Arecord tampering causes DSP crashes #489
Comments
|
@singalsu is this related to the crash that you are seeing after 46 iterations of dmic capture ? |
|
@cujomalainey |
|
@cujomalainey |
|
@singalsu Looks like another vector that may help you identifying root cause. @ZhendanYang can you try this test with SSP only to confirm it's DMIC issue. |
|
@lgirdwood I don't thinks its DMIC specific, I was able to produce something similar by running a similar script and tampering with it but instead using speaker-test. Here is the DSP crash from that. |
|
@cujomalainey ok, must be something in the pipeline or DMA related. |
|
On up2, tests are blocked by #443 . Please @keqiaozhang if it's a GLK issue. |
|
@cujomalainey @lgirdwood I am able to repro DSP crash easily with the arecord loop, however I get 'out of memory', not the original 'can't enter idle'. With the #495 applied, there is no repro on my board. No other issues like the 'idle' one appeared. |
|
I cannot reproduce this currently at master. I'm not seeing any errors in the trace buffer. These is the only errors I'm seeing in the kernel logs. |
|
@cujomalainey A PR went in today from @mmaka1 that fixed a leak in the DMA. |
|
@keqiaozhang can you verify this on Yorp? I think Zhendan's validation on UP2 can still be blocked by #443 |
|
@cujomalainey @lgirdwood |
|
@keqiaozhang I checked out the same location and ran it 10 times with tampering and it has passed every time. |
|
@stevyan @mengdonglin shouldn't this be re-opened if it's listed as related with a PM bug now ? |
|
@lgirdwood |
|
@keqiaozhang thanks, I think I copy pasted from the daily test but it looks like kernel bug with same number as this. Sorry about this. |
I am able to cause the DSP to crash pretty easily with a simple script.
while arecord -D hw:0,99 -f S32_LE -d 1; do :; doneThen by tampering with loop by using C-z (to suspend) or C-c (to interrupt) you can crash the dsp pretty easily.
Example of DSP panic by interrupting loop
a run where I simply suspend the stream (no resume)
Another run where I suspend then resume arecord (note, did not check if crash happened before resume)
the kernel output
These results are all from V0.5 drop, will continue testing as I move around the tree
The text was updated successfully, but these errors were encountered: