[Intel-gfx] Question regarding gem_alive test case in intel-gpu-tools-1.7 suite

[Chris Wilson]

On Thu, Apr 30, 2015 at 06:19:06PM +0000, Meng, David wrote:
>    Hi Team:
> 
>    Greetings.
> 
>     
> 
>    I am new to send a question to this email list.  I am working on GPU
>    driver for VMware ESXi kernel running on BDW system. The driver is based
>    on the OTC graphics driver which is released with Linux 3.19.3.
> 
>     
> 
>    When I tested the driver using the test case gem_alive from
>    intel-gpu-tools -1.7, the result is always fail. The kernel driver
>    complains that the object handle is 0.  By checking the gem_alive.c code,
>    it is found the test case dose not create any gem object but tries to  use
>     DRM_IOCTL_I915_GEM_SW_FINISH ioctl call to finish a object.  As a result,
>    it returns –ENOENT error.
>     
> 
>    I wonder is there any special kernel driver feature to be tested by this
>    case?  Or we expect it is always fail?

The clue is in the various errno that is checked by gem_alive, and
consider how they can arise from various scenarios.

In short, the test can differentiate between a kernel deadlock, a GPU
hang and a functioning interface. It answers the question of whether
"GEM is alive".
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre