ublk: fix ublk_queue_rqs() and selftests test_stress_03
Subsystems (3)
cover ublk: fix ublk_queue_rqs() and selftests test_stress_03
Hi Jens,
The 1st patch fixes ublk_queue_rqs().
The 2nd patch fixes test_stress_03.sh.
Thanks,
Ming Lei (2):
ublk: build per-io-ring-ctx batch list
selftests: ublk: don't take same backing file for more than one ublk
devices
drivers/block/ublk_drv.c | 17 +++++++++-------- tools/testing/selftests/ublk/test_stress_03.sh | 5 +++-- 2 files changed, 12 insertions(+), 10 deletions(-)
Signature · 1 line
2.47.0
PATCH 1/2 ublk: build per-io-ring-ctx batch list
ublk_queue_cmd_list() dispatches the whole batch list by scheduling task work via the tail request's io_uring_cmd, this way is fine even though more than one io_ring_ctx are involved for this batch since it is just one running context.
However, the task work handler ublk_cmd_list_tw_cb() takes issue_flags of tail uring_cmd's io_ring_ctx for completing all commands. This way is wrong if any uring_cmd is issued from different io_ring_ctx.
Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single io_ring_ctx is often taken in each daemon.
--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)
drivers/block/ublk_drv.c +9 -8
@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);
@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data; -struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd; -if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list); + cmd = this_cmd;if (ublk_prep_req (this_q, req, true) ==BLK_STS_OK )rq_list_add_tail (&submit_list, req);
@@ -1438,7 +1439,7 @@ static void ublk_queue_rqs(struct rq_list *rqlist) }if (!rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); +ublk_queue_cmd_list (cmd, &submit_list); *rqlist = requeue_list; }
Signature · 1 line
2.47.0
Re: [PATCH 1/2] ublk: build per-io-ring-ctx batch list
Quoted 48 lines · Ming Lei
ublk_queue_cmd_list()dispatches the whole batch list by scheduling task work via the tail request'sio_uring_cmd, this way is fine even though more than oneio_ring_ctxare involved for this batch since it is just one running context.However, the task work handler
ublk_cmd_list_tw_cb()takesissue_flagsof tail uring_cmd'sio_ring_ctxfor completing all commands. This way is wrong if anyuring_cmdis issued from differentio_ring_ctx.Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single
io_ring_ctxis often taken in each daemon.Fixes d796cea7b9f3 ("ublk: implement ->queue_rqs()")Fixes ab03a61c6614 ("ublk: have a per-io daemon instead of a per-queue daemon")Signed-off-by Ming Lei <[email protected]>--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)drivers/block/ublk_drv.c +2
-3@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data;
drivers/block/ublk_drv.c static void ublk_queue_rqs(struct rq_list *rqlist) · line 1432
Quoted 10 lines
drivers/block/ublk_drv.c +5
-4-struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd; -if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list);
I don't think we can assume that ublk commands submitted to the same io_uring have the same daemon task. It's possible for multiple tasks to submit to the same io_uring, even though that's not a common or performant way to use io_uring. Probably we need to check that both the task and io_ring_ctx match.
Best,
Caleb
Quoted 16 lines
drivers/block/ublk_drv.c +2
-2+ cmd = this_cmd;if (ublk_prep_req (this_q, req, true) ==BLK_STS_OK )rq_list_add_tail (&submit_list, req);@@ -1438,7 +1439,7 @@ static void ublk_queue_rqs(struct rq_list *rqlist) }if (!rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); +ublk_queue_cmd_list (cmd, &submit_list); *rqlist = requeue_list; } --2.47.0
Re: [PATCH 1/2] ublk: build per-io-ring-ctx batch list
Quoted 52 lines · Caleb Sander Mateos
Quoted 51 lines · Ming Lei
ublk_queue_cmd_list()dispatches the whole batch list by scheduling task work via the tail request'sio_uring_cmd, this way is fine even though more than oneio_ring_ctxare involved for this batch since it is just one running context.However, the task work handler
ublk_cmd_list_tw_cb()takesissue_flagsof tail uring_cmd'sio_ring_ctxfor completing all commands. This way is wrong if anyuring_cmdis issued from differentio_ring_ctx.Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single
io_ring_ctxis often taken in each daemon.Fixes d796cea7b9f3 ("ublk: implement ->queue_rqs()")Fixes ab03a61c6614 ("ublk: have a per-io daemon instead of a per-queue daemon")Signed-off-by Ming Lei <[email protected]>--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)drivers/block/ublk_drv.c +3
-4@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data; -struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd;
Quoted 13 lines
drivers/block/ublk_drv.c static void ublk_queue_rqs(struct rq_list *rqlist) · line 1432
Quoted 7 lines
drivers/block/ublk_drv.c +4
-3-if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list);I don't think we can assume that ublk commands submitted to the same
io_uringhave the same daemon task. It's possible for multiple tasks to submit to the sameio_uring, even though that's not a common or performant way to useio_uring. Probably we need to check that both the task andio_ring_ctxmatch.
Here the problem is in 'issue_flags' passed from io_uring, especially for grabbing io_ring_ctx lock.
If two uring_cmd are issued via same io_ring_ctx from two tasks, it is fine to share 'issue_flags' from one of tasks, what matters is that the io_ring_ctx lock is handled correctly when calling io_uring_cmd_done().
Thanks,
Ming
Re: [PATCH 1/2] ublk: build per-io-ring-ctx batch list
Quoted 68 lines · Ming Lei
Quoted 52 lines · Caleb Sander Mateos
Quoted 51 lines · Ming Lei
ublk_queue_cmd_list()dispatches the whole batch list by scheduling task work via the tail request'sio_uring_cmd, this way is fine even though more than oneio_ring_ctxare involved for this batch since it is just one running context.However, the task work handler
ublk_cmd_list_tw_cb()takesissue_flagsof tail uring_cmd'sio_ring_ctxfor completing all commands. This way is wrong if anyuring_cmdis issued from differentio_ring_ctx.Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single
io_ring_ctxis often taken in each daemon.Fixes d796cea7b9f3 ("ublk: implement ->queue_rqs()")Fixes ab03a61c6614 ("ublk: have a per-io daemon instead of a per-queue daemon")Signed-off-by Ming Lei <[email protected]>--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)drivers/block/ublk_drv.c +3
-4@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data; -struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd;Quoted 13 lines
drivers/block/ublk_drv.c static void ublk_queue_rqs(struct rq_list *rqlist) · line 1432
Quoted 7 lines
drivers/block/ublk_drv.c +4
-3-if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list);I don't think we can assume that ublk commands submitted to the same
io_uringhave the same daemon task. It's possible for multiple tasks to submit to the sameio_uring, even though that's not a common or performant way to useio_uring. Probably we need to check that both the task andio_ring_ctxmatch.
Quoted 6 lines
Here the problem is in '
issue_flags' passed fromio_uring, especially for grabbingio_ring_ctxlock.If two
uring_cmdare issued via sameio_ring_ctxfrom two tasks, it is fine to share 'issue_flags' from one of tasks, what matters is that theio_ring_ctxlock is handled correctly when callingio_uring_cmd_done().
Right, I understand the issue you are trying to solve. I agree it's a problem for submit_list to contain commands from multiple io_ring_ctxs. But it's also a problem if it contains commands with different daemon tasks, because ublk_queue_cmd_list() will schedule ublk_cmd_list_tw_cb() to be called in the *last command's task*. But ublk_cmd_list_tw_cb() will call ublk_dispatch_req() for all the commands in the list. So if submit_list contains commands with multiple daemon tasks, ublk_dispatch_req() will fail on the current != io->task check. So I still feel we need to call ublk_queue_cmd_list(io, &submit_list) if io->task != this_io->task (as well as if the io_ring_ctxs differ).
Best,
Caleb
Re: [PATCH 1/2] ublk: build per-io-ring-ctx batch list
Quoted 76 lines · Caleb Sander Mateos
Quoted 68 lines · Ming Lei
Quoted 52 lines · Caleb Sander Mateos
Quoted 51 lines · Ming Lei
ublk_queue_cmd_list()dispatches the whole batch list by scheduling task work via the tail request'sio_uring_cmd, this way is fine even though more than oneio_ring_ctxare involved for this batch since it is just one running context.However, the task work handler
ublk_cmd_list_tw_cb()takesissue_flagsof tail uring_cmd'sio_ring_ctxfor completing all commands. This way is wrong if anyuring_cmdis issued from differentio_ring_ctx.Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single
io_ring_ctxis often taken in each daemon.Fixes d796cea7b9f3 ("ublk: implement ->queue_rqs()")Fixes ab03a61c6614 ("ublk: have a per-io daemon instead of a per-queue daemon")Signed-off-by Ming Lei <[email protected]>--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)drivers/block/ublk_drv.c +3
-4@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data; -struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd;Quoted 13 lines
drivers/block/ublk_drv.c static void ublk_queue_rqs(struct rq_list *rqlist) · line 1432
Quoted 7 lines
drivers/block/ublk_drv.c +4
-3-if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list);I don't think we can assume that ublk commands submitted to the same
io_uringhave the same daemon task. It's possible for multiple tasks to submit to the sameio_uring, even though that's not a common or performant way to useio_uring. Probably we need to check that both the task andio_ring_ctxmatch.Quoted 6 lines
Here the problem is in '
issue_flags' passed fromio_uring, especially for grabbingio_ring_ctxlock.If two
uring_cmdare issued via sameio_ring_ctxfrom two tasks, it is fine to share 'issue_flags' from one of tasks, what matters is that theio_ring_ctxlock is handled correctly when callingio_uring_cmd_done().
Quoted 11 lines
Right, I understand the issue you are trying to solve. I agree it's a problem for
submit_listto contain commands from multipleio_ring_ctxs. But it's also a problem if it contains commands with different daemon tasks, becauseublk_queue_cmd_list()will scheduleublk_cmd_list_tw_cb()to be called in the *last command's task*. Butublk_cmd_list_tw_cb()will callublk_dispatch_req()for all the commands in the list. So ifsubmit_listcontains commands with multiple daemon tasks,ublk_dispatch_req()will fail on the current !=io->taskcheck. So I still feel we need to callublk_queue_cmd_list(io, &submit_list)ifio->task!=this_io->task(as well as if theio_ring_ctxsdiffer).
Indeed, I will send a V2 for covering different task case.
Jens, can you drop this patch?
Thanks,
Ming
Jens Axboe
maintainerRe: [PATCH 1/2] ublk: build per-io-ring-ctx batch list
Quoted 89 lines · Ming Lei
Quoted 76 lines · Caleb Sander Mateos
Quoted 68 lines · Ming Lei
Quoted 52 lines · Caleb Sander Mateos
Quoted 51 lines · Ming Lei
ublk_queue_cmd_list()dispatches the whole batch list by scheduling task work via the tail request'sio_uring_cmd, this way is fine even though more than oneio_ring_ctxare involved for this batch since it is just one running context.However, the task work handler
ublk_cmd_list_tw_cb()takesissue_flagsof tail uring_cmd'sio_ring_ctxfor completing all commands. This way is wrong if anyuring_cmdis issued from differentio_ring_ctx.Fixes it by always building per-io-ring-ctx batch list.
For typical per-queue or per-io daemon implementation, this way shouldn't make difference from performance viewpoint, because single
io_ring_ctxis often taken in each daemon.Fixes d796cea7b9f3 ("ublk: implement ->queue_rqs()")Fixes ab03a61c6614 ("ublk: have a per-io daemon instead of a per-queue daemon")Signed-off-by Ming Lei <[email protected]>--- drivers/block/ublk_drv.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)drivers/block/ublk_drv.c +3
-4@@ -1336,9 +1336,8 @@ static void ublk_cmd_list_tw_cb(struct io_uring_cmd *cmd, }while (rq); } -static void ublk_queue_cmd_list (struct ublk_io *io,struct rq_list *l) +static void ublk_queue_cmd_list (struct io_uring_cmd *cmd,struct rq_list *l) { -struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); pdu->req_list =rq_list_peek (l);@@ -1420,16 +1419,18 @@ static void ublk_queue_rqs(struct rq_list *rqlist) {struct rq_list requeue_list = { };struct rq_list submit_list = { }; -struct ublk_io *io =NULL ; +struct io_uring_cmd *cmd =NULL ;struct request *req;while ((req =rq_list_pop (rqlist))) {struct ublk_queue *this_q = req->mq_hctx->driver_data; -struct ublk_io *this_io = &this_q->ios[req->tag]; +struct io_uring_cmd *this_cmd = this_q->ios[req->tag].cmd;Quoted 13 lines
drivers/block/ublk_drv.c static void ublk_queue_rqs(struct rq_list *rqlist) · line 1432
Quoted 7 lines
drivers/block/ublk_drv.c +4
-3-if (io && io->task != this_io->task && !rq_list_empty (&submit_list)) -ublk_queue_cmd_list (io, &submit_list); - io = this_io; +if (cmd &&io_uring_cmd_ctx_handle (cmd) != +io_uring_cmd_ctx_handle (this_cmd) && + !rq_list_empty (&submit_list)) +ublk_queue_cmd_list (cmd, &submit_list);I don't think we can assume that ublk commands submitted to the same
io_uringhave the same daemon task. It's possible for multiple tasks to submit to the sameio_uring, even though that's not a common or performant way to useio_uring. Probably we need to check that both the task andio_ring_ctxmatch.Quoted 6 lines
Here the problem is in '
issue_flags' passed fromio_uring, especially for grabbingio_ring_ctxlock.If two
uring_cmdare issued via sameio_ring_ctxfrom two tasks, it is fine to share 'issue_flags' from one of tasks, what matters is that theio_ring_ctxlock is handled correctly when callingio_uring_cmd_done().Quoted 11 lines
Right, I understand the issue you are trying to solve. I agree it's a problem for
submit_listto contain commands from multipleio_ring_ctxs. But it's also a problem if it contains commands with different daemon tasks, becauseublk_queue_cmd_list()will scheduleublk_cmd_list_tw_cb()to be called in the *last command's task*. Butublk_cmd_list_tw_cb()will callublk_dispatch_req()for all the commands in the list. So ifsubmit_listcontains commands with multiple daemon tasks,ublk_dispatch_req()will fail on the current !=io->taskcheck. So I still feel we need to callublk_queue_cmd_list(io, &submit_list)ifio->task!=this_io->task(as well as if theio_ring_ctxsdiffer).
Quoted 3 lines
Indeed, I will send a V2 for covering different task case.
Jens, can you drop this patch?
Done
Signature · 1 line
Jens Axboe
PATCH 2/2 selftests: ublk: don't take same backing file for more than one ublk devices
In reply to Ming LeiDon't use same backing file for more than one ublk devices, and avoid concurrent write on same file from more ublk disks.
--- tools/testing/selftests/ublk/test_stress_03.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
tools/testing/selftests/ublk/test_stress_03.sh +3 -2
@@ -32,22 +32,23 @@ _create_backfile 2 128M ublk_io_and_remove 8G -t null -q 4 -z & ublk_io_and_remove 256M -t loop -q 4 -z "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 -z "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & +wait if _have_feature "AUTO_BUF_REG"; then ublk_io_and_remove 8G -t null -q 4 --auto_zc & ublk_io_and_remove 256M -t loop -q 4 --auto_zc "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 --auto_zc "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & ublk_io_and_remove 8G -t null -q 4 -z --auto_zc --auto_zc_fallback & + wait fi -wait if _have_feature "PER_IO_DAEMON"; then ublk_io_and_remove 8G -t null -q 4 --auto_zc --nthreads 8 --per_io_tasks & ublk_io_and_remove 256M -t loop -q 4 --auto_zc --nthreads 8 --per_io_tasks "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 --auto_zc --nthreads 8 --per_io_tasks "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & ublk_io_and_remove 8G -t null -q 4 -z --auto_zc --auto_zc_fallback --nthreads 8 --per_io_tasks & + wait fi -wait _cleanup_test "stress" _show_result $TID $ERR_CODE
Signature · 1 line
2.47.0
Re: [PATCH 2/2] selftests: ublk: don't take same backing file for more than one ublk devices
Quoted 15 lines · Ming Lei
Don't use same backing file for more than one ublk devices, and avoid concurrent write on same file from more ublk disks.
Fixes 8ccebc19ee3d ("selftests: ublk: support UBLK_F_AUTO_BUF_REG")Signed-off-by Ming Lei <[email protected]>--- tools/testing/selftests/ublk/test_stress_03.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)tools/testing/selftests/ublk/test_stress_03.sh +0
-0
tools/testing/selftests/ublk/test_stress_03.sh _create_backfile 2 128M · line 35
Quoted 4 lines
tools/testing/selftests/ublk/test_stress_03.sh +1
-0@@ -32,22 +32,23 @@ _create_backfile 2 128M ublk_io_and_remove 8G -t null -q 4 -z & ublk_io_and_remove 256M -t loop -q 4 -z "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 -z "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & +wait
Why is wait necessary here? It looks like __run_io_and_remove, which is called from run_io_and_remove, already ends with a wait. Am I missing something?
Best,
Caleb
Quoted 24 lines
tools/testing/selftests/ublk/test_stress_03.sh +2
-3if _have_feature "AUTO_BUF_REG"; then ublk_io_and_remove 8G -t null -q 4 --auto_zc & ublk_io_and_remove 256M -t loop -q 4 --auto_zc "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 --auto_zc "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & ublk_io_and_remove 8G -t null -q 4 -z --auto_zc --auto_zc_fallback & + wait fi -wait if _have_feature "PER_IO_DAEMON"; then ublk_io_and_remove 8G -t null -q 4 --auto_zc --nthreads 8 --per_io_tasks & ublk_io_and_remove 256M -t loop -q 4 --auto_zc --nthreads 8 --per_io_tasks "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 --auto_zc --nthreads 8 --per_io_tasks "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & ublk_io_and_remove 8G -t null -q 4 -z --auto_zc --auto_zc_fallback --nthreads 8 --per_io_tasks & + wait fi -wait _cleanup_test "stress" _show_result $TID $ERR_CODE --2.47.0
Re: [PATCH 2/2] selftests: ublk: don't take same backing file for more than one ublk devices
Quoted 11 lines · Caleb Sander Mateos
Quoted 10 lines · Ming Lei
Don't use same backing file for more than one ublk devices, and avoid concurrent write on same file from more ublk disks.
Fixes 8ccebc19ee3d ("selftests: ublk: support UBLK_F_AUTO_BUF_REG")Signed-off-by Ming Lei <[email protected]>--- tools/testing/selftests/ublk/test_stress_03.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
Quoted 13 lines
tools/testing/selftests/ublk/test_stress_03.sh _create_backfile 2 128M · line 35
Quoted 9 lines
tools/testing/selftests/ublk/test_stress_03.sh +1
-0@@ -32,22 +32,23 @@ _create_backfile 2 128M ublk_io_and_remove 8G -t null -q 4 -z & ublk_io_and_remove 256M -t loop -q 4 -z "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 -z "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & +waitWhy is wait necessary here? It looks like
__run_io_and_remove, which is called fromrun_io_and_remove, already ends with a wait. Am I missing something?
All tests share the three backing files, this way just avoids concurrent write to the same file from each test/ublk device.
Thanks,
Ming
Caleb Sander Mateos
Reviewed-byRe: [PATCH 2/2] selftests: ublk: don't take same backing file for more than one ublk devices
Quoted 27 lines · Ming Lei
Quoted 11 lines · Caleb Sander Mateos
Quoted 10 lines · Ming Lei
Don't use same backing file for more than one ublk devices, and avoid concurrent write on same file from more ublk disks.
Fixes 8ccebc19ee3d ("selftests: ublk: support UBLK_F_AUTO_BUF_REG")Signed-off-by Ming Lei <[email protected]>--- tools/testing/selftests/ublk/test_stress_03.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)Quoted 13 lines
tools/testing/selftests/ublk/test_stress_03.sh _create_backfile 2 128M · line 35
Quoted 9 lines
tools/testing/selftests/ublk/test_stress_03.sh +1
-0@@ -32,22 +32,23 @@ _create_backfile 2 128M ublk_io_and_remove 8G -t null -q 4 -z & ublk_io_and_remove 256M -t loop -q 4 -z "${UBLK_BACKFILES[0]}" & ublk_io_and_remove 256M -t stripe -q 4 -z "${UBLK_BACKFILES[1]}" "${UBLK_BACKFILES[2]}" & +waitWhy is wait necessary here? It looks like
__run_io_and_remove, which is called fromrun_io_and_remove, already ends with a wait. Am I missing something?
Quoted 2 lines
All tests share the three backing files, this way just avoids concurrent write to the same file from each test/ublk device.
Sorry, I didn't see the "&" at the end of the ublk_io_and_remove commands.
Quoted 8 lines · Ming Lei
The 1st patch fixes
ublk_queue_rqs().The 2nd patch fixes
test_stress_03.sh.Thanks,
[...]
Applied, thanks!
[1/2] ublk: build per-io-ring-ctx batch list
commit: e867d6c4d51a564943e88e08fb1b27bc68a81b49
[2/2] selftests: ublk: don't take same backing file for more than one ublk devices
commit: 4cb1a3d63eb7d4d0263b2d3ab4f61555263b69aaBest regards,
Signature · 1 line
Jens Axboe