* [Qemu-devel] [PATCH 0/2] Fast Thread-Local Storage support
@ 2013-07-01 9:35 Stefan Hajnoczi
2013-07-01 9:35 ` [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h Stefan Hajnoczi
2013-07-01 9:35 ` [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers Stefan Hajnoczi
0 siblings, 2 replies; 17+ messages in thread
From: Stefan Hajnoczi @ 2013-07-01 9:35 UTC (permalink / raw)
To: qemu-devel
Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, Stefan Hajnoczi, Peter Maydell
These patches by Paolo Bonzini <pbonzini@redhat.com> add thread-local storage
support for POSIX and Windows platforms. Fast native TLS mechanisms are used
when available and pthread_get/setspecific() is used as a fallback.
My dataplane block layer RFC work relies on this and it has been discussed
separately by Ed and Peter, so I wanted to share my latest edited version of
these patches.
The only change I made is to prepend "tls_" to the function names.
Paolo: Posting so you can take this back into your tree if you wish.
Paolo Bonzini (2):
exec: do not use qemu/tls.h
qemu-thread: add TLS wrappers
configure | 21 ++++++++
exec.c | 10 +++-
include/exec/cpu-all.h | 14 ++++--
include/qemu/tls.h | 125 ++++++++++++++++++++++++++++++++++++++++-------
tests/Makefile | 3 ++
tests/test-tls.c | 87 +++++++++++++++++++++++++++++++++
util/qemu-thread-win32.c | 17 +++++++
7 files changed, 253 insertions(+), 24 deletions(-)
create mode 100644 tests/test-tls.c
--
1.8.1.4
^ permalink raw reply [flat|nested] 17+ messages in thread
* [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h
2013-07-01 9:35 [Qemu-devel] [PATCH 0/2] Fast Thread-Local Storage support Stefan Hajnoczi
@ 2013-07-01 9:35 ` Stefan Hajnoczi
2013-07-01 9:51 ` Peter Maydell
2013-07-01 10:47 ` Andreas Färber
2013-07-01 9:35 ` [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers Stefan Hajnoczi
1 sibling, 2 replies; 17+ messages in thread
From: Stefan Hajnoczi @ 2013-07-01 9:35 UTC (permalink / raw)
To: qemu-devel; +Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, Peter Maydell
From: Paolo Bonzini <pbonzini@redhat.com>
The next patch will change qemu/tls.h to support more platforms, but at
some performance cost. Declare cpu_single_env directly instead of using
the tls.h abstractions.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
exec.c | 10 ++++++++--
include/exec/cpu-all.h | 14 +++++++++++---
include/qemu/tls.h | 52 --------------------------------------------------
3 files changed, 19 insertions(+), 57 deletions(-)
delete mode 100644 include/qemu/tls.h
diff --git a/exec.c b/exec.c
index c49806c..02263db 100644
--- a/exec.c
+++ b/exec.c
@@ -70,9 +70,15 @@ static MemoryRegion io_mem_unassigned;
#endif
CPUArchState *first_cpu;
+
/* current CPU in the current thread. It is only valid inside
- cpu_exec() */
-DEFINE_TLS(CPUArchState *,cpu_single_env);
+ * cpu_exec(). See comment in include/exec/cpu-all.h. */
+#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
+__thread CPUArchState *cpu_single_env;
+#else
+CPUArchState *cpu_single_env;
+#endif
+
/* 0 = Do not count executed instructions.
1 = Precise instruction counting.
2 = Adaptive rate instruction counting. */
diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
index 35bdf85..7e8df70 100644
--- a/include/exec/cpu-all.h
+++ b/include/exec/cpu-all.h
@@ -20,7 +20,6 @@
#define CPU_ALL_H
#include "qemu-common.h"
-#include "qemu/tls.h"
#include "exec/cpu-common.h"
#include "qemu/thread.h"
@@ -358,8 +357,17 @@ CPUArchState *cpu_copy(CPUArchState *env);
void QEMU_NORETURN cpu_abort(CPUArchState *env, const char *fmt, ...)
GCC_FMT_ATTR(2, 3);
extern CPUArchState *first_cpu;
-DECLARE_TLS(CPUArchState *,cpu_single_env);
-#define cpu_single_env tls_var(cpu_single_env)
+
+/* This is thread-local depending on __linux__ because:
+ * - the only -user mode supporting multiple VCPU threads is linux-user
+ * - TCG system mode is single-threaded regarding VCPUs
+ * - KVM system mode is multi-threaded but limited to Linux
+ */
+#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
+extern __thread CPUArchState *cpu_single_env;
+#else
+extern CPUArchState *cpu_single_env;
+#endif
/* Flags for use in ENV->INTERRUPT_PENDING.
diff --git a/include/qemu/tls.h b/include/qemu/tls.h
deleted file mode 100644
index b92ea9d..0000000
--- a/include/qemu/tls.h
+++ /dev/null
@@ -1,52 +0,0 @@
-/*
- * Abstraction layer for defining and using TLS variables
- *
- * Copyright (c) 2011 Red Hat, Inc
- * Copyright (c) 2011 Linaro Limited
- *
- * Authors:
- * Paolo Bonzini <pbonzini@redhat.com>
- * Peter Maydell <peter.maydell@linaro.org>
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License as
- * published by the Free Software Foundation; either version 2 of
- * the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License along
- * with this program; if not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef QEMU_TLS_H
-#define QEMU_TLS_H
-
-/* Per-thread variables. Note that we only have implementations
- * which are really thread-local on Linux; the dummy implementations
- * define plain global variables.
- *
- * This means that for the moment use should be restricted to
- * per-VCPU variables, which are OK because:
- * - the only -user mode supporting multiple VCPU threads is linux-user
- * - TCG system mode is single-threaded regarding VCPUs
- * - KVM system mode is multi-threaded but limited to Linux
- *
- * TODO: proper implementations via Win32 .tls sections and
- * POSIX pthread_getspecific.
- */
-#ifdef __linux__
-#define DECLARE_TLS(type, x) extern DEFINE_TLS(type, x)
-#define DEFINE_TLS(type, x) __thread __typeof__(type) tls__##x
-#define tls_var(x) tls__##x
-#else
-/* Dummy implementations which define plain global variables */
-#define DECLARE_TLS(type, x) extern DEFINE_TLS(type, x)
-#define DEFINE_TLS(type, x) __typeof__(type) tls__##x
-#define tls_var(x) tls__##x
-#endif
-
-#endif
--
1.8.1.4
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 9:35 [Qemu-devel] [PATCH 0/2] Fast Thread-Local Storage support Stefan Hajnoczi
2013-07-01 9:35 ` [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h Stefan Hajnoczi
@ 2013-07-01 9:35 ` Stefan Hajnoczi
2013-07-01 9:54 ` Peter Maydell
2013-07-01 18:52 ` Ed Maste
1 sibling, 2 replies; 17+ messages in thread
From: Stefan Hajnoczi @ 2013-07-01 9:35 UTC (permalink / raw)
To: qemu-devel
Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, Stefan Hajnoczi, Peter Maydell
From: Paolo Bonzini <pbonzini@redhat.com>
Fast TLS is not available on some platforms, but it is always nice to
use it. This wrapper implementation falls back to pthread_get/setspecific
on POSIX systems that lack __thread, but uses the dynamic linker's TLS
support on Linux and Windows.
The user shall call tls_alloc_foo() in every thread that needs to access
the variable---exactly once and before any access. foo is the name of
the variable as passed to DECLARE_TLS and DEFINE_TLS. Then,
tls_get_foo() will return the address of the variable. It is guaranteed
to remain the same across the lifetime of a thread, so you can cache it.
[Renamed alloc_foo()/get_foo() to tls_alloc_foo()/tls_get_foo()
-- Stefan]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
configure | 21 +++++++
include/qemu/tls.h | 139 +++++++++++++++++++++++++++++++++++++++++++++++
tests/Makefile | 3 +
tests/test-tls.c | 87 +++++++++++++++++++++++++++++
util/qemu-thread-win32.c | 17 ++++++
5 files changed, 267 insertions(+)
create mode 100644 include/qemu/tls.h
create mode 100644 tests/test-tls.c
diff --git a/configure b/configure
index 0e0adde..8ccdf2e 100755
--- a/configure
+++ b/configure
@@ -284,6 +284,7 @@ fi
ar="${AR-${cross_prefix}ar}"
as="${AS-${cross_prefix}as}"
cpp="${CPP-$cc -E}"
+nm="${NM-${cross_prefix}nm}"
objcopy="${OBJCOPY-${cross_prefix}objcopy}"
ld="${LD-${cross_prefix}ld}"
libtool="${LIBTOOL-${cross_prefix}libtool}"
@@ -3163,6 +3164,22 @@ if test "$trace_backend" = "dtrace"; then
fi
##########################################
+# check for TLS runtime
+
+# Some versions of mingw include the "magic" definitions that make
+# TLS work, some don't. Check for it.
+
+if test "$mingw32" = yes; then
+ cat > $TMPC << EOF
+int main(void) {}
+EOF
+ compile_prog "" ""
+ if $nm $TMPE | grep _tls_used > /dev/null 2>&1; then
+ mingw32_tls_runtime=yes
+ fi
+fi
+
+##########################################
# check and set a backend for coroutine
# We prefer ucontext, but it's not always possible. The fallback
@@ -3621,6 +3638,9 @@ if test "$mingw32" = "yes" ; then
version_micro=0
echo "CONFIG_FILEVERSION=$version_major,$version_minor,$version_subminor,$version_micro" >> $config_host_mak
echo "CONFIG_PRODUCTVERSION=$version_major,$version_minor,$version_subminor,$version_micro" >> $config_host_mak
+ if test "$mingw32_tls_runtime" = yes; then
+ echo "CONFIG_MINGW32_TLS_RUNTIME=y" >> $config_host_mak
+ fi
else
echo "CONFIG_POSIX=y" >> $config_host_mak
fi
@@ -4043,6 +4063,7 @@ echo "OBJCC=$objcc" >> $config_host_mak
echo "AR=$ar" >> $config_host_mak
echo "AS=$as" >> $config_host_mak
echo "CPP=$cpp" >> $config_host_mak
+echo "NM=$nm" >> $config_host_mak
echo "OBJCOPY=$objcopy" >> $config_host_mak
echo "LD=$ld" >> $config_host_mak
echo "WINDRES=$windres" >> $config_host_mak
diff --git a/include/qemu/tls.h b/include/qemu/tls.h
new file mode 100644
index 0000000..750ccb3
--- /dev/null
+++ b/include/qemu/tls.h
@@ -0,0 +1,139 @@
+/*
+ * Abstraction layer for defining and using TLS variables
+ *
+ * Copyright (c) 2011, 2013 Red Hat, Inc
+ * Copyright (c) 2011 Linaro Limited
+ *
+ * Authors:
+ * Paolo Bonzini <pbonzini@redhat.com>
+ * Peter Maydell <peter.maydell@linaro.org>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef QEMU_TLS_H
+#define QEMU_TLS_H
+
+#if defined __linux__
+#define DECLARE_TLS(type, x) \
+extern __thread typeof(type) x; \
+ \
+static inline typeof(type) *tls_get_##x(void) \
+{ \
+ return &x; \
+} \
+ \
+static inline typeof(type) *tls_alloc_##x(void) \
+{ \
+ return &x; \
+} \
+ \
+extern int dummy_##__LINE__
+
+#define DEFINE_TLS(type, x) \
+__thread typeof(type) x
+
+#elif defined CONFIG_POSIX
+typedef struct QEMUTLSValue {
+ pthread_key_t k;
+ pthread_once_t o;
+} QEMUTLSValue;
+
+#define DECLARE_TLS(type, x) \
+extern QEMUTLSValue x; \
+extern void init_##x(void); \
+ \
+static inline typeof(type) *tls_get_##x(void) \
+{ \
+ return pthread_getspecific(x.k); \
+} \
+ \
+static inline typeof(type) *tls_alloc_##x(void) \
+{ \
+ void *datum = g_malloc0(sizeof(type)); \
+ pthread_once(&x.o, init_##x); \
+ pthread_setspecific(x.k, datum); \
+ return datum; \
+} \
+ \
+extern int dummy_##__LINE__
+
+#define DEFINE_TLS(type, x) \
+void init_##x(void) { \
+ pthread_key_create(&x.k, g_free); \
+} \
+ \
+QEMUTLSValue x = { .o = PTHREAD_ONCE_INIT }
+
+#elif defined CONFIG_WIN32
+
+/* The initial contents of TLS variables are placed in the .tls section.
+ * The linker takes all section starting with ".tls$", sorts them and puts
+ * the contents in a single ".tls" section. qemu-thread-win32.c defines
+ * special symbols in .tls$000 and .tls$ZZZ that represent the beginning
+ * and end of TLS memory. The linker and run-time library then cooperate
+ * to copy memory between those symbols in the TLS area of new threads.
+ *
+ * _tls_index holds the number of our module. The executable should be
+ * zero, DLLs are numbered 1 and up. The loader fills it in for us.
+ *
+ * Thus, Teb->ThreadLocalStoragePointer[_tls_index] is the base of
+ * the TLS segment for this (thread, module) pair. Each segment has
+ * the same layout as this module's .tls segment and is initialized
+ * with the content of the .tls segment; 0 is the _tls_start variable.
+ * So, tls_get_##x passes us the offset of the passed variable relative to
+ * _tls_start, and we return that same offset plus the base of segment.
+ */
+
+typedef struct _TEB {
+ NT_TIB NtTib;
+ void *EnvironmentPointer;
+ void *x[3];
+ char **ThreadLocalStoragePointer;
+} TEB, *PTEB;
+
+extern int _tls_index;
+extern int _tls_start;
+
+static inline void *tls_var(size_t offset)
+{
+ PTEB Teb = NtCurrentTeb();
+ return (char *)(Teb->ThreadLocalStoragePointer[_tls_index]) + offset;
+}
+
+#define DECLARE_TLS(type, x) \
+extern typeof(type) tls_##x __attribute__((section(".tls$QEMU"))); \
+ \
+static inline typeof(type) *tls_get_##x(void) \
+{ \
+ return tls_var((ULONG_PTR)&(tls_##x) - (ULONG_PTR)&_tls_start); \
+} \
+ \
+static inline typeof(type) *tls_alloc_##x(void) \
+{ \
+ typeof(type) *addr = tls_get_##x(); \
+ memset((void *)addr, 0, sizeof(type)); \
+ return addr; \
+} \
+ \
+extern int dummy_##__LINE__
+
+#define DEFINE_TLS(type, x) \
+typeof(type) tls_##x __attribute__((section(".tls$QEMU")))
+
+#else
+#error No TLS abstraction available on this platform
+#endif
+
+#endif
diff --git a/tests/Makefile b/tests/Makefile
index c107489..1f5156a 100644
--- a/tests/Makefile
+++ b/tests/Makefile
@@ -42,6 +42,8 @@ check-unit-y += tests/test-xbzrle$(EXESUF)
gcov-files-test-xbzrle-y = xbzrle.c
check-unit-y += tests/test-cutils$(EXESUF)
gcov-files-test-cutils-y += util/cutils.c
+check-unit-y += tests/test-tls$(EXESUF)
+gcov-files-test-tls-y +=
check-unit-y += tests/test-mul64$(EXESUF)
gcov-files-test-mul64-y = util/host-utils.c
@@ -98,6 +100,7 @@ tests/test-hbitmap$(EXESUF): tests/test-hbitmap.o libqemuutil.a libqemustub.a
tests/test-x86-cpuid$(EXESUF): tests/test-x86-cpuid.o
tests/test-xbzrle$(EXESUF): tests/test-xbzrle.o xbzrle.o page_cache.o libqemuutil.a
tests/test-cutils$(EXESUF): tests/test-cutils.o util/cutils.o
+tests/test-tls$(EXESUF): tests/test-tls.o libqemuutil.a
tests/test-qapi-types.c tests/test-qapi-types.h :\
$(SRC_PATH)/qapi-schema-test.json $(SRC_PATH)/scripts/qapi-types.py
diff --git a/tests/test-tls.c b/tests/test-tls.c
new file mode 100644
index 0000000..54e981d
--- /dev/null
+++ b/tests/test-tls.c
@@ -0,0 +1,87 @@
+/*
+ * Unit-tests for TLS wrappers
+ *
+ * Copyright (C) 2013 Red Hat Inc.
+ *
+ * Authors:
+ * Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include <glib.h>
+#include <errno.h>
+#include <string.h>
+
+#include "qemu-common.h"
+#include "qemu/atomic.h"
+#include "qemu/thread.h"
+#include "qemu/tls.h"
+
+DECLARE_TLS(volatile long long, cnt);
+DEFINE_TLS(volatile long long, cnt);
+
+#define NUM_THREADS 10
+
+int stop;
+
+static void *test_thread(void *arg)
+{
+ volatile long long *p_cnt = alloc_cnt();
+ volatile long long **p_ret = arg;
+ long long exp = 0;
+
+ g_assert(get_cnt() == p_cnt);
+ *p_ret = p_cnt;
+ g_assert(*p_cnt == 0);
+ while (atomic_mb_read(&stop) == 0) {
+ exp++;
+ (*p_cnt)++;
+ g_assert(*get_cnt() == exp);
+ }
+
+ return NULL;
+}
+
+static void test_tls(void)
+{
+ volatile long long *addr[NUM_THREADS];
+ QemuThread t[NUM_THREADS];
+ int i;
+
+ for (i = 0; i < NUM_THREADS; i++) {
+ qemu_thread_create(&t[i], test_thread, &addr[i], QEMU_THREAD_JOINABLE);
+ }
+ g_usleep(1000000);
+ atomic_mb_set(&stop, 1);
+ for (i = 0; i < NUM_THREADS; i++) {
+ qemu_thread_join(&t[i]);
+ }
+ for (i = 1; i < NUM_THREADS; i++) {
+ g_assert(addr[i] != addr[i - 1]);
+ }
+}
+
+int main(int argc, char **argv)
+{
+ g_test_init(&argc, &argv, NULL);
+
+ g_test_add_func("/tls", test_tls);
+ return g_test_run();
+}
diff --git a/util/qemu-thread-win32.c b/util/qemu-thread-win32.c
index 517878d..f75e404 100644
--- a/util/qemu-thread-win32.c
+++ b/util/qemu-thread-win32.c
@@ -16,6 +16,23 @@
#include <assert.h>
#include <limits.h>
+/* TLS support. Some versions of mingw32 provide it, others do not. */
+
+#ifndef CONFIG_MINGW32_TLS_RUNTIME
+int __attribute__((section(".tls$AAA"))) _tls_start = 0;
+int __attribute__((section(".tls$ZZZ"))) _tls_end = 0;
+int _tls_index = 0;
+
+const IMAGE_TLS_DIRECTORY _tls_used __attribute__((used, section(".rdata$T"))) = {
+ (ULONG)(ULONG_PTR) &_tls_start, /* start of tls data */
+ (ULONG)(ULONG_PTR) &_tls_end, /* end of tls data */
+ (ULONG)(ULONG_PTR) &_tls_index, /* address of tls_index */
+ (ULONG) 0, /* pointer to callbacks */
+ (ULONG) 0, /* size of tls zero fill */
+ (ULONG) 0 /* characteristics */
+};
+#endif
+
static void error_exit(int err, const char *msg)
{
char *pstr;
--
1.8.1.4
^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h
2013-07-01 9:35 ` [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h Stefan Hajnoczi
@ 2013-07-01 9:51 ` Peter Maydell
2013-07-01 10:15 ` Paolo Bonzini
2013-07-01 10:47 ` Andreas Färber
1 sibling, 1 reply; 17+ messages in thread
From: Peter Maydell @ 2013-07-01 9:51 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, qemu-devel
On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> From: Paolo Bonzini <pbonzini@redhat.com>
>
> The next patch will change qemu/tls.h to support more platforms, but at
> some performance cost. Declare cpu_single_env directly instead of using
> the tls.h abstractions.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> exec.c | 10 ++++++++--
> include/exec/cpu-all.h | 14 +++++++++++---
> include/qemu/tls.h | 52 --------------------------------------------------
> 3 files changed, 19 insertions(+), 57 deletions(-)
> delete mode 100644 include/qemu/tls.h
>
> diff --git a/exec.c b/exec.c
> index c49806c..02263db 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -70,9 +70,15 @@ static MemoryRegion io_mem_unassigned;
> #endif
>
> CPUArchState *first_cpu;
> +
> /* current CPU in the current thread. It is only valid inside
> - cpu_exec() */
> -DEFINE_TLS(CPUArchState *,cpu_single_env);
> + * cpu_exec(). See comment in include/exec/cpu-all.h. */
> +#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
> +__thread CPUArchState *cpu_single_env;
> +#else
> +CPUArchState *cpu_single_env;
> +#endif
This is still wrong, as per my review comments on the previous
version.
-- PMM
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 9:35 ` [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers Stefan Hajnoczi
@ 2013-07-01 9:54 ` Peter Maydell
2013-07-01 10:14 ` Paolo Bonzini
2013-07-01 12:34 ` Stefan Hajnoczi
2013-07-01 18:52 ` Ed Maste
1 sibling, 2 replies; 17+ messages in thread
From: Peter Maydell @ 2013-07-01 9:54 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, qemu-devel
On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> From: Paolo Bonzini <pbonzini@redhat.com>
>
> Fast TLS is not available on some platforms, but it is always nice to
> use it. This wrapper implementation falls back to pthread_get/setspecific
> on POSIX systems that lack __thread, but uses the dynamic linker's TLS
> support on Linux and Windows.
>
> The user shall call tls_alloc_foo() in every thread that needs to access
> the variable---exactly once and before any access. foo is the name of
> the variable as passed to DECLARE_TLS and DEFINE_TLS. Then,
> tls_get_foo() will return the address of the variable. It is guaranteed
> to remain the same across the lifetime of a thread, so you can cache it.
> ##########################################
> +# check for TLS runtime
> +
> +# Some versions of mingw include the "magic" definitions that make
> +# TLS work, some don't. Check for it.
> +
> +if test "$mingw32" = yes; then
> + cat > $TMPC << EOF
> +int main(void) {}
Execution falls off the end of function without returning a value
(I would expect the compiler to issue a warning about this.)
> +#ifndef QEMU_TLS_H
> +#define QEMU_TLS_H
> +
> +#if defined __linux__
> +#define DECLARE_TLS(type, x) \
> +extern __thread typeof(type) x; \
> + \
> +static inline typeof(type) *tls_get_##x(void) \
> +{ \
> + return &x; \
> +} \
> + \
> +static inline typeof(type) *tls_alloc_##x(void) \
> +{ \
> + return &x; \
> +} \
> + \
> +extern int dummy_##__LINE__
What's this for?
thanks
-- PMM
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 9:54 ` Peter Maydell
@ 2013-07-01 10:14 ` Paolo Bonzini
2013-07-04 16:27 ` Jan Kiszka
2013-07-01 12:34 ` Stefan Hajnoczi
1 sibling, 1 reply; 17+ messages in thread
From: Paolo Bonzini @ 2013-07-01 10:14 UTC (permalink / raw)
To: Peter Maydell; +Cc: Anthony Liguori, Ed Maste, qemu-devel, Stefan Hajnoczi
Il 01/07/2013 11:54, Peter Maydell ha scritto:
> On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
>> From: Paolo Bonzini <pbonzini@redhat.com>
>>
>> Fast TLS is not available on some platforms, but it is always nice to
>> use it. This wrapper implementation falls back to pthread_get/setspecific
>> on POSIX systems that lack __thread, but uses the dynamic linker's TLS
>> support on Linux and Windows.
>>
>> The user shall call tls_alloc_foo() in every thread that needs to access
>> the variable---exactly once and before any access. foo is the name of
>> the variable as passed to DECLARE_TLS and DEFINE_TLS. Then,
>> tls_get_foo() will return the address of the variable. It is guaranteed
>> to remain the same across the lifetime of a thread, so you can cache it.
>
>> ##########################################
>> +# check for TLS runtime
>> +
>> +# Some versions of mingw include the "magic" definitions that make
>> +# TLS work, some don't. Check for it.
>> +
>> +if test "$mingw32" = yes; then
>> + cat > $TMPC << EOF
>> +int main(void) {}
>
> Execution falls off the end of function without returning a value
> (I would expect the compiler to issue a warning about this.)
>
>> +#ifndef QEMU_TLS_H
>> +#define QEMU_TLS_H
>> +
>> +#if defined __linux__
>> +#define DECLARE_TLS(type, x) \
>> +extern __thread typeof(type) x; \
>> + \
>> +static inline typeof(type) *tls_get_##x(void) \
>> +{ \
>> + return &x; \
>> +} \
>> + \
>> +static inline typeof(type) *tls_alloc_##x(void) \
>> +{ \
>> + return &x; \
>> +} \
>> + \
>> +extern int dummy_##__LINE__
>
> What's this for?
It lets you use it as
DECLARE_TLS(type, x);
Many editors impose an indent without the trailing semicolon (they parse
it as a K&R function definition).
Paolo
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h
2013-07-01 9:51 ` Peter Maydell
@ 2013-07-01 10:15 ` Paolo Bonzini
0 siblings, 0 replies; 17+ messages in thread
From: Paolo Bonzini @ 2013-07-01 10:15 UTC (permalink / raw)
To: Peter Maydell; +Cc: Anthony Liguori, Ed Maste, qemu-devel, Stefan Hajnoczi
Il 01/07/2013 11:51, Peter Maydell ha scritto:
> On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
>> From: Paolo Bonzini <pbonzini@redhat.com>
>>
>> The next patch will change qemu/tls.h to support more platforms, but at
>> some performance cost. Declare cpu_single_env directly instead of using
>> the tls.h abstractions.
>>
>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>> ---
>> exec.c | 10 ++++++++--
>> include/exec/cpu-all.h | 14 +++++++++++---
>> include/qemu/tls.h | 52 --------------------------------------------------
>> 3 files changed, 19 insertions(+), 57 deletions(-)
>> delete mode 100644 include/qemu/tls.h
>>
>> diff --git a/exec.c b/exec.c
>> index c49806c..02263db 100644
>> --- a/exec.c
>> +++ b/exec.c
>> @@ -70,9 +70,15 @@ static MemoryRegion io_mem_unassigned;
>> #endif
>>
>> CPUArchState *first_cpu;
>> +
>> /* current CPU in the current thread. It is only valid inside
>> - cpu_exec() */
>> -DEFINE_TLS(CPUArchState *,cpu_single_env);
>> + * cpu_exec(). See comment in include/exec/cpu-all.h. */
>> +#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
>> +__thread CPUArchState *cpu_single_env;
>> +#else
>> +CPUArchState *cpu_single_env;
>> +#endif
>
> This is still wrong, as per my review comments on the previous
> version.
I'm replying there.
Paolo
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h
2013-07-01 9:35 ` [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h Stefan Hajnoczi
2013-07-01 9:51 ` Peter Maydell
@ 2013-07-01 10:47 ` Andreas Färber
1 sibling, 0 replies; 17+ messages in thread
From: Andreas Färber @ 2013-07-01 10:47 UTC (permalink / raw)
To: Stefan Hajnoczi
Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, qemu-devel, Peter Maydell
Am 01.07.2013 11:35, schrieb Stefan Hajnoczi:
> From: Paolo Bonzini <pbonzini@redhat.com>
>
> The next patch will change qemu/tls.h to support more platforms, but at
> some performance cost. Declare cpu_single_env directly instead of using
> the tls.h abstractions.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> exec.c | 10 ++++++++--
> include/exec/cpu-all.h | 14 +++++++++++---
> include/qemu/tls.h | 52 --------------------------------------------------
> 3 files changed, 19 insertions(+), 57 deletions(-)
> delete mode 100644 include/qemu/tls.h
>
> diff --git a/exec.c b/exec.c
> index c49806c..02263db 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -70,9 +70,15 @@ static MemoryRegion io_mem_unassigned;
> #endif
>
> CPUArchState *first_cpu;
> +
> /* current CPU in the current thread. It is only valid inside
> - cpu_exec() */
> -DEFINE_TLS(CPUArchState *,cpu_single_env);
> + * cpu_exec(). See comment in include/exec/cpu-all.h. */
> +#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
> +__thread CPUArchState *cpu_single_env;
> +#else
> +CPUArchState *cpu_single_env;
> +#endif
As indicated elsewhere, this conflicts with my pending CPUState part 10
series, which moves, renames and changes type of cpu_single_env:
http://patchwork.ozlabs.org/patch/254825/
The latest version received no further feedback after finding a better
name than cpu_single_cpu (current_cpu), first_cpu/next_cpu is ack'ed and
so are the prereqs for current_cpu, so I will pull that in as soon as
potential conflicts with ppc and memory pulls are sorted out, since
pretty invasive and a milestone towards getting rid of CPUArchState.
I'd be happy to help with rebasing this afterwards when needed.
Regards,
Andreas
> +
> /* 0 = Do not count executed instructions.
> 1 = Precise instruction counting.
> 2 = Adaptive rate instruction counting. */
> diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
> index 35bdf85..7e8df70 100644
> --- a/include/exec/cpu-all.h
> +++ b/include/exec/cpu-all.h
> @@ -20,7 +20,6 @@
> #define CPU_ALL_H
>
> #include "qemu-common.h"
> -#include "qemu/tls.h"
> #include "exec/cpu-common.h"
> #include "qemu/thread.h"
>
> @@ -358,8 +357,17 @@ CPUArchState *cpu_copy(CPUArchState *env);
> void QEMU_NORETURN cpu_abort(CPUArchState *env, const char *fmt, ...)
> GCC_FMT_ATTR(2, 3);
> extern CPUArchState *first_cpu;
> -DECLARE_TLS(CPUArchState *,cpu_single_env);
> -#define cpu_single_env tls_var(cpu_single_env)
> +
> +/* This is thread-local depending on __linux__ because:
> + * - the only -user mode supporting multiple VCPU threads is linux-user
> + * - TCG system mode is single-threaded regarding VCPUs
> + * - KVM system mode is multi-threaded but limited to Linux
> + */
> +#if defined CONFIG_KVM || (defined CONFIG_USER_ONLY && defined CONFIG_USE_NPTL)
> +extern __thread CPUArchState *cpu_single_env;
> +#else
> +extern CPUArchState *cpu_single_env;
> +#endif
>
> /* Flags for use in ENV->INTERRUPT_PENDING.
>
> diff --git a/include/qemu/tls.h b/include/qemu/tls.h
> deleted file mode 100644
> index b92ea9d..0000000
> --- a/include/qemu/tls.h
> +++ /dev/null
> @@ -1,52 +0,0 @@
> -/*
> - * Abstraction layer for defining and using TLS variables
> - *
> - * Copyright (c) 2011 Red Hat, Inc
> - * Copyright (c) 2011 Linaro Limited
> - *
> - * Authors:
> - * Paolo Bonzini <pbonzini@redhat.com>
> - * Peter Maydell <peter.maydell@linaro.org>
> - *
> - * This program is free software; you can redistribute it and/or
> - * modify it under the terms of the GNU General Public License as
> - * published by the Free Software Foundation; either version 2 of
> - * the License, or (at your option) any later version.
> - *
> - * This program is distributed in the hope that it will be useful,
> - * but WITHOUT ANY WARRANTY; without even the implied warranty of
> - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> - * GNU General Public License for more details.
> - *
> - * You should have received a copy of the GNU General Public License along
> - * with this program; if not, see <http://www.gnu.org/licenses/>.
> - */
> -
> -#ifndef QEMU_TLS_H
> -#define QEMU_TLS_H
> -
> -/* Per-thread variables. Note that we only have implementations
> - * which are really thread-local on Linux; the dummy implementations
> - * define plain global variables.
> - *
> - * This means that for the moment use should be restricted to
> - * per-VCPU variables, which are OK because:
> - * - the only -user mode supporting multiple VCPU threads is linux-user
> - * - TCG system mode is single-threaded regarding VCPUs
> - * - KVM system mode is multi-threaded but limited to Linux
> - *
> - * TODO: proper implementations via Win32 .tls sections and
> - * POSIX pthread_getspecific.
> - */
> -#ifdef __linux__
> -#define DECLARE_TLS(type, x) extern DEFINE_TLS(type, x)
> -#define DEFINE_TLS(type, x) __thread __typeof__(type) tls__##x
> -#define tls_var(x) tls__##x
> -#else
> -/* Dummy implementations which define plain global variables */
> -#define DECLARE_TLS(type, x) extern DEFINE_TLS(type, x)
> -#define DEFINE_TLS(type, x) __typeof__(type) tls__##x
> -#define tls_var(x) tls__##x
> -#endif
> -
> -#endif
>
--
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 9:54 ` Peter Maydell
2013-07-01 10:14 ` Paolo Bonzini
@ 2013-07-01 12:34 ` Stefan Hajnoczi
1 sibling, 0 replies; 17+ messages in thread
From: Stefan Hajnoczi @ 2013-07-01 12:34 UTC (permalink / raw)
To: Peter Maydell; +Cc: Paolo Bonzini, Anthony Liguori, Ed Maste, qemu-devel
On Mon, Jul 01, 2013 at 10:54:56AM +0100, Peter Maydell wrote:
> On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > From: Paolo Bonzini <pbonzini@redhat.com>
> >
> > Fast TLS is not available on some platforms, but it is always nice to
> > use it. This wrapper implementation falls back to pthread_get/setspecific
> > on POSIX systems that lack __thread, but uses the dynamic linker's TLS
> > support on Linux and Windows.
> >
> > The user shall call tls_alloc_foo() in every thread that needs to access
> > the variable---exactly once and before any access. foo is the name of
> > the variable as passed to DECLARE_TLS and DEFINE_TLS. Then,
> > tls_get_foo() will return the address of the variable. It is guaranteed
> > to remain the same across the lifetime of a thread, so you can cache it.
>
> > ##########################################
> > +# check for TLS runtime
> > +
> > +# Some versions of mingw include the "magic" definitions that make
> > +# TLS work, some don't. Check for it.
> > +
> > +if test "$mingw32" = yes; then
> > + cat > $TMPC << EOF
> > +int main(void) {}
>
> Execution falls off the end of function without returning a value
> (I would expect the compiler to issue a warning about this.)
You are right, gcc emits a warning.
> > +#ifndef QEMU_TLS_H
> > +#define QEMU_TLS_H
> > +
> > +#if defined __linux__
> > +#define DECLARE_TLS(type, x) \
> > +extern __thread typeof(type) x; \
> > + \
> > +static inline typeof(type) *tls_get_##x(void) \
> > +{ \
> > + return &x; \
> > +} \
> > + \
> > +static inline typeof(type) *tls_alloc_##x(void) \
> > +{ \
> > + return &x; \
> > +} \
> > + \
> > +extern int dummy_##__LINE__
>
> What's this for?
It makes the DECLARE_TLS() macro use a semicolon.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 9:35 ` [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers Stefan Hajnoczi
2013-07-01 9:54 ` Peter Maydell
@ 2013-07-01 18:52 ` Ed Maste
2013-07-01 19:25 ` Peter Maydell
2013-07-02 7:50 ` Stefan Hajnoczi
1 sibling, 2 replies; 17+ messages in thread
From: Ed Maste @ 2013-07-01 18:52 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: Paolo Bonzini, qemu-devel
On 1 July 2013 05:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> From: Paolo Bonzini <pbonzini@redhat.com>
>
> Fast TLS is not available on some platforms, but it is always nice to
> use it. This wrapper implementation falls back to pthread_get/setspecific
> on POSIX systems that lack __thread, but uses the dynamic linker's TLS
> support on Linux and Windows.
The most recent version of this patch posted by Paolo that I see has:
+#if defined(__linux__) || defined(__FreeBSD__)
+#define DECLARE_TLS(type, x) \
while this one has only __linux__. Do you mind picking up that
change, if this is likely to make it in via your work?
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 18:52 ` Ed Maste
@ 2013-07-01 19:25 ` Peter Maydell
2013-07-01 20:00 ` Ed Maste
2013-07-01 20:30 ` Richard Henderson
2013-07-02 7:50 ` Stefan Hajnoczi
1 sibling, 2 replies; 17+ messages in thread
From: Peter Maydell @ 2013-07-01 19:25 UTC (permalink / raw)
To: Ed Maste; +Cc: Paolo Bonzini, qemu-devel, Stefan Hajnoczi
On 1 July 2013 19:52, Ed Maste <emaste@freebsd.org> wrote:
> The most recent version of this patch posted by Paolo that I see has:
>
> +#if defined(__linux__) || defined(__FreeBSD__)
> +#define DECLARE_TLS(type, x) \
>
> while this one has only __linux__. Do you mind picking up that
> change, if this is likely to make it in via your work?
Does any OS have a __thread which compiles but is broken, or can
we just have a configure test for this? That would let MacOSX+clang
use __thread.
-- PMM
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 19:25 ` Peter Maydell
@ 2013-07-01 20:00 ` Ed Maste
2013-07-01 20:30 ` Richard Henderson
1 sibling, 0 replies; 17+ messages in thread
From: Ed Maste @ 2013-07-01 20:00 UTC (permalink / raw)
To: Peter Maydell; +Cc: Paolo Bonzini, qemu-devel, Stefan Hajnoczi
On 1 July 2013 15:25, Peter Maydell <peter.maydell@linaro.org> wrote:
> Does any OS have a __thread which compiles but is broken, or can
> we just have a configure test for this? That would let MacOSX+clang
> use __thread.
I believe this was recently the case on NetBSD - code with __thread
would build, but fail at run time. I think this applies to NetBSD 5.0
and earlier, but I see that TLS support is listed in the 6.0 release
notes. So perhaps a __thread configure test with an override for the
known-failing case(s).
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 19:25 ` Peter Maydell
2013-07-01 20:00 ` Ed Maste
@ 2013-07-01 20:30 ` Richard Henderson
2013-07-02 7:54 ` Paolo Bonzini
1 sibling, 1 reply; 17+ messages in thread
From: Richard Henderson @ 2013-07-01 20:30 UTC (permalink / raw)
To: Peter Maydell; +Cc: Paolo Bonzini, Ed Maste, qemu-devel, Stefan Hajnoczi
On 07/01/2013 12:25 PM, Peter Maydell wrote:
> Does any OS have a __thread which compiles but is broken, or can
> we just have a configure test for this? That would let MacOSX+clang
> use __thread.
I suspect that this will work. Some targets may succeed in using gcc's
"emutls" path, which while slower than TLS is pretty much exactly the pthread
get/setcontext fallback that's been proposed elsewhere on this list.
r~
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 18:52 ` Ed Maste
2013-07-01 19:25 ` Peter Maydell
@ 2013-07-02 7:50 ` Stefan Hajnoczi
1 sibling, 0 replies; 17+ messages in thread
From: Stefan Hajnoczi @ 2013-07-02 7:50 UTC (permalink / raw)
To: Ed Maste; +Cc: Paolo Bonzini, qemu-devel, Stefan Hajnoczi
On Mon, Jul 01, 2013 at 02:52:08PM -0400, Ed Maste wrote:
> On 1 July 2013 05:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > From: Paolo Bonzini <pbonzini@redhat.com>
> >
> > Fast TLS is not available on some platforms, but it is always nice to
> > use it. This wrapper implementation falls back to pthread_get/setspecific
> > on POSIX systems that lack __thread, but uses the dynamic linker's TLS
> > support on Linux and Windows.
>
> The most recent version of this patch posted by Paolo that I see has:
>
> +#if defined(__linux__) || defined(__FreeBSD__)
> +#define DECLARE_TLS(type, x) \
>
> while this one has only __linux__. Do you mind picking up that
> change, if this is likely to make it in via your work?
Since additional changes are necessary before these patches can be
merged, I hope Paolo can pick up my "tls_" rename and send the next
revision including his improvements.
Stefan
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 20:30 ` Richard Henderson
@ 2013-07-02 7:54 ` Paolo Bonzini
0 siblings, 0 replies; 17+ messages in thread
From: Paolo Bonzini @ 2013-07-02 7:54 UTC (permalink / raw)
To: Richard Henderson; +Cc: Peter Maydell, Ed Maste, qemu-devel, Stefan Hajnoczi
Il 01/07/2013 22:30, Richard Henderson ha scritto:
> On 07/01/2013 12:25 PM, Peter Maydell wrote:
>> Does any OS have a __thread which compiles but is broken, or can
>> we just have a configure test for this? That would let MacOSX+clang
>> use __thread.
>
> I suspect that this will work. Some targets may succeed in using gcc's
> "emutls" path, which while slower than TLS is pretty much exactly the pthread
> get/setcontext fallback that's been proposed elsewhere on this list.
We do not want to hit emutls on Windows, but that can be done simply by
reordering the three implementation.
Paolo
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-01 10:14 ` Paolo Bonzini
@ 2013-07-04 16:27 ` Jan Kiszka
2013-07-04 16:38 ` Paolo Bonzini
0 siblings, 1 reply; 17+ messages in thread
From: Jan Kiszka @ 2013-07-04 16:27 UTC (permalink / raw)
To: Paolo Bonzini
Cc: Peter Maydell, Anthony Liguori, Ed Maste, qemu-devel, Stefan Hajnoczi
On 2013-07-01 12:14, Paolo Bonzini wrote:
> Il 01/07/2013 11:54, Peter Maydell ha scritto:
>> On 1 July 2013 10:35, Stefan Hajnoczi <stefanha@redhat.com> wrote:
>>> From: Paolo Bonzini <pbonzini@redhat.com>
>>>
>>> Fast TLS is not available on some platforms, but it is always nice to
>>> use it. This wrapper implementation falls back to pthread_get/setspecific
>>> on POSIX systems that lack __thread, but uses the dynamic linker's TLS
>>> support on Linux and Windows.
>>>
>>> The user shall call tls_alloc_foo() in every thread that needs to access
>>> the variable---exactly once and before any access. foo is the name of
>>> the variable as passed to DECLARE_TLS and DEFINE_TLS. Then,
>>> tls_get_foo() will return the address of the variable. It is guaranteed
>>> to remain the same across the lifetime of a thread, so you can cache it.
>>
>>> ##########################################
>>> +# check for TLS runtime
>>> +
>>> +# Some versions of mingw include the "magic" definitions that make
>>> +# TLS work, some don't. Check for it.
>>> +
>>> +if test "$mingw32" = yes; then
>>> + cat > $TMPC << EOF
>>> +int main(void) {}
>>
>> Execution falls off the end of function without returning a value
>> (I would expect the compiler to issue a warning about this.)
>>
>>> +#ifndef QEMU_TLS_H
>>> +#define QEMU_TLS_H
>>> +
>>> +#if defined __linux__
>>> +#define DECLARE_TLS(type, x) \
>>> +extern __thread typeof(type) x; \
>>> + \
>>> +static inline typeof(type) *tls_get_##x(void) \
>>> +{ \
>>> + return &x; \
>>> +} \
>>> + \
>>> +static inline typeof(type) *tls_alloc_##x(void) \
>>> +{ \
>>> + return &x; \
>>> +} \
>>> + \
>>> +extern int dummy_##__LINE__
>>
>> What's this for?
>
> It lets you use it as
>
> DECLARE_TLS(type, x);
>
> Many editors impose an indent without the trailing semicolon (they parse
> it as a K&R function definition).
This workaround causes troubles here with gcc-4.5.1:
In file included from /data/qemu/include/exec/memory.h:29:0,
from /data/qemu/include/exec/ioport.h:29,
from /data/qemu/include/hw/hw.h:11,
from /data/qemu/exec.c:30:
/data/qemu/include/qemu/rcu.h:88:339: warning: redundant redeclaration of ‘dummy___LINE__’
/data/qemu/include/exec/cpu-all.h:362:359: note: previous declaration of ‘dummy___LINE__’ was here
/data/qemu/exec.c:77:24: warning: function declaration isn’t a prototype
/data/qemu/exec.c:77:41: error: invalid storage class for function ‘tls_get_cpu_single_env’
/data/qemu/exec.c:77:41: error: conflicting types for ‘tls_get_cpu_single_env’
/data/qemu/include/exec/cpu-all.h:362:148: note: previous definition of ‘tls_get_cpu_single_env’ was here
Jan
--
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers
2013-07-04 16:27 ` Jan Kiszka
@ 2013-07-04 16:38 ` Paolo Bonzini
0 siblings, 0 replies; 17+ messages in thread
From: Paolo Bonzini @ 2013-07-04 16:38 UTC (permalink / raw)
To: Jan Kiszka
Cc: Peter Maydell, Anthony Liguori, Ed Maste, qemu-devel, Stefan Hajnoczi
Il 04/07/2013 18:27, Jan Kiszka ha scritto:
> This workaround causes troubles here with gcc-4.5.1:
>
> In file included from /data/qemu/include/exec/memory.h:29:0,
> from /data/qemu/include/exec/ioport.h:29,
> from /data/qemu/include/hw/hw.h:11,
> from /data/qemu/exec.c:30:
> /data/qemu/include/qemu/rcu.h:88:339: warning: redundant redeclaration of ‘dummy___LINE__’
> /data/qemu/include/exec/cpu-all.h:362:359: note: previous declaration of ‘dummy___LINE__’ was here
> /data/qemu/exec.c:77:24: warning: function declaration isn’t a prototype
> /data/qemu/exec.c:77:41: error: invalid storage class for function ‘tls_get_cpu_single_env’
> /data/qemu/exec.c:77:41: error: conflicting types for ‘tls_get_cpu_single_env’
> /data/qemu/include/exec/cpu-all.h:362:148: note: previous definition of ‘tls_get_cpu_single_env’ was here
Perhaps it helps to use glue(dummy, __LINE__).
Paolo
^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2013-07-04 16:38 UTC | newest]
Thread overview: 17+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-07-01 9:35 [Qemu-devel] [PATCH 0/2] Fast Thread-Local Storage support Stefan Hajnoczi
2013-07-01 9:35 ` [Qemu-devel] [PATCH 1/2] exec: do not use qemu/tls.h Stefan Hajnoczi
2013-07-01 9:51 ` Peter Maydell
2013-07-01 10:15 ` Paolo Bonzini
2013-07-01 10:47 ` Andreas Färber
2013-07-01 9:35 ` [Qemu-devel] [PATCH 2/2] qemu-thread: add TLS wrappers Stefan Hajnoczi
2013-07-01 9:54 ` Peter Maydell
2013-07-01 10:14 ` Paolo Bonzini
2013-07-04 16:27 ` Jan Kiszka
2013-07-04 16:38 ` Paolo Bonzini
2013-07-01 12:34 ` Stefan Hajnoczi
2013-07-01 18:52 ` Ed Maste
2013-07-01 19:25 ` Peter Maydell
2013-07-01 20:00 ` Ed Maste
2013-07-01 20:30 ` Richard Henderson
2013-07-02 7:54 ` Paolo Bonzini
2013-07-02 7:50 ` Stefan Hajnoczi
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.