[PATCH v2] gpu: host1x: Cancel the fence when a syncpoint wait is interrupted

From: Patrick Fischer via B4 Relay

Date: Sun Nov 22 2026 - 04:12:51 EST


From: Patrick Fischer <admin@xxxxxxxxxxxxxxxxxx>

host1x_syncpt_wait() creates its fence without the internal 30 second
timeout and cancels it only when dma_fence_wait_timeout() returns 0.
If the wait is interrupted, dma_fence_wait_timeout() returns
-ERESTARTSYS and the fence is left on the syncpoint's fence list,
holding the reference that host1x_syncpt_fence_enable_signaling() took
for the interrupt path. It is freed only once the syncpoint reaches the
threshold, which may never happen.

Callers that retry on -ERESTARTSYS allocate a new fence each time.
tegra20_channel_capture_frame() retries in a loop without delay, and
while the syncpoint does not move, for example with a camera that
delivers no frames, every round leaks one fence until the system runs
out of memory. Svyatoslav Ryhel reported this as an out-of-memory crash
in the VI capture thread; the allocations in his report come from
host1x_fence_create(), called by host1x_syncpt_wait() from
tegra20_chan_capture_kthread_start(). The DRM syncpoint wait IOCTLs end
up in host1x_syncpt_wait() as well, so a process that is signalled while
waiting leaks a fence the same way.

Cancel the fence whenever the wait did not succeed. host1x_fence_cancel()
is safe if the interrupt path has signalled the fence in the meantime,
as do_fence_timeout() then returns early.

host1x_fence_cancel() can also run before signalling was enabled on the
fence. fence->list is only initialized in host1x_intr_add_fence_locked(),
so host1x_intr_remove_fence() would then see a zeroed list head, call
list_del_init() on it and report the fence as removed, after which
do_fence_timeout() drops an interrupt path reference that was never
taken. Initialize the list head in host1x_fence_create() instead.
host1x_syncpt_wait() does not hit this today, as
dma_fence_wait_timeout() enables signalling before it checks for a
pending signal, but the cancel path should not depend on that.

The leak was found with the help of an LLM assistant reading the host1x
and dma-fence code against the call chain in the report.

Fixes: d5179020f5ce ("gpu: host1x: External timeout/cancellation for fences")
Cc: stable@xxxxxxxxxxxxxxx
Reported-by: Svyatoslav Ryhel <clamor95@xxxxxxxxx>
Signed-off-by: Patrick Fischer <admin@xxxxxxxxxxxxxxxxxx>
---
Svyatoslav reported the out-of-memory crash, and he confirmed off-list
that v1 stops the leak on his LG P895 (Tegra30). v2 changes the code,
so that confirmation does not carry over to it.

I have not run v2 on hardware. The Tegra Note 7 I work on only
captures with Svyatoslav's out-of-tree VI/CSI patches, and I have not
tried to trigger an interrupted wait there. Build-tested for arm and
arm64. The DRM IOCTL case in the changelog comes from reading the code;
I have not tried it. The same goes for cancelling a fence before
signalling was enabled, which no caller does today.

The retry loop in tegra20_channel_capture_frame() still spins without
delay as long as the wait keeps returning -ERESTARTSYS. With this patch
that costs CPU time instead of memory; how the VI driver should handle
an interrupted wait is a separate question.

Tool use (see Documentation/process/generated-content.rst): an LLM
assistant (Claude, in a chat) found the leak by reading
host1x_syncpt_wait() and the fence code against the call chain in the
OOM report. Claude Code then checked the finding against the code and
wrote the change, the changelog and this note from my notes and
instructions. For v2, the chat checked Mikko's point against the code,
and Claude Code applied the change and extended the changelog. I
reviewed the change.
---
Changes in v2:
- Initialize fence->list in host1x_fence_create() instead of
host1x_intr_add_fence_locked(), so that cancelling a fence before
signalling was enabled is safe (Mikko Perttunen)
- Link to v1: https://patch.msgid.link/20261113-upstream-host1x-fence-cancel-v1-1-89f453504222@xxxxxxxxxxxxxxxxxx
---
drivers/gpu/host1x/fence.c | 1 +
drivers/gpu/host1x/intr.c | 2 --
drivers/gpu/host1x/syncpt.c | 2 +-
3 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/host1x/fence.c b/drivers/gpu/host1x/fence.c
index 1acf579c562b..f77a391b9195 100644
--- a/drivers/gpu/host1x/fence.c
+++ b/drivers/gpu/host1x/fence.c
@@ -134,6 +134,7 @@ struct dma_fence *host1x_fence_create(struct host1x_syncpt *sp, u32 threshold,
fence->sp = sp;
fence->threshold = threshold;
fence->timeout = timeout;
+ INIT_LIST_HEAD(&fence->list);

dma_fence_init(&fence->base, &host1x_syncpt_fence_ops, &sp->fences.lock,
dma_fence_context_alloc(1), 0);
diff --git a/drivers/gpu/host1x/intr.c b/drivers/gpu/host1x/intr.c
index 30f3dff4487b..380f679b4d7f 100644
--- a/drivers/gpu/host1x/intr.c
+++ b/drivers/gpu/host1x/intr.c
@@ -47,8 +47,6 @@ void host1x_intr_add_fence_locked(struct host1x *host, struct host1x_syncpt_fenc
{
struct host1x_fence_list *fence_list = &fence->sp->fences;

- INIT_LIST_HEAD(&fence->list);
-
host1x_intr_add_fence_to_list(fence_list, fence);
host1x_intr_update_hw_state(host, fence->sp);
}
diff --git a/drivers/gpu/host1x/syncpt.c b/drivers/gpu/host1x/syncpt.c
index fe74ff6fc6f2..f9c871f218f3 100644
--- a/drivers/gpu/host1x/syncpt.c
+++ b/drivers/gpu/host1x/syncpt.c
@@ -247,7 +247,7 @@ int host1x_syncpt_wait(struct host1x_syncpt *sp, u32 thresh, long timeout,
return PTR_ERR(fence);

wait_err = dma_fence_wait_timeout(fence, true, timeout);
- if (wait_err == 0)
+ if (wait_err <= 0)
host1x_fence_cancel(fence);
dma_fence_put(fence);


---
base-commit: 9d66f26fefacebfeed15de31c1123451a17ff119
change-id: 20261113-upstream-host1x-fence-cancel-0b32753150d6

Best regards,
--
Patrick Fischer <admin@xxxxxxxxxxxxxxxxxx>