Hi,
On 2/3/20 2:12 PM, Ville Syrjälä wrote:
On Sat, Feb 01, 2020 at 10:31:59AM +0000, Chris Wilson wrote:
To write to the DISPLAY_PHY_CONTROL requires holding the powerwells,
which during early resume we have not yet acquired until later in
intel_display_power_init_hw(). So compute the initial chv_phy_control,
but leave the HW unset until we first acquire the powerwell.
<7> [120.055984] i915 0000:00:02.0: [drm:intel_power_domains_init_hw [i915]] rawclk rate: 200000 kHz
<4> [120.056381] ------------[ cut here ]------------
<4> [120.056621] i915 0000:00:02.0: Unclaimed write to register 0x1e0100
<4> [120.056924] WARNING: CPU: 1 PID: 164 at drivers/gpu/drm/i915/intel_uncore.c:1166 __unclaimed_reg_debug+0x69/0x80 [i915]
<4> [120.056935] Modules linked in: vgem snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic btusb btrtl btbcm btintel i915 bluetooth coretemp crct10dif_pclmul crc32_pclmul snd_hda_intel snd_intel_dspcfg snd_hda_codec ghash_clmulni_intel snd_hwdep ecdh_generic ecc snd_hda_core r8169 snd_pcm lpc_ich realtek pinctrl_cherryview i2c_designware_pci prime_numbers
<4> [120.057027] CPU: 1 PID: 164 Comm: kworker/u4:3 Tainted: G U 5.5.0-CI-CI_DRM_7854+ #1
<4> [120.057038] Hardware name: /NUC5CPYB, BIOS PYBSWCEL.86A.0055.2016.0812.1130 08/12/2016
<4> [120.057058] Workqueue: events_unbound async_run_entry_fn
<4> [120.057275] RIP: 0010:__unclaimed_reg_debug+0x69/0x80 [i915]
<4> [120.057289] Code: 48 8b 78 18 48 8b 5f 50 48 85 db 74 2d e8 1f a0 3f e1 45 89 e8 48 89 e9 48 89 da 48 89 c6 48 c7 c7 00 8c 48 a0 e8 67 82 df e0 <0f> 0b 83 2d ce e2 2b 00 01 5b 5d 41 5c 41 5d c3 48 8b 1f eb ce 66
<4> [120.057301] RSP: 0018:ffffc90000bcfd08 EFLAGS: 00010082
<4> [120.057315] RAX: 0000000000000000 RBX: ffff888079919b60 RCX: 0000000000000003
<4> [120.057326] RDX: 0000000080000003 RSI: 0000000000000000 RDI: 00000000ffffffff
<4> [120.057336] RBP: ffffffffa04c9f4e R08: 0000000000000000 R09: 0000000000000001
<4> [120.057348] R10: 0000000025c3d560 R11: 000000006815f798 R12: 0000000000000000
<4> [120.057359] R13: 00000000001e0100 R14: 0000000000000286 R15: ffffffff8234a76b
<4> [120.057371] FS: 0000000000000000(0000) GS:ffff888074b00000(0000) knlGS:0000000000000000
<4> [120.057382] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
<4> [120.057393] CR2: 000055f4197df0d8 CR3: 000000006f326000 CR4: 00000000001006e0
<4> [120.057404] Call Trace:
<4> [120.057635] fwtable_write32+0x114/0x1d0 [i915]
<4> [120.057892] intel_power_domains_init_hw+0x4ff/0x650 [i915]
<4> [120.058150] intel_power_domains_resume+0x3d/0x70 [i915]
<4> [120.058363] i915_drm_resume_early+0x97/0xd0 [i915]
<4> [120.058575] ? i915_resume_switcheroo+0x30/0x30 [i915]
<4> [120.058594] dpm_run_callback+0x64/0x280
<4> [120.058626] device_resume_early+0xa7/0xe0
<4> [120.058652] async_resume_early+0x14/0x40
Closes: https://gitlab.freedesktop.org/drm/intel/issues/1089
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Imre Deak <imre.deak@xxxxxxxxx>
---
drivers/gpu/drm/i915/display/intel_display_power.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_display_power.c b/drivers/gpu/drm/i915/display/intel_display_power.c
index 64943179c05e..492668d5a193 100644
--- a/drivers/gpu/drm/i915/display/intel_display_power.c
+++ b/drivers/gpu/drm/i915/display/intel_display_power.c
@@ -5163,11 +5163,10 @@ static void chv_phy_control_init(struct drm_i915_private *dev_priv)
dev_priv->chv_phy_assert[DPIO_PHY1] = true;
}
- intel_de_write(dev_priv, DISPLAY_PHY_CONTROL,
- dev_priv->chv_phy_control);
-
drm_dbg_kms(&dev_priv->drm, "Initial PHY_CONTROL=0x%08x\n",
dev_priv->chv_phy_control);
+
+ /* Defer application of initial phy_control to enabling the powerwell */
Can't recall if there was a specific reason for wanting to write this
immediately. Maybe not. At least all the asserts are after we write
the register elsewhere so should trip that stuff. I suppose the other
option would be to check that the display power well is enabled before
we write this. But this is probably OK.
Reviewed-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
What I don't undestand is what actually changed to cause this? Did we
reorganize something in the init/resume sequence that previously forced
the display power well on before this point, or did we simply not check
for the unclaimed reg access?
I have been seeing this happen occasionally for quite a while now, but it
was not always reproducible (IIRC), so I guess that we were racing with some
other code-path which did grab the power-well ? I might be completely
wrong here, but the WARN triggered by this has been on my radar for quite
a while now.
Anyways, thank you for fixing this Chris.
Regards,
Hans
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
