[pypy-commit] pypy default: Enable windows support for vmprof
fijal
pypy.commits at gmail.com
Sat Feb 13 02:48:20 EST 2016
Author: fijal
Branch:
Changeset: r82204:9dbad5cf71a6
Date: 2016-02-13 08:47 +0100
http://bitbucket.org/pypy/pypy/changeset/9dbad5cf71a6/
Log: Enable windows support for vmprof
diff --git a/pypy/config/pypyoption.py b/pypy/config/pypyoption.py
--- a/pypy/config/pypyoption.py
+++ b/pypy/config/pypyoption.py
@@ -36,13 +36,13 @@
"cStringIO", "thread", "itertools", "pyexpat", "_ssl", "cpyext", "array",
"binascii", "_multiprocessing", '_warnings', "_collections",
"_multibytecodec", "micronumpy", "_continuation", "_cffi_backend",
- "_csv", "cppyy", "_pypyjson"
+ "_csv", "cppyy", "_pypyjson", "_vmprof",
])
-if ((sys.platform.startswith('linux') or sys.platform == 'darwin')
- and os.uname()[4] == 'x86_64' and sys.maxint > 2**32):
+#if ((sys.platform.startswith('linux') or sys.platform == 'darwin')
+# and os.uname()[4] == 'x86_64' and sys.maxint > 2**32):
# it's not enough that we get x86_64
- working_modules.add('_vmprof')
+# working_modules.add('_vmprof')
translation_modules = default_modules.copy()
translation_modules.update([
diff --git a/rpython/jit/backend/llsupport/test/zrpy_vmprof_test.py b/rpython/jit/backend/llsupport/test/zrpy_vmprof_test.py
--- a/rpython/jit/backend/llsupport/test/zrpy_vmprof_test.py
+++ b/rpython/jit/backend/llsupport/test/zrpy_vmprof_test.py
@@ -3,16 +3,13 @@
from rpython.jit.backend.test.support import CCompiledMixin
from rpython.rlib.jit import JitDriver
from rpython.tool.udir import udir
+from rpython.rlib import rthread
from rpython.translator.translator import TranslationContext
from rpython.jit.backend.detect_cpu import getcpuclass
class CompiledVmprofTest(CCompiledMixin):
CPUClass = getcpuclass()
- def setup(self):
- if self.CPUClass.backend_name != 'x86_64':
- py.test.skip("vmprof only supports x86-64 CPUs at the moment")
-
def _get_TranslationContext(self):
t = TranslationContext()
t.config.translation.gc = 'incminimark'
@@ -62,6 +59,7 @@
tmpfilename = str(udir.join('test_rvmprof'))
def f(num):
+ rthread.get_ident() # register TLOFS_thread_ident
code = MyCode("py:x:foo:3")
rvmprof.register_code(code, get_name)
fd = os.open(tmpfilename, os.O_WRONLY | os.O_CREAT, 0666)
diff --git a/rpython/rlib/rvmprof/cintf.py b/rpython/rlib/rvmprof/cintf.py
--- a/rpython/rlib/rvmprof/cintf.py
+++ b/rpython/rlib/rvmprof/cintf.py
@@ -30,11 +30,11 @@
def setup():
+ compile_extra = ['-DRPYTHON_LL2CTYPES']
platform.verify_eci(ExternalCompilationInfo(
- compile_extra=['-DRPYTHON_LL2CTYPES'],
+ compile_extra=compile_extra,
**eci_kwds))
-
eci = global_eci
vmprof_init = rffi.llexternal("vmprof_init",
[rffi.INT, rffi.DOUBLE, rffi.CCHARP],
diff --git a/rpython/rlib/rvmprof/src/rvmprof.c b/rpython/rlib/rvmprof/src/rvmprof.c
--- a/rpython/rlib/rvmprof/src/rvmprof.c
+++ b/rpython/rlib/rvmprof/src/rvmprof.c
@@ -1,23 +1,21 @@
#define _GNU_SOURCE 1
-
#ifdef RPYTHON_LL2CTYPES
/* only for testing: ll2ctypes sets RPY_EXTERN from the command-line */
-# ifndef RPY_EXTERN
-# define RPY_EXTERN RPY_EXPORTED
-# endif
-# define RPY_EXPORTED extern __attribute__((visibility("default")))
-# define VMPROF_ADDR_OF_TRAMPOLINE(addr) 0
+#ifndef RPY_EXTERN
+#define RPY_EXTERN RPY_EXPORTED
+#endif
+#ifdef _WIN32
+#define RPY_EXPORTED __declspec(dllexport)
+#else
+#define RPY_EXPORTED extern __attribute__((visibility("default")))
+#endif
#else
-
# include "common_header.h"
# include "structdef.h"
# include "src/threadlocal.h"
# include "rvmprof.h"
-/*# ifndef VMPROF_ADDR_OF_TRAMPOLINE
-# error "RPython program using rvmprof, but not calling vmprof_execute_code()"
-# endif*/
#endif
diff --git a/rpython/rlib/rvmprof/src/vmprof_common.h b/rpython/rlib/rvmprof/src/vmprof_common.h
--- a/rpython/rlib/rvmprof/src/vmprof_common.h
+++ b/rpython/rlib/rvmprof/src/vmprof_common.h
@@ -7,9 +7,6 @@
static long profile_interval_usec = 0;
static int opened_profile(char *interp_name);
-#define MAX_STACK_DEPTH \
- ((SINGLE_BUF_SIZE - sizeof(struct prof_stacktrace_s)) / sizeof(void *))
-
#define MARKER_STACKTRACE '\x01'
#define MARKER_VIRTUAL_IP '\x02'
#define MARKER_TRAILER '\x03'
@@ -20,6 +17,9 @@
#define VERSION_THREAD_ID '\x01'
#define VERSION_TAG '\x02'
+#define MAX_STACK_DEPTH \
+ ((SINGLE_BUF_SIZE - sizeof(struct prof_stacktrace_s)) / sizeof(void *))
+
typedef struct prof_stacktrace_s {
char padding[sizeof(long) - 1];
char marker;
@@ -71,6 +71,43 @@
return _write_all((char*)&header, 5 * sizeof(long) + 4 + namelen);
}
+/* *************************************************************
+ * functions to dump the stack trace
+ * *************************************************************
+ */
+
+
+static int get_stack_trace(vmprof_stack_t* stack, intptr_t *result, int max_depth, intptr_t pc)
+{
+ int n = 0;
+ intptr_t addr = 0;
+ int bottom_jitted = 0;
+ // check if the pc is in JIT
+#ifdef PYPY_JIT_CODEMAP
+ if (pypy_find_codemap_at_addr((intptr_t)pc, &addr)) {
+ // the bottom part is jitted, means we can fill up the first part
+ // from the JIT
+ n = vmprof_write_header_for_jit_addr(result, n, pc, max_depth);
+ stack = stack->next; // skip the first item as it contains garbage
+ }
+#endif
+ while (n < max_depth - 1 && stack) {
+ if (stack->kind == VMPROF_CODE_TAG) {
+ result[n] = stack->kind;
+ result[n + 1] = stack->value;
+ n += 2;
+ }
+#ifdef PYPY_JIT_CODEMAP
+ else if (stack->kind == VMPROF_JITTED_TAG) {
+ pc = ((intptr_t*)(stack->value - sizeof(intptr_t)))[0];
+ n = vmprof_write_header_for_jit_addr(result, n, pc, max_depth);
+ }
+#endif
+ stack = stack->next;
+ }
+ return n;
+}
+
#ifndef RPYTHON_LL2CTYPES
static vmprof_stack_t *get_vmprof_stack(void)
{
diff --git a/rpython/rlib/rvmprof/src/vmprof_main.h b/rpython/rlib/rvmprof/src/vmprof_main.h
--- a/rpython/rlib/rvmprof/src/vmprof_main.h
+++ b/rpython/rlib/rvmprof/src/vmprof_main.h
@@ -35,6 +35,7 @@
#include "vmprof_stack.h"
#include "vmprof_getpc.h"
#include "vmprof_mt.h"
+#include "vmprof_get_custom_offset.h"
#include "vmprof_common.h"
/************************************************************/
@@ -78,46 +79,6 @@
static char atfork_hook_installed = 0;
-#include "vmprof_get_custom_offset.h"
-
-/* *************************************************************
- * functions to dump the stack trace
- * *************************************************************
- */
-
-
-static int get_stack_trace(intptr_t *result, int max_depth, intptr_t pc, ucontext_t *ucontext)
-{
- vmprof_stack_t* stack = get_vmprof_stack();
- int n = 0;
- intptr_t addr = 0;
- int bottom_jitted = 0;
- // check if the pc is in JIT
-#ifdef PYPY_JIT_CODEMAP
- if (pypy_find_codemap_at_addr((intptr_t)pc, &addr)) {
- // the bottom part is jitted, means we can fill up the first part
- // from the JIT
- n = vmprof_write_header_for_jit_addr(result, n, pc, max_depth);
- stack = stack->next; // skip the first item as it contains garbage
- }
-#endif
- while (n < max_depth - 1 && stack) {
- if (stack->kind == VMPROF_CODE_TAG) {
- result[n] = stack->kind;
- result[n + 1] = stack->value;
- n += 2;
- }
-#ifdef PYPY_JIT_CODEMAP
- else if (stack->kind == VMPROF_JITTED_TAG) {
- pc = ((intptr_t*)(stack->value - sizeof(intptr_t)))[0];
- n = vmprof_write_header_for_jit_addr(result, n, pc, max_depth);
- }
-#endif
- stack = stack->next;
- }
- return n;
-}
-
static intptr_t get_current_thread_id(void)
{
/* xxx This function is a hack on two fronts:
@@ -194,8 +155,8 @@
struct prof_stacktrace_s *st = (struct prof_stacktrace_s *)p->data;
st->marker = MARKER_STACKTRACE;
st->count = 1;
- depth = get_stack_trace(st->stack,
- MAX_STACK_DEPTH-2, GetPC((ucontext_t*)ucontext), ucontext);
+ depth = get_stack_trace(get_vmprof_stack(), st->stack,
+ MAX_STACK_DEPTH-2, GetPC((ucontext_t*)ucontext));
st->depth = depth;
st->stack[depth++] = get_current_thread_id();
p->data_offset = offsetof(struct prof_stacktrace_s, marker);
diff --git a/rpython/rlib/rvmprof/src/vmprof_main_win32.h b/rpython/rlib/rvmprof/src/vmprof_main_win32.h
--- a/rpython/rlib/rvmprof/src/vmprof_main_win32.h
+++ b/rpython/rlib/rvmprof/src/vmprof_main_win32.h
@@ -10,13 +10,30 @@
return 0;
}
+#if defined(_MSC_VER)
+#include <BaseTsd.h>
+typedef SSIZE_T ssize_t;
+#endif
+
+#include <assert.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <signal.h>
+#include <stddef.h>
+#include <sys/stat.h>
+#include <fcntl.h>
#include "vmprof_stack.h"
+#include "vmprof_get_custom_offset.h"
#include "vmprof_common.h"
#include <tlhelp32.h>
// This file has been inspired (but not copied from since the LICENSE
// would not allow it) from verysleepy profiler
+#define SINGLE_BUF_SIZE 8192
+
volatile int thread_started = 0;
volatile int enabled = 0;
@@ -55,52 +72,75 @@
return 0;
}
-int vmprof_snapshot_thread(DWORD thread_id, PyThreadState *tstate, prof_stacktrace_s *stack)
+int vmprof_snapshot_thread(struct pypy_threadlocal_s *p, prof_stacktrace_s *stack)
{
- HRESULT result;
- HANDLE hThread = OpenThread(THREAD_ALL_ACCESS, FALSE, thread_id);
- int depth;
+ void *addr;
+ vmprof_stack_t *cur;
+ long tid;
+ HANDLE hThread;
+ long depth;
+ DWORD result;
+ CONTEXT ctx;
+
+#ifdef RPYTHON_LL2CTYPES
+ return 0; // not much we can do
+#else
+#ifndef RPY_TLOFS_thread_ident
+ return 0; // we can't freeze threads, unsafe
+#else
+ hThread = OpenThread(THREAD_ALL_ACCESS, FALSE, p->thread_ident);
if (!hThread) {
return -1;
}
result = SuspendThread(hThread);
if(result == 0xffffffff)
return -1; // possible, e.g. attached debugger or thread alread suspended
- // find the correct thread
- depth = read_trace_from_cpy_frame(tstate->frame, stack->stack,
- MAX_STACK_DEPTH);
+ ctx.ContextFlags = CONTEXT_FULL;
+ if (!GetThreadContext(hThread, &ctx))
+ return -1;
+ depth = get_stack_trace(p->vmprof_tl_stack,
+ stack->stack, MAX_STACK_DEPTH-2, ctx.Eip);
stack->depth = depth;
- stack->stack[depth++] = (void*)thread_id;
+ stack->stack[depth++] = (void*)p->thread_ident;
stack->count = 1;
stack->marker = MARKER_STACKTRACE;
ResumeThread(hThread);
return depth;
+#endif
+#endif
}
long __stdcall vmprof_mainloop(void *arg)
{
+#ifndef RPYTHON_LL2CTYPES
+ struct pypy_threadlocal_s *p;
prof_stacktrace_s *stack = (prof_stacktrace_s*)malloc(SINGLE_BUF_SIZE);
- HANDLE hThreadSnap = INVALID_HANDLE_VALUE;
int depth;
- PyThreadState *tstate;
while (1) {
- Sleep(profile_interval_usec * 1000);
+ //Sleep(profile_interval_usec * 1000);
+ Sleep(10);
if (!enabled) {
continue;
}
- tstate = PyInterpreterState_Head()->tstate_head;
- while (tstate) {
- depth = vmprof_snapshot_thread(tstate->thread_id, tstate, stack);
- if (depth > 0) {
- _write_all((char*)stack + offsetof(prof_stacktrace_s, marker),
- depth * sizeof(void *) +
- sizeof(struct prof_stacktrace_s) -
- offsetof(struct prof_stacktrace_s, marker));
+ _RPython_ThreadLocals_Acquire();
+ p = _RPython_ThreadLocals_Head(); // the first one is one behind head
+ p = _RPython_ThreadLocals_Enum(p);
+ while (p) {
+ if (p->ready == 42) {
+ depth = vmprof_snapshot_thread(p, stack);
+ if (depth > 0) {
+ _write_all((char*)stack + offsetof(prof_stacktrace_s, marker),
+ depth * sizeof(void *) +
+ sizeof(struct prof_stacktrace_s) -
+ offsetof(struct prof_stacktrace_s, marker));
+ }
}
- tstate = tstate->next;
+ p = _RPython_ThreadLocals_Enum(p);
}
+ _RPython_ThreadLocals_Release();
}
+#endif
}
RPY_EXTERN
diff --git a/rpython/rlib/rvmprof/src/vmprof_stack.h b/rpython/rlib/rvmprof/src/vmprof_stack.h
--- a/rpython/rlib/rvmprof/src/vmprof_stack.h
+++ b/rpython/rlib/rvmprof/src/vmprof_stack.h
@@ -1,7 +1,11 @@
#ifndef _VMPROF_STACK_H_
#define _VMPROF_STACK_H_
+#ifdef _WIN32
+#define intptr_t long // XXX windows VC++ 2008 lacks stdint.h
+#else
#include <unistd.h>
+#endif
#define VMPROF_CODE_TAG 1 /* <- also in cintf.py */
#define VMPROF_BLACKHOLE_TAG 2
diff --git a/rpython/rlib/rvmprof/test/test_ztranslation.py b/rpython/rlib/rvmprof/test/test_ztranslation.py
--- a/rpython/rlib/rvmprof/test/test_ztranslation.py
+++ b/rpython/rlib/rvmprof/test/test_ztranslation.py
@@ -3,11 +3,10 @@
sys.path += ['../../../..'] # for subprocess in test_interpreted
import py
from rpython.tool.udir import udir
-from rpython.rlib import rvmprof
+from rpython.rlib import rvmprof, rthread
from rpython.translator.c.test.test_genc import compile
from rpython.rlib.nonconst import NonConstant
-
class MyCode:
def __init__(self, count):
self.count = count
@@ -39,6 +38,7 @@
PROF_FILE = str(udir.join('test_ztranslation.prof'))
def main(argv=[]):
+ rthread.get_ident() # force TLOFS_thread_ident
if NonConstant(False):
# Hack to give os.open() the correct annotation
os.open('foo', 1, 1)
diff --git a/rpython/translator/c/src/threadlocal.c b/rpython/translator/c/src/threadlocal.c
--- a/rpython/translator/c/src/threadlocal.c
+++ b/rpython/translator/c/src/threadlocal.c
@@ -85,6 +85,11 @@
return prev->next;
}
+struct pypy_threadlocal_s *_RPython_ThreadLocals_Head(void)
+{
+ return &linkedlist_head;
+}
+
static void _RPy_ThreadLocals_Init(void *p)
{
struct pypy_threadlocal_s *tls = (struct pypy_threadlocal_s *)p;
diff --git a/rpython/translator/c/src/threadlocal.h b/rpython/translator/c/src/threadlocal.h
--- a/rpython/translator/c/src/threadlocal.h
+++ b/rpython/translator/c/src/threadlocal.h
@@ -27,6 +27,9 @@
RPY_EXTERN struct pypy_threadlocal_s *
_RPython_ThreadLocals_Enum(struct pypy_threadlocal_s *prev);
+/* will return the head of the list */
+RPY_EXTERN struct pypy_threadlocal_s *_RPython_ThreadLocals_Head();
+
#define OP_THREADLOCALREF_ACQUIRE(r) _RPython_ThreadLocals_Acquire()
#define OP_THREADLOCALREF_RELEASE(r) _RPython_ThreadLocals_Release()
#define OP_THREADLOCALREF_ENUM(p, r) r = _RPython_ThreadLocals_Enum(p)
More information about the pypy-commit
mailing list