On Wed, Sep 24, 2008 at 5:36 PM, David Miller <davem@xxxxxxxxxxxxx> wrote: > From: "Dave Airlie" <airlied@xxxxxxxxx> > Date: Wed, 24 Sep 2008 15:45:46 +1000 > >> I'm still dubious about this, wouldn't we see other wierdass side >> effects if X was trashing the BARs on other devices? > > Sure. My theory is that it's a recent xorg change causing this, > so I've been going through GIT history for xserver, libpciaccess, > and the intel driver for the past year looking for clues. > > If there is usually a gap after the video device, there would just > be no response from the PCI bus, and the way that's handled is > chipset specific. At least a while back, most x86 systems would > silently ignore writes and return all 1's in such a case, but > they may be generating bus error events these days. I simply don't > know. The only thing I can think off then is either the pciaccess conversion of the intel Xorg driver, or maybe something going wrong since PAT support was added. > >> I think tglx is on the right path, same problem as e1000, code is >> stupid, it can reenter the nvram read/write code from irq >> context, and pwn itself. > > The e1000e side here is reproducable way too easily for it to be the > same case, as far as I see it. > > The e1000 driver has probably had this problem for years and we've > only recently had some concrete cases of it triggering. > > Also, what utility are you running on your system that is even > accessing the NVRAM on the e1000e card? Knowing that might help > us understand why this problem has appeared now. Maybe there is > some diagnostic or monitoring tool that is now becoming prevalent > in these distributions where it triggers. The driver seems quite happy to access the NVRAM, I think Thomas has some backtraces that show it clearly doing silly reentrant things... > > This problem started happening seemingly "all of a sudden", even to > people who have been keeping sort-of recent with their kernels, such > as yourself. > > Yet we can't get any sense yet what range of kernel versions are in > use when the problem triggers. I've seen it reported at least at 2.6.27-rc1 and maybe even one of Fedora's -rc0 kernels. Dave. > > I'm about to leave for a week or so in Paris for the netfilter > workshop, so I hope that someone other than myself will do some data > mining like I have instead of (merely) tossing theories around and > finger pointing. > -- To unsubscribe from this list: send the line "unsubscribe kernel-testers" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
