[PATCH i-g-t] tests/core_hotunplug: Take care of closing fence before failing

[Chris Wilson]

Quoting Janusz Krzysztofik (2020-10-13 20:16:37)
> The test was designed to keep track of open device file descriptors
> for safe driver unbind on recovery from a failed subtest.  In that
> context, fences introduced by commit 1fbd127bd4e1 ("core_hotplug:
> Teach the healthcheck how to check execution status") can affect device
> recovery as much as an open device file if not closed before unbind.
> 
> Refactor local_i915_healthcheck() so it takes care of closing fences
> before failing the health check.  While avoiding use of igt_fail() and
> friends, report actual source of the failure via igt_warn_on_f().
> 
> Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik at linux.intel.com>
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> ---
>  tests/core_hotunplug.c | 42 +++++++++++++++++++++++++++++++-----------
>  1 file changed, 31 insertions(+), 11 deletions(-)
> 
> diff --git a/tests/core_hotunplug.c b/tests/core_hotunplug.c
> index cdc07c85d..806e5b8a0 100644
> --- a/tests/core_hotunplug.c
> +++ b/tests/core_hotunplug.c
> @@ -233,9 +233,9 @@ static int merge_fences(int old, int new)
>                 return new;
>  
>         merge = sync_fence_merge(old, new);
> -       igt_assert(merge != -1);
> -       close(old);
> -       close(new);
> +       /* Assume fence close errors don't affect device close status */
> +       igt_ignore_warn(local_close(old, "old fence close failed"));
> +       igt_ignore_warn(local_close(new, "new fence close failed"));
>  
>         return merge;
>  }
> @@ -249,29 +249,49 @@ static int local_i915_healthcheck(int i915, const char *prefix)
>                 .buffer_count = 1,
>         };
>         const struct intel_execution_engine2 *engine;
> -       int fence = -1;
> +       int fence = -1, err = 0, status = 1;
>  
>         local_debug("%s%s\n", prefix, "running i915 GPU healthcheck");
> -       if (local_i915_is_wedged(i915))
> +       if (igt_warn_on_f(local_i915_is_wedged(i915), "GPU found wedged\n"))
>                 return -EIO;
>  
> +       /* Assume gem_create()/gem_write() failures are unrecoverable */
>         obj.handle = gem_create(i915, 4096);
>         gem_write(i915, obj.handle, 0, &bbe, sizeof(bbe));
>  
> +       /* As soon as a fence is open, don't fail before closing it */
>         __for_each_physical_engine(i915, engine) {
>                 execbuf.flags = engine->flags | I915_EXEC_FENCE_OUT;
> -               gem_execbuf_wr(i915, &execbuf);
> +               err = __gem_execbuf_wr(i915, &execbuf);
> +               if (igt_warn_on_f(err < 0, "__gem_execbuf_wr() failed\n"))
> +                       break;
>  
>                 fence = merge_fences(fence, execbuf.rsvd2 >> 32);
> +               if (igt_warn_on_f(fence < 0, "merge_fences() failed\n")) {
> +                       err = fence;
> +                       break;
> +               }
> +       }
> +       if (fence >= 0) {
> +               status = sync_fence_wait(fence, -1);
> +               if (igt_warn_on_f(status < 0, "sync_fence_wait() failed\n"))
> +                       err = status;
> +               if (!err)
> +                       status = sync_fence_status(fence);
> +
> +               /* Assume fence close errors don't affect device close status */
> +               igt_ignore_warn(local_close(fence, "fence close failed"));
>         }
> -       igt_assert(fence != -1);
> +
> +       /* Assume gem_close() failure is unrecoverable */
>         gem_close(i915, obj.handle);
>  
> -       igt_assert_eq(sync_fence_wait(fence, -1), 0);
> -       igt_assert_eq(sync_fence_status(fence), 1);
> -       close(fence);
> +       if (err < 0)
> +               return err;
> +       if (igt_warn_on_f(status != 1, "sync_fence_status() failed\n"))
> +               return status < 0 ? status : -EIO;

You've lost the importance of reporting the error value, and
misrepresenting the meaning of status == 0.
-Chris