[Intel-gfx] [PATCH] Don't touch the pipestat regs for detecting FIFO underrun. The kernel owns them.

[Jesse Barnes]

On Friday, January 2, 2009 6:43 pm Eric Anholt wrote:
> On Mon, 2008-12-29 at 14:07 -0800, Jesse Barnes wrote:
> > On Monday, December 29, 2008 1:45 pm Eric Anholt wrote:
> > > Since we don't perform any synchronization with the kernel on these
> > > regs, we could race with the kernel to write stale values and end up
> > > not having vblank interrupts enabled when somebody was waiting on one.
> > > ---
> > >  src/i830_display.c |    4 ----
> > >  src/i830_driver.c  |   26 --------------------------
> > >  2 files changed, 0 insertions(+), 30 deletions(-)
> >
> > If we do this (patch seems fine) we should check for it in the kernel
> > instead, maybe by enabling the corresponding error interrupt.
>
> There's preliminary stuff for it in the error-recovery branch of my
> tree, but it got backed out because it caused hangs that killed ssh and
> I didn't have time to debug.

How hard would it be to just extract the underrun reporting?  It's already 
helped us find quite a few bugs, so just getting the info would be a big help 
even if we don't recover very well from it...

-- 
Jesse Barnes, Intel Open Source Technology Center