On Friday 02 April 2010 07:51:15 Maxim Levitsky wrote:
> On Thu, 2010-04-01 at 17:45 -0400, Bob Copeland wrote:
> > On Fri, Apr 02, 2010 at 12:18:25AM +0300, Maxim Levitsky wrote:
> > > > Here, TX interrupts stop, then after a while (1~2 sec), I get the
> > > > loads of AR5K_INT_RXORN, although I *know* that card has literally
> > > > nothing to receive (this channel is free of neighbor APs), and I
> > > > send UDP stream.
> > > >
> > > > Sure, some stuff is received (beacons, broadcast) but that shouldn't
> > > > overflow the queue.
> >
> > This is the bugzilla, it sounds like the same thing.
> >
> > http://bugzilla.kernel.org/show_bug.cgi?id=15538
> >
> > Sounds likely that interrupts on the card get turned off for
> > some reason. IIRC you debugged wherever we set the IMR and we
> > weren't specifically changing it around the time things broke?
>
> I did a quick look at that by reading the regs via debugfs, and IMR
> related registers don't change a bit between working and non working
> state.
> Of course that is better done with printks, etc, but I think that the
> problem is much deeper. after all, _most_ atheros chips work just fine.
> Its that AR2425 that doesn't.
hi maxim!
i'm not sure if this is related or not, but you could try to decrease the DMA
size around line 1380 in base.c:
/*
* Set Rx/Tx DMA Configuration
*
* Set standard DMA size (128). Note that
* a DMA size of 512 causes rx overruns and tx errors
* on pci-e cards (tested on 5424 but since rx overruns
* also occur on 5416/5418 with madwifi we set 128
* for all PCI-E cards to be safe).
*
* XXX: need to check 5210 for this
* TODO: Check out tx triger level, it's always 64 on dumps but I
* guess we can tweak it and see how it goes ;-)
*/
if (ah->ah_version != AR5K_AR5210) {
AR5K_REG_WRITE_BITS(ah, AR5K_TXCFG,
AR5K_TXCFG_SDMAMR, AR5K_DMASIZE_128B);
AR5K_REG_WRITE_BITS(ah, AR5K_RXCFG,
AR5K_RXCFG_SDMAMW, AR5K_DMASIZE_128B);
}
the comments there suggest problems with pci-e cards (yours is pci-e isn't
it?) and mention similar symptoms to yours. also i have seen this bug only in
a configuration where DMA was broken (that is on a routerstation pro, with a
ar71xx CPU which has a PCI DMA bug which forces us to decrease the size to 4).
so maybe experimenting with these sizes might help... just an idea... and let
us know how it goes...
bruno
_______________________________________________
ath5k-devel mailing list
ath5k-devel@lists.ath5k.org
https://lists.ath5k.org/mailman/listinfo/ath5k-devel
