Merge pull request snabbco#66 from lukego/vmprofile

Port VMProfile sampling profiler from Snabb

src/Makefile

@@ -228,7 +228,7 @@ LJLIB_C= $(LJLIB_O:.o=.c)
 LJCORE_O= lj_gc.o lj_err.o lj_char.o lj_bc.o lj_obj.o lj_buf.o \
 	  lj_str.o lj_tab.o lj_func.o lj_udata.o lj_meta.o lj_debug.o \
 	  lj_state.o lj_dispatch.o lj_vmmath.o lj_strscan.o \
-	  lj_strfmt.o lj_strfmt_num.o lj_api.o \
+	  lj_strfmt.o lj_strfmt_num.o lj_api.o lj_vmprofile.o \
 	  lj_lex.o lj_parse.o lj_bcread.o lj_bcwrite.o lj_load.o \
 	  lj_ir.o lj_opt_mem.o lj_opt_fold.o lj_opt_narrow.o \
 	  lj_opt_dce.o lj_opt_loop.o lj_opt_split.o lj_opt_sink.o \

src/lib_jit.c

@@ -200,6 +200,29 @@ LJLIB_CF(jit_opt_start)
 
 #include "lj_libdef.h"
 
+/* -- jit.vmprofile module  ----------------------------------------------- */
+
+#define LJLIB_MODULE_jit_vmprofile
+
+LJLIB_CF(jit_vmprofile_start)
+{
+  luaJIT_vmprofile_start(L);
+  return 0;
+}
+
+LJLIB_CF(jit_vmprofile_stop)
+{
+  luaJIT_vmprofile_stop(L);
+  return 0;
+}
+
+#include "lj_libdef.h"
+
+static int luaopen_jit_vmprofile(lua_State *L)
+{
+  LJ_LIB_REG(L, NULL, jit_vmprofile);
+  return 1;
+}
 
 /* -- JIT compiler initialization ----------------------------------------- */
 
@@ -259,6 +282,8 @@ LUALIB_API int luaopen_jit(lua_State *L)
   lua_pushinteger(L, LUAJIT_VERSION_NUM);
   lua_pushliteral(L, LUAJIT_VERSION);
   LJ_LIB_REG(L, LUA_JITLIBNAME, jit);
+  lj_lib_prereg(L, LUA_JITLIBNAME ".vmprofile", luaopen_jit_vmprofile,
+                tabref(L->env));
   LJ_LIB_REG(L, "jit.opt", jit_opt);
   L->top -= 2;
   return 1;

src/lj_gc.c

@@ -648,6 +648,7 @@ int lj_gc_step(lua_State *L)
   global_State *g = G(L);
   GCSize lim;
   int32_t ostate = g->vmstate;
+  g->gcvmstate = ostate;
   setvmstate(g, GC);
   lim = (GCSTEPSIZE/100) * g->gc.stepmul;
   if (lim == 0)
@@ -698,6 +699,7 @@ void lj_gc_fullgc(lua_State *L)
 {
   global_State *g = G(L);
   int32_t ostate = g->vmstate;
+  g->gcvmstate = ostate;
   setvmstate(g, GC);
   if (g->gc.state <= GCSatomic) {  /* Caught somewhere in the middle. */
     setmref(g->gc.sweep, &g->gc.root);  /* Sweep everything (preserving it). */

src/lj_obj.h

@@ -516,7 +516,8 @@ typedef struct global_State {
   lua_Alloc allocf;	/* Memory allocator. */
   void *allocd;		/* Memory allocator data. */
   GCState gc;		/* Garbage collector. */
-  volatile int32_t vmstate;  /* VM state or current JIT code trace number. */
+  volatile int32_t vmstate;   /* VM state or current JIT code trace number. */
+  volatile int32_t gcvmstate; /* Previous VM state (only when state is GC). */
   SBuf tmpbuf;		/* Temporary string buffer. */
   GCstr strempty;	/* Empty string. */
   uint8_t stremptyz;	/* Zero terminator of empty string. */

src/lj_vmprofile.c

@@ -0,0 +1,113 @@
+/*
+** VM profiling.
+** Copyright (C) 2016 Luke Gorrie. See Copyright Notice in luajit.h
+*/
+
+#define lj_vmprofile_c
+#define LUA_CORE
+
+#define _GNU_SOURCE 1
+#include <stdio.h>
+#include <assert.h>
+#include <sys/time.h>
+#include <stdint.h>
+#include <signal.h>
+#include <ucontext.h>
+#undef _GNU_SOURCE
+
+#include "lj_obj.h"
+#include "lj_dispatch.h"
+#include "lj_jit.h"
+#include "lj_trace.h"
+#include "lj_vmprofile.h"
+
+static struct {
+  global_State *g;
+  struct sigaction oldsa;
+} state;
+
+/* -- State that the application can manage via FFI ----------------------- */
+
+static VMProfile *profile;      /* Current counters */
+
+/* How much memory to allocate for profiler counters. */
+int vmprofile_get_profile_size() {
+  return sizeof(VMProfile);
+}
+
+/* Set the memory where the next samples will be counted. 
+   Size of the memory must match vmprofile_get_profile_size(). */
+void vmprofile_set_profile(void *counters) {
+  profile = (VMProfile*)counters;
+  profile->magic = 0x1d50f007;
+  profile->major = 2;
+  profile->minor = 0;
+}
+
+/* -- Signal handler ------------------------------------------------------ */
+
+/* Signal handler: bumps one counter. */
+static void vmprofile_signal(int sig, siginfo_t *si, void *data)
+{
+  if (profile != NULL) {
+    lua_State *L = gco2th(gcref(state.g->cur_L));
+    int vmstate = state.g->vmstate;
+    int trace = ~vmstate == LJ_VMST_GC ? state.g->gcvmstate : vmstate;
+    /* Not in a trace */
+    if (trace < 0) {
+      profile->vm[~vmstate]++;
+    } else {
+      int bucket = trace > LJ_VMPROFILE_TRACE_MAX ? 0 : trace;
+      VMProfileTraceCount *count = &profile->trace[bucket];
+      GCtrace *T = traceref(L2J(L), (TraceNo)trace);
+      intptr_t ip = (intptr_t)((ucontext_t*)data)->uc_mcontext.gregs[REG_RIP];
+      ptrdiff_t mcposition = ip - (intptr_t)T->mcode;
+      if (~vmstate == LJ_VMST_GC) {
+        count->gc++;
+      } else if ((mcposition < 0) || (mcposition >= T->szmcode)) {
+        count->other++;
+      } else if ((T->mcloop != 0) && (mcposition >= T->mcloop)) {
+        count->loop++;
+      } else {
+        count->head++;
+      }
+    }
+  }
+}
+
+static void start_timer(int interval)
+{
+  struct itimerval tm;
+  struct sigaction sa;
+  tm.it_value.tv_sec = tm.it_interval.tv_sec = interval / 1000;
+  tm.it_value.tv_usec = tm.it_interval.tv_usec = (interval % 1000) * 1000;
+  setitimer(ITIMER_PROF, &tm, NULL);
+  sa.sa_flags = SA_SIGINFO | SA_RESTART;
+  sa.sa_sigaction = vmprofile_signal;
+  sigemptyset(&sa.sa_mask);
+  sigaction(SIGPROF, &sa, &state.oldsa);
+}
+
+static void stop_timer()
+{
+  struct itimerval tm;
+  tm.it_value.tv_sec = tm.it_interval.tv_sec = 0;
+  tm.it_value.tv_usec = tm.it_interval.tv_usec = 0;
+  setitimer(ITIMER_PROF, &tm, NULL);
+  sigaction(SIGPROF, NULL, &state.oldsa);
+}
+
+/* -- Lua API ------------------------------------------------------------- */
+
+LUA_API void luaJIT_vmprofile_start(lua_State *L)
+{
+  memset(&state, 0, sizeof(state));
+  state.g = G(L);
+  start_timer(1);               /* Sample every 1ms */
+}
+
+LUA_API void luaJIT_vmprofile_stop(lua_State *L)
+{
+  stop_timer();
+}
+

src/lj_vmprofile.h

@@ -0,0 +1,38 @@
+/*
+** Virtual machine profiling.
+** Copyright (C) 2017 Luke Gorrie. See Copyright Notice in luajit.h
+*/
+
+#ifndef _LJ_VMPROFILE_H
+#define _LJ_VMPROFILE_H
+
+
+/* Counters are 64-bit to avoid overflow even in long running processes. */
+typedef uint64_t VMProfileCount;
+
+/* Maximum trace number for distinct counter buckets. Traces with
+   higher numbers will be counted together in bucket zero. */
+#define LJ_VMPROFILE_TRACE_MAX 4096
+
+/* Traces have separate counters for different machine code regions. */
+typedef struct VMProfileTraceCount {
+  VMProfileCount head;          /* Head of the trace (non-looping part) */
+  VMProfileCount loop;          /* Loop of the trace */
+  VMProfileCount other;         /* Outside the trace mcode (unidentified) */
+  VMProfileCount gc;            /* Garbage collection from this trace. */
+} VMProfileTraceCount;
+
+/* Complete set of counters for VM and traces. */
+typedef struct VMProfile {
+  uint32_t magic;               /* 0x1d50f007 */
+  uint16_t major, minor;        /* 2, 0 */
+  VMProfileCount vm[LJ_VMST__MAX];
+  VMProfileTraceCount trace[LJ_VMPROFILE_TRACE_MAX+1];
+} VMProfile;
+
+/* Functions that should be accessed via FFI. */
+
+void vmprofile_set_profile(void *counters);
+int vmprofile_get_profile_size();
+
+#endif

src/luajit.h

@@ -64,6 +64,10 @@ enum {
 /* Control the JIT engine. */
 LUA_API int luaJIT_setmode(lua_State *L, int idx, int mode);
 
+/* VM profiling API. */
+LUA_API void luaJIT_vmprofile_start(lua_State *L);
+LUA_API void luaJIT_vmprofile_stop(lua_State *L);
+
 /* Enforce (dynamic) linker error for version mismatches. Call from main. */
 LUA_API void LUAJIT_VERSION_SYM(void);