VC4 GPU down

[yuanzhaowei]

Hi Eric, I got a problem which casue vc4 GPU down, error log is shown below: [ 2647.764746] [drm] Resetting GPU. [ 2648.788790] [drm] Resetting GPU. [ 2649.748808] [drm] Resetting GPU. [ 2650.772789] [drm] Resetting GPU. [ 2651.796793] [drm] Resetting GPU. ......

This is an opengl-es-cts test case names "ES2-CTS.gtf.GL2FixedTests.divide_by_zero.divide_by_zero" who is trying to set several kinds of scenrio to make GPU execute "divide by zero" caculation, include a calling to glViewport(0, 0, 0, 0) which cause GPU down.

If I modify arguments width and height to a significative vaule, the problem doesn't occur

But which strange is: 1. I generated an another simple test program with glViewport(0, 0, 0, 0), the problem doesn't occur 2. In the test case "ES2-CTS.gtf.GL2FixedTests.divide_by_zero.divide_by_zero" , there are more than one sub test cases to test if GPU can handle the situation that divided by 0, they all passed before calling to glViewport(0, 0, 0, 0) So, it seems VC4 driver and GPU have ability to handle that, but I don't understand why GPU is down

Attachment is the complete command line dumped by vc4_dump_cl(), Could you please help? Thanks CL.txt

[anholt]

I would recommend hacking out subtests from the CTS test until you get down to the minimal testcase that will hang the GPU. Then, since you've got weird viewport stuff involved, I'd pay particular attention to the arguments being passed to the kernel in submit.min/max_x/y_tile and what vc4_render_cl.c does with those.