On Tue, Oct 4, 2022 at 10:22 PM Lianbo Jiang <lijiang@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 <dwysocha@redhat.com>
> Signed-off-by: Lianbo Jiang <lijiang@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 <dwysocha@redhat.com>

Good job Lianbo!