Re: [PATCH v2 15/16] drm/i915/gen9: Reject display updates that exceed wm limitations

[Matt Roper <matthew.d.roper@intel.com>]

On Fri, Apr 22, 2016 at 10:21:20AM -0400, Lyude Paul wrote:
> Unfortunately dropping that patch just results with the screen staying blank
> starting from the moment the driver loads. 
> 
> As for the external monitors question: the only time this issue seems to happen
> with this patchset is when there's no external monitors connected. If I boot
> with just the laptop and nothing connected, it gets as far as showing systemd
> starting all the services in fbcon up to where it starts GDM, and then the
> screen just gets stuck on that framebuffer. Switching to a VT at that point
> doesn't fix it.
> 
> If I plug in another monitor to the system everything starts working fine again,
> even if I unplug the display again afterwards or switch VTs. I've recorded a
> sample of dmesg where I boot up the laptop with no displays connected, connect
> an external MST display, and then disconnect it (just grep for "rhtest unknown:"
> to find where I marked the spot before I connected the monitors, etc. ):
> 
> https://paste.fedoraproject.org/358607/61334721/
> 
> Also the contents of /sys/kernel/debug/dri/0/i915_pri_wm_latency:
> WM0 2 (2.0 usec)
> WM1 19 (19.0 usec)
> WM2 28 (28.0 usec)
> WM3 32 (32.0 usec)
> WM4 63 (63.0 usec)
> WM5 77 (77.0 usec)
> WM6 83 (83.0 usec)
> WM7 99 (99.0 usec)


Okay, I think I've figured this out.  The problem arises because with
this patch set we only re-calculate the DDB (the data buffer that
watermarks refer to) when we need to, otherwise we use the allocation
that was computed on previous modesets.  That pretty much always works
fine, except when we've just booted and we're using the DDB allocations
inherited from the BIOS; those BIOS allocations may fail to meet our
needs/expectations and mislead us into thinking there's no way we can
program watermarks safely.

I think the solution is probably to just add a flag to the driver that
forces our very first real modeset to completely recompute proper DDB
allocations, even if it otherwise wouldn't need to.  I'll put together a
patch for that shortly and see if it helps.


Matt

> 
> On Thu, 2016-04-21 at 16:14 -0700, Matt Roper wrote:
> > On Thu, Apr 21, 2016 at 05:49:52PM -0400, Lyude Paul wrote:
> > > 
> > > On a T560, this ends up rejecting valid watermark configurations so the
> > > internal
> > > display doesn't switch from fbcon to X properly:
> > Hmm.  We can leave this patch out of the series and still have a fix for
> > the WARN_ON(!wm_changed), but one of the main goals was to identify and
> > reject configurations that the platform is truly incapable of handling.
> > I'm still going to include this patch in my next spin of the series on
> > the mailing list, but we can just omit it during merge if we need more
> > time to figure out why your configuration is being rejected.  I'll also
> > add an extra DRM_DEBUG_KMS line to this patch to give us more insight
> > into what the final calculated values were and how far out of bounds
> > they were.
> > 
> > If you drop this patch but keep the rest of the series, does the system
> > still operate properly when it goes into X?  I.e., no underruns reported
> > and no flickering/corruption on screen?
> > 
> > Could you provide the contents of
> > /sys/kernel/debug/dri/0/i915_pri_wm_latency ?  I've heard that there are
> > some SKL pcode (BIOS) versions floating around that give bad memory
> > latency values which result in us calculating the wrong watermark values
> > in the end; it would be good to see if that's a contributing factor
> > here.
> > 
> > Also, I can see your eDP panel in the log below; is that your only
> > display or do you also have an external display attached?
> > 
> > 
> > 
> > Matt
> > 
> > > 
> > > 
> > > [    5.767383] [drm:intelfb_create] re-using BIOS fb
> > > [    5.767444] [drm] Initialized i915 1.6.0 20160411 for 0000:00:02.0 on
> > > minor 0
> > > [    5.767449] [drm:intelfb_create] allocated 1920x1080 fb: 0x00000000, bo
> > > ffff88021374c000
> > > [    5.767760] fbcon: inteldrmfb (fb0) is primary device
> > > [    5.768147] [drm:connected_sink_compute_bpp] [CONNECTOR:37:eDP-1]
> > > checking for sink bpp constrains
> > > [    5.768150] [drm:connected_sink_compute_bpp] clamping display bpp (was
> > > 36) to EDID reported max of 18
> > > [    5.768159] [drm:skl_update_scaler_crtc] Updating scaler for [CRTC:26]
> > > scaler_user index 0.31
> > > [    5.768162] [drm:skl_update_scaler] scaler_user index 0.31: Staged
> > > freeing scaler id 0 scaler_users = 0x0
> > > [    5.768166] [drm:intel_dp_compute_config] DP link computation with max
> > > lane count 2 max bw 270000 pixel clock 142900KHz
> > > [    5.768172] [drm:intel_dp_compute_config] DP link bw 0a rate select 00
> > > lane count 2 clock 270000 bpp 18
> > > [    5.768174] [drm:intel_dp_compute_config] DP link bw required 257220
> > > available 432000
> > > [    5.768180] [drm:intel_modeset_pipe_config] hw max bpp: 36, pipe bpp: 18,
> > > dithering: 1
> > > [    5.768193] [drm:intel_dump_pipe_config] [CRTC:26][modeset] config
> > > ffff8800d5c11000 for pipe A
> > > [    5.768200] [drm:intel_dump_pipe_config] cpu_transcoder: EDP
> > > [    5.768202] [drm:intel_dump_pipe_config] pipe bpp: 18, dithering: 1
> > > [    5.768206] [drm:intel_dump_pipe_config] fdi/pch: 0, lanes: 0, gmch_m: 0,
> > > gmch_n: 0, link_m: 0, link_n: 0, tu: 0
> > > [    5.768211] [drm:intel_dump_pipe_config] dp: 1, lanes: 2, gmch_m:
> > > 4994717, gmch_n: 8388608, link_m: 277484, link_n: 524288, tu: 64
> > > [    5.768215] [drm:intel_dump_pipe_config] dp: 1, lanes: 2, gmch_m2: 0,
> > > gmch_n2: 0, link_m2: 0, link_n2: 0, tu2: 0
> > > [    5.768217] [drm:intel_dump_pipe_config] audio: 0, infoframes: 0
> > > [    5.768219] [drm:intel_dump_pipe_config] requested mode:
> > > [    5.768226] [drm:drm_mode_debug_printmodeline] Modeline 0:"1920x1080" 60
> > > 142900 1920 1968 2000 2082 1080 1082 1087 1144 0x48 0xa
> > > [    5.768228] [drm:intel_dump_pipe_config] adjusted mode:
> > > [    5.768234] [drm:drm_mode_debug_printmodeline] Modeline 0:"1920x1080" 60
> > > 142900 1920 1968 2000 2082 1080 1082 1087 1144 0x48 0xa
> > > [    5.768239] [drm:intel_dump_crtc_timings] crtc timings: 142900 1920 1968
> > > 2000 2082 1080 1082 1087 1144, type: 0x48 flags: 0xa
> > > [    5.768241] [drm:intel_dump_pipe_config] port clock: 270000
> > > [    5.768244] [drm:intel_dump_pipe_config] pipe src size: 1920x1080
> > > [    5.768247] [drm:intel_dump_pipe_config] num_scalers: 2, scaler_users:
> > > 0x0, scaler_id: -1
> > > [    5.768251] [drm:intel_dump_pipe_config] gmch pfit: control: 0x00000000,
> > > ratios: 0x00000000, lvds border: 0x00000000
> > > [    5.768254] [drm:intel_dump_pipe_config] pch pfit: pos: 0x00000000, size:
> > > 0x00000000, disabled
> > > [    5.768256] [drm:intel_dump_pipe_config] ips: 0
> > > [    5.768258] [drm:intel_dump_pipe_config] double wide: 0
> > > [    5.768263] [drm:intel_dump_pipe_config] ddi_pll_sel: 1; dpll_hw_state:
> > > ctrl1: 0x13, cfgcr1: 0x0, cfgcr2: 0x0
> > > [    5.768266] [drm:intel_dump_pipe_config] planes on this crtc
> > > [    5.768277] [drm:intel_dump_pipe_config] STANDARD PLANE:23 plane: 0.0
> > > idx: 0 enabled
> > > [    5.768284] [drm:intel_dump_pipe_config] 	FB:58, fb = 1920x1080
> > > format = 0x34325258
> > > [    5.768285] [drm:intel_dump_pipe_config] 	scaler:-1 src (0, 0)
> > > 1920x1080 dst (0, 0) 1920x1080
> > > [    5.768290] [drm:intel_dump_pipe_config] CURSOR PLANE:25 plane: 0.1 idx:
> > > 1 disabled, scaler_id = -1
> > > [    5.768295] [drm:intel_dump_pipe_config] STANDARD PLANE:27 plane: 0.1
> > > idx: 2 disabled, scaler_id = -1
> > > [    5.768309] [drm:skl_update_scaler_plane] Updating scaler for [PLANE:23]
> > > scaler_user index 0.0
> > > [    5.768319] [drm:intel_find_shared_dpll] CRTC:26 allocated DPLL 0
> > > [    5.768323] [drm:intel_reference_shared_dpll] using DPLL 0 for pipe A
> > > [    5.768327] [drm:skl_update_scaler_crtc] Updating scaler for [CRTC:26]
> > > scaler_user index 0.31
> > > [    5.768339] [drm:skl_compute_plane_wm] Requested display configuration
> > > exceeds system watermark limitations
> > > [    5.768359] [drm:drm_fb_helper_hotplug_event] 
> > > [    5.768363] [drm:drm_helper_probe_single_connector_modes]
> > > [CONNECTOR:37:eDP-1]
> > > [    5.768367] [drm:intel_dp_detect] [CONNECTOR:37:eDP-1]
> > > [    5.768786] [drm:intel_dp_probe_oui] Sink OUI: 0022b9
> > > [    5.769177] [drm:intel_dp_probe_oui] Branch OUI: 444615
> > > [    5.769604] [drm:drm_edid_to_eld] ELD: no CEA Extension found
> > > [    5.769611] [drm:drm_helper_probe_single_connector_modes]
> > > [CONNECTOR:37:eDP-1] probed modes :
> > > [    5.769618] [drm:drm_mode_debug_printmodeline] Modeline 38:"1920x1080" 60
> > > 142900 1920 1968 2000 2082 1080 1082 1087 1144 0x48 0xa
> > > [    5.769621] [drm:drm_helper_probe_single_connector_modes]
> > > [CONNECTOR:45:DP-1]
> > > [    5.769624] [drm:intel_dp_detect] [CONNECTOR:45:DP-1]
> > > [    5.769628] [drm:intel_power_well_enable] enabling power well 2
> > > [    5.769633] [drm:skl_set_power_well] Enabling power well 2
> > > [    5.769659] [drm:intel_power_well_disable] disabling power well 2
> > > [    5.769672] [drm:skl_set_power_well] Disabling power well 2
> > > 
> > > On Tue, 2016-04-19 at 19:26 -0700, Matt Roper wrote:
> > > > 
> > > > If we can't find any valid level 0 watermark values for the requested
> > > > atomic transaction, reject the configuration before we try to start
> > > > programming the hardware.
> > > > 
> > > > Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
> > > > Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > > > ---
> > > >  drivers/gpu/drm/i915/intel_pm.c | 12 +++++++++++-
> > > >  1 file changed, 11 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/intel_pm.c
> > > > b/drivers/gpu/drm/i915/intel_pm.c
> > > > index 59d4574..96ffd54 100644
> > > > --- a/drivers/gpu/drm/i915/intel_pm.c
> > > > +++ b/drivers/gpu/drm/i915/intel_pm.c
> > > > @@ -3309,7 +3309,17 @@ static int skl_compute_plane_wm(const struct
> > > > drm_i915_private *dev_priv,
> > > >  
> > > >  	if (res_blocks >= ddb_allocation || res_lines > 31) {
> > > >  		*enabled = false;
> > > > -		return 0;
> > > > +
> > > > +		/*
> > > > +		 * If there are no valid level 0 watermarks, then we
> > > > can't
> > > > +		 * support this display configuration.
> > > > +		 */
> > > > +		if (level) {
> > > > +			return 0;
> > > > +		} else {
> > > > +			DRM_DEBUG_KMS("Requested display configuration
> > > > exceeds system watermark limitations\n");
> > > > +			return -EINVAL;
> > > > +		}
> > > >  	}
> > > >  
> > > >  	*out_blocks = res_blocks;

-- 
Matt Roper
Graphics Software Engineer
IoTG Platform Enabling & Development
Intel Corporation
(916) 356-2795
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx