Runtime PM fixes
Subsystems (2)
cover Runtime PM fixes
From: Paulo Zanoni <[email protected]>
Hi
This series contains a collection of fixes for problems I found on the runtime PM code. All these problems either result in failure of some pm_pc8 test, or result in a Kernel WARN/DRM_ERROR when running pm_pc8, or there is a patch in this series that adds a WARN that will be triggered if we ever unfix the problem. Notice that nothing comes from bugzilla reports (see below).
So the real fixes of this series are patches 1-5. Patches 6+ either only fix debugfs files, or add more WARNs to make sure the problems we solve here remain solved, or do small cleanups.
I did _not_ add Cc: [email protected] tags to the patches for the following reasons reasons: (i) I simply don't know if we want them; (ii) I have other 44 patches that should be applied on top of these, so having part of the series in -fixes and another in -next can be a mess; and (iii) runtime PM is still disabled by default (although a simple sysfs write can enable it, and powertop makes this even easier) and its autosuspend delay is set to 10s, plus 5ms of PC8 timeout, so the problems are either very hard or impossible to reproduce without i915.pc8_timeout=0 (and that's why, I believe, we don't have bug reports for these things). But if we wan to Cc stable, it's fine too: we can do this with all the patches, or just with 1-5, or more.
Thanks,
Paulo
Chris Wilson (1):
drm/i915: Accurately track when we mark the hardware as idle/busy
Paulo Zanoni (10):
drm/i915: put runtime PM only at the end of intel_mark_idle drm/i915: put runtime PM only when we actually release force_wake drm/i915: get runtime PM at intel_set_mode
drm/i915: get runtime PM while trying to detect CRT drm/i915: get/put runtime PM in more places at i915_debugfs.c drm/i915: kill dev_priv->pc8.gpu_idle
drm/i915: call assert_device_not_suspended at gen6_force_wake_work drm/i915: assert force wake is disabled when we runtime suspend drm/i915: don't get/put runtime PM at the debugfs forcewake file drm/i915: assert we're not runtime suspended when writing registers
drivers/gpu/drm/i915/i915_debugfs.c | 21 +++++++++++++--- drivers/gpu/drm/i915/i915_drv.c | 1 + drivers/gpu/drm/i915/i915_drv.h | 19 ++++++++++----- drivers/gpu/drm/i915/intel_crt.c | 26 ++++++++++++++------ drivers/gpu/drm/i915/intel_display.c | 47 ++++++++++++++---------------------- drivers/gpu/drm/i915/intel_pm.c | 3 +-- drivers/gpu/drm/i915/intel_uncore.c | 34 ++++++++++++++++++-------- 7 files changed, 94 insertions(+), 57 deletions(-)
Signature · 1 line
1.8.5.3
PATCH 1/11 drm/i915: Accurately track when we mark the hardware as idle/busy
From: Chris Wilson <[email protected]>
We currently call intel_mark_idle() too often, as we do so as a side-effect of processing the request queue. However, we the calls to intel_mark_idle() are expected to be paired with a call to intel_mark_busy() (or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.
v2: From Paulo
- Make it compile
- Drop the __i915_add_request chunk
--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)
Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
drivers/gpu/drm/i915/i915_drv.h +8 -0
@@ -1124,6 +1124,14 @@ struct i915_gem_mm {*/ bool interruptible; +/** +* Is the GPU currently considered idle, or busy executing userspace +* requests? Whilst idle, we attempt to power down the hardware and +* display clocks. In order to reduce the effect on performance, there +* is a slight delay before we do so. +*/ +bool busy; +/** Bit 6 swizzling required for X tiling */ uint32_t bit_6_swizzle_x;/** Bit 6 swizzling required for Y tiling */
drivers/gpu/drm/i915/intel_display.c +9 -0
@@ -8192,8 +8192,12 @@ void intel_mark_busy(struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; +if (dev_priv->mm.busy) +return ; +hsw_package_c8_gpu_busy (dev_priv);i915_update_gfx_val (dev_priv); + dev_priv->mm.busy = true; }void intel_mark_idle (struct drm_device *dev)
@@ -8201,6 +8205,11 @@ void intel_mark_idle(struct drm_device *dev)struct drm_i915_private *dev_priv = dev->dev_private;struct drm_crtc *crtc; +if (!dev_priv->mm.busy) +return ; + + dev_priv->mm.busy = false; +hsw_package_c8_gpu_idle (dev_priv);if (!i915.powersave)
Signature · 1 line
1.8.5.3
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
Quoted 18 lines · Paulo Zanoni
From: Chris Wilson <[email protected]>
We currently call
intel_mark_idle()too often, as we do so as a side-effect of processing the request queue. However, we the calls tointel_mark_idle()are expected to be paired with a call tointel_mark_busy()(or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.v2: From Paulo
- Make it compile
- Drop the__i915_add_requestchunkReported-by Paulo Zanoni <[email protected]>Tested-by Paulo Zanoni <[email protected]>Signed-off-by Chris Wilson <[email protected]>Cc Paulo Zanoni <[email protected]>Signed-off-by Paulo Zanoni <[email protected]>
Quoted 8 lines
--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
Didn't see them... Why have you altered the logic? -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
2014-02-21 13:55 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 25 lines · Paulo Zanoni
Quoted 24 lines
From: Chris Wilson <[email protected]>
We currently call
intel_mark_idle()too often, as we do so as a side-effect of processing the request queue. However, we the calls tointel_mark_idle()are expected to be paired with a call tointel_mark_busy()(or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.v2: From Paulo
- Make it compile
- Drop the__i915_add_requestchunkReported-by Paulo Zanoni <[email protected]>Tested-by Paulo Zanoni <[email protected]>Signed-off-by Chris Wilson <[email protected]>Cc Paulo Zanoni <[email protected]>Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)
Quoted 4 lines
Quoted 2 lines
Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
Didn't see them... Why have you altered the logic?
See the comment at the __i915_add_request chunk:
http://lists.freedesktop.org/archives/intel-gfx/2014-February/040334.html
Maybe I just broke your patch :)
If my review doesn't make sense, we can stick to your version, it should do the job, and I can retest everything easily.
Quoted 4 lines
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
Signature · 1 line
Paulo Zanoni
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
Quoted 26 lines · Paulo Zanoni
2014-02-21 13:55 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 25 lines · Paulo Zanoni
Quoted 24 lines
From: Chris Wilson <[email protected]>
We currently call
intel_mark_idle()too often, as we do so as a side-effect of processing the request queue. However, we the calls tointel_mark_idle()are expected to be paired with a call tointel_mark_busy()(or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.v2: From Paulo
- Make it compile
- Drop the__i915_add_requestchunkReported-by Paulo Zanoni <[email protected]>Tested-by Paulo Zanoni <[email protected]>Signed-off-by Chris Wilson <[email protected]>Cc Paulo Zanoni <[email protected]>Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
Didn't see them... Why have you altered the logic?
See the comment at the
__i915_add_requestchunk:http://lists.freedesktop.org/archives/intel-gfx/2014-February/040334.html
Oh, I didn't look for comments inline.
Quoted 4 lines
Maybe I just broke your patch :)
If my review doesn't make sense, we can stick to your version, it should do the job, and I can retest everything easily.
If there was a pending work item, the call to intel_mark_busy() would return false. So we can revamp the logic around there a little bit. The reason for the change should be self-evident - the previous code lost its way in the transition to multiple rings arguing over a global property. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
2014-02-21 14:27 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 38 lines · Paulo Zanoni
Quoted 26 lines
2014-02-21 13:55 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 25 lines · Paulo Zanoni
Quoted 24 lines
From: Chris Wilson <[email protected]>
We currently call
intel_mark_idle()too often, as we do so as a side-effect of processing the request queue. However, we the calls tointel_mark_idle()are expected to be paired with a call tointel_mark_busy()(or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.v2: From Paulo
- Make it compile
- Drop the__i915_add_requestchunkReported-by Paulo Zanoni <[email protected]>Tested-by Paulo Zanoni <[email protected]>Signed-off-by Chris Wilson <[email protected]>Cc Paulo Zanoni <[email protected]>Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
Didn't see them... Why have you altered the logic?
See the comment at the
__i915_add_requestchunk:http://lists.freedesktop.org/archives/intel-gfx/2014-February/040334.html
Oh, I didn't look for comments inline.
Quoted 1 line
Quoted 8 lines
Quoted 3 lines
Maybe I just broke your patch :)
If my review doesn't make sense, we can stick to your version, it should do the job, and I can retest everything easily.If there was a pending work item, the call to
intel_mark_busy()would return false. So we can revamp the logic around there a little bit. The reason for the change should be self-evident - the previous code lost its way in the transition to multiple rings arguing over a global property
Just to avoid any possible confusions when/if we merge this series:
Chris sent a new version of this patch on the original mail thread.
Thanks,
Paulo
.
Quoted 4 lines
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
Signature · 1 line
Paulo Zanoni
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
Quoted 43 lines · Paulo Zanoni
2014-02-21 14:27 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 42 lines · Paulo Zanoni
Quoted 26 lines
2014-02-21 13:55 GMT-03:00 Chris Wilson <[email protected]>:
Quoted 25 lines · Paulo Zanoni
Quoted 24 lines
From: Chris Wilson <[email protected]>
We currently call
intel_mark_idle()too often, as we do so as a side-effect of processing the request queue. However, we the calls tointel_mark_idle()are expected to be paired with a call tointel_mark_busy()(or else we try to idle the hardware by accessing registers that are already disabled). Make the idle/busy tracking explicit to prevent the multiple calls.v2: From Paulo
- Make it compile
- Drop the__i915_add_requestchunkReported-by Paulo Zanoni <[email protected]>Tested-by Paulo Zanoni <[email protected]>Signed-off-by Chris Wilson <[email protected]>Cc Paulo Zanoni <[email protected]>Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 8 ++++++++ drivers/gpu/drm/i915/intel_display.c | 9 +++++++++ 2 files changed, 17 insertions(+)Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
Chris did not reply to my review comments yet, so I just went and implemented them. We need at least an ACK form him here before merging.
Didn't see them... Why have you altered the logic?
See the comment at the
__i915_add_requestchunk:http://lists.freedesktop.org/archives/intel-gfx/2014-February/040334.html
Oh, I didn't look for comments inline.
Quoted 4 lines
Maybe I just broke your patch :)
If my review doesn't make sense, we can stick to your version, it should do the job, and I can retest everything easily.
Quoted 7 lines
Quoted 4 lines
If there was a pending work item, the call to
intel_mark_busy()would return false. So we can revamp the logic around there a little bit. The reason for the change should be self-evident - the previous code lost its way in the transition to multiple rings arguing over a global propertyJust to avoid any possible confusions when/if we merge this series:
Chris sent a new version of this patch on the original mail thread.
Just to double check: Have I merged the right version? -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 01/11] drm/i915: Accurately track when we mark the hardware as idle/busy
Quoted 1 line · Daniel Vetter
Just to double check: Have I merged the right version?
No conflicts or residual with my tree, so yes.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
PATCH 2/11 drm/i915: put runtime PM only at the end of intel_mark_idle
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Because intel_mark_idle still touches some registers: it needs the machine to be awake. If you set both the autosuspend and PC8 delays to zero, you can get a "Device suspended" WARN when gen6_rps_idle touches registers.
This is not easy to reproduce, but happens once in a while when running pm_pc8.
Testcase: igt/pm_pc8
--- drivers/gpu/drm/i915/intel_display.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_display.c +4 -3
@@ -8210,10 +8210,8 @@ void intel_mark_idle(struct drm_device *dev) dev_priv->mm.busy = false; -hsw_package_c8_gpu_idle (dev_priv); -if (!i915.powersave) -return ; + goto out;list_for_each_entry (crtc, &dev->mode_config.crtc_list, head) {if (!crtc->fb)
@@ -8224,6 +8222,9 @@ void intel_mark_idle(struct drm_device *dev)if (INTEL_INFO (dev)->gen >=6 )gen6_rps_idle (dev->dev_private); + +out: +hsw_package_c8_gpu_idle (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,
Signature · 1 line
1.8.5.3
Re: [PATCH 02/11] drm/i915: put runtime PM only at the end of intel_mark_idle
Quoted 33 lines · On Fri, 21 Feb 2014 13:52:19 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
Because
intel_mark_idlestill touches some registers: it needs the machine to be awake. If you set both the autosuspend and PC8 delays to zero, you can get a "Device suspended" WARN whengen6_rps_idletouches registers.This is not easy to reproduce, but happens once in a while when running
pm_pc8.Testcase: igt/pm_pc8
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/intel_display.c +1
-3@@ -8210,10 +8210,8 @@ void intel_mark_idle(struct drm_device *dev) dev_priv->mm.busy = false; -hsw_package_c8_gpu_idle (dev_priv); -if (!i915.powersave) -return ; + goto out;list_for_each_entry (crtc, &dev->mode_config.crtc_list, head) {if (!crtc->fb)
drivers/gpu/drm/i915/intel_display.c void intel_mark_idle(struct drm_device *dev) · line 8230
Quoted 9 lines
drivers/gpu/drm/i915/intel_display.c +3
-0@@ -8224,6 +8222,9 @@ void intel_mark_idle(struct drm_device *dev)if (INTEL_INFO (dev)->gen >=6 )gen6_rps_idle (dev->dev_private); + +out: +hsw_package_c8_gpu_idle (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
Re: [PATCH 02/11] drm/i915: put runtime PM only at the end of intel_mark_idle
Quoted 44 lines · Jesse Barnes
Quoted 41 lines · On Fri, 21 Feb 2014 13:52:19 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
Because
intel_mark_idlestill touches some registers: it needs the machine to be awake. If you set both the autosuspend and PC8 delays to zero, you can get a "Device suspended" WARN whengen6_rps_idletouches registers.This is not easy to reproduce, but happens once in a while when running
pm_pc8.Testcase: igt/pm_pc8
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/intel_display.c +4
-3@@ -8210,10 +8210,8 @@ void intel_mark_idle(struct drm_device *dev) dev_priv->mm.busy = false; -hsw_package_c8_gpu_idle (dev_priv); -if (!i915.powersave) -return ; + goto out;list_for_each_entry (crtc, &dev->mode_config.crtc_list, head) {if (!crtc->fb)@@ -8224,6 +8222,9 @@ void intel_mark_idle(struct drm_device *dev)if (INTEL_INFO (dev)->gen >=6 )gen6_rps_idle (dev->dev_private); + +out: +hsw_package_c8_gpu_idle (dev_priv); }
Quoted 3 lines
drivers/gpu/drm/i915/intel_display.c void intel_mark_idle(struct drm_device *dev) · line 8230
Quoted 1 line
drivers/gpu/drm/i915/intel_display.c +0
-0void intel_mark_fb_busy(struct drm_i915_gem_object *obj,Reviewed-by Jesse Barnes <[email protected]>
Smells like another good reason for Jesse's idea with a register range based power domain checker. Especially with suspend delays we'll have a really hard time catching these kinds of bugs otherwise ... -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 3/11 drm/i915: put runtime PM only when we actually release force_wake
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
When we call gen6_gt_force_wake_put we don't actually put force_wake, we just schedule gen6_force_wake_work through mod_delayed_work, and that will eventually release force_wake.
The problem is that we call intel_runtime_pm_put directly at gen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend while force_wake is still supposed to be enabled if the graphics autosuspend_delay_ms is too small.
Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
--- drivers/gpu/drm/i915/intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_uncore.c +6 -1
@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)
@@ -393,6 +395,7 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;
@@ -405,13 +408,15 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true;mod_delayed_work (dev_priv->wq, &dev_priv->uncore.force_wake_work,1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); -intel_runtime_pm_put (dev_priv); +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */
Signature · 1 line
1.8.5.3
Re: [PATCH 03/11] drm/i915: put runtime PM only when we actually release force_wake
Quoted 52 lines · On Fri, 21 Feb 2014 13:52:20 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
When we call
gen6_gt_force_wake_putwe don't actually putforce_wake, we just schedulegen6_force_wake_workthroughmod_delayed_work, and that will eventually releaseforce_wake.The problem is that we call
intel_runtime_pm_putdirectly atgen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend whileforce_wakeis still supposed to be enabled if the graphicsautosuspend_delay_msis too small.Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_uncore.c +4
-0@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)@@ -393,6 +395,7 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;@@ -405,13 +408,15 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true; mod_delayed_work(dev_priv->wq, &dev_priv->uncore.force_wake_work,
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 422
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-11); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); -intel_runtime_pm_put (dev_priv); +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */
Do we need this for the VLV path too?
It's a little confusing that we do this delayed thing, incrementing the count and then decrementing again in the work queue, but what you have looks correct for both cases.
So with the VLV thing addressed:
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
Re: [PATCH 03/11] drm/i915: put runtime PM only when we actually release force_wake
2014-02-21 14:34 GMT-03:00 Jesse Barnes <[email protected]>:
Quoted 59 lines
Quoted 56 lines · On Fri, 21 Feb 2014 13:52:20 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
When we call
gen6_gt_force_wake_putwe don't actually putforce_wake, we just schedulegen6_force_wake_workthroughmod_delayed_work, and that will eventually releaseforce_wake.The problem is that we call
intel_runtime_pm_putdirectly atgen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend whileforce_wakeis still supposed to be enabled if the graphicsautosuspend_delay_msis too small.Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_uncore.c +4
-0@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)@@ -393,6 +395,7 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;@@ -405,13 +408,15 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true;mod_delayed_work (dev_priv->wq, &dev_priv->uncore.force_wake_work,1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags);
Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 422
Quoted 6 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-1-intel_runtime_pm_put (dev_priv); +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */ Do we need this for the VLV path too?
Yeah, my patch is wrong for VLV due to that "return". I'll send a new version.
By the way, why doesn't VLV use the delayed work queue? I would assume the work queue is there to improve performance somehow, so it could be a good idea to use it... And maybe try to avoid special-casing VLV would be good too :)
Quoted 11 lines
It's a little confusing that we do this delayed thing, incrementing the count and then decrementing again in the work queue, but what you have looks correct for both cases.
So with the VLV thing addressed:
Reviewed-by Jesse Barnes <[email protected]>--
Jesse Barnes, Intel Open Source Technology Center
Signature · 1 line
Paulo Zanoni
Re: [PATCH 03/11] drm/i915: put runtime PM only when we actually release force_wake
Quoted 69 lines · On Fri, 21 Feb 2014 17:08:50 -0300 Paulo Zanoni <[email protected]> wrote:
2014-02-21 14:34 GMT-03:00 Jesse Barnes <[email protected]>:
Quoted 59 lines
Quoted 56 lines · On Fri, 21 Feb 2014 13:52:20 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
When we call
gen6_gt_force_wake_putwe don't actually putforce_wake, we just schedulegen6_force_wake_workthroughmod_delayed_work, and that will eventually releaseforce_wake.The problem is that we call
intel_runtime_pm_putdirectly atgen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend whileforce_wakeis still supposed to be enabled if the graphicsautosuspend_delay_msis too small.Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_uncore.c +4
-0@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)@@ -393,6 +395,7 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;@@ -405,13 +408,15 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true;mod_delayed_work (dev_priv->wq, &dev_priv->uncore.force_wake_work,1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags);Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 422
Quoted 6 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-1-intel_runtime_pm_put (dev_priv); +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */ Do we need this for the VLV path too?
Quoted 6 lines
Yeah, my patch is wrong for VLV due to that "return". I'll send a new version.
By the way, why doesn't VLV use the delayed work queue? I would assume the work queue is there to improve performance somehow, so it could be a good idea to use it... And maybe try to avoid special-casing VLV would be good too :)
I don't know why VLV has an early return there rather than just using a function pointer like everything else. ISTR reviewing that patch from Deepak and suggesting something else, but I guess Daniel merged it anyway.
But yes, it should be fixed up as well and should probably use the delayed mechanism.
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
PATCH 3/11 drm/i915: put runtime PM only when we actually release force_wake
From: Paulo Zanoni <[email protected]>
When we call gen6_gt_force_wake_put we don't actually put force_wake, we just schedule gen6_force_wake_work through mod_delayed_work, and that will eventually release force_wake.
The problem is that we call intel_runtime_pm_put directly at gen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend while force_wake is still supposed to be enabled if the graphics autosuspend_delay_ms is too small.
Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
v2: - Keep the VLV refcounts balanced (Jesse)
--- drivers/gpu/drm/i915/intel_uncore.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_uncore.c +11 -3
@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)
@@ -393,25 +395,31 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;/* Redirect to VLV specific routine */ -if (IS_VALLEYVIEW (dev_priv->dev)) -return vlv_force_wake_put (dev_priv, fw_engine); +if (IS_VALLEYVIEW (dev_priv->dev)) { +vlv_force_wake_put (dev_priv, fw_engine); + goto out; + }spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true;mod_delayed_work (dev_priv->wq, &dev_priv->uncore.force_wake_work,1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags); -intel_runtime_pm_put (dev_priv); +out: +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */
Signature · 1 line
1.8.5.3
Re: [PATCH 03/11] drm/i915: put runtime PM only when we actually release force_wake
In reply to Jesse BarnesQuoted 79 lines · Jesse Barnes
Quoted 69 lines · On Fri, 21 Feb 2014 17:08:50 -0300 Paulo Zanoni <[email protected]> wrote:
2014-02-21 14:34 GMT-03:00 Jesse Barnes <[email protected]>:
Quoted 59 lines
Quoted 56 lines · On Fri, 21 Feb 2014 13:52:20 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
When we call
gen6_gt_force_wake_putwe don't actually putforce_wake, we just schedulegen6_force_wake_workthroughmod_delayed_work, and that will eventually releaseforce_wake.The problem is that we call
intel_runtime_pm_putdirectly atgen6_gt_force_wake_put, so most of the times we put our runtime PM reference before the delayed work happens, so we may runtime suspend whileforce_wakeis still supposed to be enabled if the graphicsautosuspend_delay_msis too small.Now the nice thing about the current code is that after it triggers the delayed work function it gets a refcount, and it only triggers the delayed work function if refcount is zero. This guarantees that when we schedule the funciton, it will run before we try to schedule it again, which simplifies the problem and allows for the current solution to work properly (hopefully!).
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_uncore.c +4
-0@@ -299,6 +299,8 @@ static void gen6_force_wake_work(struct work_struct *work)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); + +intel_runtime_pm_put (dev_priv); }static void intel_uncore_forcewake_reset (struct drm_device *dev)@@ -393,6 +395,7 @@ 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) {unsigned long irqflags; +bool delayed = false;if (!dev_priv->uncore.funcs.force_wake_put)return ;@@ -405,13 +408,15 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)spin_lock_irqsave (&dev_priv->uncore.lock, irqflags);if (--dev_priv->uncore.forcewake_count ==0 ) { dev_priv->uncore.forcewake_count++; + delayed = true;mod_delayed_work (dev_priv->wq, &dev_priv->uncore.force_wake_work,1 ); }spin_unlock_irqrestore (&dev_priv->uncore.lock, irqflags);Quoted 8 lines
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 422
Quoted 6 lines
drivers/gpu/drm/i915/intel_uncore.c +2
-1-intel_runtime_pm_put (dev_priv); +if (!delayed) +intel_runtime_pm_put (dev_priv); }/* We give fast paths for the really cool registers */ Do we need this for the VLV path too?
Quoted 6 lines
Yeah, my patch is wrong for VLV due to that "return". I'll send a new version.
By the way, why doesn't VLV use the delayed work queue? I would assume the work queue is there to improve performance somehow, so it could be a good idea to use it... And maybe try to avoid special-casing VLV would be good too :)
Quoted 7 lines
I don't know why VLV has an early return there rather than just using a function pointer like everything else. ISTR reviewing that patch from Deepak and suggesting something else, but I guess Daniel merged it anyway.
But yes, it should be fixed up as well and should probably use the delayed mechanism.
The delayed work (well timer since Chris' latest patch) can only handle one forcewake domain. vlv has two ...
That reminds me: This delayed suspend stuff is kinda implemented in the linux pm domain code already. Just needed to drop this since I've predicted that in the end we'll invent all these wheels, too ;-) -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 4/11 drm/i915: get runtime PM at intel_set_mode
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Otherwise, when we run intel_modeset_check_state we may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if your autosuspend_delay_ms is low (not the default 10s).
--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)
drivers/gpu/drm/i915/intel_display.c +5 -0
@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;int ret; +intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);return ret; }
Signature · 1 line
1.8.5.3
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
Quoted 25 lines · On Fri, 21 Feb 2014 13:52:21 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +3
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;int ret; +intel_runtime_pm_get (dev_priv);
drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9763
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +2
-0+ ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);return ret; }
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
In reply to Paulo ZanoniQuoted 24 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +2
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;int ret;
drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9762
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +3
-0+intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);return ret; }
Ideally these should be done as part of a power domain get/put as some platforms will need to turn on some power wells too and on that path we do anyway a runtime PM get/put.
In the latest VLV power domain support patchset [1] I added the power domain get/put and state check to places I thought necessary. I haven't tested it on HSW but afaics the ones added for the HW state readout code would solve the issue you describe here.
--Imre
[1] http://www.spinics.net/lists/intel-gfx/msg40344.html
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
2014-02-24 8:23 GMT-03:00 Imre Deak <[email protected]>:
Quoted 40 lines · Paulo Zanoni
Quoted 24 lines
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +2
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;int ret;drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9762
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +3
-0+intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);return ret; }Ideally these should be done as part of a power domain get/put as some platforms will need to turn on some power wells too and on that path we do anyway a runtime PM get/put.
Quoted 4 lines
In the latest VLV power domain support patchset [1] I added the power domain get/put and state check to places I thought necessary. I haven't tested it on HSW but afaics the ones added for the HW state readout code would solve the issue you describe here.
Yes. I just quickly read the patches, and they seem to try to solve this problem. Due to the reasons you wrote on the first paragraph, I think in the long term we want the power domains solution. But as I mentioned in the cover letter, this series contains bug fixes and maybe we want them on -fixes and even stable Kernels, so maybe we want to merge this patch, then later merge the code that uses power domains, then remove the runitme_pm_get calls and leave just the power domain calls? I'm not sure.
Quoted 4 lines
--Imre
Signature · 1 line
Paulo Zanoni
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
Quoted 46 lines · Paulo Zanoni
2014-02-24 8:23 GMT-03:00 Imre Deak <[email protected]>:
Quoted 40 lines · Paulo Zanoni
Quoted 24 lines
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +2
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;int ret;drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9762
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +3
-0+intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);return ret; }Ideally these should be done as part of a power domain get/put as some platforms will need to turn on some power wells too and on that path we do anyway a runtime PM get/put.
Quoted 4 lines
In the latest VLV power domain support patchset [1] I added the power domain get/put and state check to places I thought necessary. I haven't tested it on HSW but afaics the ones added for the HW state readout code would solve the issue you describe here.
Quoted 8 lines
Yes. I just quickly read the patches, and they seem to try to solve this problem. Due to the reasons you wrote on the first paragraph, I think in the long term we want the power domains solution. But as I mentioned in the cover letter, this series contains bug fixes and maybe we want them on -fixes and even stable Kernels, so maybe we want to merge this patch, then later merge the code that uses power domains, then remove the
runitme_pm_getcalls and leave just the power domain calls? I'm not sure.
Ok, I can rebase my patchset based on the above.
Note that runtime PM was enabled post 3.13, so this fix is needed for -fixes, but not for stable kernels.
The same goes for patch 5/11.
--Imre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
In reply to Paulo ZanoniQuoted 22 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +2
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;
drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9760
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +3
-0int ret; +intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);
I've thought our code has the relevant power domain checks sprinkled all over the get_config/state functions already? If that's not the case I prefer we fix that, similar to my suggestion in reply to Imre's patches of moving the power domain checking into the callbacks.
Wrt -fixes: Since the default autosuspend delay will make it impossible to hit this I think we can punt. Rules for late -rc and cc: stable is that it needs to be a real-world problem with a real bug report.
And one more: To make pm_pc8 more useful, could we just set the autosuspend delay to 0 while the test is running? Then restore it again with an igt atexit handler. That should help with our coverage and hitting this tiny little issues you've fixed in this series.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 04/11] drm/i915: get runtime PM at intel_set_mode
2014-03-05 10:25 GMT-03:00 Daniel Vetter <[email protected]>:
Quoted 39 lines · Paulo Zanoni
Quoted 22 lines
From: Paulo Zanoni <[email protected]>
Otherwise, when we run
intel_modeset_check_statewe may already be runtime suspended, and our state checking code will read registers while the device is suspended. This can only happen if yourautosuspend_delay_msis low (not the default 10s).Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_display.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/gpu/drm/i915/intel_display.c +2
-0@@ -9746,13 +9746,18 @@ static int intel_set_mode(struct drm_crtc *crtc,struct drm_display_mode *mode, int x, int y, struct drm_framebuffer *fb) { + struct drm_device *dev = crtc->dev; +struct drm_i915_private *dev_priv = dev->dev_private;drivers/gpu/drm/i915/intel_display.c static int intel_set_mode(struct drm_crtc *crtc, · line 9760
Quoted 10 lines
drivers/gpu/drm/i915/intel_display.c +3
-0int ret; +intel_runtime_pm_get (dev_priv); + ret =__intel_set_mode (crtc, mode, x, y, fb);if (ret ==0 )intel_modeset_check_state (crtc->dev); +intel_runtime_pm_put (dev_priv);I've thought our code has the relevant power domain checks sprinkled all over the get_config/state functions already? If that's not the case I prefer we fix that, similar to my suggestion in reply to Imre's patches of moving the power domain checking into the callbacks.
Quoted 8 lines
Wrt -fixes: Since the default autosuspend delay will make it impossible to hit this I think we can punt. Rules for late -rc and cc: stable is that it needs to be a real-world problem with a real bug report.
And one more: To make
pm_pc8more useful, could we just set the autosuspend delay to 0 while the test is running? Then restore it again with an igt atexit handler. That should help with our coverage and hitting this tiny little issues you've fixed in this series.
We already set autosuspend_delay_msg to zero when the test is running, we just don't restore the original value later. The problem is that we currently have autosuspend_delay_ms and i915.pc8_timeout as non-zero default, and we don't have support to change i915.pc8_timeout at runtime. If we merge the series that's currently under review (00/23 Merge PC8 with runtime PM, v2), we'll just kill i915.pc8_timeout, so we should be fine, and QA will automatically start running everything with zero timeouts.
Quoted 5 lines
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
Signature · 1 line
Paulo Zanoni
PATCH 5/11 drm/i915: get runtime PM while trying to detect CRT
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Otherwise we'll read registers that return 0xffffffff, trigger some WARNs, think CRT is actually connected (because certain bits are 1), and fail the drm-resources-equal testcase!
Tested on a SNB machine with runtime PM support (which is not upstream yet, but is already on my public tree at freedesktop.org, and will hopefully eventually become upstream).
Testcase: igt/pm_pc8/drm-resources-equal
--- drivers/gpu/drm/i915/intel_crt.c | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/intel_crt.c +19 -7
@@ -630,10 +630,13 @@ static enum drm_connector_status intel_crt_detect(struct drm_connector *connector, bool force) {struct drm_device *dev = connector->dev; +struct drm_i915_private *dev_priv = dev->dev_private;struct intel_crt *crt =intel_attached_crt (connector);enum drm_connector_status status;struct intel_load_detect_pipe tmp; +intel_runtime_pm_get (dev_priv); +DRM_DEBUG_KMS ("[CONNECTOR:%d:%s] force=%d\n" , connector->base.id,drm_get_connector_name (connector), force);
@@ -645,23 +648,30 @@ intel_crt_detect(struct drm_connector *connector, bool force)*/ if (intel_crt_detect_hotplug (connector)) {DRM_DEBUG_KMS ("CRT detected via hotplug\n" ); -return connector_status_connected; + status = connector_status_connected; + goto out; }else DRM_DEBUG_KMS ("CRT not detected via hotplug\n" ); } -if (intel_crt_detect_ddc (connector)) -return connector_status_connected; +if (intel_crt_detect_ddc (connector)) { + status = connector_status_connected; + goto out; + }/* Load detection is broken on HPD capable machines. Whoever wants a * broken monitor (without edid) to work behind a broken kvm (that fails * to have the right resistors for HP detection) needs to fix this up. * For now just bail out. */ -if (I915_HAS_HOTPLUG (dev)) -return connector_status_disconnected; +if (I915_HAS_HOTPLUG (dev)) { + status = connector_status_disconnected; + goto out; + } -if (!force) -return connector->status; +if (!force) { + status = connector->status; + goto out; + }/* for pre-945g platforms use load detect */ if (intel_get_load_detect_pipe (connector,NULL , &tmp)) {
@@ -673,6 +683,8 @@ intel_crt_detect(struct drm_connector *connector, bool force) }else status = connector_status_unknown; +out: +intel_runtime_pm_put (dev_priv);return status; }
Signature · 1 line
1.8.5.3
Re: [PATCH 05/11] drm/i915: get runtime PM while trying to detect CRT
Quoted 73 lines · On Fri, 21 Feb 2014 13:52:22 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
Otherwise we'll read registers that return
0xffffffff, trigger some WARNs, think CRT is actually connected (because certain bits are 1), and fail the drm-resources-equal testcase!Tested on a SNB machine with runtime PM support (which is not upstream yet, but is already on my public tree at freedesktop.org, and will hopefully eventually become upstream).
Testcase: igt/pm_pc8/drm-resources-equal
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_crt.c | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/intel_crt.c +17
-7@@ -630,10 +630,13 @@ static enum drm_connector_status intel_crt_detect(struct drm_connector *connector, bool force) {struct drm_device *dev = connector->dev; +struct drm_i915_private *dev_priv = dev->dev_private;struct intel_crt *crt =intel_attached_crt (connector);enum drm_connector_status status;struct intel_load_detect_pipe tmp; +intel_runtime_pm_get (dev_priv); +DRM_DEBUG_KMS ("[CONNECTOR:%d:%s] force=%d\n" , connector->base.id,drm_get_connector_name (connector), force);@@ -645,23 +648,30 @@ intel_crt_detect(struct drm_connector *connector, bool force)*/ if (intel_crt_detect_hotplug (connector)) {DRM_DEBUG_KMS ("CRT detected via hotplug\n" ); -return connector_status_connected; + status = connector_status_connected; + goto out; }else DRM_DEBUG_KMS ("CRT not detected via hotplug\n" ); } -if (intel_crt_detect_ddc (connector)) -return connector_status_connected; +if (intel_crt_detect_ddc (connector)) { + status = connector_status_connected; + goto out; + }/* Load detection is broken on HPD capable machines. Whoever wants a * broken monitor (without edid) to work behind a broken kvm (that fails * to have the right resistors for HP detection) needs to fix this up. * For now just bail out. */ -if (I915_HAS_HOTPLUG (dev)) -return connector_status_disconnected; +if (I915_HAS_HOTPLUG (dev)) { + status = connector_status_disconnected; + goto out; + } -if (!force) -return connector->status; +if (!force) { + status = connector->status; + goto out; + }/* for pre-945g platforms use load detect */ if (intel_get_load_detect_pipe (connector,NULL , &tmp)) {
drivers/gpu/drm/i915/intel_crt.c intel_crt_detect(struct drm_connector *connector, bool force) · line 690
Quoted 8 lines
drivers/gpu/drm/i915/intel_crt.c +2
-0@@ -673,6 +683,8 @@ intel_crt_detect(struct drm_connector *connector, bool force) }else status = connector_status_unknown; +out: +intel_runtime_pm_put (dev_priv);return status; }
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
Re: [PATCH 05/11] drm/i915: get runtime PM while trying to detect CRT
In reply to Paulo ZanoniQuoted 73 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Otherwise we'll read registers that return
0xffffffff, trigger some WARNs, think CRT is actually connected (because certain bits are 1), and fail the drm-resources-equal testcase!Tested on a SNB machine with runtime PM support (which is not upstream yet, but is already on my public tree at freedesktop.org, and will hopefully eventually become upstream).
Testcase: igt/pm_pc8/drm-resources-equal
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_crt.c | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/intel_crt.c +17
-7@@ -630,10 +630,13 @@ static enum drm_connector_status intel_crt_detect(struct drm_connector *connector, bool force) {struct drm_device *dev = connector->dev; +struct drm_i915_private *dev_priv = dev->dev_private;struct intel_crt *crt =intel_attached_crt (connector);enum drm_connector_status status;struct intel_load_detect_pipe tmp; +intel_runtime_pm_get (dev_priv); +DRM_DEBUG_KMS ("[CONNECTOR:%d:%s] force=%d\n" , connector->base.id,drm_get_connector_name (connector), force);@@ -645,23 +648,30 @@ intel_crt_detect(struct drm_connector *connector, bool force)*/ if (intel_crt_detect_hotplug (connector)) {DRM_DEBUG_KMS ("CRT detected via hotplug\n" ); -return connector_status_connected; + status = connector_status_connected; + goto out; }else DRM_DEBUG_KMS ("CRT not detected via hotplug\n" ); } -if (intel_crt_detect_ddc (connector)) -return connector_status_connected; +if (intel_crt_detect_ddc (connector)) { + status = connector_status_connected; + goto out; + }/* Load detection is broken on HPD capable machines. Whoever wants a * broken monitor (without edid) to work behind a broken kvm (that fails * to have the right resistors for HP detection) needs to fix this up. * For now just bail out. */ -if (I915_HAS_HOTPLUG (dev)) -return connector_status_disconnected; +if (I915_HAS_HOTPLUG (dev)) { + status = connector_status_disconnected; + goto out; + } -if (!force) -return connector->status; +if (!force) { + status = connector->status; + goto out; + }/* for pre-945g platforms use load detect */ if (intel_get_load_detect_pipe (connector,NULL , &tmp)) {
drivers/gpu/drm/i915/intel_crt.c intel_crt_detect(struct drm_connector *connector, bool force) · line 689
Quoted 7 lines
drivers/gpu/drm/i915/intel_crt.c +2
-0@@ -673,6 +683,8 @@ intel_crt_detect(struct drm_connector *connector, bool force) }else status = connector_status_unknown; +out: +intel_runtime_pm_put (dev_priv);return status; }
Similarly to 04/11, this should be part of a power domain get/put for a new CRT power domain I added in the VLV power domain support patches. On top of those to solve this issue we'd also need to add an enable/disable handler for the HSW always-on power well, which would only do a hsw_disable_package_c8()/hsw_enable_package_c8().
--Imre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 05/11] drm/i915: get runtime PM while trying to detect CRT
2014-02-24 8:33 GMT-03:00 Imre Deak <[email protected]>:
Quoted 77 lines · Paulo Zanoni
Quoted 76 lines
From: Paulo Zanoni <[email protected]>
Otherwise we'll read registers that return
0xffffffff, trigger some WARNs, think CRT is actually connected (because certain bits are 1), and fail the drm-resources-equal testcase!Tested on a SNB machine with runtime PM support (which is not upstream yet, but is already on my public tree at freedesktop.org, and will hopefully eventually become upstream).
Testcase: igt/pm_pc8/drm-resources-equal
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_crt.c | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/intel_crt.c +17
-7@@ -630,10 +630,13 @@ static enum drm_connector_status intel_crt_detect(struct drm_connector *connector, bool force) {struct drm_device *dev = connector->dev; +struct drm_i915_private *dev_priv = dev->dev_private;struct intel_crt *crt =intel_attached_crt (connector);enum drm_connector_status status;struct intel_load_detect_pipe tmp; +intel_runtime_pm_get (dev_priv); +DRM_DEBUG_KMS ("[CONNECTOR:%d:%s] force=%d\n" , connector->base.id,drm_get_connector_name (connector), force);@@ -645,23 +648,30 @@ intel_crt_detect(struct drm_connector *connector, bool force)*/ if (intel_crt_detect_hotplug (connector)) {DRM_DEBUG_KMS ("CRT detected via hotplug\n" ); -return connector_status_connected; + status = connector_status_connected; + goto out; }else DRM_DEBUG_KMS ("CRT not detected via hotplug\n" ); } -if (intel_crt_detect_ddc (connector)) -return connector_status_connected; +if (intel_crt_detect_ddc (connector)) { + status = connector_status_connected; + goto out; + }/* Load detection is broken on HPD capable machines. Whoever wants a * broken monitor (without edid) to work behind a broken kvm (that fails * to have the right resistors for HP detection) needs to fix this up. * For now just bail out. */ -if (I915_HAS_HOTPLUG (dev)) -return connector_status_disconnected; +if (I915_HAS_HOTPLUG (dev)) { + status = connector_status_disconnected; + goto out; + } -if (!force) -return connector->status; +if (!force) { + status = connector->status; + goto out; + }/* for pre-945g platforms use load detect */ if (intel_get_load_detect_pipe (connector,NULL , &tmp)) {@@ -673,6 +683,8 @@ intel_crt_detect(struct drm_connector *connector, bool force) }else status = connector_status_unknown;
Quoted 10 lines
drivers/gpu/drm/i915/intel_crt.c intel_crt_detect(struct drm_connector *connector, bool force) · line 689
Quoted 4 lines
drivers/gpu/drm/i915/intel_crt.c +2
-0+out: +intel_runtime_pm_put (dev_priv);return status; }Similarly to 04/11, this should be part of a power domain get/put for a new CRT power domain I added in the VLV power domain support patches. On top of those to solve this issue we'd also need to add an enable/disable handler for the HSW always-on power well, which would only do a hsw_disable_package_c8()/hsw_enable_package_c8().
Yes, I guess your series solves the same problem. But then, again, this is a bug fix, so the same discussion of patch 4/11 applies here.
Quoted 2 lines
--Imre
Signature · 1 line
Paulo Zanoni
PATCH 6/11 drm/i915: get/put runtime PM in more places at i915_debugfs.c
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
These are places where we read (not write) registers while we're runtime suspended.
--- drivers/gpu/drm/i915/i915_debugfs.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
drivers/gpu/drm/i915/i915_debugfs.c +17 -0
@@ -1348,6 +1348,8 @@ static int i915_fbc_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (intel_fbc_enabled (dev)) {seq_puts (m,"FBC enabled\n" ); }else {
@@ -1391,6 +1393,9 @@ static int i915_fbc_status(struct seq_file *m, void *unused) }seq_putc (m,'\n' ); } + +intel_runtime_pm_put (dev_priv); +return 0 ; }
@@ -1405,11 +1410,15 @@ static int i915_ips_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (IS_BROADWELL (dev) ||I915_READ (IPS_CTL ) &IPS_ENABLE )seq_puts (m,"enabled\n" );else seq_puts (m,"disabled\n" ); +intel_runtime_pm_put (dev_priv); +return 0 ; }
@@ -1420,6 +1429,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)drm_i915_private_t *dev_priv = dev->dev_private;bool sr_enabled = false; +intel_runtime_pm_get (dev_priv); +if (HAS_PCH_SPLIT (dev)) sr_enabled =I915_READ (WM1_LP_ILK ) &WM1_LP_SR_EN ;else if (IS_CRESTLINE (dev) ||IS_I945G (dev) ||IS_I945GM (dev))
@@ -1429,6 +1440,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)else if (IS_PINEVIEW (dev)) sr_enabled =I915_READ (DSPFW3 ) &PINEVIEW_SELF_REFRESH_EN ; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"self-refresh: %s\n" , sr_enabled ?"enabled" :"disabled" );
@@ -1972,12 +1985,16 @@ static int i915_energy_uJ(struct seq_file *m, void *data)if (INTEL_INFO (dev)->gen <6 )return -ENODEV ; +intel_runtime_pm_get (dev_priv); +rdmsrl (MSR_RAPL_POWER_UNIT , power); power = (power &0x1f00 ) >>8 ; units =1000000 / (1 << power);/* convert to uJ */ power =I915_READ (MCH_SECP_NRG_STTS ); power *= units; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"%llu" , (long long unsigned )power);return 0 ;
Signature · 1 line
1.8.5.3
Re: [PATCH 06/11] drm/i915: get/put runtime PM in more places at i915_debugfs.c
Quoted 74 lines · On Fri, 21 Feb 2014 13:52:23 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
These are places where we read (not write) registers while we're runtime suspended.
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)drivers/gpu/drm/i915/i915_debugfs.c +15
-0@@ -1348,6 +1348,8 @@ static int i915_fbc_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (intel_fbc_enabled (dev)) {seq_puts (m,"FBC enabled\n" ); }else {@@ -1391,6 +1393,9 @@ static int i915_fbc_status(struct seq_file *m, void *unused) }seq_putc (m,'\n' ); } + +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1405,11 +1410,15 @@ static int i915_ips_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (IS_BROADWELL (dev) ||I915_READ (IPS_CTL ) &IPS_ENABLE )seq_puts (m,"enabled\n" );else seq_puts (m,"disabled\n" ); +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1420,6 +1429,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)drm_i915_private_t *dev_priv = dev->dev_private;bool sr_enabled = false; +intel_runtime_pm_get (dev_priv); +if (HAS_PCH_SPLIT (dev)) sr_enabled =I915_READ (WM1_LP_ILK ) &WM1_LP_SR_EN ;else if (IS_CRESTLINE (dev) ||IS_I945G (dev) ||IS_I945GM (dev))@@ -1429,6 +1440,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)else if (IS_PINEVIEW (dev)) sr_enabled =I915_READ (DSPFW3 ) &PINEVIEW_SELF_REFRESH_EN ; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"self-refresh: %s\n" , sr_enabled ?"enabled" :"disabled" );@@ -1972,12 +1985,16 @@ static int i915_energy_uJ(struct seq_file *m, void *data)if (INTEL_INFO (dev)->gen <6 )return -ENODEV ; +intel_runtime_pm_get (dev_priv); +rdmsrl (MSR_RAPL_POWER_UNIT , power);
drivers/gpu/drm/i915/i915_debugfs.c static int i915_energy_uJ(struct seq_file *m, void *data) · line 2000
Quoted 10 lines
drivers/gpu/drm/i915/i915_debugfs.c +2
-0power = (power &0x1f00 ) >>8 ; units =1000000 / (1 << power);/* convert to uJ */ power =I915_READ (MCH_SECP_NRG_STTS ); power *= units; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"%llu" , (long long unsigned )power);return 0 ;
Looks like there are more places we need this too.. wonder if it would be better to put the get into some wrapper around our sysfs files...
But these bits look correct, if not sufficient, so:
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
Re: [PATCH 06/11] drm/i915: get/put runtime PM in more places at i915_debugfs.c
2014-02-21 14:41 GMT-03:00 Jesse Barnes <[email protected]>:
Quoted 82 lines
Quoted 79 lines · On Fri, 21 Feb 2014 13:52:23 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
These are places where we read (not write) registers while we're runtime suspended.
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)drivers/gpu/drm/i915/i915_debugfs.c +15
-0@@ -1348,6 +1348,8 @@ static int i915_fbc_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (intel_fbc_enabled (dev)) {seq_puts (m,"FBC enabled\n" ); }else {@@ -1391,6 +1393,9 @@ static int i915_fbc_status(struct seq_file *m, void *unused) }seq_putc (m,'\n' ); } + +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1405,11 +1410,15 @@ static int i915_ips_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (IS_BROADWELL (dev) ||I915_READ (IPS_CTL ) &IPS_ENABLE )seq_puts (m,"enabled\n" );else seq_puts (m,"disabled\n" ); +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1420,6 +1429,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)drm_i915_private_t *dev_priv = dev->dev_private;bool sr_enabled = false; +intel_runtime_pm_get (dev_priv); +if (HAS_PCH_SPLIT (dev)) sr_enabled =I915_READ (WM1_LP_ILK ) &WM1_LP_SR_EN ;else if (IS_CRESTLINE (dev) ||IS_I945G (dev) ||IS_I945GM (dev))@@ -1429,6 +1440,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)else if (IS_PINEVIEW (dev)) sr_enabled =I915_READ (DSPFW3 ) &PINEVIEW_SELF_REFRESH_EN ; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"self-refresh: %s\n" , sr_enabled ?"enabled" :"disabled" );@@ -1972,12 +1985,16 @@ static int i915_energy_uJ(struct seq_file *m, void *data)if (INTEL_INFO (dev)->gen <6 )return -ENODEV ; +intel_runtime_pm_get (dev_priv); +rdmsrl (MSR_RAPL_POWER_UNIT , power); power = (power &0x1f00 ) >>8 ; units =1000000 / (1 << power);/* convert to uJ */ power =I915_READ (MCH_SECP_NRG_STTS ); power *= units;
Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c static int i915_energy_uJ(struct seq_file *m, void *data) · line 2000
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +2
-0+intel_runtime_pm_put (dev_priv); +seq_printf (m,"%llu" , (long long unsigned )power);return 0 ;Looks like there are more places we need this too.. wonder if it would be better to put the get into some wrapper around our sysfs files...
Well, at least the deubgfs-read test from pm_pc8 doesn't complain on my HSW and SNB. But yeah, I thought about the wrapper too. I imagine people adding new debugfs files will always forget about the runtime PM refcounts, so maybe the wrapper is safer. But I could do this in a separate patch.
I also thought about wrapping the connector->detect functions.
Thanks for the reviews!
Quoted 7 lines
But these bits look correct, if not sufficient, so:
Reviewed-by Jesse Barnes <[email protected]>--
Jesse Barnes, Intel Open Source Technology Center
Signature · 1 line
Paulo Zanoni
Re: [PATCH 06/11] drm/i915: get/put runtime PM in more places at i915_debugfs.c
Quoted 92 lines · Paulo Zanoni
2014-02-21 14:41 GMT-03:00 Jesse Barnes <[email protected]>:
Quoted 82 lines
Quoted 79 lines · On Fri, 21 Feb 2014 13:52:23 -0300 Paulo Zanoni <[email protected]> wrote:
From: Paulo Zanoni <[email protected]>
These are places where we read (not write) registers while we're runtime suspended.
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)drivers/gpu/drm/i915/i915_debugfs.c +15
-0@@ -1348,6 +1348,8 @@ static int i915_fbc_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (intel_fbc_enabled (dev)) {seq_puts (m,"FBC enabled\n" ); }else {@@ -1391,6 +1393,9 @@ static int i915_fbc_status(struct seq_file *m, void *unused) }seq_putc (m,'\n' ); } + +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1405,11 +1410,15 @@ static int i915_ips_status(struct seq_file *m, void *unused)return 0 ; } +intel_runtime_pm_get (dev_priv); +if (IS_BROADWELL (dev) ||I915_READ (IPS_CTL ) &IPS_ENABLE )seq_puts (m,"enabled\n" );else seq_puts (m,"disabled\n" ); +intel_runtime_pm_put (dev_priv); +return 0 ; }@@ -1420,6 +1429,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)drm_i915_private_t *dev_priv = dev->dev_private;bool sr_enabled = false; +intel_runtime_pm_get (dev_priv); +if (HAS_PCH_SPLIT (dev)) sr_enabled =I915_READ (WM1_LP_ILK ) &WM1_LP_SR_EN ;else if (IS_CRESTLINE (dev) ||IS_I945G (dev) ||IS_I945GM (dev))@@ -1429,6 +1440,8 @@ static int i915_sr_status(struct seq_file *m, void *unused)else if (IS_PINEVIEW (dev)) sr_enabled =I915_READ (DSPFW3 ) &PINEVIEW_SELF_REFRESH_EN ; +intel_runtime_pm_put (dev_priv); +seq_printf (m,"self-refresh: %s\n" , sr_enabled ?"enabled" :"disabled" );@@ -1972,12 +1985,16 @@ static int i915_energy_uJ(struct seq_file *m, void *data)if (INTEL_INFO (dev)->gen <6 )return -ENODEV ; +intel_runtime_pm_get (dev_priv); +rdmsrl (MSR_RAPL_POWER_UNIT , power); power = (power &0x1f00 ) >>8 ; units =1000000 / (1 << power);/* convert to uJ */ power =I915_READ (MCH_SECP_NRG_STTS ); power *= units;Quoted 8 lines
drivers/gpu/drm/i915/i915_debugfs.c static int i915_energy_uJ(struct seq_file *m, void *data) · line 2000
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +2
-0+intel_runtime_pm_put (dev_priv); +seq_printf (m,"%llu" , (long long unsigned )power);return 0 ;Looks like there are more places we need this too.. wonder if it would be better to put the get into some wrapper around our sysfs files...
Quoted 7 lines
Well, at least the deubgfs-read test from
pm_pc8doesn't complain on my HSW and SNB. But yeah, I thought about the wrapper too. I imagine people adding new debugfs files will always forget about the runtime PM refcounts, so maybe the wrapper is safer. But I could do this in a separate patch.I also thought about wrapping the
connector->detectfunctions.
I prefer explicit get/put calls instead of magic wrappers. runtime pm get/put calls can be rather heavy-weight functions when we need to actually bring up a domain again and restore piles of registers. So being explicit is imo a feature.
Even more so with our explosion of power domains - I expect that some of the debugfs files need to grab multiple different domains all over the place.
-Daniel
Quoted 19 lines
Thanks for the reviews!
Quoted 7 lines
But these bits look correct, if not sufficient, so:
Reviewed-by Jesse Barnes <[email protected]>--
Jesse Barnes, Intel Open Source Technology CenterSignature · 5 lines
Paulo Zanoni _______________________________________________ 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
PATCH 7/11 drm/i915: kill dev_priv->pc8.gpu_idle
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Since the addition of dev_priv->mm.busy, there's no more need for dev_priv->pc8.gpu_idle, so kill it.
Notice that when you remove gpu_idle, hsw_package_c8_gpu_idle and hsw_package_c8_gpu_busy become identical to hsw_enable_package_c8 and hsw_disable_package_c8, so just use them.
Also, when we boot the machine, dev_priv->mm.busy initially considers the machine as idle. This is opposed to dev_priv->pc8.gpu_idle, which considered it busy. So dev_priv->pc8.disable_count has to be initalized to 1 now.
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 10 ++++------ drivers/gpu/drm/i915/intel_display.c | 30 ++---------------------------- drivers/gpu/drm/i915/intel_pm.c | 3 +-- 4 files changed, 8 insertions(+), 37 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +1 -1
@@ -2014,7 +2014,7 @@ static int i915_pc8_status(struct seq_file *m, void *unused)mutex_lock (&dev_priv->pc8.lock);seq_printf (m,"Requirements met: %s\n" ,yesno (dev_priv->pc8.requirements_met)); -seq_printf (m,"GPU idle: %s\n" ,yesno (dev_priv->pc8.gpu_idle)); +seq_printf (m,"GPU idle: %s\n" ,yesno (!dev_priv->mm.busy));seq_printf (m,"Disable count: %d\n" , dev_priv->pc8.disable_count);seq_printf (m,"IRQs disabled: %s\n" ,yesno (dev_priv->pc8.irqs_disabled));
drivers/gpu/drm/i915/i915_drv.h +4 -6
@@ -1321,11 +1321,10 @@ struct ilk_wm_values {* Ideally every piece of our code that needs PC8+ disabled would call * hsw_disable_package_c8, which would increment disable_count and prevent the * system from reaching PC8+. But we don't have a symmetric way to do this for -* everything, so we have the requirements_met and gpu_idle variables. When we -* switch requirements_met or gpu_idle to true we decrease disable_count, and -* increase it in the opposite case. The requirements_met variable is true when -* all the CRTCs, encoders and the power well are disabled. The gpu_idle -* variable is true when the GPU is idle. +* everything, so we have the requirements_met variable. When we switch +* requirements_met to true we decrease disable_count, and increase it in the +* opposite case. The requirements_met variable is true when all the CRTCs, +* encoders and the power well are disabled. * * In addition to everything, we only actually enable PC8+ if disable_count * stays at zero for at least some seconds. This is implemented with the
@@ -1348,7 +1347,6 @@ struct ilk_wm_values {*/ struct i915_package_c8 {bool requirements_met; -bool gpu_idle;bool irqs_disabled;/* Only true after the delayed work task actually enables it. */ bool enabled;
drivers/gpu/drm/i915/intel_display.c +2 -28
@@ -6812,32 +6812,6 @@ done:mutex_unlock (&dev_priv->pc8.lock); } -static void hsw_package_c8_gpu_idle (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (!dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = true; -__hsw_enable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} - -static void hsw_package_c8_gpu_busy (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = false; -__hsw_disable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} -#define for_each_power_domain (domain, mask) \ for ((domain) = 0; (domain) < POWER_DOMAIN_NUM; (domain)++) \ if ((1 << (domain)) & (mask))
@@ -8195,7 +8169,7 @@ void intel_mark_busy(struct drm_device *dev)if (dev_priv->mm.busy)return ; -hsw_package_c8_gpu_busy (dev_priv); +hsw_disable_package_c8 (dev_priv);i915_update_gfx_val (dev_priv); dev_priv->mm.busy = true; }
@@ -8224,7 +8198,7 @@ void intel_mark_idle(struct drm_device *dev)gen6_rps_idle (dev->dev_private); out: -hsw_package_c8_gpu_idle (dev_priv); +hsw_enable_package_c8 (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,
drivers/gpu/drm/i915/intel_pm.c +1 -2
@@ -5786,10 +5786,9 @@ void intel_pm_setup(struct drm_device *dev)mutex_init (&dev_priv->pc8.lock); dev_priv->pc8.requirements_met = false; - dev_priv->pc8.gpu_idle = false; dev_priv->pc8.irqs_disabled = false; dev_priv->pc8.enabled = false; - dev_priv->pc8.disable_count =2 ;/* requirements_met + gpu_idle */ + dev_priv->pc8.disable_count =1 ;/* requirements_met */ INIT_DELAYED_WORK (&dev_priv->pc8.enable_work, hsw_enable_pc8_work);INIT_DELAYED_WORK (&dev_priv->rps.delayed_resume_work, intel_gen6_powersave_work);
Signature · 1 line
1.8.5.3
Re: [PATCH 07/11] drm/i915: kill dev_priv->pc8.gpu_idle
Quoted 123 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Since the addition of
dev_priv->mm.busy, there's no more need fordev_priv->pc8.gpu_idle, so kill it.Notice that when you remove
gpu_idle,hsw_package_c8_gpu_idleandhsw_package_c8_gpu_busybecome identical tohsw_enable_package_c8andhsw_disable_package_c8, so just use them.Also, when we boot the machine,
dev_priv->mm.busyinitially considers the machine as idle. This is opposed todev_priv->pc8.gpu_idle, which considered it busy. Sodev_priv->pc8.disable_counthas to be initalized to 1 now.Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 10 ++++------ drivers/gpu/drm/i915/intel_display.c | 30 ++---------------------------- drivers/gpu/drm/i915/intel_pm.c | 3 +-- 4 files changed, 8 insertions(+), 37 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -2014,7 +2014,7 @@ static int i915_pc8_status(struct seq_file *m, void *unused)mutex_lock (&dev_priv->pc8.lock);seq_printf (m,"Requirements met: %s\n" ,yesno (dev_priv->pc8.requirements_met)); -seq_printf (m,"GPU idle: %s\n" ,yesno (dev_priv->pc8.gpu_idle)); +seq_printf (m,"GPU idle: %s\n" ,yesno (!dev_priv->mm.busy));seq_printf (m,"Disable count: %d\n" , dev_priv->pc8.disable_count);seq_printf (m,"IRQs disabled: %s\n" ,yesno (dev_priv->pc8.irqs_disabled));drivers/gpu/drm/i915/i915_drv.h +4
-6@@ -1321,11 +1321,10 @@ struct ilk_wm_values {* Ideally every piece of our code that needs PC8+ disabled would call * hsw_disable_package_c8, which would increment disable_count and prevent the * system from reaching PC8+. But we don't have a symmetric way to do this for -* everything, so we have the requirements_met and gpu_idle variables. When we -* switch requirements_met or gpu_idle to true we decrease disable_count, and -* increase it in the opposite case. The requirements_met variable is true when -* all the CRTCs, encoders and the power well are disabled. The gpu_idle -* variable is true when the GPU is idle. +* everything, so we have the requirements_met variable. When we switch +* requirements_met to true we decrease disable_count, and increase it in the +* opposite case. The requirements_met variable is true when all the CRTCs, +* encoders and the power well are disabled. * * In addition to everything, we only actually enable PC8+ if disable_count * stays at zero for at least some seconds. This is implemented with the @@ -1348,7 +1347,6 @@ struct ilk_wm_values {*/ struct i915_package_c8 {bool requirements_met; -bool gpu_idle;bool irqs_disabled;/* Only true after the delayed work task actually enables it. */ bool enabled;drivers/gpu/drm/i915/intel_display.c +2
-28@@ -6812,32 +6812,6 @@ done:mutex_unlock (&dev_priv->pc8.lock); } -static void hsw_package_c8_gpu_idle (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (!dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = true; -__hsw_enable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} - -static void hsw_package_c8_gpu_busy (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = false; -__hsw_disable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} -#define for_each_power_domain (domain, mask) \ for ((domain) = 0; (domain) < POWER_DOMAIN_NUM; (domain)++) \ if ((1 << (domain)) & (mask))@@ -8195,7 +8169,7 @@ void intel_mark_busy(struct drm_device *dev)if (dev_priv->mm.busy)return ; -hsw_package_c8_gpu_busy (dev_priv); +hsw_disable_package_c8 (dev_priv);i915_update_gfx_val (dev_priv); dev_priv->mm.busy = true; }@@ -8224,7 +8198,7 @@ void intel_mark_idle(struct drm_device *dev)gen6_rps_idle (dev->dev_private); out: -hsw_package_c8_gpu_idle (dev_priv); +hsw_enable_package_c8 (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,drivers/gpu/drm/i915/intel_pm.c +0
-0
drivers/gpu/drm/i915/intel_pm.c void intel_pm_setup(struct drm_device *dev) · line 5791
Quoted 8 lines
drivers/gpu/drm/i915/intel_pm.c +1
-2@@ -5786,10 +5786,9 @@ void intel_pm_setup(struct drm_device *dev)mutex_init (&dev_priv->pc8.lock); dev_priv->pc8.requirements_met = false; - dev_priv->pc8.gpu_idle = false; dev_priv->pc8.irqs_disabled = false; dev_priv->pc8.enabled = false; - dev_priv->pc8.disable_count =2 ;/* requirements_met + gpu_idle */ + dev_priv->pc8.disable_count =1 ;/* requirements_met */
This looks ok, but it's part of "Merge PC8 with runtime PM, v2" along with patch 1/11, so they can be skipped from this patchset.
--Imre
drivers/gpu/drm/i915/intel_pm.c void intel_pm_setup(struct drm_device *dev) · line 5794
Quoted 3 lines
drivers/gpu/drm/i915/intel_pm.c +0
-0INIT_DELAYED_WORK (&dev_priv->pc8.enable_work, hsw_enable_pc8_work);INIT_DELAYED_WORK (&dev_priv->rps.delayed_resume_work, intel_gen6_powersave_work);
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 07/11] drm/i915: kill dev_priv->pc8.gpu_idle
2014-02-28 10:50 GMT-03:00 Imre Deak <[email protected]>:
Quoted 125 lines · Paulo Zanoni
Quoted 124 lines
From: Paulo Zanoni <[email protected]>
Since the addition of
dev_priv->mm.busy, there's no more need fordev_priv->pc8.gpu_idle, so kill it.Notice that when you remove
gpu_idle,hsw_package_c8_gpu_idleandhsw_package_c8_gpu_busybecome identical tohsw_enable_package_c8andhsw_disable_package_c8, so just use them.Also, when we boot the machine,
dev_priv->mm.busyinitially considers the machine as idle. This is opposed todev_priv->pc8.gpu_idle, which considered it busy. Sodev_priv->pc8.disable_counthas to be initalized to 1 now.Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 10 ++++------ drivers/gpu/drm/i915/intel_display.c | 30 ++---------------------------- drivers/gpu/drm/i915/intel_pm.c | 3 +-- 4 files changed, 8 insertions(+), 37 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -2014,7 +2014,7 @@ static int i915_pc8_status(struct seq_file *m, void *unused)mutex_lock (&dev_priv->pc8.lock);seq_printf (m,"Requirements met: %s\n" ,yesno (dev_priv->pc8.requirements_met)); -seq_printf (m,"GPU idle: %s\n" ,yesno (dev_priv->pc8.gpu_idle)); +seq_printf (m,"GPU idle: %s\n" ,yesno (!dev_priv->mm.busy));seq_printf (m,"Disable count: %d\n" , dev_priv->pc8.disable_count);seq_printf (m,"IRQs disabled: %s\n" ,yesno (dev_priv->pc8.irqs_disabled));drivers/gpu/drm/i915/i915_drv.h +4
-6@@ -1321,11 +1321,10 @@ struct ilk_wm_values {* Ideally every piece of our code that needs PC8+ disabled would call * hsw_disable_package_c8, which would increment disable_count and prevent the * system from reaching PC8+. But we don't have a symmetric way to do this for -* everything, so we have the requirements_met and gpu_idle variables. When we -* switch requirements_met or gpu_idle to true we decrease disable_count, and -* increase it in the opposite case. The requirements_met variable is true when -* all the CRTCs, encoders and the power well are disabled. The gpu_idle -* variable is true when the GPU is idle. +* everything, so we have the requirements_met variable. When we switch +* requirements_met to true we decrease disable_count, and increase it in the +* opposite case. The requirements_met variable is true when all the CRTCs, +* encoders and the power well are disabled. * * In addition to everything, we only actually enable PC8+ if disable_count * stays at zero for at least some seconds. This is implemented with the @@ -1348,7 +1347,6 @@ struct ilk_wm_values {*/ struct i915_package_c8 {bool requirements_met; -bool gpu_idle;bool irqs_disabled;/* Only true after the delayed work task actually enables it. */ bool enabled;drivers/gpu/drm/i915/intel_display.c +2
-28@@ -6812,32 +6812,6 @@ done:mutex_unlock (&dev_priv->pc8.lock); } -static void hsw_package_c8_gpu_idle (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (!dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = true; -__hsw_enable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} - -static void hsw_package_c8_gpu_busy (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = false; -__hsw_disable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} -#define for_each_power_domain (domain, mask) \ for ((domain) = 0; (domain) < POWER_DOMAIN_NUM; (domain)++) \ if ((1 << (domain)) & (mask))@@ -8195,7 +8169,7 @@ void intel_mark_busy(struct drm_device *dev)if (dev_priv->mm.busy)return ; -hsw_package_c8_gpu_busy (dev_priv); +hsw_disable_package_c8 (dev_priv);i915_update_gfx_val (dev_priv); dev_priv->mm.busy = true; }@@ -8224,7 +8198,7 @@ void intel_mark_idle(struct drm_device *dev)gen6_rps_idle (dev->dev_private); out: -hsw_package_c8_gpu_idle (dev_priv); +hsw_enable_package_c8 (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,drivers/gpu/drm/i915/intel_pm.c +0
-0@@ -5786,10 +5786,9 @@ void intel_pm_setup(struct drm_device *dev)
Quoted 10 lines
drivers/gpu/drm/i915/intel_pm.c void intel_pm_setup(struct drm_device *dev) · line 5791
Quoted 7 lines
drivers/gpu/drm/i915/intel_pm.c +1
-2mutex_init (&dev_priv->pc8.lock); dev_priv->pc8.requirements_met = false; - dev_priv->pc8.gpu_idle = false; dev_priv->pc8.irqs_disabled = false; dev_priv->pc8.enabled = false; - dev_priv->pc8.disable_count =2 ;/* requirements_met + gpu_idle */ + dev_priv->pc8.disable_count =1 ;/* requirements_met */ This looks ok, but it's part of "Merge PC8 with runtime PM, v2" along with patch 1/11, so they can be skipped from this patchset.
Yes. Since you spotted some potential conflicts between this series and yours, I decided to send "Merge PC8 with runtime PM v2" in a way that it's independent form this series: it just contains the fixes that are necessary, avoiding some of the conflicts you detected. So we can just merge that series instead of this, if wanted.
Quoted 8 lines
--Imre
Quoted 3 lines
drivers/gpu/drm/i915/intel_pm.c +0
-0INIT_DELAYED_WORK (&dev_priv->pc8.enable_work, hsw_enable_pc8_work);INIT_DELAYED_WORK (&dev_priv->rps.delayed_resume_work, intel_gen6_powersave_work);
Signature · 1 line
Paulo Zanoni
Re: [PATCH 07/11] drm/i915: kill dev_priv->pc8.gpu_idle
Quoted 137 lines · Paulo Zanoni
2014-02-28 10:50 GMT-03:00 Imre Deak <[email protected]>:
Quoted 125 lines · Paulo Zanoni
Quoted 124 lines
From: Paulo Zanoni <[email protected]>
Since the addition of
dev_priv->mm.busy, there's no more need fordev_priv->pc8.gpu_idle, so kill it.Notice that when you remove
gpu_idle,hsw_package_c8_gpu_idleandhsw_package_c8_gpu_busybecome identical tohsw_enable_package_c8andhsw_disable_package_c8, so just use them.Also, when we boot the machine,
dev_priv->mm.busyinitially considers the machine as idle. This is opposed todev_priv->pc8.gpu_idle, which considered it busy. Sodev_priv->pc8.disable_counthas to be initalized to 1 now.Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 10 ++++------ drivers/gpu/drm/i915/intel_display.c | 30 ++---------------------------- drivers/gpu/drm/i915/intel_pm.c | 3 +-- 4 files changed, 8 insertions(+), 37 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -2014,7 +2014,7 @@ static int i915_pc8_status(struct seq_file *m, void *unused)mutex_lock (&dev_priv->pc8.lock);seq_printf (m,"Requirements met: %s\n" ,yesno (dev_priv->pc8.requirements_met)); -seq_printf (m,"GPU idle: %s\n" ,yesno (dev_priv->pc8.gpu_idle)); +seq_printf (m,"GPU idle: %s\n" ,yesno (!dev_priv->mm.busy));seq_printf (m,"Disable count: %d\n" , dev_priv->pc8.disable_count);seq_printf (m,"IRQs disabled: %s\n" ,yesno (dev_priv->pc8.irqs_disabled));drivers/gpu/drm/i915/i915_drv.h +4
-6@@ -1321,11 +1321,10 @@ struct ilk_wm_values {* Ideally every piece of our code that needs PC8+ disabled would call * hsw_disable_package_c8, which would increment disable_count and prevent the * system from reaching PC8+. But we don't have a symmetric way to do this for -* everything, so we have the requirements_met and gpu_idle variables. When we -* switch requirements_met or gpu_idle to true we decrease disable_count, and -* increase it in the opposite case. The requirements_met variable is true when -* all the CRTCs, encoders and the power well are disabled. The gpu_idle -* variable is true when the GPU is idle. +* everything, so we have the requirements_met variable. When we switch +* requirements_met to true we decrease disable_count, and increase it in the +* opposite case. The requirements_met variable is true when all the CRTCs, +* encoders and the power well are disabled. * * In addition to everything, we only actually enable PC8+ if disable_count * stays at zero for at least some seconds. This is implemented with the @@ -1348,7 +1347,6 @@ struct ilk_wm_values {*/ struct i915_package_c8 {bool requirements_met; -bool gpu_idle;bool irqs_disabled;/* Only true after the delayed work task actually enables it. */ bool enabled;drivers/gpu/drm/i915/intel_display.c +2
-28@@ -6812,32 +6812,6 @@ done:mutex_unlock (&dev_priv->pc8.lock); } -static void hsw_package_c8_gpu_idle (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (!dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = true; -__hsw_enable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} - -static void hsw_package_c8_gpu_busy (struct drm_i915_private *dev_priv) -{ -if (!HAS_PC8 (dev_priv->dev)) -return ; - -mutex_lock (&dev_priv->pc8.lock); -if (dev_priv->pc8.gpu_idle) { - dev_priv->pc8.gpu_idle = false; -__hsw_disable_package_c8 (dev_priv); - } -mutex_unlock (&dev_priv->pc8.lock); -} -#define for_each_power_domain (domain, mask) \ for ((domain) = 0; (domain) < POWER_DOMAIN_NUM; (domain)++) \ if ((1 << (domain)) & (mask))@@ -8195,7 +8169,7 @@ void intel_mark_busy(struct drm_device *dev)if (dev_priv->mm.busy)return ; -hsw_package_c8_gpu_busy (dev_priv); +hsw_disable_package_c8 (dev_priv);i915_update_gfx_val (dev_priv); dev_priv->mm.busy = true; }@@ -8224,7 +8198,7 @@ void intel_mark_idle(struct drm_device *dev)gen6_rps_idle (dev->dev_private); out: -hsw_package_c8_gpu_idle (dev_priv); +hsw_enable_package_c8 (dev_priv); } void intel_mark_fb_busy(struct drm_i915_gem_object *obj,drivers/gpu/drm/i915/intel_pm.c +0
-0@@ -5786,10 +5786,9 @@ void intel_pm_setup(struct drm_device *dev)Quoted 10 lines
drivers/gpu/drm/i915/intel_pm.c void intel_pm_setup(struct drm_device *dev) · line 5791
Quoted 7 lines
drivers/gpu/drm/i915/intel_pm.c +1
-2mutex_init (&dev_priv->pc8.lock); dev_priv->pc8.requirements_met = false; - dev_priv->pc8.gpu_idle = false; dev_priv->pc8.irqs_disabled = false; dev_priv->pc8.enabled = false; - dev_priv->pc8.disable_count =2 ;/* requirements_met + gpu_idle */ + dev_priv->pc8.disable_count =1 ;/* requirements_met */ This looks ok, but it's part of "Merge PC8 with runtime PM, v2" along with patch 1/11, so they can be skipped from this patchset.
Quoted 5 lines
Yes. Since you spotted some potential conflicts between this series and yours, I decided to send "Merge PC8 with runtime PM v2" in a way that it's independent form this series: it just contains the fixes that are necessary, avoiding some of the conflicts you detected. So we can just merge that series instead of this, if wanted.
I'm voting for potential conflicts be damned - this entire runtime pm enabling will keep on being ugly for a while, so I'm just trying to pull in as much as possible ...
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 8/11 drm/i915: call assert_device_not_suspended at gen6_force_wake_work
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Because we shouldn't be runtime suspended when forcewake is supposed to be enabled.
This commit will trigger the WARNs because we currently have a bug with this. The next patches should fix the bug.
--- drivers/gpu/drm/i915/intel_uncore.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/intel_uncore.c +8 -7
@@ -40,6 +40,12 @@#define __raw_posting_read (dev_priv__, reg__) (void)__raw_i915_read32(dev_priv__, reg__) +static void +assert_device_not_suspended (struct drm_i915_private *dev_priv) +{ +WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, +"Device suspended\n" ); +}static void __gen6_gt_wait_for_thread_c0 (struct drm_i915_private *dev_priv) {
@@ -295,6 +301,8 @@ static void gen6_force_wake_work(struct work_struct *work)container_of (work,typeof (*dev_priv), uncore.force_wake_work.work);unsigned long irqflags; +assert_device_not_suspended (dev_priv); +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 );
@@ -451,13 +459,6 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg) } } -static void -assert_device_not_suspended (struct drm_i915_private *dev_priv) -{ -WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, -"Device suspended\n" ); -} -#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \
Signature · 1 line
1.8.5.3
Re: [PATCH 08/11] drm/i915: call assert_device_not_suspended at gen6_force_wake_work
2014-02-21 13:52 GMT-03:00 Paulo Zanoni <[email protected]>:
Quoted 7 lines
From: Paulo Zanoni <[email protected]>
Because we shouldn't be runtime suspended when forcewake is supposed to be enabled.
This commit will trigger the WARNs because we currently have a bug with this. The next patches should fix the bug.
Actually I changed the patch order and forgot to remove the sentence above. Patch 03/11 should fix it.
Quoted 49 lines
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/intel_uncore.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/intel_uncore.c +8
-8@@ -40,6 +40,12 @@#define __raw_posting_read (dev_priv__, reg__) (void)__raw_i915_read32(dev_priv__, reg__) +static void +assert_device_not_suspended (struct drm_i915_private *dev_priv) +{ +WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, +"Device suspended\n" ); +}static void __gen6_gt_wait_for_thread_c0 (struct drm_i915_private *dev_priv) {@@ -295,6 +301,8 @@ static void gen6_force_wake_work(struct work_struct *work)container_of (work,typeof (*dev_priv), uncore.force_wake_work.work);unsigned long irqflags; +assert_device_not_suspended (dev_priv); +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 );@@ -451,13 +459,6 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg) } } -static void -assert_device_not_suspended (struct drm_i915_private *dev_priv) -{ -WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, -"Device suspended\n" ); -} -#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ --1.8.5.3
Signature · 1 line
Paulo Zanoni
Imre Deak
Reviewed-byRe: [PATCH 08/11] drm/i915: call assert_device_not_suspended at gen6_force_wake_work
Quoted 14 lines · Paulo Zanoni
2014-02-21 13:52 GMT-03:00 Paulo Zanoni <[email protected]>:
Quoted 7 lines
From: Paulo Zanoni <[email protected]>
Because we shouldn't be runtime suspended when forcewake is supposed to be enabled.
This commit will trigger the WARNs because we currently have a bug with this. The next patches should fix the bug.
Actually I changed the patch order and forgot to remove the sentence above. Patch 03/11 should fix it.
Quoted 2 lines
Signed-off-by Paulo Zanoni <[email protected]>
Patch 3/11 is now part of "Merge PC8 with runtime PM, v2" and this should come after that. With the order fixed:
Quoted 34 lines
Quoted 34 lines
--- drivers/gpu/drm/i915/intel_uncore.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/intel_uncore.c +8
-0@@ -40,6 +40,12 @@#define __raw_posting_read (dev_priv__, reg__) (void)__raw_i915_read32(dev_priv__, reg__) +static void +assert_device_not_suspended (struct drm_i915_private *dev_priv) +{ +WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, +"Device suspended\n" ); +}static void __gen6_gt_wait_for_thread_c0 (struct drm_i915_private *dev_priv) {@@ -295,6 +301,8 @@ static void gen6_force_wake_work(struct work_struct *work)container_of (work,typeof (*dev_priv), uncore.force_wake_work.work);unsigned long irqflags; +assert_device_not_suspended (dev_priv); +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 );@@ -451,13 +459,6 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg) } }
Quoted 16 lines
Quoted 11 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-8-static void -assert_device_not_suspended (struct drm_i915_private *dev_priv) -{ -WARN (HAS_RUNTIME_PM (dev_priv->dev) && dev_priv->pm.suspended, -"Device suspended\n" ); -} -#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ --Quoted 2 lines
1.8.5.3
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 9/11 drm/i915: assert force wake is disabled when we runtime suspend
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Just to be sure...
--- drivers/gpu/drm/i915/i915_drv.c | 1 + drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_uncore.c | 8 ++++++++ 3 files changed, 10 insertions(+)
drivers/gpu/drm/i915/i915_drv.c +1 -0
@@ -847,6 +847,7 @@ static int i915_runtime_suspend(struct device *device)struct drm_i915_private *dev_priv = dev->dev_private;WARN_ON (!HAS_RUNTIME_PM (dev)); +assert_force_wake_inactive (dev_priv);DRM_DEBUG_KMS ("Suspending device\n" );
drivers/gpu/drm/i915/i915_drv.h +1 -0
@@ -2604,6 +2604,7 @@ extern void intel_display_print_error_state(struct drm_i915_error_state_buf *e,*/ 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 assert_force_wake_inactive (struct drm_i915_private *dev_priv);int sandybridge_pcode_read (struct drm_i915_private *dev_priv,u8 mbox,u32 *val);int sandybridge_pcode_write (struct drm_i915_private *dev_priv,u8 mbox,u32 val);
drivers/gpu/drm/i915/intel_uncore.c +8 -0
@@ -427,6 +427,14 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)intel_runtime_pm_put (dev_priv); } +void assert_force_wake_inactive (struct drm_i915_private *dev_priv) +{ +if (!dev_priv->uncore.funcs.force_wake_get) +return ; + +WARN_ON (dev_priv->uncore.forcewake_count >0 ); +} +/* We give fast paths for the really cool registers */ #define NEEDS_FORCE_WAKE (dev_priv, reg) \ ((reg) < 0x40000 && (reg) != FORCEWAKE)
Signature · 1 line
1.8.5.3
Re: [PATCH 09/11] drm/i915: assert force wake is disabled when we runtime suspend
Quoted 41 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Just to be sure...
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.c | 1 + drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_uncore.c | 8 ++++++++ 3 files changed, 10 insertions(+)drivers/gpu/drm/i915/i915_drv.c +1
-0@@ -847,6 +847,7 @@ static int i915_runtime_suspend(struct device *device)struct drm_i915_private *dev_priv = dev->dev_private;WARN_ON (!HAS_RUNTIME_PM (dev)); +assert_force_wake_inactive (dev_priv);DRM_DEBUG_KMS ("Suspending device\n" );drivers/gpu/drm/i915/i915_drv.h +1
-0@@ -2604,6 +2604,7 @@ extern void intel_display_print_error_state(struct drm_i915_error_state_buf *e,*/ 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 assert_force_wake_inactive (struct drm_i915_private *dev_priv);int sandybridge_pcode_read (struct drm_i915_private *dev_priv,u8 mbox,u32 *val);int sandybridge_pcode_write (struct drm_i915_private *dev_priv,u8 mbox,u32 val);drivers/gpu/drm/i915/intel_uncore.c +0
-0@@ -427,6 +427,14 @@ void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine)intel_runtime_pm_put (dev_priv);
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 437
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +8
-0} +void assert_force_wake_inactive (struct drm_i915_private *dev_priv) +{ +if (!dev_priv->uncore.funcs.force_wake_get) +return ; + +WARN_ON (dev_priv->uncore.forcewake_count >0 ); +} +
As patch 8/11, this one would also trigger without 3/11, so it should be applied after that.
You could also check the VLV forcewake refcounts, but I'm ok if you add that later.
With the patch order fixed:
drivers/gpu/drm/i915/intel_uncore.c void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv, int fw_engine) · line 440
Quoted 3 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-0/* We give fast paths for the really cool registers */ #define NEEDS_FORCE_WAKE (dev_priv, reg) \ ((reg) < 0x40000 && (reg) != FORCEWAKE)
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 10/11 drm/i915: don't get/put runtime PM at the debugfs forcewake file
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Because gen6_gt_force_wake_{get,put} should already be responsible for getting/putting runtime PM.
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 -- 1 file changed, 2 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +0 -2
@@ -3597,7 +3597,6 @@ 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 ;
@@ -3612,7 +3611,6 @@ 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 ; }
Signature · 1 line
1.8.5.3
Re: [PATCH 10/11] drm/i915: don't get/put runtime PM at the debugfs forcewake file
Quoted 6 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Because gen6_gt_force_wake_{get,put} should already be responsible for getting/putting runtime PM.
Signed-off-by Paulo Zanoni <[email protected]>
Quoted 17 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 -- 1 file changed, 2 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +0
-1@@ -3597,7 +3597,6 @@ 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 3616
Quoted 7 lines
drivers/gpu/drm/i915/i915_debugfs.c +0
-1@@ -3612,7 +3611,6 @@ 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 ; }
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 10/11] drm/i915: don't get/put runtime PM at the debugfs forcewake file
In reply to Paulo ZanoniQuoted 15 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
Because gen6_gt_force_wake_{get,put} should already be responsible for getting/putting runtime PM.
Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 2 -- 1 file changed, 2 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +0
-0
drivers/gpu/drm/i915/i915_debugfs.c static int i915_forcewake_open(struct inode *inode, struct file *file) · line 3600
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +0
-1@@ -3597,7 +3597,6 @@ 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 );
I agree that doing the runtime get/put from within the forcewake get/put gives us nice nesting. The with that is that we do the forcewake get/put from within register I/O functions, which means it's not simple at all to figure out where all runtime_pm_get is called, which can lead to some big surprises around locking (since our resume functions do a _lot_ of work).
This is the reason why I've asked for the exception for the forcewake -> runtime pm nesting. We could fix this properly by moving the forcewake get/put out of the register read/write functions and sprinkle it explicitly over the code, but that's a _lot_ more work.
For now I'll punt on this patch until convinced otherwise. -Daniel
Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c +0
-1return 0 ;@@ -3612,7 +3611,6 @@ 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 ; }Signature · 6 lines
1.8.5.3 _______________________________________________ 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
PATCH 11/11 drm/i915: assert we're not runtime suspended when writing registers
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
I could swear this was already happening in the current code...
Also, put the reads and writes in a generic place, so we don't forget it again when we add runtime PM support to new platforms.
--- drivers/gpu/drm/i915/intel_uncore.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/intel_uncore.c +2 -2
@@ -470,6 +470,7 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg)#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)#define REG_READ_FOOTER \
@@ -568,6 +569,7 @@ __gen4_read(64)#define REG_WRITE_HEADER \ unsigned long irqflags; \ trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)#define REG_WRITE_FOOTER \
@@ -598,7 +600,6 @@ gen6_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool traceif (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \ gen6_gt_check_fifodbg(dev_priv); \
@@ -614,7 +615,6 @@ hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace)if (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \hsw_unclaimed_reg_clear (dev_priv, reg); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \
Signature · 1 line
1.8.5.3
Re: [PATCH 11/11] drm/i915: assert we're not runtime suspended when writing registers
Quoted 8 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
I could swear this was already happening in the current code...
Also, put the reads and writes in a generic place, so we don't forget it again when we add runtime PM support to new platforms.
Signed-off-by Paulo Zanoni <[email protected]>
In subject: s/writing/reading/ . Otherwise:
Quoted 33 lines
--- drivers/gpu/drm/i915/intel_uncore.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/gpu/drm/i915/intel_uncore.c +2
-1@@ -470,6 +470,7 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg)#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)#define REG_READ_FOOTER \@@ -568,6 +569,7 @@ __gen4_read(64)#define REG_WRITE_HEADER \ unsigned long irqflags; \ trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)#define REG_WRITE_FOOTER \@@ -598,7 +600,6 @@ gen6_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool traceif (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \ gen6_gt_check_fifodbg(dev_priv); \
drivers/gpu/drm/i915/intel_uncore.c hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) · line 620
Quoted 7 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-1@@ -614,7 +615,6 @@ hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace)if (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \hsw_unclaimed_reg_clear (dev_priv, reg); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 11/11] drm/i915: assert we're not runtime suspended when writing registers
Quoted 17 lines · Imre Deak
Quoted 8 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
I could swear this was already happening in the current code...
Also, put the reads and writes in a generic place, so we don't forget it again when we add runtime PM support to new platforms.
Signed-off-by Paulo Zanoni <[email protected]>In subject: s/writing/reading/ . Otherwise:
Reviewed-by Imre Deak <[email protected]>Quoted 4 lines
--- drivers/gpu/drm/i915/intel_uncore.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg) · line 474
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-0@@ -470,6 +470,7 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg)#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)
Hm, shouldn't we do this in the generic part of the register r/w functions? Anyway, I've merged all patches in this series to dinq with the exception of the two I've complained about.
Thanks, Daniel
Quoted 34 lines
Quoted 19 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-1#define REG_READ_FOOTER \@@ -568,6 +569,7 @@ __gen4_read(64)#define REG_WRITE_HEADER \ unsigned long irqflags; \ trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)#define REG_WRITE_FOOTER \@@ -598,7 +600,6 @@ gen6_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool traceif (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \ gen6_gt_check_fifodbg(dev_priv); \drivers/gpu/drm/i915/intel_uncore.c hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) · line 620
Quoted 7 lines
drivers/gpu/drm/i915/intel_uncore.c +0
-1@@ -614,7 +615,6 @@ hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace)if (NEEDS_FORCE_WAKE ((dev_priv), (reg))) { \ __fifo_ret =__gen6_gt_wait_for_fifo (dev_priv); \ } \ -assert_device_not_suspended (dev_priv); \hsw_unclaimed_reg_clear (dev_priv, reg); \ __raw_i915_write##x(dev_priv, reg, val); \ if (unlikely(__fifo_ret)) { \_______________________________________________
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
Re: [PATCH 11/11] drm/i915: assert we're not runtime suspended when writing registers
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Imre Deak
Quoted 8 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
I could swear this was already happening in the current code...
Also, put the reads and writes in a generic place, so we don't forget it again when we add runtime PM support to new platforms.
Signed-off-by Paulo Zanoni <[email protected]>In subject: s/writing/reading/ . Otherwise:
Reviewed-by Imre Deak <[email protected]>Quoted 4 lines
--- drivers/gpu/drm/i915/intel_uncore.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 14 lines
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg) · line 474
Quoted 10 lines
drivers/gpu/drm/i915/intel_uncore.c +1
-0@@ -470,6 +470,7 @@ hsw_unclaimed_reg_check(struct drm_i915_private *dev_priv, u32 reg)#define REG_READ_HEADER (x) \ unsigned long irqflags; \ u##x val = 0; \ + assert_device_not_suspended(dev_priv); \ spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)Hm, shouldn't we do this in the generic part of the register r/w functions? Anyway, I've merged all patches in this series to dinq with the exception of the two I've complained about.
Argh, failure at reading diffs. Your patch Does The Right Thing ;-)
Sorry for the noise ...
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch