Messages in this thread | | | From | Linus Torvalds <> | Date | Sun, 16 Dec 2012 14:52:51 -0800 | Subject | Re: [GIT PULL] x86/uapi for 3.8 |
| |
On Sun, Dec 16, 2012 at 2:40 PM, Matt Fleming <matt.fleming@intel.com> wrote: > > Linus have you got a stacktrace for the oops (and maybe even a dmesg)? I > suspect it won't tell us much, but any/all info we can gather will help.
Sadly, it starts scrolling away pretty quickly, and it really wasn't very useful. None of the traces contained any of the new code.
The backtrace was something like "find_vma()" called by "do_munmap()".
I do have a phone picture of an earlier oops, which was similar. In that case, the oops happened in "unmap_single_vma()", with the call trace being
unmap_vmas exit_mmap mmput do_exit do_group_exit get_signal_to_deliver do_signal do_notify_resume retint_signal
which looks rather similar: it's some kind of nasty vma list corruption. The main difference being that the exit of the process happened as a result of a signal rather than just a plain exit() system call. Which is probably in turn the result of some corruption of user-space data structures.
In some other cases, I've got quickly scrolling stuff that I couldn't even guess at, or just a hung system at early boot.
> At this point, I'm wondering if insert_identity_mapping() is trashing > valid mappings.
It does feel like *major* data corruption. As in "not just a bitflip or two", but "megabytes of data is wrong".
Linus
| |