drm/i915: use hrtimer in wait for vblank
Subsystems (2)
PATCH v2 1/1 drm/i915: use hrtimer in wait for vblank
BZ: 178761
In wait for vblank use usleep_range, which will use hrtimers instead of msleep. Using msleep(1~20) there are more chances of sleeping for 20ms. Using usleep_range uses hrtimers and hence are precise, worst case will trigger an interrupt at the higher/max timeout.
Change-log: On replacing msleep(1) with usleep_range(1000, 2000) we have noticed the time consumed by wait for vblank is ~4ms to ~17ms.
--- drivers/gpu/drm/i915/intel_display.c | 2 +- drivers/gpu/drm/i915/intel_dp.c | 4 ++-- drivers/gpu/drm/i915/intel_drv.h | 19 ++++++++++++------- 3 files changed, 15 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/intel_display.c +1 -1
@@ -761,7 +761,7 @@ static void g4x_wait_for_vblank(struct drm_device *dev, int pipe) frame =I915_READ (frame_reg); -if (wait_for (I915_READ_NOTRACE (frame_reg) != frame,50 )) +if (wait_for_us (I915_READ_NOTRACE (frame_reg) != frame,50 ,1000 ))DRM_DEBUG_KMS ("vblank wait timed out\n" ); }
drivers/gpu/drm/i915/intel_dp.c +2 -2
@@ -1074,7 +1074,7 @@ static void wait_panel_status(struct intel_dp *intel_dp,I915_READ (pp_stat_reg), I915_READ(pp_ctrl_reg)); -if (_wait_for ((I915_READ (pp_stat_reg) & mask) == value,5000 ,10 )) { +if (wait_for_ms ((I915_READ (pp_stat_reg) & mask) == value,5000 ,10 )) {DRM_ERROR ("Panel status timeout: status %08x control %08x\n" ,I915_READ (pp_stat_reg),I915_READ (pp_ctrl_reg));
@@ -1808,7 +1808,7 @@ void intel_edp_psr_disable(struct intel_dp *intel_dp) I915_READ(EDP_PSR_CTL(dev)) & ~EDP_PSR_ENABLE);/* Wait till PSR is idle */ -if (_wait_for ((I915_READ (EDP_PSR_STATUS_CTL (dev)) & +if (wait_for_ms ((I915_READ (EDP_PSR_STATUS_CTL (dev)) &EDP_PSR_STATUS_STATE_MASK ) ==0 ,2000 ,10 ))DRM_ERROR ("Timed out waiting for PSR Idle State\n" ); }
drivers/gpu/drm/i915/intel_drv.h +12 -7
@@ -42,8 +42,8 @@* having timed out, since the timeout could be due to preemption or similar and * we've never had a chance to check the condition before the timeout. */ -#define _wait_for (COND ,MS ,W ) ({ \ - unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1; \ +#define _wait_for (COND ,TIMEOUT ,MS ,US ) ({ \ + unsigned long timeout__ = jiffies + msecs_to_jiffies(TIMEOUT) + 1;\ int ret__ = 0; \ while (!(COND)) { \ if (time_after(jiffies, timeout__)) { \
@@ -51,8 +51,11 @@ ret__ = -ETIMEDOUT ; \break ; \ } \ -if (W &&drm_can_sleep ()) { \ -msleep (W ); \ +if ((MS |US ) &&drm_can_sleep ()) { \ +if (MS ) \ +msleep (MS ); \ +else \ +usleep_range (US ,US *2 ); \ }else { \cpu_relax (); \ } \
@@ -60,10 +63,12 @@ ret__; \ }) -#define wait_for (COND ,MS ) _wait_for(COND, MS, 1) -#define wait_for_atomic (COND ,MS ) _wait_for(COND, MS, 0) +#define wait_for (COND ,TIMEOUT ) _wait_for(COND, TIMEOUT, 1, 0) +#define wait_for_ms (COND ,TIMEOUT ,MS ) _wait_for(COND, TIMEOUT, MS, 0) +#define wait_for_us (COND ,TIMEOUT ,US ) _wait_for(COND, TIMEOUT, 0, US) +#define wait_for_atomic (COND ,TIMEOUT ) _wait_for(COND, TIMEOUT, 0, 0)#define wait_for_atomic_us (COND ,US ) _wait_for((COND), \ - DIV_ROUND_UP((US), 1000), 0) + DIV_ROUND_UP((US), 1000), 0, 0)#define KHz (x) (1000 * (x))#define MHz (x) KHz(1000 * (x))
Signature · 1 line
1.7.9.5
Quoted 7 lines · Arun R Murthy
BZ: 178761
In wait for vblank use
usleep_range, which will use hrtimers instead of msleep. Using msleep(1~20) there are more chances of sleeping for 20ms. Usingusleep_rangeuses hrtimers and hence are precise, worst case will trigger an interrupt at the higher/max timeout.
Quoted 5 lines
Change-log: On replacing msleep(1) with
usleep_range(1000, 2000)we have noticed the time consumed by wait for vblank is ~4ms to ~17ms.Change-Id I6672e5697b01987a6d069ab06e76d97287b1f7aeSigned-off-by Arun R Murthy <[email protected]>
Do we actually have that many vblank waits in time critical sections still? If we can't get rid of then and they need to be faster the right approach imo is to replace the wait loops with interrupt waits (with still a fallback timeout after 50ms or so of course) using our drm vblank infrastructure.
That will wake up at exactly the time we want to, without any unnecessary wakeups in between or other ugly overhead.
So as-is with the justification of optimizing vblank waits the wait_for_us optimization is nacked.
-Daniel
Quoted 89 lines
--- drivers/gpu/drm/i915/intel_display.c | 2 +- drivers/gpu/drm/i915/intel_dp.c | 4 ++-- drivers/gpu/drm/i915/intel_drv.h | 19 ++++++++++++------- 3 files changed, 15 insertions(+), 10 deletions(-)drivers/gpu/drm/i915/intel_display.c +1
-1@@ -761,7 +761,7 @@ static void g4x_wait_for_vblank(struct drm_device *dev, int pipe) frame =I915_READ (frame_reg); -if (wait_for (I915_READ_NOTRACE (frame_reg) != frame,50 )) +if (wait_for_us (I915_READ_NOTRACE (frame_reg) != frame,50 ,1000 ))DRM_DEBUG_KMS ("vblank wait timed out\n" ); }drivers/gpu/drm/i915/intel_dp.c +2
-2@@ -1074,7 +1074,7 @@ static void wait_panel_status(struct intel_dp *intel_dp,I915_READ (pp_stat_reg), I915_READ(pp_ctrl_reg)); -if (_wait_for ((I915_READ (pp_stat_reg) & mask) == value,5000 ,10 )) { +if (wait_for_ms ((I915_READ (pp_stat_reg) & mask) == value,5000 ,10 )) {DRM_ERROR ("Panel status timeout: status %08x control %08x\n" ,I915_READ (pp_stat_reg),I915_READ (pp_ctrl_reg));@@ -1808,7 +1808,7 @@ void intel_edp_psr_disable(struct intel_dp *intel_dp) I915_READ(EDP_PSR_CTL(dev)) & ~EDP_PSR_ENABLE);/* Wait till PSR is idle */ -if (_wait_for ((I915_READ (EDP_PSR_STATUS_CTL (dev)) & +if (wait_for_ms ((I915_READ (EDP_PSR_STATUS_CTL (dev)) &EDP_PSR_STATUS_STATE_MASK ) ==0 ,2000 ,10 ))DRM_ERROR ("Timed out waiting for PSR Idle State\n" ); }drivers/gpu/drm/i915/intel_drv.h +12
-7@@ -42,8 +42,8 @@* having timed out, since the timeout could be due to preemption or similar and * we've never had a chance to check the condition before the timeout. */ -#define _wait_for (COND ,MS ,W ) ({ \ - unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1; \ +#define _wait_for (COND ,TIMEOUT ,MS ,US ) ({ \ + unsigned long timeout__ = jiffies + msecs_to_jiffies(TIMEOUT) + 1;\ int ret__ = 0; \ while (!(COND)) { \ if (time_after(jiffies, timeout__)) { \@@ -51,8 +51,11 @@ ret__ = -ETIMEDOUT ; \break ; \ } \ -if (W &&drm_can_sleep ()) { \ -msleep (W ); \ +if ((MS |US ) &&drm_can_sleep ()) { \ +if (MS ) \ +msleep (MS ); \ +else \ +usleep_range (US ,US *2 ); \ }else { \cpu_relax (); \ } \@@ -60,10 +63,12 @@ ret__; \ }) -#define wait_for (COND ,MS ) _wait_for(COND, MS, 1) -#define wait_for_atomic (COND ,MS ) _wait_for(COND, MS, 0) +#define wait_for (COND ,TIMEOUT ) _wait_for(COND, TIMEOUT, 1, 0) +#define wait_for_ms (COND ,TIMEOUT ,MS ) _wait_for(COND, TIMEOUT, MS, 0) +#define wait_for_us (COND ,TIMEOUT ,US ) _wait_for(COND, TIMEOUT, 0, US) +#define wait_for_atomic (COND ,TIMEOUT ) _wait_for(COND, TIMEOUT, 0, 0)#define wait_for_atomic_us (COND ,US ) _wait_for((COND), \ - DIV_ROUND_UP((US), 1000), 0) + DIV_ROUND_UP((US), 1000), 0, 0)#define KHz (x) (1000 * (x))#define MHz (x) KHz(1000 * (x))Signature · 1 line
1.7.9.5
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch