Re: [PATCH] drm/i915: paper over missed irq issues with force wake vodoo

[Keith Packard <keithp@keithp.com>]

[-- Attachment #1.1: Type: text/plain, Size: 1427 bytes --]

On Sat, 14 Jan 2012 01:31:40 +0100, Daniel Vetter <daniel@ffwll.ch> wrote:
> On Fri, Jan 13, 2012 at 04:11:43PM -0800, Keith Packard wrote:
> > On Sat, 14 Jan 2012 00:52:31 +0100, Daniel Vetter <daniel@ffwll.ch> wrote:
> > > acc101d drm/i915: Hold gt_lock across forcewake register reads
> > > 
> > > Imo this is a simple cleanup (reading forcewake-protected registers isn't
> > > really a fast-path for us), so material for -next.
> > 
> > The 'optimization' is just a side benefit. The fix is to prevent reads
> > From happening without forcewake being set.
> 
> I still fail to see how you can sneak a read in there without forcewake
> being asserted. And assuming I haven't understood you the last time around
> we've discussed this, you've agreed.

Yes, except during reset, where forcewake is cleared even if the lock
count is non-zero. Any reads happening while reset is going on will
return garbage. None of this is 'required' given the structure of the
code today, it just makes it all evident without having to go through
yet another long sequence of explanations.

I had patches to hold the spinlock across register writes too, as using
different locking for reading and writing seems like a bad plan, but I
didn't put those in because writes involve spinning to wait for the fifo
to drain, and that seemed like a bad thing to do while holding the
spinlock.

-- 
keith.packard@intel.com

[-- Attachment #1.2: Type: application/pgp-signature, Size: 827 bytes --]

[-- Attachment #2: Type: text/plain, Size: 159 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx