When draining a block node, we recurse to its parent and for subtree
drains also to its children. A single AIO_WAIT_WHILE() is then used to
wait for bdrv_drain_poll() to become true, which depends on all of the
nodes we recursed to. However, if the respective child or parent becomes
quiescent and calls bdrv_wakeup(), only the AioWait of the child/parent
is checked, while AIO_WAIT_WHILE() depends on the AioWait of the
original node.
Fix this by using a single AioWait for all callers of AIO_WAIT_WHILE().
This may mean that the draining thread gets a few more unnecessary
wakeups because an unrelated operation got completed, but we already
wake it up when something _could_ have changed rather than only if it
has certainly changed.
Apart from that, drain is a slow path anyway. In theory it would be
possible to use wakeups more selectively and still correctly, but the
gains are likely not worth the additional complexity. In fact, this
patch is a nice simplification for some places in the code.
Signed-off-by: Kevin Wolf <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Max Reitz <[email protected]>
return bs ? bs->aio_context : qemu_get_aio_context();
}
-AioWait *bdrv_get_aio_wait(BlockDriverState *bs)
-{
- return bs ? &bs->wait : NULL;
-}
-
void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
{
aio_co_enter(bdrv_get_aio_context(bs), co);
* Accessed with atomic ops.
*/
unsigned int in_flight;
- AioWait wait;
};
typedef struct BlockBackendAIOCB {
static void blk_dec_in_flight(BlockBackend *blk)
{
atomic_dec(&blk->in_flight);
- aio_wait_kick(&blk->wait);
+ aio_wait_kick();
}
static void error_callback_bh(void *opaque)
}
/* We may have -ENOMEDIUM completions in flight */
- AIO_WAIT_WHILE(&blk->wait,
- blk_get_aio_context(blk),
- atomic_mb_read(&blk->in_flight) > 0);
+ AIO_WAIT_WHILE(blk_get_aio_context(blk),
+ atomic_mb_read(&blk->in_flight) > 0);
if (bs) {
bdrv_drained_end(bs);
aio_context_acquire(ctx);
/* We may have -ENOMEDIUM completions in flight */
- AIO_WAIT_WHILE(&blk->wait, ctx,
- atomic_mb_read(&blk->in_flight) > 0);
+ AIO_WAIT_WHILE(ctx, atomic_mb_read(&blk->in_flight) > 0);
aio_context_release(ctx);
}
/* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
#define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
-static AioWait drain_all_aio_wait;
-
static void bdrv_parent_cb_resize(BlockDriverState *bs);
static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
int64_t offset, int bytes, BdrvRequestFlags flags);
}
/* Now poll the in-flight requests */
- AIO_WAIT_WHILE(&drain_all_aio_wait, NULL, bdrv_drain_all_poll());
+ AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll());
while ((bs = bdrv_next_all_states(bs))) {
bdrv_drain_assert_idle(bs);
void bdrv_wakeup(BlockDriverState *bs)
{
- aio_wait_kick(bdrv_get_aio_wait(bs));
- aio_wait_kick(&drain_all_aio_wait);
+ aio_wait_kick();
}
void bdrv_dec_in_flight(BlockDriverState *bs)
return 0;
}
-void block_job_wakeup_all_bdrv(BlockJob *job)
-{
- GSList *l;
-
- for (l = job->nodes; l; l = l->next) {
- BdrvChild *c = l->data;
- bdrv_wakeup(c->bs);
- }
-}
-
static void block_job_on_idle(Notifier *n, void *opaque)
{
- BlockJob *job = opaque;
- block_job_wakeup_all_bdrv(job);
+ aio_wait_kick();
}
bool block_job_is_internal(BlockJob *job)
/**
* AioWait:
*
- * An object that facilitates synchronous waiting on a condition. The main
- * loop can wait on an operation running in an IOThread as follows:
+ * An object that facilitates synchronous waiting on a condition. A single
+ * global AioWait object (global_aio_wait) is used internally.
+ *
+ * The main loop can wait on an operation running in an IOThread as follows:
*
- * AioWait *wait = ...;
* AioContext *ctx = ...;
* MyWork work = { .done = false };
* schedule_my_work_in_iothread(ctx, &work);
- * AIO_WAIT_WHILE(wait, ctx, !work.done);
+ * AIO_WAIT_WHILE(ctx, !work.done);
*
* The IOThread must call aio_wait_kick() to notify the main loop when
* work.done changes:
* {
* ...
* work.done = true;
- * aio_wait_kick(wait);
+ * aio_wait_kick();
* }
*/
typedef struct {
unsigned num_waiters;
} AioWait;
+extern AioWait global_aio_wait;
+
/**
* AIO_WAIT_WHILE:
- * @wait: the aio wait object
* @ctx: the aio context, or NULL if multiple aio contexts (for which the
* caller does not hold a lock) are involved in the polling condition.
* @cond: wait while this conditional expression is true
* wait on conditions between two IOThreads since that could lead to deadlock,
* go via the main loop instead.
*/
-#define AIO_WAIT_WHILE(wait, ctx, cond) ({ \
+#define AIO_WAIT_WHILE(ctx, cond) ({ \
bool waited_ = false; \
- AioWait *wait_ = (wait); \
+ AioWait *wait_ = &global_aio_wait; \
AioContext *ctx_ = (ctx); \
/* Increment wait_->num_waiters before evaluating cond. */ \
atomic_inc(&wait_->num_waiters); \
/**
* aio_wait_kick:
- * @wait: the aio wait object that should re-evaluate its condition
- *
* Wake up the main thread if it is waiting on AIO_WAIT_WHILE(). During
* synchronous operations performed in an IOThread, the main thread lets the
* IOThread's event loop run, waiting for the operation to complete. A
* aio_wait_kick() call will wake up the main thread.
*/
-void aio_wait_kick(AioWait *wait);
+void aio_wait_kick(void);
/**
* aio_wait_bh_oneshot:
void bdrv_drain_all_end(void);
void bdrv_drain_all(void);
-/* Returns NULL when bs == NULL */
-AioWait *bdrv_get_aio_wait(BlockDriverState *bs);
-
#define BDRV_POLL_WHILE(bs, cond) ({ \
BlockDriverState *bs_ = (bs); \
- AIO_WAIT_WHILE(bdrv_get_aio_wait(bs_), \
- bdrv_get_aio_context(bs_), \
+ AIO_WAIT_WHILE(bdrv_get_aio_context(bs_), \
cond); })
int bdrv_pdiscard(BdrvChild *child, int64_t offset, int bytes);
unsigned int in_flight;
unsigned int serialising_in_flight;
- /* Kicked to signal main loop when a request completes. */
- AioWait wait;
-
/* counter for nested bdrv_io_plug.
* Accessed with atomic ops.
*/
*/
void block_job_remove_all_bdrv(BlockJob *job);
-/**
- * block_job_wakeup_all_bdrv:
- * @job: The block job
- *
- * Calls bdrv_wakeup() for all BlockDriverStates that have been added to the
- * job. This function is to be called whenever child_job_drained_poll() would
- * go from true to false to notify waiting drain requests.
- */
-void block_job_wakeup_all_bdrv(BlockJob *job);
-
/**
* block_job_set_speed:
* @job: The job to set the speed for.
int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
{
Error *local_err = NULL;
- AioWait dummy_wait = {};
int ret;
job_ref(job);
return -EBUSY;
}
- AIO_WAIT_WHILE(&dummy_wait, job->aio_context,
+ AIO_WAIT_WHILE(job->aio_context,
(job_drain(job), !job_is_completed(job)));
ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
#include "qemu/main-loop.h"
#include "block/aio-wait.h"
+AioWait global_aio_wait;
+
static void dummy_bh_cb(void *opaque)
{
/* The point is to make AIO_WAIT_WHILE()'s aio_poll() return */
}
-void aio_wait_kick(AioWait *wait)
+void aio_wait_kick(void)
{
/* The barrier (or an atomic op) is in the caller. */
- if (atomic_read(&wait->num_waiters)) {
+ if (atomic_read(&global_aio_wait.num_waiters)) {
aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb, NULL);
}
}
typedef struct {
- AioWait wait;
bool done;
QEMUBHFunc *cb;
void *opaque;
data->cb(data->opaque);
data->done = true;
- aio_wait_kick(&data->wait);
+ aio_wait_kick();
}
void aio_wait_bh_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
assert(qemu_get_current_aio_context() == qemu_get_aio_context());
aio_bh_schedule_oneshot(ctx, aio_wait_bh, &data);
- AIO_WAIT_WHILE(&data.wait, ctx, !data.done);
+ AIO_WAIT_WHILE(ctx, !data.done);
}
projects / qemu.git / commitdiff