On 2010-09-30 08:58, Takashi Iwai wrote:

At Wed, 29 Sep 2010 10:14:49 +0200, David Henningsson wrote:

...

I'm researching a few bugs where the user claims position_fix=1 helps the problem, but adding the quirk for that model surprisingly didn't. So looking at the code, I concluded that the difference was via_dmapos_patch, and that they all had via_dmapos_patch=1, and that via_dmapos_patch=0 helped them solve the problem. Three out of five (not all of them have reported back on via_dmapos_patch=0 yet) are VIA controllers rev 10.

So we now have VIA controllers that need via_dmapos_patch=0.

I'm cc:ing Joseph here. You were the one writing via_dmapos_patch originally, could you confirm the suspicion that VIA controller rev 10 (and possibly more) actually should have via_dmapos_patch turned off?

...

I can think of a few approaches here:

1. since position_fix=1 implicitly sets via_dmapos_patch to 0 (maybe

unintentionally), we should add a position_fix=3 meaning lpib + via_dmapos_patch=1

2. figure if something has changed recently (as in "within the last

year"...) that has made via_dmapos_patch=1 work worse than before

3. figure out if there are several ATI/VIA controllers that actually

never wants the patch.

Any thoughts?

All sound as reasonable proposals. The 1 is easy. David, could you care to send a patch?'

Something like this (untested)?

2 and 3 aren't trivial, but we can start by disabling via_dmapos for recent revisions.Since it can be controlled over a module option by the fix 1, it'll be easier to check the regression.