Vi5 error： Unable to handle kernel NULL pointer dereference at virtual address 00000000

[JerryChang]

hello future.wang,

please test with single camera only for confirmation. i.e. num=1

[future.wang]

Hi Jerry,

The single camera test did not reappear for 18 hours, but there is no guarantee whether it will reoccur for a longer period of time because our product actually works in 7x24 hours without shutting down.

We hope to find the root cause and resolve it.
Any suggestions?

Thanks!

[JerryChang]

hello future.wang,

it might be race condition,
could you please give it a try to apply these two kernel patches for verification.
for instance,
867cd8c.diff (2.9 KB)
45c5511.diff (3.5 KB)

[future.wang]

Hi Jerry,

patch 867cd8c.diff has been applied to the code.
Next step,
I will apply patch 45c5511.diff to the code and continued testing.

Thanks a lot !

[future.wang]

Hi Jerry,

After applying the following two patches,:
867cd8c.diff (2.9 KB)
45c5511.diff (3.5 KB)
kernel panic caused a system restart problem, which did not recur after a week of testing.

However, the memory leak problem has not been improved, so we added the following patch,
Memory leaks are 60% slower and there are other leak points!!!

81 --- sources.orig/kernel/nvidia/drivers/media/platform/tegra/camera/fusa-capture/capture-vi.c
182 +++ sources/kernel/nvidia/drivers/media/platform/tegra/camera/fusa-capture/capture-vi.c
183 @@ -986,7 +986,10 @@ int vi_capture_release(
184             capture->requests_memoryinfo, capture->requests_memoryinfo_iova);
185         capture->requests_memoryinfo = NULL;
186     }
187 -
188 +   if (capture->unpins_list) {
189 +       vfree(capture->unpins_list);
190 +       capture->unpins_list = NULL;
191 +   }
192     ret = tegra_capture_ivc_unregister_capture_cb(capture->channel_id);
193     if (ret < 0 && err == 0) {
194         dev_err(chan->dev,  

Although the leak rate has slowed down, it still has not completely solved the problem.
Ask developers for help

1. Review whether our patch will affect normal functional logic
2. Find other leak points

Thanks!

[JerryChang]

hello future.wang,

actually, we have VI-5 memory leak fix (which to release capture requests and embedded data buffers) is now under code-review.
for example, e507583.diff (2.6 KB)

please kindly give it another try and sharing the test results.