* [PATCH] dmaengine: dmatest: terminate channel on timeout or error
@ 2026-11-18 10:08 Amita Mishra
2026-11-18 10:16 ` sashiko-bot
0 siblings, 1 reply; 2+ messages in thread
From: Amita Mishra @ 2026-11-18 10:08 UTC (permalink / raw)
To: Vinod Koul
Cc: Frank Li, Dan Williams, dmaengine, linux-kernel, stable,
Amita Mishra
In interrupt mode (!params->polled), when wait_event_freezable_timeout()
times out (!done->done) or dma_async_is_tx_complete() reports a
non-DMA_COMPLETE status, dmatest_func() jumps to error_unmap_continue and
calls dmaengine_unmap_put() without terminating the DMA channel first.
Channel termination via dmaengine_terminate_sync() only happens later
when the test thread exits, unlike the polled path which terminates the
channel immediately after dma_sync_wait().
If the DMA controller is still actively transferring when the source and
destination buffers are unmapped, an IOMMU-backed DMA engine will trigger
use-after-unmap translation faults, or overwrite the buffers while the
next iteration reinitializes them. In addition, if the timed-out transfer
completes during a subsequent iteration, its callback sets the shared
per-thread done flag and wakes done_wait prematurely, causing the next
iteration to fail with "completion busy status".
Terminate the channel with dmaengine_terminate_sync() on both the timeout
and non-DMA_COMPLETE error paths before unmapping the buffers, skipping
the redundant call on timeout when params->polled has already terminated
the channel.
Fixes: 4076e755dbec ("dmatest: convert to dmaengine_unmap_data")
Cc: stable@vger.kernel.org
Signed-off-by: Amita Mishra <amitamishra@google.com>
---
drivers/dma/dmatest.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/dma/dmatest.c b/drivers/dma/dmatest.c
index 2ae3469397f3..950b7a9e7769 100644
--- a/drivers/dma/dmatest.c
+++ b/drivers/dma/dmatest.c
@@ -852,6 +852,8 @@ static int dmatest_func(void *data)
}
if (!done->done) {
+ if (!params->polled)
+ dmaengine_terminate_sync(chan);
result("test timed out", total_tests, src->off, dst->off,
len, 0);
goto error_unmap_continue;
@@ -859,6 +861,7 @@ static int dmatest_func(void *data)
!(dma_has_cap(DMA_COMPLETION_NO_ORDER,
dev->cap_mask) &&
status == DMA_OUT_OF_ORDER)) {
+ dmaengine_terminate_sync(chan);
result(status == DMA_ERROR ?
"completion error status" :
"completion busy status", total_tests, src->off,
base-commit: aebb7274476403040d582c61478df8f4a7bf9cd3
--
2.56.1.703.gf848ed807c-goog
^ permalink raw reply related [flat|nested] 2+ messages in thread* Re: [PATCH] dmaengine: dmatest: terminate channel on timeout or error
2026-11-18 10:08 [PATCH] dmaengine: dmatest: terminate channel on timeout or error Amita Mishra
@ 2026-11-18 10:16 ` sashiko-bot
0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-18 10:16 UTC (permalink / raw)
To: Amita Mishra; +Cc: dmaengine, vkoul, Frank.Li
Thank you for your contribution! Sashiko AI review found no new issues
in this patch, but noticed 1 related pre-existing issue it does not
introduce:
- [High] dmaengine: dmatest: integer underflow in min_odd() [1]
--
commit 15e7b971cd84dccd06d2762018cf300f9236dc53
Author: Amita Mishra <amitamishra@google.com>
dmaengine: dmatest: terminate channel on timeout or error
This commit calls dmaengine_terminate_sync() before unmapping DMA buffers
when a transfer times out or completes with an error in dmatest_func().
> diff --git a/drivers/dma/dmatest.c b/drivers/dma/dmatest.c
> index 2ae3469397f37..950b7a9e77695 100644
> --- a/drivers/dma/dmatest.c
> +++ b/drivers/dma/dmatest.c
> @@ -852,6 +852,8 @@ static int dmatest_func(void *data)
> }
>
> if (!done->done) {
> + if (!params->polled)
> + dmaengine_terminate_sync(chan);
> result("test timed out", total_tests, src->off, dst->off,
> len, 0);
> goto error_unmap_continue;
> @@ -859,6 +861,7 @@ static int dmatest_func(void *data)
> !(dma_has_cap(DMA_COMPLETION_NO_ORDER,
> dev->cap_mask) &&
> status == DMA_OUT_OF_ORDER)) {
> + dmaengine_terminate_sync(chan);
> result(status == DMA_ERROR ?
> "completion error status" :
> "completion busy status", total_tests, src->off,
[Severity: High]
This problem wasn't introduced by this patch, but can min_odd() underflow
when computing src->cnt earlier in dmatest_func() [1]?
drivers/dma/dmatest.c:min_odd() {
unsigned int val = min(x, y);
return val % 2 ? val : val - 1;
}
When a DMA device reports max_xor or max_pq as 0 (or xor_sources /
pq_sources is 0), val is 0 and min_odd() returns UINT_MAX. That bypasses
the (src->cnt + dst->cnt) >= 255 check by wrapping around, after which
dmatest_alloc_test_data() passes d->cnt + 1 == 0 to kcalloc() and
dereferences ZERO_SIZE_PTR when writing d->raw[0].
[1] https://sashiko.dev/#/bug/linux-ca967586-4fc1-47a3-8616-4a9d21fe7455
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261118100839.3090172-1-amitamishra@google.com?part=1
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-11-18 10:17 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-18 10:08 [PATCH] dmaengine: dmatest: terminate channel on timeout or error Amita Mishra
2026-11-18 10:16 ` sashiko-bot
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.