Re: [Crash-utility] ppc64 specific changes to support kdump vmcore

[David Wilder <dwilder@xxxxxxxxxx>]

Dave Anderson wrote:

> Haren Myneni wrote:
>
>  
>
> > Reposting this patch as it did not made to crash mailing list. Dave,
> > Thanks for adding my different e-mail address to subscriber list.
> >
> >    
>
> No problem -- I prefer to keep this list subscriber-only to avoid the spam
> you see on other lists...
>
>  
>
> > Dave,
> >
> > Attaching a patch which contains ppc64 specific changes to read kdump
> > vmcore. As we are saving pt_regs for all cpus, it reads them from vmcore
> > instead of looking for specific symbols, used for PPC64 netdump/diskdump
> > vmcore. Also, prints regs before any active backtrace. Whereas for
> > netdump vmcore, regs will be displayed anyway as part of exception frame
> > since the 'bt' command displays from the top frame.
> > Verified with "Make Warn"
> >
> > Please let me know if you have any comments.
> >
> >    
>
> This patch looks good.  No comment...        ;-)
>
>  
>
> > As Badari mentioned, when the sparsemem is enabled, pg_dat->node_mem_map
> > member does not exists. mem_map for each node is scattered across
> > multiple sections and is not contiguous. Therefore, some detailed
> > changes are needed.  Until we fix this issue, can we include some hack
> > so that users can use other commands for vmcore analysis.
> > The fix will be:
> >   if (MEMBER_EXISTS("pglist_data", "node_mem_map"))
> >      readmem(pgdat+OFFSET(pglist_data_node_mem_map), KVADDR,
> >                    &node_mem_map, sizeof(ulong),
> >                    "node_mem_map", FAULT_ON_ERROR);
> >
> > [I noticed your response to Badari's posting. Since the sparsemem is
> > effected only for powerpc at this point, we thought, user can use the
> > limited functionality with the above change. (Ex: bt) until we have the
> > complete fix. At least we will be having crash tool available for kdump
> > vmore on ppc64. The sparsemem issue will be fixed soon. Ok, please
> > ignore this if you prefer to wait for the complete fix]
> >
> >    
>
> I was tinkering around with this by simulating the bogus setting of
> "node_mem_map" above, and I'm surprised that you can get as much
> functionality as crash does give you.  Most people probably wouldn't even
> notice.  Anyway, I was about to change my mind, and put the EXISTS
> check in for now -- but along with a non-fatal error message that would be
> annoying enough for you guys to whip up a real fix.  I'll still do that,
> unless by "fixed soon", you mean in a day or two?
>
> Dave
>
>
>  
Dave-
I am working on the sparse mem support in crash, It is going to take 
longer than a few days.
Sorry I can't give you a time estimate yet. Please go with Haren patch. 
I will update you
when I get sparse memory is working working.

>  
>
> > Thanks
> > Haren
> >
> >  ------------------------------------------------------------------------------------------------------------------------
> > diff -Naurp crash-4.0-2.20.orig/netdump.c crash-4.0-2.20/netdump.c
> > --- crash-4.0-2.20.orig/netdump.c       2006-02-11 16:50:28.000000000 -0800
> > +++ crash-4.0-2.20/netdump.c    2006-02-11 16:50:58.000000000 -0800
> > @@ -1680,7 +1680,13 @@ get_netdump_regs_ppc64(struct bt_info *b
> >        Elf64_Nhdr *note;
> >        size_t len;
> >
> > -       if (bt->task == tt->panic_task) {
> > +       if ((bt->task == tt->panic_task) ||
> > +               (is_task_active(bt->task) && nd->num_prstatus_notes > 1)) {
> > +               /*
> > +                * Registers are saved during the dump process for the
> > +                * panic task. Whereas in kdump, regs are captured for all
> > +                * CPUs if they responded to an IPI.
> > +                */
> >                 if (nd->num_prstatus_notes > 1)
> >                         note = (Elf64_Nhdr *)
> >                                 nd->nt_prstatus_percpu[bt->tc->processor];
> > diff -Naurp crash-4.0-2.20.orig/ppc64.c crash-4.0-2.20/ppc64.c
> > --- crash-4.0-2.20.orig/ppc64.c 2006-02-11 16:50:32.000000000 -0800
> > +++ crash-4.0-2.20/ppc64.c      2006-02-13 16:10:56.000000000 -0800
> > @@ -1420,20 +1420,11 @@ ppc64_check_eframe(struct ppc64_pt_regs
> >        return NULL;
> > }
> >
> > -/*
> > - *  Print exception frame information for ppc64
> > - */
> > static void
> > -ppc64_print_eframe(char *efrm_str, struct ppc64_pt_regs *regs,
> > -               struct bt_info *bt)
> > +ppc64_print_regs(struct ppc64_pt_regs *regs)
> > {
> >        int i;
> >
> > -       if (BT_REFERENCE_CHECK(bt))
> > -               return;
> > -
> > -        fprintf(fp, " %s  [%lx] exception frame:", efrm_str, regs->trap);
> > -
> >         /* print out the gprs... */
> >         for(i=0; i<32; i++) {
> >                 if(!(i % 3))
> > @@ -1465,9 +1456,66 @@ ppc64_print_eframe(char *efrm_str, struc
> >         fprintf(fp, "DAR: %016lx\n", regs->dar);
> >         fprintf(fp, " DSISR: %016lx ", regs->dsisr);
> >         fprintf(fp, "    Syscall Result: %016lx\n", regs->result);
> > +}
> > +
> > +/*
> > + * Print the exception frame information
> > + */
> > +static void
> > +ppc64_print_eframe(char *efrm_str, struct ppc64_pt_regs *regs,
> > +                       struct bt_info *bt)
> > +{
> > +       if (BT_REFERENCE_CHECK(bt))
> > +               return;
> > +
> > +       fprintf(fp, " %s  [%lx] exception frame:", efrm_str, regs->trap);
> > +       ppc64_print_regs(regs);
> >        fprintf(fp, "\n");
> > }
> >
> > +/*
> > + * get SP and IP from the saved ptregs.
> > + */
> > +static int
> > +ppc64_kdump_stack_frame(struct bt_info *bt_in, ulong *nip, ulong *ksp)
> > +{
> > +       struct ppc64_pt_regs *pt_regs;
> > +       unsigned long unip;
> > +
> > +       pt_regs = (struct ppc64_pt_regs *)bt_in->machdep;
> > +       if (!pt_regs->gpr[1]) {
> > +               /*
> > +                * Not collected regs. May be the corresponding CPU not
> > +                * responded to an IPI.
> > +                */
> > +               fprintf(fp, "%0lx: GPR1 register value (SP) was not saved\n",
> > +                       bt_in->task);
> > +               return FALSE;
> > +       }
> > +       *ksp = pt_regs->gpr[1];
> > +       readmem(*ksp+16, KVADDR, &unip, sizeof(ulong), "Regs NIP value",
> > +               FAULT_ON_ERROR);
> > +       *nip = unip;
> > +
> > +       if (bt_in->flags &&
> > +       ((BT_TEXT_SYMBOLS|BT_TEXT_SYMBOLS_PRINT|BT_TEXT_SYMBOLS_NOPRINT)))
> > +               return TRUE;
> > +
> > +       /*
> > +        * Print the collected regs for the active task
> > +        */
> > +       ppc64_print_regs(pt_regs);
> > +
> > +       fprintf(fp, " NIP [%016lx] %s\n", pt_regs->nip,
> > +               closest_symbol(pt_regs->nip));
> > +       if (unip != pt_regs->link)
> > +               fprintf(fp, " LR  [%016lx] %s\n", pt_regs->link,
> > +                       closest_symbol(pt_regs->link));
> > +
> > +       fprintf(fp, "\n");
> > +
> > +       return TRUE;
> > +}
> >
> > /*
> >  *  Get the starting point for the active cpus in a diskdump/netdump.
> > @@ -1485,12 +1533,18 @@ ppc64_get_dumpfile_stack_frame(struct bt
> >         ulong ur_ksp = 0;
> >        int check_hardirq, check_softirq;
> >        int check_intrstack = TRUE;
> > +       struct ppc64_pt_regs *pt_regs;
> > +
> > +       /*
> > +        * For the kdump vmcore, Use SP and IP values that are saved in ptregs.
> > +        */
> > +       if (pc->flags && KDUMP)
> > +               return ppc64_kdump_stack_frame(bt_in, nip, ksp);
> >
> >         bt = &bt_local;
> >         BCOPY(bt_in, bt, sizeof(struct bt_info));
> >         ms = machdep->machspec;
> >         ur_nip = ur_ksp = 0;
> > -       struct ppc64_pt_regs *pt_regs;
> >
> >        panic_task = tt->panic_task == bt->task ? TRUE : FALSE;
> >
> >
> >  ------------------------------------------------------------------------------------------------------------------------
> > --
> > Crash-utility mailing list
> > Crash-utility@xxxxxxxxxx
> > https://www.redhat.com/mailman/listinfo/crash-utility
> >    
>
> --
> Crash-utility mailing list
> Crash-utility@xxxxxxxxxx
> https://www.redhat.com/mailman/listinfo/crash-utility
>
>
>  


--
David Wilder
IBM Linux Change Team
dwilder@xxxxxxxxxx
(503)578-3789