpsr->status[i] < 0 || !mpsr->nr_pages[i])
+				continue;
+
+			/* TODO: store pgalign info in an array as well? */
+			if (mpsr->nr_pages[i] > 1) {
+				if (mpsr->nr_pages[i] * PAGE_SIZE == PTL2_SIZE)
+					pgalign = PTL2_SHIFT - PTL1_SHIFT;
+			}
+
+			dst = ihk_mc_alloc_aligned_pages_node(mpsr->nr_pages[i],
+					pgalign, IHK_MC_AP_USER, mpsr->nodes[i]);
+
+			if (!dst) {
+				mpsr->status[i] = -ENOMEM;
+				continue;
+			}
+
+			for (j = i; j < (i + mpsr->nr_pages[i]); ++j) {
+				mpsr->status[j] = mpsr->nodes[i];
+			}
+
+			mpsr->dst_phys[i] = virt_to_phys(dst);
+
+			dkprintf("%s: virt 0x%lx:%lu to node %d, pgalign: %d,"
+					" allocated phys: 0x%lx\n",
+					__FUNCTION__, mpsr->virt_addr[i],
+					mpsr->nr_pages[i] * PAGE_SIZE,
+					mpsr->nodes[i], pgalign, mpsr->dst_phys[i]);
+		}
+	}
+
+	/* Barrier */
+	ihk_atomic_inc(&mpsr->phase_done);
+	while (ihk_atomic_read(&mpsr->phase_done) <
+			(phase * nr_cpus)) {
+		cpu_pause();
+	}
+
+	if (mpsr->phase_ret != 0) {
+		goto out;
+	}
+
+	dkprintf("%s: phase %d done\n", __FUNCTION__, phase);
+	++phase;
+
+	/* Copy, PTE update, memfree in parallel */
+	for (i = i_s; i < i_e; ++i) {
+		if (!mpsr->dst_phys[i])
+			continue;
+
+		fast_memcpy(phys_to_virt(mpsr->dst_phys[i]),
+				phys_to_virt(pte_get_phys(mpsr->ptep[i])),
+				mpsr->nr_pages[i] * PAGE_SIZE);
+
+		ihk_mc_free_pages(
+				phys_to_virt(pte_get_phys(mpsr->ptep[i])),
+				mpsr->nr_pages[i]);
+
+		pte_update_phys(mpsr->ptep[i], mpsr->dst_phys[i]);
+
+		dkprintf("%s: virt 0x%lx:%lu copied and remapped to phys: 0x%lu\n",
+				__FUNCTION__, mpsr->virt_addr[i],
+				mpsr->nr_pages[i] * PAGE_SIZE,
+				mpsr->dst_phys[i]);
+	}
+
+	/* XXX: do a separate SMP call with only CPUs running threads
+	 * of this process? */
+	if (cpu_local_var(current)->proc == mpsr->proc) {
+		/* Invalidate all TLBs */
+		for (i = 0; i < mpsr->count; i++) {
+			if (!mpsr->dst_phys[i])
+				continue;
+
+			flush_tlb_single((unsigned long)mpsr->virt_addr[i]);
+		}
+	}
+
+out:
+	if (save_pt) {
+		ihk_mc_load_page_table(save_pt);
+	}
+
+	return mpsr->phase_ret;
+}
+
+time_t time(void) {
+	struct syscall_request sreq IHK_DMA_ALIGN;
+	struct thread *thread = cpu_local_var(current);
+	time_t ret;
+	sreq.number = __NR_time;
+	sreq.args[0] = (uintptr_t)NULL;
+	ret = (time_t)do_syscall(&sreq, ihk_mc_get_processor_id(), thread->proc->pid);
+	return ret;
+}
+
 /*** End of File ***/

arch/x86_64/kernel/vsyscall.c

@@ -102,7 +102,7 @@ int vsyscall_gettimeofday(struct timeval *tv, void *tz)
 			: "%rcx", "%r11", "memory");
 
 	if (error) {
-		*(int *)0 = 0;	/* i.e. raise(SIGSEGV) */
+		*(volatile int *)0 = 0;	/* i.e. raise(SIGSEGV) */
 	}
 	return error;
 } /* vsyscall_gettimeofday() */

arch/x86_64/tools/mcoverlay-create-smp-x86.sh.in

@@ -45,7 +45,6 @@ error_exit() {
 
 	exit 1
 }
-
 fi
 
 if [ ! -e /tmp/mcos ]; then 

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

@@ -19,6 +19,7 @@ ETCDIR=@ETCDIR@
 KMODDIR="${prefix}/kmod"
 KERNDIR="${prefix}/@TARGET@/kernel"
 ENABLE_MCOVERLAYFS="@ENABLE_MCOVERLAYFS@"
+MCK_BUILDID=@BUILDID@
 
 mem="512M@0"
 cpus=""
@@ -43,8 +44,12 @@ fi
 
 turbo=""
 ihk_irq=""
+safe_kernel_map=""
+umask_old=`umask`
+idle_halt=""
+allow_oversubscribe=""
 
-while getopts :tk:c:m:o:f:r:q:i:d: OPT
+while getopts :stk:c:m:o:f:r:q:i:d:e:hO OPT
 do
 	case ${OPT} in
 	f)	facility=${OPTARG}
@@ -57,21 +62,32 @@ do
 		;;
 	m) mem=${OPTARG}
 		;;
+	s) safe_kernel_map="safe_kernel_map"
+		;;
 	r) ikc_map=${OPTARG}
 		;;
 	q) ihk_irq=${OPTARG}
 		;;
 	t) turbo="turbo"
 		;;
+	e) extra_kopts=${OPTARG}
+		;;
 	d) DUMP_LEVEL=${OPTARG}
 		;;
 	i) mon_interval=${OPTARG}
 		;;
+	h) idle_halt="idle_halt"
+		;;
+	O) allow_oversubscribe="allow_oversubscribe"
+		;;
 	*)  echo "invalid option -${OPT}" >&2
 		exit 1
 	esac
 done
 
+redirect_kmsg=0
+turbo="turbo"
+
 # Start ihkmond
 pid=`pidof ihkmond`
 if [ "${pid}" != "" ]; then
@@ -80,6 +96,16 @@ fi
 if [ "${redirect_kmsg}" != "0" -o "${mon_interval}" != "-1" ]; then
     ${SBINDIR}/ihkmond -f ${facility} -k ${redirect_kmsg} -i ${mon_interval}
 fi
+
+disable_irqbalance_mck() {
+	if [ -f /etc/systemd/system/irqbalance_mck.service ]; then
+		systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null
+
+		# Invalid .service file persists so remove it
+		rm -f /etc/systemd/system/irqbalance_mck.service
+	fi
+}
+
 #
 # Revert any state that has been initialized before the error occured.
 #
@@ -87,6 +113,16 @@ error_exit() {
 	local status=$1
 
 	case $status in
+	irqbalance_mck_started)
+		if [ "${irqbalance_used}" == "yes" ]; then
+		    if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
+			if ! systemctl stop irqbalance_mck.service 2>/dev/null; then
+			    echo "warning: failed to stop irqbalance_mck" >&2
+			fi
+			disable_irqbalance_mck
+		    fi
+		fi
+		;&
 	mcos_sys_mounted)
 		if [ "$enable_mcoverlay" == "yes" ]; then
 			umount /tmp/mcos/mcos0_sys
@@ -148,25 +184,25 @@ error_exit() {
 		fi
 		;&
 	ihk_smp_loaded)
-		rmmod ihk_smp_x86 2>/dev/null || echo "warning: failed to remove ihk_smp_x86" >&2
+		rmmod ihk_smp_@ARCH@ 2>/dev/null || echo "warning: failed to remove ihk_smp_@ARCH@" >&2
 		;&
 	ihk_loaded)
 		rmmod ihk 2>/dev/null || echo "warning: failed to remove ihk" >&2
 		;&
+	smp_affinity_modified)
+		umask $umask_old
+		if [ "${irqbalance_used}" == "yes" ]; then
+		    if ! perl -e '$tmpdir="/tmp/mcreboot"; @files = grep { -f } glob "$tmpdir/proc/irq/*/smp_affinity"; foreach $file (@files) { $dest = substr($file, length($tmpdir)); if (0) {print "cp $file $dest\n";} system("cp $file $dest 2>/dev/null"); }'; then
+			echo "warning: failed to restore /proc/irq/*/smp_affinity" >&2
+		    fi
+		    if [ -e /tmp/mcreboot ]; then rm -rf /tmp/mcreboot; fi
+		fi
+		;&
 	irqbalance_stopped)
-		if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
-			if ! systemctl stop irqbalance_mck.service 2>/dev/null; then
-				echo "warning: failed to stop irqbalance_mck" >&2
-			fi
-			if ! systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null; then
-				echo "warning: failed to disable irqbalance_mck" >&2
-			fi
-			if ! etcdir=@ETCDIR@ perl -e '$etcdir=$ENV{'etcdir'}; @files = grep { -f } glob "$etcdir/proc/irq/*/smp_affinity"; foreach $file (@files) { $dest = substr($file, length($etcdir)); if(0) {print "cp $file $dest\n";} system("cp $file $dest 2>/dev/null"); }'; then
-				echo "warning: failed to restore /proc/irq/*/smp_affinity" >&2
-			fi
-			if ! systemctl start irqbalance.service; then
-				echo "warning: failed to start irqbalance" >&2;
-			fi
+		if [ "${irqbalance_used}" == "yes" ]; then
+		    if ! systemctl start irqbalance.service; then
+			echo "warning: failed to start irqbalance" >&2;
+		    fi
 		fi
 		;&
 	initial)
@@ -200,7 +236,7 @@ if [ "${ENABLE_MCOVERLAYFS}" == "yes" ]; then
 			enable_mcoverlay="yes"
 		fi
 	else
-		if [ ${linux_version_code} -eq 199168 -a ${rhel_release} -ge 327 ]; then
+		if [ ${linux_version_code} -eq 199168 -a ${rhel_release} -ge 327 -a ${rhel_release} -le 693 ]; then
 			enable_mcoverlay="yes"
 		fi
 		if [ ${linux_version_code} -ge 262144 -a ${linux_version_code} -lt 262400 ]; then
@@ -236,9 +272,9 @@ if [ "${irqbalance_used}" == "yes" ]; then
 		exit 1
     fi;
 
-    if ! etcdir=@ETCDIR@ perl -e 'use File::Copy qw(copy); $etcdir=$ENV{'etcdir'}; @files = grep { -f } glob "/proc/irq/*/smp_affinity"; foreach $file (@files) { $rel = substr($file, 1); $dir=substr($rel, 0, length($rel)-length("/smp_affinity")); if(0) { print "cp $file $etcdir/$rel\n";} if(system("mkdir -p $etcdir/$dir")){ exit 1;} if(!copy($file,"$etcdir/$rel")){ exit 1;} }'; then
+    if ! perl -e 'use File::Copy qw(copy); $tmpdir="/tmp/mcreboot"; @files = grep { -f } glob "/proc/irq/*/smp_affinity"; foreach $file (@files) { $rel = substr($file, 1); $dir = substr($rel, 0, length($rel) - length("/smp_affinity")); if (system("mkdir -p $tmpdir/$dir")) { exit 1; } if (0) { print "cp $file $tmpdir/$rel\n"; } if (!copy($file,"$tmpdir/$rel")) { exit 1; } }'; then
 		echo "error: saving /proc/irq/*/smp_affinity" >&2
-		error_exit "mcos_sys_mounted"
+		error_exit "irqbalance_stopped"
 	fi;
 
 # Prevent /proc/irq/*/smp_affinity from getting zero after offlining
@@ -252,32 +288,45 @@ if [ "${irqbalance_used}" == "yes" ]; then
 
     if ! ncpus=$ncpus smp_affinity_mask=$smp_affinity_mask perl -e '@dirs = grep { -d } glob "/proc/irq/*"; foreach $dir (@dirs) { $hit = 0; $affinity_str = `cat $dir/smp_affinity`; chomp $affinity_str; @int32strs = split /,/, $affinity_str; @int32strs_mask=split /,/, $ENV{'smp_affinity_mask'}; for($i=0;$i <= $#int32strs_mask; $i++) { $int32strs_inv[$i] = sprintf("%08x",hex($int32strs_mask[$i])^0xffffffff); if($i == 0) { $len = int((($ENV{'ncpus'}%32)+3)/4); if($len != 0) { $int32strs_inv[$i] = substr($int32strs_inv[$i], -$len, $len); } } } $inv = join(",", @int32strs_inv); $nint32s = int(($ENV{'ncpus'}+31)/32); for($j = $nint32s - 1; $j >= 0; $j--) { if(hex($int32strs[$nint32s - 1 - $j]) & hex($int32strs_mask[$nint32s - 1 - $j])) { $hit = 1; }} if($hit == 1) { $cmd = "echo $inv > $dir/smp_affinity 2>/dev/null"; system $cmd;}}'; then
 		echo "error: modifying /proc/irq/*/smp_affinity" >&2
-		error_exit "mcos_sys_mounted"
+		error_exit "irqbalance_stopped"
 	fi
-
 fi
 
+# Set umask so that proc/sys files/directories created by
+# mcctrl.ko and mcreboot.sh have appropriate permission bits
+umask_dec=$(( 8#${umask_old} & 8#0002 ))
+umask 0`printf "%o" ${umask_dec}`
+
 # Load IHK if not loaded
 if ! grep -E 'ihk\s' /proc/modules &>/dev/null; then
 	if ! taskset -c 0 insmod ${KMODDIR}/ihk.ko 2>/dev/null; then
 		echo "error: loading ihk" >&2
-		error_exit "irqbalance_stopped"
+		error_exit "smp_affinity_modified"
 	fi
 fi
 
-# Increase swappiness so that we have better chance to allocate memory for IHK
-echo 100 > /proc/sys/vm/swappiness
+# Copy modules under /tmp to avoid loading from shared FS
+if mkdir -p /tmp/mcos-kmod; then
+	cp ${KMODDIR}/* /tmp/mcos-kmod/
+	KMODDIR="/tmp/mcos-kmod/"
+fi
 
-# Drop Linux caches to free memory
-sync && echo 3 > /proc/sys/vm/drop_caches
+# Fujitsu drops caches for us in between jobs so don't do it on OFP
+if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" == "" ]; then
+	# Increase swappiness so that we have better chance to allocate memory for IHK
+	echo 100 > /proc/sys/vm/swappiness
 
-# Merge free memory areas into large, physically contigous ones
-echo 1 > /proc/sys/vm/compact_memory 2>/dev/null
+	# Drop Linux caches to free memory
+	sync && echo 3 > /proc/sys/vm/drop_caches
 
-sync
+	# Merge free memory areas into large, physically contigous ones
+	echo 1 > /proc/sys/vm/compact_memory 2>/dev/null
+
+	sync
+fi
 
 # Load IHK-SMP if not loaded and reserve CPUs and memory
-if ! grep ihk_smp_x86 /proc/modules &>/dev/null; then
+if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
 	if [ "$ihk_irq" == "" ]; then
 		for i in `seq 64 255`; do
 			if [ ! -d /proc/irq/$i ] && [ "`cat /proc/interrupts | grep ":" | awk '{print $1}' | grep -o '[0-9]*' | grep -e '^$i$'`" == "" ]; then
@@ -290,30 +339,46 @@ if ! grep ihk_smp_x86 /proc/modules &>/dev/null; then
 			error_exit "ihk_loaded"
 		fi
 	fi
-	if ! taskset -c 0 insmod ${KMODDIR}/ihk-smp-x86.ko ihk_start_irq=$ihk_irq ihk_ikc_irq_core=$ihk_ikc_irq_core 2>/dev/null; then
-		echo "error: loading ihk-smp-x86" >&2
+	if ! taskset -c 0 insmod ${KMODDIR}/ihk-smp-@ARCH@.ko ihk_start_irq=$ihk_irq ihk_ikc_irq_core=$ihk_ikc_irq_core 2>/dev/null; then
+		echo "error: loading ihk-smp-@ARCH@" >&2
 		error_exit "ihk_loaded"
 	fi
 
-	# Offline-reonline RAM (special case for OFP SNC-4 mode)
-	if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" != "" ] && [ "`cat /sys/devices/system/node/online`" == "0-7" ]; then
-		for i in  0 1 2 3; do
-			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
-				echo 0 > $f 2>&1 > /dev/null;
-			done
-			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
-				echo 1 > $f 2>&1 > /dev/null;
-			done
-		done
-		for i in 4 5 6 7; do
-			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
-				echo 0 > $f 2>&1 > /dev/null;
-			done
-			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
-				echo 1 > $f 2>&1 > /dev/null;
-			done
-		done
-	fi
+# 	# Offline-reonline RAM (special case for OFP SNC-4 flat mode)
+# 	if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" != "" ] && [ "`cat /sys/devices/system/node/online`" == "0-7" ]; then
+# 		for i in  0 1 2 3; do
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 0 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 1 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 		done
+# 		for i in 4 5 6 7; do
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 0 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 		done
+# 		for i in 4 5 6 7; do
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 1 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 		done
+# 	fi
+#
+# 	# Offline-reonline RAM (special case for OFP Quadrant flat mode)
+# 	if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" != "" ] && [ "`cat /sys/devices/system/node/online`" == "0-1" ]; then
+# 		for i in 1; do
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 0 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 		done
+# 		for i in 1; do
+# 			find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
+# 				echo 1 | tee $f 2>/dev/null 1>/dev/null
+# 			done
+# 		done
+# 	fi
 
 	if ! ${SBINDIR}/ihkconfig 0 reserve mem ${mem}; then
 		echo "error: reserving memory" >&2
@@ -333,13 +398,20 @@ if ! grep mcctrl /proc/modules &>/dev/null; then
 	fi
 fi
 
+# Check that different versions of binaries/scripts are not mixed
+IHK_BUILDID=`${SBINDIR}/ihkconfig 0 get buildid`
+if [  "${IHK_BUILDID}" != "${MCK_BUILDID}" ]; then
+	echo "IHK build-id (${IHK_BUILDID}) didn't match McKernel build-id (${MCK_BUILDID})." >&2
+	exit 1
+fi
+
 # Destroy all LWK instances
 if ls /dev/mcos* 1>/dev/null 2>&1; then
 	for i in /dev/mcos*; do
 		ind=`echo $i|cut -c10-`;
 		# Retry when conflicting with ihkmond
 		nretry=0
-		until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -lt 4 ]; do
+		until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -ge 4 ]; do
 		    sleep 0.25
 		    nretry=$[ $nretry + 1 ]
 		done
@@ -383,7 +455,7 @@ if ! ${SBINDIR}/ihkosctl 0 load ${KERNDIR}/mckernel.img; then
 fi
 
 # Set kernel arguments
-if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo dump_level=${DUMP_LEVEL}"; then
+if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo $safe_kernel_map $idle_halt dump_level=${DUMP_LEVEL} $extra_kopts $allow_oversubscribe"; then
 	echo "error: setting kernel arguments" >&2
 	error_exit "os_created"
 fi
@@ -409,7 +481,9 @@ if [ "${irqbalance_used}" == "yes" ]; then
     banirq=`cat /proc/interrupts| perl -e 'while(<>) { if(/^\s*(\d+).*IHK\-SMP\s*$/) {print $1;}}'`
 
     sed "s/%mask%/$smp_affinity_mask/g" $ETCDIR/irqbalance_mck.in | sed "s/%banirq%/$banirq/g" > /tmp/irqbalance_mck
-	systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null
+
+	disable_irqbalance_mck
+
 	if ! systemctl link $ETCDIR/irqbalance_mck.service >/dev/null 2>/dev/null; then
 		echo "error: linking irqbalance_mck" >&2
 		error_exit "mcos_sys_mounted"
@@ -421,3 +495,8 @@ if [ "${irqbalance_used}" == "yes" ]; then
 	fi
 #    echo cpus=$cpus ncpus=$ncpus banirq=$banirq
 fi
+
+# Restore umask
+umask ${umask_old}
+
+exit 0

arch/x86_64/tools/mcstop+release-smp-x86.sh.in

@@ -18,17 +18,24 @@ mem=""
 cpus=""
 irqbalance_used=""
 
+disable_irqbalance_mck() {
+	if [ -f /etc/systemd/system/irqbalance_mck.service ]; then
+		systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null
+
+		# Invalid .service file persists so remove it
+		rm -f /etc/systemd/system/irqbalance_mck.service
+	fi
+}
+
 # No SMP module? Exit.
-if ! grep ihk_smp_x86 /proc/modules &>/dev/null; then exit 0; fi
+if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then exit 0; fi
 
 if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
 	irqbalance_used="yes"
 	if ! systemctl stop irqbalance_mck.service 2>/dev/null; then
 		echo "warning: failed to stop irqbalance_mck" >&2
 	fi
-	if ! systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null; then
-		echo "warning: failed to disable irqbalance_mck" >&2
-	fi
+	disable_irqbalance_mck
 fi
 
 # Destroy all LWK instances
@@ -37,7 +44,7 @@ if ls /dev/mcos* 1>/dev/null 2>&1; then
 		ind=`echo $i|cut -c10-`;
 		# Retry when conflicting with ihkmond
 		nretry=0
-		until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -lt 4 ]; do
+		until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -ge 4 ]; do
 		    sleep 0.25
 		    nretry=$[ $nretry + 1 ]
 		done
@@ -48,6 +55,9 @@ if ls /dev/mcos* 1>/dev/null 2>&1; then
 	done
 fi
 
+# Allow ihkmond to flush kmsg buffer
+sleep 2.0
+
 # Query IHK-SMP resources and release them
 if ! ${SBINDIR}/ihkconfig 0 query cpu > /dev/null; then
 	echo "error: querying cpus" >&2
@@ -62,17 +72,23 @@ if [ "${cpus}" != "" ]; then
 	fi
 fi
 
-if ! ${SBINDIR}/ihkconfig 0 query mem > /dev/null; then
-	echo "error: querying memory" >&2
-	exit 1
-fi
+#if ! ${SBINDIR}/ihkconfig 0 query mem > /dev/null; then
+#	echo "error: querying memory" >&2
+#	exit 1
+#fi
+#
+#mem=`${SBINDIR}/ihkconfig 0 query mem`
+#if [ "${mem}" != "" ]; then
+#	if ! ${SBINDIR}/ihkconfig 0 release mem $mem > /dev/null; then
+#		echo "error: releasing memory" >&2
+#		exit 1
+#	fi
+#fi
 
-mem=`${SBINDIR}/ihkconfig 0 query mem`
-if [ "${mem}" != "" ]; then
-	if ! ${SBINDIR}/ihkconfig 0 release mem $mem > /dev/null; then
-		echo "error: releasing memory" >&2
-		exit 1
-	fi
+# Release all memory
+if ! ${SBINDIR}/ihkconfig 0 release mem "all" > /dev/null; then
+	echo "error: releasing memory" >&2
+	exit 1
 fi
 
 # Remove delegator if loaded
@@ -87,9 +103,9 @@ fi
 . ${SBINDIR}/mcoverlay-destroy.sh
 
 # Remove SMP module
-if grep ihk_smp_x86 /proc/modules &>/dev/null; then
-	if ! rmmod ihk_smp_x86 2>/dev/null; then
-		echo "error: removing ihk_smp_x86" >&2
+if grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
+	if ! rmmod ihk_smp_@ARCH@ 2>/dev/null; then
+		echo "error: removing ihk_smp_@ARCH@" >&2
 		exit 1
 	fi
 fi
@@ -110,9 +126,10 @@ fi
 
 # Start irqbalance with the original settings
 if [ "${irqbalance_used}" != "" ]; then
-	if ! etcdir=@ETCDIR@ perl -e '$etcdir=$ENV{'etcdir'}; @files = grep { -f } glob "$etcdir/proc/irq/*/smp_affinity"; foreach $file (@files) { $dest = substr($file, length($etcdir)); if(0) {print "cp $file $dest\n";} system("cp $file $dest 2>/dev/null"); }'; then
+	if ! perl -e '$tmpdir="/tmp/mcreboot"; @files = grep { -f } glob "$tmpdir/proc/irq/*/smp_affinity"; foreach $file (@files) { $dest = substr($file, length($tmpdir)); if (0) {print "cp $file $dest\n";} system("cp $file $dest 2>/dev/null"); }'; then
 		echo "warning: failed to restore /proc/irq/*/smp_affinity" >&2
 	fi
+	if [ -e /tmp/mcreboot ]; then rm -rf /tmp/mcreboot; fi
 	if ! systemctl start irqbalance.service; then
 		echo "warning: failed to start irqbalance" >&2;
 	fi

arch/x86_64/tools/mpimcexec.1in

@@ -0,0 +1,60 @@
+.\" Man page for mpimcexec
+.\"
+
+.TH MPIMCEXEC 1 "@MCKERNEL_RELEASE_DATE@" "Version @MCKERNEL_VERSION@" MCKERNEL @MCKERNEL_VERSION@"
+.SH NAME
+mpimcexec \- run an MPI application on McKernel
+.\"
+
+.\" ----------------------------  SYNOPSIS ----------------------------
+.SH SYNOPSIS
+.B mpimcexec \fR [\fIoptions\fR] \fI<command>\fR
+
+.\" ----------------------------  DESCRIPTION ----------------------------
+.SH DESCRIPTION
+mpimcexec is a wrapper script for running MPI applications on McKernel.
+It internally calls mpiexec to spawn mcexec on compute nodes, which in
+turn runs \fI<command>\fR on McKernel. mpimcexec specifies a number of
+mcexec arguments that enable high performance execution.
+
+.\" ----------------------------  OPTIONS ----------------------------
+.SH OPTIONS
+
+.TP
+.B	-ppn N, --ppn N, --ranks-per-node N
+Specify the number of MPI ranks per node. This argument is required.
+.TP
+.B	-n N, --n N, --ranks N
+Specify the number of total MPI ranks.
+e.g.,
+   $ mpimcexec -n 32 -ppn 4 ./a.out
+.br
+In the above example, 32 MPI processes are invoked
+on eight compute nodes each of which has four processes.
+.TP
+.B	--nodes N
+Specify the number of compute nodes.
+By default, all nodes, specified by "PJM --mpi proc" option, are used.
+.TP
+.B	--env, -env 
+Pass an additional environment variable
+.TP
+.B	-m N, --numa N
+Specify preferred NUMA node.
+.TP
+.B	-h <file name>, ---hostfile <file name>
+Specify a host file for MPI.
+.TP
+.B	--help
+Show help message.
+
+.PP
+.\" ----------------------------  SEE ALSO ----------------------------
+.SH SEE ALSO
+\fBmcexec\fR (1), \fBmpiexec\fR (1)
+
+.\" ----------------------------  AUTHORS ----------------------------
+.SH AUTHORS
+Copyright (C) 2018 McKernel Development Team, RIKEN, Japan
+
+

arch/x86_64/tools/mpimcexec.in

@@ -0,0 +1,147 @@
+#!/bin/bash
+#
+# OFP McKernel MPI wrapper script
+# author: Balazs Gerofi <bgerofi@riken.jp>
+#      Copyright (C) 2018  RIKEN R-CCS
+#
+
+prefix="@prefix@"
+BINDIR="${prefix}/bin"
+
+if [ "${BASH_VERSINFO[0]}" -lt 4 ]; then
+	echo "You need at least bash-4.0 to run this script." >&2
+	exit 1
+fi
+
+RANKS=""
+NODES=""
+PPN=""
+MPI_ENV=""
+COMMAND=""
+NUMA=""
+HOSTFILE=""
+
+if [ ! -z "${PJM_PROC_BY_NODE}" ]; then
+	PPN=${PJM_PROC_BY_NODE}
+elif [ ! -z "${MPI_LOCALNRANKS}" ]; then
+	PPN=${MPI_LOCALNRANKS}
+fi
+
+help_exit() {
+	echo ""
+	echo "Spawn an McKernel MPI job on Oakforest-PACS."
+	echo "usage: `basename $0` -ppn ranks_per_node [--nodes nodes] [-n ranks] [--env additional_environment]... command"
+	echo ""
+	echo "   -ppn | --ppn | --ranks-per-node         Number of MPI ranks per node (required)"
+	echo "   -n | --n | --ranks                      Total number of MPI ranks in the job"
+	echo "   --nodes                                 Number of nodes to be used"
+	echo "   --env | -env                            Pass an additional environment variable"
+	echo "   -m | --numa                             Preferred NUMA node(s)"
+	echo "   -h | --hostfile                         Host file for MPI"
+	echo "   --help                                  Show help message"
+	exit 1
+}
+
+# Parse options
+while true; do
+	case $1 in
+		-ppn | --ppn | --ranks-per-node )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs an interger value for -ppn, --ppn, or --ranks-per-node option"
+			    help_exit
+			fi
+			PPN=$2
+			shift 2
+			;;
+		-n | --n | --ranks )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs an interger value for -n, --n, or --ranks option"
+			    help_exit
+			fi
+			RANKS=$2
+			shift 2
+			;;
+		-m | --numa )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs an interger value for -m or --numa option"
+			    help_exit
+			fi
+			NUMA="-m $2"
+			shift 2
+			;;
+		--nodes )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs an interger value for --nodes option"
+			    help_exit
+			fi
+			NODES=$2
+			shift 2
+			;;
+		--env | -env )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs an environment variable name for -env or --env option"
+			    help_exit
+			fi
+			if [ -z "`echo $2 | grep I_MPI_PIN`" ]; then
+				MPI_ENV=`echo "${MPI_ENV} -env $2" | xargs`
+			fi
+			shift 2
+			;;
+		-h | --hostfile )
+			if [ $# -lt 2 ]; then
+			    echo "error: needs a file name for -h or --hostfile option"
+			    help_exit
+			fi
+			HOSTFILE="-hostfile $2"
+			shift 2
+			;;
+		--help )
+			help_exit
+			;;
+		* )
+			COMMAND=$@
+			break
+			;;
+	esac
+done
+
+if [ -z ${PPN} ]; then
+	echo "error: please specify the number of ranks per node"
+	help_exit
+fi
+
+# Unless explicitly specified, use Fujitsu inherited value
+if [ -z ${NODES} ]; then
+	NODES=${PJM_VNODES}
+fi
+
+if [ -z ${RANKS} ] && [ -z ${NODES} ]; then
+	echo "error: please specify the total number of ranks or the number of nodes"
+	help_exit
+fi
+
+if [ "x${COMMAND}" = "x" ]; then
+	echo "error: please specify command"
+	help_exit
+fi
+
+# Calculate total job size if not specified
+if [ -z ${RANKS} ]; then
+	let RANKS=(${PPN}*${NODES})
+fi
+
+# Support direct SSH when not executed from Fujitsu job system
+if [ -z ${PJM_VNODES} ]; then
+	HOSTFILE="-launcher-exec ssh ${HOSTFILE}"
+fi
+
+export I_MPI_PIN=off
+export PSM2_RCVTHREAD=0
+export HFI_NO_CPUAFFINITY=1
+export I_MPI_COLL_INTRANODE_SHM_THRESHOLD=4194304
+export PSM2_MQ_RNDV_HFI_WINDOW=4194304
+export PSM2_MQ_EAGER_SDMA_SZ=65536
+export PSM2_MQ_RNDV_HFI_THRESH=200000
+
+mpirun ${HOSTFILE} -n ${RANKS} -ppn ${PPN} ${MPI_ENV} ${BINDIR}/mcexec -n ${PPN} ${NUMA} --enable-hfi1 --mpol-threshold=1M --stack-premap=4M,4G --extend-heap-by=8M --disable-sched-yield --mpol-shm-premap ${COMMAND}
+

config.h.in

@@ -3,13 +3,19 @@
 /* Path of install directory for binary */
 #undef BINDIR
 
+/* IHK build-id to confirm IHK and McKernel built at the same time are used */
+#undef BUILDID
+
 /* whether mcoverlayfs is enabled */
 #undef ENABLE_MCOVERLAYFS
 
 /* whether memdump feature is enabled */
 #undef ENABLE_MEMDUMP
 
-/* whether mcoverlayfs is enabled */
+/* whether perf is enabled */
+#undef ENABLE_PERF
+
+/* whether qlmpi is enabled */
 #undef ENABLE_QLMPI
 
 /* whether rusage is enabled */

configure

@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.69 for mckernel 0.9.0.
+# Generated by GNU Autoconf 2.69 for mckernel 1.5.1-knl+hfi.
 #
 #
 # Copyright (C) 1992-1996, 1998-2012 Free Software Foundation, Inc.
@@ -577,8 +577,8 @@ MAKEFLAGS=
 # Identity of this package.
 PACKAGE_NAME='mckernel'
 PACKAGE_TARNAME='mckernel'
-PACKAGE_VERSION='0.9.0'
-PACKAGE_STRING='mckernel 0.9.0'
+PACKAGE_VERSION='1.5.1-knl+hfi'
+PACKAGE_STRING='mckernel 1.5.1-knl+hfi'
 PACKAGE_BUGREPORT=''
 PACKAGE_URL=''
 
@@ -645,6 +645,7 @@ TARGET
 UNAME_R
 KDIR
 ARCH
+BUILDID
 XCC
 FGREP
 EGREP
@@ -708,6 +709,7 @@ enable_dcfa
 enable_memdump
 enable_mcoverlayfs
 enable_rusage
+enable_perf
 enable_qlmpi
 with_uname_r
 '
@@ -1260,7 +1262,7 @@ if test "$ac_init_help" = "long"; then
   # Omit some internal or obsolete options to make the list less imposing.
   # This message is too long to be a string in the A/UX 3.1 sh.
   cat <<_ACEOF
-\`configure' configures mckernel 0.9.0 to adapt to many kinds of systems.
+\`configure' configures mckernel 1.5.1-knl+hfi to adapt to many kinds of systems.
 
 Usage: $0 [OPTION]... [VAR=VALUE]...
 
@@ -1321,7 +1323,7 @@ fi
 
 if test -n "$ac_init_help"; then
   case $ac_init_help in
-     short | recursive ) echo "Configuration of mckernel 0.9.0:";;
+     short | recursive ) echo "Configuration of mckernel 1.5.1-knl+hfi:";;
    esac
   cat <<\_ACEOF
 
@@ -1333,6 +1335,7 @@ Optional Features:
   --enable-memdump        enable dumping memory and analyzing a dump
   --enable-mcoverlayfs    enable mcoverlayfs implementation
   --enable-rusage         enable rusage implementation
+  --enable-perf           enable perf_event implementation
   --enable-qlmpi          enable qlmpi implementation
 
 Optional Packages:
@@ -1428,7 +1431,7 @@ fi
 test -n "$ac_init_help" && exit $ac_status
 if $ac_init_version; then
   cat <<\_ACEOF
-mckernel configure 0.9.0
+mckernel configure 1.5.1-knl+hfi
 generated by GNU Autoconf 2.69
 
 Copyright (C) 2012 Free Software Foundation, Inc.
@@ -1726,7 +1729,7 @@ cat >config.log <<_ACEOF
 This file contains any messages produced by compilers while
 running configure, to aid debugging if configure makes a mistake.
 
-It was created by mckernel $as_me 0.9.0, which was
+It was created by mckernel $as_me 1.5.1-knl+hfi, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   $ $0 $@
@@ -2079,12 +2082,12 @@ ac_compiler_gnu=$ac_cv_c_compiler_gnu
 
 
 
-IHK_VERSION=0.9.0
-MCKERNEL_VERSION=0.9.0
-DCFA_VERSION=0.9.0
-IHK_RELEASE_DATE=2013-11-18
-MCKERNEL_RELEASE_DATE=2013-11-18
-DCFA_RELEASE_DATE=2013-11-18
+IHK_VERSION=1.5.1-knl+hfi
+MCKERNEL_VERSION=1.5.1-knl+hfi
+DCFA_VERSION=DCFA_VERSION_m4
+IHK_RELEASE_DATE=2019-05-14
+MCKERNEL_RELEASE_DATE=2019-05-14
+DCFA_RELEASE_DATE=DCFA_RELEASE_DATE_m4
 
 
 
@@ -3568,6 +3571,14 @@ else
 fi
 
 
+# Check whether --enable-perf was given.
+if test "${enable_perf+set}" = set; then :
+  enableval=$enable_perf; ENABLE_PERF=$enableval
+else
+  ENABLE_PERF=yes
+fi
+
+
 # Check whether --enable-qlmpi was given.
 if test "${enable_qlmpi+set}" = set; then :
   enableval=$enable_qlmpi; ENABLE_QLMPI=$enableval
@@ -4275,7 +4286,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     builtin-mic)
@@ -4292,7 +4303,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/attached/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/attached/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     builtin-x86)
@@ -4309,7 +4320,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/attached/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     smp-x86)
@@ -4341,7 +4352,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/smp-x86/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     smp-arm64)
@@ -4366,11 +4377,14 @@ case $WITH_TARGET in
 	if test "X$ETCDIR" = X; then
 		ETCDIR="$prefix/etc"
 	fi
+	if test "X$INCLUDEDIR" = X; then
+		INCLUDEDIR="$prefix/include"
+	fi
 	if test "X$KMODDIR" = X; then
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/smp-arm64/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     *)
@@ -4961,6 +4975,17 @@ else
 $as_echo "$as_me: rusage is disabled" >&6;}
 fi
 
+if test "x$ENABLE_PERF" = "xyes" ; then
+
+$as_echo "#define ENABLE_PERF 1" >>confdefs.h
+
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: perf is enabled" >&5
+$as_echo "$as_me: perf is enabled" >&6;}
+else
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: perf is disabled" >&5
+$as_echo "$as_me: perf is disabled" >&6;}
+fi
+
 if test "x$MCKERNEL_INCDIR" != "x" ; then
 
 cat >>confdefs.h <<_ACEOF
@@ -4988,6 +5013,20 @@ cat >>confdefs.h <<_ACEOF
 _ACEOF
 
 
+ABS_SRCDIR=$( cd $( dirname $0 ); pwd )
+IHK_ABS_SRCDIR=${ABS_SRCDIR}/../ihk
+BUILDID=$( cd $IHK_ABS_SRCDIR; if [ ! -d .git ]; then echo $IHK_VERSION; else bash -c 'git rev-list -1 HEAD | cut -c1-8'; fi )
+{ $as_echo "$as_me:${as_lineno-$LINENO}: BUILDID=$BUILDID" >&5
+$as_echo "$as_me: BUILDID=$BUILDID" >&6;}
+if test "x$BUILDID" != "x" ; then
+
+cat >>confdefs.h <<_ACEOF
+#define BUILDID "$BUILDID"
+_ACEOF
+
+fi
+
+
 
 
 
@@ -5021,9 +5060,14 @@ ac_config_headers="$ac_config_headers config.h"
 
 # POSTK_DEBUG_ARCH_DEP_37
 # AC_CONFIG_FILES arch dependfiles separate
-ac_config_files="$ac_config_files Makefile executer/user/Makefile executer/user/mcexec.1:executer/user/mcexec.1in executer/user/vmcore2mckdump executer/user/arch/$ARCH/Makefile executer/user/arch/x86_64/Makefile executer/kernel/mcctrl/Makefile executer/kernel/mcctrl/arch/$ARCH/Makefile executer/kernel/mcoverlayfs/Makefile executer/kernel/mcoverlayfs/linux-3.10.0-327.36.1.el7/Makefile executer/kernel/mcoverlayfs/linux-4.0.9/Makefile executer/kernel/mcoverlayfs/linux-4.6.7/Makefile executer/include/qlmpilib.h kernel/Makefile kernel/Makefile.build kernel/include/swapfmt.h arch/x86/tools/mcreboot-attached-mic.sh arch/x86/tools/mcshutdown-attached-mic.sh arch/x86/tools/mcreboot-builtin-x86.sh arch/x86/tools/mcreboot-smp-x86.sh arch/x86/tools/mcstop+release-smp-x86.sh arch/x86/tools/mcoverlay-destroy-smp-x86.sh arch/x86/tools/mcoverlay-create-smp-x86.sh arch/x86/tools/eclair-dump-backtrace.exp arch/x86/tools/mcshutdown-builtin-x86.sh arch/x86/tools/mcreboot.1:arch/x86/tools/mcreboot.1in arch/x86/tools/irqbalance_mck.service arch/x86/tools/irqbalance_mck.in"
+ac_config_files="$ac_config_files Makefile executer/user/Makefile executer/user/mcexec.1:executer/user/mcexec.1in executer/user/vmcore2mckdump executer/user/arch/$ARCH/Makefile executer/user/arch/x86_64/Makefile executer/kernel/mcctrl/Makefile executer/kernel/mcctrl/arch/$ARCH/Makefile executer/kernel/mcoverlayfs/Makefile executer/kernel/mcoverlayfs/linux-3.10.0-327.36.1.el7/Makefile executer/kernel/mcoverlayfs/linux-4.0.9/Makefile executer/kernel/mcoverlayfs/linux-4.6.7/Makefile executer/include/qlmpilib.h kernel/Makefile kernel/Makefile.build kernel/include/swapfmt.h arch/x86_64/tools/mcreboot-attached-mic.sh arch/x86_64/tools/mcshutdown-attached-mic.sh arch/x86_64/tools/mcreboot-builtin-x86.sh arch/x86_64/tools/mcreboot-smp-x86.sh arch/x86_64/tools/mcstop+release-smp-x86.sh arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh arch/x86_64/tools/mcoverlay-create-smp-x86.sh arch/x86_64/tools/eclair-dump-backtrace.exp arch/x86_64/tools/mcshutdown-builtin-x86.sh arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in arch/x86_64/tools/mpimcexec arch/x86_64/tools/mpimcexec.1:arch/x86_64/tools/mpimcexec.1in arch/x86_64/tools/irqbalance_mck.service arch/x86_64/tools/irqbalance_mck.in tools/mcstat/Makefile"
 
 
+if test "$TARGET" = "smp-x86"; then
+ac_config_files="$ac_config_files arch/x86_64/kernel/Makefile.arch"
+
+fi
+
 if test "$TARGET" = "smp-arm64"; then
 ac_config_files="$ac_config_files kernel/config/config.smp-arm64 arch/arm64/kernel/vdso/Makefile arch/arm64/kernel/Makefile.arch"
 
@@ -5541,7 +5585,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 # report actual input values of CONFIG_FILES etc. instead of their
 # values after options handling.
 ac_log="
-This file was extended by mckernel $as_me 0.9.0, which was
+This file was extended by mckernel $as_me 1.5.1-knl+hfi, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   CONFIG_FILES    = $CONFIG_FILES
@@ -5603,7 +5647,7 @@ _ACEOF
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
 ac_cs_version="\\
-mckernel config.status 0.9.0
+mckernel config.status 1.5.1-knl+hfi
 configured by $0, generated by GNU Autoconf 2.69,
   with options \\"\$ac_cs_config\\"
 
@@ -5741,18 +5785,22 @@ do
     "kernel/Makefile") CONFIG_FILES="$CONFIG_FILES kernel/Makefile" ;;
     "kernel/Makefile.build") CONFIG_FILES="$CONFIG_FILES kernel/Makefile.build" ;;
     "kernel/include/swapfmt.h") CONFIG_FILES="$CONFIG_FILES kernel/include/swapfmt.h" ;;
-    "arch/x86/tools/mcreboot-attached-mic.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcreboot-attached-mic.sh" ;;
-    "arch/x86/tools/mcshutdown-attached-mic.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcshutdown-attached-mic.sh" ;;
-    "arch/x86/tools/mcreboot-builtin-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcreboot-builtin-x86.sh" ;;
-    "arch/x86/tools/mcreboot-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcreboot-smp-x86.sh" ;;
-    "arch/x86/tools/mcstop+release-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcstop+release-smp-x86.sh" ;;
-    "arch/x86/tools/mcoverlay-destroy-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcoverlay-destroy-smp-x86.sh" ;;
-    "arch/x86/tools/mcoverlay-create-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcoverlay-create-smp-x86.sh" ;;
-    "arch/x86/tools/eclair-dump-backtrace.exp") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/eclair-dump-backtrace.exp" ;;
-    "arch/x86/tools/mcshutdown-builtin-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcshutdown-builtin-x86.sh" ;;
-    "arch/x86/tools/mcreboot.1") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/mcreboot.1:arch/x86/tools/mcreboot.1in" ;;
-    "arch/x86/tools/irqbalance_mck.service") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/irqbalance_mck.service" ;;
-    "arch/x86/tools/irqbalance_mck.in") CONFIG_FILES="$CONFIG_FILES arch/x86/tools/irqbalance_mck.in" ;;
+    "arch/x86_64/tools/mcreboot-attached-mic.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcreboot-attached-mic.sh" ;;
+    "arch/x86_64/tools/mcshutdown-attached-mic.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcshutdown-attached-mic.sh" ;;
+    "arch/x86_64/tools/mcreboot-builtin-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcreboot-builtin-x86.sh" ;;
+    "arch/x86_64/tools/mcreboot-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcreboot-smp-x86.sh" ;;
+    "arch/x86_64/tools/mcstop+release-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcstop+release-smp-x86.sh" ;;
+    "arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh" ;;
+    "arch/x86_64/tools/mcoverlay-create-smp-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcoverlay-create-smp-x86.sh" ;;
+    "arch/x86_64/tools/eclair-dump-backtrace.exp") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/eclair-dump-backtrace.exp" ;;
+    "arch/x86_64/tools/mcshutdown-builtin-x86.sh") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcshutdown-builtin-x86.sh" ;;
+    "arch/x86_64/tools/mcreboot.1") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in" ;;
+    "arch/x86_64/tools/mpimcexec") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mpimcexec" ;;
+    "arch/x86_64/tools/mpimcexec.1") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/mpimcexec.1:arch/x86_64/tools/mpimcexec.1in" ;;
+    "arch/x86_64/tools/irqbalance_mck.service") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/irqbalance_mck.service" ;;
+    "arch/x86_64/tools/irqbalance_mck.in") CONFIG_FILES="$CONFIG_FILES arch/x86_64/tools/irqbalance_mck.in" ;;
+    "tools/mcstat/Makefile") CONFIG_FILES="$CONFIG_FILES tools/mcstat/Makefile" ;;
+    "arch/x86_64/kernel/Makefile.arch") CONFIG_FILES="$CONFIG_FILES arch/x86_64/kernel/Makefile.arch" ;;
     "kernel/config/config.smp-arm64") CONFIG_FILES="$CONFIG_FILES kernel/config/config.smp-arm64" ;;
     "arch/arm64/kernel/vdso/Makefile") CONFIG_FILES="$CONFIG_FILES arch/arm64/kernel/vdso/Makefile" ;;
     "arch/arm64/kernel/Makefile.arch") CONFIG_FILES="$CONFIG_FILES arch/arm64/kernel/Makefile.arch" ;;

configure.ac

@@ -1,11 +1,9 @@
 # configure.ac COPYRIGHT FUJITSU LIMITED 2015-2016
 AC_PREREQ(2.63)
-m4_define([IHK_VERSION_m4],[0.9.0])dnl
-m4_define([MCKERNEL_VERSION_m4],[0.9.0])dnl
-m4_define([DCFA_VERSION_m4],[0.9.0])dnl
-m4_define([IHK_RELEASE_DATE_m4],[2013-11-18])dnl
-m4_define([MCKERNEL_RELEASE_DATE_m4],[2013-11-18])dnl
-m4_define([DCFA_RELEASE_DATE_m4],[2013-11-18])dnl
+m4_define([IHK_VERSION_m4],[1.5.1-knl+hfi])dnl
+m4_define([MCKERNEL_VERSION_m4],[1.5.1-knl+hfi])dnl
+m4_define([IHK_RELEASE_DATE_m4],[2019-05-14])dnl
+m4_define([MCKERNEL_RELEASE_DATE_m4],[2019-05-14])dnl
 
 AC_INIT([mckernel], MCKERNEL_VERSION_m4)
 
@@ -134,6 +132,12 @@ AC_ARG_ENABLE([rusage],
 	      [ENABLE_RUSAGE=$enableval],
 	      [ENABLE_RUSAGE=yes])
 
+AC_ARG_ENABLE([perf],
+	      AC_HELP_STRING([--enable-perf],
+			     [enable perf_event implementation]),
+	      [ENABLE_PERF=$enableval],
+	      [ENABLE_PERF=yes])
+
 AC_ARG_ENABLE([qlmpi],
 	      AC_HELP_STRING([--enable-qlmpi],
 			     [enable qlmpi implementation]),
@@ -225,7 +229,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     builtin-mic)
@@ -242,7 +246,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/attached/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/attached/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     builtin-x86)
@@ -259,7 +263,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/attached/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     smp-x86)
@@ -291,7 +295,7 @@ case $WITH_TARGET in
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/smp-x86/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     smp-arm64)
@@ -316,11 +320,14 @@ case $WITH_TARGET in
 	if test "X$ETCDIR" = X; then
 		ETCDIR="$prefix/etc"
 	fi
+	if test "X$INCLUDEDIR" = X; then
+		INCLUDEDIR="$prefix/include"
+	fi
 	if test "X$KMODDIR" = X; then
 		KMODDIR="$prefix/kmod"
 	fi
 	if test "X$MANDIR" = X; then
-		MANDIR="$prefix/smp-arm64/man"
+		MANDIR="$prefix/share/man"
 	fi
 	;;
     *)
@@ -451,7 +458,7 @@ else
 fi
 
 if test "x$ENABLE_QLMPI" = "xyes" ; then
-	AC_DEFINE([ENABLE_QLMPI],[1],[whether mcoverlayfs is enabled])
+	AC_DEFINE([ENABLE_QLMPI],[1],[whether qlmpi is enabled])
 	AC_MSG_NOTICE([qlmpi is enabled])
 else
 	AC_MSG_NOTICE([qlmpi is disabled])
@@ -475,6 +482,13 @@ else
 	AC_MSG_NOTICE([rusage is disabled])
 fi
 
+if test "x$ENABLE_PERF" = "xyes" ; then
+	AC_DEFINE([ENABLE_PERF],[1],[whether perf is enabled])
+	AC_MSG_NOTICE([perf is enabled])
+else
+	AC_MSG_NOTICE([perf is disabled])
+fi
+
 if test "x$MCKERNEL_INCDIR" != "x" ; then
    AC_DEFINE_UNQUOTED(MCKERNEL_INCDIR,"$MCKERNEL_INCDIR",[McKernel specific headers])
 fi
@@ -486,6 +500,15 @@ fi
 AC_DEFINE_UNQUOTED(BINDIR,"$BINDIR",[Path of install directory for binary])
 AC_DEFINE_UNQUOTED(SBINDIR,"$SBINDIR",[Path of install directory for system binary])
 
+ABS_SRCDIR=$( cd $( dirname $0 ); pwd )
+IHK_ABS_SRCDIR=${ABS_SRCDIR}/../ihk
+BUILDID=$( cd $IHK_ABS_SRCDIR; if @<:@ ! -d .git @:>@; then echo $IHK_VERSION; else bash -c 'git rev-list -1 HEAD | cut -c1-8'; fi )
+AC_MSG_NOTICE([BUILDID=$BUILDID])
+if test "x$BUILDID" != "x" ; then
+   AC_DEFINE_UNQUOTED(BUILDID,"$BUILDID",[IHK build-id to confirm IHK and McKernel built at the same time are used])
+fi
+AC_SUBST(BUILDID)
+
 AC_SUBST(CC)
 AC_SUBST(XCC)
 AC_SUBST(ARCH)
@@ -535,20 +558,29 @@ AC_CONFIG_FILES([
 	kernel/Makefile
 	kernel/Makefile.build
 	kernel/include/swapfmt.h
-	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/mcoverlay-destroy-smp-x86.sh
-	arch/x86/tools/mcoverlay-create-smp-x86.sh
-	arch/x86/tools/eclair-dump-backtrace.exp
-	arch/x86/tools/mcshutdown-builtin-x86.sh
-	arch/x86/tools/mcreboot.1:arch/x86/tools/mcreboot.1in
-	arch/x86/tools/irqbalance_mck.service
-	arch/x86/tools/irqbalance_mck.in
+	arch/x86_64/tools/mcreboot-attached-mic.sh
+	arch/x86_64/tools/mcshutdown-attached-mic.sh
+	arch/x86_64/tools/mcreboot-builtin-x86.sh
+	arch/x86_64/tools/mcreboot-smp-x86.sh
+	arch/x86_64/tools/mcstop+release-smp-x86.sh
+	arch/x86_64/tools/mcoverlay-destroy-smp-x86.sh
+	arch/x86_64/tools/mcoverlay-create-smp-x86.sh
+	arch/x86_64/tools/eclair-dump-backtrace.exp
+	arch/x86_64/tools/mcshutdown-builtin-x86.sh
+	arch/x86_64/tools/mcreboot.1:arch/x86_64/tools/mcreboot.1in
+	arch/x86_64/tools/mpimcexec
+	arch/x86_64/tools/mpimcexec.1:arch/x86_64/tools/mpimcexec.1in
+	arch/x86_64/tools/irqbalance_mck.service
+	arch/x86_64/tools/irqbalance_mck.in
+	tools/mcstat/Makefile
 ])
 
+if test "$TARGET" = "smp-x86"; then
+AC_CONFIG_FILES([
+	arch/x86_64/kernel/Makefile.arch
+])
+fi
+
 if test "$TARGET" = "smp-arm64"; then
 AC_CONFIG_FILES([
 	kernel/config/config.smp-arm64

executer/include/ihklib_rusage.h

@@ -5,6 +5,10 @@
 #define IHK_MAX_NUM_NUMA_NODES 1024
 #define IHK_MAX_NUM_CPUS 1024
 
+#define IHK_OS_PGSIZE_4KB 0
+#define IHK_OS_PGSIZE_2MB 1
+#define IHK_OS_PGSIZE_1GB 2
+
 struct mckernel_rusage {
 	unsigned long memory_stat_rss[IHK_MAX_NUM_PGSIZES];
 	unsigned long memory_stat_mapped_file[IHK_MAX_NUM_PGSIZES];

executer/include/uprotocol.h

@@ -91,6 +91,7 @@ struct program_image_section {
 
 struct get_cpu_set_arg {
 	int nr_processes;
+	int *process_rank;
 	void *cpu_set;
 	size_t cpu_set_size;	// Size in bytes
 	int *target_core;
@@ -109,6 +110,8 @@ typedef unsigned long __cpu_set_unit;
 #define MPOL_NO_BSS               0x04
 #define MPOL_SHM_PREMAP           0x08
 
+#define MCEXEC_HFI1               0x01
+
 struct program_load_desc {
 	int num_sections;
 	int status;
@@ -137,10 +140,14 @@ struct program_load_desc {
 	unsigned long envs_len;
 	struct rlimit rlimit[MCK_RLIM_MAX];
 	unsigned long interp_align;
+	unsigned long mcexec_flags;
 	unsigned long mpol_flags;
 	unsigned long mpol_threshold;
 	unsigned long heap_extension;
+	long stack_premap;
+	unsigned long mpol_bind_mask;
 	int nr_processes;
+	int process_rank;
 	char shell_path[SHELL_PATH_MAX_LEN];
 	__cpu_set_unit cpu_set[PLD_CPU_SET_SIZE];
 	int profile;
@@ -187,6 +194,7 @@ struct syscall_response {
 	long ret;
 	unsigned long fault_address;
 	unsigned long fault_reason;
+	void *private_data;
 };
 
 struct syscall_ret_desc {

executer/kernel/mcctrl/Makefile.in

@@ -9,13 +9,22 @@ IHK_BASE=$(src)/../../../../ihk
 
 obj-m += mcctrl.o
 
-# POSTK_DEBUG_ARCH_DEP_1, arch depend "-mcmodel"
 # POSTK_DEBUG_ARCH_DEP_83, arch depend translate_rva_to_rpa() move
-ifeq ($(ARCH), arm64)
-ccflags-y := -I$(IHK_BASE)/linux/include -I$(IHK_BASE)/linux/include/ihk/arch/$(ARCH) -I$(IHK_BASE)/ikc/include -I$(IHK_BASE)/ikc/include/ikc/arch/$(ARCH) -I$(IHK_BASE)/include -I$(IHK_BASE)/include/arch/$(ARCH) -I$(src)/../../include -I$(src)/arch/$(ARCH)/include -DMCEXEC_PATH=\"$(BINDIR)/mcexec\" -I@abs_builddir@
-else
-ccflags-y := -I$(IHK_BASE)/linux/include -I$(IHK_BASE)/linux/include/ihk/arch/$(ARCH) -I$(IHK_BASE)/ikc/include -I$(IHK_BASE)/ikc/include/ikc/arch/$(ARCH) -I$(IHK_BASE)/include -I$(IHK_BASE)/include/arch/$(ARCH) -I$(src)/../../../kernel/include -I$(src)/../../include -mcmodel=kernel -mno-red-zone -DMCEXEC_PATH=\"$(BINDIR)/mcexec\" -I@abs_builddir@ -I@abs_builddir@/../../../
-endif
+ccflags-y := -I$(IHK_BASE)/linux/include \
+             -I$(IHK_BASE)/linux/include/ihk/arch/$(ARCH) \
+             -I$(IHK_BASE)/ikc/include \
+             -I$(IHK_BASE)/ikc/include/ikc/arch/$(ARCH) \
+             -I$(IHK_BASE)/include \
+             -I$(IHK_BASE)/include/arch/$(ARCH) \
+             -I$(src)/../../include \
+             -I$(src)/arch/$(ARCH)/include \
+             -I@abs_builddir@ \
+             -I@abs_builddir@/../../../ \
+             -I$(src)/../../../kernel/include \
+             -DMCEXEC_PATH=\"$(BINDIR)/mcexec\"
+
+# depending arch
+include @abs_builddir@/arch/$(ARCH)/Makefile
 
 mcctrl-y := driver.o control.o ikc.o syscall.o procfs.o binfmt_mcexec.o
 mcctrl-y += sysfs.o sysfs_files.o arch/$(ARCH)/archdeps.o

executer/kernel/mcctrl/arch/x86_64/Makefile.in

@@ -1 +1 @@
-# dummy file
+ccflags-y += -mno-red-zone -mcmodel=kernel

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

@@ -327,6 +327,14 @@ int translate_rva_to_rpa(ihk_os_t os, unsigned long rpt, unsigned long rva,
 			pgsize = 1UL << offsh;
 			rpa = pt[ix] & ((1UL << 52) - 1) & ~(pgsize - 1);
 			rpa |= rva & (pgsize - 1);
+
+			/* For GB pages, just report regular 2MB page */
+			if (offsh == 30) {
+				pgsize = 1UL << 21;
+				dprintk("%s: GB page translated 0x%lx -> 0x%lx, pgsize: %lu\n",
+						__FUNCTION__, rva, rpa, pgsize);
+			}
+
 			ihk_device_unmap_virtual(ihk_os_to_dev(os), pt, PAGE_SIZE);
 			ihk_device_unmap_memory(ihk_os_to_dev(os), phys, PAGE_SIZE);
 			error = 0;

executer/kernel/mcctrl/control.c

@@ -190,7 +190,11 @@ static long mcexec_prepare_image(ihk_os_t os,
 	
 	pdesc->status = 0;
 	mb();
-	mcctrl_ikc_send(os, pdesc->cpu, &isp);
+	ret = mcctrl_ikc_send(os, pdesc->cpu, &isp);
+	if(ret < 0) {
+		printk("%s: ERROR mcctrl_ikc_send: %d\n", __FUNCTION__, ret);
+		goto put_and_free_out;
+	}
 
 	ret = wait_event_interruptible(ppd->wq_prepare, pdesc->status);
 	if (ret < 0) {
@@ -363,7 +367,7 @@ static long mcexec_debug_log(ihk_os_t os, unsigned long arg)
 }
 
 int mcexec_close_exec(ihk_os_t os);
-int mcexec_destroy_per_process_data(ihk_os_t os);
+int mcexec_destroy_per_process_data(ihk_os_t os, int pid);
 
 static void release_handler(ihk_os_t os, void *param)
 {
@@ -383,7 +387,7 @@ static void release_handler(ihk_os_t os, void *param)
 
 	mcexec_close_exec(os);
 
-	mcexec_destroy_per_process_data(os);
+	mcexec_destroy_per_process_data(os, info->pid);
 
 	memset(&isp, '\0', sizeof isp);
 	isp.msg = SCD_MSG_CLEANUP_PROCESS;
@@ -431,6 +435,7 @@ static long mcexec_start_image(ihk_os_t os,
 	struct mcctrl_channel *c;
 	struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
 	struct mcos_handler_info *info;
+	int ret = 0;
 
 	desc = kmalloc(sizeof(*desc), GFP_KERNEL);
 	if (!desc) {
@@ -441,17 +446,18 @@ static long mcexec_start_image(ihk_os_t os,
 
 	if (copy_from_user(desc, udesc,
 	                   sizeof(struct program_load_desc))) {
-		kfree(desc);
-		return -EFAULT;
+		ret = -EFAULT;
+		goto out;
 	}
 
 	info = new_mcos_handler_info(os, file);
 #ifdef POSTK_DEBUG_TEMP_FIX_64 /* host process is SIGKILLed fix. */
 	if (info == NULL) {
-		kfree(desc);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out;
 	}
 #endif /* POSTK_DEBUG_TEMP_FIX_64 */
+
 	info->pid = desc->pid;
 	info->cpu = desc->cpu;
 	ihk_os_register_release_handler(file, release_handler, info);
@@ -467,10 +473,14 @@ static long mcexec_start_image(ihk_os_t os,
 	isp.ref = desc->cpu;
 	isp.arg = desc->rprocess;
 
-	mcctrl_ikc_send(os, desc->cpu, &isp);
+	ret = mcctrl_ikc_send(os, desc->cpu, &isp);
+	if (ret < 0) {
+		printk("%s: error: sending IKC msg\n", __FUNCTION__);
+	}
 
+out:
 	kfree(desc);
-	return 0;
+	return ret;
 }
 
 static DECLARE_WAIT_QUEUE_HEAD(signalq);
@@ -628,6 +638,7 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 
 	pli->task = current;
 	pli->ready = 0;
+	pli->timeout = 0;
 	init_waitqueue_head(&pli->pli_wq);
 
 	pli_next = NULL;
@@ -681,6 +692,7 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 		wake_up_interruptible(&pli_next->pli_wq);
 		/* Reset process counter */
 		pe->nr_processes_left = pe->nr_processes;
+		pe->process_rank = 0;
 	}
 
 	/* Wait for the rest if not the last or if the last but
@@ -689,11 +701,50 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
 		dprintk("%s: pid: %d, waiting in list\n",
 				__FUNCTION__, task_tgid_vnr(current));
 		mutex_unlock(&pe->lock);
-		ret = wait_event_interruptible(pli->pli_wq, pli->ready);
+		/* Timeout period: 10 secs + (#procs * 0.1sec) */
+		ret = wait_event_interruptible_timeout(pli->pli_wq,
+				pli->ready,
+				msecs_to_jiffies(10000 + req.nr_processes * 100));
 		mutex_lock(&pe->lock);
-		if (ret != 0) {
+
+		/* First timeout task? Wake up everyone else,
+		 * but tell them we timed out */
+		if (ret == 0) {
+			printk("%s: error: pid: %d, timed out, waking everyone\n",
+					__FUNCTION__, task_tgid_vnr(current));
+			while (!list_empty(&pe->pli_list)) {
+				pli_next = list_first_entry(&pe->pli_list,
+						struct process_list_item, list);
+				list_del(&pli_next->list);
+				pli_next->ready = 1;
+				pli_next->timeout = 1;
+				wake_up_interruptible(&pli_next->pli_wq);
+			}
+
+			/* Reset process counter to start state */
+			pe->nr_processes = -1;
+			ret = -ETIMEDOUT;
 			goto put_and_unlock_out;
 		}
+
+		/* Interrupted or woken up by someone else due to time out? */
+		if (ret < 0 || pli->timeout) {
+			if (ret > 0) {
+				printk("%s: error: pid: %d, job startup timed out\n",
+						__FUNCTION__, task_tgid_vnr(current));
+				ret = -ETIMEDOUT;
+			}
+			goto put_and_unlock_out;
+		}
+
+		/* Incorrect wakeup state? */
+		if (!pli->ready) {
+			printk("%s: error: pid: %d, not ready but woken?\n",
+					__FUNCTION__, task_tgid_vnr(current));
+			ret = -EINVAL;
+			goto put_and_unlock_out;
+		}
+
 		dprintk("%s: pid: %d, woken up\n",
 				__FUNCTION__, task_tgid_vnr(current));
 	}
@@ -873,6 +924,15 @@ next_cpu:
 		goto put_and_unlock_out;
 	}
 
+	/* Copy rank */
+	if (copy_to_user(req.process_rank, &pe->process_rank,
+				sizeof(int))) {
+		printk("%s: error copying process rank to user\n",
+				__FUNCTION__);
+		ret = -EINVAL;
+		goto put_and_unlock_out;
+	}
+
 	/* mcexec NUMA to bind to */
 	mcexec_linux_numa = cpu_to_node(mckernel_cpu_2_linux_cpu(udp, cpu));
 	if (copy_to_user(req.mcexec_linux_numa, &mcexec_linux_numa,
@@ -920,6 +980,7 @@ next_cpu:
 	}
 	/* Otherwise wake up next process in list */
 	else {
+		++pe->process_rank;
 		pli_next = list_first_entry(&pe->pli_list,
 			struct process_list_item, list);
 		list_del(&pli_next->list);
@@ -1012,7 +1073,6 @@ out:
 	return ret;
 }
 
-
 /* NOTE: per-process data is refcounted.
  * For every get call the user should call put. */
 struct mcctrl_per_proc_data *mcctrl_get_per_proc_data(
@@ -1142,7 +1202,7 @@ int mcexec_syscall(struct mcctrl_usrdata *ud, struct ikc_scd_packet *packet)
 		return -1;
 	}
 
-	dprintk("%s: (packet_handler) rtid: %d, ttid: %d, sys nr: %d\n",
+	dprintk("%s: (packet_handler) rtid: %d, ttid: %d, sys nr: %lu\n",
 			__FUNCTION__,
 			packet->req.rtid,
 			packet->req.ttid,
@@ -1197,8 +1257,8 @@ retry_alloc:
 
 	wqhln->packet = packet;
 	wqhln->req = 1;
-	ihk_ikc_spinlock_unlock(&ppd->wq_list_lock, flags);
 	wake_up(&wqhln->wq_syscall);
+	ihk_ikc_spinlock_unlock(&ppd->wq_list_lock, flags);
 
 	mcctrl_put_per_proc_data(ppd);
 
@@ -1312,7 +1372,7 @@ retry_alloc:
 	}
 
 	packet->req.valid = 0; /* ack */
-	dprintk("%s: system call: %d, args[0]: %lu, args[1]: %lu, args[2]: %lu, "
+	dprintk("%s: system call: %lu, args[0]: %lu, args[1]: %lu, args[2]: %lu, "
 			"args[3]: %lu, args[4]: %lu, args[5]: %lu\n",
 			__FUNCTION__,
 			packet->req.number,
@@ -1340,7 +1400,6 @@ retry_alloc:
 			goto put_ppd_out;
 		}
 
-#ifdef POSTK_DEBUG_ARCH_DEP_46 /* user area direct access fix. */
 		if (copy_to_user(&req->cpu, &packet->ref, sizeof(req->cpu))) {
 			if (mcctrl_delete_per_thread_data(ppd, current) < 0) {
 				kprintf("%s: error deleting per-thread data\n", __FUNCTION__);
@@ -1348,9 +1407,6 @@ retry_alloc:
 			ret = -EINVAL;
 			goto put_ppd_out;
 		}
-#else /* POSTK_DEBUG_ARCH_DEP_46 */
-		req->cpu = packet->ref;
-#endif /* POSTK_DEBUG_ARCH_DEP_46 */
 
 		ret = 0;
 		goto put_ppd_out;
@@ -1441,7 +1497,7 @@ long mcexec_load_syscall(ihk_os_t os, struct syscall_load_desc *__user arg)
 	rpm = ihk_device_map_virtual(ihk_os_to_dev(os), phys, desc.size, NULL, 0);
 #endif
 
-	dprintk("mcexec_load_syscall: %s (desc.size: %d)\n", rpm, desc.size);
+	dprintk("mcexec_load_syscall: %p (desc.size: %lu)\n", rpm, desc.size);
 
 	if (copy_to_user((void *__user)desc.dest, rpm, desc.size)) {
 		return -EFAULT;
@@ -1676,12 +1732,12 @@ int mcexec_create_per_process_data(ihk_os_t os)
 	return 0;
 }
 
-int mcexec_destroy_per_process_data(ihk_os_t os)
+int mcexec_destroy_per_process_data(ihk_os_t os, int pid)
 {
 	struct mcctrl_usrdata *usrdata = ihk_host_os_get_usrdata(os);
 	struct mcctrl_per_proc_data *ppd = NULL;
 
-	ppd = mcctrl_get_per_proc_data(usrdata, task_tgid_vnr(current));
+	ppd = mcctrl_get_per_proc_data(usrdata, pid);
 
 	if (ppd) {
 		/* One for the reference and one for deallocation.
@@ -2427,7 +2483,9 @@ mcexec_terminate_thread(ihk_os_t os, unsigned long *param, struct file *file)
 	mcctrl_delete_per_thread_data(ppd, tsk);
 	__return_syscall(usrdata->os, packet, param[2], tid);
 	ihk_ikc_release_packet((struct ihk_ikc_free_packet *)packet,
-	                       (usrdata->channels + packet->ref)->c);
+						   (usrdata->ikc2linux[smp_processor_id()] ?
+							usrdata->ikc2linux[smp_processor_id()] :
+							usrdata->ikc2linux[0]));
 err:
 	if(ppd)
 		mcctrl_put_per_proc_data(ppd);

executer/kernel/mcctrl/driver.c

@@ -27,6 +27,7 @@
 #include <linux/miscdevice.h>
 #include <linux/slab.h>
 #include <linux/device.h>
+#include <linux/delay.h>
 #include "mcctrl.h"
 #include <ihk/ihk_host_user.h>
 
@@ -169,6 +170,14 @@ error_cleanup_channels:
 int mcctrl_os_shutdown_notifier(int os_index)
 {
 	if (os[os_index]) {
+		/* Wait for os running */
+		if (ihk_os_wait_for_status(os[os_index], IHK_OS_STATUS_RUNNING, 0, 200) != 0) {
+			printk("IHK: OS does not become RUNNING in shutdown. Force shutdown.\n");
+			/* send nmi to force shutdown */
+			ihk_os_send_nmi(os[os_index], 3);
+			mdelay(200);
+		}
+
 		sysfsm_cleanup(os[os_index]);
 		free_topology_info(os[os_index]);
 		ihk_os_unregister_user_call_handlers(os[os_index], mcctrl_uc + os_index);

executer/kernel/mcctrl/mcctrl.h

@@ -304,6 +304,7 @@ struct node_topology {
 
 struct process_list_item {
 	int ready;
+	int timeout;
 	struct task_struct *task;
 	struct list_head list;
 	wait_queue_head_t pli_wq;
@@ -313,6 +314,7 @@ struct mcctrl_part_exec {
 	struct mutex lock;	
 	int nr_processes;
 	int nr_processes_left;
+	int process_rank;
 	cpumask_t cpus_used;
 	struct list_head pli_list;
 };

executer/kernel/mcctrl/procfs.c

@@ -1019,7 +1019,8 @@ static const struct procfs_entry tid_entry_stuff[] = {
 
 static const struct procfs_entry pid_entry_stuff[] = {
 	PROC_REG("auxv",       S_IRUSR, NULL),
-	PROC_REG("cgroup",     S_IXUSR, NULL),
+	/* Support the case where McKernel process retrieves its job-id under the Fujitsu TCS suite. */
+//	PROC_REG("cgroup",     S_IXUSR, NULL),
 //	PROC_REG("clear_refs", S_IWUSR, NULL),
 	PROC_REG("cmdline",    S_IRUGO, NULL),
 //	PROC_REG("comm",       S_IRUGO|S_IWUSR, NULL),

executer/kernel/mcctrl/syscall.c

@@ -222,6 +222,14 @@ int translate_rva_to_rpa(ihk_os_t os, unsigned long rpt, unsigned long rva,
 			pgsize = 1UL << offsh;
 			rpa = pt[ix] & ((1UL << 52) - 1) & ~(pgsize - 1);
 			rpa |= rva & (pgsize - 1);
+
+			/* For GB pages, just report regular 2MB page */
+			if (offsh == 30) {
+				pgsize = 1UL << 21;
+				dprintk("%s: GB page translated 0x%lx -> 0x%lx, pgsize: %lu\n",
+						__FUNCTION__, rva, rpa, pgsize);
+			}
+
 			ihk_device_unmap_virtual(ihk_os_to_dev(os), pt, PAGE_SIZE);
 			ihk_device_unmap_memory(ihk_os_to_dev(os), phys, PAGE_SIZE);
 			error = 0;
@@ -799,7 +807,7 @@ static int rus_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	packet = (struct ikc_scd_packet *)mcctrl_get_per_thread_data(ppd, current);
 	if (!packet) {
-		error = -ENOENT;
+		ret = VM_FAULT_SIGBUS;
 		printk("%s: no packet registered for TID %d\n",
 				__FUNCTION__, task_pid_vnr(current));
 		goto put_and_out;
@@ -1174,6 +1182,7 @@ struct pager_create_result {
 	int		maxprot;
 	uint32_t flags;
 	size_t size;
+	char path[PATH_MAX];
 };
 
 enum {
@@ -1192,6 +1201,33 @@ enum {
 	MF_END
 };
 
+static int pager_get_path(struct file *file, char *path) {
+	int error = 0;
+	char *pathbuf, *fullpath;
+
+	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
+	if (!pathbuf) {
+		printk("%s: ERROR: allocating path\n", __FUNCTION__);
+		error = -ENOMEM;
+		goto out;
+	}
+
+	fullpath = d_path(&file->f_path, pathbuf, PATH_MAX);
+	if (!IS_ERR(fullpath)) {
+		memcpy(path, fullpath, strlen(fullpath));
+	}
+	else {
+		path[0] = 0;
+	}
+
+out:
+	if (pathbuf) {
+		kfree(pathbuf);
+	}
+	return error;
+}
+
+
 static int pager_req_create(ihk_os_t os, int fd, uintptr_t result_pa)
 {
 	ihk_device_t dev = ihk_os_to_dev(os);
@@ -1286,7 +1322,10 @@ static int pager_req_create(ihk_os_t os, int fd, uintptr_t result_pa)
 							dprintk("%s: filename: %s, premap & zerofill\n",
 									__FUNCTION__, fullpath);
 						}
-						else if (strstr(fullpath, "libmpi") != NULL) {
+						else if (strstr(fullpath, "libmpi") ||
+								strstr(fullpath, "libiomp") ||
+								strstr(fullpath, "libpthread") ||
+								strstr(fullpath, "libc.so")) {
 							mf_flags = MF_PREFETCH;
 							dprintk("%s: filename: %s, prefetch\n",
 									__FUNCTION__, fullpath);
@@ -1325,6 +1364,7 @@ found:
 	phys = ihk_device_map_memory(dev, result_pa, sizeof(*resp));
 	resp = ihk_device_map_virtual(dev, phys, sizeof(*resp), NULL, 0);
 	if (!resp) {
+		ihk_device_unmap_memory(dev, phys, sizeof(*resp));
 		printk("%s: ERROR: invalid response structure address\n",
 			__FUNCTION__);
 		error = -EINVAL;
@@ -1335,10 +1375,18 @@ found:
 	resp->maxprot = maxprot;
 	resp->flags = mf_flags;
 	resp->size = st.size;
+
+	error = pager_get_path(file, resp->path);
+	if (error) {
+		goto out_unmap;
+	}
+
+	error = 0;
+
+out_unmap:
 	ihk_device_unmap_virtual(dev, resp, sizeof(*resp));
 	ihk_device_unmap_memory(dev, phys, sizeof(*resp));
 
-	error = 0;
 out:
 	if (newpager) {
 		kfree(newpager);
@@ -1570,6 +1618,7 @@ struct pager_map_result {
 	uintptr_t	handle;
 	int		maxprot;
 	int8_t		padding[4];
+    char path[PATH_MAX];
 };
 
 static int pager_req_map(ihk_os_t os, int fd, size_t len, off_t off,
@@ -1624,20 +1673,22 @@ static int pager_req_map(ihk_os_t os, int fd, size_t len, off_t off,
 		maxprot |= PROT_EXEC;
 	}
 
-	down_write(&current->mm->mmap_sem);
+	prot_and_flags = MAP_SHARED |
+		(prot_and_flags & (MAP_POPULATE | MAP_LOCKED));
+
 #define	ANY_WHERE 0
 	if (prot_and_flags & MAP_LOCKED) prot_and_flags |= MAP_POPULATE;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(3,5,0)
+	down_write(&current->mm->mmap_sem);
+
 	va = do_mmap_pgoff(file, ANY_WHERE, len, maxprot, 
-			MAP_SHARED | (prot_and_flags & (MAP_POPULATE | MAP_LOCKED)), pgoff);
-#endif	
+			prot_and_flags, pgoff);
 
 	up_write(&current->mm->mmap_sem);
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,5,0)
-	va = vm_mmap(file, ANY_WHERE, len, maxprot, MAP_SHARED | 
-			(prot_and_flags & (MAP_POPULATE | MAP_LOCKED)), pgoff << PAGE_SHIFT);
+#else
+	va = vm_mmap(file, ANY_WHERE, len, maxprot,
+			prot_and_flags, pgoff << PAGE_SHIFT);
 #endif
 
 	if (IS_ERR_VALUE(va)) {
@@ -1657,6 +1708,7 @@ static int pager_req_map(ihk_os_t os, int fd, size_t len, off_t off,
 	phys = ihk_device_map_memory(dev, result_rpa, sizeof(*resp));
 	resp = ihk_device_map_virtual(dev, phys, sizeof(*resp), NULL, 0);
 	if (!resp) {
+		ihk_device_unmap_memory(dev, phys, sizeof(*resp));
 		printk("%s: ERROR: invalid response structure address\n",
 			__FUNCTION__);
 		error = -EINVAL;
@@ -1665,13 +1717,16 @@ static int pager_req_map(ihk_os_t os, int fd, size_t len, off_t off,
 
 	resp->handle = (uintptr_t)pager;
 	resp->maxprot = maxprot;
-	ihk_device_unmap_virtual(dev, resp, sizeof(*resp));
-	ihk_device_unmap_memory(dev, phys, sizeof(*resp));
+
+	error = pager_get_path(file, resp->path);
+	if (error) {
+		goto out_unmap;
+	}
 
 	error = down_interruptible(&ppd->devobj_pager_lock);
 	if (error) {
 		error = -EINTR;
-		goto out;
+		goto out_unmap;
 	}
 
 	list_add_tail(&pager->list, &ppd->devobj_pager_list);
@@ -1680,6 +1735,10 @@ static int pager_req_map(ihk_os_t os, int fd, size_t len, off_t off,
 	pager = 0;
 	error = 0;
 
+out_unmap:
+	ihk_device_unmap_virtual(dev, resp, sizeof(*resp));
+	ihk_device_unmap_memory(dev, phys, sizeof(*resp));
+
 out:
 	if (file) {
 		fput(file);
@@ -1854,6 +1913,7 @@ static int pager_req_unmap(ihk_os_t os, uintptr_t handle)
 	kfree(pager);
 
 out:
+	mcctrl_put_per_proc_data(ppd);
 	return error;
 }
 
@@ -2005,6 +2065,17 @@ void __return_syscall(ihk_os_t os, struct ikc_scd_packet *packet,
 	/* Map response structure and notify offloading thread */
 	res->ret = ret;
 	res->stid = stid;
+	res->private_data = 0;
+
+	/* Special case for open() to return private_data */
+	if (packet->req.number == __NR_open && ret > 0) {
+		struct fd f;
+		f = fdget(ret);
+		if (f.file) {
+			res->private_data = f.file->private_data;
+			fdput(f);
+		}
+	}
 
 	if (__notify_syscall_requester(os, packet, res) < 0) {
 		printk("%s: WARNING: failed to notify PID %d\n",

executer/kernel/mcctrl/sysfs.c

@@ -1207,7 +1207,7 @@ sysfsm_unlink(struct sysfsm_data *sdp, const char *path0, int flags)
 		goto out;
 	}
 
-	if (!flags & SYSFS_UNLINK_KEEP_ANCESTOR) {
+	if (!(flags & SYSFS_UNLINK_KEEP_ANCESTOR)) {
 		cleanup_ancestor(dirp);
 	}
 

executer/kernel/mcoverlayfs/Makefile.in

@@ -16,7 +16,7 @@ ifeq ($(BUILD_MODULE),none)
 BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -ge 262144 -a ${LINUX_VERSION_CODE} -lt 262400 ]; then echo "linux-4.0.9"; else echo "none"; fi)
 endif
 ifeq ($(BUILD_MODULE),none)
-BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -ge 243680 -a ${LINUX_VERSION_CODE} -lt 263936 ]; then echo "linux-4.6.7"; else echo "none"; fi)
+BUILD_MODULE=$(shell if [ ${LINUX_VERSION_CODE} -ge 263680 -a ${LINUX_VERSION_CODE} -lt 263936 ]; then echo "linux-4.6.7"; else echo "none"; fi)
 endif
 endif
 ifeq ($(BUILD_MODULE_TMP),rhel)
@@ -33,7 +33,7 @@ endif
 
 modules:
 ifneq ($(BUILD_MODULE),none)
-	@(cd $(BUILD_MODULE); make modules)
+	+@(cd $(BUILD_MODULE); make modules)
 endif
 
 clean:

executer/user/Makefile.in

@@ -10,16 +10,16 @@ MANDIR=@MANDIR@
 MCKERNEL_INCDIR=@MCKERNEL_INCDIR@
 MCKERNEL_LIBDIR=@MCKERNEL_LIBDIR@
 KDIR ?= @KDIR@
-CFLAGS=-Wall -O -I. -I$(VPATH)/arch/${ARCH}
+ARCH=@ARCH@
+CFLAGS=-Wall -O -I. -I$(VPATH)/arch/${ARCH} -I${IHKDIR} -I@abs_builddir@/../../../ihk/linux/include
 LDFLAGS=@LDFLAGS@
 RPATH=$(shell echo $(LDFLAGS)|awk '{for(i=1;i<=NF;i++){if($$i~/^-L/){w=$$i;sub(/^-L/,"-Wl,-rpath,",w);print w}}}')
 VPATH=@abs_srcdir@
 TARGET=mcexec libsched_yield ldump2mcdump.so
 @uncomment_if_ENABLE_MEMDUMP@TARGET+=eclair
 LIBS=@LIBS@
-ARCH=@ARCH@
 IHKDIR ?= $(VPATH)/../../../ihk/linux/include/
-MCEXEC_LIBS=-lmcexec -lrt -lnuma -pthread
+MCEXEC_LIBS=-lmcexec -lrt -lnuma -pthread -L@abs_builddir@/../../../ihk/linux/user -lihk -Wl,-rpath,$(MCKERNEL_LIBDIR)
 ENABLE_QLMPI=@ENABLE_QLMPI@
 
 ifeq ($(ENABLE_QLMPI),yes)
@@ -40,10 +40,10 @@ mcexec: mcexec.c libmcexec.a
 # POSTK_DEBUG_ARCH_DEP_34, eclair arch depend separate.
 ifeq ($(ARCH), arm64)
 eclair: eclair.c arch/$(ARCH)/arch-eclair.c
-	$(CC) -I.. -I. -I./arch/$(ARCH)/include -I$(VPATH)/.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include -I${IHKDIR} $(CFLAGS) -o $@ $^ $(LIBS)
+	$(CC) -I.. -I. -I./arch/$(ARCH)/include -I$(VPATH)/.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include $(CFLAGS) -o $@ $^ $(LIBS)
 else
-eclair: eclair.c
-	$(CC) $(CFLAGS) -I${IHKDIR} -o $@ $^ $(LIBS)
+eclair: eclair.c arch/$(ARCH)/arch-eclair.c
+	$(CC) -I.. -I$(VPATH) -I$(VPATH)/arch/$(ARCH)/include $(CFLAGS) -o $@ $^ $(LIBS)
 endif
 
 ldump2mcdump.so: ldump2mcdump.c
@@ -53,7 +53,7 @@ libsched_yield: libsched_yield.c
 	$(CC) -shared -fPIC -Wl,-soname,sched_yield.so.1 -o libsched_yield.so.1.0.0 $^ -lc -ldl
 
 libmcexec.a::
-	(cd arch/${ARCH}; make)
+	+(cd arch/${ARCH}; $(MAKE))
 
 libqlmpi.so: qlmpilib.c
 	$(MCC) $(CFLAGS) $(LDFLAGS) -shared -fPIC -o $@ $<
@@ -77,18 +77,19 @@ ql_talker: ql_talker.o
 	$(CC) $^ $(CFLAGS) -o $@
 
 clean::
-	(cd arch/${ARCH}; make clean)
+	(cd arch/${ARCH}; $(MAKE) clean)
 	$(RM) $(TARGET) *.o
 
 .PHONY: all clean install
 
 install::
-	(cd arch/${ARCH}; make install)
+	(cd arch/${ARCH}; $(MAKE) install)
 	mkdir -p -m 755 $(BINDIR)
 	install -m 755 mcexec $(BINDIR)
 	mkdir -p -m 755 $(MCKERNEL_LIBDIR)
 	install -m 755 ldump2mcdump.so $(MCKERNEL_LIBDIR)
 	install -m 755 libsched_yield.so.1.0.0 $(MCKERNEL_LIBDIR)
+	mkdir -p -m 755 $(MANDIR)/man1
 	install -m 644 mcexec.1 $(MANDIR)/man1/mcexec.1
 ifeq ($(ENABLE_QLMPI),yes)
 	install -m 644 ../include/qlmpilib.h $(MCKERNEL_INCDIR)

executer/user/arch/arm64/Makefile.in

@@ -9,12 +9,15 @@ LIBS=@LIBS@
 
 all: $(TARGET)
 
-../../libmcexec.a: archdep.o
-	$(AR) cr ../../libmcexec.a archdep.o 
+../../libmcexec.a: archdep.o arch_syscall.o
+	$(AR) cr ../../libmcexec.a archdep.o arch_syscall.o
 
 archdep.o: archdep.S
 	$(CC) -c -I${KDIR} $(CFLAGS) $(EXTRA_CFLAGS) -fPIE -pie -pthread $<
 
+arch_syscall.o: arch_syscall.c
+	$(CC) -c -I${KDIR} $(CFLAGS) $(EXTRA_CFLAGS) -fPIE -pie -pthread $<
+
 clean:
 	$(RM) $(TARGET) *.o