Re: [Crash-utility] [PATCH v2 4/6] ppc64: handle backtrace when CPU is in an emerency stack

On 2022/07/04 14:25, Hari Bathini wrote: s/paca_ptrs/paca_ptr/ ppc64.c: In function ‘ppc64_init_paca_info’: ppc64.c:1219:8: error: ‘paca_ptrs’ undeclared (first use in this function); did you mean ‘paca_ptr’? if (!(paca_ptrs = (ulong *)calloc(kt->cpus, sizeof(ulong)))) ^~~~~~~~~ paca_ptr ppc64.c:1219:8: note: each undeclared identifier is reported only once for each function it appears in Ditto. Ditto. But these typos can be fixed when merging, no need to respin for this. Thanks, Kazu > +} > + > /* > * Verify that the kernel has made the vmemmap list available, > * and if so, stash the relevant data required to make vtop > @@ -1755,6 +1826,11 @@ ppc64_eframe_search(struct bt_info *bt_in) > addr = bt->stackbase + > roundup(SIZE(thread_info), sizeof(ulong)); > } else if (!INSTACK(addr, bt)) { > + enum emergency_stack_type estype; > + > + if ((estype = ppc64_in_emergency_stack(bt->tc->processor, addr, false))) > + ppc64_set_bt_emergency_stack(estype, bt); > + > /* > * If the user specified SP is in HW interrupt stack > * (only for tasks running on other CPUs and in 2.4 > @@ -1856,6 +1932,84 @@ ppc64_in_irqstack(ulong addr) > return 0; > } > > +/* > + * Check if the CPU is running in any of its emergency stacks. > + * Returns > + * NONE_STACK : if input is invalid or addr is not within any emergency stack. > + * EMERGENCY_STACK : if the addr is within emergency stack. > + * NMI_EMERGENCY_STACK : if the addr is within NMI emergency stack. > + * MC_EMERGENCY_STACK : if the addr is within machine check emergency stack. > + */ > +static enum emergency_stack_type > +ppc64_in_emergency_stack(int cpu, ulong addr, bool verbose) > +{ > + struct machine_specific *ms = machdep->machspec; > + ulong base, top; > + > + if (cpu < 0 || cpu >= kt->cpus) > + return NONE_STACK; > + > + if (ms->emergency_sp) { > + top = ms->emergency_sp[cpu]; > + base = top - STACKSIZE(); > + if (addr >= base && addr < top) { > + if (verbose) > + fprintf(fp, "---<Emergency Stack>---\n"); > + return EMERGENCY_STACK; > + } > + } > + > + if (ms->nmi_emergency_sp) { > + top = ms->nmi_emergency_sp[cpu]; > + base = top - STACKSIZE(); > + if (addr >= base && addr < top) { > + if (verbose) > + fprintf(fp, "---<NMI Emergency Stack>---\n"); > + return NMI_EMERGENCY_STACK; > + } > + } > + > + if (ms->mc_emergency_sp) { > + top = ms->mc_emergency_sp[cpu]; > + base = top - STACKSIZE(); > + if (addr >= base && addr < top) { > + if (verbose) > + fprintf(fp, "---<Machine Check Emergency Stack>---\n"); > + return MC_EMERGENCY_STACK; > + } > + } > + > + return NONE_STACK; > +} > + > +static void > +ppc64_set_bt_emergency_stack(enum emergency_stack_type type, struct bt_info *bt) > +{ > + struct machine_specific *ms = machdep->machspec; > + ulong top; > + > + switch (type) { > + case EMERGENCY_STACK: > + top = ms->emergency_sp[bt->tc->processor]; > + break; > + case NMI_EMERGENCY_STACK: > + top = ms->nmi_emergency_sp[bt->tc->processor]; > + break; > + case MC_EMERGENCY_STACK: > + top = ms->mc_emergency_sp[bt->tc->processor]; > + break; > + default: > + top = 0; > + break; > + } > + > + if (top) { > + bt->stackbase = top - STACKSIZE(); > + bt->stacktop = top; > + alter_stackbuf(bt); > + } > +} > + > /* > * Unroll a kernel stack. > */ > @@ -1936,10 +2090,13 @@ ppc64_back_trace_cmd(struct bt_info *bt) > static void > ppc64_back_trace(struct gnu_request *req, struct bt_info *bt) > { > - int frame = 0; > - ulong lr = 0; /* hack...need to pass in initial lr reg */ > + enum emergency_stack_type estype; > ulong newpc = 0, newsp, marker; > + int c = bt->tc->processor; > + ulong nmi_sp = 0; > int eframe_found; > + int frame = 0; > + ulong lr = 0; /* hack...need to pass in initial lr reg */ > > if (!INSTACK(req->sp, bt)) { > ulong irqstack; > @@ -1949,6 +2106,10 @@ ppc64_back_trace(struct gnu_request *req, struct bt_info *bt) > bt->stackbase = irqstack; > bt->stacktop = bt->stackbase + STACKSIZE(); > alter_stackbuf(bt); > + } else if ((estype = ppc64_in_emergency_stack(c, req->sp, true))) { > + if (estype == NMI_EMERGENCY_STACK) > + nmi_sp = req->sp; > + ppc64_set_bt_emergency_stack(estype, bt); > } else if (ms->hwintrstack) { > bt->stacktop = ms->hwintrstack[bt->tc->processor] + > sizeof(ulong); > @@ -1957,9 +2118,7 @@ ppc64_back_trace(struct gnu_request *req, struct bt_info *bt) > bt->stackbuf = ms->hwstackbuf; > alter_stackbuf(bt); > } else { > - if (CRASHDEBUG(1)) { > - fprintf(fp, "cannot find the stack info.\n"); > - } > + fprintf(fp, "cannot find the stack info.\n"); > return; > } > } > @@ -1989,13 +2148,20 @@ ppc64_back_trace(struct gnu_request *req, struct bt_info *bt) > newsp = > *(ulong *)&bt->stackbuf[newsp - bt->stackbase]; > if (!INSTACK(newsp, bt)) { > - /* > - * Switch HW interrupt stack to process's stack. > - */ > - bt->stackbase = GET_STACKBASE(bt->task); > - bt->stacktop = GET_STACKTOP(bt->task); > - alter_stackbuf(bt); > - } > + if ((estype = ppc64_in_emergency_stack(c, newsp, true))) { > + if (!nmi_sp && estype == NMI_EMERGENCY_STACK) > + nmi_sp = newsp; > + ppc64_set_bt_emergency_stack(estype, bt); > + } else { > + /* > + * Switch HW interrupt stack or emergency stack > + * to process's stack. > + */ > + bt->stackbase = GET_STACKBASE(bt->task); > + bt->stacktop = GET_STACKTOP(bt->task); > + alter_stackbuf(bt); > + } > + } > if (IS_KVADDR(newsp) && INSTACK(newsp, bt)) > newpc = *(ulong *)&bt->stackbuf[newsp + 16 - > bt->stackbase]; > @@ -2039,6 +2205,16 @@ ppc64_back_trace(struct gnu_request *req, struct bt_info *bt) > } > } > > + /* > + * NMI stack may not be re-entrant. In so, an SP in the NMI stack > + * is likely to point back to an SP within the NMI stack, in case > + * of a nested NMI. > + */ > + if (nmi_sp && nmi_sp == newsp) { > + fprintf(fp, "---<Nested NMI>---\n"); > + break; > + } > + > /* > * Some Linux 3.7 kernel threads have been seen to have > * their end-of-trace stack linkage pointer pointing > @@ -2416,6 +2592,9 @@ ppc64_get_dumpfile_stack_frame(struct bt_info *bt_in, ulong *nip, ulong *ksp) > pt_regs = (struct ppc64_pt_regs *)bt->machdep; > ur_nip = pt_regs->nip; > ur_ksp = pt_regs->gpr[1]; > + /* Print the collected regs for panic task. */ > + ppc64_print_regs(pt_regs); > + ppc64_print_nip_lr(pt_regs, 1); > } else if ((pc->flags & KDUMP) || > ((pc->flags & DISKDUMP) && > (*diskdump_flags & KDUMP_CMPRS_LOCAL))) {