On Tue, Oct 4, 2022 at 10:22 PM Lianbo Jiang <liji...@redhat.com> wrote:
>
> Currently crash will fail and then exit, if the initialization of
> the emergency stacks information fails. In real customer environments,
> sometimes, a vmcore may be partially damaged, although such vmcores
> are rare. For example:
>
> # ./crash ../3.10.0-1127.18.2.el7.ppc64le/vmcore
> ../3.10.0-1127.18.2.el7.ppc64le/vmlinux -s
> crash: invalid kernel virtual address: 38 type: "paca->emergency_sp"
> #
>
> Lets try to keep loading vmcore if such issues happen, so call
> the readmem() with the RETURN_ON_ERROR instead of FAULT_ON_ERROR,
> which allows the crash move on.
>
> Reported-by: Dave Wysochanski <dwyso...@redhat.com>
> Signed-off-by: Lianbo Jiang <liji...@redhat.com>
> ---
> ppc64.c | 12 ++++++------
> 1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/ppc64.c b/ppc64.c
> index 4ea1f7c0c6f8..f94b402ec64d 100644
> --- a/ppc64.c
> +++ b/ppc64.c
> @@ -1224,13 +1224,13 @@ ppc64_init_paca_info(void)
> ulong paca_loc;
>
> readmem(symbol_value("paca_ptrs"), KVADDR, &paca_loc,
> sizeof(void *),
> - "paca double pointer", FAULT_ON_ERROR);
> + "paca double pointer", RETURN_ON_ERROR);
> readmem(paca_loc, KVADDR, paca_ptr, sizeof(void *) * kt->cpus,
> - "paca pointers", FAULT_ON_ERROR);
> + "paca pointers", RETURN_ON_ERROR);
> } else if (symbol_exists("paca") &&
> (get_symbol_type("paca", NULL, NULL) == TYPE_CODE_PTR)) {
> readmem(symbol_value("paca"), KVADDR, paca_ptr, sizeof(void
> *) * kt->cpus,
> - "paca pointers", FAULT_ON_ERROR);
> + "paca pointers", RETURN_ON_ERROR);
> } else {
> free(paca_ptr);
> return;
> @@ -1245,7 +1245,7 @@ ppc64_init_paca_info(void)
> for (i = 0; i < kt->cpus; i++)
> readmem(paca_ptr[i] + offset, KVADDR,
> &ms->emergency_sp[i],
> sizeof(void *), "paca->emergency_sp",
> - FAULT_ON_ERROR);
> + RETURN_ON_ERROR);
> }
>
> if (MEMBER_EXISTS("paca_struct", "nmi_emergency_sp")) {
> @@ -1256,7 +1256,7 @@ ppc64_init_paca_info(void)
> for (i = 0; i < kt->cpus; i++)
> readmem(paca_ptr[i] + offset, KVADDR,
> &ms->nmi_emergency_sp[i],
> sizeof(void *), "paca->nmi_emergency_sp",
> - FAULT_ON_ERROR);
> + RETURN_ON_ERROR);
> }
>
> if (MEMBER_EXISTS("paca_struct", "mc_emergency_sp")) {
> @@ -1267,7 +1267,7 @@ ppc64_init_paca_info(void)
> for (i = 0; i < kt->cpus; i++)
> readmem(paca_ptr[i] + offset, KVADDR,
> &ms->mc_emergency_sp[i],
> sizeof(void *), "paca->mc_emergency_sp",
> - FAULT_ON_ERROR);
> + RETURN_ON_ERROR);
> }
>
> free(paca_ptr);
> --
> 2.37.1
>
Consider adding a 'Fixes' tag for the patch that introduced this
problem, per the bisect
in https://bugzilla.redhat.com/show_bug.cgi?id=2127525
Fixes: cdd57e8b16ab ("ppc64: handle backtrace when CPU is in an
emergency stack")
Other than that, I tested this and now on the vmcore in question,
crash loads ok.
I get a lot of the below "invalid kernel virtual address", but I think
it is fine:
...
crash: invalid kernel virtual address: 2d0 type: "paca->mc_emergency_sp"
crash> quit
Tested-and-Reviewed-by: Dave Wysochanski <dwyso...@redhat.com>
Good job Lianbo!
--
Crash-utility mailing list
Crash-utility@redhat.com
https://listman.redhat.com/mailman/listinfo/crash-utility
Contribution Guidelines: https://github.com/crash-utility/crash/wiki
