Re: crashme fault

[Randy Dunlap <randy.dunlap@oracle.com>]

On Fri, 14 Sep 2007 21:28:12 -0700 (PDT) Linus Torvalds wrote:

> On Wed, 12 Sep 2007, Randy Dunlap wrote:
> >
> > I run almost-daily kernel testing.  I haven't seen 'crashme' cause a
> > kernel fault until today, and now I've seen it twice on 2.6.23-rc6-git2,
> > x86_64.  After the first fault, I ran 'crashme' about 10 more times
> > to get the second fault (usually for 10 minutes, one time for 30
> > minutes).
> 
> Interesting. If this is reproducible for you, can you try to narrow down 
> (with bit-bisect) roughly when it started.

It doesn't happen quite on demand.  Only 2 times in several hours
of testing of the last few days.

> > There is very little helpful info.  RIP is strange, e.g.: 000000000051b446
> > The call stack is not printed.  No kernel symbols are printed,
> > even though I have CONFIG_KALLSYMS{_ALL}=y.
> 
> It looks like it's a page fault as a result of a *user*space* access, and 
> most likely your machine would continue happily, except you have 
> "panic_on_oops" set, so when the oops happens, it shuts the system down.

I see.

> Now, the reason I say it looks like a user space access is that you have
> 
> 	RIP: 0033:[<0000000000510eea>]
> 	RSP: 002b:00007fffc9a8ec10
> 
> which are all user space segments. So the register contents clearly say 
> "page fault in user space".
> 
> However, what makes the kernel oops (rather than just send a SIGSEGV) is 
> that the page fault "error code" is zero (that's the number that is 
> printed out just after the "Oops" string). For a normal user space access, 
> you should have bit #2 set in the error code.
> 
> So the kernel thinks it's a kernel page fault, because the page fault 
> error code says so. But everything else seems to indicate that it's really 
> user mode.. It would be very interesting to hear when this started 
> happening.
> 
> Even if you cannot bisect it down all the way, since you say that you do 
> almost daily kernel testing, is this really new to 2.6.23-rc6-git2, and 
> 2.6.23-rc6-git1 was fine? 

I'll check back thru my test logs.

> Andi, anything comes to mind?
> 
> 			Linus
> 
> ---
> > [ 7487.208128] Unable to handle kernel paging request at 00000000ff019b53 RIP:
> > [ 7487.212752]  [<0000000000510eea>]
> > [ 7487.218537] PGD 10c1a2067 PUD 0
> > [ 7487.221811] Oops: 0000 [1] SMP
> > [ 7487.224989] CPU 2
> > [ 7487.227024] Modules linked in: loop
> > [ 7487.230550] Pid: 19139, comm: crashme Not tainted 2.6.23-rc6-git2 #1
> > [ 7487.236896] RIP: 0033:[<0000000000510eea>]  [<0000000000510eea>]
> > [ 7487.242925] RSP: 002b:00007fffc9a8ec10  EFLAGS: 00010e83
> > [ 7487.248234] RAX: 00000000ffff8c4a RBX: 00000000004014f1 RCX: 00002b20e11c8b37
> > [ 7487.255361] RDX: 0000000000510ee0 RSI: 0000000000000000 RDI: 000000000000000a
> > [ 7487.262489] RBP: 00007fffc9a8ec10 R08: 00007fffc9a8eb60 R09: 0000000000000000
> > [ 7487.269616] R10: 0000000000000008 R11: 0000000000000612 R12: 0000000000000000
> > [ 7487.276743] R13: 00007fffc9a8ee00 R14: 0000000000000000 R15: 0000000000000000
> > [ 7487.283871] FS:  00002b20e13676d0(0000) GS:ffff81011fc75840(0000) knlGS:0000000000000000
> > [ 7487.291952] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> > [ 7487.297693] CR2: 00000000ff019b53 CR3: 000000005be6b000 CR4: 00000000000006e0
> > [ 7487.304821] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [ 7487.311949] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> > [ 7487.319076] Process crashme (pid: 19139, threadinfo ffff810106830000, task ffff810102cf5040)
> > [ 7487.327511]
> > [ 7487.329009] RIP  [<0000000000510eea>]
> > [ 7487.332690]  RSP <00007fffc9a8ec10>
> > [ 7487.336180] CR2: 00000000ff019b53
> > [ 7487.339810] Kernel panic - not syncing: Fatal exception
> -

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***
-
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/