Messages in this thread | | | Date | Mon, 8 Dec 2003 09:12:49 -0800 (PST) | From | Linus Torvalds <> | Subject | Re: 2.6.test11 bug |
| |
Rafael - you had the wrong address for the kernel mailing list, and I didn't notice when I replied.
Here is my comment again for the mailing list in case somebody else can think of something. The more heads, the merrier.
Linus
On Mon, 8 Dec 2003, Linus Torvalds wrote: > > On Mon, 8 Dec 2003, Rafal Skoczylas wrote: > > > > > Rafal - how consistent is the second form of the oops? > > > Have you seen that trace more than once? > > > > Not exactly the same, but there are some similarities (If I understand > > this log correctly). I ripped those oopses out of the logs so maybe you > > could look yourself and see something I don't see: > > http://secprog.org/who/rs/linux/2.6-test11-log.txt > > Oh wow. > > They all look to be (except for the odd last "bad page state" one, which > is likely just a result of some _other_ earlier corruption) due to the > high bit being cleared. And it's consistent across reboots too, so it's > not just some corruption that stayed around in memory. > > In all cases it looks like a pointer that should have been of the form > "0xDxxxxxxx", and was corrupted into the form "0x5xxxxxxx". > > And every time it's "mlnetd" - which may just be a coincidence (possibly > brought on by that being the most commonly run thing on your box), but it > certainly looks like it could also be an indication of the source of the > corruption. > > I find it interesting that it seems to be bit 31 that is corrupt: > admittedly that is also the only bit that is almost guaranteed to cause > oopses when it corrupts (because it changes the pointer "the most"), but > it's also to some degree one of the least used bits for bit operations. A > lot of things use bit #0 for locking etc. > > I'll have to think about this, but quite frankly I'm also hoping to see > more of a pattern about what this is all about. Can you keep your oopses > up somewhere? Maybe opening a bug on bugme.osdl.org? Even though I don't > use bugme personally, it's good to keep the record around when we don't > immediately see the reason for something.. > > Linus > - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |