2013/12/20 Jani Nikula <jani.nikula@xxxxxxxxxxxxxxx>: > On Thu, 19 Dec 2013, Paulo Zanoni <przanoni@xxxxxxxxx> wrote: >> From: Paulo Zanoni <paulo.r.zanoni@xxxxxxxxx> >> >> The eDP spec defines some points where after you do action A, you have >> to wait some time before action B. The thing is that in our driver >> action B does not happen exactly after action A, but we still use >> msleep() calls directly. What this patch does is that we record the >> timestamp of when action A happened, then, just before action B, we >> look at how much time has passed and only sleep the remaining amount >> needed. >> >> With this change, I am able to save about 5-20ms (out of the total >> 200ms) of the backlight_off delay and completely skip the 1ms >> backlight_on delay. The 600ms vdd_off delay doesn't happen during >> normal usage anymore due to a previous patch. >> >> v2: - Rename ironlake_wait_jiffies_delay to intel_wait_until_after and >> move it to intel_display.c >> - Fix the msleep call: diff is in jiffies >> v3: - Use "tmp_jiffies" so we don't need to worry about the value of >> "jiffies" advancing while we're doing the math. >> v4: - Rename function again. >> - Move function to i915_drv.h. >> - Store last_power_cycle at edp_panel_off too. >> - Use msecs_to_jiffies_timeout, then replace the msleep with an >> open-coded version that avoids the extra +1 jiffy. >> - Try to add units to every variable name so we don't confuse >> jiffies with milliseconds. >> >> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@xxxxxxxxx> >> --- >> drivers/gpu/drm/i915/i915_drv.h | 29 +++++++++++++++++++++++++++++ >> drivers/gpu/drm/i915/intel_dp.c | 27 ++++++++++++++++++++++++--- >> drivers/gpu/drm/i915/intel_drv.h | 3 +++ >> 3 files changed, 56 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h >> index cc8afff..7e9b436 100644 >> --- a/drivers/gpu/drm/i915/i915_drv.h >> +++ b/drivers/gpu/drm/i915/i915_drv.h >> @@ -2636,4 +2636,33 @@ timespec_to_jiffies_timeout(const struct timespec *value) >> return min_t(unsigned long, MAX_JIFFY_OFFSET, j + 1); >> } >> >> +/* >> + * If you need to wait X milliseconds between events A and B, but event B >> + * doesn't happen exactly after event A, you record the timestamp (jiffies) of >> + * when event A happened, then just before event B you call this function and >> + * pass the timestamp as the first argument, and X as the second argument. >> + */ >> +static inline void >> +wait_remaining_ms_from_jiffies(unsigned long timestamp_jiffies, int to_wait_ms) >> +{ >> + unsigned long target_jiffies, tmp_jiffies; >> + unsigned int remaining_ms; >> + >> + /* >> + * Don't re-read the value of "jiffies" every time since it may change >> + * behind our back and break the math. >> + */ >> + tmp_jiffies = jiffies; >> + target_jiffies = timestamp_jiffies + >> + msecs_to_jiffies_timeout(to_wait_ms); > > Is it possible this gets called before the last_* timestamp fields are > initialized, resulting to timestamp_jiffies being 0? Do we need to > special case that? If not, perhaps add a WARN_ON(timestamp_jiffies == 0) > so we maintain this precondition. The jiffies variable keeps running forever, it can be 0 at any time. In the worst case we will wait more than we should, and this case is really rare, so it's fine IMHO. And when booting, jiffies usually has a super-high value. > >> + >> + if (time_after(target_jiffies, tmp_jiffies)) { >> + remaining_ms = jiffies_to_msecs((long)target_jiffies - >> + (long)tmp_jiffies); >> + while (remaining_ms) >> + remaining_ms = >> + schedule_timeout_uninterruptible(remaining_ms); > > schedule_timeout_uninterruptible takes jiffies, not ms. I think you > should do the measurements again after fixing this. Thanks for catching it! I have CONFIG_HZ=1000, so I guess 1ms == 1 jiffy here. > >> + } >> +} >> + >> #endif >> diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c >> index 9d96447..2f82af4 100644 >> --- a/drivers/gpu/drm/i915/intel_dp.c >> +++ b/drivers/gpu/drm/i915/intel_dp.c >> @@ -1057,9 +1057,26 @@ static void ironlake_wait_panel_off(struct intel_dp *intel_dp) >> static void ironlake_wait_panel_power_cycle(struct intel_dp *intel_dp) >> { >> DRM_DEBUG_KMS("Wait for panel power cycle\n"); >> + >> + /* When we disable the VDD override bit last we have to do the manual >> + * wait. */ >> + wait_remaining_ms_from_jiffies(intel_dp->last_power_cycle, >> + intel_dp->panel_power_cycle_delay); >> + >> ironlake_wait_panel_status(intel_dp, IDLE_CYCLE_MASK, IDLE_CYCLE_VALUE); >> } >> >> +static void ironlake_wait_backlight_on(struct intel_dp *intel_dp) >> +{ >> + wait_remaining_ms_from_jiffies(intel_dp->last_power_on, >> + intel_dp->backlight_on_delay); >> +} >> + >> +static void ironlake_edp_wait_backlight_off(struct intel_dp *intel_dp) >> +{ >> + wait_remaining_ms_from_jiffies(intel_dp->last_backlight_off, >> + intel_dp->backlight_off_delay); >> +} >> >> /* Read the current pp_control value, unlocking the register if it >> * is locked >> @@ -1147,7 +1164,7 @@ static void ironlake_panel_vdd_off_sync(struct intel_dp *intel_dp) >> I915_READ(pp_stat_reg), I915_READ(pp_ctrl_reg)); >> >> if ((pp & POWER_TARGET_ON) == 0) >> - msleep(intel_dp->panel_power_cycle_delay); >> + intel_dp->last_power_cycle = jiffies; >> >> intel_runtime_pm_put(dev_priv); >> } >> @@ -1222,6 +1239,7 @@ void ironlake_edp_panel_on(struct intel_dp *intel_dp) >> POSTING_READ(pp_ctrl_reg); >> >> ironlake_wait_panel_on(intel_dp); >> + intel_dp->last_power_on = jiffies; >> >> if (IS_GEN5(dev)) { >> pp |= PANEL_POWER_RESET; /* restore panel reset bit */ >> @@ -1242,6 +1260,8 @@ void ironlake_edp_panel_off(struct intel_dp *intel_dp) >> >> DRM_DEBUG_KMS("Turn eDP power off\n"); >> >> + ironlake_edp_wait_backlight_off(intel_dp); >> + >> pp = ironlake_get_pp_control(intel_dp); >> /* We need to switch off panel power _and_ force vdd, for otherwise some >> * panels get very unhappy and cease to work. */ >> @@ -1252,6 +1272,7 @@ void ironlake_edp_panel_off(struct intel_dp *intel_dp) >> I915_WRITE(pp_ctrl_reg, pp); >> POSTING_READ(pp_ctrl_reg); >> >> + intel_dp->last_power_cycle = jiffies; >> ironlake_wait_panel_off(intel_dp); >> } >> >> @@ -1273,7 +1294,7 @@ void ironlake_edp_backlight_on(struct intel_dp *intel_dp) >> * link. So delay a bit to make sure the image is solid before >> * allowing it to appear. >> */ >> - msleep(intel_dp->backlight_on_delay); >> + ironlake_wait_backlight_on(intel_dp); >> pp = ironlake_get_pp_control(intel_dp); >> pp |= EDP_BLC_ENABLE; >> >> @@ -1305,7 +1326,7 @@ void ironlake_edp_backlight_off(struct intel_dp *intel_dp) >> >> I915_WRITE(pp_ctrl_reg, pp); >> POSTING_READ(pp_ctrl_reg); >> - msleep(intel_dp->backlight_off_delay); >> + intel_dp->last_backlight_off = jiffies; >> } >> >> static void ironlake_edp_pll_on(struct intel_dp *intel_dp) >> diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h >> index 46aea6c..92de688 100644 >> --- a/drivers/gpu/drm/i915/intel_drv.h >> +++ b/drivers/gpu/drm/i915/intel_drv.h >> @@ -485,6 +485,9 @@ struct intel_dp { >> int backlight_off_delay; >> struct delayed_work panel_vdd_work; >> bool want_panel_vdd; >> + unsigned long last_power_cycle; >> + unsigned long last_power_on; >> + unsigned long last_backlight_off; > > The delays and the timestamps could be grouped into an unnamed struct > within intel_dp, but this can be left as a follow-up patch. > > BR, > Jani. > >> bool psr_setup_done; >> struct intel_connector *attached_connector; >> }; >> -- >> 1.8.3.1 >> >> _______________________________________________ >> Intel-gfx mailing list >> Intel-gfx@xxxxxxxxxxxxxxxxxxxxx >> http://lists.freedesktop.org/mailman/listinfo/intel-gfx > > -- > Jani Nikula, Intel Open Source Technology Center -- Paulo Zanoni _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx