[Intel-gfx] [PATCH] drm/i915/perf: Use GTT when saving/restoring engine GPR

[Chris Wilson]

Quoting Umesh Nerlige Ramappa (2020-07-09 20:58:37)
> MI_STORE_REGISTER_MEM and MI_LOAD_REGISTER_MEM need to know which
> translation to use when saving restoring the engine general purpose
> registers to and from the GT scratch. Since GT scratch is mapped to
> ggtt, we need to set an additional bit in the command to use GTT.
> 
> Fixes: daed3e44396d17 ("drm/i915/perf: implement active wait for noa configurations")
> Suggested-by: Prathap Kumar Valsan <prathap.kumar.valsan at intel.com>
> Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_perf.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> index de69d430b1ed..c6f6370283cf 100644
> --- a/drivers/gpu/drm/i915/i915_perf.c
> +++ b/drivers/gpu/drm/i915/i915_perf.c
> @@ -1592,6 +1592,7 @@ static u32 *save_restore_register(struct i915_perf_stream *stream, u32 *cs,
>         u32 d;
>  
>         cmd = save ? MI_STORE_REGISTER_MEM : MI_LOAD_REGISTER_MEM;
> +       cmd |= MI_SRM_LRM_GLOBAL_GTT;

Indeed.

For bonus points, please write a selftest to verify that the user's GPR
are restored. Something as simple as live_noa_delay, but instead of
measuring the time; submit a request to write telltales into the GPR,
submit a request to run noa_wait; then readback the telltales from a
final request.

Now since the noa_wait is being run from the GGTT, the address space
selector should be reading from the GGTT. So I am actually curious as to
whether we have a bug or not.
-Chris