-----Original Message-----
> Calls to x86_64_exception_frame() with combined items set in the flags
> argument that include EFRAME_VERIFY do not have the EFRAME_VERIFY
> operation performed. I have some cores where multiple cases of
> attempting to read a not-present pt_regs end a single PID backtrace with
> a failure. One instance has the pt_regs read overrunning stacktop
> because the pt_regs is not present and the level's stack position is
> closer to stacktop than the size of a pt_regs. That results in a
> backtrace failing before complete with a "seek error" at the start of
> page after stacktop:
>
> crash> bt 7456
> PID: 7456 TASK: ffff933fdb960000 CPU: 0 COMMAND: "sh"
> #0 [fffffe0000009e58] crash_nmi_callback at ffffffff93260e93
> ...
> #9 [ffffaea5c0003f80] hrtimer_interrupt at ffffffff933313d5
> bt: seek error: kernel virtual address: ffffaea5c0004000 type: "pt_regs"
> crash>
>
> The correct backtrace would reach level #12 with no seek error.
>
> The condition to perform the EFRAME_VERIFY operation tests if the flags
> value equals EFRAME_VERIFY, not if the value includes EFRAME_VERIFY. The
> call to x86_64_exception_frame() in x86_64_print_stack_entry() performed
> when eframe_check >= 0 supplies a flags value of EFRAME_PRINT |
> EFRAME_VERIFY.
>
> In the bt example above backtrace reaches level #9, 128 bytes from the
> top of the current stack's pages in an IRQ stack and with a function
> name ending in "_interrupt". This leads to x86_64_print_stack_entry()
> setting eframe_check to zero and x86_64_exception_frame() being called
> with flags EFRAME_PRINT | EFRAME_VERIFY. x86_64_exception_frame()
> doesn't perform the verify because flags is not just EFRAME_VERIFY. An
> attempt is made to read 168 bytes (SIZE(pt_regs) - 8 bytes) from a
> position 128 bytes from the top of the stack. The stack in question is
> followed by a not-present page and the read fails attempting to read
> from the page following stacktop.
>
> Signed-off-by: David Mair <dmair@xxxxxxxx>
Thanks for the patch.
but it seems that it stops "bt -e" and "bt -E" options working, can we
keep them?
Also, I could not apply the patch as is because of a few format issues.
I think your MTA or something replaced tabs with spaces and inserted a
new line between the "ulong kvaddr," and "char *local,". We can fix
this when applying but if you update the patch for the above comment,
please fix it as well (if possible). It would be helpful.
Thanks,
Kazu
> ---
> diff --git a/x86_64.c b/x86_64.c
> index fc05e8a..cc870e0 100644
> --- a/x86_64.c
> +++ b/x86_64.c
> @@ -4418,7 +4418,7 @@ x86_64_exception_frame(ulong flags, ulong kvaddr,
> char *local,
> long err;
> char buf[BUFSIZE];
>
> - if (flags == EFRAME_VERIFY) {
> + if (flags & EFRAME_VERIFY) {
> if (!accessible(kvaddr) ||
> !accessible(kvaddr + SIZE(pt_regs) - sizeof(long)))
> return FALSE;
>
> --
> Crash-utility mailing list
> Crash-utility@xxxxxxxxxx
> https://www.redhat.com/mailman/listinfo/crash-utility
--
Crash-utility mailing list
Crash-utility@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/crash-utility
