drm/i915/vlv: Not reallocating VLV PCTX upon every suspend/resume
Subsystems (2)
PATCH 3/7 drm/i915/vlv: Not reallocating VLV PCTX upon every suspend/resume
From: Akash Goel <[email protected]>
VLV PCTX will come from stolen memory. Upon every suspend/resume cycle, this is being deallocated/reallocated. Given that PCTX has to be there at a constant stolen memory offset, doing this is not required. Also there is a chance of it getting corrupted, if this range gets used for some other allocation(on resume), which can result in GPU hangs.
--- drivers/gpu/drm/i915/i915_gem_stolen.c | 5 +++++ drivers/gpu/drm/i915/intel_pm.c | 9 ++++----- 2 files changed, 9 insertions(+), 5 deletions(-)
drivers/gpu/drm/i915/i915_gem_stolen.c +5 -0
@@ -213,6 +213,11 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)if (!drm_mm_initialized (&dev_priv->mm.stolen))return ; +if (dev_priv->vlv_pctx) { +drm_gem_object_unreference (&dev_priv->vlv_pctx->base); + dev_priv->vlv_pctx =NULL ; + } +i915_gem_stolen_cleanup_compression (dev);drm_mm_takedown (&dev_priv->mm.stolen); }
drivers/gpu/drm/i915/intel_pm.c +4 -5
@@ -3138,11 +3138,6 @@ static void valleyview_disable_rps(struct drm_device *dev)I915_WRITE (GEN6_RC_CONTROL ,0 );gen6_disable_rps_interrupts (dev); - -if (dev_priv->vlv_pctx) { -drm_gem_object_unreference (&dev_priv->vlv_pctx->base); - dev_priv->vlv_pctx =NULL ; - } }static void intel_print_rc6_info (struct drm_device *dev,u32 mode)
@@ -3508,6 +3503,10 @@ static void valleyview_setup_pctx(struct drm_device *dev)u32 pcbr;int pctx_size =24 *1024 ; +/* If PC Context is already there, then bail out*/ +if (dev_priv->vlv_pctx) +return ; + pcbr =I915_READ (VLV_PCBR );if (pcbr) {/* BIOS set it up already, grab the pre-alloc'd space */
Signature · 1 line
1.8.5.2
Quoted 19 lines · On Thu, Jan 09, 2014 at 11:00:42AM +0530, [email protected] wrote:
From: Akash Goel <[email protected]>
VLV PCTX will come from stolen memory. Upon every suspend/resume cycle, this is being deallocated/reallocated. Given that PCTX has to be there at a constant stolen memory offset, doing this is not required. Also there is a chance of it getting corrupted, if this range gets used for some other allocation(on resume), which can result in GPU hangs.
Signed-off-by Akash Goel <[email protected]>--- drivers/gpu/drm/i915/i915_gem_stolen.c | 5 +++++ drivers/gpu/drm/i915/intel_pm.c | 9 ++++----- 2 files changed, 9 insertions(+), 5 deletions(-)drivers/gpu/drm/i915/i915_gem_stolen.c +0
-0
drivers/gpu/drm/i915/i915_gem_stolen.c void i915_gem_cleanup_stolen(struct drm_device *dev) · line 219
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem_stolen.c +4
-0@@ -213,6 +213,11 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)if (!drm_mm_initialized (&dev_priv->mm.stolen))return ; +if (dev_priv->vlv_pctx) { +drm_gem_object_unreference (&dev_priv->vlv_pctx->base); + dev_priv->vlv_pctx =NULL ; + }
This should be extracted into a intel_pm_teardown to mirror intel_pm_setup. I know that we already cleanup the fbc allocations here, but for sane driver teardown code we should move all those deallocations out eventually and just have a WARN_ON(drm_mm_not_empty) in here before tearing down the stolen mem subsystem.
-Daniel
Quoted 38 lines
drivers/gpu/drm/i915/i915_gem_stolen.c +1
-0+i915_gem_stolen_cleanup_compression (dev);drm_mm_takedown (&dev_priv->mm.stolen); }drivers/gpu/drm/i915/intel_pm.c +4
-5@@ -3138,11 +3138,6 @@ static void valleyview_disable_rps(struct drm_device *dev)I915_WRITE (GEN6_RC_CONTROL ,0 );gen6_disable_rps_interrupts (dev); - -if (dev_priv->vlv_pctx) { -drm_gem_object_unreference (&dev_priv->vlv_pctx->base); - dev_priv->vlv_pctx =NULL ; - } }static void intel_print_rc6_info (struct drm_device *dev,u32 mode)@@ -3508,6 +3503,10 @@ static void valleyview_setup_pctx(struct drm_device *dev)u32 pcbr;int pctx_size =24 *1024 ; +/* If PC Context is already there, then bail out*/ +if (dev_priv->vlv_pctx) +return ; + pcbr =I915_READ (VLV_PCBR );if (pcbr) {/* BIOS set it up already, grab the pre-alloc'd space */ Signature · 6 lines
1.8.5.2 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Quoted 11 lines
drivers/gpu/drm/i915/i915_gem_stolen.c +4
-0@@ -213,6 +213,11 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)if (!drm_mm_initialized (&dev_priv->mm.stolen))return ; +if (dev_priv->vlv_pctx) { +drm_gem_object_unreference (&dev_priv->vlv_pctx->base); + dev_priv->vlv_pctx =NULL ; + }Quoted 2 lines
This should be extracted into a
intel_pm_teardownto mirrorintel_pm_setup. I know that we already cleanup the fbc allocations here, but for sane driver teardown code we should move all those deallocations out eventually and just have aWARN_ON(drm_mm_not_empty)in here before tearing down the stolen mem subsystem.
Actually as of now there isn't any such function intel_pm_teardown defined to mirror intel_pm_setup.
Although there is a intel_enable_gt_powersave & intel_disable_gt_powersave pair, but we couldn't use the intel_disable_gt_powersave for this, as we wanted to clean up the PCTX only at the time of driver unload. So the most appropriate place for us to do so was i915_gem_cleanup_stolen function.
Best Regards
Akash
-----Original Message-----
From: Daniel Vetter [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Thursday, January 09, 2014 12:34 PM
To: Goel, Akash
Cc: [email protected]
Subject: Re: [Intel-gfx] [PATCH 3/7] drm/i915/vlv: Not reallocating VLV PCTX upon every suspend/resume
Quoted 20 lines · On Thu, Jan 09, 2014 at 11:00:42AM +0530, [email protected] wrote:
From: Akash Goel <[email protected]>
VLV PCTX will come from stolen memory. Upon every suspend/resume cycle, this is being deallocated/reallocated. Given that PCTX has to be there at a constant stolen memory offset, doing this is not required. Also there is a chance of it getting corrupted, if this range gets used for some other allocation(on resume), which can result in GPU hangs.
Signed-off-by Akash Goel <[email protected]>--- drivers/gpu/drm/i915/i915_gem_stolen.c | 5 +++++ drivers/gpu/drm/i915/intel_pm.c | 9 ++++----- 2 files changed, 9 insertions(+), 5 deletions(-)drivers/gpu/drm/i915/i915_gem_stolen.c +0
-0
drivers/gpu/drm/i915/i915_gem_stolen.c void i915_gem_cleanup_stolen(struct drm_device *dev) · line 219
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem_stolen.c +4
-0@@ -213,6 +213,11 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)if (!drm_mm_initialized (&dev_priv->mm.stolen))return ; +if (dev_priv->vlv_pctx) { +drm_gem_object_unreference (&dev_priv->vlv_pctx->base); + dev_priv->vlv_pctx =NULL ; + }
This should be extracted into a intel_pm_teardown to mirror intel_pm_setup. I know that we already cleanup the fbc allocations here, but for sane driver teardown code we should move all those deallocations out eventually and just have a WARN_ON(drm_mm_not_empty) in here before tearing down the stolen mem subsystem. -Daniel
Quoted 34 lines
drivers/gpu/drm/i915/i915_gem_stolen.c +1
-0+i915_gem_stolen_cleanup_compression (dev);drm_mm_takedown (&dev_priv->mm.stolen); }(quoted diff) +0
-0
b/drivers/gpu/drm/i915/intel_pm.cindex 469170c..369cc73 100644drivers/gpu/drm/i915/intel_pm.c +4
-7@@ -3138,11 +3138,6 @@ static void valleyview_disable_rps(struct drm_device *dev)I915_WRITE (GEN6_RC_CONTROL ,0 );gen6_disable_rps_interrupts (dev); - -if (dev_priv->vlv_pctx) { -drm_gem_object_unreference (&dev_priv->vlv_pctx->base); - dev_priv->vlv_pctx =NULL ; - } } static void intel_print_rc6_info(struct drm_device *dev, u32 mode) @@ -3508,6 +3503,10 @@ static void valleyview_setup_pctx(struct drm_device *dev) u32 pcbr; int pctx_size = 24*1024; +/* If PC Context is already there, then bail out*/ +if (dev_priv->vlv_pctx) +return ; + pcbr =I915_READ (VLV_PCBR );if (pcbr) {/* BIOS set it up already, grab the pre-alloc'd space */ --1.8.5.2
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch