drm/i915: add cherryview specfic forcewake in execlists_elsp_write
Subsystems (2)
PATCH 1/1 drm/i915: add cherryview specfic forcewake in execlists_elsp_write
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into __vlv_force_wake_get.
--- drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/intel_lrc.c +25 -4
@@ -300,8 +300,18 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,* Instead, we do the runtime_pm_get/put when creating/destroying requests. */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_RENDER ); +if (dev_priv->uncore.fw_mediacount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_MEDIA ); + }else { +if (dev_priv->uncore.forcewake_count++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_ALL ); + }spin_unlock_irqrestore (&dev_priv->uncore.lock, flags);I915_WRITE (RING_ELSP (ring), desc[1 ]);
@@ -315,8 +325,19 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,/* Release Force Wakeup (see the big comment above). */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (--dev_priv->uncore.fw_rendercount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_RENDER ); +if (--dev_priv->uncore.fw_mediacount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_MEDIA ); + }else { +if (--dev_priv->uncore.forcewake_count ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_ALL ); + } +spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); }
Signature · 1 line
1.9.1
Quoted 19 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)drivers/gpu/drm/i915/intel_lrc.c +0
-0@@ -300,8 +300,18 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,* Instead, we do the runtime_pm_get/put when creating/destroying requests. */
drivers/gpu/drm/i915/intel_lrc.c static void execlists_elsp_write(struct intel_engine_cs *ring, · line 309
Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +7
-2spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_RENDER ); +if (dev_priv->uncore.fw_mediacount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_MEDIA );
This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 32 lines
drivers/gpu/drm/i915/intel_lrc.c +18
-2+ }else { +if (dev_priv->uncore.forcewake_count++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_ALL ); + }spin_unlock_irqrestore (&dev_priv->uncore.lock, flags);I915_WRITE (RING_ELSP (ring), desc[1 ]);@@ -315,8 +325,19 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,/* Release Force Wakeup (see the big comment above). */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (--dev_priv->uncore.fw_rendercount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_RENDER ); +if (--dev_priv->uncore.fw_mediacount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_MEDIA ); + }else { +if (--dev_priv->uncore.forcewake_count ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_ALL ); + } +spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); }Signature · 1 line
1.9.1
Signature · 2 lines
Ville Syrjälä Intel OTC
Quoted 25 lines · Ville Syrjälä
Quoted 24 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)drivers/gpu/drm/i915/intel_lrc.c +2
-2@@ -300,8 +300,18 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,* Instead, we do the runtime_pm_get/put when creating/destroying requests. */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 )
Quoted 8 lines
drivers/gpu/drm/i915/intel_lrc.c static void execlists_elsp_write(struct intel_engine_cs *ring, · line 309
Quoted 5 lines
drivers/gpu/drm/i915/intel_lrc.c +5
-0+ dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_RENDER ); +if (dev_priv->uncore.fw_mediacount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_MEDIA );This will wake both wells. Is that needed or should we just pick one based on the ring?
Also unlike the comment says runtime_pm_get() can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to call gen6_gt_force_wake_get() here, but maybe that would trigger a might_sleep() warning. the current force wake code duplication (esp. outside intel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outside gen6_gt_force_wake_get()? I never really liked hiding it there anyway.
Quoted 37 lines
Quoted 33 lines
drivers/gpu/drm/i915/intel_lrc.c +18
-2+ }else { +if (dev_priv->uncore.forcewake_count++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_ALL ); + }spin_unlock_irqrestore (&dev_priv->uncore.lock, flags);I915_WRITE (RING_ELSP (ring), desc[1 ]);@@ -315,8 +325,19 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,/* Release Force Wakeup (see the big comment above). */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (--dev_priv->uncore.fw_rendercount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_RENDER ); +if (--dev_priv->uncore.fw_mediacount ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_MEDIA ); + }else { +if (--dev_priv->uncore.forcewake_count ==0 ) + dev_priv->uncore.funcs.force_wake_put (dev_priv, +FORCEWAKE_ALL ); + } +spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); }Signature · 1 line
1.9.1Signature · 2 lines
Ville Syrjälä Intel OTC
Signature · 2 lines
Ville Syrjälä Intel OTC
Quoted 35 lines · Ville Syrjälä
Quoted 25 lines · Ville Syrjälä
Quoted 24 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)drivers/gpu/drm/i915/intel_lrc.c +2
-2@@ -300,8 +300,18 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,* Instead, we do the runtime_pm_get/put when creating/destroying requests. */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 )Quoted 8 lines
drivers/gpu/drm/i915/intel_lrc.c static void execlists_elsp_write(struct intel_engine_cs *ring, · line 309
Quoted 5 lines
drivers/gpu/drm/i915/intel_lrc.c +5
-0+ dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_RENDER ); +if (dev_priv->uncore.fw_mediacount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_MEDIA );This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 8 lines
Also unlike the comment says
runtime_pm_get()can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to callgen6_gt_force_wake_get()here, but maybe that would trigger amight_sleep()warning. the current force wake code duplication (esp. outsideintel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outsidegen6_gt_force_wake_get()? I never really liked hiding it there anyway.
Yeah this is just broken design. And if you look at the other wheel to track outstanding gpu work (requests) then it's not needed at all.
But I'm not sure what's the priority of the "rework execlists to use requests" task is and when (if ever that will happen). Jesse is the arbiter for this stuff anyway, so adding him.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Quoted 42 lines · Daniel Vetter
Quoted 33 lines · Ville Syrjälä
Quoted 24 lines · Ville Syrjälä
Quoted 23 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)drivers/gpu/drm/i915/intel_lrc.c +1
-2@@ -300,8 +300,18 @@ static void execlists_elsp_write(struct intel_engine_cs *ring,* Instead, we do the runtime_pm_get/put when creating/destroying requests. */ spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); +if (IS_CHERRYVIEW (dev_priv->dev)) {Quoted 8 lines
drivers/gpu/drm/i915/intel_lrc.c static void execlists_elsp_write(struct intel_engine_cs *ring, · line 309
Quoted 6 lines
drivers/gpu/drm/i915/intel_lrc.c +6
-0+if (dev_priv->uncore.fw_rendercount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_RENDER ); +if (dev_priv->uncore.fw_mediacount++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_MEDIA );This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 8 lines
Also unlike the comment says
runtime_pm_get()can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to callgen6_gt_force_wake_get()here, but maybe that would trigger amight_sleep()warning. the current force wake code duplication (esp. outsideintel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outsidegen6_gt_force_wake_get()? I never really liked hiding it there anyway.
Quoted 7 lines
Yeah this is just broken design. And if you look at the other wheel to track outstanding gpu work (requests) then it's not needed at all.
But I'm not sure what's the priority of the "rework execlists to use requests" task is and when (if ever that will happen). Jesse is the arbiter for this stuff anyway, so adding him.
-Daniel
hmm , agreed do we have a reworked execlist? The reason why added this, on chv when i enable execlist, due to incorrect forcewake count is causing multiple entries to forcewake_get resulting in "*ERROR*: Timed out waiting for forcewake old ack to clear" "and Hang.
Quoted 62 lines · On Tue, 09 Sep 2014 21:45:08 +0530 Deepak S <[email protected]> wrote:
Quoted 51 lines · Daniel Vetter
Quoted 41 lines · Ville Syrjälä
Quoted 32 lines · Ville Syrjälä
Quoted 30 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29+++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)
(quoted diff) +0
-0
b/drivers/gpu/drm/i915/intel_lrc.cindex bd1b28d..bafd38b 100644drivers/gpu/drm/i915/intel_lrc.c +0
-0@@ -300,8 +300,18 @@ static void execlists_elsp_write(structintel_engine_cs *ring, * Instead, we do the runtime_pm_get/put whencreating/destroying requests. */
spin_lock_irqsave(&dev_priv->uncore.lock, flags);drivers/gpu/drm/i915/intel_lrc.c +0
-2-if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get(dev_priv,
FORCEWAKE_ALL);drivers/gpu/drm/i915/intel_lrc.c +3
-0+if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+Quoted 8 lines
Quoted 6 lines
FORCEWAKE_RENDER);drivers/gpu/drm/i915/intel_lrc.c +2
-0+if (dev_priv->uncore.fw_mediacount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+FORCEWAKE_MEDIA);This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 9 lines
Also unlike the comment says
runtime_pm_get()can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to callgen6_gt_force_wake_get()here, but maybe that would trigger amight_sleep()warning. the current force wake code duplication (esp. outsideintel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outsidegen6_gt_force_wake_get()? I never really liked hiding it there anyway.Quoted 8 lines
Yeah this is just broken design. And if you look at the other wheel to track outstanding gpu work (requests) then it's not needed at all.
But I'm not sure what's the priority of the "rework execlists to use requests" task is and when (if ever that will happen). Jesse is the arbiter for this stuff anyway, so adding him.
-Daniel
Quoted 4 lines
hmm , agreed do we have a reworked execlist? The reason why added this, on chv when i enable execlist, due to incorrect forcewake count is causing multiple entries to
forcewake_getresulting in "*ERROR*:
Timed out waiting for forcewake old ack to clear" "and Hang.
I'm hoping we can get execlists reworked on top of the request/seqno stuff shortly after it lands, but I don't think that's a reason to block this fix, since Chris is still busy fixing up the request changes.
Jesse
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Daniel Vetter
AppliedQuoted 70 lines · Jesse Barnes
Quoted 62 lines · On Tue, 09 Sep 2014 21:45:08 +0530 Deepak S <[email protected]> wrote:
Quoted 51 lines · Daniel Vetter
Quoted 41 lines · Ville Syrjälä
Quoted 32 lines · Ville Syrjälä
Quoted 30 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29+++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)
(quoted diff) +0
-0
b/drivers/gpu/drm/i915/intel_lrc.cindex bd1b28d..bafd38b 100644drivers/gpu/drm/i915/intel_lrc.c +0
-0@@ -300,8 +300,18 @@ static void execlists_elsp_write(structintel_engine_cs *ring, * Instead, we do the runtime_pm_get/put whencreating/destroying requests. */
spin_lock_irqsave(&dev_priv->uncore.lock, flags);drivers/gpu/drm/i915/intel_lrc.c +0
-2-if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get(dev_priv,
FORCEWAKE_ALL);drivers/gpu/drm/i915/intel_lrc.c +3
-0+if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+Quoted 8 lines
Quoted 6 lines
FORCEWAKE_RENDER);drivers/gpu/drm/i915/intel_lrc.c +2
-0+if (dev_priv->uncore.fw_mediacount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+FORCEWAKE_MEDIA);This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 9 lines
Also unlike the comment says
runtime_pm_get()can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to callgen6_gt_force_wake_get()here, but maybe that would trigger amight_sleep()warning. the current force wake code duplication (esp. outsideintel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outsidegen6_gt_force_wake_get()? I never really liked hiding it there anyway.Quoted 8 lines
Yeah this is just broken design. And if you look at the other wheel to track outstanding gpu work (requests) then it's not needed at all.
But I'm not sure what's the priority of the "rework execlists to use requests" task is and when (if ever that will happen). Jesse is the arbiter for this stuff anyway, so adding him.
-DanielQuoted 4 lines
hmm , agreed do we have a reworked execlist? The reason why added this, on chv when i enable execlist, due to incorrect forcewake count is causing multiple entries to
forcewake_getresulting in "*ERROR*:
Timed out waiting for forcewake old ack to clear" "and Hang.
Quoted 4 lines
I'm hoping we can get execlists reworked on top of the request/seqno stuff shortly after it lands, but I don't think that's a reason to block this fix, since Chris is still busy fixing up the request changes.
Queued for -next, thanks for the patch.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Quoted 71 lines · Daniel Vetter
Quoted 70 lines · Jesse Barnes
Quoted 66 lines · On Tue, 09 Sep 2014 21:45:08 +0530 Deepak S <[email protected]> wrote:
Quoted 51 lines · Daniel Vetter
Quoted 41 lines · Ville Syrjälä
Quoted 32 lines · Ville Syrjälä
Quoted 30 lines · On Tue, Sep 09, 2014 at 07:14:16PM +0530, [email protected] wrote:
From: Deepak S <[email protected]>
In chv, we have two power wells Render & Media. We need to use corresponsing forcewake count. If we dont follow this we are getting error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to multiple entry into
__vlv_force_wake_get.Signed-off-by Deepak S <[email protected]>--- drivers/gpu/drm/i915/intel_lrc.c | 29+++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)
(quoted diff) +0
-0
b/drivers/gpu/drm/i915/intel_lrc.cindex bd1b28d..bafd38b 100644drivers/gpu/drm/i915/intel_lrc.c +0
-0@@ -300,8 +300,18 @@ static void execlists_elsp_write(structintel_engine_cs *ring, * Instead, we do the runtime_pm_get/put whencreating/destroying requests. */
spin_lock_irqsave(&dev_priv->uncore.lock, flags);drivers/gpu/drm/i915/intel_lrc.c +0
-2-if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get(dev_priv,
FORCEWAKE_ALL);drivers/gpu/drm/i915/intel_lrc.c +3
-0+if (IS_CHERRYVIEW (dev_priv->dev)) { +if (dev_priv->uncore.fw_rendercount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+Quoted 8 lines
Quoted 6 lines
FORCEWAKE_RENDER);drivers/gpu/drm/i915/intel_lrc.c +2
-0+if (dev_priv->uncore.fw_mediacount++ ==0 ) +dev_priv->uncore.funcs.force_wake_get(dev_priv,
+FORCEWAKE_MEDIA);This will wake both wells. Is that needed or should we just pick one based on the ring?
Quoted 9 lines
Also unlike the comment says
runtime_pm_get()can't sleep since someone must already be holding a reference, othwewise we surely can't go writing any registers. So in theory we should be able to callgen6_gt_force_wake_get()here, but maybe that would trigger amight_sleep()warning. the current force wake code duplication (esp. outsideintel_uncore.c) is rather unfortunate and I'd like to see it killed off. Maybe we just need to pull the rpm get/put outsidegen6_gt_force_wake_get()? I never really liked hiding it there anyway.Quoted 8 lines
Yeah this is just broken design. And if you look at the other wheel to track outstanding gpu work (requests) then it's not needed at all.
But I'm not sure what's the priority of the "rework execlists to use requests" task is and when (if ever that will happen). Jesse is the arbiter for this stuff anyway, so adding him.
-Danielhmm , agreed do we have a reworked execlist? The reason why added this, on chv when i enable execlist, due to incorrect forcewake count is causing multiple entries to
forcewake_getresulting in "*ERROR*:
Timed out waiting for forcewake old ack to clear" "and Hang.
Quoted 6 lines
Quoted 4 lines
I'm hoping we can get execlists reworked on top of the request/seqno stuff shortly after it lands, but I don't think that's a reason to block this fix, since Chris is still busy fixing up the request changes.
Queued for -next, thanks for the patch.
Aside if someone gets bored and wants to apply some polish: And __ variant of force_wake_get/put which only asserts that the device isn't runtime suspended instead of doing the runtime_pm_get/put would be cute - we have one other user in the hsw/bdw rpm code already.
The current force_wake_get/put functions would then just wrap those raw versions.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Quoted 7 lines · Daniel Vetter
Aside if someone gets bored and wants to apply some polish: And __ variant of force_wake_get/put which only asserts that the device isn't runtime suspended instead of doing the runtime_pm_get/put would be cute - we have one other user in the hsw/bdw rpm code already.
The current force_wake_get/put functions would then just wrap those raw versions.
I fail to see the reason why they take it at all. Conceptually, gen6_gt_force_wake_get() is just as lowlevel as the register access they wrap. The only one that breaks the rule is i915_debugfs.c and that seems easy enough to wrap with its own intel_runtime_pm_get/_put.
And once you've finish with that you can then sort out the mess that is the vlv variants. And now skl continues in the cut'n'paste'n'paste vein. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Hmm Ok Daniel. Let me try and clean up the forcewake. :)
Chris, most of the debugfs is already covered with runtime_get/put. If anything is missed we can add that
Thanks
Deepak
-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Wednesday, September 10, 2014 9:21 PM
To: Daniel Vetter
Cc: Jesse Barnes; S, Deepak; [email protected]
Subject: Re: [Intel-gfx] [PATCH] drm/i915: add cherryview specfic forcewake in execlists_elsp_write
Quoted 7 lines · Daniel Vetter
Aside if someone gets bored and wants to apply some polish: And __ variant of force_wake_get/put which only asserts that the device isn't runtime suspended instead of doing the runtime_pm_get/put would be cute - we have one other user in the hsw/bdw rpm code already.
The current force_wake_get/put functions would then just wrap those raw versions.
I fail to see the reason why they take it at all. Conceptually,gen6_gt_force_wake_get() is just as lowlevel as the register access they wrap. The only one that breaks the rule is i915_debugfs.c and that seems easy enough to wrap with its own intel_runtime_pm_get/_put.
And once you've finish with that you can then sort out the mess that is the vlv variants. And now skl continues in the cut'n'paste'n'paste vein. -Chris
--
Chris Wilson, Intel Open Source Technology Centre
PATCH 1/1 rpm
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 ++ drivers/gpu/drm/i915/intel_display.c | 20 ++------------ drivers/gpu/drm/i915/intel_lrc.c | 21 ++------------- drivers/gpu/drm/i915/intel_uncore.c | 52 +++++++++++++++--------------------- 4 files changed, 27 insertions(+), 68 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +2 -0
@@ -4148,6 +4148,7 @@ static int i915_forcewake_open(struct inode *inode, struct file *file)if (INTEL_INFO (dev)->gen <6 )return 0 ; +intel_runtime_pm_get (dev_priv);gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );return 0 ;
@@ -4162,6 +4163,7 @@ static int i915_forcewake_release(struct inode *inode, struct file *file)return 0 ;gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); +intel_runtime_pm_put (dev_priv);return 0 ; }
drivers/gpu/drm/i915/intel_display.c +2 -18
@@ -7596,7 +7596,6 @@ static void hsw_disable_lcpll(struct drm_i915_private *dev_priv,static void hsw_restore_lcpll (struct drm_i915_private *dev_priv) {uint32_t val; -unsigned long irqflags; val =I915_READ (LCPLL_CTL );
@@ -7607,19 +7606,8 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)/* * Make sure we're not on PC8 state before disabling PC8, otherwise * we'll hang the machine. To prevent PC8 state, just enable force_wake. -* -* The other problem is that hsw_restore_lcpll() is called as part of -* the runtime PM resume sequence, so we can't just call -* gen6_gt_force_wake_get() because that function calls -* intel_runtime_pm_get(), and we can't change the runtime PM refcount -* while we are on the resume sequence. So to solve this problem we have -* to call special forcewake code that doesn't touch runtime PM and -* doesn't enable the forcewake delayed work. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );if (val &LCPLL_POWER_DOWN_ALLOW ) { val &= ~LCPLL_POWER_DOWN_ALLOW ;
@@ -7649,11 +7637,7 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)DRM_ERROR ("Switching back to LCPLL failed\n" ); } -/* See the big comment above. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); }/*
drivers/gpu/drm/i915/intel_lrc.c +2 -19
@@ -243,7 +243,6 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,struct drm_i915_private *dev_priv = engine->i915;uint64_t tmp;uint32_t desc[4 ]; -unsigned long flags;/* XXX: You must always write both descriptors in the order below. */
@@ -260,18 +259,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine, desc[1 ] =upper_32_bits (tmp); desc[0 ] =lower_32_bits (tmp); -/* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes -* are in progress. -* -* The other problem is that we can't just call gen6_gt_force_wake_get() -* because that function calls intel_runtime_pm_get(), which might sleep. -* Instead, we do the runtime_pm_get/put when creating/destroying requests. -*/ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); - +gen6_gt_force_wake_get (dev_priv, engine->power_domains);I915_WRITE (RING_ELSP (engine), desc[1 ]);I915_WRITE (RING_ELSP (engine), desc[0 ]);I915_WRITE (RING_ELSP (engine), desc[3 ]);
@@ -280,12 +268,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,/* ELSP is a wo register, so use another nearby reg for posting instead */ POSTING_READ (RING_EXECLIST_STATUS (engine)); - -/* Release Force Wakeup (see the big comment above). */ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); +gen6_gt_force_wake_put (dev_priv, engine->power_domains); }static u16 next_tag (struct intel_engine_cs *engine)
drivers/gpu/drm/i915/intel_uncore.c +21 -31
101 lines not on this page · raw message
Signature · 1 line
1.9.1
Re: [PATCH] rpm
2014-09-10 13:43 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 20 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 ++ drivers/gpu/drm/i915/intel_display.c | 20 ++------------ drivers/gpu/drm/i915/intel_lrc.c | 21 ++------------- drivers/gpu/drm/i915/intel_uncore.c | 52 +++++++++++++++--------------------- 4 files changed, 27 insertions(+), 68 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4148,6 +4148,7 @@ static int i915_forcewake_open(struct inode *inode, struct file *file)if (INTEL_INFO (dev)->gen <6 )return 0 ; +intel_runtime_pm_get (dev_priv);gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );return 0 ;
drivers/gpu/drm/i915/i915_debugfs.c static int i915_forcewake_release(struct inode *inode, struct file *file) · line 4169
Quoted 7 lines
drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4162,6 +4163,7 @@ static int i915_forcewake_release(struct inode *inode, struct file *file)return 0 ;gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); +intel_runtime_pm_put (dev_priv);return 0 ; }
Just a minor comment on the chunk above. I didn't look at the rest of the patch.
We used to have exactly the code that you rewrote, but we decided to remove it because, without it, we can test that gen6_gt_force_wake_get actually wakes up the HW and keeps it awake until someone calls gen6_gt_force_wake_put.
If we add these get/put calls above, we won't really detect any problems related with the actual gen6_gt_force_wake_get/put functions, so we may hide problems.
Quoted 210 lines
drivers/gpu/drm/i915/intel_display.c +2
-18@@ -7596,7 +7596,6 @@ static void hsw_disable_lcpll(struct drm_i915_private *dev_priv,static void hsw_restore_lcpll (struct drm_i915_private *dev_priv) {uint32_t val; -unsigned long irqflags; val =I915_READ (LCPLL_CTL );@@ -7607,19 +7606,8 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)/* * Make sure we're not on PC8 state before disabling PC8, otherwise * we'll hang the machine. To prevent PC8 state, just enable force_wake. -* -* The other problem is that hsw_restore_lcpll() is called as part of -* the runtime PM resume sequence, so we can't just call -* gen6_gt_force_wake_get() because that function calls -* intel_runtime_pm_get(), and we can't change the runtime PM refcount -* while we are on the resume sequence. So to solve this problem we have -* to call special forcewake code that doesn't touch runtime PM and -* doesn't enable the forcewake delayed work. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );if (val &LCPLL_POWER_DOWN_ALLOW ) { val &= ~LCPLL_POWER_DOWN_ALLOW ;@@ -7649,11 +7637,7 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)DRM_ERROR ("Switching back to LCPLL failed\n" ); } -/* See the big comment above. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); }/* drivers/gpu/drm/i915/intel_lrc.c +2
-19@@ -243,7 +243,6 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,struct drm_i915_private *dev_priv = engine->i915;uint64_t tmp;uint32_t desc[4 ]; -unsigned long flags;/* XXX: You must always write both descriptors in the order below. */ @@ -260,18 +259,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine, desc[1 ] =upper_32_bits (tmp); desc[0 ] =lower_32_bits (tmp); -/* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes -* are in progress. -* -* The other problem is that we can't just call gen6_gt_force_wake_get() -* because that function calls intel_runtime_pm_get(), which might sleep. -* Instead, we do the runtime_pm_get/put when creating/destroying requests. -*/ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); - +gen6_gt_force_wake_get (dev_priv, engine->power_domains);I915_WRITE (RING_ELSP (engine), desc[1 ]);I915_WRITE (RING_ELSP (engine), desc[0 ]);I915_WRITE (RING_ELSP (engine), desc[3 ]);@@ -280,12 +268,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,/* ELSP is a wo register, so use another nearby reg for posting instead */ POSTING_READ (RING_EXECLIST_STATUS (engine)); - -/* Release Force Wakeup (see the big comment above). */ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); +gen6_gt_force_wake_put (dev_priv, engine->power_domains); }static u16 next_tag (struct intel_engine_cs *engine)drivers/gpu/drm/i915/intel_uncore.c +21
-32@@ -24,6 +24,8 @@#include "i915_drv.h" #include "intel_drv.h" +#include <linux/pm_runtime.h> +#define FORCEWAKE_ACK_TIMEOUT_MS 2#define __raw_i915_read8 (dev_priv__, reg__) readb((dev_priv__)->regs + (reg__))@@ -258,10 +260,6 @@ static void __vlv_force_wake_put(struct drm_i915_private *dev_priv,static void vlv_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER && dev_priv->uncore.fw_rendercount++ !=0 ) fw_engine &= ~FORCEWAKE_RENDER ;@@ -271,16 +269,10 @@ static void vlv_force_wake_get(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void vlv_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER ) {WARN_ON (!dev_priv->uncore.fw_rendercount);if (--dev_priv->uncore.fw_rendercount !=0 )@@ -295,8 +287,6 @@ static void vlv_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_put (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void gen6_force_wake_timer (unsigned long arg)@@ -406,15 +396,18 @@ void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv, int fw_engine)if (!dev_priv->uncore.funcs.force_wake_get)return ; -intel_runtime_pm_get (dev_priv); - -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) -return vlv_force_wake_get (dev_priv, fw_engine); +WARN_ON (!pm_runtime_active (&dev_priv->dev->pdev->dev));spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); + +/* Redirect to VLV specific routine */ +if (IS_VALLEYVIEW (dev_priv->dev)) { +vlv_force_wake_get (dev_priv, fw_engine); + }else { +if (dev_priv->uncore.forcewake_count++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_ALL ); + }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }@@ -428,25 +421,22 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)if (!dev_priv->uncore.funcs.force_wake_put)return ; +spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); +/* Redirect to VLV specific routine */ if (IS_VALLEYVIEW (dev_priv->dev)) {vlv_force_wake_put (dev_priv, fw_engine); - goto out; - } - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -WARN_ON (!dev_priv->uncore.forcewake_count); - -if (--dev_priv->uncore.forcewake_count ==0 ) { - dev_priv->uncore.forcewake_count++; -mod_timer_pinned (&dev_priv->uncore.force_wake_timer, - jiffies +1 ); + }else { +WARN_ON (!dev_priv->uncore.forcewake_count); +if (--dev_priv->uncore.forcewake_count ==0 ) { + dev_priv->uncore.forcewake_count++; +mod_timer_pinned (&dev_priv->uncore.force_wake_timer, + jiffies +1 ); + } }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); -out: -intel_runtime_pm_put (dev_priv); }void assert_force_wake_inactive (struct drm_i915_private *dev_priv) --1.9.1
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 1 line
Paulo Zanoni
Re: [PATCH] rpm
Quoted 31 lines · Paulo Zanoni
2014-09-10 13:43 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 20 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 ++ drivers/gpu/drm/i915/intel_display.c | 20 ++------------ drivers/gpu/drm/i915/intel_lrc.c | 21 ++------------- drivers/gpu/drm/i915/intel_uncore.c | 52 +++++++++++++++--------------------- 4 files changed, 27 insertions(+), 68 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4148,6 +4148,7 @@ static int i915_forcewake_open(struct inode *inode, struct file *file)if (INTEL_INFO (dev)->gen <6 )return 0 ; +intel_runtime_pm_get (dev_priv);gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );return 0 ;drivers/gpu/drm/i915/i915_debugfs.c static int i915_forcewake_release(struct inode *inode, struct file *file) · line 4169
Quoted 7 lines
drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4162,6 +4163,7 @@ static int i915_forcewake_release(struct inode *inode, struct file *file)return 0 ;gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); +intel_runtime_pm_put (dev_priv);return 0 ; }Just a minor comment on the chunk above. I didn't look at the rest of the patch.
Quoted 4 lines
We used to have exactly the code that you rewrote, but we decided to remove it because, without it, we can test that
gen6_gt_force_wake_getactually wakes up the HW and keeps it awake until someone callsgen6_gt_force_wake_put.
But why? gen6_gt_force_wake_get() should never be called outside of a rpm context - it is lowlevel register access.
Quoted 3 lines
If we add these get/put calls above, we won't really detect any problems related with the actual gen6_gt_force_wake_get/put functions, so we may hide problems.
See above. That's how the design got broken in the first place and we have very lowlevel code being copied and pasted throughout the driver. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH] rpm
In reply to Paulo ZanoniQuoted 31 lines · Paulo Zanoni
2014-09-10 13:43 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 20 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 ++ drivers/gpu/drm/i915/intel_display.c | 20 ++------------ drivers/gpu/drm/i915/intel_lrc.c | 21 ++------------- drivers/gpu/drm/i915/intel_uncore.c | 52 +++++++++++++++--------------------- 4 files changed, 27 insertions(+), 68 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4148,6 +4148,7 @@ static int i915_forcewake_open(struct inode *inode, struct file *file)if (INTEL_INFO (dev)->gen <6 )return 0 ; +intel_runtime_pm_get (dev_priv);gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );return 0 ;drivers/gpu/drm/i915/i915_debugfs.c static int i915_forcewake_release(struct inode *inode, struct file *file) · line 4169
Quoted 7 lines
drivers/gpu/drm/i915/i915_debugfs.c +1
-0@@ -4162,6 +4163,7 @@ static int i915_forcewake_release(struct inode *inode, struct file *file)return 0 ;gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); +intel_runtime_pm_put (dev_priv);return 0 ; }Just a minor comment on the chunk above. I didn't look at the rest of the patch.
Quoted 8 lines
We used to have exactly the code that you rewrote, but we decided to remove it because, without it, we can test that
gen6_gt_force_wake_getactually wakes up the HW and keeps it awake until someone callsgen6_gt_force_wake_put.If we add these get/put calls above, we won't really detect any problems related with the actual gen6_gt_force_wake_get/put functions, so we may hide problems.
Well, I think the patch is a great. With the rpm stuff killed from forcewake handling we could actually use the delayed forcewake put in the register access macros, which is where it might actually do some good. IIRC that was even the original intention when the timer got added. In the current form I'm pretty sure the timer is practically useless.
Quoted 220 lines
Quoted 211 lines
drivers/gpu/drm/i915/intel_display.c +2
-18@@ -7596,7 +7596,6 @@ static void hsw_disable_lcpll(struct drm_i915_private *dev_priv,static void hsw_restore_lcpll (struct drm_i915_private *dev_priv) {uint32_t val; -unsigned long irqflags; val =I915_READ (LCPLL_CTL );@@ -7607,19 +7606,8 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)/* * Make sure we're not on PC8 state before disabling PC8, otherwise * we'll hang the machine. To prevent PC8 state, just enable force_wake. -* -* The other problem is that hsw_restore_lcpll() is called as part of -* the runtime PM resume sequence, so we can't just call -* gen6_gt_force_wake_get() because that function calls -* intel_runtime_pm_get(), and we can't change the runtime PM refcount -* while we are on the resume sequence. So to solve this problem we have -* to call special forcewake code that doesn't touch runtime PM and -* doesn't enable the forcewake delayed work. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );if (val &LCPLL_POWER_DOWN_ALLOW ) { val &= ~LCPLL_POWER_DOWN_ALLOW ;@@ -7649,11 +7637,7 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)DRM_ERROR ("Switching back to LCPLL failed\n" ); } -/* See the big comment above. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); }/* drivers/gpu/drm/i915/intel_lrc.c +2
-19@@ -243,7 +243,6 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,struct drm_i915_private *dev_priv = engine->i915;uint64_t tmp;uint32_t desc[4 ]; -unsigned long flags;/* XXX: You must always write both descriptors in the order below. */ @@ -260,18 +259,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine, desc[1 ] =upper_32_bits (tmp); desc[0 ] =lower_32_bits (tmp); -/* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes -* are in progress. -* -* The other problem is that we can't just call gen6_gt_force_wake_get() -* because that function calls intel_runtime_pm_get(), which might sleep. -* Instead, we do the runtime_pm_get/put when creating/destroying requests. -*/ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); - +gen6_gt_force_wake_get (dev_priv, engine->power_domains);I915_WRITE (RING_ELSP (engine), desc[1 ]);I915_WRITE (RING_ELSP (engine), desc[0 ]);I915_WRITE (RING_ELSP (engine), desc[3 ]);@@ -280,12 +268,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,/* ELSP is a wo register, so use another nearby reg for posting instead */ POSTING_READ (RING_EXECLIST_STATUS (engine)); - -/* Release Force Wakeup (see the big comment above). */ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); +gen6_gt_force_wake_put (dev_priv, engine->power_domains); }static u16 next_tag (struct intel_engine_cs *engine)drivers/gpu/drm/i915/intel_uncore.c +21
-32@@ -24,6 +24,8 @@#include "i915_drv.h" #include "intel_drv.h" +#include <linux/pm_runtime.h> +#define FORCEWAKE_ACK_TIMEOUT_MS 2#define __raw_i915_read8 (dev_priv__, reg__) readb((dev_priv__)->regs + (reg__))@@ -258,10 +260,6 @@ static void __vlv_force_wake_put(struct drm_i915_private *dev_priv,static void vlv_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER && dev_priv->uncore.fw_rendercount++ !=0 ) fw_engine &= ~FORCEWAKE_RENDER ;@@ -271,16 +269,10 @@ static void vlv_force_wake_get(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void vlv_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER ) {WARN_ON (!dev_priv->uncore.fw_rendercount);if (--dev_priv->uncore.fw_rendercount !=0 )@@ -295,8 +287,6 @@ static void vlv_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_put (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void gen6_force_wake_timer (unsigned long arg)@@ -406,15 +396,18 @@ void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv, int fw_engine)if (!dev_priv->uncore.funcs.force_wake_get)return ; -intel_runtime_pm_get (dev_priv); - -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) -return vlv_force_wake_get (dev_priv, fw_engine); +WARN_ON (!pm_runtime_active (&dev_priv->dev->pdev->dev));spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); + +/* Redirect to VLV specific routine */ +if (IS_VALLEYVIEW (dev_priv->dev)) { +vlv_force_wake_get (dev_priv, fw_engine); + }else { +if (dev_priv->uncore.forcewake_count++ ==0 ) + dev_priv->uncore.funcs.force_wake_get (dev_priv, +FORCEWAKE_ALL ); + }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }@@ -428,25 +421,22 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)if (!dev_priv->uncore.funcs.force_wake_put)return ; +spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); +/* Redirect to VLV specific routine */ if (IS_VALLEYVIEW (dev_priv->dev)) {vlv_force_wake_put (dev_priv, fw_engine); - goto out; - } - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -WARN_ON (!dev_priv->uncore.forcewake_count); - -if (--dev_priv->uncore.forcewake_count ==0 ) { - dev_priv->uncore.forcewake_count++; -mod_timer_pinned (&dev_priv->uncore.force_wake_timer, - jiffies +1 ); + }else { +WARN_ON (!dev_priv->uncore.forcewake_count); +if (--dev_priv->uncore.forcewake_count ==0 ) { + dev_priv->uncore.forcewake_count++; +mod_timer_pinned (&dev_priv->uncore.force_wake_timer, + jiffies +1 ); + } }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); -out: -intel_runtime_pm_put (dev_priv); }void assert_force_wake_inactive (struct drm_i915_private *dev_priv) --1.9.1
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfxSignature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 2 lines
Ville Syrjälä Intel OTC
Re: [PATCH] rpm
In reply to Chris WilsonQuoted 161 lines · Chris Wilson
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 ++ drivers/gpu/drm/i915/intel_display.c | 20 ++------------ drivers/gpu/drm/i915/intel_lrc.c | 21 ++------------- drivers/gpu/drm/i915/intel_uncore.c | 52 +++++++++++++++--------------------- 4 files changed, 27 insertions(+), 68 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-0@@ -4148,6 +4148,7 @@ static int i915_forcewake_open(struct inode *inode, struct file *file)if (INTEL_INFO (dev)->gen <6 )return 0 ; +intel_runtime_pm_get (dev_priv);gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );return 0 ;@@ -4162,6 +4163,7 @@ static int i915_forcewake_release(struct inode *inode, struct file *file)return 0 ;gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); +intel_runtime_pm_put (dev_priv);return 0 ; }drivers/gpu/drm/i915/intel_display.c +2
-18@@ -7596,7 +7596,6 @@ static void hsw_disable_lcpll(struct drm_i915_private *dev_priv,static void hsw_restore_lcpll (struct drm_i915_private *dev_priv) {uint32_t val; -unsigned long irqflags; val =I915_READ (LCPLL_CTL );@@ -7607,19 +7606,8 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)/* * Make sure we're not on PC8 state before disabling PC8, otherwise * we'll hang the machine. To prevent PC8 state, just enable force_wake. -* -* The other problem is that hsw_restore_lcpll() is called as part of -* the runtime PM resume sequence, so we can't just call -* gen6_gt_force_wake_get() because that function calls -* intel_runtime_pm_get(), and we can't change the runtime PM refcount -* while we are on the resume sequence. So to solve this problem we have -* to call special forcewake code that doesn't touch runtime PM and -* doesn't enable the forcewake delayed work. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_get (dev_priv,FORCEWAKE_ALL );if (val &LCPLL_POWER_DOWN_ALLOW ) { val &= ~LCPLL_POWER_DOWN_ALLOW ;@@ -7649,11 +7637,7 @@ static void hsw_restore_lcpll(struct drm_i915_private *dev_priv)DRM_ERROR ("Switching back to LCPLL failed\n" ); } -/* See the big comment above. */ -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +gen6_gt_force_wake_put (dev_priv,FORCEWAKE_ALL ); }/* drivers/gpu/drm/i915/intel_lrc.c +2
-19@@ -243,7 +243,6 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,struct drm_i915_private *dev_priv = engine->i915;uint64_t tmp;uint32_t desc[4 ]; -unsigned long flags;/* XXX: You must always write both descriptors in the order below. */ @@ -260,18 +259,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine, desc[1 ] =upper_32_bits (tmp); desc[0 ] =lower_32_bits (tmp); -/* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes -* are in progress. -* -* The other problem is that we can't just call gen6_gt_force_wake_get() -* because that function calls intel_runtime_pm_get(), which might sleep. -* Instead, we do the runtime_pm_get/put when creating/destroying requests. -*/ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); - +gen6_gt_force_wake_get (dev_priv, engine->power_domains);I915_WRITE (RING_ELSP (engine), desc[1 ]);I915_WRITE (RING_ELSP (engine), desc[0 ]);I915_WRITE (RING_ELSP (engine), desc[3 ]);@@ -280,12 +268,7 @@ static void execlists_submit_pair(struct intel_engine_cs *engine,/* ELSP is a wo register, so use another nearby reg for posting instead */ POSTING_READ (RING_EXECLIST_STATUS (engine)); - -/* Release Force Wakeup (see the big comment above). */ -spin_lock_irqsave (&dev_priv->uncore.lock, flags); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, flags); +gen6_gt_force_wake_put (dev_priv, engine->power_domains); }static u16 next_tag (struct intel_engine_cs *engine)drivers/gpu/drm/i915/intel_uncore.c +2
-10@@ -24,6 +24,8 @@#include "i915_drv.h" #include "intel_drv.h" +#include <linux/pm_runtime.h> +#define FORCEWAKE_ACK_TIMEOUT_MS 2#define __raw_i915_read8 (dev_priv__, reg__) readb((dev_priv__)->regs + (reg__))@@ -258,10 +260,6 @@ static void __vlv_force_wake_put(struct drm_i915_private *dev_priv,static void vlv_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER && dev_priv->uncore.fw_rendercount++ !=0 ) fw_engine &= ~FORCEWAKE_RENDER ;@@ -271,16 +269,10 @@ static void vlv_force_wake_get(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void vlv_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) { -unsigned long irqflags; - -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -if (fw_engine &FORCEWAKE_RENDER ) {WARN_ON (!dev_priv->uncore.fw_rendercount);if (--dev_priv->uncore.fw_rendercount !=0 )
drivers/gpu/drm/i915/intel_uncore.c static void vlv_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 292
Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-2@@ -295,8 +287,6 @@ static void vlv_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)if (fw_engine) dev_priv->uncore.funcs.force_wake_put (dev_priv, fw_engine); - -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }static void gen6_force_wake_timer (unsigned long arg)
Looks like you forgot to kill the rpm_put() from the timer. And then we also need to make sure the timer is approriately cancelled when runtime suspending the device.
Signature · 2 lines
Ville Syrjälä Intel OTC
Re: [PATCH] rpm
Quoted 6 lines · Ville Syrjälä
Quoted 1 line · Chris Wilson
static void
gen6_force_wake_timer(unsigned long arg)Looks like you forgot to kill the
rpm_put()from the timer. And then we also need to make sure the timer is approriately cancelled when runtime suspending the device.
Or maybe I have that already fixed in my tree so you don't see it in this patch ;-)
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
PATCH 1/1 drm/i915: Reduce duplicated forcewake logic
In reply to Chris WilsonIntroduce a structure to track the individual forcewake domains and use that to eliminated duplicate logic.
--- drivers/gpu/drm/i915/i915_debugfs.c | 41 +++--- drivers/gpu/drm/i915/i915_drv.h | 32 +++-- drivers/gpu/drm/i915/intel_uncore.c | 265 ++++++++++++++++-------------------- 3 files changed, 157 insertions(+), 181 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +16 -25
@@ -1317,7 +1317,6 @@ static int vlv_drpc_info(struct seq_file *m)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;u32 rpmodectl1, rcctl1; -unsigned fw_rendercount =0 , fw_mediacount =0 ;intel_runtime_pm_get (dev_priv);
@@ -1350,22 +1349,12 @@ static int vlv_drpc_info(struct seq_file *m)seq_printf (m,"Media RC6 residency since boot: %u\n" ,I915_READ (VLV_GT_MEDIA_RC6 )); -spin_lock_irq (&dev_priv->uncore.lock); - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; -spin_unlock_irq (&dev_priv->uncore.lock); - -seq_printf (m,"Forcewake Render Count = %u\n" , fw_rendercount); -seq_printf (m,"Forcewake Media Count = %u\n" , fw_mediacount); - - -return 0 ; +return i915_gen6_forcewake_count_info (m, data); }static int gen6_drpc_info (struct seq_file *m) { -struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;
@@ -1379,7 +1368,7 @@ static int gen6_drpc_info(struct seq_file *m)intel_runtime_pm_get (dev_priv);spin_lock_irq (&dev_priv->uncore.lock); - forcewake_count = dev_priv->uncore.forcewake_count; + forcewake_count = dev_priv->uncore.fw_domain[FW_DOMAIN_RENDER ].wake_count;spin_unlock_irq (&dev_priv->uncore.lock);if (forcewake_count) {
@@ -1976,21 +1965,23 @@ static int i915_gen6_forcewake_count_info(struct seq_file *m, void *data)struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private; -unsigned forcewake_count =0 , fw_rendercount =0 , fw_mediacount =0 ; +const char *domain_names[] = { +"render" , +"media" , + }; +int i;spin_lock_irq (&dev_priv->uncore.lock); -if (IS_VALLEYVIEW (dev)) { - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; - }else - forcewake_count = dev_priv->uncore.forcewake_count; -spin_unlock_irq (&dev_priv->uncore.lock); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((dev_priv->uncore.fw_domains & (1 << i)) ==0 ) +continue ; -if (IS_VALLEYVIEW (dev)) { -seq_printf (m,"fw_rendercount = %u\n" , fw_rendercount); -seq_printf (m,"fw_mediacount = %u\n" , fw_mediacount); - }else -seq_printf (m,"forcewake count = %u\n" , forcewake_count); +seq_printf (m,"%s.wake_count = %u\n" , + domain_names[i], + dev_priv->uncore.fw_domain[i].wake_count); + } + +spin_unlock_irq (&dev_priv->uncore.lock);return 0 ; }
drivers/gpu/drm/i915/i915_drv.h +21 -11
@@ -528,18 +528,30 @@ struct intel_uncore_funcs { uint64_t val, bool trace); }; +enum { +FW_DOMAIN_RENDER =0 , +FW_DOMAIN_MEDIA , + +FW_DOMAIN_COUNT +}; +struct intel_uncore {spinlock_t lock;/** lock is also taken in irq contexts. */ struct intel_uncore_funcs funcs;unsigned fifo_count; -unsigned forcewake_count; - -unsigned fw_rendercount; -unsigned fw_mediacount; +unsigned fw_domains; -struct timer_list force_wake_timer; +struct intel_uncore_forcewake_domain { +struct drm_i915_private *i915; +int id; +unsigned wake_count; +struct timer_list timer; + } fw_domain[FW_DOMAIN_COUNT ]; +#define FORCEWAKE_RENDER (1 << FW_DOMAIN_RENDER) +#define FORCEWAKE_MEDIA (1 << FW_DOMAIN_MEDIA) +#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) };#define DEV_INFO_FOR_EACH_FLAG (func, sep) \
@@ -2948,8 +2960,10 @@ extern void intel_display_print_error_state(struct drm_i915_error_state_buf *e,* must be set to prevent GT core from power down and stale values being * returned. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine); -void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine); +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains); +void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, +unsigned fw_domains);void assert_force_wake_inactive (struct drm_i915_private *dev_priv);int sandybridge_pcode_read (struct drm_i915_private *dev_priv,u8 mbox,u32 *val);
@@ -2981,10 +2995,6 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val);int vlv_gpu_freq (struct drm_i915_private *dev_priv,int val);int vlv_freq_opcode (struct drm_i915_private *dev_priv,int val); -#define FORCEWAKE_RENDER (1 << 0) -#define FORCEWAKE_MEDIA (1 << 1) -#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) -#define I915_READ8 (reg) dev_priv->uncore.funcs.mmio_readb(dev_priv, (reg), true)#define I915_WRITE8 (reg, val) dev_priv->uncore.funcs.mmio_writeb(dev_priv, (reg), (val), true)
drivers/gpu/drm/i915/intel_uncore.c +120 -145
439 lines not on this page · raw message
Signature · 1 line
1.9.1
Re: [PATCH] drm/i915: Reduce duplicated forcewake logic
Quoted 7 lines · Chris Wilson
Introduce a structure to track the individual forcewake domains and use that to eliminated duplicate logic.
--- drivers/gpu/drm/i915/i915_debugfs.c | 41 +++--- drivers/gpu/drm/i915/i915_drv.h | 32 +++-- drivers/gpu/drm/i915/intel_uncore.c | 265 ++++++++++++++++-------------------- 3 files changed, 157 insertions(+), 181 deletions(-)
As said yesterday I like the idea, some comments and questions inline.
Quoted 56 lines
drivers/gpu/drm/i915/i915_debugfs.c +7
-14@@ -1317,7 +1317,6 @@ static int vlv_drpc_info(struct seq_file *m)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;u32 rpmodectl1, rcctl1; -unsigned fw_rendercount =0 , fw_mediacount =0 ;intel_runtime_pm_get (dev_priv);@@ -1350,22 +1349,12 @@ static int vlv_drpc_info(struct seq_file *m)seq_printf (m,"Media RC6 residency since boot: %u\n" ,I915_READ (VLV_GT_MEDIA_RC6 )); -spin_lock_irq (&dev_priv->uncore.lock); - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; -spin_unlock_irq (&dev_priv->uncore.lock); - -seq_printf (m,"Forcewake Render Count = %u\n" , fw_rendercount); -seq_printf (m,"Forcewake Media Count = %u\n" , fw_mediacount); - - -return 0 ; +return i915_gen6_forcewake_count_info (m, data); }static int gen6_drpc_info (struct seq_file *m) { -struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;@@ -1379,7 +1368,7 @@ static int gen6_drpc_info(struct seq_file *m)intel_runtime_pm_get (dev_priv);spin_lock_irq (&dev_priv->uncore.lock); - forcewake_count = dev_priv->uncore.forcewake_count; + forcewake_count = dev_priv->uncore.fw_domain[FW_DOMAIN_RENDER ].wake_count;spin_unlock_irq (&dev_priv->uncore.lock);if (forcewake_count) {@@ -1976,21 +1965,23 @@ static int i915_gen6_forcewake_count_info(struct seq_file *m, void *data)struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private; -unsigned forcewake_count =0 , fw_rendercount =0 , fw_mediacount =0 ; +const char *domain_names[] = { +"render" , +"media" , + }; +int i;
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gen6_forcewake_count_info(struct seq_file *m, void *data) · line 1977
Quoted 10 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-6spin_lock_irq (&dev_priv->uncore.lock); -if (IS_VALLEYVIEW (dev)) { - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; - }else - forcewake_count = dev_priv->uncore.forcewake_count; -spin_unlock_irq (&dev_priv->uncore.lock); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((dev_priv->uncore.fw_domains & (1 << i)) ==0 ) +continue ;
What do you think about something like for_each_forcewake_domain (and perhaps for_each_possible_forcewake_domain) for readability since there is a fair number of these loops?
Quoted 19 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-5-if (IS_VALLEYVIEW (dev)) { -seq_printf (m,"fw_rendercount = %u\n" , fw_rendercount); -seq_printf (m,"fw_mediacount = %u\n" , fw_mediacount); - }else -seq_printf (m,"forcewake count = %u\n" , forcewake_count); +seq_printf (m,"%s.wake_count = %u\n" , + domain_names[i], + dev_priv->uncore.fw_domain[i].wake_count); + } + +spin_unlock_irq (&dev_priv->uncore.lock);return 0 ; }drivers/gpu/drm/i915/i915_drv.h +0
-0
drivers/gpu/drm/i915/i915_drv.h struct intel_uncore_funcs { · line 536
Quoted 9 lines
drivers/gpu/drm/i915/i915_drv.h +6
-0@@ -528,18 +528,30 @@ struct intel_uncore_funcs { uint64_t val, bool trace); }; +enum { +FW_DOMAIN_RENDER =0 , +FW_DOMAIN_MEDIA , + +FW_DOMAIN_COUNT +};
Would there be any need to attempt hiding these somehow so people don't try to pass these instead of FORCEWAKE_... ones into relevant functions? No idea how though. Maybe gcc will complain if passing in enum into int?
Quoted 6 lines
drivers/gpu/drm/i915/i915_drv.h +1
-0+struct intel_uncore {spinlock_t lock;/** lock is also taken in irq contexts. */ struct intel_uncore_funcs funcs;
drivers/gpu/drm/i915/i915_drv.h struct intel_uncore_funcs { · line 547
Quoted 10 lines
drivers/gpu/drm/i915/i915_drv.h +3
-5unsigned fifo_count; -unsigned forcewake_count; - -unsigned fw_rendercount; -unsigned fw_mediacount; +unsigned fw_domains; -struct timer_list force_wake_timer; +struct intel_uncore_forcewake_domain { +struct drm_i915_private *i915;
Should this be named dev_priv for consistency?
Quoted 84 lines
drivers/gpu/drm/i915/i915_drv.h +11
-6+int id; +unsigned wake_count; +struct timer_list timer; + } fw_domain[FW_DOMAIN_COUNT ]; +#define FORCEWAKE_RENDER (1 << FW_DOMAIN_RENDER) +#define FORCEWAKE_MEDIA (1 << FW_DOMAIN_MEDIA) +#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) };#define DEV_INFO_FOR_EACH_FLAG (func, sep) \@@ -2948,8 +2960,10 @@ extern void intel_display_print_error_state(struct drm_i915_error_state_buf *e,* must be set to prevent GT core from power down and stale values being * returned. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine); -void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine); +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains); +void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, +unsigned fw_domains);void assert_force_wake_inactive (struct drm_i915_private *dev_priv);int sandybridge_pcode_read (struct drm_i915_private *dev_priv,u8 mbox,u32 *val);@@ -2981,10 +2995,6 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val);int vlv_gpu_freq (struct drm_i915_private *dev_priv,int val);int vlv_freq_opcode (struct drm_i915_private *dev_priv,int val); -#define FORCEWAKE_RENDER (1 << 0) -#define FORCEWAKE_MEDIA (1 << 1) -#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) -#define I915_READ8 (reg) dev_priv->uncore.funcs.mmio_readb(dev_priv, (reg), true)#define I915_WRITE8 (reg, val) dev_priv->uncore.funcs.mmio_writeb(dev_priv, (reg), (val), true)drivers/gpu/drm/i915/intel_uncore.c +5
-5@@ -73,7 +73,7 @@ static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv) }static void __gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {if (wait_for_atomic ((__raw_i915_read32 (dev_priv,FORCEWAKE_ACK ) &1 ) ==0 ,FORCEWAKE_ACK_TIMEOUT_MS ))@@ -99,7 +99,7 @@ static void __gen7_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv) }static void __gen7_gt_force_wake_mt_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {u32 forcewake_ack;@@ -136,7 +136,7 @@ static void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv) }static void __gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {__raw_i915_write32 (dev_priv,FORCEWAKE ,0 );/* something from same cacheline, but !FORCEWAKE */ @@ -145,7 +145,7 @@ static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, }static void __gen7_gt_force_wake_mt_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {__raw_i915_write32 (dev_priv,FORCEWAKE_MT ,_MASKED_BIT_DISABLE (FORCEWAKE_KERNEL ));@@ -194,7 +194,7 @@ static void vlv_force_wake_reset(struct drm_i915_private *dev_priv) }static void __vlv_force_wake_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {/* Check for Render Engine */ if (FORCEWAKE_RENDER & fw_engine) {
drivers/gpu/drm/i915/intel_uncore.c static void __vlv_force_wake_get(struct drm_i915_private *dev_priv, · line 242
Quoted 6 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-1@@ -238,9 +238,8 @@ static void __vlv_force_wake_get(struct drm_i915_private *dev_priv, }static void __vlv_force_wake_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {
What's up with the above series of indentation changes?
Quoted 104 lines
drivers/gpu/drm/i915/intel_uncore.c +20
-51-/* Check for Render Engine */ if (FORCEWAKE_RENDER & fw_engine) __raw_i915_write32(dev_priv, FORCEWAKE_VLV,@@ -258,59 +257,35 @@ static void __vlv_force_wake_put(struct drm_i915_private *dev_priv,gen6_gt_check_fifodbg (dev_priv); } -static void vlv_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) -{ -if (fw_engine &FORCEWAKE_RENDER && - dev_priv->uncore.fw_rendercount++ !=0 ) - fw_engine &= ~FORCEWAKE_RENDER ; -if (fw_engine &FORCEWAKE_MEDIA && - dev_priv->uncore.fw_mediacount++ !=0 ) - fw_engine &= ~FORCEWAKE_MEDIA ; - -if (fw_engine) - dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_engine); -} - -static void vlv_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) -{ -if (fw_engine &FORCEWAKE_RENDER ) { -WARN_ON (!dev_priv->uncore.fw_rendercount); -if (--dev_priv->uncore.fw_rendercount !=0 ) - fw_engine &= ~FORCEWAKE_RENDER ; - } - -if (fw_engine &FORCEWAKE_MEDIA ) { -WARN_ON (!dev_priv->uncore.fw_mediacount); -if (--dev_priv->uncore.fw_mediacount !=0 ) - fw_engine &= ~FORCEWAKE_MEDIA ; - } - -if (fw_engine) - dev_priv->uncore.funcs.force_wake_put (dev_priv, fw_engine); -} -static void gen6_force_wake_timer (unsigned long arg) { -struct drm_i915_private *dev_priv = (void *)arg; +struct intel_uncore_forcewake_domain *domain = (void *)arg;unsigned long irqflags; -assert_device_not_suspended (dev_priv); +assert_device_not_suspended (domain->i915); -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -WARN_ON (!dev_priv->uncore.forcewake_count); +spin_lock_irqsave (&domain->i915->uncore.lock, irqflags); +WARN_ON (!domain->wake_count); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +if (--domain->wake_count ==0 ) + domain->i915->uncore.funcs.force_wake_put (domain->i915, +1 << domain->id); +spin_unlock_irqrestore (&domain->i915->uncore.lock, irqflags); }void intel_uncore_forcewake_reset (struct drm_device *dev,bool restore) {struct drm_i915_private *dev_priv = dev->dev_private;unsigned long irqflags; +int i; -if (del_timer_sync (&dev_priv->uncore.force_wake_timer)) -gen6_force_wake_timer ((unsigned long )dev_priv); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++ ) { +if ((dev_priv->uncore.fw_domains & (1 << i)) ==0 ) +continue ; + +if (del_timer_sync (&dev_priv->uncore.fw_domain[i].timer)) +gen6_force_wake_timer ((unsigned long )&dev_priv->uncore.fw_domain[i]); + }/* Hold uncore.lock across reset to prevent any register access * with forcewake not set correctly @@ -327,18 +302,12 @@ void intel_uncore_forcewake_reset(struct drm_device *dev, bool restore)if (restore) {/* If reset with a user forcewake, try to restore */ unsigned fw =0 ; +int i; -if (IS_VALLEYVIEW (dev)) { -if (dev_priv->uncore.fw_rendercount) - fw |=FORCEWAKE_RENDER ; - -if (dev_priv->uncore.fw_mediacount) - fw |=FORCEWAKE_MEDIA ; - }else { -if (dev_priv->uncore.forcewake_count) - fw =FORCEWAKE_ALL ; +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if (dev_priv->uncore.fw_domain[i].wake_count) + fw |=1 << i; } -if (fw) dev_priv->uncore.funcs.force_wake_get (dev_priv, fw);
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_sanitize(struct drm_device *dev) · line 362
Quoted 6 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-1@@ -389,50 +358,60 @@ void intel_uncore_sanitize(struct drm_device *dev)* be called at the beginning of the sequence followed by a call to * gen6_gt_force_wake_put() at the end of the sequence. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains)
Is gen6_ still the best prefix for these two?
Should you also use the opportunity to add kerneldoc now that Daniel is making a bit push in that direction?
drivers/gpu/drm/i915/intel_uncore.c static void __vlv_force_wake_put(struct drm_i915_private *dev_priv, · line 280
Quoted 3 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-0{unsigned long irqflags; +int i;
Not sure if it makes any difference nowadays to use unsigned for loops like this.
Quoted 55 lines
drivers/gpu/drm/i915/intel_uncore.c +15
-17if (!dev_priv->uncore.funcs.force_wake_get)return ;WARN_ON (!pm_runtime_active (&dev_priv->dev->pdev->dev)); + fw_domains &= dev_priv->uncore.fw_domains; +spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) { -vlv_force_wake_get (dev_priv, fw_engine); - }else { -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv, -FORCEWAKE_ALL ); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +if (dev_priv->uncore.fw_domain[i].wake_count++) + fw_domains &= ~(1 << i); } + +if (fw_domains) + dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_domains); +spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }/* * see gen6_gt_force_wake_get() */ -void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) +void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, +unsigned fw_domains) {unsigned long irqflags; +int i;if (!dev_priv->uncore.funcs.force_wake_put)return ;spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) { -vlv_force_wake_put (dev_priv, fw_engine); - }else { -WARN_ON (!dev_priv->uncore.forcewake_count); -if (--dev_priv->uncore.forcewake_count ==0 ) { - dev_priv->uncore.forcewake_count++; -mod_timer_pinned (&dev_priv->uncore.force_wake_timer, - jiffies +1 );
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_sanitize(struct drm_device *dev) · line 412
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +9
-1- } +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +WARN_ON (!dev_priv->uncore.fw_domain[i].wake_count); +if (--dev_priv->uncore.fw_domain[i].wake_count) +continue ; + + dev_priv->uncore.fw_domain[i].wake_count++;
I would put a comment here about why we are incrementing after decrementing.
Possibly also what is the purpose of the timer if not already documented somewhere.
Quoted 21 lines
drivers/gpu/drm/i915/intel_uncore.c +6
-1+mod_timer_pinned (&dev_priv->uncore.fw_domain[i].timer, + jiffies +1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags);@@ -441,10 +420,13 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)void assert_force_wake_inactive (struct drm_i915_private *dev_priv) { +int i; +if (!dev_priv->uncore.funcs.force_wake_get)return ; -WARN_ON (dev_priv->uncore.forcewake_count >0 ); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) +WARN_ON (dev_priv->uncore.fw_domain[i].wake_count >0 ); }/* We give fast paths for the really cool registers */
drivers/gpu/drm/i915/intel_uncore.c __gen2_read(64) · line 564
Quoted 5 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-0@@ -578,19 +560,38 @@ __gen2_read(64)trace_i915_reg_rw (false, reg, val,sizeof (val), trace); \ return val +static inline void __force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains)
This function is to be used by what type of callers compared to gen6_gt_force_wake_get?
Quoted 53 lines
drivers/gpu/drm/i915/intel_uncore.c +25
-15+{ +int i; + +/* Ideally GCC would be constant-fold and eliminate this loop */ + +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +if (dev_priv->uncore.fw_domain[i].wake_count) { + fw_domains &= ~(1 << i); +continue ; + } + + dev_priv->uncore.fw_domain[i].wake_count++; +mod_timer_pinned (&dev_priv->uncore.fw_domain[i].timer, + jiffies +1 ); + } + +if (fw_domains) + dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_domains); +} +#define __gen6_read (x) \ static u##x \ gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ GEN6_READ_HEADER(x); \ hsw_unclaimed_reg_debug(dev_priv, reg, true, true); \ - if (dev_priv->uncore.forcewake_count == 0 && \ - NEEDS_FORCE_WAKE((dev_priv), (reg))) { \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, \ - FORCEWAKE_ALL); \ - dev_priv->uncore.forcewake_count++; \ - mod_timer_pinned(&dev_priv->uncore.force_wake_timer, \ - jiffies + 1); \ - } \ + if (NEEDS_FORCE_WAKE((dev_priv), (reg))) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ val = __raw_i915_read##x(dev_priv, reg); \ hsw_unclaimed_reg_debug(dev_priv, reg, true, false); \ GEN6_READ_FOOTER; \@@ -599,45 +600,27 @@ gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \#define __vlv_read (x) \ static u##x \ vlv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ - unsigned fwengine = 0; \ GEN6_READ_HEADER(x); \ - if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \
drivers/gpu/drm/i915/intel_uncore.c gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ · line 615
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +4
-5- } \ -if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get (dev_priv, fwengine); \ +if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET (reg)) \ +__force_wake_get (dev_priv,FORCEWAKE_RENDER ); \ +else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET (reg)) \ +__force_wake_get (dev_priv,FORCEWAKE_MEDIA ); \ val = __raw_i915_read##x(dev_priv, reg); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \
This changes from immediate to delayed put - would it make sense to move it into a follow up patch?
Quoted 127 lines
drivers/gpu/drm/i915/intel_uncore.c +23
-48GEN6_READ_FOOTER ; \ }#define __chv_read (x) \ static u##x \ chv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ - unsigned fwengine = 0; \ GEN6_READ_HEADER(x); \ - if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \ - } else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine |= FORCEWAKE_RENDER; \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine |= FORCEWAKE_MEDIA; \ - } \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, fwengine); \ + if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_MEDIA); \ + else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, \ + FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \ val = __raw_i915_read##x(dev_priv, reg); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \ GEN6_READ_FOOTER; \ }@@ -766,17 +749,9 @@ static void \ gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \ GEN6_WRITE_HEADER; \ hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \ - if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) { \ - if (dev_priv->uncore.forcewake_count == 0) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, \ - FORCEWAKE_ALL); \ - __raw_i915_write##x(dev_priv, reg, val); \ - if (dev_priv->uncore.forcewake_count == 0) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, \ - FORCEWAKE_ALL); \ - } else { \ - __raw_i915_write##x(dev_priv, reg, val); \ - } \ + if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + __raw_i915_write##x(dev_priv, reg, val); \ hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \ hsw_unclaimed_reg_detect(dev_priv); \ GEN6_WRITE_FOOTER; \@@ -785,28 +760,17 @@ gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace#define __chv_write (x) \ static void \ chv_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \ - unsigned fwengine = 0; \ bool shadowed = is_gen8_shadowed(dev_priv, reg); \ GEN6_WRITE_HEADER; \ if (!shadowed) { \ - if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \ - } else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine |= FORCEWAKE_RENDER; \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine |= FORCEWAKE_MEDIA; \ - } \ + if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_MEDIA); \ + else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \ } \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, fwengine); \ __raw_i915_write##x(dev_priv, reg, val); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \ GEN6_WRITE_FOOTER; \ }@@ -837,18 +801,18 @@ __gen6_write(64)void intel_uncore_init (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; - -setup_timer (&dev_priv->uncore.force_wake_timer, - gen6_force_wake_timer, (unsigned long )dev_priv); +int i;intel_uncore_early_sanitize (dev, false);if (IS_VALLEYVIEW (dev)) { dev_priv->uncore.funcs.force_wake_get = __vlv_force_wake_get; dev_priv->uncore.funcs.force_wake_put = __vlv_force_wake_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER |FORCEWAKE_MEDIA ; }else if (IS_HASWELL (dev) ||IS_BROADWELL (dev)) { dev_priv->uncore.funcs.force_wake_get = __gen7_gt_force_wake_mt_get; dev_priv->uncore.funcs.force_wake_put = __gen7_gt_force_wake_mt_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; }else if (IS_IVYBRIDGE (dev)) {u32 ecobus;@@ -880,11 +844,22 @@ void intel_uncore_init(struct drm_device *dev) dev_priv->uncore.funcs.force_wake_put = __gen6_gt_force_wake_put; } + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; }else if (IS_GEN6 (dev)) { dev_priv->uncore.funcs.force_wake_get = __gen6_gt_force_wake_get; dev_priv->uncore.funcs.force_wake_put = __gen6_gt_force_wake_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; + } +
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_init(struct drm_device *dev) · line 865
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +7
-0+for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { + dev_priv->uncore.fw_domain[i].i915 = dev_priv; + dev_priv->uncore.fw_domain[i].id = i; + +setup_timer (&dev_priv->uncore.fw_domain[i].timer, + gen6_force_wake_timer, + (unsigned long )&dev_priv->uncore.fw_domain[i]); }switch (INTEL_INFO (dev)->gen) {
Regards,
Tvrtko
Re: [PATCH] drm/i915: Reduce duplicated forcewake logic
Quoted 11 lines · Tvrtko Ursulin
Quoted 8 lines · Chris Wilson
Introduce a structure to track the individual forcewake domains and use that to eliminated duplicate logic.
--- drivers/gpu/drm/i915/i915_debugfs.c | 41 +++--- drivers/gpu/drm/i915/i915_drv.h | 32 +++-- drivers/gpu/drm/i915/intel_uncore.c | 265 ++++++++++++++++-------------------- 3 files changed, 157 insertions(+), 181 deletions(-)
Quoted 1 line
As said yesterday I like the idea, some comments and questions inline.
Forgot about intel_lrc.c which looks like wouldn't compile after this patch. :)
Regards,
Tvrtko
Re: [PATCH] drm/i915: Reduce duplicated forcewake logic
In reply to Tvrtko UrsulinQuoted 5 lines · Tvrtko Ursulin
drivers/gpu/drm/i915/i915_drv.h struct intel_uncore_funcs { · line 547
Quoted 3 lines
drivers/gpu/drm/i915/i915_drv.h +2
-1-struct timer_list force_wake_timer; +struct intel_uncore_forcewake_domain { +struct drm_i915_private *i915;Should this be named
dev_privfor consistency?
Depends on whose tree you are looking at. ;-)
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH] drm/i915: Reduce duplicated forcewake logic
In reply to Chris WilsonQuoted 54 lines · Chris Wilson
Introduce a structure to track the individual forcewake domains and use that to eliminated duplicate logic.
--- drivers/gpu/drm/i915/i915_debugfs.c | 41 +++--- drivers/gpu/drm/i915/i915_drv.h | 32 +++-- drivers/gpu/drm/i915/intel_uncore.c | 265 ++++++++++++++++-------------------- 3 files changed, 157 insertions(+), 181 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-13@@ -1317,7 +1317,6 @@ static int vlv_drpc_info(struct seq_file *m)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;u32 rpmodectl1, rcctl1; -unsigned fw_rendercount =0 , fw_mediacount =0 ;intel_runtime_pm_get (dev_priv);@@ -1350,22 +1349,12 @@ static int vlv_drpc_info(struct seq_file *m)seq_printf (m,"Media RC6 residency since boot: %u\n" ,I915_READ (VLV_GT_MEDIA_RC6 )); -spin_lock_irq (&dev_priv->uncore.lock); - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; -spin_unlock_irq (&dev_priv->uncore.lock); - -seq_printf (m,"Forcewake Render Count = %u\n" , fw_rendercount); -seq_printf (m,"Forcewake Media Count = %u\n" , fw_mediacount); - - -return 0 ; +return i915_gen6_forcewake_count_info (m, data); }static int gen6_drpc_info (struct seq_file *m) { -struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;@@ -1379,7 +1368,7 @@ static int gen6_drpc_info(struct seq_file *m)intel_runtime_pm_get (dev_priv);spin_lock_irq (&dev_priv->uncore.lock); - forcewake_count = dev_priv->uncore.forcewake_count; + forcewake_count = dev_priv->uncore.fw_domain[FW_DOMAIN_RENDER ].wake_count;spin_unlock_irq (&dev_priv->uncore.lock);if (forcewake_count) {
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gen6_forcewake_count_info(struct seq_file *m, void *data) · line 1968
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -1976,21 +1965,23 @@ static int i915_gen6_forcewake_count_info(struct seq_file *m, void *data)struct drm_info_node *node = m->private;struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private; -unsigned forcewake_count =0 , fw_rendercount =0 , fw_mediacount =0 ; +const char *domain_names[] = {
static const char * const domain_names[] ?
Hmm. Maybe someone should fire up coccinelle to go through the entire codebase with this...
Quoted 296 lines
drivers/gpu/drm/i915/i915_debugfs.c +13
-11+"render" , + "media", + }; +int i;spin_lock_irq (&dev_priv->uncore.lock); -if (IS_VALLEYVIEW (dev)) { - fw_rendercount = dev_priv->uncore.fw_rendercount; - fw_mediacount = dev_priv->uncore.fw_mediacount; - }else - forcewake_count = dev_priv->uncore.forcewake_count; -spin_unlock_irq (&dev_priv->uncore.lock); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((dev_priv->uncore.fw_domains & (1 << i)) ==0 ) +continue ; -if (IS_VALLEYVIEW (dev)) { -seq_printf (m,"fw_rendercount = %u\n" , fw_rendercount); -seq_printf (m,"fw_mediacount = %u\n" , fw_mediacount); - }else -seq_printf (m,"forcewake count = %u\n" , forcewake_count); +seq_printf (m,"%s.wake_count = %u\n" , + domain_names[i], + dev_priv->uncore.fw_domain[i].wake_count); + } + +spin_unlock_irq (&dev_priv->uncore.lock);return 0 ; }drivers/gpu/drm/i915/i915_drv.h +21
-11@@ -528,18 +528,30 @@ struct intel_uncore_funcs { uint64_t val, bool trace); }; +enum { +FW_DOMAIN_RENDER =0 , +FW_DOMAIN_MEDIA , + +FW_DOMAIN_COUNT +}; +struct intel_uncore {spinlock_t lock;/** lock is also taken in irq contexts. */ struct intel_uncore_funcs funcs;unsigned fifo_count; -unsigned forcewake_count; - -unsigned fw_rendercount; -unsigned fw_mediacount; +unsigned fw_domains; -struct timer_list force_wake_timer; +struct intel_uncore_forcewake_domain { +struct drm_i915_private *i915; +int id; +unsigned wake_count; +struct timer_list timer; + } fw_domain[FW_DOMAIN_COUNT ]; +#define FORCEWAKE_RENDER (1 << FW_DOMAIN_RENDER) +#define FORCEWAKE_MEDIA (1 << FW_DOMAIN_MEDIA) +#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) };#define DEV_INFO_FOR_EACH_FLAG (func, sep) \@@ -2948,8 +2960,10 @@ extern void intel_display_print_error_state(struct drm_i915_error_state_buf *e,* must be set to prevent GT core from power down and stale values being * returned. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine); -void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine); +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains); +void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, +unsigned fw_domains);void assert_force_wake_inactive (struct drm_i915_private *dev_priv);int sandybridge_pcode_read (struct drm_i915_private *dev_priv,u8 mbox,u32 *val);@@ -2981,10 +2995,6 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val);int vlv_gpu_freq (struct drm_i915_private *dev_priv,int val);int vlv_freq_opcode (struct drm_i915_private *dev_priv,int val); -#define FORCEWAKE_RENDER (1 << 0) -#define FORCEWAKE_MEDIA (1 << 1) -#define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) -#define I915_READ8 (reg) dev_priv->uncore.funcs.mmio_readb(dev_priv, (reg), true)#define I915_WRITE8 (reg, val) dev_priv->uncore.funcs.mmio_writeb(dev_priv, (reg), (val), true)drivers/gpu/drm/i915/intel_uncore.c +41
-65@@ -73,7 +73,7 @@ static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv) }static void __gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {if (wait_for_atomic ((__raw_i915_read32 (dev_priv,FORCEWAKE_ACK ) &1 ) ==0 ,FORCEWAKE_ACK_TIMEOUT_MS ))@@ -99,7 +99,7 @@ static void __gen7_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv) }static void __gen7_gt_force_wake_mt_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {u32 forcewake_ack;@@ -136,7 +136,7 @@ static void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv) }static void __gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {__raw_i915_write32 (dev_priv,FORCEWAKE ,0 );/* something from same cacheline, but !FORCEWAKE */ @@ -145,7 +145,7 @@ static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, }static void __gen7_gt_force_wake_mt_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {__raw_i915_write32 (dev_priv,FORCEWAKE_MT ,_MASKED_BIT_DISABLE (FORCEWAKE_KERNEL ));@@ -194,7 +194,7 @@ static void vlv_force_wake_reset(struct drm_i915_private *dev_priv) }static void __vlv_force_wake_get (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) {/* Check for Render Engine */ if (FORCEWAKE_RENDER & fw_engine) {@@ -238,9 +238,8 @@ static void __vlv_force_wake_get(struct drm_i915_private *dev_priv, }static void __vlv_force_wake_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine) { -/* Check for Render Engine */ if (FORCEWAKE_RENDER & fw_engine) __raw_i915_write32(dev_priv, FORCEWAKE_VLV,@@ -258,59 +257,35 @@ static void __vlv_force_wake_put(struct drm_i915_private *dev_priv,gen6_gt_check_fifodbg (dev_priv); } -static void vlv_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) -{ -if (fw_engine &FORCEWAKE_RENDER && - dev_priv->uncore.fw_rendercount++ !=0 ) - fw_engine &= ~FORCEWAKE_RENDER ; -if (fw_engine &FORCEWAKE_MEDIA && - dev_priv->uncore.fw_mediacount++ !=0 ) - fw_engine &= ~FORCEWAKE_MEDIA ; - -if (fw_engine) - dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_engine); -} - -static void vlv_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) -{ -if (fw_engine &FORCEWAKE_RENDER ) { -WARN_ON (!dev_priv->uncore.fw_rendercount); -if (--dev_priv->uncore.fw_rendercount !=0 ) - fw_engine &= ~FORCEWAKE_RENDER ; - } - -if (fw_engine &FORCEWAKE_MEDIA ) { -WARN_ON (!dev_priv->uncore.fw_mediacount); -if (--dev_priv->uncore.fw_mediacount !=0 ) - fw_engine &= ~FORCEWAKE_MEDIA ; - } - -if (fw_engine) - dev_priv->uncore.funcs.force_wake_put (dev_priv, fw_engine); -} -static void gen6_force_wake_timer (unsigned long arg) { -struct drm_i915_private *dev_priv = (void *)arg; +struct intel_uncore_forcewake_domain *domain = (void *)arg;unsigned long irqflags; -assert_device_not_suspended (dev_priv); +assert_device_not_suspended (domain->i915); -spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -WARN_ON (!dev_priv->uncore.forcewake_count); +spin_lock_irqsave (&domain->i915->uncore.lock, irqflags); +WARN_ON (!domain->wake_count); -if (--dev_priv->uncore.forcewake_count ==0 ) - dev_priv->uncore.funcs.force_wake_put (dev_priv,FORCEWAKE_ALL ); -spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); +if (--domain->wake_count ==0 ) + domain->i915->uncore.funcs.force_wake_put (domain->i915, +1 << domain->id); +spin_unlock_irqrestore (&domain->i915->uncore.lock, irqflags); }void intel_uncore_forcewake_reset (struct drm_device *dev,bool restore) {struct drm_i915_private *dev_priv = dev->dev_private;unsigned long irqflags; +int i; -if (del_timer_sync (&dev_priv->uncore.force_wake_timer)) -gen6_force_wake_timer ((unsigned long )dev_priv); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++ ) { +if ((dev_priv->uncore.fw_domains & (1 << i)) ==0 ) +continue ; + +if (del_timer_sync (&dev_priv->uncore.fw_domain[i].timer)) +gen6_force_wake_timer ((unsigned long )&dev_priv->uncore.fw_domain[i]); + }/* Hold uncore.lock across reset to prevent any register access * with forcewake not set correctly @@ -327,18 +302,12 @@ void intel_uncore_forcewake_reset(struct drm_device *dev, bool restore)if (restore) {/* If reset with a user forcewake, try to restore */ unsigned fw =0 ; +int i; -if (IS_VALLEYVIEW (dev)) { -if (dev_priv->uncore.fw_rendercount) - fw |=FORCEWAKE_RENDER ; - -if (dev_priv->uncore.fw_mediacount) - fw |=FORCEWAKE_MEDIA ; - }else { -if (dev_priv->uncore.forcewake_count) - fw =FORCEWAKE_ALL ; +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if (dev_priv->uncore.fw_domain[i].wake_count) + fw |=1 << i; } -if (fw) dev_priv->uncore.funcs.force_wake_get (dev_priv, fw);@@ -389,50 +358,60 @@ void intel_uncore_sanitize(struct drm_device *dev)* be called at the beginning of the sequence followed by a call to * gen6_gt_force_wake_put() at the end of the sequence. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains) {unsigned long irqflags; +int i;if (!dev_priv->uncore.funcs.force_wake_get)return ;WARN_ON (!pm_runtime_active (&dev_priv->dev->pdev->dev)); + fw_domains &= dev_priv->uncore.fw_domains; +spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) { -vlv_force_wake_get (dev_priv, fw_engine); - }else { -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv, -FORCEWAKE_ALL ); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +if (dev_priv->uncore.fw_domain[i].wake_count++) + fw_domains &= ~(1 << i); } + +if (fw_domains) + dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_domains); +spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }/* * see gen6_gt_force_wake_get()
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_sanitize(struct drm_device *dev) · line 400
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +3
-1*/ -void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv,int fw_engine) +void gen6_gt_force_wake_put (struct drm_i915_private *dev_priv, +unsigned fw_domains) {unsigned long irqflags; +int i;if (!dev_priv->uncore.funcs.force_wake_put)return ;
Aren't we missing this
fw_domains &= dev_priv->uncore.fw_domains;
here? I would expect to see the WARN triggering below due to wake_count==0 when the domain doesn't exist.
Quoted 11 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-8spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) { -vlv_force_wake_put (dev_priv, fw_engine); - }else { -WARN_ON (!dev_priv->uncore.forcewake_count); -if (--dev_priv->uncore.forcewake_count ==0 ) { - dev_priv->uncore.forcewake_count++; - mod_timer_pinned(&dev_priv->uncore.force_wake_timer,
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_sanitize(struct drm_device *dev) · line 411
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +8
-2- jiffies + 1); - } +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +WARN_ON (!dev_priv->uncore.fw_domain[i].wake_count); +if (--dev_priv->uncore.fw_domain[i].wake_count) +continue ; +
Maybe WARN_ON(timer_pending()) here?
Quoted 31 lines
drivers/gpu/drm/i915/intel_uncore.c +13
-1+ dev_priv->uncore.fw_domain[i].wake_count++; +mod_timer_pinned (&dev_priv->uncore.fw_domain[i].timer, + jiffies +1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags);@@ -441,10 +420,13 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)void assert_force_wake_inactive (struct drm_i915_private *dev_priv) { +int i; +if (!dev_priv->uncore.funcs.force_wake_get)return ; -WARN_ON (dev_priv->uncore.forcewake_count >0 ); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) +WARN_ON (dev_priv->uncore.fw_domain[i].wake_count >0 ); }/* We give fast paths for the really cool registers */ @@ -578,19 +560,38 @@ __gen2_read(64)trace_i915_reg_rw (false, reg, val,sizeof (val), trace); \ return val +static inline void __force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains) +{ +int i; + +/* Ideally GCC would be constant-fold and eliminate this loop */
drivers/gpu/drm/i915/intel_uncore.c __gen2_read(64) · line 578
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +10
-0+ +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +if (dev_priv->uncore.fw_domain[i].wake_count) { + fw_domains &= ~(1 << i); +continue ; + } +
And here?
I think I had some concerns about the wake_count vs. timer handling when I saw the first version of this patch, but now that I look at this I can't see anything wrong with it.
Quoted 11 lines
drivers/gpu/drm/i915/intel_uncore.c +9
-0+ dev_priv->uncore.fw_domain[i].wake_count++; +mod_timer_pinned (&dev_priv->uncore.fw_domain[i].timer, + jiffies +1 ); + } + +if (fw_domains) + dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_domains); +} +#define __gen6_read (x) \ static u##x \
drivers/gpu/drm/i915/intel_uncore.c __gen2_read(64) · line 592
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-7gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ GEN6_READ_HEADER(x); \ hsw_unclaimed_reg_debug(dev_priv, reg, true, true); \ - if (dev_priv->uncore.forcewake_count == 0 && \ - NEEDS_FORCE_WAKE((dev_priv), (reg))) { \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, \ - FORCEWAKE_ALL); \ - dev_priv->uncore.forcewake_count++; \ - mod_timer_pinned(&dev_priv->uncore.force_wake_timer, \ - jiffies + 1); \
This seems to be against some other tree where the timer is already used for register accesses. That's not the case in nightly AFAICS.
I'm a bit worried about making the timer change at the same time as the code refactoring. I would suggest refactoring first, and tossing in another patch on top which expands the use of the timer to include register access. That way we could at least revert the timer part if necessary without losing the much cleaner code.
Also the code in nightly still has the runtime pm stuff in the forcewake code. I guess you cleaned those out in your tree, but I don't remeber seeing a recent patch on that front.
Otherwise this lookd very nice IMO, so would be great to get it in finally.
Quoted 171 lines
drivers/gpu/drm/i915/intel_uncore.c +36
-61- } \ +if (NEEDS_FORCE_WAKE ((dev_priv), (reg))) \ +__force_wake_get (dev_priv,FORCEWAKE_RENDER ); \ val = __raw_i915_read##x(dev_priv, reg); \ hsw_unclaimed_reg_debug(dev_priv, reg, true, false); \ GEN6_READ_FOOTER; \@@ -599,45 +600,27 @@ gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \#define __vlv_read (x) \ static u##x \ vlv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ - unsigned fwengine = 0; \ GEN6_READ_HEADER(x); \ - if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \ - } \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, fwengine); \ + if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_MEDIA); \ val = __raw_i915_read##x(dev_priv, reg); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \ GEN6_READ_FOOTER; \ }#define __chv_read (x) \ static u##x \ chv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \ - unsigned fwengine = 0; \ GEN6_READ_HEADER(x); \ - if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \ - } else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine |= FORCEWAKE_RENDER; \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine |= FORCEWAKE_MEDIA; \ - } \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, fwengine); \ + if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_MEDIA); \ + else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, \ + FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \ val = __raw_i915_read##x(dev_priv, reg); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \ GEN6_READ_FOOTER; \ }@@ -766,17 +749,9 @@ static void \ gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \ GEN6_WRITE_HEADER; \ hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \ - if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) { \ - if (dev_priv->uncore.forcewake_count == 0) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, \ - FORCEWAKE_ALL); \ - __raw_i915_write##x(dev_priv, reg, val); \ - if (dev_priv->uncore.forcewake_count == 0) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, \ - FORCEWAKE_ALL); \ - } else { \ - __raw_i915_write##x(dev_priv, reg, val); \ - } \ + if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + __raw_i915_write##x(dev_priv, reg, val); \ hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \ hsw_unclaimed_reg_detect(dev_priv); \ GEN6_WRITE_FOOTER; \@@ -785,28 +760,17 @@ gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace#define __chv_write (x) \ static void \ chv_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \ - unsigned fwengine = 0; \ bool shadowed = is_gen8_shadowed(dev_priv, reg); \ GEN6_WRITE_HEADER; \ if (!shadowed) { \ - if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine = FORCEWAKE_RENDER; \ - } else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine = FORCEWAKE_MEDIA; \ - } else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) { \ - if (dev_priv->uncore.fw_rendercount == 0) \ - fwengine |= FORCEWAKE_RENDER; \ - if (dev_priv->uncore.fw_mediacount == 0) \ - fwengine |= FORCEWAKE_MEDIA; \ - } \ + if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER); \ + else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_MEDIA); \ + else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \ + __force_wake_get(dev_priv, FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \ } \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_get(dev_priv, fwengine); \ __raw_i915_write##x(dev_priv, reg, val); \ - if (fwengine) \ - dev_priv->uncore.funcs.force_wake_put(dev_priv, fwengine); \ GEN6_WRITE_FOOTER; \ }@@ -837,18 +801,18 @@ __gen6_write(64)void intel_uncore_init (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; - -setup_timer (&dev_priv->uncore.force_wake_timer, - gen6_force_wake_timer, (unsigned long )dev_priv); +int i;intel_uncore_early_sanitize (dev, false);if (IS_VALLEYVIEW (dev)) { dev_priv->uncore.funcs.force_wake_get = __vlv_force_wake_get; dev_priv->uncore.funcs.force_wake_put = __vlv_force_wake_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER |FORCEWAKE_MEDIA ; }else if (IS_HASWELL (dev) ||IS_BROADWELL (dev)) { dev_priv->uncore.funcs.force_wake_get = __gen7_gt_force_wake_mt_get; dev_priv->uncore.funcs.force_wake_put = __gen7_gt_force_wake_mt_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; }else if (IS_IVYBRIDGE (dev)) {u32 ecobus;@@ -880,11 +844,22 @@ void intel_uncore_init(struct drm_device *dev) dev_priv->uncore.funcs.force_wake_put = __gen6_gt_force_wake_put; } + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; }else if (IS_GEN6 (dev)) { dev_priv->uncore.funcs.force_wake_get = __gen6_gt_force_wake_get; dev_priv->uncore.funcs.force_wake_put = __gen6_gt_force_wake_put; + dev_priv->uncore.fw_domains =FORCEWAKE_RENDER ; + } + +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { + dev_priv->uncore.fw_domain[i].i915 = dev_priv; + dev_priv->uncore.fw_domain[i].id = i; + +setup_timer (&dev_priv->uncore.fw_domain[i].timer, + gen6_force_wake_timer, + (unsigned long )&dev_priv->uncore.fw_domain[i]); }switch (INTEL_INFO (dev)->gen) {Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä Intel OTC _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH] drm/i915: Reduce duplicated forcewake logic
Quoted 8 lines · Ville Syrjälä
Quoted 7 lines · Chris Wilson
Introduce a structure to track the individual forcewake domains and use that to eliminated duplicate logic.
--- drivers/gpu/drm/i915/i915_debugfs.c | 41 +++--- drivers/gpu/drm/i915/i915_drv.h | 32 +++-- drivers/gpu/drm/i915/intel_uncore.c | 265 ++++++++++++++++-------------------- 3 files changed, 157 insertions(+), 181 deletions(-)
Hi Chris,
this looks useful -- I was looking at refactoring the VLV vs GEN9 versions of forcewake too. A few comments below:
Quoted 11 lines
drivers/gpu/drm/i915/intel_uncore.c static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, · line 148
Quoted 11 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-1@@ -145,7 +145,7 @@ static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, }static void __gen7_gt_force_wake_mt_put (struct drm_i915_private *dev_priv, -int fw_engine) +int fw_engine)
Here and in all the other versions: how about renaming the parameter to "fw_domains" in line with your new enum and struct naming? Thus making it clearer that it can have multiple bits set ...
Quoted 32 lines
Quoted 32 lines
drivers/gpu/drm/i915/intel_uncore.c +10
-8@@ -389,50 +358,60 @@ void intel_uncore_sanitize(struct drm_device *dev)* be called at the beginning of the sequence followed by a call to * gen6_gt_force_wake_put() at the end of the sequence. */ -void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv,int fw_engine) +void gen6_gt_force_wake_get (struct drm_i915_private *dev_priv, +unsigned fw_domains) {unsigned long irqflags; +int i;if (!dev_priv->uncore.funcs.force_wake_get)return ;WARN_ON (!pm_runtime_active (&dev_priv->dev->pdev->dev)); + fw_domains &= dev_priv->uncore.fw_domains; +spin_lock_irqsave (&dev_priv->uncore.lock, irqflags); -/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) { -vlv_force_wake_get (dev_priv, fw_engine); - }else { -if (dev_priv->uncore.forcewake_count++ ==0 ) - dev_priv->uncore.funcs.force_wake_get (dev_priv, -FORCEWAKE_ALL ); +for (i =0 ; i <FW_DOMAIN_COUNT ; i++) { +if ((fw_domains & (1 << i)) ==0 ) +continue ; + +if (dev_priv->uncore.fw_domain[i].wake_count++)
Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c void intel_uncore_sanitize(struct drm_device *dev) · line 388
Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c +5
-0+ fw_domains &= ~(1 << i); } + +if (fw_domains) + dev_priv->uncore.funcs.force_wake_get (dev_priv, fw_domains); +spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); }
Nice to get rid of the "Redirect to XXX specific routine" stuff -- one of the things I don't like about the current forcewake code is that the VLV and GEN9 versions of the outer wrapper are inconsistent w.r.t. how (and how many times) to call the ->force_wait_get() vfunc -- vlv collects all the bits and makes one call, gen9 makes one call for each bit separately. This code would handle all gens consistently, which is much nicer. And I prefer a single call with multiple bits, so that the low-level code, which knows where the corresponding h/w bits are, can deal efficiently with multiple bits mapping to the same register, or (for example) issue all the forcewake writes first and then poll them all for completion, rather than doing the write/poll pairs sequentially.
[snip]
Quoted 4 lines
Also the code in nightly still has the runtime pm stuff in the forcewake code. I guess you cleaned those out in your tree, but I don't remeber seeing a recent patch on that front.
Yes, this has been another area we wanted clarified and/or tidied up too
- I posted a message earlier today asking about whether the PM calls were necessary and a patch that allowed them to be avoided. But if they can be removed completely that would be better still :)
Quoted 2 lines
Otherwise this lookd very nice IMO, so would be great to get it in finally.
Agreed -
.Dave.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx