[Crash-utility] [RFC PATCH 08/15] Support "mod -d|-D" options

[lijiang]

On Fri, Jun 2, 2023 at 10:18 AM HAGIO KAZUHITO(萩尾 一仁) <k-hagio-ab at nec.com>
wrote:

> On 2023/05/30 21:30, lijiang wrote:
> > On Mon, May 29, 2023 at 9:03 AM HAGIO KAZUHITO(萩尾 一仁) <
> k-hagio-ab at nec.com>
> > wrote:
> >
> >>>
> >>> +                                       for (j = MOD_TEXT; j <
> >>>> MOD_MEM_NUM_TYPES; j++) {
> >>>> +                                               lm->load_symtable[j] =
> >>>> NULL;
> >>>> +                                               lm->load_symend[j] =
> >> NULL;
> >>>> +                                       }
> >>>> +                               } else { /* somehow this function runs
> >> for
> >>>> unloaded modules */
> >>>>
> >>>
> >>> Could you please explain why it may get into the "else" code path? And
> >> why
> >>> do we need to handle this situation for now? But not needed before the
> >>> module memory changes.
> >>
> >> This path is used by "mod -d|-D" and reinit_modules() for unloaded
> modules.
> >>
> >> The old procedure always does the cleanup and reinstall of the hash
> >> entries, so kept it as it is.
> >>
> >
> > I tried to add some debug information in the above code path, but did not
> > observe any output(via the mod -d/-D). It's strange.
>
> With this change,
>
> --- a/symbols.c
> +++ b/symbols.c
> @@ -13951,6 +13951,7 @@ delete_load_module(ulong base_addr)
>                                                  lm->load_symend[j] = NULL;
>                                          }
>                                  } else { /* somehow this function runs
> for unloaded modules */
> +                                       fprintf(fp, "in the else block
> (-D)\n");
>                                          for (j = MOD_TEXT; j <
> MOD_MEM_NUM_TYPES; j++) {
>                                                  if (!lm->symtable[j])
>                                                          continue;
> @@ -14016,6 +14017,7 @@ delete_load_module(ulong base_addr)
>                                                  lm->load_symend[j] = NULL;
>                                          }
>                                  } else { /* somehow this function runs
> for unloaded modules */
> +                                       fprintf(fp, "in the else block
> (-d)\n");
>                                          for (j = MOD_TEXT; j <
> MOD_MEM_NUM_TYPES; j++) {
>                                                  if (!lm->symtable[j])
>                                                          continue;
>
> These should be printed:
>
> crash-6.4> mod -s dm_mod
>       MODULE       NAME                     TEXT_BASE         SIZE  OBJECT
> FILE
> ffffffffc042fd00  dm_mod                ffffffffc0418000   237568
> dm-mod.ko
> crash-6.4> mod -d dm_mod
> crash-6.4> mod -d dm_mod
>

Ah, I know why I did not see any output(via the mod -d/-D), because I only
ran the "mod -d/-D" *once*. And for now I can see the debug information
after running the mod -d/-D *twice*.

Thank you for the explanation, Kazu.


> in the else block (-d)
> crash-6.4> mod -d dm_log
> in the else block (-d)
> crash-6.4> mod -D | head
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
> in the else block (-D)
>
> Thanks,
> Kazu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/crash-utility/attachments/20230602/f567ed00/attachment.htm>