[Crash-utility] [PATCH v5 3/4] arm64: correct a PC shown in bt

[AKASHI, Takahiro]

On 29 June 2016 at 04:48, Dave Anderson <anderson at redhat.com> wrote:
>
> > On arm64, the link register (LR) holds a return address, which is the
one
> > just after a branch instruction. So using a saved lr as PC for
backtracing
> > might cause some confusion.
> > For example, in kernel/entry.S,
> > work_resched:
> >     ...
> >     bl schedule
> >
> > ret_to_user:
> >     ...
> >
> > The current code shows "ret_o_user", instead of "work_resched",
> > as a caller of schedule().
>
> I see...

​FYI, this can potentially be seen in the kernel's ftrace-based stack
tracer or
panic dump.
In the past, ​PC calculation was modified to be "LR-4", but this patch
was reverted. See
    commit 9702970c7bd3(Revert "ARM64: unwind: Fix PC calculation")

​So I said that adding this patch was a discussion topic.​

> >
> > This patch corrects a PC by decrementing it by 4.
> > But please note that this change may also make people a bit confused
> > because a value of LR in the stack dump of "bt -f" doesn't match with
> > an address in one-line summary.
> >
> >  #2 [ffffcc7511407eb0] schedule at ffff0000d628aee0
> >     ffffcc7511407eb0: ffffcc6d22f23080 ffff0000d5b44d6c  <= LR
> >     ffffcc7511407ec0: ffffcc7511407ed0 0000000000000000
> >  #3 [ffffcc7511407ed0] work_resched at ffff0000d5b44d68  <= correcrted
PC
>
> That's probably OK -- and at least "bt -F" would clarify it somewhat.

​and "bt -Fs"​

​-Takahiro AKASHI​


> Thanks,
>   Dave
>
>
> >
> > Signed-off-by: AKASHI Takahiro <takahiro.akashi at linaro.org>
> > ---
> >  arm64.c | 23 ++++++++++++++---------
> >  1 file changed, 14 insertions(+), 9 deletions(-)
> >
> > diff --git a/arm64.c b/arm64.c
> > index b74d2c4..bad36aa 100644
> > --- a/arm64.c
> > +++ b/arm64.c
> > @@ -1504,35 +1504,40 @@ static int
> >  arm64_print_stackframe_entry(struct bt_info *bt, int level, struct
> >  arm64_stackframe *frame, FILE *ofp)
> >  {
> >       char *name, *name_plus_offset;
> > -     ulong symbol_offset;
> > +     ulong pc, symbol_offset;
> >       struct syment *sp;
> >       struct load_module *lm;
> >       char buf[BUFSIZE];
> >
> > -        name = closest_symbol(frame->pc);
> > +     /*
> > +      * if pc comes from a saved lr, it actually points to an
instruction
> > +      * after branch. To avoid any confusion, decrement pc by 4.
> > +      * See, for example, "bl schedule" before ret_to_user().
> > +      */
> > +     pc = frame->pc - 0x4;
> > +     name = closest_symbol(pc);
> >          name_plus_offset = NULL;
> >
> >          if (bt->flags & BT_SYMBOL_OFFSET) {
> > -                sp = value_search(frame->pc, &symbol_offset);
> > +             sp = value_search(pc, &symbol_offset);
> >                  if (sp && symbol_offset)
> > -                        name_plus_offset =
> > -                                value_to_symstr(frame->pc, buf,
bt->radix);
> > +                     name_plus_offset = value_to_symstr(pc, buf,
bt->radix);
> >          }
> >
> >          fprintf(ofp, "%s#%d [%8lx] %s at %lx", level < 10 ? " " : "",
level,
> >                  frame->fp ? frame->fp : bt->stacktop -
USER_EFRAME_OFFSET,
> > -             name_plus_offset ? name_plus_offset : name, frame->pc);
> > +             name_plus_offset ? name_plus_offset : name, pc);
> >
> >       if (BT_REFERENCE_CHECK(bt))
> > -             arm64_do_bt_reference_check(bt, frame->pc, name);
> > +             arm64_do_bt_reference_check(bt, pc, name);
> >
> > -     if (module_symbol(frame->pc, NULL, &lm, NULL, 0))
> > +     if (module_symbol(pc, NULL, &lm, NULL, 0))
> >               fprintf(ofp, " [%s]", lm->mod_name);
> >
> >       fprintf(ofp, "\n");
> >
> >       if (bt->flags & BT_LINE_NUMBERS) {
> > -             get_line_number(frame->pc, buf, FALSE);
> > +             get_line_number(pc, buf, FALSE);
> >               if (strlen(buf))
> >                       fprintf(ofp, "    %s\n", buf);
> >       }
> > --
> > 2.9.0
> >
> > --
> > Crash-utility mailing list
> > Crash-utility at redhat.com
> > https://www.redhat.com/mailman/listinfo/crash-utility
> >
>
> --
> Crash-utility mailing list
> Crash-utility at redhat.com
> https://www.redhat.com/mailman/listinfo/crash-utility
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/crash-utility/attachments/20160629/f4bda3b8/attachment.htm>