* [PATCH] accel/qaic: drain the control workqueue before the device is freed
@ 2026-11-18 7:24 Yaohan Li
2026-11-18 7:41 ` sashiko-bot
0 siblings, 1 reply; 2+ messages in thread
From: Yaohan Li @ 2026-11-18 7:24 UTC (permalink / raw)
To: jeff.hugo, ogabbay
Cc: carl.vanderlip, linux-arm-msm, dri-devel, linux-kernel,
jacek.lawrynowicz, quic_pkanojiy, stanislaw.gruszka
qaic_pci_remove() tears the MHI controller down and destroys the DRM
device, but nothing drains qdev->cntl_wq. qdev itself is allocated with
devm_kzalloc_flex() and is freed as soon as the PCI device is unbound,
while cntl_wq is a DRM-managed resource that survives until the last
open file drops its reference to the DRM device.
resp_work is queued from qaic_mhi_dl_xfer_cb() and resp_worker()
dereferences qdev, so a work item that is still queued, or still
running, when the driver instance goes away touches freed memory.
mhi_power_down() moves the controller to the shutdown state, disables
the interrupt and flushes the state worker that destroys the channel
devices, and mhi_unregister_controller() removes the controller, so no
data event can reach qaic_mhi_dl_xfer_cb() once
qaic_mhi_free_controller() has returned. Flushing the workqueue after
it therefore covers every work item that can still be queued.
Fixes: c501ca23a6a3 ("accel/qaic: Add uapi and core driver file")
Signed-off-by: Yaohan Li <shunsheng1218@gmail.com>
---
Prepared against v7.4-rc1-1969-g666539eecdf4.
AI assistance: the analysis, the change and its review notes were
prepared with Claude (Anthropic) as a coding assistant. The author
reviewed the reasoning and takes responsibility for the content.
Testing: compile tested only (arm64, gcc 11.4). No QAIC hardware was
available, so the teardown race was not run.
drivers/accel/qaic/qaic_drv.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/accel/qaic/qaic_drv.c b/drivers/accel/qaic/qaic_drv.c
index 90c0d31e414a..c4c3f450e67b 100644
--- a/drivers/accel/qaic/qaic_drv.c
+++ b/drivers/accel/qaic/qaic_drv.c
@@ -625,6 +625,15 @@ static void qaic_pci_remove(struct pci_dev *pdev)
qaic_dev_reset_clean_local_state(qdev);
qaic_mhi_free_controller(qdev->mhi_cntrl, link_up);
+
+ /*
+ * The control channel is gone, so nothing can queue resp_work any
+ * more. Drain it before qdev is freed: the workqueue belongs to the
+ * DRM device and may outlive this driver instance, and resp_worker()
+ * dereferences qdev.
+ */
+ flush_workqueue(qdev->cntl_wq);
+
qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 2+ messages in thread* Re: [PATCH] accel/qaic: drain the control workqueue before the device is freed
2026-11-18 7:24 [PATCH] accel/qaic: drain the control workqueue before the device is freed Yaohan Li
@ 2026-11-18 7:41 ` sashiko-bot
0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-18 7:41 UTC (permalink / raw)
To: Yaohan Li; +Cc: dri-devel
Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Low] accel/qaic: Fixes trailer cites c501ca23a6a3 instead of
22ec9a3b73dd
--
commit 4e36def68e52df8a1a1cb9f08c76e856b9e13fb0
Author: Yaohan Li <shunsheng1218@gmail.com>
> accel/qaic: drain the control workqueue before the device is freed
>
> qaic_pci_remove() tears the MHI controller down and destroys the DRM
> device, but nothing drains qdev->cntl_wq. qdev itself is allocated with
> devm_kzalloc_flex() and is freed as soon as the PCI device is unbound,
> while cntl_wq is a DRM-managed resource that survives until the last
> open file drops its reference to the DRM device.
>
> resp_work is queued from qaic_mhi_dl_xfer_cb() and resp_worker()
> dereferences qdev, so a work item that is still queued, or still
> running, when the driver instance goes away touches freed memory.
>
> mhi_power_down() moves the controller to the shutdown state, disables
> the interrupt and flushes the state worker that destroys the channel
> devices, and mhi_unregister_controller() removes the controller, so no
> data event can reach qaic_mhi_dl_xfer_cb() once
> qaic_mhi_free_controller() has returned. Flushing the workqueue after
> it therefore covers every work item that can still be queued.
>
> Fixes: c501ca23a6a3 ("accel/qaic: Add uapi and core driver file")
> Signed-off-by: Yaohan Li <shunsheng1218@gmail.com>
[Severity: Low]
In c501ca23a6a3 ("accel/qaic: Add uapi and core driver file"),
qaic_pci_remove() called cleanup_qdev(), which destroyed qdev->cntl_wq
synchronously via destroy_workqueue() before qdev was freed. The
lifetime mismatch was introduced in 22ec9a3b73dd ("accel/qaic: Leverage
DRM managed APIs to release resources"), which switched qdev->cntl_wq to
qaicm_wq_init() and dropped cleanup_qdev() from qaic_pci_remove().
Should the Fixes tag reference that commit instead:
Fixes: 22ec9a3b73dd ("accel/qaic: Leverage DRM managed APIs to release resources")
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261118072448.1259909-1-shunsheng1218@gmail.com?part=1
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-11-18 7:41 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-18 7:24 [PATCH] accel/qaic: drain the control workqueue before the device is freed Yaohan Li
2026-11-18 7:41 ` 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.