----- Original Message -----
> Hi Dave et. al.,
>
> For ARM {k,net}dumps panic tasks we go to great lengths to locate
> the prstatus data (crash registers) in
>
> netdump.c:get_netdump_regs_arm()
>
> In fact we rely so hard on them having been read from the CORE
> ELF notes and put in either
>
> nd->nt_prstatus_percpu
>
> or
>
> nd->nt_prstatus
>
> that there's a segfault if that's not the case. The thing is,
> once we've gone through all that trouble of finding them, and
> making sure bt->machdep points to them,
> arm.c:arm_get_stack_frame() is called, which ends up getting the
> crash registers from
>
> machdep->machspec->crash_task_regs
>
> anyway. These are read directly from within the PT_LOAD segments
> ignoring the contents any ELF notes.
>
> I suggest changing get_netdump_regs_arm() to:
>
> static void
> get_netdump_regs_arm(struct bt_info *bt, ulong *eip, ulong *esp)
> {
> machdep->get_stack_frame(bt, eip, esp);
> }
>
> If we're not going to use the ELF notes, why fail if they're not
> there?
I'm not sure what you mean about reading "directly from within the
PT_LOAD segments", but I agree with you that the preliminary "if"
section of get_netdump_regs_arm() which sets bt->machdep is a
no-op -- given that arm_get_dumpfile_stack_frame() just ignores
it, and resets bt->machdep from that ARM-specific ms->crash_task_regs
area.
Mika, you agree?
Dave
> Regards
> Per
--
Crash-utility mailing list
Crash-utility@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/crash-utility
