block: seriously improve savevm performance
v3 cover block: seriously improve savevm performance
This series do standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to QCOW2 image,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.
In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations with non-cached operations. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters.
This patch series is an implementation of idea discussed in the RFC posted by Denis Plotnikov
https://lists.gnu.org/archive/html/qemu-devel/2020-04/msg01925.html
Results with this series over NVME are better than original code
original rfc this cached: 1.79s 2.38s 1.27s non-cached: 3.29s 1.31s 0.81s
Changes from v2:
- code moved from QCOW2 level to generic block level
- created bdrv_flush_vmstate helper to fix 022, 029 tests
- added recursive for bs->file in bdrv_co_flush_vmstate (fix 267)
- fixed blk_save_vmstate helper
- fixed coroutine wait as Vladimir suggested with waiting fixes from me
Changes from v1:
- patchew warning fixed
- fixed validation that only 1 waiter is allowed in patch 1
PATCH v3 1/4 migration/savevm: respect qemu_fclose() error code in save_snapshot()
qemu_fclose() could return error, f.e. if bdrv_co_flush() will return the error.
This validation will become more important once we will start waiting of asynchronous IO operations, started from bdrv_write_vmstate(), which are coming soon.
--- migration/savevm.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
migration/savevm.c +6 -2
@@ -2628,7 +2628,7 @@ int save_snapshot(const char *name, Error **errp) {BlockDriverState *bs, *bs1;QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; -int ret =-1 ; +int ret =-1 , ret2;QEMUFile *f;int saved_vm_running;uint64_t vm_state_size;
@@ -2712,10 +2712,14 @@ int save_snapshot(const char *name, Error **errp) } ret =qemu_savevm_state (f, errp); vm_state_size =qemu_ftell (f); -qemu_fclose (f); + ret2 =qemu_fclose (f);if (ret <0 ) { goto the_end; } +if (ret2 <0 ) { + ret = ret2; + goto the_end; + }/* The bdrv_all_create_snapshot() call that follows acquires the AioContext * for itself. BDRV_POLL_WHILE() does not support nested locking because
Signature · 1 line
2.17.1
Re: [PATCH 1/4] migration/savevm: respect qemu_fclose() error code in save_snapshot()
Quoted 7 lines · Denis V. Lunev
qemu_fclose()could return error, f.e. ifbdrv_co_flush()will return the error.This validation will become more important once we will start waiting of asynchronous IO operations, started from
bdrv_write_vmstate(), which are coming soon.
Quoted 9 lines
Signed-off-by Denis V. Lunev<[email protected]>CC Kevin Wolf<[email protected]>CC Max Reitz<[email protected]>CC Stefan Hajnoczi<[email protected]>CC Fam Zheng<[email protected]>CC Juan Quintela<[email protected]>CC "Dr. David Alan Gilbert"<[email protected]>CC Vladimir Sementsov-Ogievskiy<[email protected]>CC Denis Plotnikov<[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH 1/4] migration/savevm: respect qemu_fclose() error code in save_snapshot()
In reply to Denis V. LunevQuoted 7 lines · * Denis V. Lunev ([email protected]) wrote:
qemu_fclose()could return error, f.e. ifbdrv_co_flush()will return the error.This validation will become more important once we will start waiting of asynchronous IO operations, started from
bdrv_write_vmstate(), which are coming soon.
Quoted 9 lines
Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>
We check the return value in very few other places; I think in the migration case we do flushes and assume that if the flushes work we were OK; then most of the closes happen on error paths or after points we think we're done.
Quoted 36 lines
--- migration/savevm.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)migration/savevm.c +6
-2@@ -2628,7 +2628,7 @@ int save_snapshot(const char *name, Error **errp) {BlockDriverState *bs, *bs1;QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; -int ret =-1 ; +int ret =-1 , ret2;QEMUFile *f;int saved_vm_running;uint64_t vm_state_size;@@ -2712,10 +2712,14 @@ int save_snapshot(const char *name, Error **errp) } ret =qemu_savevm_state (f, errp); vm_state_size =qemu_ftell (f); -qemu_fclose (f); + ret2 =qemu_fclose (f);if (ret <0 ) { goto the_end; } +if (ret2 <0 ) { + ret = ret2; + goto the_end; + }/* The bdrv_all_create_snapshot() call that follows acquires the AioContext * for itself. BDRV_POLL_WHILE() does not support nested locking because Signature · 1 line
2.17.1
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
PATCH v3 2/4 block/aio_task: allow start/wait task from any coroutine
In reply to Denis V. LunevFrom: Vladimir Sementsov-Ogievskiy <[email protected]>
Currently, aio task pool assumes that there is a main coroutine, which creates tasks and wait for them. Let's remove the restriction by using CoQueue. Code becomes clearer, interface more obvious.
--- block/aio_task.c | 21 ++++++--------------- 1 file changed, 6 insertions(+), 15 deletions(-)
block/aio_task.c +6 -15
@@ -27,11 +27,10 @@#include "block/aio_task.h" struct AioTaskPool { -Coroutine *main_co;int status;int max_busy_tasks;int busy_tasks; -bool waiting; +CoQueue waiters; }; static void coroutine_fn aio_task_co(void *opaque)
@@ -52,31 +51,23 @@ static void coroutine_fn aio_task_co(void *opaque)g_free (task); -if (pool->waiting) { - pool->waiting = false; -aio_co_wake (pool->main_co); - } +qemu_co_queue_restart_all (&pool->waiters); } void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool) {assert (pool->busy_tasks >0 ); -assert (qemu_coroutine_self () == pool->main_co); - pool->waiting = true; -qemu_coroutine_yield (); +qemu_co_queue_wait (&pool->waiters,NULL ); -assert (!pool->waiting);assert (pool->busy_tasks < pool->max_busy_tasks); } void coroutine_fn aio_task_pool_wait_slot(AioTaskPool *pool) { -if (pool->busy_tasks < pool->max_busy_tasks) { -return ; +while (pool->busy_tasks >= pool->max_busy_tasks) { +aio_task_pool_wait_one (pool); } - -aio_task_pool_wait_one (pool); } void coroutine_fn aio_task_pool_wait_all(AioTaskPool *pool)
@@ -98,8 +89,8 @@ AioTaskPool *coroutine_fn aio_task_pool_new(int max_busy_tasks) {AioTaskPool *pool =g_new0 (AioTaskPool,1 ); - pool->main_co =qemu_coroutine_self (); pool->max_busy_tasks = max_busy_tasks; +qemu_co_queue_init (&pool->waiters);return pool; }
Signature · 1 line
2.17.1
Re: [PATCH 2/4] block/aio_task: allow start/wait task from any coroutine
Quoted 49 lines · Denis V. Lunev
From: Vladimir Sementsov-Ogievskiy <[email protected]>
Currently, aio task pool assumes that there is a main coroutine, which creates tasks and wait for them. Let's remove the restriction by using CoQueue. Code becomes clearer, interface more obvious.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>--- block/aio_task.c | 21 ++++++--------------- 1 file changed, 6 insertions(+), 15 deletions(-)block/aio_task.c +2
-6@@ -27,11 +27,10 @@#include "block/aio_task.h" struct AioTaskPool { -Coroutine *main_co;int status;int max_busy_tasks;int busy_tasks; -bool waiting; +CoQueue waiters; }; static void coroutine_fn aio_task_co(void *opaque)@@ -52,31 +51,23 @@ static void coroutine_fn aio_task_co(void *opaque)g_free (task); -if (pool->waiting) { - pool->waiting = false; -aio_co_wake (pool->main_co); - } +qemu_co_queue_restart_all (&pool->waiters); } void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool)
block/aio_task.c static void coroutine_fn aio_task_co(void *opaque) · line 63
Quoted 10 lines
block/aio_task.c +1
-4{assert (pool->busy_tasks >0 ); -assert (qemu_coroutine_self () == pool->main_co); - pool->waiting = true; -qemu_coroutine_yield (); +qemu_co_queue_wait (&pool->waiters,NULL ); -assert (!pool->waiting);assert (pool->busy_tasks < pool->max_busy_tasks);
As we wake up several coroutines now, I'm afraid this assertion may start to fire.
And aio_task_pool_wait_one() becomes useless as a public API (still, it's used only locally, so we can make it static).
I'll send updated patch after reviewing the rest of the series.
Quoted 25 lines
block/aio_task.c +3
-5} void coroutine_fn aio_task_pool_wait_slot(AioTaskPool *pool) { -if (pool->busy_tasks < pool->max_busy_tasks) { -return ; +while (pool->busy_tasks >= pool->max_busy_tasks) { +aio_task_pool_wait_one (pool); } - -aio_task_pool_wait_one (pool); } void coroutine_fn aio_task_pool_wait_all(AioTaskPool *pool)@@ -98,8 +89,8 @@ AioTaskPool *coroutine_fn aio_task_pool_new(int max_busy_tasks) {AioTaskPool *pool =g_new0 (AioTaskPool,1 ); - pool->main_co =qemu_coroutine_self (); pool->max_busy_tasks = max_busy_tasks; +qemu_co_queue_init (&pool->waiters);return pool; }
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH 2/4] block/aio_task: allow start/wait task from any coroutine
Quoted 57 lines · Vladimir Sementsov-Ogievskiy
Quoted 42 lines · Denis V. Lunev
From: Vladimir Sementsov-Ogievskiy <[email protected]>
Currently, aio task pool assumes that there is a main coroutine, which creates tasks and wait for them. Let's remove the restriction by using CoQueue. Code becomes clearer, interface more obvious.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>---
block/aio_task.c| 21 ++++++---------------1 file changed, 6 insertions(+), 15 deletions(-)block/aio_task.c +2
-6@@ -27,11 +27,10 @@ #include "block/aio_task.h" struct AioTaskPool { - Coroutine *main_co; int status; int max_busy_tasks; int busy_tasks; - bool waiting; + CoQueue waiters; }; static void coroutine_fn aio_task_co(void *opaque)@@ -52,31 +51,23 @@ static void coroutine_fn aio_task_co(void *opaque) g_free(task); - if (pool->waiting) { - pool->waiting = false; - aio_co_wake(pool->main_co); - } + qemu_co_queue_restart_all(&pool->waiters);block/aio_task.c static void coroutine_fn aio_task_co(void *opaque) · line 58
Quoted 10 lines
block/aio_task.c +1
-4} void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool) { assert(pool->busy_tasks > 0); - assert(qemu_coroutine_self() == pool->main_co); - pool->waiting = true; - qemu_coroutine_yield(); + qemu_co_queue_wait(&pool->waiters, NULL); - assert(!pool->waiting); assert(pool->busy_tasks < pool->max_busy_tasks);As we wake up several coroutines now, I'm afraid this assertion may start to fire.
Andaio_task_pool_wait_one()becomes useless as a public API (still, it's used only locally, so we can make it static).
Quoted 2 lines
I'll send updated patch after reviewing the rest of the series.
Hm, OK, we have two kinds of waiters: waiting for a slot and for all tasks to finish. So, either we need two queues, or do like this patch (one queue, but wake-up all waiters, for them to check does their condition satisfied or not).
I'm OK with this patch with the following squashed-in:
include/block/aio_task.h +0 -1
@@ -48,7 +48,6 @@ bool aio_task_pool_empty(AioTaskPool *pool); void coroutine_fn aio_task_pool_start_task(AioTaskPool *pool, AioTask *task); void coroutine_fn aio_task_pool_wait_slot(AioTaskPool *pool); -void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool); void coroutine_fn aio_task_pool_wait_all(AioTaskPool *pool);#endif /* BLOCK_AIO_TASK_H */
block/aio_task.c +2 -11
@@ -54,26 +54,17 @@ static void coroutine_fn aio_task_co(void *opaque)qemu_co_queue_restart_all (&pool->waiters); } -void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool) -{ -assert (pool->busy_tasks >0 ); - -qemu_co_queue_wait (&pool->waiters,NULL ); - -assert (pool->busy_tasks < pool->max_busy_tasks); -} - void coroutine_fn aio_task_pool_wait_slot(AioTaskPool *pool) {while (pool->busy_tasks >= pool->max_busy_tasks) { -aio_task_pool_wait_one (pool); +qemu_co_queue_wait (&pool->waiters,NULL ); } } void coroutine_fn aio_task_pool_wait_all(AioTaskPool *pool) {while (pool->busy_tasks >0 ) { -aio_task_pool_wait_one (pool); +qemu_co_queue_wait (&pool->waiters,NULL ); } }
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH 2/4] block/aio_task: allow start/wait task from any coroutine
Quoted 108 lines · Vladimir Sementsov-Ogievskiy
Quoted 59 lines · Vladimir Sementsov-Ogievskiy
Quoted 42 lines · Denis V. Lunev
From: Vladimir Sementsov-Ogievskiy <[email protected]>
Currently, aio task pool assumes that there is a main coroutine, which creates tasks and wait for them. Let's remove the restriction by using CoQueue. Code becomes clearer, interface more obvious.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>---
block/aio_task.c| 21 ++++++---------------1 file changed, 6 insertions(+), 15 deletions(-)block/aio_task.c +2
-6@@ -27,11 +27,10 @@ #include "block/aio_task.h" struct AioTaskPool { - Coroutine *main_co; int status; int max_busy_tasks; int busy_tasks; - bool waiting; + CoQueue waiters; }; static void coroutine_fn aio_task_co(void *opaque)@@ -52,31 +51,23 @@ static void coroutine_fn aio_task_co(void *opaque) g_free(task); - if (pool->waiting) { - pool->waiting = false; - aio_co_wake(pool->main_co); - } + qemu_co_queue_restart_all(&pool->waiters);block/aio_task.c static void coroutine_fn aio_task_co(void *opaque) · line 58
Quoted 10 lines
block/aio_task.c +1
-4} void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool) { assert(pool->busy_tasks > 0); - assert(qemu_coroutine_self() == pool->main_co); - pool->waiting = true; - qemu_coroutine_yield(); + qemu_co_queue_wait(&pool->waiters, NULL); - assert(!pool->waiting); assert(pool->busy_tasks < pool->max_busy_tasks);As we wake up several coroutines now, I'm afraid this assertion may start to fire.
Andaio_task_pool_wait_one()becomes useless as a public API (still, it's used only locally, so we can make it static).Quoted 2 lines
I'll send updated patch after reviewing the rest of the series.
Hm, OK, we have two kinds of waiters: waiting for a slot and for all tasks to finish. So, either we need two queues, or do like this patch (one queue, but wake-up all waiters, for them to check does their condition satisfied or not).
I'm OK with this patch with the following squashed-in:
include/block/aio_task.h +0
-0@@ -48,7 +48,6 @@ bool aio_task_pool_empty(AioTaskPool *pool);void
coroutine_fnaio_task_pool_start_task(AioTaskPool *pool, AioTask *task);
voidcoroutine_fnaio_task_pool_wait_slot(AioTaskPool *pool); -voidcoroutine_fnaio_task_pool_wait_one(AioTaskPool *pool); voidcoroutine_fnaio_task_pool_wait_all(AioTaskPool *pool);
#endif /*BLOCK_AIO_TASK_H*/block/aio_task.c +0
-0@@ -54,26 +54,17 @@ static void coroutine_fn aio_task_co(void *opaque) qemu_co_queue_restart_all(&pool->waiters); }block/aio_task.c +0
-9-void coroutine_fn aio_task_pool_wait_one(AioTaskPool *pool) -{ - assert(pool->busy_tasks > 0); - - qemu_co_queue_wait(&pool->waiters, NULL); - - assert(pool->busy_tasks < pool->max_busy_tasks); -} -void
coroutine_fnaio_task_pool_wait_slot(AioTaskPool *pool){
while (pool->busy_tasks>=pool->max_busy_tasks) { -aio_task_pool_wait_one(pool);
+qemu_co_queue_wait(&pool->waiters, NULL); }
}
Quoted 11 lines
void
coroutine_fnaio_task_pool_wait_all(AioTaskPool *pool){
while (pool->busy_tasks> 0) {
-aio_task_pool_wait_one(pool);
+qemu_co_queue_wait(&pool->waiters, NULL); }
}
I'd better make this separate
Den
PATCH v3 3/4 block, migration: add bdrv_flush_vmstate helper
In reply to Denis V. LunevRight now bdrv_fclose() is just calling bdrv_flush().
The problem is that migration code is working inefficently from black layer terms and are frequently called for very small pieces of not properly aligned data. Block layer is capable to work this way, but this is very slow.
This patch is a preparation for the introduction of the intermediate buffer at block driver state. It would be beneficial to separate conventional bdrv_flush() from closing QEMU file from migration code.
The patch also forces bdrv_flush_vmstate() operation inside synchronous blk_save_vmstate() operation. This helper is used from qemu-io only.
--- block/block-backend.c | 6 +++++- block/io.c | 39 +++++++++++++++++++++++++++++++++++++++ include/block/block.h | 1 + migration/savevm.c | 3 +++ 4 files changed, 48 insertions(+), 1 deletion(-)
block/block-backend.c +5 -1
@@ -2177,7 +2177,7 @@ int blk_truncate(BlockBackend *blk, int64_t offset, bool exact,int blk_save_vmstate (BlockBackend *blk,const uint8_t *buf,int64_t pos,int size) { -int ret; +int ret, ret2;if (!blk_is_available (blk)) {return -ENOMEDIUM ;
@@ -2187,6 +2187,10 @@ int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,if (ret <0 ) {return ret; } + ret2 =bdrv_flush_vmstate (blk_bs (blk)); +if (ret2 <0 ) { +return ret; + }if (ret == size && !blk->enable_write_cache) { ret =bdrv_flush (blk_bs (blk));
block/io.c +39 -0
@@ -2725,6 +2725,45 @@ int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)return bdrv_rw_vmstate (bs, qiov, pos, true); } + +typedef struct FlushVMStateCo { +BlockDriverState *bs; +int ret; +}FlushVMStateCo ; + +static int coroutine_fn bdrv_co_flush_vmstate(BlockDriverState *bs) +{ +return 0 ; +} + +static void coroutine_fn bdrv_flush_vmstate_co_entry(void *opaque) +{ +FlushVMStateCo *rwco = opaque; + + rwco->ret =bdrv_co_flush_vmstate (rwco->bs); +aio_wait_kick (); +} + +int bdrv_flush_vmstate (BlockDriverState *bs) +{ +Coroutine *co; +FlushVMStateCo flush_co = { + .bs = bs, + .ret =NOT_DONE , + }; + +if (qemu_in_coroutine ()) { +/* Fast-path if already in coroutine context */ +bdrv_flush_vmstate_co_entry (&flush_co); + }else { + co =qemu_coroutine_create (bdrv_flush_vmstate_co_entry, &flush_co); +bdrv_coroutine_enter (bs, co); +BDRV_POLL_WHILE (bs, flush_co.ret ==NOT_DONE ); + } + +return flush_co.ret; +} +/**************************************************************/ /* async I/Os */
include/block/block.h +1 -0
@@ -572,6 +572,7 @@ int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,int bdrv_load_vmstate (BlockDriverState *bs,uint8_t *buf,int64_t pos,int size); +int bdrv_flush_vmstate (BlockDriverState *bs); void bdrv_img_create(const char *filename, const char *fmt, const char *base_filename, const char *base_fmt,
migration/savevm.c +3 -0
@@ -150,6 +150,9 @@ static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,static int bdrv_fclose (void *opaque,Error **errp) { +int err =bdrv_flush_vmstate (opaque); +if (err <0 ) +return err;return bdrv_flush (opaque); }
Signature · 1 line
2.17.1
Re: [PATCH 3/4] block, migration: add bdrv_flush_vmstate helper
Quoted 91 lines · Denis V. Lunev
Right now
bdrv_fclose()is just callingbdrv_flush().The problem is that migration code is working inefficently from black layer terms and are frequently called for very small pieces of not properly aligned data. Block layer is capable to work this way, but this is very slow.
This patch is a preparation for the introduction of the intermediate buffer at block driver state. It would be beneficial to separate conventional
bdrv_flush()from closing QEMU file from migration code.The patch also forces
bdrv_flush_vmstate()operation inside synchronousblk_save_vmstate()operation. This helper is used from qemu-io only.Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>--- block/block-backend.c | 6 +++++- block/io.c | 39 +++++++++++++++++++++++++++++++++++++++ include/block/block.h | 1 + migration/savevm.c | 3 +++ 4 files changed, 48 insertions(+), 1 deletion(-)block/block-backend.c +5
-1@@ -2177,7 +2177,7 @@ int blk_truncate(BlockBackend *blk, int64_t offset, bool exact,int blk_save_vmstate (BlockBackend *blk,const uint8_t *buf,int64_t pos,int size) { -int ret; +int ret, ret2;if (!blk_is_available (blk)) {return -ENOMEDIUM ;@@ -2187,6 +2187,10 @@ int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,if (ret <0 ) {return ret; } + ret2 =bdrv_flush_vmstate (blk_bs (blk)); +if (ret2 <0 ) { +return ret; + }if (ret == size && !blk->enable_write_cache) { ret =bdrv_flush (blk_bs (blk));block/io.c +28
-0@@ -2725,6 +2725,45 @@ int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)return bdrv_rw_vmstate (bs, qiov, pos, true); } + +typedef struct FlushVMStateCo { +BlockDriverState *bs; +int ret; +}FlushVMStateCo ; + +static int coroutine_fn bdrv_co_flush_vmstate(BlockDriverState *bs) +{ +return 0 ; +} + +static void coroutine_fn bdrv_flush_vmstate_co_entry(void *opaque) +{ +FlushVMStateCo *rwco = opaque; + + rwco->ret =bdrv_co_flush_vmstate (rwco->bs); +aio_wait_kick (); +} + +int bdrv_flush_vmstate (BlockDriverState *bs) +{ +Coroutine *co; +FlushVMStateCo flush_co = { + .bs = bs, + .ret =NOT_DONE , + }; + +if (qemu_in_coroutine ()) {
block/io.c int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) · line 2765
Quoted 10 lines
block/io.c +10
-0+/* Fast-path if already in coroutine context */ +bdrv_flush_vmstate_co_entry (&flush_co); + }else { + co =qemu_coroutine_create (bdrv_flush_vmstate_co_entry, &flush_co); +bdrv_coroutine_enter (bs, co); +BDRV_POLL_WHILE (bs, flush_co.ret ==NOT_DONE ); + } + +return flush_co.ret; +}
In block/io.c, since 7d2410cea154bf, these coroutine wrappers are using bdrv_run_co() instead.
I hope, it's an intermediate state, and my series with auto-generated wrappers will be applied, still, now we should use bdrv_run_co()-based approach for consistency.
Otherwise, patch looks OK for me, just add a new interface, doing nothing for now. Still, would be good to add a comment in block.h, that bdrv_flush_vmsate is needed after bdrv_save_vmstate.
Quoted 31 lines
block/io.c +1
-0+/**************************************************************/ /* async I/Os */ include/block/block.h +1
-0@@ -572,6 +572,7 @@ int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,int bdrv_load_vmstate (BlockDriverState *bs,uint8_t *buf,int64_t pos,int size); +int bdrv_flush_vmstate (BlockDriverState *bs); void bdrv_img_create(const char *filename, const char *fmt, const char *base_filename, const char *base_fmt,migration/savevm.c +3
-0@@ -150,6 +150,9 @@ static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,static int bdrv_fclose (void *opaque,Error **errp) { +int err =bdrv_flush_vmstate (opaque); +if (err <0 ) +return err;return bdrv_flush (opaque); }
Signature · 2 lines
Best regards, Vladimir
PATCH v3 4/4 block/io: improve savevm performance
In reply to Denis V. LunevThis patch does 2 standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to block driver,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.
Thus bdrv_vmstate_write() is becoming asynchronous. All pending operations completion are performed in newly invented bdrv_flush_vmstate().
In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations if target file descriptor is opened with O_DIRECT. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters even on cached file descriptors.
Snapshot creation time (2 GB Fedora-31 VM running over NVME storage):
original fixed cached: 1.79s 1.27s non-cached: 3.29s 0.81s
The difference over HDD would be more significant :)
--- block/io.c | 121 +++++++++++++++++++++++++++++++++++++- include/block/block_int.h | 8 +++ 2 files changed, 127 insertions(+), 2 deletions(-)
block/io.c +119 -2
@@ -26,6 +26,7 @@#include "trace.h" #include "sysemu/block-backend.h" #include "block/aio-wait.h" +#include "block/aio_task.h" #include "block/blockjob.h" #include "block/blockjob_int.h" #include "block/block_int.h"
@@ -2633,6 +2634,102 @@ typedef struct BdrvVmstateCo {int ret; } BdrvVmstateCo; +typedef struct BdrvVMStateTask { +AioTask task; + +BlockDriverState *bs; +int64_t offset; +void *buf; +size_t bytes; +}BdrvVMStateTask ; + +typedef struct BdrvSaveVMState { +AioTaskPool *pool; +BdrvVMStateTask *t; +}BdrvSaveVMState ; + + +static coroutine_fn int bdrv_co_vmstate_save_task_entry(AioTask *task) +{ +int err =0 ; +BdrvVMStateTask *t =container_of (task, BdrvVMStateTask, task); + +if (t->bytes !=0 ) { +QEMUIOVector local_qiov; +qemu_iovec_init_buf (&local_qiov, t->buf, t->bytes); + +bdrv_inc_in_flight (t->bs); + err = t->bs->drv->bdrv_save_vmstate (t->bs, &local_qiov, t->offset); +bdrv_dec_in_flight (t->bs); + } + +qemu_vfree (t->buf); +return err; +} + +static BdrvVMStateTask *bdrv_vmstate_task_create (BlockDriverState *bs, +int64_t pos,size_t size) +{ +BdrvVMStateTask *t =g_new (BdrvVMStateTask,1 ); + + *t = (BdrvVMStateTask ) { + .task.func = bdrv_co_vmstate_save_task_entry, + .buf =qemu_blockalign (bs, size), + .offset = pos, + .bs = bs, + }; + +return t; +} + +static int bdrv_co_do_save_vmstate (BlockDriverState *bs,QEMUIOVector *qiov, +int64_t pos) +{ +BdrvSaveVMState *state = bs->savevm_state; +BdrvVMStateTask *t; +size_t buf_size =MAX (bdrv_get_cluster_size (bs),1 * MiB); +size_t to_copy; +size_t off; + +if (state ==NULL ) { + state =g_new (BdrvSaveVMState,1 ); + *state = (BdrvSaveVMState ) { + .pool =aio_task_pool_new (BDRV_VMSTATE_WORKERS_MAX ), + .t =bdrv_vmstate_task_create (bs, pos, buf_size), + }; + + bs->savevm_state = state; + } + +if (aio_task_pool_status (state->pool) <0 ) { +return aio_task_pool_status (state->pool); + } + + t = state->t; +if (t->offset + t->bytes != pos) { +/* Normally this branch is not reachable from migration */ +return bs->drv->bdrv_save_vmstate (bs, qiov, pos); + } + + off =0 ; +while (1 ) { + to_copy =MIN (qiov->size - off, buf_size - t->bytes); +qemu_iovec_to_buf (qiov, off, t->buf + t->bytes, to_copy); + t->bytes += to_copy; +if (t->bytes < buf_size) { +return qiov->size; + } + +aio_task_pool_start_task (state->pool, &t->task); + + pos += to_copy; + off += to_copy; + state->t = t =bdrv_vmstate_task_create (bs, pos, buf_size); + } + +return qiov->size; +} + static int coroutine_fnbdrv_co_rw_vmstate (BlockDriverState *bs,QEMUIOVector *qiov,int64_t pos,bool is_read)
@@ -2648,7 +2745,7 @@ bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,if (is_read) { ret = drv->bdrv_load_vmstate (bs, qiov, pos); }else { - ret = drv->bdrv_save_vmstate (bs, qiov, pos); + ret =bdrv_co_do_save_vmstate (bs, qiov, pos); } }else if (bs->file) { ret =bdrv_co_rw_vmstate (bs->file->bs, qiov, pos, is_read);
@@ -2733,7 +2830,27 @@ typedef struct FlushVMStateCo { static int coroutine_fn bdrv_co_flush_vmstate(BlockDriverState *bs) { -return 0 ; +int err; +BdrvSaveVMState *state = bs->savevm_state; + +if (bs->drv->bdrv_save_vmstate ==NULL && bs->file !=NULL ) { +return bdrv_co_flush_vmstate (bs->file->bs); + } +if (state ==NULL ) { +return 0 ; + } + +aio_task_pool_start_task (state->pool, &state->t->task); + +aio_task_pool_wait_all (state->pool); + err =aio_task_pool_status (state->pool); + +aio_task_pool_free (state->pool); +g_free (state); + + bs->savevm_state =NULL ; + +return err; } static void coroutine_fn bdrv_flush_vmstate_co_entry(void *opaque)
include/block/block_int.h +8 -0
@@ -61,6 +61,8 @@#define BLOCK_PROBE_BUF_SIZE 512 +#define BDRV_VMSTATE_WORKERS_MAX 8 +enum BdrvTrackedRequestType {BDRV_TRACKED_READ ,BDRV_TRACKED_WRITE ,
@@ -784,6 +786,9 @@ struct BdrvChild {QLIST_ENTRY (BdrvChild) next_parent; }; + +typedef struct BdrvSaveVMState BdrvSaveVMState ; +/* * Note: the function bdrv_append() copies and swaps contents of * BlockDriverStates, so if you add new fields to this struct, please
@@ -947,6 +952,9 @@ struct BlockDriverState {/* BdrvChild links to this node may never be frozen */ bool never_freeze; + +/* Intermediate buffer for VM state saving from snapshot creation code */ +BdrvSaveVMState *savevm_state; };struct BlockBackendRootState {
Signature · 1 line
2.17.1
Re: [PATCH 4/4] block/io: improve savevm performance
Quoted 67 lines · Denis V. Lunev
This patch does 2 standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to block driver,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.Thus
bdrv_vmstate_write()is becoming asynchronous. All pending operations completion are performed in newly inventedbdrv_flush_vmstate().In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations if target file descriptor is opened with
O_DIRECT. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters even on cached file descriptors.Snapshot creation time (2 GB Fedora-31 VM running over NVME storage):
original fixedcached: 1.79s 1.27s non-cached: 3.29s 0.81sThe difference over HDD would be more significant :)
Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>--- block/io.c | 121 +++++++++++++++++++++++++++++++++++++- include/block/block_int.h | 8 +++ 2 files changed, 127 insertions(+), 2 deletions(-)block/io.c +14
-0@@ -26,6 +26,7 @@#include "trace.h" #include "sysemu/block-backend.h" #include "block/aio-wait.h" +#include "block/aio_task.h" #include "block/blockjob.h" #include "block/blockjob_int.h" #include "block/block_int.h" @@ -2633,6 +2634,102 @@ typedef struct BdrvVmstateCo {int ret; } BdrvVmstateCo; +typedef struct BdrvVMStateTask { +AioTask task; + +BlockDriverState *bs; +int64_t offset; +void *buf; +size_t bytes; +}BdrvVMStateTask ; + +typedef struct BdrvSaveVMState { +AioTaskPool *pool; +BdrvVMStateTask *t; +}BdrvSaveVMState ;
block/io.c typedef struct BdrvVmstateCo { · line 2659
Quoted 10 lines
block/io.c +10
-0+ + +static coroutine_fn int bdrv_co_vmstate_save_task_entry(AioTask *task) +{ +int err =0 ; +BdrvVMStateTask *t =container_of (task, BdrvVMStateTask, task); + +if (t->bytes !=0 ) { +QEMUIOVector local_qiov; +qemu_iovec_init_buf (&local_qiov, t->buf, t->bytes);
Consider special oneliner for this case:
QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, t->buf, t->bytes);
Quoted 36 lines
block/io.c +36
-0+ +bdrv_inc_in_flight (t->bs); + err = t->bs->drv->bdrv_save_vmstate (t->bs, &local_qiov, t->offset); +bdrv_dec_in_flight (t->bs); + } + +qemu_vfree (t->buf); +return err; +} + +static BdrvVMStateTask *bdrv_vmstate_task_create (BlockDriverState *bs, +int64_t pos,size_t size) +{ +BdrvVMStateTask *t =g_new (BdrvVMStateTask,1 ); + + *t = (BdrvVMStateTask ) { + .task.func = bdrv_co_vmstate_save_task_entry, + .buf =qemu_blockalign (bs, size), + .offset = pos, + .bs = bs, + }; + +return t; +} + +static int bdrv_co_do_save_vmstate (BlockDriverState *bs,QEMUIOVector *qiov, +int64_t pos) +{ +BdrvSaveVMState *state = bs->savevm_state; +BdrvVMStateTask *t; +size_t buf_size =MAX (bdrv_get_cluster_size (bs),1 * MiB); +size_t to_copy; +size_t off; + +if (state ==NULL ) { + state =g_new (BdrvSaveVMState,1 );
block/io.c typedef struct BdrvVmstateCo { · line 2705
Quoted 10 lines
block/io.c +10
-0+ *state = (BdrvSaveVMState ) { + .pool =aio_task_pool_new (BDRV_VMSTATE_WORKERS_MAX ), + .t =bdrv_vmstate_task_create (bs, pos, buf_size), + }; + + bs->savevm_state = state; + } + +if (aio_task_pool_status (state->pool) <0 ) { +return aio_task_pool_status (state->pool);
So, on failure, we still need flush, to cleanup the state. I think better is to cleanup it immediately here (goto fail, etc.).
Aha, actually in blk_save_vmstate(), you don't do bdrv_flush_vmstate() if bdrv_save_vmstate() failed.
block/io.c typedef struct BdrvVmstateCo { · line 2710
Quoted 5 lines
block/io.c +5
-0+ } + + t = state->t; +if (t->offset + t->bytes != pos) { +/* Normally this branch is not reachable from migration */
Aha, like a cache-miss. OK
Quoted 10 lines
block/io.c +10
-0+return bs->drv->bdrv_save_vmstate (bs, qiov, pos); + } + + off =0 ; +while (1 ) { + to_copy =MIN (qiov->size - off, buf_size - t->bytes); +qemu_iovec_to_buf (qiov, off, t->buf + t->bytes, to_copy); + t->bytes += to_copy; +if (t->bytes < buf_size) { +return qiov->size;
block/io.c typedef struct BdrvVmstateCo { · line 2730
Quoted 10 lines
block/io.c +10
-0+ } + +aio_task_pool_start_task (state->pool, &t->task); + + pos += to_copy; + off += to_copy; + state->t = t =bdrv_vmstate_task_create (bs, pos, buf_size); + } + +return qiov->size;
this is unreachable actually. So, I'd drop it or do "break" in a loop instead of return.
Quoted 20 lines
block/io.c +4
-2+} + static int coroutine_fnbdrv_co_rw_vmstate (BlockDriverState *bs,QEMUIOVector *qiov,int64_t pos,bool is_read)@@ -2648,7 +2745,7 @@ bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,if (is_read) { ret = drv->bdrv_load_vmstate (bs, qiov, pos); }else { - ret = drv->bdrv_save_vmstate (bs, qiov, pos); + ret =bdrv_co_do_save_vmstate (bs, qiov, pos); } }else if (bs->file) { ret =bdrv_co_rw_vmstate (bs->file->bs, qiov, pos, is_read);@@ -2733,7 +2830,27 @@ typedef struct FlushVMStateCo { static int coroutine_fn bdrv_co_flush_vmstate(BlockDriverState *bs) { -return 0 ; +int err;
block/io.c typedef struct FlushVMStateCo { · line 2843
Quoted 10 lines
block/io.c +10
-0+BdrvSaveVMState *state = bs->savevm_state; + +if (bs->drv->bdrv_save_vmstate ==NULL && bs->file !=NULL ) { +return bdrv_co_flush_vmstate (bs->file->bs); + } +if (state ==NULL ) { +return 0 ; + } + +aio_task_pool_start_task (state->pool, &state->t->task);
We probably shouldn't do it, if pool_status is already bad. But it shouldn't hurt as is.
Quoted 47 lines
block/io.c +10
-0+ +aio_task_pool_wait_all (state->pool); + err =aio_task_pool_status (state->pool); + +aio_task_pool_free (state->pool); +g_free (state); + + bs->savevm_state =NULL ; + +return err; } static void coroutine_fn bdrv_flush_vmstate_co_entry(void *opaque)include/block/block_int.h +8
-0@@ -61,6 +61,8 @@#define BLOCK_PROBE_BUF_SIZE 512 +#define BDRV_VMSTATE_WORKERS_MAX 8 +enum BdrvTrackedRequestType {BDRV_TRACKED_READ ,BDRV_TRACKED_WRITE ,@@ -784,6 +786,9 @@ struct BdrvChild {QLIST_ENTRY (BdrvChild) next_parent; }; + +typedef struct BdrvSaveVMState BdrvSaveVMState ; +/* * Note: the function bdrv_append() copies and swaps contents of * BlockDriverStates, so if you add new fields to this struct, please @@ -947,6 +952,9 @@ struct BlockDriverState {/* BdrvChild links to this node may never be frozen */ bool never_freeze; + +/* Intermediate buffer for VM state saving from snapshot creation code */ +BdrvSaveVMState *savevm_state; };struct BlockBackendRootState {
Signature · 2 lines
Best regards, Vladimir
Quoted 6 lines · * Denis V. Lunev ([email protected]) wrote:
This series do standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to QCOW2 image,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.
Quoted 6 lines
In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations with non-cached operations. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters.
It surprises me a little that you're not benefiting from the buffer internal to qemu-file.c
Dave
Quoted 19 lines
This patch series is an implementation of idea discussed in the RFC posted by Denis Plotnikov
https://lists.gnu.org/archive/html/qemu-devel/2020-04/msg01925.html
Results with this series over NVME are better than original codeoriginal rfc this cached: 1.79s 2.38s 1.27s non-cached: 3.29s 1.31s 0.81sChanges from v2:
- code moved from QCOW2 level to generic block level
- createdbdrv_flush_vmstatehelper to fix 022, 029 tests
- added recursive forbs->fileinbdrv_co_flush_vmstate(fix 267)
- fixedblk_save_vmstatehelper
- fixed coroutine wait as Vladimir suggested with waiting fixes from meChanges from v1:
- patchew warning fixed
- fixed validation that only 1 waiter is allowed in patch 1
Quoted 11 lines
Signed-off-by Denis V. Lunev <[email protected]>CC Kevin Wolf <[email protected]>CC Max Reitz <[email protected]>CC Stefan Hajnoczi <[email protected]>CC Fam Zheng <[email protected]>CC Juan Quintela <[email protected]>CC "Dr. David Alan Gilbert" <[email protected]>CC Vladimir Sementsov-Ogievskiy <[email protected]>CC Denis Plotnikov <[email protected]>
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Quoted 13 lines · Dr. David Alan Gilbert
Quoted 6 lines · * Denis V. Lunev ([email protected]) wrote:
This series do standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to QCOW2 image,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.Quoted 6 lines
In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations with non-cached operations. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters.
Quoted 4 lines
It surprises me a little that you're not benefiting from the buffer internal to
qemu-file.cDave
There are a lot of problems with this buffer:
Flushes to block driver state are performed in the abstract places, pushing
a) small IO
b) non-aligned IO both to
1) page size
2) cluster size
It should also be noted that buffer in QEMU file is quite small and all IO operations with it are synchronous. IO, like ethernet, wants good queues.
The difference is on the table.
Den
Quoted 20 lines · * Denis V. Lunev ([email protected]) wrote:
Quoted 13 lines · Dr. David Alan Gilbert
Quoted 6 lines · * Denis V. Lunev ([email protected]) wrote:
This series do standard basic things:
- it creates intermediate buffer for all writes from QEMU migration code to QCOW2 image,
- this buffer is sent to disk asynchronously, allowing several writes to run in parallel.Quoted 6 lines
In general, migration code is fantastically inefficent (by observation), buffers are not aligned and sent with arbitrary pieces, a lot of time less than 100 bytes at a chunk, which results in read-modify-write operations with non-cached operations. It should also be noted that all operations are performed into unallocated image blocks, which also suffer due to partial writes to such new clusters.
Quoted 4 lines
It surprises me a little that you're not benefiting from the buffer internal to
qemu-file.cDave
There are a lot of problems with this buffer:
Quoted 9 lines
Flushes to block driver state are performed in the abstract places, pushing
 a) small IO
 b) non-aligned IO both to
      1) page size
      2) cluster size
It should also be noted that buffer in QEMU file is quite small and all IO operations with it are synchronous. IO, like ethernet, wants good queues.
Yeh, for ethernet we immediately get the kernels buffer so it's not too bad; and I guess the async page writes are easier as well.
Dave
Quoted 4 lines
The difference is on the table.
Den
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK