[PATCH] drm/i915: Actually make per-fd contexts useful

[PATCH] drm/i915: Actually make per-fd contexts useful

[Daniel Vetter]

This goes back to when there was just one global context, when
restoring kinda didn't make sense.

Also with execlist we already have these semantics, so better to be a
bit more consistent.

Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_context.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 8603bf48d3ee..e20cfa1ac88c 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -633,7 +633,7 @@ static int do_switch(struct intel_engine_cs *ring,
 			goto unpin_out;
 	}
 
-	if (!to->legacy_hw_ctx.initialized || i915_gem_context_is_default(to))
+	if (!to->legacy_hw_ctx.initialized)
 		hw_flags |= MI_RESTORE_INHIBIT;
 
 	ret = mi_set_context(ring, to, hw_flags);
-- 
2.1.4

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915: Actually make per-fd contexts useful

[Chris Wilson]

On Wed, Jan 21, 2015 at 12:11:11PM +0100, Daniel Vetter wrote:
> This goes back to when there was just one global context, when
> restoring kinda didn't make sense.
> 
> Also with execlist we already have these semantics, so better to be a
> bit more consistent.

Where are the benchmark results? The context switch has angered QA
before...

Do we even have an igt context switch microbenchmark?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915: Actually make per-fd contexts useful

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> On Wed, Jan 21, 2015 at 12:11:11PM +0100, Daniel Vetter wrote:
>> This goes back to when there was just one global context, when
>> restoring kinda didn't make sense.
>> 
>> Also with execlist we already have these semantics, so better to be a
>> bit more consistent.
>
> Where are the benchmark results? The context switch has angered QA
> before...
>

What if we isolate fds and context by default but then let clients
choose, by using per-context parameter, if they are ok with leftover
state. From other context owned by them, or optionally even from
other clients.

- Mika

> Do we even have an igt context switch microbenchmark?
> -Chris
>
> -- 
> Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915: Actually make per-fd contexts useful

[Dave Gordon]

On 21/01/15 13:13, Mika Kuoppala wrote:
> Chris Wilson <chris@chris-wilson.co.uk> writes:
> 
>> On Wed, Jan 21, 2015 at 12:11:11PM +0100, Daniel Vetter wrote:
>>> This goes back to when there was just one global context, when
>>> restoring kinda didn't make sense.
>>>
>>> Also with execlist we already have these semantics, so better to be a
>>> bit more consistent.
>>
>> Where are the benchmark results? The context switch has angered QA
>> before...
>>
> 
> What if we isolate fds and context by default but then let clients
> choose, by using per-context parameter, if they are ok with leftover
> state. From other context owned by them, or optionally even from
> other clients.
> 
> - Mika

And you'd also want to ask them whether they mind /their/ data leaking
out into /other/ clients' contexts :-!

.Dave.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx