array index is out of range - Crashed with Family_100_bubblesort_flag #897
Comments
|
Concerning the null pointer, below are log messages based on the instrumentation I mentioned earlier. I didn't check c->type so that is interesting, it appears the array index is also corrupted. |
|
Complete Crash BT: |
|
Thanks for this report @malaya-parida. Would you be able to share with us the generated shader that caused this issue by attaching it to this report? Also, would you be willing to try build the latest GraphicsFuzz from source and try to reproduce the issue again? If you are able to reproduce it, could you please send us the exact commands you used to generate this shader? Also, could you try running glsl-reduce on the shader to try to reduce the crash? |
|
Hi Alastair, Thanq for the response. |
|
Hi Alastair, We are facing infrastructure issue to build from latest GraphicsFuzz src. Could you help to share apk built on top of latest src. |
|
GraphicsFuzz webUI shows the status of the test as CRASH in the 6 six cases for the entire suite of 100_families. below is the sample from one family (family_100_bubblesort_flag)
Is this behavior correct, as the values generated in the test content (sample shaders) are expected to crash..? Can you confirm this ? |
|
Thanks for running GraphicsFuzz @malaya-parida GraphicsFuzz generates shaders to find bugs in graphics drivers. Every graphics driver might behave differently. The "CRASH" here indicates either:
It is probably a bug in your graphics driver. This is a good thing, as it means this bug can be reported (e.g. to Intel) and the bug can be fixed. Can I ask: what are you trying to achieve? Are you trying out GraphicsFuzz for fun? Are you hoping to find bugs in a graphics driver to report/fix? Also note: our current priority is testing Vulkan drivers. And our recommended way to try this is gfauto. This will use GraphicsFuzz behind the scenes to test your Vulkan driver. |
|
We are trying to test Graphics Fuzz on Intel System & found crash in i965-dri.so. We saw array index out of range in MESA & put code to return whenever array index out of range. But now crash in MESA not seen but we see rendering issue & hence the test is failed in webUI (As seen in above pic). Practically we can't handle array index out of range & can't proceed gracefully like rendering etc. In this case, does crash expected ? |
|
With attached original shaders, we saw crash in application.
Can GraphicsFuzz check the buggy shaders ? |
|
You will need to perform a "no image reduction" on the CRASH as described in the walkthrough. For the "Error string to look for in log", try using something like: ir_constant::ir_constant |
|
We have fixed in MESA to handle array index out of range coming from Shaders (https://gitlab.freedesktop.org/mesa/mesa/issues/2604). But with this patch there is no crash in MESA but crash now seen in GraphicsFuzz application. Can we not review the shaders from GraphicsFuzz to handle the crash in application. |
|
@malaya-parida It looks like you are not using release 1.3 of GraphicsFuzz but perhaps an older release. First the shader that was called "bubblesort_flag" is called "stable_bubblesort_flag" in GraphicsFuzz 1.3. Second, the GraphicsFuzz 1.3 test generator ensures that this macro is present in every generated shader: I don't see that macro in the variant shaders you filed in the Mesa bug report. Can you please check that you're definitely using the 1.3 release, which is here: https://github.com/google/graphicsfuzz/releases/tag/v1.3 Thanks! |
|
Closing this as there's not been discussion for a year, but please re-open or open a new issue if you'd like to discuss more. |
While we test GraphicsFuzz (v1.3) & some specific test (e.g. Family_100_bubblesort_flag), we saw random crashes (2/10 times).
Right before Crash, we see array index is out of range (//DEBUG TEST: i=4294947814)
The text was updated successfully, but these errors were encountered: