[Bug 101517] [BAT][BYT,BSW] WARN_ON(dev_priv->uncore.funcs.mmio_readl(dev_priv, (((const i915_reg_t){ .reg = (0x180000 + 0x650C) })), true) & (1 << 27))

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=101517

--- Comment #4 from Ville Syrjala <ville.syrjala at linux.intel.com> ---
(In reply to krisman from comment #3)
> (In reply to Martin Peres from comment #0)
> > Starting from CI_DRM_2743, executing the tests
> > igt at kms_pipe_crc_basic@suspend-read-crc-pipe-[ac] on our baytrails and
> > braswells may lead to the following WARNING:
> > 
> > [  579.071526] [drm:chv_set_cdclk [i915]] *ERROR* timed out waiting for
> > CDclk change
> > [  579.071942] WARN_ON(dev_priv->uncore.funcs.mmio_readl(dev_priv, (((const
> 
> In the code, this WARN_ON carries the following comment:
> 
>   /*
>    * FIXME is this guaranteed to clear
>    * immediately or should we poll for it?
>   */
>   WARN_ON(I915_READ(GCI_CONTROL) & PFI_CREDIT_RESEND);
> 
> Maybe we should consider polling? :-)
> 
> Anyway, that WARN_ON can be a side effect of the *ERROR* right before it. 
> I'm Looking into that.  Also, I don't see much change in the commit log
> between CI_DRM_2742 and CI_DRM_2743 that would explain it, but I will take a
> deeper look.

The simple solution would be to just revert commit 63ff30442519 ("drm/i915:
Nuke the VLV/CHV PFI programming power domain workaround")

I failed to consider that the firmware likes to reset the cdclk to some
non-minimum frequency during suspend, and then we apparently perform a modeset
without actually modesetting anything and thus only end up reprogramming the
cdclk without the power well being enabled.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20170620/549a4554/attachment-0001.html>