[Intel-gfx] Graphics on thinkpad x270 after dock/undock works only for the first time (CPU pipe B FIFO underrun)

[Chris Wilson]

Quoting Rodrigo Vivi (2018-01-02 19:21:08)
> On Sat, Dec 30, 2017 at 12:53:58PM +0000, Jiri Kosina wrote:
> > On Sat, 30 Dec 2017, Jiri Kosina wrote:
> > 
> > > Seems like disabling RC6 on the kernel command line works this around, and 
> > > I can dock / undock several times in a row with the image always coming 
> > > up properly on the external display.
> > > 
> > > On the first undock, the WARN_ONCE() below triggers, so I believe each 
> > > undock leaks memory.
> > > 
> > > [   38.755084] Failed to release pages: bind_count=1, pages_pin_count=1, pin_global=0
> > > [   38.755138] WARNING: CPU: 3 PID: 96 at ../drivers/gpu/drm/i915/i915_gem_userptr.c:89 cancel_userptr+0xe5/0xf0 [i915]

Not a leak, just a warning the shadow pte are stale due to someone
pinning a (gfx) vma.

> > OK, I am seeing this warning with current Linus' tree (5aa90a845) even 
> > without any attempt to dock/undock, so it's probably unrelated to external 
> > outputs and it only by coincidence appeared originally at the same time I 
> > docked the machine.
> > 
> > So there are two separate issues on this machine with latest kernel 
> > (neither of them probably being regression):
> > 
> > - I have to disable i915 RC6 at the kernel cmdline, otherwise external 
> >   (dock) display gets output only randomly (seems like always only on 
> >   first dock)
> 
> Joonas, Chris, time to bring rc6_enable back on next-fixes before we
> remove this support entirely?

No. It's precisely because of this mistake (thinking that rc6 has
anything to do with the issue when it's the consequential shotgun
disabling of rpm) that the modparam is not coming back.
-Chris