* [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 11:09 ` Fam Zheng
2017-01-18 14:33 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 02/16] block-backend: allow blk_prw from coroutine context Paolo Bonzini
` (17 subsequent siblings)
18 siblings, 2 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
aio_co_wake provides the infrastructure to start a coroutine on a "home"
AioContext. It will be used by CoMutex and CoQueue, so that coroutines
don't jump from one context to another when they go to sleep on a
mutex or waitqueue. However, it can also be used as a more efficient
alternative to one-shot bottom halves, and saves the effort of tracking
which AioContext a coroutine is running on.
aio_co_schedule is the part of aio_co_wake that starts a coroutine
on a remove AioContext, but it is also useful to implement e.g.
bdrv_set_aio_context callbacks.
The implementation of aio_co_schedule is based on a lock-free
multiple-producer, single-consumer queue. The multiple producers use
cmpxchg to add to a LIFO stack. The consumer (a per-AioContext bottom
half) grabs all items added so far, inverts the list to make it FIFO,
and goes through it one item at a time until it's empty. The data
structure was inspired by OSv, which uses it in the very code we'll
"port" to QEMU for the thread-safe CoMutex.
Most of the new code is really tests.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
async.c | 65 +++++++++++++
include/block/aio.h | 32 +++++++
include/qemu/coroutine_int.h | 10 +-
tests/Makefile.include | 13 ++-
tests/iothread.c | 91 ++++++++++++++++++
tests/iothread.h | 25 +++++
tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++++
tests/test-vmstate.c | 11 ---
trace-events | 4 +
util/qemu-coroutine.c | 8 ++
10 files changed, 456 insertions(+), 16 deletions(-)
create mode 100644 tests/iothread.c
create mode 100644 tests/iothread.h
create mode 100644 tests/test-aio-multithread.c
diff --git a/async.c b/async.c
index 0d218ab..1338682 100644
--- a/async.c
+++ b/async.c
@@ -30,6 +30,8 @@
#include "qemu/main-loop.h"
#include "qemu/atomic.h"
#include "block/raw-aio.h"
+#include "trace/generated-tracers.h"
+#include "qemu/coroutine_int.h"
/***********************************************************/
/* bottom halves (can be seen as timers which expire ASAP) */
@@ -274,6 +276,9 @@ aio_ctx_finalize(GSource *source)
}
#endif
+ assert(QSLIST_EMPTY(&ctx->scheduled_coroutines));
+ qemu_bh_delete(ctx->co_schedule_bh);
+
qemu_lockcnt_lock(&ctx->list_lock);
assert(!qemu_lockcnt_count(&ctx->list_lock));
while (ctx->first_bh) {
@@ -363,6 +368,28 @@ static bool event_notifier_poll(void *opaque)
return atomic_read(&ctx->notified);
}
+static void co_schedule_bh_cb(void *opaque)
+{
+ AioContext *ctx = opaque;
+ QSLIST_HEAD(, Coroutine) straight, reversed;
+
+ QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
+ QSLIST_INIT(&straight);
+
+ while (!QSLIST_EMPTY(&reversed)) {
+ Coroutine *co = QSLIST_FIRST(&reversed);
+ QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
+ QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
+ }
+
+ while (!QSLIST_EMPTY(&straight)) {
+ Coroutine *co = QSLIST_FIRST(&straight);
+ QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
+ trace_aio_co_schedule_bh_cb(ctx, co);
+ qemu_coroutine_enter(co);
+ }
+}
+
AioContext *aio_context_new(Error **errp)
{
int ret;
@@ -378,6 +405,10 @@ AioContext *aio_context_new(Error **errp)
}
g_source_set_can_recurse(&ctx->source, true);
qemu_lockcnt_init(&ctx->list_lock);
+
+ ctx->co_schedule_bh = aio_bh_new(ctx, co_schedule_bh_cb, ctx);
+ QSLIST_INIT(&ctx->scheduled_coroutines);
+
aio_set_event_notifier(ctx, &ctx->notifier,
false,
(EventNotifierHandler *)
@@ -401,6 +432,40 @@ fail:
return NULL;
}
+void aio_co_schedule(AioContext *ctx, Coroutine *co)
+{
+ trace_aio_co_schedule(ctx, co);
+ QSLIST_INSERT_HEAD_ATOMIC(&ctx->scheduled_coroutines,
+ co, co_scheduled_next);
+ qemu_bh_schedule(ctx->co_schedule_bh);
+}
+
+void aio_co_wake(struct Coroutine *co)
+{
+ AioContext *ctx;
+
+ /* Read coroutine before co->ctx. Matches smp_wmb in
+ * qemu_coroutine_enter.
+ */
+ smp_read_barrier_depends();
+ ctx = atomic_read(&co->ctx);
+
+ if (ctx != qemu_get_current_aio_context()) {
+ aio_co_schedule(ctx, co);
+ return;
+ }
+
+ if (qemu_in_coroutine()) {
+ Coroutine *self = qemu_coroutine_self();
+ assert(self != co);
+ QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
+ } else {
+ aio_context_acquire(ctx);
+ qemu_coroutine_enter(co);
+ aio_context_release(ctx);
+ }
+}
+
void aio_context_ref(AioContext *ctx)
{
g_source_ref(&ctx->source);
diff --git a/include/block/aio.h b/include/block/aio.h
index 7df271d..614cbc6 100644
--- a/include/block/aio.h
+++ b/include/block/aio.h
@@ -47,6 +47,7 @@ typedef void QEMUBHFunc(void *opaque);
typedef bool AioPollFn(void *opaque);
typedef void IOHandler(void *opaque);
+struct Coroutine;
struct ThreadPool;
struct LinuxAioState;
@@ -108,6 +109,9 @@ struct AioContext {
bool notified;
EventNotifier notifier;
+ QSLIST_HEAD(, Coroutine) scheduled_coroutines;
+ QEMUBH *co_schedule_bh;
+
/* Thread pool for performing work and receiving completion callbacks.
* Has its own locking.
*/
@@ -483,6 +487,34 @@ static inline bool aio_node_check(AioContext *ctx, bool is_external)
}
/**
+ * aio_co_schedule:
+ * @ctx: the aio context
+ * @co: the coroutine
+ *
+ * Start a coroutine on a remote AioContext.
+ *
+ * The coroutine must not be entered by anyone else while aio_co_schedule()
+ * is active. In addition the coroutine must have yielded unless ctx
+ * is the context in which the coroutine is running (i.e. the value of
+ * qemu_get_current_aio_context() from the coroutine itself).
+ */
+void aio_co_schedule(AioContext *ctx, struct Coroutine *co);
+
+/**
+ * aio_co_wake:
+ * @co: the coroutine
+ *
+ * Restart a coroutine on the AioContext where it was running last, thus
+ * preventing coroutines from jumping from one context to another when they
+ * go to sleep.
+ *
+ * aio_co_wake may be executed either in coroutine or non-coroutine
+ * context. The coroutine must not be entered by anyone else while
+ * aio_co_wake() is active.
+ */
+void aio_co_wake(struct Coroutine *co);
+
+/**
* Return the AioContext whose event loop runs in the current thread.
*
* If called from an IOThread this will be the IOThread's AioContext. If
diff --git a/include/qemu/coroutine_int.h b/include/qemu/coroutine_int.h
index 14d4f1d..1efa356 100644
--- a/include/qemu/coroutine_int.h
+++ b/include/qemu/coroutine_int.h
@@ -40,12 +40,20 @@ struct Coroutine {
CoroutineEntry *entry;
void *entry_arg;
Coroutine *caller;
+
+ /* Only used when the coroutine has terminated. */
QSLIST_ENTRY(Coroutine) pool_next;
size_t locks_held;
- /* Coroutines that should be woken up when we yield or terminate */
+ /* Coroutines that should be woken up when we yield or terminate.
+ * Only used when the coroutine is running.
+ */
QSIMPLEQ_HEAD(, Coroutine) co_queue_wakeup;
+
+ /* Only used when the coroutine is sleeping. */
+ AioContext *ctx;
QSIMPLEQ_ENTRY(Coroutine) co_queue_next;
+ QSLIST_ENTRY(Coroutine) co_scheduled_next;
};
Coroutine *qemu_coroutine_new(void);
diff --git a/tests/Makefile.include b/tests/Makefile.include
index f776404..3b8ed9d 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -45,9 +45,13 @@ check-unit-y += tests/test-visitor-serialization$(EXESUF)
check-unit-y += tests/test-iov$(EXESUF)
gcov-files-test-iov-y = util/iov.c
check-unit-y += tests/test-aio$(EXESUF)
+gcov-files-test-aio-y = async-posix.c
+gcov-files-test-aio-$(CONFIG_WIN32) += aio-win32.c
+gcov-files-test-aio-$(CONFIG_POSIX) += aio-posix.c
+check-unit-y += tests/test-aio-multithread$(EXESUF)
+gcov-files-test-aio-multithread-y = $(gcov-files-test-aio-y)
+gcov-files-test-aio-multithread-y += util/qemu-coroutine.c tests/iothread.c
check-unit-y += tests/test-throttle$(EXESUF)
-gcov-files-test-aio-$(CONFIG_WIN32) = aio-win32.c
-gcov-files-test-aio-$(CONFIG_POSIX) = aio-posix.c
check-unit-y += tests/test-thread-pool$(EXESUF)
gcov-files-test-thread-pool-y = thread-pool.c
gcov-files-test-hbitmap-y = util/hbitmap.c
@@ -476,8 +480,8 @@ test-qapi-obj-y = tests/test-qapi-visit.o tests/test-qapi-types.o \
tests/test-qapi-event.o tests/test-qmp-introspect.o \
$(test-qom-obj-y)
test-crypto-obj-y = $(crypto-obj-y) $(test-qom-obj-y)
-test-io-obj-y = $(io-obj-y) $(test-crypto-obj-y)
-test-block-obj-y = $(block-obj-y) $(test-io-obj-y)
+test-io-obj-y = $(io-obj-y) $(test-crypto-obj-y) $(block-obj-y) tests/iothread.o
+test-block-obj-y = $(test-io-obj-y)
tests/check-qint$(EXESUF): tests/check-qint.o $(test-util-obj-y)
tests/check-qstring$(EXESUF): tests/check-qstring.o $(test-util-obj-y)
@@ -492,6 +496,7 @@ tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
+tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y)
tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y)
tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y)
tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y)
diff --git a/tests/iothread.c b/tests/iothread.c
new file mode 100644
index 0000000..777d9ee
--- /dev/null
+++ b/tests/iothread.c
@@ -0,0 +1,91 @@
+/*
+ * Event loop thread implementation for unit tests
+ *
+ * Copyright Red Hat Inc., 2013, 2016
+ *
+ * Authors:
+ * Stefan Hajnoczi <stefanha@redhat.com>
+ * Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "block/aio.h"
+#include "qemu/main-loop.h"
+#include "qemu/rcu.h"
+#include "iothread.h"
+
+struct IOThread {
+ AioContext *ctx;
+
+ QemuThread thread;
+ QemuMutex init_done_lock;
+ QemuCond init_done_cond; /* is thread initialization done? */
+ bool stopping;
+};
+
+static __thread IOThread *my_iothread;
+
+AioContext *qemu_get_current_aio_context(void)
+{
+ return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
+}
+
+static void *iothread_run(void *opaque)
+{
+ IOThread *iothread = opaque;
+
+ rcu_register_thread();
+
+ my_iothread = iothread;
+ qemu_mutex_lock(&iothread->init_done_lock);
+ iothread->ctx = aio_context_new(&error_abort);
+ qemu_cond_signal(&iothread->init_done_cond);
+ qemu_mutex_unlock(&iothread->init_done_lock);
+
+ while (!atomic_read(&iothread->stopping)) {
+ aio_poll(iothread->ctx, true);
+ }
+
+ rcu_unregister_thread();
+ return NULL;
+}
+
+void iothread_join(IOThread *iothread)
+{
+ iothread->stopping = true;
+ aio_notify(iothread->ctx);
+ qemu_thread_join(&iothread->thread);
+ qemu_cond_destroy(&iothread->init_done_cond);
+ qemu_mutex_destroy(&iothread->init_done_lock);
+ aio_context_unref(iothread->ctx);
+ g_free(iothread);
+}
+
+IOThread *iothread_new(void)
+{
+ IOThread *iothread = g_new0(IOThread, 1);
+
+ qemu_mutex_init(&iothread->init_done_lock);
+ qemu_cond_init(&iothread->init_done_cond);
+ qemu_thread_create(&iothread->thread, NULL, iothread_run,
+ iothread, QEMU_THREAD_JOINABLE);
+
+ /* Wait for initialization to complete */
+ qemu_mutex_lock(&iothread->init_done_lock);
+ while (iothread->ctx == NULL) {
+ qemu_cond_wait(&iothread->init_done_cond,
+ &iothread->init_done_lock);
+ }
+ qemu_mutex_unlock(&iothread->init_done_lock);
+ return iothread;
+}
+
+AioContext *iothread_get_aio_context(IOThread *iothread)
+{
+ return iothread->ctx;
+}
diff --git a/tests/iothread.h b/tests/iothread.h
new file mode 100644
index 0000000..4877cea
--- /dev/null
+++ b/tests/iothread.h
@@ -0,0 +1,25 @@
+/*
+ * Event loop thread implementation for unit tests
+ *
+ * Copyright Red Hat Inc., 2013, 2016
+ *
+ * Authors:
+ * Stefan Hajnoczi <stefanha@redhat.com>
+ * Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+#ifndef TEST_IOTHREAD_H
+#define TEST_IOTHREAD_H
+
+#include "block/aio.h"
+#include "qemu/thread.h"
+
+typedef struct IOThread IOThread;
+
+IOThread *iothread_new(void);
+void iothread_join(IOThread *iothread);
+AioContext *iothread_get_aio_context(IOThread *iothread);
+
+#endif
diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
new file mode 100644
index 0000000..17e81f9
--- /dev/null
+++ b/tests/test-aio-multithread.c
@@ -0,0 +1,213 @@
+/*
+ * AioContext multithreading tests
+ *
+ * Copyright Red Hat, Inc. 2016
+ *
+ * Authors:
+ * Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
+ * See the COPYING.LIB file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include <glib.h>
+#include "block/aio.h"
+#include "qapi/error.h"
+#include "qemu/coroutine.h"
+#include "qemu/thread.h"
+#include "qemu/error-report.h"
+#include "iothread.h"
+
+/* AioContext management */
+
+#define NUM_CONTEXTS 5
+
+static IOThread *threads[NUM_CONTEXTS];
+static AioContext *ctx[NUM_CONTEXTS];
+static __thread int id = -1;
+
+static QemuEvent done_event;
+
+/* Run a function synchronously on a remote iothread. */
+
+typedef struct CtxRunData {
+ QEMUBHFunc *cb;
+ void *arg;
+} CtxRunData;
+
+static void ctx_run_bh_cb(void *opaque)
+{
+ CtxRunData *data = opaque;
+
+ data->cb(data->arg);
+ qemu_event_set(&done_event);
+}
+
+static void ctx_run(int i, QEMUBHFunc *cb, void *opaque)
+{
+ CtxRunData data = {
+ .cb = cb,
+ .arg = opaque
+ };
+
+ qemu_event_reset(&done_event);
+ aio_bh_schedule_oneshot(ctx[i], ctx_run_bh_cb, &data);
+ qemu_event_wait(&done_event);
+}
+
+/* Starting the iothreads. */
+
+static void set_id_cb(void *opaque)
+{
+ int *i = opaque;
+
+ id = *i;
+}
+
+static void create_aio_contexts(void)
+{
+ int i;
+
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ threads[i] = iothread_new();
+ ctx[i] = iothread_get_aio_context(threads[i]);
+ }
+
+ qemu_event_init(&done_event, false);
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ ctx_run(i, set_id_cb, &i);
+ }
+}
+
+/* Stopping the iothreads. */
+
+static void join_aio_contexts(void)
+{
+ int i;
+
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ aio_context_ref(ctx[i]);
+ }
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ iothread_join(threads[i]);
+ }
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ aio_context_unref(ctx[i]);
+ }
+ qemu_event_destroy(&done_event);
+}
+
+/* Basic test for the stuff above. */
+
+static void test_lifecycle(void)
+{
+ create_aio_contexts();
+ join_aio_contexts();
+}
+
+/* aio_co_schedule test. */
+
+static Coroutine *to_schedule[NUM_CONTEXTS];
+
+static bool now_stopping;
+
+static int count_retry;
+static int count_here;
+static int count_other;
+
+static bool schedule_next(int n)
+{
+ Coroutine *co;
+
+ co = atomic_xchg(&to_schedule[n], NULL);
+ if (!co) {
+ atomic_inc(&count_retry);
+ return false;
+ }
+
+ if (n == id) {
+ atomic_inc(&count_here);
+ } else {
+ atomic_inc(&count_other);
+ }
+
+ aio_co_schedule(ctx[n], co);
+ return true;
+}
+
+static void finish_cb(void *opaque)
+{
+ schedule_next(id);
+}
+
+static void test_multi_co_schedule_entry(void *opaque)
+{
+ g_assert(to_schedule[id] == NULL);
+ atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
+
+ while (!atomic_mb_read(&now_stopping)) {
+ int n;
+
+ n = g_test_rand_int_range(0, NUM_CONTEXTS);
+ schedule_next(n);
+ qemu_coroutine_yield();
+
+ g_assert(to_schedule[id] == NULL);
+ atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
+ }
+}
+
+
+static void test_multi_co_schedule(int seconds)
+{
+ int i;
+
+ count_here = count_other = count_retry = 0;
+ now_stopping = false;
+
+ create_aio_contexts();
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ Coroutine *co1 = qemu_coroutine_create(test_multi_co_schedule_entry, NULL);
+ aio_co_schedule(ctx[i], co1);
+ }
+
+ g_usleep(seconds * 1000000);
+
+ atomic_mb_set(&now_stopping, true);
+ for (i = 0; i < NUM_CONTEXTS; i++) {
+ ctx_run(i, finish_cb, NULL);
+ to_schedule[i] = NULL;
+ }
+
+ join_aio_contexts();
+ g_test_message("scheduled %d, queued %d, retry %d, total %d\n",
+ count_other, count_here, count_retry,
+ count_here + count_other + count_retry);
+}
+
+static void test_multi_co_schedule_1(void)
+{
+ test_multi_co_schedule(1);
+}
+
+static void test_multi_co_schedule_10(void)
+{
+ test_multi_co_schedule(10);
+}
+
+/* End of tests. */
+
+int main(int argc, char **argv)
+{
+ init_clocks();
+
+ g_test_init(&argc, &argv, NULL);
+ g_test_add_func("/aio/multi/lifecycle", test_lifecycle);
+ if (g_test_quick()) {
+ g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
+ } else {
+ g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
+ }
+ return g_test_run();
+}
diff --git a/tests/test-vmstate.c b/tests/test-vmstate.c
index d2f529b..cb87997 100644
--- a/tests/test-vmstate.c
+++ b/tests/test-vmstate.c
@@ -33,17 +33,6 @@
static char temp_file[] = "/tmp/vmst.test.XXXXXX";
static int temp_fd;
-/* Fake yield_until_fd_readable() implementation so we don't have to pull the
- * coroutine code as dependency.
- */
-void yield_until_fd_readable(int fd)
-{
- fd_set fds;
- FD_ZERO(&fds);
- FD_SET(fd, &fds);
- select(fd + 1, &fds, NULL, NULL, NULL);
-}
-
/* Duplicate temp_fd and seek to the beginning of the file */
static QEMUFile *open_test_file(bool write)
diff --git a/trace-events b/trace-events
index 1181486..e9e31e9 100644
--- a/trace-events
+++ b/trace-events
@@ -85,6 +85,10 @@ xen_map_cache(uint64_t phys_addr) "want %#"PRIx64
xen_remap_bucket(uint64_t index) "index %#"PRIx64
xen_map_cache_return(void* ptr) "%p"
+# async.c
+aio_co_schedule(void *ctx, void *co) "ctx %p co %p"
+aio_co_schedule_bh_cb(void *ctx, void *co) "ctx %p co %p"
+
# monitor.c
handle_qmp_command(void *mon, const char *cmd_name) "mon %p cmd_name \"%s\""
monitor_protocol_event_handler(uint32_t event, void *qdict) "event=%d data=%p"
diff --git a/util/qemu-coroutine.c b/util/qemu-coroutine.c
index a5d2f6c..415600d 100644
--- a/util/qemu-coroutine.c
+++ b/util/qemu-coroutine.c
@@ -19,6 +19,7 @@
#include "qemu/atomic.h"
#include "qemu/coroutine.h"
#include "qemu/coroutine_int.h"
+#include "block/aio.h"
enum {
POOL_BATCH_SIZE = 64,
@@ -114,6 +115,13 @@ void qemu_coroutine_enter(Coroutine *co)
}
co->caller = self;
+ co->ctx = qemu_get_current_aio_context();
+
+ /* Store co->ctx before anything that stores co. Matches
+ * barrier in aio_co_wake.
+ */
+ smp_wmb();
+
ret = qemu_coroutine_switch(self, co, COROUTINE_ENTER);
qemu_co_queue_run_restart(co);
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-13 13:17 ` [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake Paolo Bonzini
@ 2017-01-16 11:09 ` Fam Zheng
2017-01-16 12:19 ` Paolo Bonzini
2017-01-18 14:33 ` Stefan Hajnoczi
1 sibling, 1 reply; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 11:09 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> aio_co_wake provides the infrastructure to start a coroutine on a "home"
> AioContext. It will be used by CoMutex and CoQueue, so that coroutines
> don't jump from one context to another when they go to sleep on a
> mutex or waitqueue. However, it can also be used as a more efficient
> alternative to one-shot bottom halves, and saves the effort of tracking
> which AioContext a coroutine is running on.
>
> aio_co_schedule is the part of aio_co_wake that starts a coroutine
> on a remove AioContext, but it is also useful to implement e.g.
s/remove/remote/ and maybe s/but/and/ ?
> bdrv_set_aio_context callbacks.
>
> The implementation of aio_co_schedule is based on a lock-free
> multiple-producer, single-consumer queue. The multiple producers use
> cmpxchg to add to a LIFO stack. The consumer (a per-AioContext bottom
> half) grabs all items added so far, inverts the list to make it FIFO,
> and goes through it one item at a time until it's empty. The data
> structure was inspired by OSv, which uses it in the very code we'll
> "port" to QEMU for the thread-safe CoMutex.
>
> Most of the new code is really tests.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> async.c | 65 +++++++++++++
> include/block/aio.h | 32 +++++++
> include/qemu/coroutine_int.h | 10 +-
> tests/Makefile.include | 13 ++-
> tests/iothread.c | 91 ++++++++++++++++++
> tests/iothread.h | 25 +++++
> tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++++
> tests/test-vmstate.c | 11 ---
> trace-events | 4 +
> util/qemu-coroutine.c | 8 ++
> 10 files changed, 456 insertions(+), 16 deletions(-)
> create mode 100644 tests/iothread.c
> create mode 100644 tests/iothread.h
> create mode 100644 tests/test-aio-multithread.c
>
> diff --git a/async.c b/async.c
> index 0d218ab..1338682 100644
> --- a/async.c
> +++ b/async.c
> @@ -30,6 +30,8 @@
> #include "qemu/main-loop.h"
> #include "qemu/atomic.h"
> #include "block/raw-aio.h"
> +#include "trace/generated-tracers.h"
> +#include "qemu/coroutine_int.h"
>
> /***********************************************************/
> /* bottom halves (can be seen as timers which expire ASAP) */
> @@ -274,6 +276,9 @@ aio_ctx_finalize(GSource *source)
> }
> #endif
>
> + assert(QSLIST_EMPTY(&ctx->scheduled_coroutines));
> + qemu_bh_delete(ctx->co_schedule_bh);
> +
> qemu_lockcnt_lock(&ctx->list_lock);
> assert(!qemu_lockcnt_count(&ctx->list_lock));
> while (ctx->first_bh) {
> @@ -363,6 +368,28 @@ static bool event_notifier_poll(void *opaque)
> return atomic_read(&ctx->notified);
> }
>
> +static void co_schedule_bh_cb(void *opaque)
> +{
> + AioContext *ctx = opaque;
> + QSLIST_HEAD(, Coroutine) straight, reversed;
> +
> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
> + QSLIST_INIT(&straight);
Worth special casing 1 element case?
> +
> + while (!QSLIST_EMPTY(&reversed)) {
> + Coroutine *co = QSLIST_FIRST(&reversed);
> + QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
> + QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
> + }
> +
> + while (!QSLIST_EMPTY(&straight)) {
> + Coroutine *co = QSLIST_FIRST(&straight);
> + QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
> + trace_aio_co_schedule_bh_cb(ctx, co);
> + qemu_coroutine_enter(co);
> + }
> +}
> +
> diff --git a/tests/iothread.c b/tests/iothread.c
> new file mode 100644
> index 0000000..777d9ee
> --- /dev/null
> +++ b/tests/iothread.c
> @@ -0,0 +1,91 @@
> +/*
> + * Event loop thread implementation for unit tests
Curious: what is preventing from (perhaps enhancing and then) using the top
iothread.c implementation?
> + *
> + * Copyright Red Hat Inc., 2013, 2016
> + *
> + * Authors:
> + * Stefan Hajnoczi <stefanha@redhat.com>
> + * Paolo Bonzini <pbonzini@redhat.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#include "qemu/osdep.h"
> +#include "qapi/error.h"
> +#include "block/aio.h"
> +#include "qemu/main-loop.h"
> +#include "qemu/rcu.h"
> +#include "iothread.h"
> +
> +struct IOThread {
> + AioContext *ctx;
> +
> + QemuThread thread;
> + QemuMutex init_done_lock;
> + QemuCond init_done_cond; /* is thread initialization done? */
> + bool stopping;
> +};
> +
> +static __thread IOThread *my_iothread;
> +
> +AioContext *qemu_get_current_aio_context(void)
> +{
> + return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
> +}
> +
> +static void *iothread_run(void *opaque)
> +{
> + IOThread *iothread = opaque;
> +
> + rcu_register_thread();
> +
> + my_iothread = iothread;
> + qemu_mutex_lock(&iothread->init_done_lock);
> + iothread->ctx = aio_context_new(&error_abort);
> + qemu_cond_signal(&iothread->init_done_cond);
> + qemu_mutex_unlock(&iothread->init_done_lock);
> +
> + while (!atomic_read(&iothread->stopping)) {
> + aio_poll(iothread->ctx, true);
> + }
> +
> + rcu_unregister_thread();
> + return NULL;
> +}
> +
> +void iothread_join(IOThread *iothread)
> +{
> + iothread->stopping = true;
> + aio_notify(iothread->ctx);
> + qemu_thread_join(&iothread->thread);
> + qemu_cond_destroy(&iothread->init_done_cond);
> + qemu_mutex_destroy(&iothread->init_done_lock);
> + aio_context_unref(iothread->ctx);
> + g_free(iothread);
> +}
> +
> +IOThread *iothread_new(void)
> +{
> + IOThread *iothread = g_new0(IOThread, 1);
> +
> + qemu_mutex_init(&iothread->init_done_lock);
> + qemu_cond_init(&iothread->init_done_cond);
> + qemu_thread_create(&iothread->thread, NULL, iothread_run,
> + iothread, QEMU_THREAD_JOINABLE);
> +
> + /* Wait for initialization to complete */
> + qemu_mutex_lock(&iothread->init_done_lock);
> + while (iothread->ctx == NULL) {
> + qemu_cond_wait(&iothread->init_done_cond,
> + &iothread->init_done_lock);
> + }
> + qemu_mutex_unlock(&iothread->init_done_lock);
> + return iothread;
> +}
> +
> +AioContext *iothread_get_aio_context(IOThread *iothread)
> +{
> + return iothread->ctx;
> +}
> diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
> new file mode 100644
> index 0000000..17e81f9
> --- /dev/null
> +++ b/tests/test-aio-multithread.c
> @@ -0,0 +1,213 @@
> +/*
> + * AioContext multithreading tests
> + *
> + * Copyright Red Hat, Inc. 2016
> + *
> + * Authors:
> + * Paolo Bonzini <pbonzini@redhat.com>
> + *
> + * This work is licensed under the terms of the GNU LGPL, version 2 or later.
> + * See the COPYING.LIB file in the top-level directory.
> + */
> +
> +#include "qemu/osdep.h"
> +#include <glib.h>
> +#include "block/aio.h"
> +#include "qapi/error.h"
> +#include "qemu/coroutine.h"
> +#include "qemu/thread.h"
> +#include "qemu/error-report.h"
> +#include "iothread.h"
> +
> +/* AioContext management */
> +
> +#define NUM_CONTEXTS 5
> +
> +static IOThread *threads[NUM_CONTEXTS];
> +static AioContext *ctx[NUM_CONTEXTS];
> +static __thread int id = -1;
> +
> +static QemuEvent done_event;
> +
> +/* Run a function synchronously on a remote iothread. */
> +
> +typedef struct CtxRunData {
> + QEMUBHFunc *cb;
> + void *arg;
> +} CtxRunData;
> +
> +static void ctx_run_bh_cb(void *opaque)
> +{
> + CtxRunData *data = opaque;
> +
> + data->cb(data->arg);
> + qemu_event_set(&done_event);
> +}
> +
> +static void ctx_run(int i, QEMUBHFunc *cb, void *opaque)
> +{
> + CtxRunData data = {
> + .cb = cb,
> + .arg = opaque
> + };
> +
> + qemu_event_reset(&done_event);
> + aio_bh_schedule_oneshot(ctx[i], ctx_run_bh_cb, &data);
> + qemu_event_wait(&done_event);
> +}
> +
> +/* Starting the iothreads. */
> +
> +static void set_id_cb(void *opaque)
> +{
> + int *i = opaque;
> +
> + id = *i;
> +}
> +
> +static void create_aio_contexts(void)
> +{
> + int i;
> +
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + threads[i] = iothread_new();
> + ctx[i] = iothread_get_aio_context(threads[i]);
> + }
> +
> + qemu_event_init(&done_event, false);
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + ctx_run(i, set_id_cb, &i);
> + }
> +}
> +
> +/* Stopping the iothreads. */
> +
> +static void join_aio_contexts(void)
> +{
> + int i;
> +
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + aio_context_ref(ctx[i]);
> + }
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + iothread_join(threads[i]);
> + }
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + aio_context_unref(ctx[i]);
> + }
> + qemu_event_destroy(&done_event);
> +}
> +
> +/* Basic test for the stuff above. */
> +
> +static void test_lifecycle(void)
> +{
> + create_aio_contexts();
> + join_aio_contexts();
> +}
> +
> +/* aio_co_schedule test. */
> +
> +static Coroutine *to_schedule[NUM_CONTEXTS];
> +
> +static bool now_stopping;
> +
> +static int count_retry;
> +static int count_here;
> +static int count_other;
> +
> +static bool schedule_next(int n)
> +{
> + Coroutine *co;
> +
> + co = atomic_xchg(&to_schedule[n], NULL);
> + if (!co) {
> + atomic_inc(&count_retry);
> + return false;
> + }
> +
> + if (n == id) {
> + atomic_inc(&count_here);
> + } else {
> + atomic_inc(&count_other);
> + }
> +
> + aio_co_schedule(ctx[n], co);
> + return true;
> +}
> +
> +static void finish_cb(void *opaque)
> +{
> + schedule_next(id);
> +}
> +
> +static void test_multi_co_schedule_entry(void *opaque)
> +{
> + g_assert(to_schedule[id] == NULL);
> + atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
> +
> + while (!atomic_mb_read(&now_stopping)) {
> + int n;
> +
> + n = g_test_rand_int_range(0, NUM_CONTEXTS);
> + schedule_next(n);
> + qemu_coroutine_yield();
> +
> + g_assert(to_schedule[id] == NULL);
> + atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
> + }
> +}
> +
> +
> +static void test_multi_co_schedule(int seconds)
> +{
> + int i;
> +
> + count_here = count_other = count_retry = 0;
> + now_stopping = false;
> +
> + create_aio_contexts();
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + Coroutine *co1 = qemu_coroutine_create(test_multi_co_schedule_entry, NULL);
> + aio_co_schedule(ctx[i], co1);
> + }
> +
> + g_usleep(seconds * 1000000);
> +
> + atomic_mb_set(&now_stopping, true);
> + for (i = 0; i < NUM_CONTEXTS; i++) {
> + ctx_run(i, finish_cb, NULL);
> + to_schedule[i] = NULL;
> + }
> +
> + join_aio_contexts();
> + g_test_message("scheduled %d, queued %d, retry %d, total %d\n",
> + count_other, count_here, count_retry,
> + count_here + count_other + count_retry);
> +}
> +
> +static void test_multi_co_schedule_1(void)
> +{
> + test_multi_co_schedule(1);
> +}
> +
> +static void test_multi_co_schedule_10(void)
> +{
> + test_multi_co_schedule(10);
> +}
> +
> +/* End of tests. */
> +
> +int main(int argc, char **argv)
> +{
> + init_clocks();
> +
> + g_test_init(&argc, &argv, NULL);
> + g_test_add_func("/aio/multi/lifecycle", test_lifecycle);
> + if (g_test_quick()) {
> + g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
> + } else {
> + g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
> + }
> + return g_test_run();
> +}
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-16 11:09 ` Fam Zheng
@ 2017-01-16 12:19 ` Paolo Bonzini
2017-01-16 12:44 ` Fam Zheng
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 12:19 UTC (permalink / raw)
To: Fam Zheng; +Cc: qemu-devel, stefanha
On 16/01/2017 12:09, Fam Zheng wrote:
> On Fri, 01/13 14:17, Paolo Bonzini wrote:
>> aio_co_wake provides the infrastructure to start a coroutine on a "home"
>> AioContext. It will be used by CoMutex and CoQueue, so that coroutines
>> don't jump from one context to another when they go to sleep on a
>> mutex or waitqueue. However, it can also be used as a more efficient
>> alternative to one-shot bottom halves, and saves the effort of tracking
>> which AioContext a coroutine is running on.
>>
>> aio_co_schedule is the part of aio_co_wake that starts a coroutine
>> on a remove AioContext, but it is also useful to implement e.g.
>
> s/remove/remote/ and maybe s/but/and/ ?
>
>> bdrv_set_aio_context callbacks.
>>
>> The implementation of aio_co_schedule is based on a lock-free
>> multiple-producer, single-consumer queue. The multiple producers use
>> cmpxchg to add to a LIFO stack. The consumer (a per-AioContext bottom
>> half) grabs all items added so far, inverts the list to make it FIFO,
>> and goes through it one item at a time until it's empty. The data
>> structure was inspired by OSv, which uses it in the very code we'll
>> "port" to QEMU for the thread-safe CoMutex.
>>
>> Most of the new code is really tests.
>>
>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>> ---
>> async.c | 65 +++++++++++++
>> include/block/aio.h | 32 +++++++
>> include/qemu/coroutine_int.h | 10 +-
>> tests/Makefile.include | 13 ++-
>> tests/iothread.c | 91 ++++++++++++++++++
>> tests/iothread.h | 25 +++++
>> tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++++
>> tests/test-vmstate.c | 11 ---
>> trace-events | 4 +
>> util/qemu-coroutine.c | 8 ++
>> 10 files changed, 456 insertions(+), 16 deletions(-)
>> create mode 100644 tests/iothread.c
>> create mode 100644 tests/iothread.h
>> create mode 100644 tests/test-aio-multithread.c
>>
>> diff --git a/async.c b/async.c
>> index 0d218ab..1338682 100644
>> --- a/async.c
>> +++ b/async.c
>> @@ -30,6 +30,8 @@
>> #include "qemu/main-loop.h"
>> #include "qemu/atomic.h"
>> #include "block/raw-aio.h"
>> +#include "trace/generated-tracers.h"
>> +#include "qemu/coroutine_int.h"
>>
>> /***********************************************************/
>> /* bottom halves (can be seen as timers which expire ASAP) */
>> @@ -274,6 +276,9 @@ aio_ctx_finalize(GSource *source)
>> }
>> #endif
>>
>> + assert(QSLIST_EMPTY(&ctx->scheduled_coroutines));
>> + qemu_bh_delete(ctx->co_schedule_bh);
>> +
>> qemu_lockcnt_lock(&ctx->list_lock);
>> assert(!qemu_lockcnt_count(&ctx->list_lock));
>> while (ctx->first_bh) {
>> @@ -363,6 +368,28 @@ static bool event_notifier_poll(void *opaque)
>> return atomic_read(&ctx->notified);
>> }
>>
>> +static void co_schedule_bh_cb(void *opaque)
>> +{
>> + AioContext *ctx = opaque;
>> + QSLIST_HEAD(, Coroutine) straight, reversed;
>> +
>> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
>> + QSLIST_INIT(&straight);
>
> Worth special casing 1 element case?
Sounds like premature optimization; the QSLIST_MOVE_ATOMIC is going to
be pretty expensive anyway. Do you mean something like:
if (QSLIST_EMPTY(&reversed)) {
return;
}
Coroutine *co = QSLIST_FIRST(&reversed);
if (!QSLIST_NEXT(co, co_scheduled_next)) {
straight = reversed;
} else {
do {
...
} while (!QSLIST_EMPTY(&reversed);
}
do {
...
} while (!QSLIST_EMPTY(&straight);
? Looks a but busy. However, removing the QSLIST_EMPTY case and then
using do/while may be a nice middle.
Paolo
>> +
>> + while (!QSLIST_EMPTY(&reversed)) {
>> + Coroutine *co = QSLIST_FIRST(&reversed);
>> + QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
>> + QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
>> + }
>> +
>> + while (!QSLIST_EMPTY(&straight)) {
>> + Coroutine *co = QSLIST_FIRST(&straight);
>> + QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
>> + trace_aio_co_schedule_bh_cb(ctx, co);
>> + qemu_coroutine_enter(co);
>> + }
>> +}
>> +
>> diff --git a/tests/iothread.c b/tests/iothread.c
>> new file mode 100644
>> index 0000000..777d9ee
>> --- /dev/null
>> +++ b/tests/iothread.c
>> @@ -0,0 +1,91 @@
>> +/*
>> + * Event loop thread implementation for unit tests
>
> Curious: what is preventing from (perhaps enhancing and then) using the top
> iothread.c implementation?
Mostly the dependency of iothread.c on QOM. iothread_new is much
simpler than creating a new object, adding it to the QOM tree, calling
user_creatable_complete, etc. A wrapper wouldn't be much smaller than
this file.
Paolo
>> + *
>> + * Copyright Red Hat Inc., 2013, 2016
>> + *
>> + * Authors:
>> + * Stefan Hajnoczi <stefanha@redhat.com>
>> + * Paolo Bonzini <pbonzini@redhat.com>
>> + *
>> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
>> + * See the COPYING file in the top-level directory.
>> + *
>> + */
>> +
>> +#include "qemu/osdep.h"
>> +#include "qapi/error.h"
>> +#include "block/aio.h"
>> +#include "qemu/main-loop.h"
>> +#include "qemu/rcu.h"
>> +#include "iothread.h"
>> +
>> +struct IOThread {
>> + AioContext *ctx;
>> +
>> + QemuThread thread;
>> + QemuMutex init_done_lock;
>> + QemuCond init_done_cond; /* is thread initialization done? */
>> + bool stopping;
>> +};
>> +
>> +static __thread IOThread *my_iothread;
>> +
>> +AioContext *qemu_get_current_aio_context(void)
>> +{
>> + return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
>> +}
>> +
>> +static void *iothread_run(void *opaque)
>> +{
>> + IOThread *iothread = opaque;
>> +
>> + rcu_register_thread();
>> +
>> + my_iothread = iothread;
>> + qemu_mutex_lock(&iothread->init_done_lock);
>> + iothread->ctx = aio_context_new(&error_abort);
>> + qemu_cond_signal(&iothread->init_done_cond);
>> + qemu_mutex_unlock(&iothread->init_done_lock);
>> +
>> + while (!atomic_read(&iothread->stopping)) {
>> + aio_poll(iothread->ctx, true);
>> + }
>> +
>> + rcu_unregister_thread();
>> + return NULL;
>> +}
>> +
>> +void iothread_join(IOThread *iothread)
>> +{
>> + iothread->stopping = true;
>> + aio_notify(iothread->ctx);
>> + qemu_thread_join(&iothread->thread);
>> + qemu_cond_destroy(&iothread->init_done_cond);
>> + qemu_mutex_destroy(&iothread->init_done_lock);
>> + aio_context_unref(iothread->ctx);
>> + g_free(iothread);
>> +}
>> +
>> +IOThread *iothread_new(void)
>> +{
>> + IOThread *iothread = g_new0(IOThread, 1);
>> +
>> + qemu_mutex_init(&iothread->init_done_lock);
>> + qemu_cond_init(&iothread->init_done_cond);
>> + qemu_thread_create(&iothread->thread, NULL, iothread_run,
>> + iothread, QEMU_THREAD_JOINABLE);
>> +
>> + /* Wait for initialization to complete */
>> + qemu_mutex_lock(&iothread->init_done_lock);
>> + while (iothread->ctx == NULL) {
>> + qemu_cond_wait(&iothread->init_done_cond,
>> + &iothread->init_done_lock);
>> + }
>> + qemu_mutex_unlock(&iothread->init_done_lock);
>> + return iothread;
>> +}
>> +
>> +AioContext *iothread_get_aio_context(IOThread *iothread)
>> +{
>> + return iothread->ctx;
>> +}
>> diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
>> new file mode 100644
>> index 0000000..17e81f9
>> --- /dev/null
>> +++ b/tests/test-aio-multithread.c
>> @@ -0,0 +1,213 @@
>> +/*
>> + * AioContext multithreading tests
>> + *
>> + * Copyright Red Hat, Inc. 2016
>> + *
>> + * Authors:
>> + * Paolo Bonzini <pbonzini@redhat.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2 or later.
>> + * See the COPYING.LIB file in the top-level directory.
>> + */
>> +
>> +#include "qemu/osdep.h"
>> +#include <glib.h>
>> +#include "block/aio.h"
>> +#include "qapi/error.h"
>> +#include "qemu/coroutine.h"
>> +#include "qemu/thread.h"
>> +#include "qemu/error-report.h"
>> +#include "iothread.h"
>> +
>> +/* AioContext management */
>> +
>> +#define NUM_CONTEXTS 5
>> +
>> +static IOThread *threads[NUM_CONTEXTS];
>> +static AioContext *ctx[NUM_CONTEXTS];
>> +static __thread int id = -1;
>> +
>> +static QemuEvent done_event;
>> +
>> +/* Run a function synchronously on a remote iothread. */
>> +
>> +typedef struct CtxRunData {
>> + QEMUBHFunc *cb;
>> + void *arg;
>> +} CtxRunData;
>> +
>> +static void ctx_run_bh_cb(void *opaque)
>> +{
>> + CtxRunData *data = opaque;
>> +
>> + data->cb(data->arg);
>> + qemu_event_set(&done_event);
>> +}
>> +
>> +static void ctx_run(int i, QEMUBHFunc *cb, void *opaque)
>> +{
>> + CtxRunData data = {
>> + .cb = cb,
>> + .arg = opaque
>> + };
>> +
>> + qemu_event_reset(&done_event);
>> + aio_bh_schedule_oneshot(ctx[i], ctx_run_bh_cb, &data);
>> + qemu_event_wait(&done_event);
>> +}
>> +
>> +/* Starting the iothreads. */
>> +
>> +static void set_id_cb(void *opaque)
>> +{
>> + int *i = opaque;
>> +
>> + id = *i;
>> +}
>> +
>> +static void create_aio_contexts(void)
>> +{
>> + int i;
>> +
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + threads[i] = iothread_new();
>> + ctx[i] = iothread_get_aio_context(threads[i]);
>> + }
>> +
>> + qemu_event_init(&done_event, false);
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + ctx_run(i, set_id_cb, &i);
>> + }
>> +}
>> +
>> +/* Stopping the iothreads. */
>> +
>> +static void join_aio_contexts(void)
>> +{
>> + int i;
>> +
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + aio_context_ref(ctx[i]);
>> + }
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + iothread_join(threads[i]);
>> + }
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + aio_context_unref(ctx[i]);
>> + }
>> + qemu_event_destroy(&done_event);
>> +}
>> +
>> +/* Basic test for the stuff above. */
>> +
>> +static void test_lifecycle(void)
>> +{
>> + create_aio_contexts();
>> + join_aio_contexts();
>> +}
>> +
>> +/* aio_co_schedule test. */
>> +
>> +static Coroutine *to_schedule[NUM_CONTEXTS];
>> +
>> +static bool now_stopping;
>> +
>> +static int count_retry;
>> +static int count_here;
>> +static int count_other;
>> +
>> +static bool schedule_next(int n)
>> +{
>> + Coroutine *co;
>> +
>> + co = atomic_xchg(&to_schedule[n], NULL);
>> + if (!co) {
>> + atomic_inc(&count_retry);
>> + return false;
>> + }
>> +
>> + if (n == id) {
>> + atomic_inc(&count_here);
>> + } else {
>> + atomic_inc(&count_other);
>> + }
>> +
>> + aio_co_schedule(ctx[n], co);
>> + return true;
>> +}
>> +
>> +static void finish_cb(void *opaque)
>> +{
>> + schedule_next(id);
>> +}
>> +
>> +static void test_multi_co_schedule_entry(void *opaque)
>> +{
>> + g_assert(to_schedule[id] == NULL);
>> + atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
>> +
>> + while (!atomic_mb_read(&now_stopping)) {
>> + int n;
>> +
>> + n = g_test_rand_int_range(0, NUM_CONTEXTS);
>> + schedule_next(n);
>> + qemu_coroutine_yield();
>> +
>> + g_assert(to_schedule[id] == NULL);
>> + atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
>> + }
>> +}
>> +
>> +
>> +static void test_multi_co_schedule(int seconds)
>> +{
>> + int i;
>> +
>> + count_here = count_other = count_retry = 0;
>> + now_stopping = false;
>> +
>> + create_aio_contexts();
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + Coroutine *co1 = qemu_coroutine_create(test_multi_co_schedule_entry, NULL);
>> + aio_co_schedule(ctx[i], co1);
>> + }
>> +
>> + g_usleep(seconds * 1000000);
>> +
>> + atomic_mb_set(&now_stopping, true);
>> + for (i = 0; i < NUM_CONTEXTS; i++) {
>> + ctx_run(i, finish_cb, NULL);
>> + to_schedule[i] = NULL;
>> + }
>> +
>> + join_aio_contexts();
>> + g_test_message("scheduled %d, queued %d, retry %d, total %d\n",
>> + count_other, count_here, count_retry,
>> + count_here + count_other + count_retry);
>> +}
>> +
>> +static void test_multi_co_schedule_1(void)
>> +{
>> + test_multi_co_schedule(1);
>> +}
>> +
>> +static void test_multi_co_schedule_10(void)
>> +{
>> + test_multi_co_schedule(10);
>> +}
>> +
>> +/* End of tests. */
>> +
>> +int main(int argc, char **argv)
>> +{
>> + init_clocks();
>> +
>> + g_test_init(&argc, &argv, NULL);
>> + g_test_add_func("/aio/multi/lifecycle", test_lifecycle);
>> + if (g_test_quick()) {
>> + g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
>> + } else {
>> + g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
>> + }
>> + return g_test_run();
>> +}
>
> Fam
>
>
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-16 12:19 ` Paolo Bonzini
@ 2017-01-16 12:44 ` Fam Zheng
0 siblings, 0 replies; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 12:44 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Mon, 01/16 13:19, Paolo Bonzini wrote:
> >> +static void co_schedule_bh_cb(void *opaque)
> >> +{
> >> + AioContext *ctx = opaque;
> >> + QSLIST_HEAD(, Coroutine) straight, reversed;
> >> +
> >> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
> >> + QSLIST_INIT(&straight);
> >
> > Worth special casing 1 element case?
>
> Sounds like premature optimization; the QSLIST_MOVE_ATOMIC is going to
> be pretty expensive anyway. Do you mean something like:
>
> if (QSLIST_EMPTY(&reversed)) {
> return;
> }
> Coroutine *co = QSLIST_FIRST(&reversed);
> if (!QSLIST_NEXT(co, co_scheduled_next)) {
> straight = reversed;
> } else {
> do {
> ...
> } while (!QSLIST_EMPTY(&reversed);
> }
>
> do {
> ...
> } while (!QSLIST_EMPTY(&straight);
>
> ? Looks a but busy. However, removing the QSLIST_EMPTY case and then
> using do/while may be a nice middle.
I think QSLIST_EMPTY is very unusual. I don't know if these are premature or
not, just asked because the !QSLIST_NEXT() case will be the most common one.
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-13 13:17 ` [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake Paolo Bonzini
2017-01-16 11:09 ` Fam Zheng
@ 2017-01-18 14:33 ` Stefan Hajnoczi
2017-01-18 15:40 ` Paolo Bonzini
1 sibling, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:33 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 1690 bytes --]
On Fri, Jan 13, 2017 at 02:17:16PM +0100, Paolo Bonzini wrote:
> +static void co_schedule_bh_cb(void *opaque)
> +{
> + AioContext *ctx = opaque;
> + QSLIST_HEAD(, Coroutine) straight, reversed;
> +
> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
> + QSLIST_INIT(&straight);
> +
> + while (!QSLIST_EMPTY(&reversed)) {
> + Coroutine *co = QSLIST_FIRST(&reversed);
> + QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
> + QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
> + }
> +
> + while (!QSLIST_EMPTY(&straight)) {
> + Coroutine *co = QSLIST_FIRST(&straight);
> + QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
> + trace_aio_co_schedule_bh_cb(ctx, co);
> + qemu_coroutine_enter(co);
> + }
> +}
ctx->scheduled_coroutines is a specialized CoQueue. Was there no way to
modify and then use CoQueue instead of open coding it?
> +void aio_co_wake(struct Coroutine *co)
> +{
> + AioContext *ctx;
> +
> + /* Read coroutine before co->ctx. Matches smp_wmb in
> + * qemu_coroutine_enter.
> + */
> + smp_read_barrier_depends();
> + ctx = atomic_read(&co->ctx);
> +
> + if (ctx != qemu_get_current_aio_context()) {
> + aio_co_schedule(ctx, co);
> + return;
> + }
> +
> + if (qemu_in_coroutine()) {
> + Coroutine *self = qemu_coroutine_self();
> + assert(self != co);
> + QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
> + } else {
> + aio_context_acquire(ctx);
> + qemu_coroutine_enter(co);
> + aio_context_release(ctx);
Why is it necessary to acquire AioContext here? We're already in ctx.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-18 14:33 ` Stefan Hajnoczi
@ 2017-01-18 15:40 ` Paolo Bonzini
2017-01-19 16:49 ` Stefan Hajnoczi
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-18 15:40 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: qemu-devel, famz
On 18/01/2017 15:33, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 02:17:16PM +0100, Paolo Bonzini wrote:
>> +static void co_schedule_bh_cb(void *opaque)
>> +{
>> + AioContext *ctx = opaque;
>> + QSLIST_HEAD(, Coroutine) straight, reversed;
>> +
>> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
>> + QSLIST_INIT(&straight);
>> +
>> + while (!QSLIST_EMPTY(&reversed)) {
>> + Coroutine *co = QSLIST_FIRST(&reversed);
>> + QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
>> + QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
>> + }
>> +
>> + while (!QSLIST_EMPTY(&straight)) {
>> + Coroutine *co = QSLIST_FIRST(&straight);
>> + QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
>> + trace_aio_co_schedule_bh_cb(ctx, co);
>> + qemu_coroutine_enter(co);
>> + }
>> +}
>
> ctx->scheduled_coroutines is a specialized CoQueue. Was there no way to
> modify and then use CoQueue instead of open coding it?
First of all, I'm trying to avoid a circular dependency when CoQueue can
use aio_co_schedule (indirectly through aio_co_wake) after patch 7.
Secondarily, co_schedule_bh_cb can perform a single pass on
ctx->scheduled_coroutines because it will be rescheduled by
aio_co_schedule. The same is not true for qemu_co_queue_restart_all.
Also, CoQueue can have multiple consumers, while scheduled_coroutines
cannot. Currently, CoQueue needs no thread-safety because it's
protected by AioContext and/or by the non-preemptive nature of
coroutines. Later, it's going to be protected by an external CoMutex,
just like a mutex/condvar pair. scheduled_coroutines is different in
this respect.
>> +void aio_co_wake(struct Coroutine *co)
>> +{
>> + AioContext *ctx;
>> +
>> + /* Read coroutine before co->ctx. Matches smp_wmb in
>> + * qemu_coroutine_enter.
>> + */
>> + smp_read_barrier_depends();
>> + ctx = atomic_read(&co->ctx);
>> +
>> + if (ctx != qemu_get_current_aio_context()) {
>> + aio_co_schedule(ctx, co);
>> + return;
>> + }
>> +
>> + if (qemu_in_coroutine()) {
>> + Coroutine *self = qemu_coroutine_self();
>> + assert(self != co);
>> + QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
>> + } else {
>> + aio_context_acquire(ctx);
>> + qemu_coroutine_enter(co);
>> + aio_context_release(ctx);
>
> Why is it necessary to acquire AioContext here? We're already in ctx.
We're in its thread, but we've not necessarily acquired it yet.
aio_co_wake is called "aio_*" because it's a central place for
AioContext to acquire itself for coroutines. This way, coroutines only
care about CoMutexes, and not about AioContext. This was "highly
recommended" :) by Kevin last year and it's the main change since the
previous posting
(https://lists.gnu.org/archive/html/qemu-devel/2015-11/msg05416.html for
example).
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake
2017-01-18 15:40 ` Paolo Bonzini
@ 2017-01-19 16:49 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-19 16:49 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 3329 bytes --]
On Wed, Jan 18, 2017 at 04:40:29PM +0100, Paolo Bonzini wrote:
> On 18/01/2017 15:33, Stefan Hajnoczi wrote:
> > On Fri, Jan 13, 2017 at 02:17:16PM +0100, Paolo Bonzini wrote:
> >> +static void co_schedule_bh_cb(void *opaque)
> >> +{
> >> + AioContext *ctx = opaque;
> >> + QSLIST_HEAD(, Coroutine) straight, reversed;
> >> +
> >> + QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
> >> + QSLIST_INIT(&straight);
> >> +
> >> + while (!QSLIST_EMPTY(&reversed)) {
> >> + Coroutine *co = QSLIST_FIRST(&reversed);
> >> + QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
> >> + QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
> >> + }
> >> +
> >> + while (!QSLIST_EMPTY(&straight)) {
> >> + Coroutine *co = QSLIST_FIRST(&straight);
> >> + QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
> >> + trace_aio_co_schedule_bh_cb(ctx, co);
> >> + qemu_coroutine_enter(co);
> >> + }
> >> +}
> >
> > ctx->scheduled_coroutines is a specialized CoQueue. Was there no way to
> > modify and then use CoQueue instead of open coding it?
>
> First of all, I'm trying to avoid a circular dependency when CoQueue can
> use aio_co_schedule (indirectly through aio_co_wake) after patch 7.
>
> Secondarily, co_schedule_bh_cb can perform a single pass on
> ctx->scheduled_coroutines because it will be rescheduled by
> aio_co_schedule. The same is not true for qemu_co_queue_restart_all.
>
> Also, CoQueue can have multiple consumers, while scheduled_coroutines
> cannot. Currently, CoQueue needs no thread-safety because it's
> protected by AioContext and/or by the non-preemptive nature of
> coroutines. Later, it's going to be protected by an external CoMutex,
> just like a mutex/condvar pair. scheduled_coroutines is different in
> this respect.
Thanks for explaining.
> >> +void aio_co_wake(struct Coroutine *co)
> >> +{
> >> + AioContext *ctx;
> >> +
> >> + /* Read coroutine before co->ctx. Matches smp_wmb in
> >> + * qemu_coroutine_enter.
> >> + */
> >> + smp_read_barrier_depends();
> >> + ctx = atomic_read(&co->ctx);
> >> +
> >> + if (ctx != qemu_get_current_aio_context()) {
> >> + aio_co_schedule(ctx, co);
> >> + return;
> >> + }
> >> +
> >> + if (qemu_in_coroutine()) {
> >> + Coroutine *self = qemu_coroutine_self();
> >> + assert(self != co);
> >> + QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
> >> + } else {
> >> + aio_context_acquire(ctx);
> >> + qemu_coroutine_enter(co);
> >> + aio_context_release(ctx);
> >
> > Why is it necessary to acquire AioContext here? We're already in ctx.
>
> We're in its thread, but we've not necessarily acquired it yet.
> aio_co_wake is called "aio_*" because it's a central place for
> AioContext to acquire itself for coroutines. This way, coroutines only
> care about CoMutexes, and not about AioContext. This was "highly
> recommended" :) by Kevin last year and it's the main change since the
> previous posting
> (https://lists.gnu.org/archive/html/qemu-devel/2015-11/msg05416.html for
> example).
I see. That also explains why we don't need to acquire AioContext if
we're already in a coroutine.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 02/16] block-backend: allow blk_prw from coroutine context
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
2017-01-13 13:17 ` [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 14:35 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 03/16] test-thread-pool: use generic AioContext infrastructure Paolo Bonzini
` (16 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
qcow2_create2 calls this. Do not run a nested event loop, as that
breaks when aio_co_wake tries to queue the coroutine on the co_queue_wakeup
list of the currently running one.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
block/block-backend.c | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)
diff --git a/block/block-backend.c b/block/block-backend.c
index efbf398..1177598 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -880,7 +880,6 @@ static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
{
QEMUIOVector qiov;
struct iovec iov;
- Coroutine *co;
BlkRwCo rwco;
iov = (struct iovec) {
@@ -897,9 +896,14 @@ static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
.ret = NOT_DONE,
};
- co = qemu_coroutine_create(co_entry, &rwco);
- qemu_coroutine_enter(co);
- BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
+ if (qemu_in_coroutine()) {
+ /* Fast-path if already in coroutine context */
+ co_entry(&rwco);
+ } else {
+ Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
+ qemu_coroutine_enter(co);
+ BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
+ }
return rwco.ret;
}
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 02/16] block-backend: allow blk_prw from coroutine context
2017-01-13 13:17 ` [Qemu-devel] [PATCH 02/16] block-backend: allow blk_prw from coroutine context Paolo Bonzini
@ 2017-01-18 14:35 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:35 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 466 bytes --]
On Fri, Jan 13, 2017 at 02:17:17PM +0100, Paolo Bonzini wrote:
> qcow2_create2 calls this. Do not run a nested event loop, as that
> breaks when aio_co_wake tries to queue the coroutine on the co_queue_wakeup
> list of the currently running one.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> block/block-backend.c | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 03/16] test-thread-pool: use generic AioContext infrastructure
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
2017-01-13 13:17 ` [Qemu-devel] [PATCH 01/16] aio: introduce aio_co_schedule and aio_co_wake Paolo Bonzini
2017-01-13 13:17 ` [Qemu-devel] [PATCH 02/16] block-backend: allow blk_prw from coroutine context Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 14:35 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
` (15 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Once the thread pool starts using aio_co_wake, it will also need
qemu_get_current_aio_context(). Make test-thread-pool create
an AioContext with qemu_init_main_loop, so that stubs/iothread.c
and tests/iothread.c can provide the rest.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
tests/test-thread-pool.c | 12 +++---------
1 file changed, 3 insertions(+), 9 deletions(-)
diff --git a/tests/test-thread-pool.c b/tests/test-thread-pool.c
index 8dbf66a..91b4ec5 100644
--- a/tests/test-thread-pool.c
+++ b/tests/test-thread-pool.c
@@ -6,6 +6,7 @@
#include "qapi/error.h"
#include "qemu/timer.h"
#include "qemu/error-report.h"
+#include "qemu/main-loop.h"
static AioContext *ctx;
static ThreadPool *pool;
@@ -224,15 +225,9 @@ static void test_cancel_async(void)
int main(int argc, char **argv)
{
int ret;
- Error *local_error = NULL;
- init_clocks();
-
- ctx = aio_context_new(&local_error);
- if (!ctx) {
- error_reportf_err(local_error, "Failed to create AIO Context: ");
- exit(1);
- }
+ qemu_init_main_loop(&error_abort);
+ ctx = qemu_get_current_aio_context();
pool = aio_get_thread_pool(ctx);
g_test_init(&argc, &argv, NULL);
@@ -245,6 +240,5 @@ int main(int argc, char **argv)
ret = g_test_run();
- aio_context_unref(ctx);
return ret;
}
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 03/16] test-thread-pool: use generic AioContext infrastructure
2017-01-13 13:17 ` [Qemu-devel] [PATCH 03/16] test-thread-pool: use generic AioContext infrastructure Paolo Bonzini
@ 2017-01-18 14:35 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:35 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 529 bytes --]
On Fri, Jan 13, 2017 at 02:17:18PM +0100, Paolo Bonzini wrote:
> Once the thread pool starts using aio_co_wake, it will also need
> qemu_get_current_aio_context(). Make test-thread-pool create
> an AioContext with qemu_init_main_loop, so that stubs/iothread.c
> and tests/iothread.c can provide the rest.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> tests/test-thread-pool.c | 12 +++---------
> 1 file changed, 3 insertions(+), 9 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (2 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 03/16] test-thread-pool: use generic AioContext infrastructure Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 11:31 ` Fam Zheng
` (3 more replies)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts Paolo Bonzini
` (14 subsequent siblings)
18 siblings, 4 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
This is in preparation for making qio_channel_yield work on
AioContexts other than the main one.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
include/io/channel.h | 30 ++++++++++++++++++++++++++++++
io/channel-command.c | 13 +++++++++++++
io/channel-file.c | 11 +++++++++++
io/channel-socket.c | 16 +++++++++++-----
io/channel-tls.c | 12 ++++++++++++
io/channel-watch.c | 6 ++++++
io/channel.c | 11 +++++++++++
7 files changed, 94 insertions(+), 5 deletions(-)
diff --git a/include/io/channel.h b/include/io/channel.h
index 32a9470..665edd7 100644
--- a/include/io/channel.h
+++ b/include/io/channel.h
@@ -23,6 +23,7 @@
#include "qemu-common.h"
#include "qom/object.h"
+#include "block/aio.h"
#define TYPE_QIO_CHANNEL "qio-channel"
#define QIO_CHANNEL(obj) \
@@ -58,6 +59,8 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
GIOCondition condition,
gpointer data);
+typedef struct QIOChannelRestart QIOChannelRestart;
+
/**
* QIOChannel:
*
@@ -80,6 +83,9 @@ struct QIOChannel {
Object parent;
unsigned int features; /* bitmask of QIOChannelFeatures */
char *name;
+ AioContext *ctx;
+ QIOChannelRestart *read_coroutine;
+ QIOChannelRestart *write_coroutine;
#ifdef _WIN32
HANDLE event; /* For use with GSource on Win32 */
#endif
@@ -132,6 +138,11 @@ struct QIOChannelClass {
off_t offset,
int whence,
Error **errp);
+ void (*io_set_aio_fd_handler)(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque);
};
/* General I/O handling functions */
@@ -525,4 +536,23 @@ void qio_channel_yield(QIOChannel *ioc,
void qio_channel_wait(QIOChannel *ioc,
GIOCondition condition);
+/**
+ * qio_channel_set_aio_fd_handler:
+ * @ioc: the channel object
+ * @ctx: the AioContext to set the handlers on
+ * @io_read: the read handler
+ * @io_write: the write handler
+ * @opaque: the opaque value passed to the handler
+ *
+ * This is used internally by qio_channel_yield(). It can
+ * be used by channel implementations to forward the handlers
+ * to another channel (e.g. from #QIOChannelTLS to the
+ * underlying socket).
+ */
+void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque);
+
#endif /* QIO_CHANNEL_H */
diff --git a/io/channel-command.c b/io/channel-command.c
index ad25313..4000b61 100644
--- a/io/channel-command.c
+++ b/io/channel-command.c
@@ -328,6 +328,18 @@ static int qio_channel_command_close(QIOChannel *ioc,
}
+static void qio_channel_command_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque)
+{
+ QIOChannelCommand *cioc = QIO_CHANNEL_COMMAND(ioc);
+ aio_set_fd_handler(ctx, cioc->readfd, false, io_read, NULL, NULL, opaque);
+ aio_set_fd_handler(ctx, cioc->writefd, false, NULL, io_write, NULL, opaque);
+}
+
+
static GSource *qio_channel_command_create_watch(QIOChannel *ioc,
GIOCondition condition)
{
@@ -349,6 +361,7 @@ static void qio_channel_command_class_init(ObjectClass *klass,
ioc_klass->io_set_blocking = qio_channel_command_set_blocking;
ioc_klass->io_close = qio_channel_command_close;
ioc_klass->io_create_watch = qio_channel_command_create_watch;
+ ioc_klass->io_set_aio_fd_handler = qio_channel_command_set_aio_fd_handler;
}
static const TypeInfo qio_channel_command_info = {
diff --git a/io/channel-file.c b/io/channel-file.c
index e1da243..b383273 100644
--- a/io/channel-file.c
+++ b/io/channel-file.c
@@ -186,6 +186,16 @@ static int qio_channel_file_close(QIOChannel *ioc,
}
+static void qio_channel_file_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque)
+{
+ QIOChannelFile *fioc = QIO_CHANNEL_FILE(ioc);
+ aio_set_fd_handler(ctx, fioc->fd, false, io_read, io_write, NULL, opaque);
+}
+
static GSource *qio_channel_file_create_watch(QIOChannel *ioc,
GIOCondition condition)
{
@@ -206,6 +216,7 @@ static void qio_channel_file_class_init(ObjectClass *klass,
ioc_klass->io_seek = qio_channel_file_seek;
ioc_klass->io_close = qio_channel_file_close;
ioc_klass->io_create_watch = qio_channel_file_create_watch;
+ ioc_klass->io_set_aio_fd_handler = qio_channel_file_set_aio_fd_handler;
}
static const TypeInfo qio_channel_file_info = {
diff --git a/io/channel-socket.c b/io/channel-socket.c
index d7e03f6..3909f65 100644
--- a/io/channel-socket.c
+++ b/io/channel-socket.c
@@ -661,11 +661,6 @@ qio_channel_socket_set_blocking(QIOChannel *ioc,
qemu_set_block(sioc->fd);
} else {
qemu_set_nonblock(sioc->fd);
-#ifdef WIN32
- WSAEventSelect(sioc->fd, ioc->event,
- FD_READ | FD_ACCEPT | FD_CLOSE |
- FD_CONNECT | FD_WRITE | FD_OOB);
-#endif
}
return 0;
}
@@ -745,6 +740,16 @@ qio_channel_socket_shutdown(QIOChannel *ioc,
return 0;
}
+static void qio_channel_socket_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque)
+{
+ QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
+ aio_set_fd_handler(ctx, sioc->fd, false, io_read, io_write, NULL, opaque);
+}
+
static GSource *qio_channel_socket_create_watch(QIOChannel *ioc,
GIOCondition condition)
{
@@ -767,6 +772,7 @@ static void qio_channel_socket_class_init(ObjectClass *klass,
ioc_klass->io_set_cork = qio_channel_socket_set_cork;
ioc_klass->io_set_delay = qio_channel_socket_set_delay;
ioc_klass->io_create_watch = qio_channel_socket_create_watch;
+ ioc_klass->io_set_aio_fd_handler = qio_channel_socket_set_aio_fd_handler;
}
static const TypeInfo qio_channel_socket_info = {
diff --git a/io/channel-tls.c b/io/channel-tls.c
index d24dc8c..fa3f93e 100644
--- a/io/channel-tls.c
+++ b/io/channel-tls.c
@@ -349,6 +349,17 @@ static int qio_channel_tls_close(QIOChannel *ioc,
return qio_channel_close(tioc->master, errp);
}
+static void qio_channel_tls_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque)
+{
+ QIOChannelTLS *tioc = QIO_CHANNEL_TLS(ioc);
+
+ qio_channel_set_aio_fd_handler(tioc->master, ctx, io_read, io_write, opaque);
+}
+
static GSource *qio_channel_tls_create_watch(QIOChannel *ioc,
GIOCondition condition)
{
@@ -376,6 +387,7 @@ static void qio_channel_tls_class_init(ObjectClass *klass,
ioc_klass->io_close = qio_channel_tls_close;
ioc_klass->io_shutdown = qio_channel_tls_shutdown;
ioc_klass->io_create_watch = qio_channel_tls_create_watch;
+ ioc_klass->io_set_aio_fd_handler = qio_channel_tls_set_aio_fd_handler;
}
static const TypeInfo qio_channel_tls_info = {
diff --git a/io/channel-watch.c b/io/channel-watch.c
index cf1cdff..8640d1c 100644
--- a/io/channel-watch.c
+++ b/io/channel-watch.c
@@ -285,6 +285,12 @@ GSource *qio_channel_create_socket_watch(QIOChannel *ioc,
GSource *source;
QIOChannelSocketSource *ssource;
+#ifdef WIN32
+ WSAEventSelect(socket, ioc->event,
+ FD_READ | FD_ACCEPT | FD_CLOSE |
+ FD_CONNECT | FD_WRITE | FD_OOB);
+#endif
+
source = g_source_new(&qio_channel_socket_source_funcs,
sizeof(QIOChannelSocketSource));
ssource = (QIOChannelSocketSource *)source;
diff --git a/io/channel.c b/io/channel.c
index 80924c1..ce470d7 100644
--- a/io/channel.c
+++ b/io/channel.c
@@ -154,6 +154,17 @@ GSource *qio_channel_create_watch(QIOChannel *ioc,
}
+void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
+ AioContext *ctx,
+ IOHandler *io_read,
+ IOHandler *io_write,
+ void *opaque)
+{
+ QIOChannelClass *klass = QIO_CHANNEL_GET_CLASS(ioc);
+
+ klass->io_set_aio_fd_handler(ioc, ctx, io_read, io_write, opaque);
+}
+
guint qio_channel_add_watch(QIOChannel *ioc,
GIOCondition condition,
QIOChannelFunc func,
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
@ 2017-01-16 11:31 ` Fam Zheng
2017-01-16 12:52 ` Daniel P. Berrange
` (2 subsequent siblings)
3 siblings, 0 replies; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 11:31 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> This is in preparation for making qio_channel_yield work on
> AioContexts other than the main one.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> include/io/channel.h | 30 ++++++++++++++++++++++++++++++
> io/channel-command.c | 13 +++++++++++++
> io/channel-file.c | 11 +++++++++++
> io/channel-socket.c | 16 +++++++++++-----
> io/channel-tls.c | 12 ++++++++++++
> io/channel-watch.c | 6 ++++++
> io/channel.c | 11 +++++++++++
> 7 files changed, 94 insertions(+), 5 deletions(-)
>
> diff --git a/include/io/channel.h b/include/io/channel.h
> index 32a9470..665edd7 100644
> --- a/include/io/channel.h
> +++ b/include/io/channel.h
> @@ -23,6 +23,7 @@
>
> #include "qemu-common.h"
> #include "qom/object.h"
> +#include "block/aio.h"
>
> #define TYPE_QIO_CHANNEL "qio-channel"
> #define QIO_CHANNEL(obj) \
> @@ -58,6 +59,8 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
> GIOCondition condition,
> gpointer data);
>
> +typedef struct QIOChannelRestart QIOChannelRestart;
> +
> /**
> * QIOChannel:
> *
> @@ -80,6 +83,9 @@ struct QIOChannel {
> Object parent;
> unsigned int features; /* bitmask of QIOChannelFeatures */
> char *name;
> + AioContext *ctx;
> + QIOChannelRestart *read_coroutine;
> + QIOChannelRestart *write_coroutine;
> #ifdef _WIN32
> HANDLE event; /* For use with GSource on Win32 */
> #endif
> @@ -132,6 +138,11 @@ struct QIOChannelClass {
> off_t offset,
> int whence,
> Error **errp);
> + void (*io_set_aio_fd_handler)(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque);
> };
>
> /* General I/O handling functions */
> @@ -525,4 +536,23 @@ void qio_channel_yield(QIOChannel *ioc,
> void qio_channel_wait(QIOChannel *ioc,
> GIOCondition condition);
>
> +/**
> + * qio_channel_set_aio_fd_handler:
> + * @ioc: the channel object
> + * @ctx: the AioContext to set the handlers on
> + * @io_read: the read handler
> + * @io_write: the write handler
> + * @opaque: the opaque value passed to the handler
> + *
> + * This is used internally by qio_channel_yield(). It can
> + * be used by channel implementations to forward the handlers
> + * to another channel (e.g. from #QIOChannelTLS to the
> + * underlying socket).
> + */
> +void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque);
> +
> #endif /* QIO_CHANNEL_H */
> diff --git a/io/channel-command.c b/io/channel-command.c
> index ad25313..4000b61 100644
> --- a/io/channel-command.c
> +++ b/io/channel-command.c
> @@ -328,6 +328,18 @@ static int qio_channel_command_close(QIOChannel *ioc,
> }
>
>
> +static void qio_channel_command_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
Alignment is a bit off.
> +{
> + QIOChannelCommand *cioc = QIO_CHANNEL_COMMAND(ioc);
> + aio_set_fd_handler(ctx, cioc->readfd, false, io_read, NULL, NULL, opaque);
> + aio_set_fd_handler(ctx, cioc->writefd, false, NULL, io_write, NULL, opaque);
> +}
> +
> +
> static GSource *qio_channel_command_create_watch(QIOChannel *ioc,
> GIOCondition condition)
> {
> @@ -349,6 +361,7 @@ static void qio_channel_command_class_init(ObjectClass *klass,
> ioc_klass->io_set_blocking = qio_channel_command_set_blocking;
> ioc_klass->io_close = qio_channel_command_close;
> ioc_klass->io_create_watch = qio_channel_command_create_watch;
> + ioc_klass->io_set_aio_fd_handler = qio_channel_command_set_aio_fd_handler;
> }
>
> static const TypeInfo qio_channel_command_info = {
> diff --git a/io/channel-file.c b/io/channel-file.c
> index e1da243..b383273 100644
> --- a/io/channel-file.c
> +++ b/io/channel-file.c
> @@ -186,6 +186,16 @@ static int qio_channel_file_close(QIOChannel *ioc,
> }
>
>
> +static void qio_channel_file_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
> +{
> + QIOChannelFile *fioc = QIO_CHANNEL_FILE(ioc);
> + aio_set_fd_handler(ctx, fioc->fd, false, io_read, io_write, NULL, opaque);
> +}
> +
> static GSource *qio_channel_file_create_watch(QIOChannel *ioc,
> GIOCondition condition)
> {
> @@ -206,6 +216,7 @@ static void qio_channel_file_class_init(ObjectClass *klass,
> ioc_klass->io_seek = qio_channel_file_seek;
> ioc_klass->io_close = qio_channel_file_close;
> ioc_klass->io_create_watch = qio_channel_file_create_watch;
> + ioc_klass->io_set_aio_fd_handler = qio_channel_file_set_aio_fd_handler;
> }
>
> static const TypeInfo qio_channel_file_info = {
> diff --git a/io/channel-socket.c b/io/channel-socket.c
> index d7e03f6..3909f65 100644
> --- a/io/channel-socket.c
> +++ b/io/channel-socket.c
> @@ -661,11 +661,6 @@ qio_channel_socket_set_blocking(QIOChannel *ioc,
> qemu_set_block(sioc->fd);
> } else {
> qemu_set_nonblock(sioc->fd);
> -#ifdef WIN32
> - WSAEventSelect(sioc->fd, ioc->event,
> - FD_READ | FD_ACCEPT | FD_CLOSE |
> - FD_CONNECT | FD_WRITE | FD_OOB);
> -#endif
> }
> return 0;
> }
> @@ -745,6 +740,16 @@ qio_channel_socket_shutdown(QIOChannel *ioc,
> return 0;
> }
>
> +static void qio_channel_socket_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
> +{
> + QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
> + aio_set_fd_handler(ctx, sioc->fd, false, io_read, io_write, NULL, opaque);
> +}
> +
> static GSource *qio_channel_socket_create_watch(QIOChannel *ioc,
> GIOCondition condition)
> {
> @@ -767,6 +772,7 @@ static void qio_channel_socket_class_init(ObjectClass *klass,
> ioc_klass->io_set_cork = qio_channel_socket_set_cork;
> ioc_klass->io_set_delay = qio_channel_socket_set_delay;
> ioc_klass->io_create_watch = qio_channel_socket_create_watch;
> + ioc_klass->io_set_aio_fd_handler = qio_channel_socket_set_aio_fd_handler;
> }
>
> static const TypeInfo qio_channel_socket_info = {
> diff --git a/io/channel-tls.c b/io/channel-tls.c
> index d24dc8c..fa3f93e 100644
> --- a/io/channel-tls.c
> +++ b/io/channel-tls.c
> @@ -349,6 +349,17 @@ static int qio_channel_tls_close(QIOChannel *ioc,
> return qio_channel_close(tioc->master, errp);
> }
>
> +static void qio_channel_tls_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
> +{
> + QIOChannelTLS *tioc = QIO_CHANNEL_TLS(ioc);
> +
> + qio_channel_set_aio_fd_handler(tioc->master, ctx, io_read, io_write, opaque);
> +}
> +
> static GSource *qio_channel_tls_create_watch(QIOChannel *ioc,
> GIOCondition condition)
> {
> @@ -376,6 +387,7 @@ static void qio_channel_tls_class_init(ObjectClass *klass,
> ioc_klass->io_close = qio_channel_tls_close;
> ioc_klass->io_shutdown = qio_channel_tls_shutdown;
> ioc_klass->io_create_watch = qio_channel_tls_create_watch;
> + ioc_klass->io_set_aio_fd_handler = qio_channel_tls_set_aio_fd_handler;
> }
>
> static const TypeInfo qio_channel_tls_info = {
> diff --git a/io/channel-watch.c b/io/channel-watch.c
> index cf1cdff..8640d1c 100644
> --- a/io/channel-watch.c
> +++ b/io/channel-watch.c
> @@ -285,6 +285,12 @@ GSource *qio_channel_create_socket_watch(QIOChannel *ioc,
> GSource *source;
> QIOChannelSocketSource *ssource;
>
> +#ifdef WIN32
> + WSAEventSelect(socket, ioc->event,
> + FD_READ | FD_ACCEPT | FD_CLOSE |
> + FD_CONNECT | FD_WRITE | FD_OOB);
> +#endif
> +
Not sure about this code movement (previously it is only for nonblocking),
otherwise looks good.
Fam
> source = g_source_new(&qio_channel_socket_source_funcs,
> sizeof(QIOChannelSocketSource));
> ssource = (QIOChannelSocketSource *)source;
> diff --git a/io/channel.c b/io/channel.c
> index 80924c1..ce470d7 100644
> --- a/io/channel.c
> +++ b/io/channel.c
> @@ -154,6 +154,17 @@ GSource *qio_channel_create_watch(QIOChannel *ioc,
> }
>
>
> +void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
> +{
> + QIOChannelClass *klass = QIO_CHANNEL_GET_CLASS(ioc);
> +
> + klass->io_set_aio_fd_handler(ioc, ctx, io_read, io_write, opaque);
> +}
> +
> guint qio_channel_add_watch(QIOChannel *ioc,
> GIOCondition condition,
> QIOChannelFunc func,
> --
> 2.9.3
>
>
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
2017-01-16 11:31 ` Fam Zheng
@ 2017-01-16 12:52 ` Daniel P. Berrange
2017-01-16 12:54 ` Daniel P. Berrange
2017-01-18 14:47 ` Stefan Hajnoczi
3 siblings, 0 replies; 69+ messages in thread
From: Daniel P. Berrange @ 2017-01-16 12:52 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz, stefanha
On Fri, Jan 13, 2017 at 02:17:19PM +0100, Paolo Bonzini wrote:
> This is in preparation for making qio_channel_yield work on
> AioContexts other than the main one.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
> diff --git a/io/channel-command.c b/io/channel-command.c
> index ad25313..4000b61 100644
> --- a/io/channel-command.c
> +++ b/io/channel-command.c
> @@ -328,6 +328,18 @@ static int qio_channel_command_close(QIOChannel *ioc,
> }
>
>
> +static void qio_channel_command_set_aio_fd_handler(QIOChannel *ioc,
> + AioContext *ctx,
> + IOHandler *io_read,
> + IOHandler *io_write,
> + void *opaque)
nitpick alignment
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
2017-01-16 11:31 ` Fam Zheng
2017-01-16 12:52 ` Daniel P. Berrange
@ 2017-01-16 12:54 ` Daniel P. Berrange
2017-01-18 14:47 ` Stefan Hajnoczi
3 siblings, 0 replies; 69+ messages in thread
From: Daniel P. Berrange @ 2017-01-16 12:54 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz, stefanha
On Fri, Jan 13, 2017 at 02:17:19PM +0100, Paolo Bonzini wrote:
> This is in preparation for making qio_channel_yield work on
> AioContexts other than the main one.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> include/io/channel.h | 30 ++++++++++++++++++++++++++++++
> io/channel-command.c | 13 +++++++++++++
> io/channel-file.c | 11 +++++++++++
> io/channel-socket.c | 16 +++++++++++-----
> io/channel-tls.c | 12 ++++++++++++
> io/channel-watch.c | 6 ++++++
> io/channel.c | 11 +++++++++++
> 7 files changed, 94 insertions(+), 5 deletions(-)
>
> diff --git a/include/io/channel.h b/include/io/channel.h
> index 32a9470..665edd7 100644
> --- a/include/io/channel.h
> +++ b/include/io/channel.h
> @@ -80,6 +83,9 @@ struct QIOChannel {
> Object parent;
> unsigned int features; /* bitmask of QIOChannelFeatures */
> char *name;
> + AioContext *ctx;
> + QIOChannelRestart *read_coroutine;
> + QIOChannelRestart *write_coroutine;
This hunk belongs in the next patch since its not used here and you also
change the data types in the next patch.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
` (2 preceding siblings ...)
2017-01-16 12:54 ` Daniel P. Berrange
@ 2017-01-18 14:47 ` Stefan Hajnoczi
3 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:47 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 656 bytes --]
On Fri, Jan 13, 2017 at 02:17:19PM +0100, Paolo Bonzini wrote:
> This is in preparation for making qio_channel_yield work on
> AioContexts other than the main one.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> include/io/channel.h | 30 ++++++++++++++++++++++++++++++
> io/channel-command.c | 13 +++++++++++++
> io/channel-file.c | 11 +++++++++++
> io/channel-socket.c | 16 +++++++++++-----
> io/channel-tls.c | 12 ++++++++++++
> io/channel-watch.c | 6 ++++++
> io/channel.c | 11 +++++++++++
> 7 files changed, 94 insertions(+), 5 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (3 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 04/16] io: add methods to set I/O handlers on AioContext Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 11:38 ` Fam Zheng
` (2 more replies)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads Paolo Bonzini
` (13 subsequent siblings)
18 siblings, 3 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Support separate coroutines for reading and writing, and place the
read/write handlers on the AioContext that the QIOChannel is registered
with.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
include/io/channel.h | 37 ++++++++++++++++++----
io/channel.c | 86 ++++++++++++++++++++++++++++++++++++++------------
tests/Makefile.include | 2 +-
3 files changed, 96 insertions(+), 29 deletions(-)
diff --git a/include/io/channel.h b/include/io/channel.h
index 665edd7..d7bad94 100644
--- a/include/io/channel.h
+++ b/include/io/channel.h
@@ -23,6 +23,7 @@
#include "qemu-common.h"
#include "qom/object.h"
+#include "qemu/coroutine.h"
#include "block/aio.h"
#define TYPE_QIO_CHANNEL "qio-channel"
@@ -59,8 +60,6 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
GIOCondition condition,
gpointer data);
-typedef struct QIOChannelRestart QIOChannelRestart;
-
/**
* QIOChannel:
*
@@ -84,8 +83,8 @@ struct QIOChannel {
unsigned int features; /* bitmask of QIOChannelFeatures */
char *name;
AioContext *ctx;
- QIOChannelRestart *read_coroutine;
- QIOChannelRestart *write_coroutine;
+ Coroutine *read_coroutine;
+ Coroutine *write_coroutine;
#ifdef _WIN32
HANDLE event; /* For use with GSource on Win32 */
#endif
@@ -508,13 +507,37 @@ guint qio_channel_add_watch(QIOChannel *ioc,
/**
+ * qio_channel_set_aio_context:
+ * @ioc: the channel object
+ * @ctx: the #AioContext to set the handlers on
+ *
+ * Request that qio_channel_yield() sets I/O handlers on
+ * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
+ * uses QEMU's main thread event loop.
+ */
+void qio_channel_set_aio_context(QIOChannel *ioc,
+ AioContext *ctx);
+
+/**
+ * qio_channel_detach_aio_context:
+ * @ioc: the channel object
+ *
+ * Disable any I/O handlers set by qio_channel_yield(). With the
+ * help of aio_co_schedule(), this allows moving a coroutine that was
+ * paused by qio_channel_yield() to another context.
+ */
+void qio_channel_detach_aio_context(QIOChannel *ioc);
+
+/**
* qio_channel_yield:
* @ioc: the channel object
* @condition: the I/O condition to wait for
*
- * Yields execution from the current coroutine until
- * the condition indicated by @condition becomes
- * available.
+ * Yields execution from the current coroutine until the condition
+ * indicated by @condition becomes available. @condition must
+ * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
+ * addition, no two coroutine can be waiting on the same condition
+ * and channel at the same time.
*
* This must only be called from coroutine context
*/
diff --git a/io/channel.c b/io/channel.c
index ce470d7..1e043bf 100644
--- a/io/channel.c
+++ b/io/channel.c
@@ -21,7 +21,7 @@
#include "qemu/osdep.h"
#include "io/channel.h"
#include "qapi/error.h"
-#include "qemu/coroutine.h"
+#include "qemu/main-loop.h"
bool qio_channel_has_feature(QIOChannel *ioc,
QIOChannelFeature feature)
@@ -238,36 +238,80 @@ off_t qio_channel_io_seek(QIOChannel *ioc,
}
-typedef struct QIOChannelYieldData QIOChannelYieldData;
-struct QIOChannelYieldData {
- QIOChannel *ioc;
- Coroutine *co;
-};
+static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
+
+static void qio_channel_restart_read(void *opaque)
+{
+ QIOChannel *ioc = opaque;
+ Coroutine *co = ioc->read_coroutine;
+ ioc->read_coroutine = NULL;
+ qio_channel_set_aio_fd_handlers(ioc);
+ aio_co_wake(co);
+}
-static gboolean qio_channel_yield_enter(QIOChannel *ioc,
- GIOCondition condition,
- gpointer opaque)
+static void qio_channel_restart_write(void *opaque)
{
- QIOChannelYieldData *data = opaque;
- qemu_coroutine_enter(data->co);
- return FALSE;
+ QIOChannel *ioc = opaque;
+ Coroutine *co = ioc->write_coroutine;
+
+ ioc->write_coroutine = NULL;
+ qio_channel_set_aio_fd_handlers(ioc);
+ aio_co_wake(co);
}
+static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
+{
+ IOHandler *rd_handler = NULL, *wr_handler = NULL;
+ AioContext *ctx;
+
+ if (ioc->read_coroutine) {
+ rd_handler = qio_channel_restart_read;
+ }
+ if (ioc->write_coroutine) {
+ rd_handler = qio_channel_restart_write;
+ }
+
+ ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
+ qio_channel_set_aio_fd_handler(ioc, ctx, rd_handler, wr_handler, ioc);
+}
+
+void qio_channel_set_aio_context(QIOChannel *ioc,
+ AioContext *ctx)
+{
+ AioContext *old_ctx;
+ if (ioc->ctx == ctx) {
+ return;
+ }
+
+ old_ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
+ qio_channel_set_aio_fd_handler(ioc, old_ctx, NULL, NULL, NULL);
+ ioc->ctx = ctx;
+ qio_channel_set_aio_fd_handlers(ioc);
+}
+
+void qio_channel_detach_aio_context(QIOChannel *ioc)
+{
+ ioc->read_coroutine = NULL;
+ ioc->write_coroutine = NULL;
+ qio_channel_set_aio_fd_handlers(ioc);
+ ioc->ctx = NULL;
+}
void coroutine_fn qio_channel_yield(QIOChannel *ioc,
GIOCondition condition)
{
- QIOChannelYieldData data;
-
assert(qemu_in_coroutine());
- data.ioc = ioc;
- data.co = qemu_coroutine_self();
- qio_channel_add_watch(ioc,
- condition,
- qio_channel_yield_enter,
- &data,
- NULL);
+ if (condition == G_IO_IN) {
+ assert(!ioc->read_coroutine);
+ ioc->read_coroutine = qemu_coroutine_self();
+ } else if (condition == G_IO_OUT) {
+ assert(!ioc->write_coroutine);
+ ioc->write_coroutine = qemu_coroutine_self();
+ } else {
+ abort();
+ }
+ qio_channel_set_aio_fd_handlers(ioc);
qemu_coroutine_yield();
}
diff --git a/tests/Makefile.include b/tests/Makefile.include
index 3b8ed9d..7d11bbb 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -493,7 +493,7 @@ tests/check-qjson$(EXESUF): tests/check-qjson.o $(test-util-obj-y)
tests/check-qom-interface$(EXESUF): tests/check-qom-interface.o $(test-qom-obj-y)
tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
-tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
+tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y)
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-13 13:17 ` [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts Paolo Bonzini
@ 2017-01-16 11:38 ` Fam Zheng
2017-01-16 12:24 ` Paolo Bonzini
2017-01-16 12:55 ` Daniel P. Berrange
2017-01-16 12:58 ` Daniel P. Berrange
2017-01-18 14:58 ` Stefan Hajnoczi
2 siblings, 2 replies; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 11:38 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> Support separate coroutines for reading and writing, and place the
> read/write handlers on the AioContext that the QIOChannel is registered
> with.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> include/io/channel.h | 37 ++++++++++++++++++----
> io/channel.c | 86 ++++++++++++++++++++++++++++++++++++++------------
> tests/Makefile.include | 2 +-
> 3 files changed, 96 insertions(+), 29 deletions(-)
>
> diff --git a/include/io/channel.h b/include/io/channel.h
> index 665edd7..d7bad94 100644
> --- a/include/io/channel.h
> +++ b/include/io/channel.h
> @@ -23,6 +23,7 @@
>
> #include "qemu-common.h"
> #include "qom/object.h"
> +#include "qemu/coroutine.h"
> #include "block/aio.h"
>
> #define TYPE_QIO_CHANNEL "qio-channel"
> @@ -59,8 +60,6 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
> GIOCondition condition,
> gpointer data);
>
> -typedef struct QIOChannelRestart QIOChannelRestart;
> -
> /**
> * QIOChannel:
> *
> @@ -84,8 +83,8 @@ struct QIOChannel {
> unsigned int features; /* bitmask of QIOChannelFeatures */
> char *name;
> AioContext *ctx;
> - QIOChannelRestart *read_coroutine;
> - QIOChannelRestart *write_coroutine;
> + Coroutine *read_coroutine;
> + Coroutine *write_coroutine;
> #ifdef _WIN32
> HANDLE event; /* For use with GSource on Win32 */
> #endif
> @@ -508,13 +507,37 @@ guint qio_channel_add_watch(QIOChannel *ioc,
>
>
> /**
> + * qio_channel_set_aio_context:
> + * @ioc: the channel object
> + * @ctx: the #AioContext to set the handlers on
> + *
> + * Request that qio_channel_yield() sets I/O handlers on
> + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
> + * uses QEMU's main thread event loop.
> + */
> +void qio_channel_set_aio_context(QIOChannel *ioc,
> + AioContext *ctx);
> +
> +/**
> + * qio_channel_detach_aio_context:
> + * @ioc: the channel object
> + *
> + * Disable any I/O handlers set by qio_channel_yield(). With the
> + * help of aio_co_schedule(), this allows moving a coroutine that was
> + * paused by qio_channel_yield() to another context.
> + */
> +void qio_channel_detach_aio_context(QIOChannel *ioc);
> +
> +/**
> * qio_channel_yield:
> * @ioc: the channel object
> * @condition: the I/O condition to wait for
> *
> - * Yields execution from the current coroutine until
> - * the condition indicated by @condition becomes
> - * available.
> + * Yields execution from the current coroutine until the condition
> + * indicated by @condition becomes available. @condition must
> + * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
> + * addition, no two coroutine can be waiting on the same condition
> + * and channel at the same time.
> *
> * This must only be called from coroutine context
> */
> diff --git a/io/channel.c b/io/channel.c
> index ce470d7..1e043bf 100644
> --- a/io/channel.c
> +++ b/io/channel.c
> @@ -21,7 +21,7 @@
> #include "qemu/osdep.h"
> #include "io/channel.h"
> #include "qapi/error.h"
> -#include "qemu/coroutine.h"
> +#include "qemu/main-loop.h"
>
> bool qio_channel_has_feature(QIOChannel *ioc,
> QIOChannelFeature feature)
> @@ -238,36 +238,80 @@ off_t qio_channel_io_seek(QIOChannel *ioc,
> }
>
>
> -typedef struct QIOChannelYieldData QIOChannelYieldData;
> -struct QIOChannelYieldData {
> - QIOChannel *ioc;
> - Coroutine *co;
> -};
> +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
> +
> +static void qio_channel_restart_read(void *opaque)
> +{
> + QIOChannel *ioc = opaque;
> + Coroutine *co = ioc->read_coroutine;
>
> + ioc->read_coroutine = NULL;
> + qio_channel_set_aio_fd_handlers(ioc);
> + aio_co_wake(co);
> +}
>
> -static gboolean qio_channel_yield_enter(QIOChannel *ioc,
> - GIOCondition condition,
> - gpointer opaque)
> +static void qio_channel_restart_write(void *opaque)
> {
> - QIOChannelYieldData *data = opaque;
> - qemu_coroutine_enter(data->co);
> - return FALSE;
> + QIOChannel *ioc = opaque;
> + Coroutine *co = ioc->write_coroutine;
> +
> + ioc->write_coroutine = NULL;
> + qio_channel_set_aio_fd_handlers(ioc);
> + aio_co_wake(co);
> }
>
> +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
> +{
> + IOHandler *rd_handler = NULL, *wr_handler = NULL;
> + AioContext *ctx;
> +
> + if (ioc->read_coroutine) {
> + rd_handler = qio_channel_restart_read;
s/\t/ /
> + }
> + if (ioc->write_coroutine) {
> + rd_handler = qio_channel_restart_write;
s/\t/ /
> + }
> +
> + ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> + qio_channel_set_aio_fd_handler(ioc, ctx, rd_handler, wr_handler, ioc);
> +}
> +
> +void qio_channel_set_aio_context(QIOChannel *ioc,
> + AioContext *ctx)
> +{
> + AioContext *old_ctx;
> + if (ioc->ctx == ctx) {
> + return;
> + }
> +
> + old_ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> + qio_channel_set_aio_fd_handler(ioc, old_ctx, NULL, NULL, NULL);
> + ioc->ctx = ctx;
> + qio_channel_set_aio_fd_handlers(ioc);
> +}
> +
> +void qio_channel_detach_aio_context(QIOChannel *ioc)
> +{
> + ioc->read_coroutine = NULL;
> + ioc->write_coroutine = NULL;
> + qio_channel_set_aio_fd_handlers(ioc);
Why is qio_channel_set_aio_fd_handler not needed here?
> + ioc->ctx = NULL;
> +}
>
> void coroutine_fn qio_channel_yield(QIOChannel *ioc,
> GIOCondition condition)
> {
> - QIOChannelYieldData data;
> -
> assert(qemu_in_coroutine());
> - data.ioc = ioc;
> - data.co = qemu_coroutine_self();
> - qio_channel_add_watch(ioc,
> - condition,
> - qio_channel_yield_enter,
> - &data,
> - NULL);
> + if (condition == G_IO_IN) {
> + assert(!ioc->read_coroutine);
> + ioc->read_coroutine = qemu_coroutine_self();
> + } else if (condition == G_IO_OUT) {
> + assert(!ioc->write_coroutine);
> + ioc->write_coroutine = qemu_coroutine_self();
> + } else {
> + abort();
> + }
> + qio_channel_set_aio_fd_handlers(ioc);
> qemu_coroutine_yield();
> }
>
> diff --git a/tests/Makefile.include b/tests/Makefile.include
> index 3b8ed9d..7d11bbb 100644
> --- a/tests/Makefile.include
> +++ b/tests/Makefile.include
> @@ -493,7 +493,7 @@ tests/check-qjson$(EXESUF): tests/check-qjson.o $(test-util-obj-y)
> tests/check-qom-interface$(EXESUF): tests/check-qom-interface.o $(test-qom-obj-y)
> tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
>
> -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
> +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
I guess this is a hint for moving coroutine code into a lower level library like
util.
Fam
> tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
> tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
> tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y)
> --
> 2.9.3
>
>
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 11:38 ` Fam Zheng
@ 2017-01-16 12:24 ` Paolo Bonzini
2017-01-16 12:47 ` Fam Zheng
2017-01-16 12:55 ` Daniel P. Berrange
1 sibling, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 12:24 UTC (permalink / raw)
To: Fam Zheng; +Cc: qemu-devel, stefanha
On 16/01/2017 12:38, Fam Zheng wrote:
>> +void qio_channel_detach_aio_context(QIOChannel *ioc)
>> +{
>> + ioc->read_coroutine = NULL;
>> + ioc->write_coroutine = NULL;
>> + qio_channel_set_aio_fd_handlers(ioc);
>> + ioc->ctx = NULL;
>
> Why is qio_channel_set_aio_fd_handler not needed here?
Because there are no read_coroutine and write_coroutine anymore. The
caller needs to schedule them on the right AioContext after calling
qio_channel_set_aio_context. See nbd_client_attach_aio_context in the
next patch for an example.
>> -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
>> +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
>
> I guess this is a hint for moving coroutine code into a lower level library like
> util.
Coroutine, or AioContext? The reason for this is that io/ now uses
aio_co_wake.
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 12:24 ` Paolo Bonzini
@ 2017-01-16 12:47 ` Fam Zheng
2017-01-16 12:59 ` Daniel P. Berrange
0 siblings, 1 reply; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 12:47 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Mon, 01/16 13:24, Paolo Bonzini wrote:
>
>
> On 16/01/2017 12:38, Fam Zheng wrote:
> >> +void qio_channel_detach_aio_context(QIOChannel *ioc)
> >> +{
> >> + ioc->read_coroutine = NULL;
> >> + ioc->write_coroutine = NULL;
> >> + qio_channel_set_aio_fd_handlers(ioc);
> >> + ioc->ctx = NULL;
> >
> > Why is qio_channel_set_aio_fd_handler not needed here?
>
> Because there are no read_coroutine and write_coroutine anymore. The
> caller needs to schedule them on the right AioContext after calling
> qio_channel_set_aio_context. See nbd_client_attach_aio_context in the
> next patch for an example.
>
> >> -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
> >> +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
> >
> > I guess this is a hint for moving coroutine code into a lower level library like
> > util.
>
> Coroutine, or AioContext? The reason for this is that io/ now uses
> aio_co_wake.
Or both. It just feels a bit odd to see *char* depend on *block*, maybe there're
more such dependencies to come even outside tests/?
Not necessarily for this series, of course.
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 12:47 ` Fam Zheng
@ 2017-01-16 12:59 ` Daniel P. Berrange
0 siblings, 0 replies; 69+ messages in thread
From: Daniel P. Berrange @ 2017-01-16 12:59 UTC (permalink / raw)
To: Fam Zheng; +Cc: Paolo Bonzini, qemu-devel, stefanha
On Mon, Jan 16, 2017 at 08:47:28PM +0800, Fam Zheng wrote:
> On Mon, 01/16 13:24, Paolo Bonzini wrote:
> >
> >
> > On 16/01/2017 12:38, Fam Zheng wrote:
> > >> +void qio_channel_detach_aio_context(QIOChannel *ioc)
> > >> +{
> > >> + ioc->read_coroutine = NULL;
> > >> + ioc->write_coroutine = NULL;
> > >> + qio_channel_set_aio_fd_handlers(ioc);
> > >> + ioc->ctx = NULL;
> > >
> > > Why is qio_channel_set_aio_fd_handler not needed here?
> >
> > Because there are no read_coroutine and write_coroutine anymore. The
> > caller needs to schedule them on the right AioContext after calling
> > qio_channel_set_aio_context. See nbd_client_attach_aio_context in the
> > next patch for an example.
> >
> > >> -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
> > >> +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
> > >
> > > I guess this is a hint for moving coroutine code into a lower level library like
> > > util.
> >
> > Coroutine, or AioContext? The reason for this is that io/ now uses
> > aio_co_wake.
>
> Or both. It just feels a bit odd to see *char* depend on *block*, maybe there're
> more such dependencies to come even outside tests/?
>
> Not necessarily for this series, of course.
On the contrary, I think it should be in this series - we shouldn't
introduce a dependancy on the block layer from io layer.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 11:38 ` Fam Zheng
2017-01-16 12:24 ` Paolo Bonzini
@ 2017-01-16 12:55 ` Daniel P. Berrange
2017-01-18 14:48 ` Stefan Hajnoczi
1 sibling, 1 reply; 69+ messages in thread
From: Daniel P. Berrange @ 2017-01-16 12:55 UTC (permalink / raw)
To: Fam Zheng; +Cc: Paolo Bonzini, qemu-devel, stefanha
On Mon, Jan 16, 2017 at 07:38:24PM +0800, Fam Zheng wrote:
> On Fri, 01/13 14:17, Paolo Bonzini wrote:
> > Support separate coroutines for reading and writing, and place the
> > read/write handlers on the AioContext that the QIOChannel is registered
> > with.
> >
> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > ---
> > include/io/channel.h | 37 ++++++++++++++++++----
> > io/channel.c | 86 ++++++++++++++++++++++++++++++++++++++------------
> > tests/Makefile.include | 2 +-
> > 3 files changed, 96 insertions(+), 29 deletions(-)
> >
> > diff --git a/include/io/channel.h b/include/io/channel.h
> > index 665edd7..d7bad94 100644
> > --- a/include/io/channel.h
> > +++ b/include/io/channel.h
> > @@ -23,6 +23,7 @@
> >
> > #include "qemu-common.h"
> > #include "qom/object.h"
> > +#include "qemu/coroutine.h"
> > #include "block/aio.h"
> >
> > #define TYPE_QIO_CHANNEL "qio-channel"
> > @@ -59,8 +60,6 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
> > GIOCondition condition,
> > gpointer data);
> >
> > -typedef struct QIOChannelRestart QIOChannelRestart;
> > -
> > /**
> > * QIOChannel:
> > *
> > @@ -84,8 +83,8 @@ struct QIOChannel {
> > unsigned int features; /* bitmask of QIOChannelFeatures */
> > char *name;
> > AioContext *ctx;
> > - QIOChannelRestart *read_coroutine;
> > - QIOChannelRestart *write_coroutine;
> > + Coroutine *read_coroutine;
> > + Coroutine *write_coroutine;
> > #ifdef _WIN32
> > HANDLE event; /* For use with GSource on Win32 */
> > #endif
> > @@ -508,13 +507,37 @@ guint qio_channel_add_watch(QIOChannel *ioc,
> >
> >
> > /**
> > + * qio_channel_set_aio_context:
> > + * @ioc: the channel object
> > + * @ctx: the #AioContext to set the handlers on
> > + *
> > + * Request that qio_channel_yield() sets I/O handlers on
> > + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
> > + * uses QEMU's main thread event loop.
> > + */
> > +void qio_channel_set_aio_context(QIOChannel *ioc,
> > + AioContext *ctx);
> > +
> > +/**
> > + * qio_channel_detach_aio_context:
> > + * @ioc: the channel object
> > + *
> > + * Disable any I/O handlers set by qio_channel_yield(). With the
> > + * help of aio_co_schedule(), this allows moving a coroutine that was
> > + * paused by qio_channel_yield() to another context.
> > + */
> > +void qio_channel_detach_aio_context(QIOChannel *ioc);
> > +
> > +/**
> > * qio_channel_yield:
> > * @ioc: the channel object
> > * @condition: the I/O condition to wait for
> > *
> > - * Yields execution from the current coroutine until
> > - * the condition indicated by @condition becomes
> > - * available.
> > + * Yields execution from the current coroutine until the condition
> > + * indicated by @condition becomes available. @condition must
> > + * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
> > + * addition, no two coroutine can be waiting on the same condition
> > + * and channel at the same time.
> > *
> > * This must only be called from coroutine context
> > */
> > diff --git a/io/channel.c b/io/channel.c
> > index ce470d7..1e043bf 100644
> > --- a/io/channel.c
> > +++ b/io/channel.c
> > @@ -21,7 +21,7 @@
> > #include "qemu/osdep.h"
> > #include "io/channel.h"
> > #include "qapi/error.h"
> > -#include "qemu/coroutine.h"
> > +#include "qemu/main-loop.h"
> >
> > bool qio_channel_has_feature(QIOChannel *ioc,
> > QIOChannelFeature feature)
> > @@ -238,36 +238,80 @@ off_t qio_channel_io_seek(QIOChannel *ioc,
> > }
> >
> >
> > -typedef struct QIOChannelYieldData QIOChannelYieldData;
> > -struct QIOChannelYieldData {
> > - QIOChannel *ioc;
> > - Coroutine *co;
> > -};
> > +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
> > +
> > +static void qio_channel_restart_read(void *opaque)
> > +{
> > + QIOChannel *ioc = opaque;
> > + Coroutine *co = ioc->read_coroutine;
> >
> > + ioc->read_coroutine = NULL;
> > + qio_channel_set_aio_fd_handlers(ioc);
> > + aio_co_wake(co);
> > +}
> >
> > -static gboolean qio_channel_yield_enter(QIOChannel *ioc,
> > - GIOCondition condition,
> > - gpointer opaque)
> > +static void qio_channel_restart_write(void *opaque)
> > {
> > - QIOChannelYieldData *data = opaque;
> > - qemu_coroutine_enter(data->co);
> > - return FALSE;
> > + QIOChannel *ioc = opaque;
> > + Coroutine *co = ioc->write_coroutine;
> > +
> > + ioc->write_coroutine = NULL;
> > + qio_channel_set_aio_fd_handlers(ioc);
> > + aio_co_wake(co);
> > }
> >
> > +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
> > +{
> > + IOHandler *rd_handler = NULL, *wr_handler = NULL;
> > + AioContext *ctx;
> > +
> > + if (ioc->read_coroutine) {
> > + rd_handler = qio_channel_restart_read;
>
> s/\t/ /
>
> > + }
> > + if (ioc->write_coroutine) {
> > + rd_handler = qio_channel_restart_write;
>
> s/\t/ /
>
> > + }
> > +
> > + ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> > + qio_channel_set_aio_fd_handler(ioc, ctx, rd_handler, wr_handler, ioc);
> > +}
> > +
> > +void qio_channel_set_aio_context(QIOChannel *ioc,
> > + AioContext *ctx)
> > +{
> > + AioContext *old_ctx;
> > + if (ioc->ctx == ctx) {
> > + return;
> > + }
> > +
> > + old_ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> > + qio_channel_set_aio_fd_handler(ioc, old_ctx, NULL, NULL, NULL);
> > + ioc->ctx = ctx;
> > + qio_channel_set_aio_fd_handlers(ioc);
> > +}
> > +
> > +void qio_channel_detach_aio_context(QIOChannel *ioc)
> > +{
> > + ioc->read_coroutine = NULL;
> > + ioc->write_coroutine = NULL;
> > + qio_channel_set_aio_fd_handlers(ioc);
>
> Why is qio_channel_set_aio_fd_handler not needed here?
>
> > + ioc->ctx = NULL;
> > +}
> >
> > void coroutine_fn qio_channel_yield(QIOChannel *ioc,
> > GIOCondition condition)
> > {
> > - QIOChannelYieldData data;
> > -
> > assert(qemu_in_coroutine());
> > - data.ioc = ioc;
> > - data.co = qemu_coroutine_self();
> > - qio_channel_add_watch(ioc,
> > - condition,
> > - qio_channel_yield_enter,
> > - &data,
> > - NULL);
> > + if (condition == G_IO_IN) {
> > + assert(!ioc->read_coroutine);
> > + ioc->read_coroutine = qemu_coroutine_self();
> > + } else if (condition == G_IO_OUT) {
> > + assert(!ioc->write_coroutine);
> > + ioc->write_coroutine = qemu_coroutine_self();
> > + } else {
> > + abort();
> > + }
> > + qio_channel_set_aio_fd_handlers(ioc);
> > qemu_coroutine_yield();
> > }
> >
> > diff --git a/tests/Makefile.include b/tests/Makefile.include
> > index 3b8ed9d..7d11bbb 100644
> > --- a/tests/Makefile.include
> > +++ b/tests/Makefile.include
> > @@ -493,7 +493,7 @@ tests/check-qjson$(EXESUF): tests/check-qjson.o $(test-util-obj-y)
> > tests/check-qom-interface$(EXESUF): tests/check-qom-interface.o $(test-qom-obj-y)
> > tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
> >
> > -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
> > +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
>
> I guess this is a hint for moving coroutine code into a lower level library like
> util.
The coroutine code is already in util/, so I'm assuming this is actually
for the AioContext stuff. Yes, though, AioContext ought to be moved into
util/ as part of this series IMHO, since the io/ channel code shouldn't
have a dependancy on block/ layer.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 12:55 ` Daniel P. Berrange
@ 2017-01-18 14:48 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:48 UTC (permalink / raw)
To: Daniel P. Berrange; +Cc: Fam Zheng, Paolo Bonzini, qemu-devel
[-- Attachment #1: Type: text/plain, Size: 8929 bytes --]
On Mon, Jan 16, 2017 at 12:55:59PM +0000, Daniel P. Berrange wrote:
> On Mon, Jan 16, 2017 at 07:38:24PM +0800, Fam Zheng wrote:
> > On Fri, 01/13 14:17, Paolo Bonzini wrote:
> > > Support separate coroutines for reading and writing, and place the
> > > read/write handlers on the AioContext that the QIOChannel is registered
> > > with.
> > >
> > > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > > ---
> > > include/io/channel.h | 37 ++++++++++++++++++----
> > > io/channel.c | 86 ++++++++++++++++++++++++++++++++++++++------------
> > > tests/Makefile.include | 2 +-
> > > 3 files changed, 96 insertions(+), 29 deletions(-)
> > >
> > > diff --git a/include/io/channel.h b/include/io/channel.h
> > > index 665edd7..d7bad94 100644
> > > --- a/include/io/channel.h
> > > +++ b/include/io/channel.h
> > > @@ -23,6 +23,7 @@
> > >
> > > #include "qemu-common.h"
> > > #include "qom/object.h"
> > > +#include "qemu/coroutine.h"
> > > #include "block/aio.h"
> > >
> > > #define TYPE_QIO_CHANNEL "qio-channel"
> > > @@ -59,8 +60,6 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
> > > GIOCondition condition,
> > > gpointer data);
> > >
> > > -typedef struct QIOChannelRestart QIOChannelRestart;
> > > -
> > > /**
> > > * QIOChannel:
> > > *
> > > @@ -84,8 +83,8 @@ struct QIOChannel {
> > > unsigned int features; /* bitmask of QIOChannelFeatures */
> > > char *name;
> > > AioContext *ctx;
> > > - QIOChannelRestart *read_coroutine;
> > > - QIOChannelRestart *write_coroutine;
> > > + Coroutine *read_coroutine;
> > > + Coroutine *write_coroutine;
> > > #ifdef _WIN32
> > > HANDLE event; /* For use with GSource on Win32 */
> > > #endif
> > > @@ -508,13 +507,37 @@ guint qio_channel_add_watch(QIOChannel *ioc,
> > >
> > >
> > > /**
> > > + * qio_channel_set_aio_context:
> > > + * @ioc: the channel object
> > > + * @ctx: the #AioContext to set the handlers on
> > > + *
> > > + * Request that qio_channel_yield() sets I/O handlers on
> > > + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
> > > + * uses QEMU's main thread event loop.
> > > + */
> > > +void qio_channel_set_aio_context(QIOChannel *ioc,
> > > + AioContext *ctx);
> > > +
> > > +/**
> > > + * qio_channel_detach_aio_context:
> > > + * @ioc: the channel object
> > > + *
> > > + * Disable any I/O handlers set by qio_channel_yield(). With the
> > > + * help of aio_co_schedule(), this allows moving a coroutine that was
> > > + * paused by qio_channel_yield() to another context.
> > > + */
> > > +void qio_channel_detach_aio_context(QIOChannel *ioc);
> > > +
> > > +/**
> > > * qio_channel_yield:
> > > * @ioc: the channel object
> > > * @condition: the I/O condition to wait for
> > > *
> > > - * Yields execution from the current coroutine until
> > > - * the condition indicated by @condition becomes
> > > - * available.
> > > + * Yields execution from the current coroutine until the condition
> > > + * indicated by @condition becomes available. @condition must
> > > + * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
> > > + * addition, no two coroutine can be waiting on the same condition
> > > + * and channel at the same time.
> > > *
> > > * This must only be called from coroutine context
> > > */
> > > diff --git a/io/channel.c b/io/channel.c
> > > index ce470d7..1e043bf 100644
> > > --- a/io/channel.c
> > > +++ b/io/channel.c
> > > @@ -21,7 +21,7 @@
> > > #include "qemu/osdep.h"
> > > #include "io/channel.h"
> > > #include "qapi/error.h"
> > > -#include "qemu/coroutine.h"
> > > +#include "qemu/main-loop.h"
> > >
> > > bool qio_channel_has_feature(QIOChannel *ioc,
> > > QIOChannelFeature feature)
> > > @@ -238,36 +238,80 @@ off_t qio_channel_io_seek(QIOChannel *ioc,
> > > }
> > >
> > >
> > > -typedef struct QIOChannelYieldData QIOChannelYieldData;
> > > -struct QIOChannelYieldData {
> > > - QIOChannel *ioc;
> > > - Coroutine *co;
> > > -};
> > > +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
> > > +
> > > +static void qio_channel_restart_read(void *opaque)
> > > +{
> > > + QIOChannel *ioc = opaque;
> > > + Coroutine *co = ioc->read_coroutine;
> > >
> > > + ioc->read_coroutine = NULL;
> > > + qio_channel_set_aio_fd_handlers(ioc);
> > > + aio_co_wake(co);
> > > +}
> > >
> > > -static gboolean qio_channel_yield_enter(QIOChannel *ioc,
> > > - GIOCondition condition,
> > > - gpointer opaque)
> > > +static void qio_channel_restart_write(void *opaque)
> > > {
> > > - QIOChannelYieldData *data = opaque;
> > > - qemu_coroutine_enter(data->co);
> > > - return FALSE;
> > > + QIOChannel *ioc = opaque;
> > > + Coroutine *co = ioc->write_coroutine;
> > > +
> > > + ioc->write_coroutine = NULL;
> > > + qio_channel_set_aio_fd_handlers(ioc);
> > > + aio_co_wake(co);
> > > }
> > >
> > > +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
> > > +{
> > > + IOHandler *rd_handler = NULL, *wr_handler = NULL;
> > > + AioContext *ctx;
> > > +
> > > + if (ioc->read_coroutine) {
> > > + rd_handler = qio_channel_restart_read;
> >
> > s/\t/ /
> >
> > > + }
> > > + if (ioc->write_coroutine) {
> > > + rd_handler = qio_channel_restart_write;
> >
> > s/\t/ /
> >
> > > + }
> > > +
> > > + ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> > > + qio_channel_set_aio_fd_handler(ioc, ctx, rd_handler, wr_handler, ioc);
> > > +}
> > > +
> > > +void qio_channel_set_aio_context(QIOChannel *ioc,
> > > + AioContext *ctx)
> > > +{
> > > + AioContext *old_ctx;
> > > + if (ioc->ctx == ctx) {
> > > + return;
> > > + }
> > > +
> > > + old_ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
> > > + qio_channel_set_aio_fd_handler(ioc, old_ctx, NULL, NULL, NULL);
> > > + ioc->ctx = ctx;
> > > + qio_channel_set_aio_fd_handlers(ioc);
> > > +}
> > > +
> > > +void qio_channel_detach_aio_context(QIOChannel *ioc)
> > > +{
> > > + ioc->read_coroutine = NULL;
> > > + ioc->write_coroutine = NULL;
> > > + qio_channel_set_aio_fd_handlers(ioc);
> >
> > Why is qio_channel_set_aio_fd_handler not needed here?
> >
> > > + ioc->ctx = NULL;
> > > +}
> > >
> > > void coroutine_fn qio_channel_yield(QIOChannel *ioc,
> > > GIOCondition condition)
> > > {
> > > - QIOChannelYieldData data;
> > > -
> > > assert(qemu_in_coroutine());
> > > - data.ioc = ioc;
> > > - data.co = qemu_coroutine_self();
> > > - qio_channel_add_watch(ioc,
> > > - condition,
> > > - qio_channel_yield_enter,
> > > - &data,
> > > - NULL);
> > > + if (condition == G_IO_IN) {
> > > + assert(!ioc->read_coroutine);
> > > + ioc->read_coroutine = qemu_coroutine_self();
> > > + } else if (condition == G_IO_OUT) {
> > > + assert(!ioc->write_coroutine);
> > > + ioc->write_coroutine = qemu_coroutine_self();
> > > + } else {
> > > + abort();
> > > + }
> > > + qio_channel_set_aio_fd_handlers(ioc);
> > > qemu_coroutine_yield();
> > > }
> > >
> > > diff --git a/tests/Makefile.include b/tests/Makefile.include
> > > index 3b8ed9d..7d11bbb 100644
> > > --- a/tests/Makefile.include
> > > +++ b/tests/Makefile.include
> > > @@ -493,7 +493,7 @@ tests/check-qjson$(EXESUF): tests/check-qjson.o $(test-util-obj-y)
> > > tests/check-qom-interface$(EXESUF): tests/check-qom-interface.o $(test-qom-obj-y)
> > > tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
> > >
> > > -tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y)
> > > +tests/test-char$(EXESUF): tests/test-char.o qemu-char.o qemu-timer.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(test-block-obj-y)
> >
> > I guess this is a hint for moving coroutine code into a lower level library like
> > util.
>
> The coroutine code is already in util/, so I'm assuming this is actually
> for the AioContext stuff. Yes, though, AioContext ought to be moved into
> util/ as part of this series IMHO, since the io/ channel code shouldn't
> have a dependancy on block/ layer.
That makes sense to me. I think the move is overdue since coroutines
are widely useful outside the block layer and aren't specific to block
I/O.
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-13 13:17 ` [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts Paolo Bonzini
2017-01-16 11:38 ` Fam Zheng
@ 2017-01-16 12:58 ` Daniel P. Berrange
2017-01-16 14:18 ` Paolo Bonzini
2017-01-18 14:58 ` Stefan Hajnoczi
2 siblings, 1 reply; 69+ messages in thread
From: Daniel P. Berrange @ 2017-01-16 12:58 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz, stefanha
On Fri, Jan 13, 2017 at 02:17:20PM +0100, Paolo Bonzini wrote:
> Support separate coroutines for reading and writing, and place the
> read/write handlers on the AioContext that the QIOChannel is registered
> with.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> include/io/channel.h | 37 ++++++++++++++++++----
> io/channel.c | 86 ++++++++++++++++++++++++++++++++++++++------------
> tests/Makefile.include | 2 +-
> 3 files changed, 96 insertions(+), 29 deletions(-)
>
> diff --git a/include/io/channel.h b/include/io/channel.h
> index 665edd7..d7bad94 100644
> --- a/include/io/channel.h
> +++ b/include/io/channel.h
> @@ -23,6 +23,7 @@
>
> #include "qemu-common.h"
> #include "qom/object.h"
> +#include "qemu/coroutine.h"
> #include "block/aio.h"
>
> #define TYPE_QIO_CHANNEL "qio-channel"
> @@ -59,8 +60,6 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
> GIOCondition condition,
> gpointer data);
>
> -typedef struct QIOChannelRestart QIOChannelRestart;
> -
> /**
> * QIOChannel:
> *
> @@ -84,8 +83,8 @@ struct QIOChannel {
> unsigned int features; /* bitmask of QIOChannelFeatures */
> char *name;
> AioContext *ctx;
> - QIOChannelRestart *read_coroutine;
> - QIOChannelRestart *write_coroutine;
> + Coroutine *read_coroutine;
> + Coroutine *write_coroutine;
Need to squash in part of previous patch here.
> #ifdef _WIN32
> HANDLE event; /* For use with GSource on Win32 */
> #endif
> @@ -508,13 +507,37 @@ guint qio_channel_add_watch(QIOChannel *ioc,
>
>
> /**
> + * qio_channel_set_aio_context:
> + * @ioc: the channel object
> + * @ctx: the #AioContext to set the handlers on
> + *
> + * Request that qio_channel_yield() sets I/O handlers on
> + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
> + * uses QEMU's main thread event loop.
> + */
Can you note that it is explicitly permitted to call this while
inside a qio_channel_yield().
> +void qio_channel_set_aio_context(QIOChannel *ioc,
> + AioContext *ctx);
> +
> +/**
> + * qio_channel_detach_aio_context:
> + * @ioc: the channel object
> + *
> + * Disable any I/O handlers set by qio_channel_yield(). With the
> + * help of aio_co_schedule(), this allows moving a coroutine that was
> + * paused by qio_channel_yield() to another context.
> + */
> +void qio_channel_detach_aio_context(QIOChannel *ioc);
> diff --git a/io/channel.c b/io/channel.c
> index ce470d7..1e043bf 100644
> --- a/io/channel.c
> +++ b/io/channel.c
> @@ -21,7 +21,7 @@
> #include "qemu/osdep.h"
> #include "io/channel.h"
> #include "qapi/error.h"
> -#include "qemu/coroutine.h"
> +#include "qemu/main-loop.h"
>
> bool qio_channel_has_feature(QIOChannel *ioc,
> QIOChannelFeature feature)
> @@ -238,36 +238,80 @@ off_t qio_channel_io_seek(QIOChannel *ioc,
> }
>
>
> -typedef struct QIOChannelYieldData QIOChannelYieldData;
> -struct QIOChannelYieldData {
> - QIOChannel *ioc;
> - Coroutine *co;
> -};
> +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
> +
> +static void qio_channel_restart_read(void *opaque)
> +{
> + QIOChannel *ioc = opaque;
> + Coroutine *co = ioc->read_coroutine;
>
> + ioc->read_coroutine = NULL;
> + qio_channel_set_aio_fd_handlers(ioc);
> + aio_co_wake(co);
> +}
>
> -static gboolean qio_channel_yield_enter(QIOChannel *ioc,
> - GIOCondition condition,
> - gpointer opaque)
> +static void qio_channel_restart_write(void *opaque)
> {
> - QIOChannelYieldData *data = opaque;
> - qemu_coroutine_enter(data->co);
> - return FALSE;
> + QIOChannel *ioc = opaque;
> + Coroutine *co = ioc->write_coroutine;
> +
> + ioc->write_coroutine = NULL;
> + qio_channel_set_aio_fd_handlers(ioc);
> + aio_co_wake(co);
> }
>
> +static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
> +{
> + IOHandler *rd_handler = NULL, *wr_handler = NULL;
> + AioContext *ctx;
> +
> + if (ioc->read_coroutine) {
> + rd_handler = qio_channel_restart_read;
> + }
> + if (ioc->write_coroutine) {
> + rd_handler = qio_channel_restart_write;
> + }
Tab damage.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-16 12:58 ` Daniel P. Berrange
@ 2017-01-16 14:18 ` Paolo Bonzini
0 siblings, 0 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 14:18 UTC (permalink / raw)
To: Daniel P. Berrange; +Cc: famz, qemu-devel, stefanha
On 16/01/2017 13:58, Daniel P. Berrange wrote:
>> + * Request that qio_channel_yield() sets I/O handlers on
>> + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
>> + * uses QEMU's main thread event loop.
>> + */
> Can you note that it is explicitly permitted to call this while
> inside a qio_channel_yield().
Yes:
* You can move a #QIOChannel from an #AioContext to another even if
* I/O handlers are set for a coroutine. However, #QIOChannel provides
* no synchronization between the calls to qio_channel_yield() and
* qio_channel_set_aio_context().
*
* Therefore you should first call qio_channel_detach_aio_context()
* to ensure that the coroutine is not entered concurrently. Then,
* while the coroutine has yielded, call qio_channel_set_aio_context(),
* and then aio_co_schedule() to place the coroutine on the new
* #AioContext. The calls to qio_channel_detach_aio_context()
* and qio_channel_set_aio_context() should be protected with
* aio_context_acquire() and aio_context_release().
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-13 13:17 ` [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts Paolo Bonzini
2017-01-16 11:38 ` Fam Zheng
2017-01-16 12:58 ` Daniel P. Berrange
@ 2017-01-18 14:58 ` Stefan Hajnoczi
2017-01-18 16:43 ` Paolo Bonzini
2 siblings, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 14:58 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 1748 bytes --]
On Fri, Jan 13, 2017 at 02:17:20PM +0100, Paolo Bonzini wrote:
> /**
> + * qio_channel_set_aio_context:
> + * @ioc: the channel object
> + * @ctx: the #AioContext to set the handlers on
> + *
> + * Request that qio_channel_yield() sets I/O handlers on
> + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
> + * uses QEMU's main thread event loop.
> + */
> +void qio_channel_set_aio_context(QIOChannel *ioc,
> + AioContext *ctx);
> +
> +/**
> + * qio_channel_detach_aio_context:
> + * @ioc: the channel object
> + *
> + * Disable any I/O handlers set by qio_channel_yield(). With the
> + * help of aio_co_schedule(), this allows moving a coroutine that was
> + * paused by qio_channel_yield() to another context.
> + */
> +void qio_channel_detach_aio_context(QIOChannel *ioc);
The block layer's bdrv_set_aio_context() has different semantics. It
invokes .detach()/.attach() callbacks and does AioContext locking so the
function can be called safely even while the block driver is waiting for
events.
It's unfortunate to that the block and io channel APIs act differently
despite having similar names. Was there a reason to choose different
semantics?
> +
> +/**
> * qio_channel_yield:
> * @ioc: the channel object
> * @condition: the I/O condition to wait for
> *
> - * Yields execution from the current coroutine until
> - * the condition indicated by @condition becomes
> - * available.
> + * Yields execution from the current coroutine until the condition
> + * indicated by @condition becomes available. @condition must
> + * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
> + * addition, no two coroutine can be waiting on the same condition
s/coroutine/coroutines/
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts
2017-01-18 14:58 ` Stefan Hajnoczi
@ 2017-01-18 16:43 ` Paolo Bonzini
2017-01-18 17:22 ` Eric Blake
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-18 16:43 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 3378 bytes --]
On 18/01/2017 15:58, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 02:17:20PM +0100, Paolo Bonzini wrote:
>> /**
>> + * qio_channel_set_aio_context:
>> + * @ioc: the channel object
>> + * @ctx: the #AioContext to set the handlers on
>> + *
>> + * Request that qio_channel_yield() sets I/O handlers on
>> + * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
>> + * uses QEMU's main thread event loop.
>> + */
>> +void qio_channel_set_aio_context(QIOChannel *ioc,
>> + AioContext *ctx);
>> +
>> +/**
>> + * qio_channel_detach_aio_context:
>> + * @ioc: the channel object
>> + *
>> + * Disable any I/O handlers set by qio_channel_yield(). With the
>> + * help of aio_co_schedule(), this allows moving a coroutine that was
>> + * paused by qio_channel_yield() to another context.
>> + */
>> +void qio_channel_detach_aio_context(QIOChannel *ioc);
>
> The block layer's bdrv_set_aio_context() has different semantics. It
> invokes .detach()/.attach() callbacks and does AioContext locking so the
> function can be called safely even while the block driver is waiting for
> events.
>
> It's unfortunate to that the block and io channel APIs act differently
> despite having similar names. Was there a reason to choose different
> semantics?
Hmm, it's true. I had forgotten that bdrv_set_aio_context exists.
set_aio_context can be called from the block layer attach callback, but
it's not enough alone (you need aio_co_schedule too) so I didn't want to
call the function qio_channel_attach_aio_context. But maybe it *is* a
better name, I'll go for it in v2.
By the way, v2 will have a better comment on how to use the API:
+ * You can move a #QIOChannel from an #AioContext to another even if
+ * I/O handlers are set for a coroutine. However, #QIOChannel provides
+ * no synchronization between the calls to qio_channel_yield() and
+ * qio_channel_set_aio_context().
+ *
+ * Therefore you should first call qio_channel_detach_aio_context()
+ * to ensure that the coroutine is not entered concurrently. Then,
+ * while the coroutine has yielded, call qio_channel_set_aio_context(),
+ * and then aio_co_schedule() to place the coroutine on the new
+ * #AioContext. The calls to qio_channel_detach_aio_context()
+ * and qio_channel_set_aio_context() should be protected with
+ * aio_context_acquire() and aio_context_release().
The "while the coroutine has yielded" part is currently handled with
aio_context_acquire/aio_context_release (the coroutine cannot run at all
between aio_context_acquire and release).
When they will be gone, some kind of BDRV_POLL_WHILE at the end of
bdrv_detach_aio_context should be enough to ensure that the event loop
is quiescent.
Paolo
>> +
>> +/**
>> * qio_channel_yield:
>> * @ioc: the channel object
>> * @condition: the I/O condition to wait for
>> *
>> - * Yields execution from the current coroutine until
>> - * the condition indicated by @condition becomes
>> - * available.
>> + * Yields execution from the current coroutine until the condition
>> + * indicated by @condition becomes available. @condition must
>> + * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
>> + * addition, no two coroutine can be waiting on the same condition
>
> s/coroutine/coroutines/
>
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (4 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 05/16] io: make qio_channel_yield aware of AioContexts Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 12:52 ` Fam Zheng
2017-01-18 15:24 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 07/16] coroutine-lock: reschedule coroutine on the AioContext it was running on Paolo Bonzini
` (12 subsequent siblings)
18 siblings, 2 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Read the replies from a coroutine, switching the read side between the
"read header" coroutine and the I/O coroutine that reads the body of
the reply.
qio_channel_yield is used so that the right coroutine is restarted
automatically, eliminating the need for send_coroutine in
NBDClientSession.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
block/nbd-client.c | 108 +++++++++++++++++++++--------------------------------
block/nbd-client.h | 2 +-
nbd/client.c | 2 +-
nbd/common.c | 9 +----
4 files changed, 45 insertions(+), 76 deletions(-)
diff --git a/block/nbd-client.c b/block/nbd-client.c
index 06f1532..eacc7a5 100644
--- a/block/nbd-client.c
+++ b/block/nbd-client.c
@@ -33,8 +33,9 @@
#define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
#define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs))
-static void nbd_recv_coroutines_enter_all(NBDClientSession *s)
+static void nbd_recv_coroutines_enter_all(BlockDriverState *bs)
{
+ NBDClientSession *s = nbd_get_client_session(bs);
int i;
for (i = 0; i < MAX_NBD_REQUESTS; i++) {
@@ -42,6 +43,7 @@ static void nbd_recv_coroutines_enter_all(NBDClientSession *s)
qemu_coroutine_enter(s->recv_coroutine[i]);
}
}
+ BDRV_POLL_WHILE(bs, s->read_reply_co);
}
static void nbd_teardown_connection(BlockDriverState *bs)
@@ -56,7 +58,7 @@ static void nbd_teardown_connection(BlockDriverState *bs)
qio_channel_shutdown(client->ioc,
QIO_CHANNEL_SHUTDOWN_BOTH,
NULL);
- nbd_recv_coroutines_enter_all(client);
+ nbd_recv_coroutines_enter_all(bs);
nbd_client_detach_aio_context(bs);
object_unref(OBJECT(client->sioc));
@@ -65,54 +67,34 @@ static void nbd_teardown_connection(BlockDriverState *bs)
client->ioc = NULL;
}
-static void nbd_reply_ready(void *opaque)
+static void nbd_read_reply_entry(void *opaque)
{
- BlockDriverState *bs = opaque;
- NBDClientSession *s = nbd_get_client_session(bs);
+ NBDClientSession *s = opaque;
uint64_t i;
int ret;
- if (!s->ioc) { /* Already closed */
- return;
- }
-
- if (s->reply.handle == 0) {
- /* No reply already in flight. Fetch a header. It is possible
- * that another thread has done the same thing in parallel, so
- * the socket is not readable anymore.
- */
+ for (;;) {
+ assert(s->reply.handle == 0);
ret = nbd_receive_reply(s->ioc, &s->reply);
- if (ret == -EAGAIN) {
- return;
- }
if (ret < 0) {
- s->reply.handle = 0;
- goto fail;
+ break;
}
- }
-
- /* There's no need for a mutex on the receive side, because the
- * handler acts as a synchronization point and ensures that only
- * one coroutine is called until the reply finishes. */
- i = HANDLE_TO_INDEX(s, s->reply.handle);
- if (i >= MAX_NBD_REQUESTS) {
- goto fail;
- }
- if (s->recv_coroutine[i]) {
- qemu_coroutine_enter(s->recv_coroutine[i]);
- return;
- }
-
-fail:
- nbd_teardown_connection(bs);
-}
+ /* There's no need for a mutex on the receive side, because the
+ * handler acts as a synchronization point and ensures that only
+ * one coroutine is called until the reply finishes.
+ */
+ i = HANDLE_TO_INDEX(s, s->reply.handle);
+ if (i >= MAX_NBD_REQUESTS || !s->recv_coroutine[i]) {
+ break;
+ }
-static void nbd_restart_write(void *opaque)
-{
- BlockDriverState *bs = opaque;
+ aio_co_wake(s->recv_coroutine[i]);
- qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine);
+ /* We're woken up by the recv_coroutine itself. */
+ qemu_coroutine_yield();
+ }
+ s->read_reply_co = NULL;
}
static int nbd_co_send_request(BlockDriverState *bs,
@@ -120,7 +102,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
QEMUIOVector *qiov)
{
NBDClientSession *s = nbd_get_client_session(bs);
- AioContext *aio_context;
int rc, ret, i;
qemu_co_mutex_lock(&s->send_mutex);
@@ -141,11 +122,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
return -EPIPE;
}
- s->send_coroutine = qemu_coroutine_self();
- aio_context = bdrv_get_aio_context(bs);
-
- aio_set_fd_handler(aio_context, s->sioc->fd, false,
- nbd_reply_ready, nbd_restart_write, NULL, bs);
if (qiov) {
qio_channel_set_cork(s->ioc, true);
rc = nbd_send_request(s->ioc, request);
@@ -160,9 +136,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
} else {
rc = nbd_send_request(s->ioc, request);
}
- aio_set_fd_handler(aio_context, s->sioc->fd, false,
- nbd_reply_ready, NULL, NULL, bs);
- s->send_coroutine = NULL;
qemu_co_mutex_unlock(&s->send_mutex);
return rc;
}
@@ -174,8 +147,7 @@ static void nbd_co_receive_reply(NBDClientSession *s,
{
int ret;
- /* Wait until we're woken up by the read handler. TODO: perhaps
- * peek at the next reply and avoid yielding if it's ours? */
+ /* Wait until we're woken up by nbd_read_reply_entry. */
qemu_coroutine_yield();
*reply = s->reply;
if (reply->handle != request->handle ||
@@ -209,14 +181,18 @@ static void nbd_coroutine_start(NBDClientSession *s,
/* s->recv_coroutine[i] is set as soon as we get the send_lock. */
}
-static void nbd_coroutine_end(NBDClientSession *s,
+static void nbd_coroutine_end(BlockDriverState *bs,
NBDRequest *request)
{
+ NBDClientSession *s = nbd_get_client_session(bs);
int i = HANDLE_TO_INDEX(s, request->handle);
+
s->recv_coroutine[i] = NULL;
- if (s->in_flight-- == MAX_NBD_REQUESTS) {
- qemu_co_queue_next(&s->free_sema);
- }
+ s->in_flight--;
+ qemu_co_queue_next(&s->free_sema);
+
+ /* Kick the read_reply_co to get the next reply. */
+ aio_co_wake(s->read_reply_co);
}
int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
@@ -241,7 +217,7 @@ int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
} else {
nbd_co_receive_reply(client, &request, &reply, qiov);
}
- nbd_coroutine_end(client, &request);
+ nbd_coroutine_end(bs, &request);
return -reply.error;
}
@@ -271,7 +247,7 @@ int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
} else {
nbd_co_receive_reply(client, &request, &reply, NULL);
}
- nbd_coroutine_end(client, &request);
+ nbd_coroutine_end(bs, &request);
return -reply.error;
}
@@ -306,7 +282,7 @@ int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
} else {
nbd_co_receive_reply(client, &request, &reply, NULL);
}
- nbd_coroutine_end(client, &request);
+ nbd_coroutine_end(bs, &request);
return -reply.error;
}
@@ -331,7 +307,7 @@ int nbd_client_co_flush(BlockDriverState *bs)
} else {
nbd_co_receive_reply(client, &request, &reply, NULL);
}
- nbd_coroutine_end(client, &request);
+ nbd_coroutine_end(bs, &request);
return -reply.error;
}
@@ -357,23 +333,23 @@ int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
} else {
nbd_co_receive_reply(client, &request, &reply, NULL);
}
- nbd_coroutine_end(client, &request);
+ nbd_coroutine_end(bs, &request);
return -reply.error;
}
void nbd_client_detach_aio_context(BlockDriverState *bs)
{
- aio_set_fd_handler(bdrv_get_aio_context(bs),
- nbd_get_client_session(bs)->sioc->fd,
- false, NULL, NULL, NULL, NULL);
+ NBDClientSession *client = nbd_get_client_session(bs);
+ qio_channel_detach_aio_context(QIO_CHANNEL(client->sioc));
}
void nbd_client_attach_aio_context(BlockDriverState *bs,
AioContext *new_context)
{
- aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sioc->fd,
- false, nbd_reply_ready, NULL, NULL, bs);
+ NBDClientSession *client = nbd_get_client_session(bs);
+ qio_channel_set_aio_context(QIO_CHANNEL(client->sioc), new_context);
+ aio_co_schedule(new_context, client->read_reply_co);
}
void nbd_client_close(BlockDriverState *bs)
@@ -434,7 +410,7 @@ int nbd_client_init(BlockDriverState *bs,
/* Now that we're connected, set the socket to be non-blocking and
* kick the reply mechanism. */
qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
-
+ client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
logout("Established connection with NBD server\n");
diff --git a/block/nbd-client.h b/block/nbd-client.h
index f8d6006..8cdfc92 100644
--- a/block/nbd-client.h
+++ b/block/nbd-client.h
@@ -25,7 +25,7 @@ typedef struct NBDClientSession {
CoMutex send_mutex;
CoQueue free_sema;
- Coroutine *send_coroutine;
+ Coroutine *read_reply_co;
int in_flight;
Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
diff --git a/nbd/client.c b/nbd/client.c
index ffb0743..5c9dee3 100644
--- a/nbd/client.c
+++ b/nbd/client.c
@@ -778,7 +778,7 @@ ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply)
ssize_t ret;
ret = read_sync(ioc, buf, sizeof(buf));
- if (ret < 0) {
+ if (ret <= 0) {
return ret;
}
diff --git a/nbd/common.c b/nbd/common.c
index b583a4f..805dbdc 100644
--- a/nbd/common.c
+++ b/nbd/common.c
@@ -43,14 +43,7 @@ ssize_t nbd_wr_syncv(QIOChannel *ioc,
}
if (len == QIO_CHANNEL_ERR_BLOCK) {
if (qemu_in_coroutine()) {
- /* XXX figure out if we can create a variant on
- * qio_channel_yield() that works with AIO contexts
- * and consider using that in this branch */
- qemu_coroutine_yield();
- } else if (done) {
- /* XXX this is needed by nbd_reply_ready. */
- qio_channel_wait(ioc,
- do_read ? G_IO_IN : G_IO_OUT);
+ qio_channel_yield(ioc, do_read ? G_IO_IN : G_IO_OUT);
} else {
return -EAGAIN;
}
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads
2017-01-13 13:17 ` [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads Paolo Bonzini
@ 2017-01-16 12:52 ` Fam Zheng
2017-01-16 13:31 ` Paolo Bonzini
2017-01-18 15:24 ` Stefan Hajnoczi
1 sibling, 1 reply; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 12:52 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> Read the replies from a coroutine, switching the read side between the
> "read header" coroutine and the I/O coroutine that reads the body of
> the reply.
>
> qio_channel_yield is used so that the right coroutine is restarted
> automatically, eliminating the need for send_coroutine in
> NBDClientSession.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> block/nbd-client.c | 108 +++++++++++++++++++++--------------------------------
> block/nbd-client.h | 2 +-
> nbd/client.c | 2 +-
> nbd/common.c | 9 +----
> 4 files changed, 45 insertions(+), 76 deletions(-)
>
> @@ -65,54 +67,34 @@ static void nbd_teardown_connection(BlockDriverState *bs)
> client->ioc = NULL;
> }
>
> -static void nbd_reply_ready(void *opaque)
> +static void nbd_read_reply_entry(void *opaque)
> {
> - BlockDriverState *bs = opaque;
> - NBDClientSession *s = nbd_get_client_session(bs);
> + NBDClientSession *s = opaque;
> uint64_t i;
> int ret;
>
> - if (!s->ioc) { /* Already closed */
> - return;
> - }
> -
> - if (s->reply.handle == 0) {
> - /* No reply already in flight. Fetch a header. It is possible
> - * that another thread has done the same thing in parallel, so
> - * the socket is not readable anymore.
> - */
> + for (;;) {
> + assert(s->reply.handle == 0);
> ret = nbd_receive_reply(s->ioc, &s->reply);
> - if (ret == -EAGAIN) {
> - return;
> - }
> if (ret < 0) {
> - s->reply.handle = 0;
> - goto fail;
> + break;
> }
> - }
> -
> - /* There's no need for a mutex on the receive side, because the
> - * handler acts as a synchronization point and ensures that only
> - * one coroutine is called until the reply finishes. */
> - i = HANDLE_TO_INDEX(s, s->reply.handle);
> - if (i >= MAX_NBD_REQUESTS) {
> - goto fail;
> - }
>
> - if (s->recv_coroutine[i]) {
> - qemu_coroutine_enter(s->recv_coroutine[i]);
> - return;
> - }
> -
> -fail:
> - nbd_teardown_connection(bs);
> -}
> + /* There's no need for a mutex on the receive side, because the
> + * handler acts as a synchronization point and ensures that only
> + * one coroutine is called until the reply finishes.
> + */
> + i = HANDLE_TO_INDEX(s, s->reply.handle);
> + if (i >= MAX_NBD_REQUESTS || !s->recv_coroutine[i]) {
> + break;
> + }
>
> -static void nbd_restart_write(void *opaque)
> -{
> - BlockDriverState *bs = opaque;
> + aio_co_wake(s->recv_coroutine[i]);
>
> - qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine);
> + /* We're woken up by the recv_coroutine itself. */
"Wait until we're woken by ..." ?
> + qemu_coroutine_yield();
> + }
> + s->read_reply_co = NULL;
> }
>
> static int nbd_co_send_request(BlockDriverState *bs,
> @@ -120,7 +102,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
> QEMUIOVector *qiov)
> {
> NBDClientSession *s = nbd_get_client_session(bs);
> - AioContext *aio_context;
> int rc, ret, i;
>
> qemu_co_mutex_lock(&s->send_mutex);
> @@ -141,11 +122,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
> return -EPIPE;
> }
>
> - s->send_coroutine = qemu_coroutine_self();
> - aio_context = bdrv_get_aio_context(bs);
> -
> - aio_set_fd_handler(aio_context, s->sioc->fd, false,
> - nbd_reply_ready, nbd_restart_write, NULL, bs);
> if (qiov) {
> qio_channel_set_cork(s->ioc, true);
> rc = nbd_send_request(s->ioc, request);
> @@ -160,9 +136,6 @@ static int nbd_co_send_request(BlockDriverState *bs,
> } else {
> rc = nbd_send_request(s->ioc, request);
> }
> - aio_set_fd_handler(aio_context, s->sioc->fd, false,
> - nbd_reply_ready, NULL, NULL, bs);
> - s->send_coroutine = NULL;
> qemu_co_mutex_unlock(&s->send_mutex);
> return rc;
> }
> @@ -174,8 +147,7 @@ static void nbd_co_receive_reply(NBDClientSession *s,
> {
> int ret;
>
> - /* Wait until we're woken up by the read handler. TODO: perhaps
> - * peek at the next reply and avoid yielding if it's ours? */
> + /* Wait until we're woken up by nbd_read_reply_entry. */
> qemu_coroutine_yield();
> *reply = s->reply;
> if (reply->handle != request->handle ||
> @@ -209,14 +181,18 @@ static void nbd_coroutine_start(NBDClientSession *s,
> /* s->recv_coroutine[i] is set as soon as we get the send_lock. */
> }
>
> -static void nbd_coroutine_end(NBDClientSession *s,
> +static void nbd_coroutine_end(BlockDriverState *bs,
> NBDRequest *request)
> {
> + NBDClientSession *s = nbd_get_client_session(bs);
> int i = HANDLE_TO_INDEX(s, request->handle);
> +
> s->recv_coroutine[i] = NULL;
> - if (s->in_flight-- == MAX_NBD_REQUESTS) {
> - qemu_co_queue_next(&s->free_sema);
> - }
> + s->in_flight--;
> + qemu_co_queue_next(&s->free_sema);
> +
> + /* Kick the read_reply_co to get the next reply. */
> + aio_co_wake(s->read_reply_co);
Can't s->read_reply_co be NULL? nbd_read_reply_entry unsets it. (Surprisingly
this file is rather unfamiliar to me, it's possible I'm missing something.)
> }
>
> int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
> void nbd_client_attach_aio_context(BlockDriverState *bs,
> AioContext *new_context)
> {
> - aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sioc->fd,
> - false, nbd_reply_ready, NULL, NULL, bs);
> + NBDClientSession *client = nbd_get_client_session(bs);
> + qio_channel_set_aio_context(QIO_CHANNEL(client->sioc), new_context);
> + aio_co_schedule(new_context, client->read_reply_co);
Like above, is client->read_reply_co possibly NULL?
> }
>
> void nbd_client_close(BlockDriverState *bs)
> @@ -434,7 +410,7 @@ int nbd_client_init(BlockDriverState *bs,
> /* Now that we're connected, set the socket to be non-blocking and
> * kick the reply mechanism. */
> qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
> -
> + client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
> nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
>
> logout("Established connection with NBD server\n");
> diff --git a/block/nbd-client.h b/block/nbd-client.h
> index f8d6006..8cdfc92 100644
> --- a/block/nbd-client.h
> +++ b/block/nbd-client.h
> @@ -25,7 +25,7 @@ typedef struct NBDClientSession {
>
> CoMutex send_mutex;
> CoQueue free_sema;
> - Coroutine *send_coroutine;
> + Coroutine *read_reply_co;
> int in_flight;
>
> Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
> diff --git a/nbd/client.c b/nbd/client.c
> index ffb0743..5c9dee3 100644
> --- a/nbd/client.c
> +++ b/nbd/client.c
> @@ -778,7 +778,7 @@ ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply)
> ssize_t ret;
>
> ret = read_sync(ioc, buf, sizeof(buf));
> - if (ret < 0) {
> + if (ret <= 0) {
Not sure why this belongs to this patch, but it also looks harmless.
> return ret;
> }
>
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads
2017-01-16 12:52 ` Fam Zheng
@ 2017-01-16 13:31 ` Paolo Bonzini
0 siblings, 0 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 13:31 UTC (permalink / raw)
To: Fam Zheng; +Cc: qemu-devel, stefanha
On 16/01/2017 13:52, Fam Zheng wrote:
> + /* Kick the read_reply_co to get the next reply. */
> + aio_co_wake(s->read_reply_co);
>
> Can't s->read_reply_co be NULL? nbd_read_reply_entry unsets it. (Surprisingly
> this file is rather unfamiliar to me, it's possible I'm missing something.)
Yes, that can happen depending on how the coroutines are scheduled when
the server goes down.
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads
2017-01-13 13:17 ` [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads Paolo Bonzini
2017-01-16 12:52 ` Fam Zheng
@ 2017-01-18 15:24 ` Stefan Hajnoczi
2017-01-18 16:43 ` Paolo Bonzini
1 sibling, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:24 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 309 bytes --]
On Fri, Jan 13, 2017 at 02:17:21PM +0100, Paolo Bonzini wrote:
> @@ -65,54 +67,34 @@ static void nbd_teardown_connection(BlockDriverState *bs)
> client->ioc = NULL;
> }
>
> -static void nbd_reply_ready(void *opaque)
> +static void nbd_read_reply_entry(void *opaque)
Please use coroutine_fn.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads
2017-01-18 15:24 ` Stefan Hajnoczi
@ 2017-01-18 16:43 ` Paolo Bonzini
0 siblings, 0 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-18 16:43 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: famz, qemu-devel
On 18/01/2017 16:24, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 02:17:21PM +0100, Paolo Bonzini wrote:
>> @@ -65,54 +67,34 @@ static void nbd_teardown_connection(BlockDriverState *bs)
>> client->ioc = NULL;
>> }
>>
>> -static void nbd_reply_ready(void *opaque)
>> +static void nbd_read_reply_entry(void *opaque)
>
> Please use coroutine_fn.
Yes, you requested that in the RFC as well.
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 07/16] coroutine-lock: reschedule coroutine on the AioContext it was running on
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (5 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 06/16] nbd: do not block on partial reply header reads Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:26 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 08/16] qed: introduce qed_aio_start_io and qed_aio_next_io_cb Paolo Bonzini
` (11 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
As a small step towards the introduction of multiqueue, we want
coroutines to remain on the same AioContext that started them,
unless they are moved explicitly with e.g. aio_co_schedule. This patch
avoids that coroutines switch AioContext when they use a CoMutex.
For now it does not make much of a difference, because the CoMutex
is not thread-safe and the AioContext itself is used to protect the
CoMutex from concurrent access. However, this is going to change.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
util/qemu-coroutine-lock.c | 5 ++---
util/trace-events | 1 -
2 files changed, 2 insertions(+), 4 deletions(-)
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
index 14cf9ce..e6afd1a 100644
--- a/util/qemu-coroutine-lock.c
+++ b/util/qemu-coroutine-lock.c
@@ -27,6 +27,7 @@
#include "qemu/coroutine.h"
#include "qemu/coroutine_int.h"
#include "qemu/queue.h"
+#include "block/aio.h"
#include "trace.h"
void qemu_co_queue_init(CoQueue *queue)
@@ -63,7 +64,6 @@ void qemu_co_queue_run_restart(Coroutine *co)
static bool qemu_co_queue_do_restart(CoQueue *queue, bool single)
{
- Coroutine *self = qemu_coroutine_self();
Coroutine *next;
if (QSIMPLEQ_EMPTY(&queue->entries)) {
@@ -72,8 +72,7 @@ static bool qemu_co_queue_do_restart(CoQueue *queue, bool single)
while ((next = QSIMPLEQ_FIRST(&queue->entries)) != NULL) {
QSIMPLEQ_REMOVE_HEAD(&queue->entries, co_queue_next);
- QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, next, co_queue_next);
- trace_qemu_co_queue_next(next);
+ aio_co_wake(next);
if (single) {
break;
}
diff --git a/util/trace-events b/util/trace-events
index 2b8aa30..65705c4 100644
--- a/util/trace-events
+++ b/util/trace-events
@@ -13,7 +13,6 @@ qemu_coroutine_terminate(void *co) "self %p"
# util/qemu-coroutine-lock.c
qemu_co_queue_run_restart(void *co) "co %p"
-qemu_co_queue_next(void *nxt) "next %p"
qemu_co_mutex_lock_entry(void *mutex, void *self) "mutex %p self %p"
qemu_co_mutex_lock_return(void *mutex, void *self) "mutex %p self %p"
qemu_co_mutex_unlock_entry(void *mutex, void *self) "mutex %p self %p"
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 07/16] coroutine-lock: reschedule coroutine on the AioContext it was running on
2017-01-13 13:17 ` [Qemu-devel] [PATCH 07/16] coroutine-lock: reschedule coroutine on the AioContext it was running on Paolo Bonzini
@ 2017-01-18 15:26 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:26 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 802 bytes --]
On Fri, Jan 13, 2017 at 02:17:22PM +0100, Paolo Bonzini wrote:
> As a small step towards the introduction of multiqueue, we want
> coroutines to remain on the same AioContext that started them,
> unless they are moved explicitly with e.g. aio_co_schedule. This patch
> avoids that coroutines switch AioContext when they use a CoMutex.
> For now it does not make much of a difference, because the CoMutex
> is not thread-safe and the AioContext itself is used to protect the
> CoMutex from concurrent access. However, this is going to change.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> util/qemu-coroutine-lock.c | 5 ++---
> util/trace-events | 1 -
> 2 files changed, 2 insertions(+), 4 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 08/16] qed: introduce qed_aio_start_io and qed_aio_next_io_cb
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (6 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 07/16] coroutine-lock: reschedule coroutine on the AioContext it was running on Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:27 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 09/16] aio: push aio_context_acquire/release down to dispatching Paolo Bonzini
` (10 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
qed_aio_start_io and qed_aio_next_io will not have to acquire/release
the AioContext, while qed_aio_next_io_cb will. Split the functionality
and gain a little type-safety in the process.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
block/qed.c | 39 +++++++++++++++++++++++++--------------
1 file changed, 25 insertions(+), 14 deletions(-)
diff --git a/block/qed.c b/block/qed.c
index 1a7ef0a..7f1c508 100644
--- a/block/qed.c
+++ b/block/qed.c
@@ -273,7 +273,19 @@ static CachedL2Table *qed_new_l2_table(BDRVQEDState *s)
return l2_table;
}
-static void qed_aio_next_io(void *opaque, int ret);
+static void qed_aio_next_io(QEDAIOCB *acb, int ret);
+
+static void qed_aio_start_io(QEDAIOCB *acb)
+{
+ qed_aio_next_io(acb, 0);
+}
+
+static void qed_aio_next_io_cb(void *opaque, int ret)
+{
+ QEDAIOCB *acb = opaque;
+
+ qed_aio_next_io(acb, ret);
+}
static void qed_plug_allocating_write_reqs(BDRVQEDState *s)
{
@@ -292,7 +304,7 @@ static void qed_unplug_allocating_write_reqs(BDRVQEDState *s)
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
if (acb) {
- qed_aio_next_io(acb, 0);
+ qed_aio_start_io(acb);
}
}
@@ -959,7 +971,7 @@ static void qed_aio_complete(QEDAIOCB *acb, int ret)
QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next);
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
if (acb) {
- qed_aio_next_io(acb, 0);
+ qed_aio_start_io(acb);
} else if (s->header.features & QED_F_NEED_CHECK) {
qed_start_need_check_timer(s);
}
@@ -984,7 +996,7 @@ static void qed_commit_l2_update(void *opaque, int ret)
acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
assert(acb->request.l2_table != NULL);
- qed_aio_next_io(opaque, ret);
+ qed_aio_next_io(acb, ret);
}
/**
@@ -1032,11 +1044,11 @@ static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset)
if (need_alloc) {
/* Write out the whole new L2 table */
qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true,
- qed_aio_write_l1_update, acb);
+ qed_aio_write_l1_update, acb);
} else {
/* Write out only the updated part of the L2 table */
qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false,
- qed_aio_next_io, acb);
+ qed_aio_next_io_cb, acb);
}
return;
@@ -1088,7 +1100,7 @@ static void qed_aio_write_main(void *opaque, int ret)
}
if (acb->find_cluster_ret == QED_CLUSTER_FOUND) {
- next_fn = qed_aio_next_io;
+ next_fn = qed_aio_next_io_cb;
} else {
if (s->bs->backing) {
next_fn = qed_aio_write_flush_before_l2_update;
@@ -1201,7 +1213,7 @@ static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len)
if (acb->flags & QED_AIOCB_ZERO) {
/* Skip ahead if the clusters are already zero */
if (acb->find_cluster_ret == QED_CLUSTER_ZERO) {
- qed_aio_next_io(acb, 0);
+ qed_aio_start_io(acb);
return;
}
@@ -1321,18 +1333,18 @@ static void qed_aio_read_data(void *opaque, int ret,
/* Handle zero cluster and backing file reads */
if (ret == QED_CLUSTER_ZERO) {
qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size);
- qed_aio_next_io(acb, 0);
+ qed_aio_start_io(acb);
return;
} else if (ret != QED_CLUSTER_FOUND) {
qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov,
- &acb->backing_qiov, qed_aio_next_io, acb);
+ &acb->backing_qiov, qed_aio_next_io_cb, acb);
return;
}
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE,
&acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE,
- qed_aio_next_io, acb);
+ qed_aio_next_io_cb, acb);
return;
err:
@@ -1342,9 +1354,8 @@ err:
/**
* Begin next I/O or complete the request
*/
-static void qed_aio_next_io(void *opaque, int ret)
+static void qed_aio_next_io(QEDAIOCB *acb, int ret)
{
- QEDAIOCB *acb = opaque;
BDRVQEDState *s = acb_to_s(acb);
QEDFindClusterFunc *io_fn = (acb->flags & QED_AIOCB_WRITE) ?
qed_aio_write_data : qed_aio_read_data;
@@ -1400,7 +1411,7 @@ static BlockAIOCB *qed_aio_setup(BlockDriverState *bs,
qemu_iovec_init(&acb->cur_qiov, qiov->niov);
/* Start request */
- qed_aio_next_io(acb, 0);
+ qed_aio_start_io(acb);
return &acb->common;
}
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 08/16] qed: introduce qed_aio_start_io and qed_aio_next_io_cb
2017-01-13 13:17 ` [Qemu-devel] [PATCH 08/16] qed: introduce qed_aio_start_io and qed_aio_next_io_cb Paolo Bonzini
@ 2017-01-18 15:27 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:27 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 495 bytes --]
On Fri, Jan 13, 2017 at 02:17:23PM +0100, Paolo Bonzini wrote:
> qed_aio_start_io and qed_aio_next_io will not have to acquire/release
> the AioContext, while qed_aio_next_io_cb will. Split the functionality
> and gain a little type-safety in the process.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> block/qed.c | 39 +++++++++++++++++++++++++--------------
> 1 file changed, 25 insertions(+), 14 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 09/16] aio: push aio_context_acquire/release down to dispatching
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (7 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 08/16] qed: introduce qed_aio_start_io and qed_aio_next_io_cb Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:29 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it Paolo Bonzini
` (9 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
The AioContext data structures are now protected by list_lock and/or
they are walked with FOREACH_RCU primitives. There is no need anymore
to acquire the AioContext for the entire duration of aio_dispatch.
Instead, just acquire it before and after invoking the callbacks.
The next step is then to push it further down.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
aio-posix.c | 25 +++++++++++--------------
aio-win32.c | 15 +++++++--------
async.c | 2 ++
3 files changed, 20 insertions(+), 22 deletions(-)
diff --git a/aio-posix.c b/aio-posix.c
index 9453d83..3fd64fb 100644
--- a/aio-posix.c
+++ b/aio-posix.c
@@ -402,7 +402,9 @@ static bool aio_dispatch_handlers(AioContext *ctx)
(revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) &&
aio_node_check(ctx, node->is_external) &&
node->io_read) {
+ aio_context_acquire(ctx);
node->io_read(node->opaque);
+ aio_context_release(ctx);
/* aio_notify() does not count as progress */
if (node->opaque != &ctx->notifier) {
@@ -413,7 +415,9 @@ static bool aio_dispatch_handlers(AioContext *ctx)
(revents & (G_IO_OUT | G_IO_ERR)) &&
aio_node_check(ctx, node->is_external) &&
node->io_write) {
+ aio_context_acquire(ctx);
node->io_write(node->opaque);
+ aio_context_release(ctx);
progress = true;
}
@@ -450,7 +454,9 @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
}
/* Run our timers */
+ aio_context_acquire(ctx);
progress |= timerlistgroup_run_timers(&ctx->tlg);
+ aio_context_release(ctx);
return progress;
}
@@ -596,9 +602,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
int64_t timeout;
int64_t start = 0;
- aio_context_acquire(ctx);
- progress = false;
-
/* aio_notify can avoid the expensive event_notifier_set if
* everything (file descriptors, bottom halves, timers) will
* be re-evaluated before the next blocking poll(). This is
@@ -616,9 +619,11 @@ bool aio_poll(AioContext *ctx, bool blocking)
start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
}
- if (try_poll_mode(ctx, blocking)) {
- progress = true;
- } else {
+ aio_context_acquire(ctx);
+ progress = try_poll_mode(ctx, blocking);
+ aio_context_release(ctx);
+
+ if (!progress) {
assert(npfd == 0);
/* fill pollfds */
@@ -635,9 +640,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
timeout = blocking ? aio_compute_timeout(ctx) : 0;
/* wait until next event */
- if (timeout) {
- aio_context_release(ctx);
- }
if (aio_epoll_check_poll(ctx, pollfds, npfd, timeout)) {
AioHandler epoll_handler;
@@ -649,9 +651,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
} else {
ret = qemu_poll_ns(pollfds, npfd, timeout);
}
- if (timeout) {
- aio_context_acquire(ctx);
- }
}
if (blocking) {
@@ -716,8 +715,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
progress = true;
}
- aio_context_release(ctx);
-
return progress;
}
diff --git a/aio-win32.c b/aio-win32.c
index 900524c..ab6d0e5 100644
--- a/aio-win32.c
+++ b/aio-win32.c
@@ -266,7 +266,9 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
(revents || event_notifier_get_handle(node->e) == event) &&
node->io_notify) {
node->pfd.revents = 0;
+ aio_context_acquire(ctx);
node->io_notify(node->e);
+ aio_context_release(ctx);
/* aio_notify() does not count as progress */
if (node->e != &ctx->notifier) {
@@ -278,11 +280,15 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
(node->io_read || node->io_write)) {
node->pfd.revents = 0;
if ((revents & G_IO_IN) && node->io_read) {
+ aio_context_acquire(ctx);
node->io_read(node->opaque);
+ aio_context_release(ctx);
progress = true;
}
if ((revents & G_IO_OUT) && node->io_write) {
+ aio_context_acquire(ctx);
node->io_write(node->opaque);
+ aio_context_release(ctx);
progress = true;
}
@@ -329,7 +335,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
int count;
int timeout;
- aio_context_acquire(ctx);
progress = false;
/* aio_notify can avoid the expensive event_notifier_set if
@@ -371,17 +376,11 @@ bool aio_poll(AioContext *ctx, bool blocking)
timeout = blocking && !have_select_revents
? qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)) : 0;
- if (timeout) {
- aio_context_release(ctx);
- }
ret = WaitForMultipleObjects(count, events, FALSE, timeout);
if (blocking) {
assert(first);
atomic_sub(&ctx->notify_me, 2);
}
- if (timeout) {
- aio_context_acquire(ctx);
- }
if (first) {
aio_notify_accept(ctx);
@@ -404,8 +403,8 @@ bool aio_poll(AioContext *ctx, bool blocking)
progress |= aio_dispatch_handlers(ctx, event);
} while (count > 0);
+ aio_context_acquire(ctx);
progress |= timerlistgroup_run_timers(&ctx->tlg);
-
aio_context_release(ctx);
return progress;
}
diff --git a/async.c b/async.c
index 1338682..ccb70e7 100644
--- a/async.c
+++ b/async.c
@@ -113,7 +113,9 @@ int aio_bh_poll(AioContext *ctx)
ret = 1;
}
bh->idle = 0;
+ aio_context_acquire(ctx);
aio_bh_call(bh);
+ aio_context_release(ctx);
}
if (bh->deleted) {
deleted = true;
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 09/16] aio: push aio_context_acquire/release down to dispatching
2017-01-13 13:17 ` [Qemu-devel] [PATCH 09/16] aio: push aio_context_acquire/release down to dispatching Paolo Bonzini
@ 2017-01-18 15:29 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:29 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 735 bytes --]
On Fri, Jan 13, 2017 at 02:17:24PM +0100, Paolo Bonzini wrote:
> The AioContext data structures are now protected by list_lock and/or
> they are walked with FOREACH_RCU primitives. There is no need anymore
> to acquire the AioContext for the entire duration of aio_dispatch.
> Instead, just acquire it before and after invoking the callbacks.
> The next step is then to push it further down.
>
> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> aio-posix.c | 25 +++++++++++--------------
> aio-win32.c | 15 +++++++--------
> async.c | 2 ++
> 3 files changed, 20 insertions(+), 22 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (8 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 09/16] aio: push aio_context_acquire/release down to dispatching Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 13:07 ` Fam Zheng
2017-01-18 15:43 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks " Paolo Bonzini
` (8 subsequent siblings)
18 siblings, 2 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
aio-posix.c | 2 --
aio-win32.c | 2 --
block/curl.c | 2 ++
block/iscsi.c | 8 ++++++--
block/null.c | 4 ++++
block/qed.c | 12 ++++++++++++
block/qed.h | 3 +++
block/throttle-groups.c | 2 ++
util/qemu-coroutine-sleep.c | 2 +-
9 files changed, 30 insertions(+), 7 deletions(-)
diff --git a/aio-posix.c b/aio-posix.c
index 3fd64fb..8d79cf3 100644
--- a/aio-posix.c
+++ b/aio-posix.c
@@ -454,9 +454,7 @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
}
/* Run our timers */
- aio_context_acquire(ctx);
progress |= timerlistgroup_run_timers(&ctx->tlg);
- aio_context_release(ctx);
return progress;
}
diff --git a/aio-win32.c b/aio-win32.c
index ab6d0e5..810e1c6 100644
--- a/aio-win32.c
+++ b/aio-win32.c
@@ -403,9 +403,7 @@ bool aio_poll(AioContext *ctx, bool blocking)
progress |= aio_dispatch_handlers(ctx, event);
} while (count > 0);
- aio_context_acquire(ctx);
progress |= timerlistgroup_run_timers(&ctx->tlg);
- aio_context_release(ctx);
return progress;
}
diff --git a/block/curl.c b/block/curl.c
index 792fef8..65e6da1 100644
--- a/block/curl.c
+++ b/block/curl.c
@@ -424,9 +424,11 @@ static void curl_multi_timeout_do(void *arg)
return;
}
+ aio_context_acquire(s->aio_context);
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
curl_multi_check_completion(s);
+ aio_context_release(s->aio_context);
#else
abort();
#endif
diff --git a/block/iscsi.c b/block/iscsi.c
index 6aeeb9e..e1f10d6 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -174,7 +174,7 @@ static void iscsi_retry_timer_expired(void *opaque)
struct IscsiTask *iTask = opaque;
iTask->complete = 1;
if (iTask->co) {
- qemu_coroutine_enter(iTask->co);
+ aio_co_wake(iTask->co);
}
}
@@ -1388,16 +1388,20 @@ static void iscsi_nop_timed_event(void *opaque)
{
IscsiLun *iscsilun = opaque;
+ aio_context_acquire(iscsilun->aio_context);
if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
error_report("iSCSI: NOP timeout. Reconnecting...");
iscsilun->request_timed_out = true;
} else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
- return;
+ goto out;
}
timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
iscsi_set_events(iscsilun);
+
+out:
+ aio_context_release(iscsilun->aio_context);
}
static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
diff --git a/block/null.c b/block/null.c
index b300390..356209a 100644
--- a/block/null.c
+++ b/block/null.c
@@ -141,7 +141,11 @@ static void null_bh_cb(void *opaque)
static void null_timer_cb(void *opaque)
{
NullAIOCB *acb = opaque;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
+
+ aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, 0);
+ aio_context_release(ctx);
timer_deinit(&acb->timer);
qemu_aio_unref(acb);
}
diff --git a/block/qed.c b/block/qed.c
index 7f1c508..a21d025 100644
--- a/block/qed.c
+++ b/block/qed.c
@@ -345,10 +345,22 @@ static void qed_need_check_timer_cb(void *opaque)
trace_qed_need_check_timer_cb(s);
+ qed_acquire(s);
qed_plug_allocating_write_reqs(s);
/* Ensure writes are on disk before clearing flag */
bdrv_aio_flush(s->bs->file->bs, qed_clear_need_check, s);
+ qed_release(s);
+}
+
+void qed_acquire(BDRVQEDState *s)
+{
+ aio_context_acquire(bdrv_get_aio_context(s->bs));
+}
+
+void qed_release(BDRVQEDState *s)
+{
+ aio_context_release(bdrv_get_aio_context(s->bs));
}
static void qed_start_need_check_timer(BDRVQEDState *s)
diff --git a/block/qed.h b/block/qed.h
index 9676ab9..ce8c314 100644
--- a/block/qed.h
+++ b/block/qed.h
@@ -198,6 +198,9 @@ enum {
*/
typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
+void qed_acquire(BDRVQEDState *s);
+void qed_release(BDRVQEDState *s);
+
/**
* Generic callback for chaining async callbacks
*/
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
index 17b2efb..aade5de 100644
--- a/block/throttle-groups.c
+++ b/block/throttle-groups.c
@@ -416,7 +416,9 @@ static void timer_cb(BlockBackend *blk, bool is_write)
qemu_mutex_unlock(&tg->lock);
/* Run the request that was waiting for this timer */
+ aio_context_acquire(blk_get_aio_context(blk));
empty_queue = !qemu_co_enter_next(&blkp->throttled_reqs[is_write]);
+ aio_context_release(blk_get_aio_context(blk));
/* If the request queue was empty then we have to take care of
* scheduling the next one */
diff --git a/util/qemu-coroutine-sleep.c b/util/qemu-coroutine-sleep.c
index 25de3ed..9c56550 100644
--- a/util/qemu-coroutine-sleep.c
+++ b/util/qemu-coroutine-sleep.c
@@ -25,7 +25,7 @@ static void co_sleep_cb(void *opaque)
{
CoSleepCB *sleep_cb = opaque;
- qemu_coroutine_enter(sleep_cb->co);
+ aio_co_wake(sleep_cb->co);
}
void coroutine_fn co_aio_sleep_ns(AioContext *ctx, QEMUClockType type,
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-13 13:17 ` [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it Paolo Bonzini
@ 2017-01-16 13:07 ` Fam Zheng
2017-01-16 13:32 ` Paolo Bonzini
2017-01-18 15:43 ` Stefan Hajnoczi
1 sibling, 1 reply; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 13:07 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> diff --git a/block/qed.c b/block/qed.c
> index 7f1c508..a21d025 100644
> --- a/block/qed.c
> +++ b/block/qed.c
> @@ -345,10 +345,22 @@ static void qed_need_check_timer_cb(void *opaque)
>
> trace_qed_need_check_timer_cb(s);
>
> + qed_acquire(s);
> qed_plug_allocating_write_reqs(s);
>
> /* Ensure writes are on disk before clearing flag */
> bdrv_aio_flush(s->bs->file->bs, qed_clear_need_check, s);
> + qed_release(s);
> +}
> +
> +void qed_acquire(BDRVQEDState *s)
> +{
> + aio_context_acquire(bdrv_get_aio_context(s->bs));
> +}
> +
> +void qed_release(BDRVQEDState *s)
> +{
> + aio_context_release(bdrv_get_aio_context(s->bs));
> }
>
> static void qed_start_need_check_timer(BDRVQEDState *s)
> diff --git a/block/qed.h b/block/qed.h
> index 9676ab9..ce8c314 100644
> --- a/block/qed.h
> +++ b/block/qed.h
> @@ -198,6 +198,9 @@ enum {
> */
> typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
>
> +void qed_acquire(BDRVQEDState *s);
> +void qed_release(BDRVQEDState *s);
> +
Why cannot these be local (static) functions, in block/qed.c?
> /**
> * Generic callback for chaining async callbacks
> */
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-16 13:07 ` Fam Zheng
@ 2017-01-16 13:32 ` Paolo Bonzini
2017-01-16 13:50 ` Fam Zheng
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 13:32 UTC (permalink / raw)
To: Fam Zheng; +Cc: qemu-devel, stefanha
On 16/01/2017 14:07, Fam Zheng wrote:
> On Fri, 01/13 14:17, Paolo Bonzini wrote:
>> diff --git a/block/qed.c b/block/qed.c
>> index 7f1c508..a21d025 100644
>> --- a/block/qed.c
>> +++ b/block/qed.c
>> @@ -345,10 +345,22 @@ static void qed_need_check_timer_cb(void *opaque)
>>
>> trace_qed_need_check_timer_cb(s);
>>
>> + qed_acquire(s);
>> qed_plug_allocating_write_reqs(s);
>>
>> /* Ensure writes are on disk before clearing flag */
>> bdrv_aio_flush(s->bs->file->bs, qed_clear_need_check, s);
>> + qed_release(s);
>> +}
>> +
>> +void qed_acquire(BDRVQEDState *s)
>> +{
>> + aio_context_acquire(bdrv_get_aio_context(s->bs));
>> +}
>> +
>> +void qed_release(BDRVQEDState *s)
>> +{
>> + aio_context_release(bdrv_get_aio_context(s->bs));
>> }
>>
>> static void qed_start_need_check_timer(BDRVQEDState *s)
>> diff --git a/block/qed.h b/block/qed.h
>> index 9676ab9..ce8c314 100644
>> --- a/block/qed.h
>> +++ b/block/qed.h
>> @@ -198,6 +198,9 @@ enum {
>> */
>> typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
>>
>> +void qed_acquire(BDRVQEDState *s);
>> +void qed_release(BDRVQEDState *s);
>> +
>
> Why cannot these be local (static) functions, in block/qed.c?
Patch 13 uses them elsewhere. Should I put them in a separate patch?
Paolo
>> /**
>> * Generic callback for chaining async callbacks
>> */
>
> Fam
>
>
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-13 13:17 ` [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it Paolo Bonzini
2017-01-16 13:07 ` Fam Zheng
@ 2017-01-18 15:43 ` Stefan Hajnoczi
2017-01-18 16:44 ` Paolo Bonzini
1 sibling, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:43 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 595 bytes --]
On Fri, Jan 13, 2017 at 02:17:25PM +0100, Paolo Bonzini wrote:
> diff --git a/block/null.c b/block/null.c
> index b300390..356209a 100644
> --- a/block/null.c
> +++ b/block/null.c
> @@ -141,7 +141,11 @@ static void null_bh_cb(void *opaque)
> static void null_timer_cb(void *opaque)
> {
> NullAIOCB *acb = opaque;
> + AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
> +
> + aio_context_acquire(ctx);
> acb->common.cb(acb->common.opaque, 0);
> + aio_context_release(ctx);
> timer_deinit(&acb->timer);
> qemu_aio_unref(acb);
Is qemu_aio_unref() thread-safe?
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-18 15:43 ` Stefan Hajnoczi
@ 2017-01-18 16:44 ` Paolo Bonzini
2017-01-19 16:59 ` Stefan Hajnoczi
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-18 16:44 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 1083 bytes --]
On 18/01/2017 16:43, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 02:17:25PM +0100, Paolo Bonzini wrote:
>> diff --git a/block/null.c b/block/null.c
>> index b300390..356209a 100644
>> --- a/block/null.c
>> +++ b/block/null.c
>> @@ -141,7 +141,11 @@ static void null_bh_cb(void *opaque)
>> static void null_timer_cb(void *opaque)
>> {
>> NullAIOCB *acb = opaque;
>> + AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
>> +
>> + aio_context_acquire(ctx);
>> acb->common.cb(acb->common.opaque, 0);
>> + aio_context_release(ctx);
>> timer_deinit(&acb->timer);
>> qemu_aio_unref(acb);
>
> Is qemu_aio_unref() thread-safe?
qemu_aio_ref()/qemu_aio_unref() is only used by bdrv_aio_cancel, which
in turn is not used by dataplane. So in the multithreaded case
qemu_aio_unref() is effectively qemu_aio_free().
Probably needs more documentation, or a different implementation of
bdrv_aio_cancel (e.g. replacing the reference counting with a
NotifierList of some kind). Let me know what you prefer for v2.
Paolo
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it
2017-01-18 16:44 ` Paolo Bonzini
@ 2017-01-19 16:59 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-19 16:59 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 1351 bytes --]
On Wed, Jan 18, 2017 at 05:44:24PM +0100, Paolo Bonzini wrote:
> On 18/01/2017 16:43, Stefan Hajnoczi wrote:
> > On Fri, Jan 13, 2017 at 02:17:25PM +0100, Paolo Bonzini wrote:
> >> diff --git a/block/null.c b/block/null.c
> >> index b300390..356209a 100644
> >> --- a/block/null.c
> >> +++ b/block/null.c
> >> @@ -141,7 +141,11 @@ static void null_bh_cb(void *opaque)
> >> static void null_timer_cb(void *opaque)
> >> {
> >> NullAIOCB *acb = opaque;
> >> + AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
> >> +
> >> + aio_context_acquire(ctx);
> >> acb->common.cb(acb->common.opaque, 0);
> >> + aio_context_release(ctx);
> >> timer_deinit(&acb->timer);
> >> qemu_aio_unref(acb);
> >
> > Is qemu_aio_unref() thread-safe?
>
> qemu_aio_ref()/qemu_aio_unref() is only used by bdrv_aio_cancel, which
> in turn is not used by dataplane. So in the multithreaded case
> qemu_aio_unref() is effectively qemu_aio_free().
>
> Probably needs more documentation, or a different implementation of
> bdrv_aio_cancel (e.g. replacing the reference counting with a
> NotifierList of some kind). Let me know what you prefer for v2.
Documentation is fine. I just checked and see that virtio-scsi
dataplane uses blk_aio_cancel_async() so the aio refcount is never
touched - no race.
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks that need it
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (9 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 10/16] block: explicitly acquire aiocontext in timers that need it Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-16 13:36 ` Fam Zheng
2017-01-18 15:49 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 12/16] block: explicitly acquire aiocontext in bottom halves " Paolo Bonzini
` (7 subsequent siblings)
18 siblings, 2 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
This covers both file descriptor callbacks and polling callbacks,
since they execute related code.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
aio-posix.c | 7 -------
aio-win32.c | 6 ------
block/curl.c | 16 +++++++++++++---
block/iscsi.c | 4 ++++
block/linux-aio.c | 4 ++++
block/nfs.c | 6 ++++++
block/sheepdog.c | 29 +++++++++++++++--------------
block/ssh.c | 29 +++++++++--------------------
block/win32-aio.c | 10 ++++++----
hw/block/virtio-blk.c | 5 ++++-
hw/scsi/virtio-scsi.c | 6 ++++++
nbd/server.c | 4 ++++
12 files changed, 71 insertions(+), 55 deletions(-)
diff --git a/aio-posix.c b/aio-posix.c
index 8d79cf3..6beebcd 100644
--- a/aio-posix.c
+++ b/aio-posix.c
@@ -402,9 +402,7 @@ static bool aio_dispatch_handlers(AioContext *ctx)
(revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) &&
aio_node_check(ctx, node->is_external) &&
node->io_read) {
- aio_context_acquire(ctx);
node->io_read(node->opaque);
- aio_context_release(ctx);
/* aio_notify() does not count as progress */
if (node->opaque != &ctx->notifier) {
@@ -415,9 +413,7 @@ static bool aio_dispatch_handlers(AioContext *ctx)
(revents & (G_IO_OUT | G_IO_ERR)) &&
aio_node_check(ctx, node->is_external) &&
node->io_write) {
- aio_context_acquire(ctx);
node->io_write(node->opaque);
- aio_context_release(ctx);
progress = true;
}
@@ -617,10 +613,7 @@ bool aio_poll(AioContext *ctx, bool blocking)
start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
}
- aio_context_acquire(ctx);
progress = try_poll_mode(ctx, blocking);
- aio_context_release(ctx);
-
if (!progress) {
assert(npfd == 0);
diff --git a/aio-win32.c b/aio-win32.c
index 810e1c6..20b63ce 100644
--- a/aio-win32.c
+++ b/aio-win32.c
@@ -266,9 +266,7 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
(revents || event_notifier_get_handle(node->e) == event) &&
node->io_notify) {
node->pfd.revents = 0;
- aio_context_acquire(ctx);
node->io_notify(node->e);
- aio_context_release(ctx);
/* aio_notify() does not count as progress */
if (node->e != &ctx->notifier) {
@@ -280,15 +278,11 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
(node->io_read || node->io_write)) {
node->pfd.revents = 0;
if ((revents & G_IO_IN) && node->io_read) {
- aio_context_acquire(ctx);
node->io_read(node->opaque);
- aio_context_release(ctx);
progress = true;
}
if ((revents & G_IO_OUT) && node->io_write) {
- aio_context_acquire(ctx);
node->io_write(node->opaque);
- aio_context_release(ctx);
progress = true;
}
diff --git a/block/curl.c b/block/curl.c
index 65e6da1..05b9ca3 100644
--- a/block/curl.c
+++ b/block/curl.c
@@ -386,9 +386,8 @@ static void curl_multi_check_completion(BDRVCURLState *s)
}
}
-static void curl_multi_do(void *arg)
+static void curl_multi_do_locked(CURLState *s)
{
- CURLState *s = (CURLState *)arg;
CURLSocket *socket, *next_socket;
int running;
int r;
@@ -406,12 +405,23 @@ static void curl_multi_do(void *arg)
}
}
+static void curl_multi_do(void *arg)
+{
+ CURLState *s = (CURLState *)arg;
+
+ aio_context_acquire(s->s->aio_context);
+ curl_multi_do_locked(s);
+ aio_context_release(s->s->aio_context);
+}
+
static void curl_multi_read(void *arg)
{
CURLState *s = (CURLState *)arg;
- curl_multi_do(arg);
+ aio_context_acquire(s->s->aio_context);
+ curl_multi_do_locked(s);
curl_multi_check_completion(s->s);
+ aio_context_release(s->s->aio_context);
}
static void curl_multi_timeout_do(void *arg)
diff --git a/block/iscsi.c b/block/iscsi.c
index e1f10d6..54d1381 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -394,8 +394,10 @@ iscsi_process_read(void *arg)
IscsiLun *iscsilun = arg;
struct iscsi_context *iscsi = iscsilun->iscsi;
+ aio_context_acquire(iscsilun->aio_context);
iscsi_service(iscsi, POLLIN);
iscsi_set_events(iscsilun);
+ aio_context_release(iscsilun->aio_context);
}
static void
@@ -404,8 +406,10 @@ iscsi_process_write(void *arg)
IscsiLun *iscsilun = arg;
struct iscsi_context *iscsi = iscsilun->iscsi;
+ aio_context_acquire(iscsilun->aio_context);
iscsi_service(iscsi, POLLOUT);
iscsi_set_events(iscsilun);
+ aio_context_release(iscsilun->aio_context);
}
static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
diff --git a/block/linux-aio.c b/block/linux-aio.c
index 03ab741..277c016 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -251,7 +251,9 @@ static void qemu_laio_completion_cb(EventNotifier *e)
LinuxAioState *s = container_of(e, LinuxAioState, e);
if (event_notifier_test_and_clear(&s->e)) {
+ aio_context_acquire(s->aio_context);
qemu_laio_process_completions_and_submit(s);
+ aio_context_release(s->aio_context);
}
}
@@ -265,7 +267,9 @@ static bool qemu_laio_poll_cb(void *opaque)
return false;
}
+ aio_context_acquire(s->aio_context);
qemu_laio_process_completions_and_submit(s);
+ aio_context_release(s->aio_context);
return true;
}
diff --git a/block/nfs.c b/block/nfs.c
index a564340..803faf9 100644
--- a/block/nfs.c
+++ b/block/nfs.c
@@ -207,15 +207,21 @@ static void nfs_set_events(NFSClient *client)
static void nfs_process_read(void *arg)
{
NFSClient *client = arg;
+
+ aio_context_acquire(client->aio_context);
nfs_service(client->context, POLLIN);
nfs_set_events(client);
+ aio_context_release(client->aio_context);
}
static void nfs_process_write(void *arg)
{
NFSClient *client = arg;
+
+ aio_context_acquire(client->aio_context);
nfs_service(client->context, POLLOUT);
nfs_set_events(client);
+ aio_context_release(client->aio_context);
}
static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
diff --git a/block/sheepdog.c b/block/sheepdog.c
index f757157..32c4e4c 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -575,13 +575,6 @@ static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
return ret;
}
-static void restart_co_req(void *opaque)
-{
- Coroutine *co = opaque;
-
- qemu_coroutine_enter(co);
-}
-
typedef struct SheepdogReqCo {
int sockfd;
BlockDriverState *bs;
@@ -592,12 +585,19 @@ typedef struct SheepdogReqCo {
unsigned int *rlen;
int ret;
bool finished;
+ Coroutine *co;
} SheepdogReqCo;
+static void restart_co_req(void *opaque)
+{
+ SheepdogReqCo *srco = opaque;
+
+ aio_co_wake(srco->co);
+}
+
static coroutine_fn void do_co_req(void *opaque)
{
int ret;
- Coroutine *co;
SheepdogReqCo *srco = opaque;
int sockfd = srco->sockfd;
SheepdogReq *hdr = srco->hdr;
@@ -605,9 +605,9 @@ static coroutine_fn void do_co_req(void *opaque)
unsigned int *wlen = srco->wlen;
unsigned int *rlen = srco->rlen;
- co = qemu_coroutine_self();
+ srco->co = qemu_coroutine_self();
aio_set_fd_handler(srco->aio_context, sockfd, false,
- NULL, restart_co_req, NULL, co);
+ NULL, restart_co_req, NULL, srco);
ret = send_co_req(sockfd, hdr, data, wlen);
if (ret < 0) {
@@ -615,7 +615,7 @@ static coroutine_fn void do_co_req(void *opaque)
}
aio_set_fd_handler(srco->aio_context, sockfd, false,
- restart_co_req, NULL, NULL, co);
+ restart_co_req, NULL, NULL, srco);
ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
if (ret != sizeof(*hdr)) {
@@ -643,6 +643,7 @@ out:
aio_set_fd_handler(srco->aio_context, sockfd, false,
NULL, NULL, NULL, NULL);
+ srco->co = NULL;
srco->ret = ret;
srco->finished = true;
if (srco->bs) {
@@ -866,7 +867,7 @@ static void coroutine_fn aio_read_response(void *opaque)
* We've finished all requests which belong to the AIOCB, so
* we can switch back to sd_co_readv/writev now.
*/
- qemu_coroutine_enter(acb->coroutine);
+ aio_co_wake(acb->coroutine);
}
return;
@@ -883,14 +884,14 @@ static void co_read_response(void *opaque)
s->co_recv = qemu_coroutine_create(aio_read_response, opaque);
}
- qemu_coroutine_enter(s->co_recv);
+ aio_co_wake(s->co_recv);
}
static void co_write_request(void *opaque)
{
BDRVSheepdogState *s = opaque;
- qemu_coroutine_enter(s->co_send);
+ aio_co_wake(s->co_send);
}
/*
diff --git a/block/ssh.c b/block/ssh.c
index e0edf20..835932e 100644
--- a/block/ssh.c
+++ b/block/ssh.c
@@ -889,10 +889,14 @@ static void restart_coroutine(void *opaque)
DPRINTF("co=%p", co);
- qemu_coroutine_enter(co);
+ aio_co_wake(co);
}
-static coroutine_fn void set_fd_handler(BDRVSSHState *s, BlockDriverState *bs)
+/* A non-blocking call returned EAGAIN, so yield, ensuring the
+ * handlers are set up so that we'll be rescheduled when there is an
+ * interesting event on the socket.
+ */
+static coroutine_fn void co_yield(BDRVSSHState *s, BlockDriverState *bs)
{
int r;
IOHandler *rd_handler = NULL, *wr_handler = NULL;
@@ -912,25 +916,10 @@ static coroutine_fn void set_fd_handler(BDRVSSHState *s, BlockDriverState *bs)
aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock,
false, rd_handler, wr_handler, NULL, co);
-}
-
-static coroutine_fn void clear_fd_handler(BDRVSSHState *s,
- BlockDriverState *bs)
-{
- DPRINTF("s->sock=%d", s->sock);
- aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock,
- false, NULL, NULL, NULL, NULL);
-}
-
-/* A non-blocking call returned EAGAIN, so yield, ensuring the
- * handlers are set up so that we'll be rescheduled when there is an
- * interesting event on the socket.
- */
-static coroutine_fn void co_yield(BDRVSSHState *s, BlockDriverState *bs)
-{
- set_fd_handler(s, bs);
qemu_coroutine_yield();
- clear_fd_handler(s, bs);
+ DPRINTF("s->sock=%d - back", s->sock);
+ aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock, false,
+ NULL, NULL, NULL, NULL);
}
/* SFTP has a function `libssh2_sftp_seek64' which seeks to a position
diff --git a/block/win32-aio.c b/block/win32-aio.c
index 8cdf73b..c3f8f1a 100644
--- a/block/win32-aio.c
+++ b/block/win32-aio.c
@@ -41,7 +41,7 @@ struct QEMUWin32AIOState {
HANDLE hIOCP;
EventNotifier e;
int count;
- bool is_aio_context_attached;
+ AioContext *aio_ctx;
};
typedef struct QEMUWin32AIOCB {
@@ -88,7 +88,9 @@ static void win32_aio_process_completion(QEMUWin32AIOState *s,
}
+ aio_context_acquire(s->aio_ctx);
waiocb->common.cb(waiocb->common.opaque, ret);
+ aio_context_release(s->aio_ctx);
qemu_aio_unref(waiocb);
}
@@ -176,13 +178,13 @@ void win32_aio_detach_aio_context(QEMUWin32AIOState *aio,
AioContext *old_context)
{
aio_set_event_notifier(old_context, &aio->e, false, NULL, NULL);
- aio->is_aio_context_attached = false;
+ aio->aio_ctx = NULL;
}
void win32_aio_attach_aio_context(QEMUWin32AIOState *aio,
AioContext *new_context)
{
- aio->is_aio_context_attached = true;
+ aio->aio_ctx = new_context;
aio_set_event_notifier(new_context, &aio->e, false,
win32_aio_completion_cb, NULL);
}
@@ -212,7 +214,7 @@ out_free_state:
void win32_aio_cleanup(QEMUWin32AIOState *aio)
{
- assert(!aio->is_aio_context_attached);
+ assert(!aio->aio_ctx);
CloseHandle(aio->hIOCP);
event_notifier_cleanup(&aio->e);
g_free(aio);
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 702eda8..a00ee38 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -150,7 +150,8 @@ static void virtio_blk_ioctl_complete(void *opaque, int status)
{
VirtIOBlockIoctlReq *ioctl_req = opaque;
VirtIOBlockReq *req = ioctl_req->req;
- VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
+ VirtIOBlock *s = req->dev;
+ VirtIODevice *vdev = VIRTIO_DEVICE(s);
struct virtio_scsi_inhdr *scsi;
struct sg_io_hdr *hdr;
@@ -586,6 +587,7 @@ void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
VirtIOBlockReq *req;
MultiReqBuffer mrb = {};
+ aio_context_acquire(blk_get_aio_context(s->blk));
blk_io_plug(s->blk);
do {
@@ -607,6 +609,7 @@ void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
}
blk_io_unplug(s->blk);
+ aio_context_release(blk_get_aio_context(s->blk));
}
static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index ce19eff..5d9718a 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -440,9 +440,11 @@ void virtio_scsi_handle_ctrl_vq(VirtIOSCSI *s, VirtQueue *vq)
{
VirtIOSCSIReq *req;
+ virtio_scsi_acquire(s);
while ((req = virtio_scsi_pop_req(s, vq))) {
virtio_scsi_handle_ctrl_req(s, req);
}
+ virtio_scsi_release(s);
}
static void virtio_scsi_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
@@ -598,6 +600,7 @@ void virtio_scsi_handle_cmd_vq(VirtIOSCSI *s, VirtQueue *vq)
QTAILQ_HEAD(, VirtIOSCSIReq) reqs = QTAILQ_HEAD_INITIALIZER(reqs);
+ virtio_scsi_acquire(s);
do {
virtio_queue_set_notification(vq, 0);
@@ -624,6 +627,7 @@ void virtio_scsi_handle_cmd_vq(VirtIOSCSI *s, VirtQueue *vq)
QTAILQ_FOREACH_SAFE(req, &reqs, next, next) {
virtio_scsi_handle_cmd_req_submit(s, req);
}
+ virtio_scsi_release(s);
}
static void virtio_scsi_handle_cmd(VirtIODevice *vdev, VirtQueue *vq)
@@ -754,9 +758,11 @@ out:
void virtio_scsi_handle_event_vq(VirtIOSCSI *s, VirtQueue *vq)
{
+ virtio_scsi_acquire(s);
if (s->events_dropped) {
virtio_scsi_push_event(s, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
}
+ virtio_scsi_release(s);
}
static void virtio_scsi_handle_event(VirtIODevice *vdev, VirtQueue *vq)
diff --git a/nbd/server.c b/nbd/server.c
index efe5cb8..08fb720 100644
--- a/nbd/server.c
+++ b/nbd/server.c
@@ -1366,6 +1366,10 @@ static void nbd_restart_write(void *opaque)
static void nbd_set_handlers(NBDClient *client)
{
if (client->exp && client->exp->ctx) {
+ /* Note that the handlers do not expect any concurrency; qemu-nbd
+ * does not instantiate multiple AioContexts yet, nor does it call
+ * aio_poll/aio_dispatch from multiple threads.
+ */
aio_set_fd_handler(client->exp->ctx, client->sioc->fd, true,
client->can_read ? nbd_read : NULL,
client->send_coroutine ? nbd_restart_write : NULL,
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks that need it
2017-01-13 13:17 ` [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks " Paolo Bonzini
@ 2017-01-16 13:36 ` Fam Zheng
2017-01-16 14:49 ` Paolo Bonzini
2017-01-18 15:49 ` Stefan Hajnoczi
1 sibling, 1 reply; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 13:36 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> diff --git a/nbd/server.c b/nbd/server.c
> index efe5cb8..08fb720 100644
> --- a/nbd/server.c
> +++ b/nbd/server.c
> @@ -1366,6 +1366,10 @@ static void nbd_restart_write(void *opaque)
> static void nbd_set_handlers(NBDClient *client)
> {
> if (client->exp && client->exp->ctx) {
> + /* Note that the handlers do not expect any concurrency; qemu-nbd
> + * does not instantiate multiple AioContexts yet, nor does it call
> + * aio_poll/aio_dispatch from multiple threads.
> + */
> aio_set_fd_handler(client->exp->ctx, client->sioc->fd, true,
> client->can_read ? nbd_read : NULL,
> client->send_coroutine ? nbd_restart_write : NULL,
What about the built-in server (QMP nbd_server_start)?
> --
> 2.9.3
>
>
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks that need it
2017-01-16 13:36 ` Fam Zheng
@ 2017-01-16 14:49 ` Paolo Bonzini
0 siblings, 0 replies; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-16 14:49 UTC (permalink / raw)
To: Fam Zheng; +Cc: qemu-devel, stefanha, Max Reitz
On 16/01/2017 14:36, Fam Zheng wrote:
>> + /* Note that the handlers do not expect any concurrency; qemu-nbd
>> + * does not instantiate multiple AioContexts yet, nor does it call
>> + * aio_poll/aio_dispatch from multiple threads.
>> + */
>> aio_set_fd_handler(client->exp->ctx, client->sioc->fd, true,
>> client->can_read ? nbd_read : NULL,
>> client->send_coroutine ? nbd_restart_write : NULL,
>
> What about the built-in server (QMP nbd_server_start)?
Will fix. :)
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks that need it
2017-01-13 13:17 ` [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks " Paolo Bonzini
2017-01-16 13:36 ` Fam Zheng
@ 2017-01-18 15:49 ` Stefan Hajnoczi
1 sibling, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:49 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 854 bytes --]
On Fri, Jan 13, 2017 at 02:17:26PM +0100, Paolo Bonzini wrote:
> This covers both file descriptor callbacks and polling callbacks,
> since they execute related code.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> aio-posix.c | 7 -------
> aio-win32.c | 6 ------
> block/curl.c | 16 +++++++++++++---
> block/iscsi.c | 4 ++++
> block/linux-aio.c | 4 ++++
> block/nfs.c | 6 ++++++
> block/sheepdog.c | 29 +++++++++++++++--------------
> block/ssh.c | 29 +++++++++--------------------
> block/win32-aio.c | 10 ++++++----
> hw/block/virtio-blk.c | 5 ++++-
> hw/scsi/virtio-scsi.c | 6 ++++++
> nbd/server.c | 4 ++++
> 12 files changed, 71 insertions(+), 55 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 12/16] block: explicitly acquire aiocontext in bottom halves that need it
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (10 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 11/16] block: explicitly acquire aiocontext in callbacks " Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:54 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 13/16] block: explicitly acquire aiocontext in aio callbacks " Paolo Bonzini
` (6 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
async.c | 4 ++--
block/archipelago.c | 3 +++
block/blkdebug.c | 9 +--------
block/blkreplay.c | 2 +-
block/block-backend.c | 6 ++++++
block/curl.c | 26 ++++++++++++++++++--------
block/gluster.c | 9 +--------
block/io.c | 6 +++++-
block/iscsi.c | 6 +++++-
block/linux-aio.c | 15 +++++++++------
block/nfs.c | 3 ++-
block/null.c | 4 ++++
block/qed.c | 3 +++
block/rbd.c | 4 ++++
dma-helpers.c | 2 ++
hw/block/virtio-blk.c | 2 ++
hw/scsi/scsi-bus.c | 2 ++
thread-pool.c | 2 ++
18 files changed, 72 insertions(+), 36 deletions(-)
diff --git a/async.c b/async.c
index ccb70e7..0243ca9 100644
--- a/async.c
+++ b/async.c
@@ -113,9 +113,7 @@ int aio_bh_poll(AioContext *ctx)
ret = 1;
}
bh->idle = 0;
- aio_context_acquire(ctx);
aio_bh_call(bh);
- aio_context_release(ctx);
}
if (bh->deleted) {
deleted = true;
@@ -388,7 +386,9 @@ static void co_schedule_bh_cb(void *opaque)
Coroutine *co = QSLIST_FIRST(&straight);
QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
trace_aio_co_schedule_bh_cb(ctx, co);
+ aio_context_acquire(ctx);
qemu_coroutine_enter(co);
+ aio_context_release(ctx);
}
}
diff --git a/block/archipelago.c b/block/archipelago.c
index 2449cfc..a624390 100644
--- a/block/archipelago.c
+++ b/block/archipelago.c
@@ -310,8 +310,11 @@ static void qemu_archipelago_complete_aio(void *opaque)
{
AIORequestData *reqdata = (AIORequestData *) opaque;
ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb;
+ AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs);
+ aio_context_acquire(ctx);
aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret);
+ aio_context_release(ctx);
aio_cb->status = 0;
qemu_aio_unref(aio_cb);
diff --git a/block/blkdebug.c b/block/blkdebug.c
index acccf85..259369d 100644
--- a/block/blkdebug.c
+++ b/block/blkdebug.c
@@ -405,12 +405,6 @@ out:
return ret;
}
-static void error_callback_bh(void *opaque)
-{
- Coroutine *co = opaque;
- qemu_coroutine_enter(co);
-}
-
static int inject_error(BlockDriverState *bs, BlkdebugRule *rule)
{
BDRVBlkdebugState *s = bs->opaque;
@@ -423,8 +417,7 @@ static int inject_error(BlockDriverState *bs, BlkdebugRule *rule)
}
if (!immediately) {
- aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), error_callback_bh,
- qemu_coroutine_self());
+ aio_co_schedule(bdrv_get_aio_context(bs), qemu_coroutine_self());
qemu_coroutine_yield();
}
diff --git a/block/blkreplay.c b/block/blkreplay.c
index a741654..cfc8c5b 100755
--- a/block/blkreplay.c
+++ b/block/blkreplay.c
@@ -60,7 +60,7 @@ static int64_t blkreplay_getlength(BlockDriverState *bs)
static void blkreplay_bh_cb(void *opaque)
{
Request *req = opaque;
- qemu_coroutine_enter(req->co);
+ aio_co_wake(req->co);
qemu_bh_delete(req->bh);
g_free(req);
}
diff --git a/block/block-backend.c b/block/block-backend.c
index 1177598..bfc0e6b 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -939,9 +939,12 @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
static void error_callback_bh(void *opaque)
{
struct BlockBackendAIOCB *acb = opaque;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
bdrv_dec_in_flight(acb->common.bs);
+ aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, acb->ret);
+ aio_context_release(ctx);
qemu_aio_unref(acb);
}
@@ -983,9 +986,12 @@ static void blk_aio_complete(BlkAioEmAIOCB *acb)
static void blk_aio_complete_bh(void *opaque)
{
BlkAioEmAIOCB *acb = opaque;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
assert(acb->has_returned);
+ aio_context_acquire(ctx);
blk_aio_complete(acb);
+ aio_context_release(ctx);
}
static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
diff --git a/block/curl.c b/block/curl.c
index 05b9ca3..f3f063b 100644
--- a/block/curl.c
+++ b/block/curl.c
@@ -796,13 +796,18 @@ static void curl_readv_bh_cb(void *p)
{
CURLState *state;
int running;
+ int ret = -EINPROGRESS;
CURLAIOCB *acb = p;
- BDRVCURLState *s = acb->common.bs->opaque;
+ BlockDriverState *bs = acb->common.bs;
+ BDRVCURLState *s = bs->opaque;
+ AioContext *ctx = bdrv_get_aio_context(bs);
size_t start = acb->sector_num * BDRV_SECTOR_SIZE;
size_t end;
+ aio_context_acquire(ctx);
+
// In case we have the requested data already (e.g. read-ahead),
// we can just call the callback and be done.
switch (curl_find_buf(s, start, acb->nb_sectors * BDRV_SECTOR_SIZE, acb)) {
@@ -810,7 +815,7 @@ static void curl_readv_bh_cb(void *p)
qemu_aio_unref(acb);
// fall through
case FIND_RET_WAIT:
- return;
+ goto out;
default:
break;
}
@@ -818,9 +823,8 @@ static void curl_readv_bh_cb(void *p)
// No cache found, so let's start a new request
state = curl_init_state(acb->common.bs, s);
if (!state) {
- acb->common.cb(acb->common.opaque, -EIO);
- qemu_aio_unref(acb);
- return;
+ ret = -EIO;
+ goto out;
}
acb->start = 0;
@@ -834,9 +838,8 @@ static void curl_readv_bh_cb(void *p)
state->orig_buf = g_try_malloc(state->buf_len);
if (state->buf_len && state->orig_buf == NULL) {
curl_clean_state(state);
- acb->common.cb(acb->common.opaque, -ENOMEM);
- qemu_aio_unref(acb);
- return;
+ ret = -ENOMEM;
+ goto out;
}
state->acb[0] = acb;
@@ -849,6 +852,13 @@ static void curl_readv_bh_cb(void *p)
/* Tell curl it needs to kick things off */
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
+
+out:
+ if (ret != -EINPROGRESS) {
+ acb->common.cb(acb->common.opaque, ret);
+ qemu_aio_unref(acb);
+ }
+ aio_context_release(ctx);
}
static BlockAIOCB *curl_aio_readv(BlockDriverState *bs,
diff --git a/block/gluster.c b/block/gluster.c
index 1a22f29..56b4abe 100644
--- a/block/gluster.c
+++ b/block/gluster.c
@@ -698,13 +698,6 @@ static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
return qemu_gluster_glfs_init(gconf, errp);
}
-static void qemu_gluster_complete_aio(void *opaque)
-{
- GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
-
- qemu_coroutine_enter(acb->coroutine);
-}
-
/*
* AIO callback routine called from GlusterFS thread.
*/
@@ -720,7 +713,7 @@ static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
acb->ret = -EIO; /* Partial read/write - fail it */
}
- aio_bh_schedule_oneshot(acb->aio_context, qemu_gluster_complete_aio, acb);
+ aio_co_schedule(acb->aio_context, acb->coroutine);
}
static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
diff --git a/block/io.c b/block/io.c
index c42b34a..800fcf8 100644
--- a/block/io.c
+++ b/block/io.c
@@ -189,7 +189,7 @@ static void bdrv_co_drain_bh_cb(void *opaque)
bdrv_dec_in_flight(bs);
bdrv_drained_begin(bs);
data->done = true;
- qemu_coroutine_enter(co);
+ aio_co_wake(co);
}
static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs)
@@ -2147,9 +2147,13 @@ static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
static void bdrv_co_em_bh(void *opaque)
{
BlockAIOCBCoroutine *acb = opaque;
+ BlockDriverState *bs = acb->common.bs;
+ AioContext *ctx = bdrv_get_aio_context(bs);
assert(!acb->need_bh);
+ aio_context_acquire(ctx);
bdrv_co_complete(acb);
+ aio_context_release(ctx);
}
static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
diff --git a/block/iscsi.c b/block/iscsi.c
index 54d1381..5de5d93 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -136,13 +136,16 @@ static void
iscsi_bh_cb(void *p)
{
IscsiAIOCB *acb = p;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
qemu_bh_delete(acb->bh);
g_free(acb->buf);
acb->buf = NULL;
+ aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, acb->status);
+ aio_context_release(ctx);
if (acb->task != NULL) {
scsi_free_scsi_task(acb->task);
@@ -165,8 +168,9 @@ iscsi_schedule_bh(IscsiAIOCB *acb)
static void iscsi_co_generic_bh_cb(void *opaque)
{
struct IscsiTask *iTask = opaque;
+
iTask->complete = 1;
- qemu_coroutine_enter(iTask->co);
+ aio_co_wake(iTask->co);
}
static void iscsi_retry_timer_expired(void *opaque)
diff --git a/block/linux-aio.c b/block/linux-aio.c
index 277c016..f7ae38a 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -54,10 +54,10 @@ struct LinuxAioState {
io_context_t ctx;
EventNotifier e;
- /* io queue for submit at batch */
+ /* io queue for submit at batch. Protected by AioContext lock. */
LaioQueue io_q;
- /* I/O completion processing */
+ /* I/O completion processing. Only runs in I/O thread. */
QEMUBH *completion_bh;
int event_idx;
int event_max;
@@ -75,6 +75,7 @@ static inline ssize_t io_event_ret(struct io_event *ev)
*/
static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
{
+ LinuxAioState *s = laiocb->ctx;
int ret;
ret = laiocb->ret;
@@ -93,6 +94,7 @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
}
laiocb->ret = ret;
+ aio_context_acquire(s->aio_context);
if (laiocb->co) {
/* If the coroutine is already entered it must be in ioq_submit() and
* will notice laio->ret has been filled in when it eventually runs
@@ -106,6 +108,7 @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
laiocb->common.cb(laiocb->common.opaque, ret);
qemu_aio_unref(laiocb);
}
+ aio_context_release(s->aio_context);
}
/**
@@ -234,9 +237,12 @@ static void qemu_laio_process_completions(LinuxAioState *s)
static void qemu_laio_process_completions_and_submit(LinuxAioState *s)
{
qemu_laio_process_completions(s);
+
+ aio_context_acquire(s->aio_context);
if (!s->io_q.plugged && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
ioq_submit(s);
}
+ aio_context_release(s->aio_context);
}
static void qemu_laio_completion_bh(void *opaque)
@@ -251,9 +257,7 @@ static void qemu_laio_completion_cb(EventNotifier *e)
LinuxAioState *s = container_of(e, LinuxAioState, e);
if (event_notifier_test_and_clear(&s->e)) {
- aio_context_acquire(s->aio_context);
qemu_laio_process_completions_and_submit(s);
- aio_context_release(s->aio_context);
}
}
@@ -267,9 +271,7 @@ static bool qemu_laio_poll_cb(void *opaque)
return false;
}
- aio_context_acquire(s->aio_context);
qemu_laio_process_completions_and_submit(s);
- aio_context_release(s->aio_context);
return true;
}
@@ -459,6 +461,7 @@ void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context)
{
aio_set_event_notifier(old_context, &s->e, false, NULL, NULL);
qemu_bh_delete(s->completion_bh);
+ s->aio_context = NULL;
}
void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context)
diff --git a/block/nfs.c b/block/nfs.c
index 803faf9..32631bb 100644
--- a/block/nfs.c
+++ b/block/nfs.c
@@ -236,8 +236,9 @@ static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
static void nfs_co_generic_bh_cb(void *opaque)
{
NFSRPC *task = opaque;
+
task->complete = 1;
- qemu_coroutine_enter(task->co);
+ aio_co_wake(task->co);
}
static void
diff --git a/block/null.c b/block/null.c
index 356209a..5eb2038 100644
--- a/block/null.c
+++ b/block/null.c
@@ -134,7 +134,11 @@ static const AIOCBInfo null_aiocb_info = {
static void null_bh_cb(void *opaque)
{
NullAIOCB *acb = opaque;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
+
+ aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, 0);
+ aio_context_release(ctx);
qemu_aio_unref(acb);
}
diff --git a/block/qed.c b/block/qed.c
index a21d025..db8295d 100644
--- a/block/qed.c
+++ b/block/qed.c
@@ -942,6 +942,7 @@ static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index,
static void qed_aio_complete_bh(void *opaque)
{
QEDAIOCB *acb = opaque;
+ BDRVQEDState *s = acb_to_s(acb);
BlockCompletionFunc *cb = acb->common.cb;
void *user_opaque = acb->common.opaque;
int ret = acb->bh_ret;
@@ -949,7 +950,9 @@ static void qed_aio_complete_bh(void *opaque)
qemu_aio_unref(acb);
/* Invoke callback */
+ qed_acquire(s);
cb(user_opaque, ret);
+ qed_release(s);
}
static void qed_aio_complete(QEDAIOCB *acb, int ret)
diff --git a/block/rbd.c b/block/rbd.c
index a57b3e3..2cb2cb4 100644
--- a/block/rbd.c
+++ b/block/rbd.c
@@ -413,6 +413,7 @@ shutdown:
static void qemu_rbd_complete_aio(RADOSCB *rcb)
{
RBDAIOCB *acb = rcb->acb;
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
int64_t r;
r = rcb->ret;
@@ -445,7 +446,10 @@ static void qemu_rbd_complete_aio(RADOSCB *rcb)
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
}
qemu_vfree(acb->bounce);
+
+ aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
+ aio_context_release(ctx);
qemu_aio_unref(acb);
}
diff --git a/dma-helpers.c b/dma-helpers.c
index 6f9d47c..39d4802 100644
--- a/dma-helpers.c
+++ b/dma-helpers.c
@@ -166,8 +166,10 @@ static void dma_blk_cb(void *opaque, int ret)
QEMU_ALIGN_DOWN(dbs->iov.size, dbs->align));
}
+ aio_context_acquire(dbs->ctx);
dbs->acb = dbs->io_func(dbs->offset, &dbs->iov,
dma_blk_cb, dbs, dbs->io_func_opaque);
+ aio_context_release(dbs->ctx);
assert(dbs->acb);
}
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index a00ee38..af652f3 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -639,6 +639,7 @@ static void virtio_blk_dma_restart_bh(void *opaque)
s->rq = NULL;
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
while (req) {
VirtIOBlockReq *next = req->next;
if (virtio_blk_handle_request(req, &mrb)) {
@@ -659,6 +660,7 @@ static void virtio_blk_dma_restart_bh(void *opaque)
if (mrb.num_reqs) {
virtio_blk_submit_multireq(s->blk, &mrb);
}
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
}
static void virtio_blk_dma_restart_cb(void *opaque, int running,
diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
index 297216d..dddeee3 100644
--- a/hw/scsi/scsi-bus.c
+++ b/hw/scsi/scsi-bus.c
@@ -105,6 +105,7 @@ static void scsi_dma_restart_bh(void *opaque)
qemu_bh_delete(s->bh);
s->bh = NULL;
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
scsi_req_ref(req);
if (req->retry) {
@@ -122,6 +123,7 @@ static void scsi_dma_restart_bh(void *opaque)
}
scsi_req_unref(req);
}
+ aio_context_release(blk_get_aio_context(s->conf.blk));
}
void scsi_req_retry(SCSIRequest *req)
diff --git a/thread-pool.c b/thread-pool.c
index 6fba913..7c9cec5 100644
--- a/thread-pool.c
+++ b/thread-pool.c
@@ -165,6 +165,7 @@ static void thread_pool_completion_bh(void *opaque)
ThreadPool *pool = opaque;
ThreadPoolElement *elem, *next;
+ aio_context_acquire(pool->ctx);
restart:
QLIST_FOREACH_SAFE(elem, &pool->head, all, next) {
if (elem->state != THREAD_DONE) {
@@ -191,6 +192,7 @@ restart:
qemu_aio_unref(elem);
}
}
+ aio_context_release(pool->ctx);
}
static void thread_pool_cancel(BlockAIOCB *acb)
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* [Qemu-devel] [PATCH 13/16] block: explicitly acquire aiocontext in aio callbacks that need it
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (11 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 12/16] block: explicitly acquire aiocontext in bottom halves " Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:58 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 14/16] aio-posix: partially inline aio_dispatch into aio_poll Paolo Bonzini
` (5 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
block/archipelago.c | 3 ---
block/block-backend.c | 7 -------
block/curl.c | 2 +-
block/io.c | 6 +-----
block/iscsi.c | 3 ---
block/linux-aio.c | 5 +----
block/mirror.c | 12 +++++++++---
block/null.c | 8 --------
block/qed-cluster.c | 2 ++
block/qed-table.c | 12 ++++++++++--
block/qed.c | 4 ++--
block/rbd.c | 4 ----
block/win32-aio.c | 3 ---
hw/block/virtio-blk.c | 12 +++++++++++-
hw/scsi/scsi-disk.c | 15 +++++++++++++++
hw/scsi/scsi-generic.c | 20 +++++++++++++++++---
thread-pool.c | 4 +++-
17 files changed, 72 insertions(+), 50 deletions(-)
diff --git a/block/archipelago.c b/block/archipelago.c
index a624390..2449cfc 100644
--- a/block/archipelago.c
+++ b/block/archipelago.c
@@ -310,11 +310,8 @@ static void qemu_archipelago_complete_aio(void *opaque)
{
AIORequestData *reqdata = (AIORequestData *) opaque;
ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb;
- AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs);
- aio_context_acquire(ctx);
aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret);
- aio_context_release(ctx);
aio_cb->status = 0;
qemu_aio_unref(aio_cb);
diff --git a/block/block-backend.c b/block/block-backend.c
index bfc0e6b..819f272 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -939,12 +939,9 @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
static void error_callback_bh(void *opaque)
{
struct BlockBackendAIOCB *acb = opaque;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
bdrv_dec_in_flight(acb->common.bs);
- aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, acb->ret);
- aio_context_release(ctx);
qemu_aio_unref(acb);
}
@@ -986,12 +983,8 @@ static void blk_aio_complete(BlkAioEmAIOCB *acb)
static void blk_aio_complete_bh(void *opaque)
{
BlkAioEmAIOCB *acb = opaque;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
-
assert(acb->has_returned);
- aio_context_acquire(ctx);
blk_aio_complete(acb);
- aio_context_release(ctx);
}
static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
diff --git a/block/curl.c b/block/curl.c
index f3f063b..2939cc7 100644
--- a/block/curl.c
+++ b/block/curl.c
@@ -854,11 +854,11 @@ static void curl_readv_bh_cb(void *p)
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
out:
+ aio_context_release(ctx);
if (ret != -EINPROGRESS) {
acb->common.cb(acb->common.opaque, ret);
qemu_aio_unref(acb);
}
- aio_context_release(ctx);
}
static BlockAIOCB *curl_aio_readv(BlockDriverState *bs,
diff --git a/block/io.c b/block/io.c
index 800fcf8..3a879cc 100644
--- a/block/io.c
+++ b/block/io.c
@@ -813,7 +813,7 @@ static void bdrv_co_io_em_complete(void *opaque, int ret)
CoroutineIOCompletion *co = opaque;
co->ret = ret;
- qemu_coroutine_enter(co->coroutine);
+ aio_co_wake(co->coroutine);
}
static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
@@ -2147,13 +2147,9 @@ static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
static void bdrv_co_em_bh(void *opaque)
{
BlockAIOCBCoroutine *acb = opaque;
- BlockDriverState *bs = acb->common.bs;
- AioContext *ctx = bdrv_get_aio_context(bs);
assert(!acb->need_bh);
- aio_context_acquire(ctx);
bdrv_co_complete(acb);
- aio_context_release(ctx);
}
static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
diff --git a/block/iscsi.c b/block/iscsi.c
index 5de5d93..c19c838 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -136,16 +136,13 @@ static void
iscsi_bh_cb(void *p)
{
IscsiAIOCB *acb = p;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
qemu_bh_delete(acb->bh);
g_free(acb->buf);
acb->buf = NULL;
- aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, acb->status);
- aio_context_release(ctx);
if (acb->task != NULL) {
scsi_free_scsi_task(acb->task);
diff --git a/block/linux-aio.c b/block/linux-aio.c
index f7ae38a..88b8d55 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -75,7 +75,6 @@ static inline ssize_t io_event_ret(struct io_event *ev)
*/
static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
{
- LinuxAioState *s = laiocb->ctx;
int ret;
ret = laiocb->ret;
@@ -94,7 +93,6 @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
}
laiocb->ret = ret;
- aio_context_acquire(s->aio_context);
if (laiocb->co) {
/* If the coroutine is already entered it must be in ioq_submit() and
* will notice laio->ret has been filled in when it eventually runs
@@ -102,13 +100,12 @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
* that!
*/
if (!qemu_coroutine_entered(laiocb->co)) {
- qemu_coroutine_enter(laiocb->co);
+ aio_co_wake(laiocb->co);
}
} else {
laiocb->common.cb(laiocb->common.opaque, ret);
qemu_aio_unref(laiocb);
}
- aio_context_release(s->aio_context);
}
/**
diff --git a/block/mirror.c b/block/mirror.c
index 301ba92..698a54e 100644
--- a/block/mirror.c
+++ b/block/mirror.c
@@ -132,6 +132,8 @@ static void mirror_write_complete(void *opaque, int ret)
{
MirrorOp *op = opaque;
MirrorBlockJob *s = op->s;
+
+ aio_context_acquire(blk_get_aio_context(s->common.blk));
if (ret < 0) {
BlockErrorAction action;
@@ -142,12 +144,15 @@ static void mirror_write_complete(void *opaque, int ret)
}
}
mirror_iteration_done(op, ret);
+ aio_context_release(blk_get_aio_context(s->common.blk));
}
static void mirror_read_complete(void *opaque, int ret)
{
MirrorOp *op = opaque;
MirrorBlockJob *s = op->s;
+
+ aio_context_acquire(blk_get_aio_context(s->common.blk));
if (ret < 0) {
BlockErrorAction action;
@@ -158,10 +163,11 @@ static void mirror_read_complete(void *opaque, int ret)
}
mirror_iteration_done(op, ret);
- return;
+ } else {
+ blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov,
+ 0, mirror_write_complete, op);
}
- blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov,
- 0, mirror_write_complete, op);
+ aio_context_release(blk_get_aio_context(s->common.blk));
}
static inline void mirror_clip_sectors(MirrorBlockJob *s,
diff --git a/block/null.c b/block/null.c
index 5eb2038..b300390 100644
--- a/block/null.c
+++ b/block/null.c
@@ -134,22 +134,14 @@ static const AIOCBInfo null_aiocb_info = {
static void null_bh_cb(void *opaque)
{
NullAIOCB *acb = opaque;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
-
- aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, 0);
- aio_context_release(ctx);
qemu_aio_unref(acb);
}
static void null_timer_cb(void *opaque)
{
NullAIOCB *acb = opaque;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
-
- aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, 0);
- aio_context_release(ctx);
timer_deinit(&acb->timer);
qemu_aio_unref(acb);
}
diff --git a/block/qed-cluster.c b/block/qed-cluster.c
index c24e756..8f5da74 100644
--- a/block/qed-cluster.c
+++ b/block/qed-cluster.c
@@ -83,6 +83,7 @@ static void qed_find_cluster_cb(void *opaque, int ret)
unsigned int index;
unsigned int n;
+ qed_acquire(s);
if (ret) {
goto out;
}
@@ -109,6 +110,7 @@ static void qed_find_cluster_cb(void *opaque, int ret)
out:
find_cluster_cb->cb(find_cluster_cb->opaque, ret, offset, len);
+ qed_release(s);
g_free(find_cluster_cb);
}
diff --git a/block/qed-table.c b/block/qed-table.c
index ed443e2..b12c298 100644
--- a/block/qed-table.c
+++ b/block/qed-table.c
@@ -31,6 +31,7 @@ static void qed_read_table_cb(void *opaque, int ret)
{
QEDReadTableCB *read_table_cb = opaque;
QEDTable *table = read_table_cb->table;
+ BDRVQEDState *s = read_table_cb->s;
int noffsets = read_table_cb->qiov.size / sizeof(uint64_t);
int i;
@@ -40,13 +41,15 @@ static void qed_read_table_cb(void *opaque, int ret)
}
/* Byteswap offsets */
+ qed_acquire(s);
for (i = 0; i < noffsets; i++) {
table->offsets[i] = le64_to_cpu(table->offsets[i]);
}
+ qed_release(s);
out:
/* Completion */
- trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret);
+ trace_qed_read_table_cb(s, read_table_cb->table, ret);
gencb_complete(&read_table_cb->gencb, ret);
}
@@ -84,8 +87,9 @@ typedef struct {
static void qed_write_table_cb(void *opaque, int ret)
{
QEDWriteTableCB *write_table_cb = opaque;
+ BDRVQEDState *s = write_table_cb->s;
- trace_qed_write_table_cb(write_table_cb->s,
+ trace_qed_write_table_cb(s,
write_table_cb->orig_table,
write_table_cb->flush,
ret);
@@ -97,8 +101,10 @@ static void qed_write_table_cb(void *opaque, int ret)
if (write_table_cb->flush) {
/* We still need to flush first */
write_table_cb->flush = false;
+ qed_acquire(s);
bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb,
write_table_cb);
+ qed_release(s);
return;
}
@@ -213,6 +219,7 @@ static void qed_read_l2_table_cb(void *opaque, int ret)
CachedL2Table *l2_table = request->l2_table;
uint64_t l2_offset = read_l2_table_cb->l2_offset;
+ qed_acquire(s);
if (ret) {
/* can't trust loaded L2 table anymore */
qed_unref_l2_cache_entry(l2_table);
@@ -228,6 +235,7 @@ static void qed_read_l2_table_cb(void *opaque, int ret)
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
assert(request->l2_table != NULL);
}
+ qed_release(s);
gencb_complete(&read_l2_table_cb->gencb, ret);
}
diff --git a/block/qed.c b/block/qed.c
index db8295d..0b62c77 100644
--- a/block/qed.c
+++ b/block/qed.c
@@ -745,7 +745,7 @@ static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t l
}
if (cb->co) {
- qemu_coroutine_enter(cb->co);
+ aio_co_wake(cb->co);
}
}
@@ -1462,7 +1462,7 @@ static void coroutine_fn qed_co_pwrite_zeroes_cb(void *opaque, int ret)
cb->done = true;
cb->ret = ret;
if (cb->co) {
- qemu_coroutine_enter(cb->co);
+ aio_co_wake(cb->co);
}
}
diff --git a/block/rbd.c b/block/rbd.c
index 2cb2cb4..a57b3e3 100644
--- a/block/rbd.c
+++ b/block/rbd.c
@@ -413,7 +413,6 @@ shutdown:
static void qemu_rbd_complete_aio(RADOSCB *rcb)
{
RBDAIOCB *acb = rcb->acb;
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
int64_t r;
r = rcb->ret;
@@ -446,10 +445,7 @@ static void qemu_rbd_complete_aio(RADOSCB *rcb)
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
}
qemu_vfree(acb->bounce);
-
- aio_context_acquire(ctx);
acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
- aio_context_release(ctx);
qemu_aio_unref(acb);
}
diff --git a/block/win32-aio.c b/block/win32-aio.c
index c3f8f1a..3be8f45 100644
--- a/block/win32-aio.c
+++ b/block/win32-aio.c
@@ -87,10 +87,7 @@ static void win32_aio_process_completion(QEMUWin32AIOState *s,
qemu_vfree(waiocb->buf);
}
-
- aio_context_acquire(s->aio_ctx);
waiocb->common.cb(waiocb->common.opaque, ret);
- aio_context_release(s->aio_ctx);
qemu_aio_unref(waiocb);
}
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index af652f3..39516e8 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -89,7 +89,9 @@ static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
static void virtio_blk_rw_complete(void *opaque, int ret)
{
VirtIOBlockReq *next = opaque;
+ VirtIOBlock *s = next->dev;
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
while (next) {
VirtIOBlockReq *req = next;
next = req->mr_next;
@@ -122,21 +124,27 @@ static void virtio_blk_rw_complete(void *opaque, int ret)
block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
virtio_blk_free_request(req);
}
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
}
static void virtio_blk_flush_complete(void *opaque, int ret)
{
VirtIOBlockReq *req = opaque;
+ VirtIOBlock *s = req->dev;
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
if (ret) {
if (virtio_blk_handle_rw_error(req, -ret, 0)) {
- return;
+ goto out;
}
}
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
virtio_blk_free_request(req);
+
+out:
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
}
#ifdef __linux__
@@ -183,8 +191,10 @@ static void virtio_blk_ioctl_complete(void *opaque, int status)
virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
out:
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
virtio_blk_req_complete(req, status);
virtio_blk_free_request(req);
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
g_free(ioctl_req);
}
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index bdd1e5f..baf66b5 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -207,6 +207,7 @@ static void scsi_aio_complete(void *opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (scsi_disk_req_check_error(r, ret, true)) {
goto done;
}
@@ -215,6 +216,7 @@ static void scsi_aio_complete(void *opaque, int ret)
scsi_req_complete(&r->req, GOOD);
done:
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
scsi_req_unref(&r->req);
}
@@ -290,12 +292,14 @@ static void scsi_dma_complete(void *opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (ret < 0) {
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
} else {
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
}
scsi_dma_complete_noio(r, ret);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
static void scsi_read_complete(void * opaque, int ret)
@@ -306,6 +310,7 @@ static void scsi_read_complete(void * opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (scsi_disk_req_check_error(r, ret, true)) {
goto done;
}
@@ -320,6 +325,7 @@ static void scsi_read_complete(void * opaque, int ret)
done:
scsi_req_unref(&r->req);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
/* Actually issue a read to the block device. */
@@ -364,12 +370,14 @@ static void scsi_do_read_cb(void *opaque, int ret)
assert (r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (ret < 0) {
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
} else {
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
}
scsi_do_read(opaque, ret);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
/* Read more data from scsi device into buffer. */
@@ -489,12 +497,14 @@ static void scsi_write_complete(void * opaque, int ret)
assert (r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (ret < 0) {
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
} else {
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
}
scsi_write_complete_noio(r, ret);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
static void scsi_write_data(SCSIRequest *req)
@@ -1625,11 +1635,14 @@ static void scsi_unmap_complete(void *opaque, int ret)
{
UnmapCBData *data = opaque;
SCSIDiskReq *r = data->r;
+ SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
scsi_unmap_complete_noio(data, ret);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
@@ -1696,6 +1709,7 @@ static void scsi_write_same_complete(void *opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
if (scsi_disk_req_check_error(r, ret, true)) {
goto done;
}
@@ -1724,6 +1738,7 @@ done:
scsi_req_unref(&r->req);
qemu_vfree(data->iov.iov_base);
g_free(data);
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
}
static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
diff --git a/hw/scsi/scsi-generic.c b/hw/scsi/scsi-generic.c
index 7a588a7..fdf9168 100644
--- a/hw/scsi/scsi-generic.c
+++ b/hw/scsi/scsi-generic.c
@@ -143,10 +143,14 @@ done:
static void scsi_command_complete(void *opaque, int ret)
{
SCSIGenericReq *r = (SCSIGenericReq *)opaque;
+ SCSIDevice *s = r->req.dev;
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
scsi_command_complete_noio(r, ret);
+ aio_context_release(blk_get_aio_context(s->conf.blk));
}
static int execute_command(BlockBackend *blk,
@@ -182,9 +186,11 @@ static void scsi_read_complete(void * opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
+
if (ret || r->req.io_canceled) {
scsi_command_complete_noio(r, ret);
- return;
+ goto done;
}
len = r->io_header.dxfer_len - r->io_header.resid;
@@ -193,7 +199,7 @@ static void scsi_read_complete(void * opaque, int ret)
r->len = -1;
if (len == 0) {
scsi_command_complete_noio(r, 0);
- return;
+ goto done;
}
/* Snoop READ CAPACITY output to set the blocksize. */
@@ -237,6 +243,9 @@ static void scsi_read_complete(void * opaque, int ret)
}
scsi_req_data(&r->req, len);
scsi_req_unref(&r->req);
+
+done:
+ aio_context_release(blk_get_aio_context(s->conf.blk));
}
/* Read more data from scsi device into buffer. */
@@ -272,9 +281,11 @@ static void scsi_write_complete(void * opaque, int ret)
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
+
if (ret || r->req.io_canceled) {
scsi_command_complete_noio(r, ret);
- return;
+ goto done;
}
if (r->req.cmd.buf[0] == MODE_SELECT && r->req.cmd.buf[4] == 12 &&
@@ -284,6 +295,9 @@ static void scsi_write_complete(void * opaque, int ret)
}
scsi_command_complete_noio(r, ret);
+
+done:
+ aio_context_release(blk_get_aio_context(s->conf.blk));
}
/* Write data to a scsi device. Returns nonzero on failure.
diff --git a/thread-pool.c b/thread-pool.c
index 7c9cec5..ce6cd30 100644
--- a/thread-pool.c
+++ b/thread-pool.c
@@ -185,7 +185,9 @@ restart:
*/
qemu_bh_schedule(pool->completion_bh);
+ aio_context_release(pool->ctx);
elem->common.cb(elem->common.opaque, elem->ret);
+ aio_context_acquire(pool->ctx);
qemu_aio_unref(elem);
goto restart;
} else {
@@ -269,7 +271,7 @@ static void thread_pool_co_cb(void *opaque, int ret)
ThreadPoolCo *co = opaque;
co->ret = ret;
- qemu_coroutine_enter(co->co);
+ aio_co_wake(co->co);
}
int coroutine_fn thread_pool_submit_co(ThreadPool *pool, ThreadPoolFunc *func,
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* [Qemu-devel] [PATCH 14/16] aio-posix: partially inline aio_dispatch into aio_poll
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (12 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 13/16] block: explicitly acquire aiocontext in aio callbacks " Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 15:59 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 15/16] async: remove unnecessary inc/dec pairs Paolo Bonzini
` (4 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
This patch prepares for the removal of unnecessary lockcnt inc/dec pairs.
Extract the dispatching loop for file descriptor handlers into a new
function aio_dispatch_handlers, and then inline aio_dispatch into
aio_poll.
aio_dispatch can now become void.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
aio-posix.c | 44 ++++++++++++++------------------------------
aio-win32.c | 13 ++++---------
async.c | 2 +-
include/block/aio.h | 6 +-----
4 files changed, 20 insertions(+), 45 deletions(-)
diff --git a/aio-posix.c b/aio-posix.c
index 6beebcd..51e92b8 100644
--- a/aio-posix.c
+++ b/aio-posix.c
@@ -386,12 +386,6 @@ static bool aio_dispatch_handlers(AioContext *ctx)
AioHandler *node, *tmp;
bool progress = false;
- /*
- * We have to walk very carefully in case aio_set_fd_handler is
- * called while we're walking.
- */
- qemu_lockcnt_inc(&ctx->list_lock);
-
QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
int revents;
@@ -426,33 +420,18 @@ static bool aio_dispatch_handlers(AioContext *ctx)
}
}
- qemu_lockcnt_dec(&ctx->list_lock);
return progress;
}
-/*
- * Note that dispatch_fds == false has the side-effect of post-poning the
- * freeing of deleted handlers.
- */
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
+void aio_dispatch(AioContext *ctx)
{
- bool progress;
-
- /*
- * If there are callbacks left that have been queued, we need to call them.
- * Do not call select in this case, because it is possible that the caller
- * does not need a complete flush (as is the case for aio_poll loops).
- */
- progress = aio_bh_poll(ctx);
+ aio_bh_poll(ctx);
- if (dispatch_fds) {
- progress |= aio_dispatch_handlers(ctx);
- }
-
- /* Run our timers */
- progress |= timerlistgroup_run_timers(&ctx->tlg);
+ qemu_lockcnt_inc(&ctx->list_lock);
+ aio_dispatch_handlers(ctx);
+ qemu_lockcnt_dec(&ctx->list_lock);
- return progress;
+ timerlistgroup_run_timers(&ctx->tlg);
}
/* These thread-local variables are used only in a small part of aio_poll
@@ -701,11 +680,16 @@ bool aio_poll(AioContext *ctx, bool blocking)
npfd = 0;
qemu_lockcnt_dec(&ctx->list_lock);
- /* Run dispatch even if there were no readable fds to run timers */
- if (aio_dispatch(ctx, ret > 0)) {
- progress = true;
+ progress |= aio_bh_poll(ctx);
+
+ if (ret > 0) {
+ qemu_lockcnt_inc(&ctx->list_lock);
+ progress |= aio_dispatch_handlers(ctx);
+ qemu_lockcnt_dec(&ctx->list_lock);
}
+ progress |= timerlistgroup_run_timers(&ctx->tlg);
+
return progress;
}
diff --git a/aio-win32.c b/aio-win32.c
index 20b63ce..442a179 100644
--- a/aio-win32.c
+++ b/aio-win32.c
@@ -309,16 +309,11 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
return progress;
}
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
+void aio_dispatch(AioContext *ctx)
{
- bool progress;
-
- progress = aio_bh_poll(ctx);
- if (dispatch_fds) {
- progress |= aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
- }
- progress |= timerlistgroup_run_timers(&ctx->tlg);
- return progress;
+ aio_bh_poll(ctx);
+ aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
+ timerlistgroup_run_timers(&ctx->tlg);
}
bool aio_poll(AioContext *ctx, bool blocking)
diff --git a/async.c b/async.c
index 0243ca9..1839aa5 100644
--- a/async.c
+++ b/async.c
@@ -257,7 +257,7 @@ aio_ctx_dispatch(GSource *source,
AioContext *ctx = (AioContext *) source;
assert(callback == NULL);
- aio_dispatch(ctx, true);
+ aio_dispatch(ctx);
return true;
}
diff --git a/include/block/aio.h b/include/block/aio.h
index 614cbc6..677b6ff 100644
--- a/include/block/aio.h
+++ b/include/block/aio.h
@@ -310,12 +310,8 @@ bool aio_pending(AioContext *ctx);
/* Dispatch any pending callbacks from the GSource attached to the AioContext.
*
* This is used internally in the implementation of the GSource.
- *
- * @dispatch_fds: true to process fds, false to skip them
- * (can be used as an optimization by callers that know there
- * are no fds ready)
*/
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds);
+void aio_dispatch(AioContext *ctx);
/* Progress in completing AIO work to occur. This can issue new pending
* aio as a result of executing I/O completion or bh callbacks.
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 14/16] aio-posix: partially inline aio_dispatch into aio_poll
2017-01-13 13:17 ` [Qemu-devel] [PATCH 14/16] aio-posix: partially inline aio_dispatch into aio_poll Paolo Bonzini
@ 2017-01-18 15:59 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 15:59 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 695 bytes --]
On Fri, Jan 13, 2017 at 02:17:29PM +0100, Paolo Bonzini wrote:
> This patch prepares for the removal of unnecessary lockcnt inc/dec pairs.
> Extract the dispatching loop for file descriptor handlers into a new
> function aio_dispatch_handlers, and then inline aio_dispatch into
> aio_poll.
>
> aio_dispatch can now become void.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> aio-posix.c | 44 ++++++++++++++------------------------------
> aio-win32.c | 13 ++++---------
> async.c | 2 +-
> include/block/aio.h | 6 +-----
> 4 files changed, 20 insertions(+), 45 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* [Qemu-devel] [PATCH 15/16] async: remove unnecessary inc/dec pairs
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (13 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 14/16] aio-posix: partially inline aio_dispatch into aio_poll Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 16:00 ` Stefan Hajnoczi
2017-01-13 13:17 ` [Qemu-devel] [PATCH 16/16] block: document fields protected by AioContext lock Paolo Bonzini
` (3 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Pull the increment/decrement pair out of aio_bh_poll and into the
callers.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
aio-posix.c | 8 +++-----
aio-win32.c | 8 ++++----
async.c | 12 ++++++------
3 files changed, 13 insertions(+), 15 deletions(-)
diff --git a/aio-posix.c b/aio-posix.c
index 51e92b8..2537bcd 100644
--- a/aio-posix.c
+++ b/aio-posix.c
@@ -425,9 +425,8 @@ static bool aio_dispatch_handlers(AioContext *ctx)
void aio_dispatch(AioContext *ctx)
{
- aio_bh_poll(ctx);
-
qemu_lockcnt_inc(&ctx->list_lock);
+ aio_bh_poll(ctx);
aio_dispatch_handlers(ctx);
qemu_lockcnt_dec(&ctx->list_lock);
@@ -678,16 +677,15 @@ bool aio_poll(AioContext *ctx, bool blocking)
}
npfd = 0;
- qemu_lockcnt_dec(&ctx->list_lock);
progress |= aio_bh_poll(ctx);
if (ret > 0) {
- qemu_lockcnt_inc(&ctx->list_lock);
progress |= aio_dispatch_handlers(ctx);
- qemu_lockcnt_dec(&ctx->list_lock);
}
+ qemu_lockcnt_dec(&ctx->list_lock);
+
progress |= timerlistgroup_run_timers(&ctx->tlg);
return progress;
diff --git a/aio-win32.c b/aio-win32.c
index 442a179..bca496a 100644
--- a/aio-win32.c
+++ b/aio-win32.c
@@ -253,8 +253,6 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
bool progress = false;
AioHandler *tmp;
- qemu_lockcnt_inc(&ctx->list_lock);
-
/*
* We have to walk very carefully in case aio_set_fd_handler is
* called while we're walking.
@@ -305,14 +303,15 @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
}
}
- qemu_lockcnt_dec(&ctx->list_lock);
return progress;
}
void aio_dispatch(AioContext *ctx)
{
+ qemu_lockcnt_inc(&ctx->list_lock);
aio_bh_poll(ctx);
aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
+ qemu_lockcnt_dec(&ctx->list_lock);
timerlistgroup_run_timers(&ctx->tlg);
}
@@ -349,7 +348,6 @@ bool aio_poll(AioContext *ctx, bool blocking)
}
}
- qemu_lockcnt_dec(&ctx->list_lock);
first = true;
/* ctx->notifier is always registered. */
@@ -392,6 +390,8 @@ bool aio_poll(AioContext *ctx, bool blocking)
progress |= aio_dispatch_handlers(ctx, event);
} while (count > 0);
+ qemu_lockcnt_dec(&ctx->list_lock);
+
progress |= timerlistgroup_run_timers(&ctx->tlg);
return progress;
}
diff --git a/async.c b/async.c
index 1839aa5..c471b1e 100644
--- a/async.c
+++ b/async.c
@@ -89,15 +89,16 @@ void aio_bh_call(QEMUBH *bh)
bh->cb(bh->opaque);
}
-/* Multiple occurrences of aio_bh_poll cannot be called concurrently */
+/* Multiple occurrences of aio_bh_poll cannot be called concurrently.
+ * The count in ctx->list_lock is incremented before the call, and is
+ * not affected by the call.
+ */
int aio_bh_poll(AioContext *ctx)
{
QEMUBH *bh, **bhp, *next;
int ret;
bool deleted = false;
- qemu_lockcnt_inc(&ctx->list_lock);
-
ret = 0;
for (bh = atomic_rcu_read(&ctx->first_bh); bh; bh = next) {
next = atomic_rcu_read(&bh->next);
@@ -122,11 +123,10 @@ int aio_bh_poll(AioContext *ctx)
/* remove deleted bhs */
if (!deleted) {
- qemu_lockcnt_dec(&ctx->list_lock);
return ret;
}
- if (qemu_lockcnt_dec_and_lock(&ctx->list_lock)) {
+ if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
bhp = &ctx->first_bh;
while (*bhp) {
bh = *bhp;
@@ -137,7 +137,7 @@ int aio_bh_poll(AioContext *ctx)
bhp = &bh->next;
}
}
- qemu_lockcnt_unlock(&ctx->list_lock);
+ qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
}
return ret;
}
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* [Qemu-devel] [PATCH 16/16] block: document fields protected by AioContext lock
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (14 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 15/16] async: remove unnecessary inc/dec pairs Paolo Bonzini
@ 2017-01-13 13:17 ` Paolo Bonzini
2017-01-18 16:01 ` Stefan Hajnoczi
2017-01-16 16:26 ` [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Fam Zheng
` (2 subsequent siblings)
18 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-13 13:17 UTC (permalink / raw)
To: qemu-devel; +Cc: famz, stefanha
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
include/block/block_int.h | 64 +++++++++++++++++++++++++-----------------
include/sysemu/block-backend.h | 14 ++++++---
2 files changed, 49 insertions(+), 29 deletions(-)
diff --git a/include/block/block_int.h b/include/block/block_int.h
index 2d92d7e..1670941 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -430,8 +430,9 @@ struct BdrvChild {
* copied as well.
*/
struct BlockDriverState {
- int64_t total_sectors; /* if we are reading a disk image, give its
- size in sectors */
+ /* Protected by big QEMU lock or read-only after opening. No special
+ * locking needed during I/O...
+ */
int open_flags; /* flags used to open the file, re-used for re-open */
bool read_only; /* if true, the media is read only */
bool encrypted; /* if true, the media is encrypted */
@@ -439,14 +440,6 @@ struct BlockDriverState {
bool sg; /* if true, the device is a /dev/sg* */
bool probed; /* if true, format was probed rather than specified */
- int copy_on_read; /* if nonzero, copy read backing sectors into image.
- note this is a reference count */
-
- CoQueue flush_queue; /* Serializing flush queue */
- bool active_flush_req; /* Flush request in flight? */
- unsigned int write_gen; /* Current data generation */
- unsigned int flushed_gen; /* Flushed write generation */
-
BlockDriver *drv; /* NULL means no media */
void *opaque;
@@ -468,18 +461,6 @@ struct BlockDriverState {
BdrvChild *backing;
BdrvChild *file;
- /* Callback before write request is processed */
- NotifierWithReturnList before_write_notifiers;
-
- /* number of in-flight requests; overall and serialising */
- unsigned int in_flight;
- unsigned int serialising_in_flight;
-
- bool wakeup;
-
- /* Offset after the highest byte written to */
- uint64_t wr_highest_offset;
-
/* I/O Limits */
BlockLimits bl;
@@ -497,11 +478,8 @@ struct BlockDriverState {
QTAILQ_ENTRY(BlockDriverState) bs_list;
/* element of the list of monitor-owned BDS */
QTAILQ_ENTRY(BlockDriverState) monitor_list;
- QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
int refcnt;
- QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
-
/* operation blockers */
QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
@@ -522,6 +500,31 @@ struct BlockDriverState {
/* The error object in use for blocking operations on backing_hd */
Error *backing_blocker;
+ /* Protected by AioContext lock */
+
+ /* If true, copy read backing sectors into image. Can be >1 if more
+ * than one client has requested copy-on-read.
+ */
+ int copy_on_read;
+
+ /* If we are reading a disk image, give its size in sectors.
+ * Generally read-only; it is written to by load_vmstate and save_vmstate,
+ * but the block layer is quiescent during those.
+ */
+ int64_t total_sectors;
+
+ /* Callback before write request is processed */
+ NotifierWithReturnList before_write_notifiers;
+
+ /* number of in-flight requests; overall and serialising */
+ unsigned int in_flight;
+ unsigned int serialising_in_flight;
+
+ bool wakeup;
+
+ /* Offset after the highest byte written to */
+ uint64_t wr_highest_offset;
+
/* threshold limit for writes, in bytes. "High water mark". */
uint64_t write_threshold_offset;
NotifierWithReturn write_threshold_notifier;
@@ -529,6 +532,17 @@ struct BlockDriverState {
/* counter for nested bdrv_io_plug */
unsigned io_plugged;
+ QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
+ CoQueue flush_queue; /* Serializing flush queue */
+ bool active_flush_req; /* Flush request in flight? */
+ unsigned int write_gen; /* Current data generation */
+ unsigned int flushed_gen; /* Flushed write generation */
+
+ QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
+
+ /* do we need to tell the quest if we have a volatile write cache? */
+ int enable_write_cache;
+
int quiesce_counter;
};
diff --git a/include/sysemu/block-backend.h b/include/sysemu/block-backend.h
index 6444e41..f365a51 100644
--- a/include/sysemu/block-backend.h
+++ b/include/sysemu/block-backend.h
@@ -64,14 +64,20 @@ typedef struct BlockDevOps {
* fields that must be public. This is in particular for QLIST_ENTRY() and
* friends so that BlockBackends can be kept in lists outside block-backend.c */
typedef struct BlockBackendPublic {
- /* I/O throttling.
- * throttle_state tells us if this BlockBackend has I/O limits configured.
- * io_limits_disabled tells us if they are currently being enforced */
+ /* I/O throttling has its own locking, but also some fields are
+ * protected by the AioContext lock.
+ */
+
+ /* Protected by AioContext lock. */
CoQueue throttled_reqs[2];
+
+ /* Nonzero if the I/O limits are currently being ignored; generally
+ * it is zero. */
unsigned int io_limits_disabled;
/* The following fields are protected by the ThrottleGroup lock.
- * See the ThrottleGroup documentation for details. */
+ * See the ThrottleGroup documentation for details.
+ * throttle_state tells us if I/O limits are configured. */
ThrottleState *throttle_state;
ThrottleTimers throttle_timers;
unsigned pending_reqs[2];
--
2.9.3
^ permalink raw reply related [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (15 preceding siblings ...)
2017-01-13 13:17 ` [Qemu-devel] [PATCH 16/16] block: document fields protected by AioContext lock Paolo Bonzini
@ 2017-01-16 16:26 ` Fam Zheng
2017-01-18 16:02 ` Stefan Hajnoczi
2017-01-18 16:03 ` Stefan Hajnoczi
18 siblings, 0 replies; 69+ messages in thread
From: Fam Zheng @ 2017-01-16 16:26 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, stefanha
On Fri, 01/13 14:17, Paolo Bonzini wrote:
> Paolo Bonzini (16):
> aio: introduce aio_co_schedule and aio_co_wake
> block-backend: allow blk_prw from coroutine context
> test-thread-pool: use generic AioContext infrastructure
> io: add methods to set I/O handlers on AioContext
> io: make qio_channel_yield aware of AioContexts
> nbd: do not block on partial reply header reads
> coroutine-lock: reschedule coroutine on the AioContext it was running
> on
> qed: introduce qed_aio_start_io and qed_aio_next_io_cb
> aio: push aio_context_acquire/release down to dispatching
> block: explicitly acquire aiocontext in timers that need it
> block: explicitly acquire aiocontext in callbacks that need it
> block: explicitly acquire aiocontext in bottom halves that need it
> block: explicitly acquire aiocontext in aio callbacks that need it
> aio-posix: partially inline aio_dispatch into aio_poll
> async: remove unnecessary inc/dec pairs
> block: document fields protected by AioContext lock
I could only spot a few minor issues as noted in individual replies, otherwise
looks sane!
Fam
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (16 preceding siblings ...)
2017-01-16 16:26 ` [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Fam Zheng
@ 2017-01-18 16:02 ` Stefan Hajnoczi
2017-01-18 16:07 ` Paolo Bonzini
2017-01-18 16:03 ` Stefan Hajnoczi
18 siblings, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 16:02 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 365 bytes --]
On Fri, Jan 13, 2017 at 02:17:15PM +0100, Paolo Bonzini wrote:
> hw/block/virtio-blk.c | 19 +++-
> hw/scsi/scsi-bus.c | 2 +
> hw/scsi/scsi-disk.c | 15 +++
> hw/scsi/scsi-generic.c | 20 +++-
> hw/scsi/virtio-scsi.c | 6 ++
Are all the other device models safe because they don't support
IOThread?
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-13 13:17 [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2 Paolo Bonzini
` (17 preceding siblings ...)
2017-01-18 16:02 ` Stefan Hajnoczi
@ 2017-01-18 16:03 ` Stefan Hajnoczi
2017-01-18 16:31 ` Paolo Bonzini
18 siblings, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-18 16:03 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 6460 bytes --]
On Fri, Jan 13, 2017 at 02:17:15PM +0100, Paolo Bonzini wrote:
> This series pushes down aio_context_acquire/release to the point
> where we can actually reason on using different fine-grained mutexes.
>
> The main infrastructure is introduced in patch 1. The new API aio_co_wake
> starts a coroutine with aio_context_acquire/release protection, which
> requires tracking each coroutine's "home" AioContext. aio_co_schedule
> instead takes care of moving a sleeping coroutine to a different
> AioContext, also ensuring that it runs under aio_context_acquire/release.
> This is useful to implement bdrv_set_aio_context, as a simpler alternative
> to bottom halves. Even though one-shot BHs are already simpler than
> what we had before, after this patch aio_co_wake and aio_co_schedule
> save you from having to do aio_context_acquire/release explicitly.
>
> After patch 2 and 3, which are just small preparatory changes, patches
> 4 to 7 provide an example of how to use the new API. In particular patch
> 4 to 6 implement a new organization of coroutines in the NBD client,
> which allows not blocking on partial reply header reads.
>
> Patch 8 introduces helpers for AioContext locking in QED, which is
> the most complex AIO-based driver left. Then the actual meat of the
> series runs from patch 9 to patch 13, followed by small optimizations
> in patches 14 and 15.
>
> The patches do some back and forth in adding/removing
> aio_context_acquire/release calls in block/*.c but ultimately a small
> number of aio_context_acquire/release pairs are added after the pushdown.
> These are mostly in drivers that use external libraries (where they
> actually could already be replaced by QemuMutex) and in device models.
>
> Notably, coroutines need not care about aio_context_acquire/release.
> The device models ensure that the first creation of the coroutine has
> the AioContext, while aio_co_wake/aio_co_schedule do the same after
> they yield. Therefore, most of the files only need to use those two
> functions instead of, respectively, qemu_coroutine_enter and
> aio_bh_schedule_oneshot.
>
> However, this is only an intermediate step which is needed because the
> block layer and qemu-coroutine locks are thread-unsafe. So the next
> part will add separate locking, independent of AioContext, to block.c and
> mostly block/io.c---this includes making CoMutex thread-safe. Patch 16
> therefore already documents the current locking policies block.h to
> prepare for the next series.
>
> Paolo
>
> Paolo Bonzini (16):
> aio: introduce aio_co_schedule and aio_co_wake
> block-backend: allow blk_prw from coroutine context
> test-thread-pool: use generic AioContext infrastructure
> io: add methods to set I/O handlers on AioContext
> io: make qio_channel_yield aware of AioContexts
> nbd: do not block on partial reply header reads
> coroutine-lock: reschedule coroutine on the AioContext it was running
> on
> qed: introduce qed_aio_start_io and qed_aio_next_io_cb
> aio: push aio_context_acquire/release down to dispatching
> block: explicitly acquire aiocontext in timers that need it
> block: explicitly acquire aiocontext in callbacks that need it
> block: explicitly acquire aiocontext in bottom halves that need it
> block: explicitly acquire aiocontext in aio callbacks that need it
> aio-posix: partially inline aio_dispatch into aio_poll
> async: remove unnecessary inc/dec pairs
> block: document fields protected by AioContext lock
>
> aio-posix.c | 60 +++---------
> aio-win32.c | 30 ++----
> async.c | 81 ++++++++++++++--
> block/blkdebug.c | 9 +-
> block/blkreplay.c | 2 +-
> block/block-backend.c | 13 ++-
> block/curl.c | 44 ++++++---
> block/gluster.c | 9 +-
> block/io.c | 4 +-
> block/iscsi.c | 15 ++-
> block/linux-aio.c | 10 +-
> block/mirror.c | 12 ++-
> block/nbd-client.c | 108 ++++++++-------------
> block/nbd-client.h | 2 +-
> block/nfs.c | 9 +-
> block/qed-cluster.c | 2 +
> block/qed-table.c | 12 ++-
> block/qed.c | 58 +++++++----
> block/qed.h | 3 +
> block/sheepdog.c | 29 +++---
> block/ssh.c | 29 ++----
> block/throttle-groups.c | 2 +
> block/win32-aio.c | 9 +-
> dma-helpers.c | 2 +
> hw/block/virtio-blk.c | 19 +++-
> hw/scsi/scsi-bus.c | 2 +
> hw/scsi/scsi-disk.c | 15 +++
> hw/scsi/scsi-generic.c | 20 +++-
> hw/scsi/virtio-scsi.c | 6 ++
> include/block/aio.h | 38 +++++++-
> include/block/block_int.h | 64 ++++++++-----
> include/io/channel.h | 59 +++++++++++-
> include/qemu/coroutine_int.h | 10 +-
> include/sysemu/block-backend.h | 14 ++-
> io/channel-command.c | 13 +++
> io/channel-file.c | 11 +++
> io/channel-socket.c | 16 +++-
> io/channel-tls.c | 12 +++
> io/channel-watch.c | 6 ++
> io/channel.c | 97 +++++++++++++++----
> nbd/client.c | 2 +-
> nbd/common.c | 9 +-
> nbd/server.c | 4 +
> tests/Makefile.include | 15 ++-
> tests/iothread.c | 91 ++++++++++++++++++
> tests/iothread.h | 25 +++++
> tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++
> tests/test-thread-pool.c | 12 +--
> tests/test-vmstate.c | 11 ---
> thread-pool.c | 6 +-
> trace-events | 4 +
> util/qemu-coroutine-lock.c | 5 +-
> util/qemu-coroutine-sleep.c | 2 +-
> util/qemu-coroutine.c | 8 ++
> util/trace-events | 1 -
> 55 files changed, 1012 insertions(+), 352 deletions(-)
> create mode 100644 tests/iothread.c
> create mode 100644 tests/iothread.h
> create mode 100644 tests/test-aio-multithread.c
This is a big and somewhat risky change. Have you run any performance
benchmarks?
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-18 16:03 ` Stefan Hajnoczi
@ 2017-01-18 16:31 ` Paolo Bonzini
2017-01-19 17:01 ` Stefan Hajnoczi
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-18 16:31 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 7452 bytes --]
On 18/01/2017 17:03, Stefan Hajnoczi wrote:
> On Fri, Jan 13, 2017 at 02:17:15PM +0100, Paolo Bonzini wrote:
>> This series pushes down aio_context_acquire/release to the point
>> where we can actually reason on using different fine-grained mutexes.
>>
>> The main infrastructure is introduced in patch 1. The new API aio_co_wake
>> starts a coroutine with aio_context_acquire/release protection, which
>> requires tracking each coroutine's "home" AioContext. aio_co_schedule
>> instead takes care of moving a sleeping coroutine to a different
>> AioContext, also ensuring that it runs under aio_context_acquire/release.
>> This is useful to implement bdrv_set_aio_context, as a simpler alternative
>> to bottom halves. Even though one-shot BHs are already simpler than
>> what we had before, after this patch aio_co_wake and aio_co_schedule
>> save you from having to do aio_context_acquire/release explicitly.
>>
>> After patch 2 and 3, which are just small preparatory changes, patches
>> 4 to 7 provide an example of how to use the new API. In particular patch
>> 4 to 6 implement a new organization of coroutines in the NBD client,
>> which allows not blocking on partial reply header reads.
>>
>> Patch 8 introduces helpers for AioContext locking in QED, which is
>> the most complex AIO-based driver left. Then the actual meat of the
>> series runs from patch 9 to patch 13, followed by small optimizations
>> in patches 14 and 15.
>>
>> The patches do some back and forth in adding/removing
>> aio_context_acquire/release calls in block/*.c but ultimately a small
>> number of aio_context_acquire/release pairs are added after the pushdown.
>> These are mostly in drivers that use external libraries (where they
>> actually could already be replaced by QemuMutex) and in device models.
>>
>> Notably, coroutines need not care about aio_context_acquire/release.
>> The device models ensure that the first creation of the coroutine has
>> the AioContext, while aio_co_wake/aio_co_schedule do the same after
>> they yield. Therefore, most of the files only need to use those two
>> functions instead of, respectively, qemu_coroutine_enter and
>> aio_bh_schedule_oneshot.
>>
>> However, this is only an intermediate step which is needed because the
>> block layer and qemu-coroutine locks are thread-unsafe. So the next
>> part will add separate locking, independent of AioContext, to block.c and
>> mostly block/io.c---this includes making CoMutex thread-safe. Patch 16
>> therefore already documents the current locking policies block.h to
>> prepare for the next series.
>>
>> Paolo
>>
>> Paolo Bonzini (16):
>> aio: introduce aio_co_schedule and aio_co_wake
>> block-backend: allow blk_prw from coroutine context
>> test-thread-pool: use generic AioContext infrastructure
>> io: add methods to set I/O handlers on AioContext
>> io: make qio_channel_yield aware of AioContexts
>> nbd: do not block on partial reply header reads
>> coroutine-lock: reschedule coroutine on the AioContext it was running
>> on
>> qed: introduce qed_aio_start_io and qed_aio_next_io_cb
>> aio: push aio_context_acquire/release down to dispatching
>> block: explicitly acquire aiocontext in timers that need it
>> block: explicitly acquire aiocontext in callbacks that need it
>> block: explicitly acquire aiocontext in bottom halves that need it
>> block: explicitly acquire aiocontext in aio callbacks that need it
>> aio-posix: partially inline aio_dispatch into aio_poll
>> async: remove unnecessary inc/dec pairs
>> block: document fields protected by AioContext lock
>>
>> aio-posix.c | 60 +++---------
>> aio-win32.c | 30 ++----
>> async.c | 81 ++++++++++++++--
>> block/blkdebug.c | 9 +-
>> block/blkreplay.c | 2 +-
>> block/block-backend.c | 13 ++-
>> block/curl.c | 44 ++++++---
>> block/gluster.c | 9 +-
>> block/io.c | 4 +-
>> block/iscsi.c | 15 ++-
>> block/linux-aio.c | 10 +-
>> block/mirror.c | 12 ++-
>> block/nbd-client.c | 108 ++++++++-------------
>> block/nbd-client.h | 2 +-
>> block/nfs.c | 9 +-
>> block/qed-cluster.c | 2 +
>> block/qed-table.c | 12 ++-
>> block/qed.c | 58 +++++++----
>> block/qed.h | 3 +
>> block/sheepdog.c | 29 +++---
>> block/ssh.c | 29 ++----
>> block/throttle-groups.c | 2 +
>> block/win32-aio.c | 9 +-
>> dma-helpers.c | 2 +
>> hw/block/virtio-blk.c | 19 +++-
>> hw/scsi/scsi-bus.c | 2 +
>> hw/scsi/scsi-disk.c | 15 +++
>> hw/scsi/scsi-generic.c | 20 +++-
>> hw/scsi/virtio-scsi.c | 6 ++
>> include/block/aio.h | 38 +++++++-
>> include/block/block_int.h | 64 ++++++++-----
>> include/io/channel.h | 59 +++++++++++-
>> include/qemu/coroutine_int.h | 10 +-
>> include/sysemu/block-backend.h | 14 ++-
>> io/channel-command.c | 13 +++
>> io/channel-file.c | 11 +++
>> io/channel-socket.c | 16 +++-
>> io/channel-tls.c | 12 +++
>> io/channel-watch.c | 6 ++
>> io/channel.c | 97 +++++++++++++++----
>> nbd/client.c | 2 +-
>> nbd/common.c | 9 +-
>> nbd/server.c | 4 +
>> tests/Makefile.include | 15 ++-
>> tests/iothread.c | 91 ++++++++++++++++++
>> tests/iothread.h | 25 +++++
>> tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++
>> tests/test-thread-pool.c | 12 +--
>> tests/test-vmstate.c | 11 ---
>> thread-pool.c | 6 +-
>> trace-events | 4 +
>> util/qemu-coroutine-lock.c | 5 +-
>> util/qemu-coroutine-sleep.c | 2 +-
>> util/qemu-coroutine.c | 8 ++
>> util/trace-events | 1 -
>> 55 files changed, 1012 insertions(+), 352 deletions(-)
>> create mode 100644 tests/iothread.c
>> create mode 100644 tests/iothread.h
>> create mode 100644 tests/test-aio-multithread.c
>
> This is a big and somewhat risky change. Have you run any performance
> benchmarks?
Not recently; I ran them a year ago and there was no measurable difference.
However, this is also an intermediate state; the design is such that in
the end the performance-critical case (virtio-blk + linux-aio) will run
without taking any mutex at all(*). It will only use thread-local data,
or simple BDS statistics that can be updated with atomic_add or
atomic_cmpxchg. Fam's NVMe driver could also be extended to use
multiple hardware queues, one per QEMU thread, and avoid synchronization
completely.
Paolo
(*) Well, almost. tracked_request_begin/tracked_request_end still need
protection in the multiqueue case. But they can use a spinlock
since the critical section is very short; the expensive
CoMutex/CoQueue synchronization only happens for copy-on-read and
misaligned writes).
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-18 16:31 ` Paolo Bonzini
@ 2017-01-19 17:01 ` Stefan Hajnoczi
2017-01-20 16:39 ` Paolo Bonzini
0 siblings, 1 reply; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-19 17:01 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, famz
[-- Attachment #1: Type: text/plain, Size: 7074 bytes --]
On Wed, Jan 18, 2017 at 05:31:03PM +0100, Paolo Bonzini wrote:
>
>
> On 18/01/2017 17:03, Stefan Hajnoczi wrote:
> > On Fri, Jan 13, 2017 at 02:17:15PM +0100, Paolo Bonzini wrote:
> >> This series pushes down aio_context_acquire/release to the point
> >> where we can actually reason on using different fine-grained mutexes.
> >>
> >> The main infrastructure is introduced in patch 1. The new API aio_co_wake
> >> starts a coroutine with aio_context_acquire/release protection, which
> >> requires tracking each coroutine's "home" AioContext. aio_co_schedule
> >> instead takes care of moving a sleeping coroutine to a different
> >> AioContext, also ensuring that it runs under aio_context_acquire/release.
> >> This is useful to implement bdrv_set_aio_context, as a simpler alternative
> >> to bottom halves. Even though one-shot BHs are already simpler than
> >> what we had before, after this patch aio_co_wake and aio_co_schedule
> >> save you from having to do aio_context_acquire/release explicitly.
> >>
> >> After patch 2 and 3, which are just small preparatory changes, patches
> >> 4 to 7 provide an example of how to use the new API. In particular patch
> >> 4 to 6 implement a new organization of coroutines in the NBD client,
> >> which allows not blocking on partial reply header reads.
> >>
> >> Patch 8 introduces helpers for AioContext locking in QED, which is
> >> the most complex AIO-based driver left. Then the actual meat of the
> >> series runs from patch 9 to patch 13, followed by small optimizations
> >> in patches 14 and 15.
> >>
> >> The patches do some back and forth in adding/removing
> >> aio_context_acquire/release calls in block/*.c but ultimately a small
> >> number of aio_context_acquire/release pairs are added after the pushdown.
> >> These are mostly in drivers that use external libraries (where they
> >> actually could already be replaced by QemuMutex) and in device models.
> >>
> >> Notably, coroutines need not care about aio_context_acquire/release.
> >> The device models ensure that the first creation of the coroutine has
> >> the AioContext, while aio_co_wake/aio_co_schedule do the same after
> >> they yield. Therefore, most of the files only need to use those two
> >> functions instead of, respectively, qemu_coroutine_enter and
> >> aio_bh_schedule_oneshot.
> >>
> >> However, this is only an intermediate step which is needed because the
> >> block layer and qemu-coroutine locks are thread-unsafe. So the next
> >> part will add separate locking, independent of AioContext, to block.c and
> >> mostly block/io.c---this includes making CoMutex thread-safe. Patch 16
> >> therefore already documents the current locking policies block.h to
> >> prepare for the next series.
> >>
> >> Paolo
> >>
> >> Paolo Bonzini (16):
> >> aio: introduce aio_co_schedule and aio_co_wake
> >> block-backend: allow blk_prw from coroutine context
> >> test-thread-pool: use generic AioContext infrastructure
> >> io: add methods to set I/O handlers on AioContext
> >> io: make qio_channel_yield aware of AioContexts
> >> nbd: do not block on partial reply header reads
> >> coroutine-lock: reschedule coroutine on the AioContext it was running
> >> on
> >> qed: introduce qed_aio_start_io and qed_aio_next_io_cb
> >> aio: push aio_context_acquire/release down to dispatching
> >> block: explicitly acquire aiocontext in timers that need it
> >> block: explicitly acquire aiocontext in callbacks that need it
> >> block: explicitly acquire aiocontext in bottom halves that need it
> >> block: explicitly acquire aiocontext in aio callbacks that need it
> >> aio-posix: partially inline aio_dispatch into aio_poll
> >> async: remove unnecessary inc/dec pairs
> >> block: document fields protected by AioContext lock
> >>
> >> aio-posix.c | 60 +++---------
> >> aio-win32.c | 30 ++----
> >> async.c | 81 ++++++++++++++--
> >> block/blkdebug.c | 9 +-
> >> block/blkreplay.c | 2 +-
> >> block/block-backend.c | 13 ++-
> >> block/curl.c | 44 ++++++---
> >> block/gluster.c | 9 +-
> >> block/io.c | 4 +-
> >> block/iscsi.c | 15 ++-
> >> block/linux-aio.c | 10 +-
> >> block/mirror.c | 12 ++-
> >> block/nbd-client.c | 108 ++++++++-------------
> >> block/nbd-client.h | 2 +-
> >> block/nfs.c | 9 +-
> >> block/qed-cluster.c | 2 +
> >> block/qed-table.c | 12 ++-
> >> block/qed.c | 58 +++++++----
> >> block/qed.h | 3 +
> >> block/sheepdog.c | 29 +++---
> >> block/ssh.c | 29 ++----
> >> block/throttle-groups.c | 2 +
> >> block/win32-aio.c | 9 +-
> >> dma-helpers.c | 2 +
> >> hw/block/virtio-blk.c | 19 +++-
> >> hw/scsi/scsi-bus.c | 2 +
> >> hw/scsi/scsi-disk.c | 15 +++
> >> hw/scsi/scsi-generic.c | 20 +++-
> >> hw/scsi/virtio-scsi.c | 6 ++
> >> include/block/aio.h | 38 +++++++-
> >> include/block/block_int.h | 64 ++++++++-----
> >> include/io/channel.h | 59 +++++++++++-
> >> include/qemu/coroutine_int.h | 10 +-
> >> include/sysemu/block-backend.h | 14 ++-
> >> io/channel-command.c | 13 +++
> >> io/channel-file.c | 11 +++
> >> io/channel-socket.c | 16 +++-
> >> io/channel-tls.c | 12 +++
> >> io/channel-watch.c | 6 ++
> >> io/channel.c | 97 +++++++++++++++----
> >> nbd/client.c | 2 +-
> >> nbd/common.c | 9 +-
> >> nbd/server.c | 4 +
> >> tests/Makefile.include | 15 ++-
> >> tests/iothread.c | 91 ++++++++++++++++++
> >> tests/iothread.h | 25 +++++
> >> tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++
> >> tests/test-thread-pool.c | 12 +--
> >> tests/test-vmstate.c | 11 ---
> >> thread-pool.c | 6 +-
> >> trace-events | 4 +
> >> util/qemu-coroutine-lock.c | 5 +-
> >> util/qemu-coroutine-sleep.c | 2 +-
> >> util/qemu-coroutine.c | 8 ++
> >> util/trace-events | 1 -
> >> 55 files changed, 1012 insertions(+), 352 deletions(-)
> >> create mode 100644 tests/iothread.c
> >> create mode 100644 tests/iothread.h
> >> create mode 100644 tests/test-aio-multithread.c
> >
> > This is a big and somewhat risky change. Have you run any performance
> > benchmarks?
>
> Not recently; I ran them a year ago and there was no measurable difference.
Good, please re-run for the final version.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-19 17:01 ` Stefan Hajnoczi
@ 2017-01-20 16:39 ` Paolo Bonzini
2017-01-23 10:36 ` Stefan Hajnoczi
0 siblings, 1 reply; 69+ messages in thread
From: Paolo Bonzini @ 2017-01-20 16:39 UTC (permalink / raw)
To: Stefan Hajnoczi; +Cc: famz, qemu-devel
On 19/01/2017 18:01, Stefan Hajnoczi wrote:
>>> This is a big and somewhat risky change. Have you run any performance
>>> benchmarks?
>>
>> Not recently; I ran them a year ago and there was no measurable difference.
>
> Good, please re-run for the final version.
Done, went better than expected. :)
base
4k-1 null READ: io=6646.8MB, aggrb=226852KB/s, minb=226852KB/s, maxb=226852KB/s, mint=30000msec, maxt=30000msec
4k-1 nvme READ: io=3086.4MB, aggrb=105342KB/s, minb=105342KB/s, maxb=105342KB/s, mint=30001msec, maxt=30001msec
4k-32 nvme READ: io=14246MB, aggrb=486241KB/s, minb=486241KB/s, maxb=486241KB/s, mint=30001msec, maxt=30001msec
patched:
4k-1 null READ: io=7044.5MB, aggrb=240436KB/s, minb=240436KB/s, maxb=240436KB/s, mint=30000msec, maxt=30000msec
4k-1 nvme READ: io=3463.3MB, aggrb=118208KB/s, minb=118208KB/s, maxb=118208KB/s, mint=30001msec, maxt=30001msec
4k-32 nvme READ: io=15217MB, aggrb=519378KB/s, minb=519378KB/s, maxb=519378KB/s, mint=30001msec, maxt=30001msec
null is null-co://, nvme is a fast (180kIOPS on bare-metal) NVMe SSD.
So it seems to be some 5-10% faster. No idea why so I would not give too
much importance on the numbers, but at least it's not worse.
Paolo
^ permalink raw reply [flat|nested] 69+ messages in thread
* Re: [Qemu-devel] [PATCH 00/16] aio_context_acquire/release pushdown, part 2
2017-01-20 16:39 ` Paolo Bonzini
@ 2017-01-23 10:36 ` Stefan Hajnoczi
0 siblings, 0 replies; 69+ messages in thread
From: Stefan Hajnoczi @ 2017-01-23 10:36 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: Stefan Hajnoczi, famz, qemu-devel
[-- Attachment #1: Type: text/plain, Size: 1381 bytes --]
On Fri, Jan 20, 2017 at 05:39:40PM +0100, Paolo Bonzini wrote:
>
>
> On 19/01/2017 18:01, Stefan Hajnoczi wrote:
> >>> This is a big and somewhat risky change. Have you run any performance
> >>> benchmarks?
> >>
> >> Not recently; I ran them a year ago and there was no measurable difference.
> >
> > Good, please re-run for the final version.
>
> Done, went better than expected. :)
>
> base
> 4k-1 null READ: io=6646.8MB, aggrb=226852KB/s, minb=226852KB/s, maxb=226852KB/s, mint=30000msec, maxt=30000msec
> 4k-1 nvme READ: io=3086.4MB, aggrb=105342KB/s, minb=105342KB/s, maxb=105342KB/s, mint=30001msec, maxt=30001msec
> 4k-32 nvme READ: io=14246MB, aggrb=486241KB/s, minb=486241KB/s, maxb=486241KB/s, mint=30001msec, maxt=30001msec
>
> patched:
> 4k-1 null READ: io=7044.5MB, aggrb=240436KB/s, minb=240436KB/s, maxb=240436KB/s, mint=30000msec, maxt=30000msec
> 4k-1 nvme READ: io=3463.3MB, aggrb=118208KB/s, minb=118208KB/s, maxb=118208KB/s, mint=30001msec, maxt=30001msec
> 4k-32 nvme READ: io=15217MB, aggrb=519378KB/s, minb=519378KB/s, maxb=519378KB/s, mint=30001msec, maxt=30001msec
>
> null is null-co://, nvme is a fast (180kIOPS on bare-metal) NVMe SSD.
>
> So it seems to be some 5-10% faster. No idea why so I would not give too
> much importance on the numbers, but at least it's not worse.
Nice, thank you!
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]
^ permalink raw reply [flat|nested] 69+ messages in thread