]> git.karo-electronics.de Git - linux-beck.git/commitdiff
drm/i915: VLV/CHV PSR: Increase wait delay time before active PSR.
authorRodrigo Vivi <rodrigo.vivi@intel.com>
Fri, 31 Jul 2015 00:07:55 +0000 (17:07 -0700)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 5 Aug 2015 08:07:44 +0000 (10:07 +0200)
Since active function on VLV immediately activate PSR let's give more
time for idleness. Different from core platforms where we have idle_frames
count.

Also kms_psr_sink_crc now is automated and always get this:

[drm:intel_enable_pipe] enabling pipe A
[drm:intel_edp_backlight_on]
[drm:intel_panel_enable_backlight] pipe
[drm:intel_panel_enable_backlight] pipe A
[drm:intel_panel_actually_set_backlight] set backlight PWM = 7812

PSR gets enabled somewhere here after backlight.

[drm:intel_get_hpd_pins] hotplug event received, stat 0x00000000, dig 0x0
[drm:vlv_pipe_set_fifo_size] Pipe A FIFO split 511 / 511 / 511
[drm:vlv_update_wm] Setting FIFO watermarks - A: plane=391, cursor=63, sp

PSR gets flushed around here by intel_atomic_commit

[drm:vlv_pipe_set_fifo_size] Pipe A FIFO split 511 / 511 / 511
[drm:vlv_update_wm] Setting FIFO watermarks - A: plane=391, cursor=63, sp
[drm:intel_set_memory_cxsr] memory self-refresh is enabled
[drm:intel_connector_check_state] [CONNECTOR:39:eDP-1]
[drm:check_encoder_state] [ENCODER:30:DAC-30]
[drm:check_encoder_state] [ENCODER:31:TMDS-31]
[drm:check_encoder_state] [ENCODER:36:TMDS-36]
[drm:check_encoder_state] [ENCODER:38:TMDS-38]
[drm:check_crtc_state] [CRTC:21]
[drm:check_crtc_state] [CRTC:26]
[drm:intel_psr_activate [i915]] *ERROR* PSR Active
[drm:intel_get_hpd_pins] hotplug event received, stat 0x00000000, dig 0x
[drm:intel_set_cpu_fifo_underrun_reporting [i915]] *ERROR* pipe A underrun
[drm:intel_cpu_fifo_underrun_irq_handler [i915]] *ERROR* CPU pipe A FIFO
Underrun.

It is true that in a product we won't keep disabling and enabling planes so
frequently, but for safeness let's stay conservative.

It is also true that 500ms is an etternity. But PSR is anyway a power saving
feature for idle scenario. So if it is idle feature stays on and 500ms to get
it reanabled is not that insane.

v2: Rebase over intel_psr.c and fix typo.
v3: Revival: Manual tests indicated that this is needed. With a short delay
    there is a huge risk of getting blank screens when planes are being enabled.
v4: Revival 2 with reasonable delay. 1/2 sec instead of 5. VBT is 10 sec but
    actually time for link training what we aren't doing, but with only 100 sec
    in some cases kms_psr_sink_crc manual was showing blank screen,
    so let's use this for now. Also changed comment by a FIXME.
v5: Rebase after a long time, remove FIXME and update comment above.
v6: msecs_to_jiffies is already on delay. remove duplication.
v7: use msecs_to_jiffies on schedule_delayed_work call.

Reviewed-by: Durgadoss R <durgadoss.r@intel.com> (v4)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Tested-By: Intel Graphics QA PRTS (Patch Regression Test System Contact: shuang.he@intel.com)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/intel_psr.c

index acd8ec859f71b8bd5a22dcaa8809b7bc552f29e4..a04b4dc5ed9b459ce42de39c12de74539a620066 100644 (file)
@@ -698,6 +698,7 @@ void intel_psr_flush(struct drm_device *dev,
        struct drm_i915_private *dev_priv = dev->dev_private;
        struct drm_crtc *crtc;
        enum pipe pipe;
+       int delay_ms = HAS_DDI(dev) ? 100 : 500;
 
        mutex_lock(&dev_priv->psr.lock);
        if (!dev_priv->psr.enabled) {
@@ -733,7 +734,7 @@ void intel_psr_flush(struct drm_device *dev,
 
        if (!dev_priv->psr.active && !dev_priv->psr.busy_frontbuffer_bits)
                schedule_delayed_work(&dev_priv->psr.work,
-                                     msecs_to_jiffies(100));
+                                     msecs_to_jiffies(delay_ms));
        mutex_unlock(&dev_priv->psr.lock);
 }