atomisp_vidioc_default() calls atomisp_exp_id_unlock(),
atomisp_exp_id_capture(), and atomisp_inject_a_fake_event() without
holding isp->mutex. However, all three functions have
lockdep_assert_held(&isp->mutex) indicating callers must hold this lock.

Other ioctl handlers in the same driver (e.g., atomisp_start_streaming)
properly acquire the mutex before operating on ISP state.

Add mutex_lock()/mutex_unlock() around these three ioctl cases to fix the
missing lock protection.

Signed-off-by: Ziyi Guo <n7l8m4@u.northwestern.edu>
---
 drivers/staging/media/atomisp/pci/atomisp_ioctl.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/staging/media/atomisp/pci/atomisp_ioctl.c b/drivers/staging/media/atomisp/pci/atomisp_ioctl.c
index bb8b2f2213b0..ab13630c8149 100644
--- a/drivers/staging/media/atomisp/pci/atomisp_ioctl.c
+++ b/drivers/staging/media/atomisp/pci/atomisp_ioctl.c
@@ -1521,10 +1521,14 @@ static long atomisp_vidioc_default(struct file *file, void *fh,
 		break;
 
 	case ATOMISP_IOC_EXP_ID_UNLOCK:
+		mutex_lock(&asd->isp->mutex);
 		err = atomisp_exp_id_unlock(asd, arg);
+		mutex_unlock(&asd->isp->mutex);
 		break;
 	case ATOMISP_IOC_EXP_ID_CAPTURE:
+		mutex_lock(&asd->isp->mutex);
 		err = atomisp_exp_id_capture(asd, arg);
+		mutex_unlock(&asd->isp->mutex);
 		break;
 	case ATOMISP_IOC_S_ENABLE_DZ_CAPT_PIPE:
 		err = atomisp_enable_dz_capt_pipe(asd, arg);
@@ -1537,7 +1541,9 @@ static long atomisp_vidioc_default(struct file *file, void *fh,
 		err = atomisp_formats(asd, 1, arg);
 		break;
 	case ATOMISP_IOC_INJECT_A_FAKE_EVENT:
+		mutex_lock(&asd->isp->mutex);
 		err = atomisp_inject_a_fake_event(asd, arg);
+		mutex_unlock(&asd->isp->mutex);
 		break;
 	case ATOMISP_IOC_S_ARRAY_RESOLUTION:
 		err = atomisp_set_array_res(asd, arg);
-- 
2.34.1

On Thu, Feb 5, 2026 at 6:37 AM Ziyi Guo <n7l8m4@u.northwestern.edu> wrote:
>
> atomisp_vidioc_default() calls atomisp_exp_id_unlock(),
> atomisp_exp_id_capture(), and atomisp_inject_a_fake_event() without
> holding isp->mutex. However, all three functions have
> lockdep_assert_held(&isp->mutex) indicating callers must hold this lock.

Did you get a splat on real HW?

> Other ioctl handlers in the same driver (e.g., atomisp_start_streaming)
> properly acquire the mutex before operating on ISP state.
>
> Add mutex_lock()/mutex_unlock() around these three ioctl cases to fix the
> missing lock protection.

How did you find the issue?

-- 
With Best Regards,
Andy Shevchenko