Hi Tao and kuan-ying, I think the reason why my vmlinux does not contain symbol "union thread_union", may be that it was optimized during compilation. I have tried to add this line in "cpufreq_schedutil.c": > + volatile union thread_union test_for_reserve_union;
And use gdb to parser vmlinux it generated, can get the union thread_union: > (gdb) ptype union thread_union > type = union thread_union { > struct task_struct task; > unsigned long stack[4096]; >} However, when I delete that line of code, the vmlinux will not contain union thread_union: > (gdb) ptype union thread_union > No union type named thread_union. Thanks, Yeping.zheng Tao Liu <l...@redhat.com> 于2024年7月17日周三 14:21写道: > Hi Yeping, > > Thanks for the patch. > > I re-thought about the issue. I want to ask a few questions: > > 1) What's your kernel version? Because I checked the latest kernel > source 6.10, that the related code doesn't change: > > * #define IRQ_STACK_SIZE THREAD_SIZE > * > * and finding a solid usage of THREAD_SIZE is hard, but: > * > * union thread_union { > * ... > * unsigned long stack[THREAD_SIZE/sizeof(long)]; > * }; > > So I guess we can still get the THREAD_SIZE by the old crash code right? > > if (MEMBER_EXISTS("thread_union", "stack")) { > if ((sz = MEMBER_SIZE("thread_union", "stack")) > 0) > ms->irq_stack_size = sz; > > Could you please debug with your vmcore why code doesn't enter the > above if conditions, so it goes into your arm64_set_irq_stack_size() > check? > > 2) The commit message seems mis-formatted in your attachment patch. > > Thanks, > Tao Liu > > On Wed, Jul 17, 2024 at 5:16 PM yp z <wonder...@gmail.com> wrote: > > > > Hi Tao, > > Thank you for your suggestions for patch. I followed your suggestion > and provided a new patch. Please help to review it, thanks. > > > > From 831701099a7097662ddec9de464131ad50c7134b Mon Sep 17 00:00:00 2001 > > From: wonderzyp <wonder...@qq.com> > > Date: Wed, 17 Jul 2024 11:02:06 +0800 > > Subject: [PATCH] A segfault issue was observed on KASAN enabled arm64 > kernel > > due to the incorrect irq_stack_size, see the following stack trace: > > (gdb) > > bt > #0 0x00005635ac2b166b in arm64_unwind_frame (frame=0x7ffdaf35cb70, > > bt=0x7ffdaf35d430) > at arm64.c:2821 > #1 arm64_back_trace_cmd > > (bt=0x7ffdaf35d430) at arm64.c:3306 > #2 0x00005635ac27b108 in > back_trace > > (bt=bt@entry=0x7ffdaf35d430) at kernel.c:3239 > #3 0x00005635ac2880ae > in > > cmd_bt () at kernel.c:2863 > #4 0x00005635ac1f16dc in exec_command () > at > > main.c:893 > #5 0x00005635ac1f192a in main_loop () at main.c:840 > #6 > > 0x00005635ac50df81 in captured_main (data=<optimized out>) at > main.c:1284 > > > #7 gdb_main (args=<optimized out>) at main.c:1313 > #8 > 0x00005635ac50e000 > > in gdb_main_entry (argc=<optimized out>, argv=<optimized out>) > at > > main.c:1338 > #9 0x00005635ac1ea2a5 in main (argc=5, > argv=0x7ffdaf35dde8) at > > main.c:721 > > > > The issue was caused by not setting irq_stack_size correctly, and this > patch will fix the issue by two ways: > > 1. if CONFIG_IKCONFIG is set, calculate the irq_stack_size according to > kernel source code > > 2. if CONFIG_IKCONFIG is not set, get THREAD_SHIFT value by > disassembling the tbnz instruction to calculate the irq_stack_size > > > > Signed-off-by: Yeping.Zheng <yeping.zh...@nio.com> > > --- > > arm64.c | 86 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-- > > 1 file changed, 84 insertions(+), 2 deletions(-) > > > > diff --git a/arm64.c b/arm64.c > > index b3040d7..8347ba1 100644 > > --- a/arm64.c > > +++ b/arm64.c > > @@ -93,6 +93,7 @@ static void arm64_calc_VA_BITS(void); > > static int arm64_is_uvaddr(ulong, struct task_context *); > > static void arm64_calc_KERNELPACMASK(void); > > static int arm64_get_vmcoreinfo(unsigned long *vaddr, const char > *label, int base); > > +static ulong arm64_set_irq_stack_size(struct machine_specific *ms); > > > > struct kernel_range { > > unsigned long modules_vaddr, modules_end; > > @@ -2223,8 +2224,10 @@ arm64_irq_stack_init(void) > > if (MEMBER_EXISTS("thread_union", "stack")) { > > if ((sz = MEMBER_SIZE("thread_union", "stack")) > 0) > > ms->irq_stack_size = sz; > > - } else > > - ms->irq_stack_size = ARM64_IRQ_STACK_SIZE; > > + } else { > > + ulong res = arm64_set_irq_stack_size(ms); > > + ms->irq_stack_size = (res > 0) ? res : ARM64_IRQ_STACK_SIZE; > > + } > > > > machdep->flags |= IRQ_STACKS; > > > > @@ -4921,6 +4924,85 @@ static void arm64_calc_KERNELPACMASK(void) > > } > > } > > > > +static ulong arm64_set_irq_stack_size(struct machine_specific *ms) > > +{ > > + char *string; > > + int ret; > > + int kasan_thread_shift = 0; > > + int min_thread_shift; > > + ulong arm64_page_shift; > > + ulong thread_shift = 0; > > + ulong thread_size; > > + struct syment *sp; > > + const char* tbnz_str = "tbnz"; > > + > > + if (kt->ikconfig_flags & IKCONFIG_AVAIL) { > > + if ((ret = get_kernel_config("CONFIG_KASAN_GENERIC", NULL) == > IKCONFIG_Y) || > > + (ret = get_kernel_config("CONFIG_KASAN_SW_TAGS", NULL) == IKCONFIG_Y)) > { > > + kasan_thread_shift = 1; > > + } > > + min_thread_shift = 14 + kasan_thread_shift; > > + > > + if ((ret = get_kernel_config("CONFIG_VMAP_STACK", NULL)) == > IKCONFIG_Y){ > > + if ((ret = get_kernel_config("CONFIG_ARM64_PAGE_SHIFT", &string)) == > IKCONFIG_STR){ > > + arm64_page_shift = atol(string); > > + } > > + if (min_thread_shift < arm64_page_shift){ > > + thread_shift = arm64_page_shift; > > + } else { > > + thread_shift = min_thread_shift; > > + } > > + } > > + } else { > > + > > + if (!(sp = symbol_search("vectors"))) { > > + return -1; > > + } > > + > > + const char* tbnz_str = "tbnz"; > > + struct gnu_request *req; > > + req = (struct gnu_request *)GETBUF(sizeof(struct gnu_request)); > > + req->command = GNU_PASS_THROUGH; > > + req->buf = GETBUF(BUFSIZE); > > + strcat(req->buf, "x/1024i "); > > + > > + char tmp[100]; > > + sprintf(tmp, "0x%lx", sp->value); > > + strcat(req->buf, tmp); > > + req->flags = (GNU_RETURN_ON_ERROR); > > + open_tmpfile(); > > + req->fp = pc->tmpfile; > > + gdb_interface(req); > > + > > + > > + rewind(pc->tmpfile); > > + char line[BUFSIZE]; > > + while (fgets(line, BUFSIZE, pc->tmpfile) != NULL) { > > + // printf("the buffer is not null\n"); > > + char* tbnz_pos = strstr(line, tbnz_str); > > + if (tbnz_pos != NULL) { > > + char* thread_shift_pos = strstr(tbnz_pos, "#"); > > + if (sscanf(thread_shift_pos + 1, "%ld", &thread_shift) == 1) { > > + if (CRASHDEBUG(1)){ > > + error(INFO, "Detect thread shift via tbnz %ld\n", thread_shift); > > + } > > + break; > > + } > > + } > > + } > > + close_tmpfile(); > > + FREEBUF(req->buf); > > + FREEBUF(req); > > + } > > + > > + if (thread_shift == 0) { > > + return -1; > > + } > > + > > + thread_size = ((1UL) << thread_shift); > > + return thread_size; > > +} > > + > > #endif /* ARM64 */ > > > > > > -- > > 2.25.1 > > > > > > Tao Liu <l...@redhat.com> 于2024年7月16日周二 16:22写道: > >> > >> Hi Yeping, > >> > >> Thanks for the fix. > >> > >> On Thu, Jul 11, 2024 at 1:38 PM <wonder...@gmail.com> wrote: > >> > > >> > When using the crash tool to parse the ARM64 dump file with KASAN > enabled, I found that using the bt -a command will cause this tool to > crash, the following is the backtrace infomation. > >> > > >> > (gdb) bt > >> > #0 0x00005635ac2b166b in arm64_unwind_frame (frame=0x7ffdaf35cb70, > bt=0x7ffdaf35d430) > >> > at arm64.c:2821 > >> > #1 arm64_back_trace_cmd (bt=0x7ffdaf35d430) at arm64.c:3306 > >> > #2 0x00005635ac27b108 in back_trace (bt=bt@entry=0x7ffdaf35d430) at > kernel.c:3239 > >> > #3 0x00005635ac2880ae in cmd_bt () at kernel.c:2863 > >> > #4 0x00005635ac1f16dc in exec_command () at main.c:893 > >> > #5 0x00005635ac1f192a in main_loop () at main.c:840 > >> > #6 0x00005635ac50df81 in captured_main (data=<optimized out>) at > main.c:1284 > >> > #7 gdb_main (args=<optimized out>) at main.c:1313 > >> > #8 0x00005635ac50e000 in gdb_main_entry (argc=<optimized out>, > argv=<optimized out>) > >> > at main.c:1338 > >> > #9 0x00005635ac1ea2a5 in main (argc=5, argv=0x7ffdaf35dde8) at > main.c:721 > >> > Eventually, I found that it was may caused by not setting > irq_stack_size properly, and provide this patch to solve it. > >> > > >> > >> Could you please re-draft your commit message? The original one looks > >> informal. E.g: > >> > >> A segfault issue was observed on KASAN enabled arm64 kernel due to the > >> incorrect irq_stack_size, see the following stack trace: > >> ... > >> The issue was caused by ...., and this patch will fix the issue by .... > >> > >> > > >> > From 34b28aa8c11e77d20adec4f7705a14d239c8a55f Mon Sep 17 00:00:00 2001 > >> > From: wonderzyp <wonder...@qq.com> > >> > Date: Mon, 8 Jul 2024 20:11:38 +0800 > >> > Subject: [PATCH 1131/1131] set_arm64_irq_stack_size > >> > > >> > Signed-off-by: Yeping Zheng <wonder...@gmail.com> > >> > --- > >> > arm64.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++-- > >> > 1 file changed, 47 insertions(+), 2 deletions(-) > >> > > >> > diff --git a/arm64.c b/arm64.c > >> > index b3040d7..39d891b 100644 > >> > --- a/arm64.c > >> > +++ b/arm64.c > >> > @@ -93,6 +93,7 @@ static void arm64_calc_VA_BITS(void); > >> > static int arm64_is_uvaddr(ulong, struct task_context *); > >> > static void arm64_calc_KERNELPACMASK(void); > >> > static int arm64_get_vmcoreinfo(unsigned long *vaddr, const char > *label, int base); > >> > +static ulong arm64_set_irq_stack_size(struct machine_specific *ms); > >> > > >> > struct kernel_range { > >> > unsigned long modules_vaddr, modules_end; > >> > @@ -2223,8 +2224,14 @@ arm64_irq_stack_init(void) > >> > if (MEMBER_EXISTS("thread_union", "stack")) { > >> > if ((sz = MEMBER_SIZE("thread_union", > "stack")) > 0) > >> > ms->irq_stack_size = sz; > >> > - } else > >> > - ms->irq_stack_size = ARM64_IRQ_STACK_SIZE; > >> > + } else { > >> > + ulong res = arm64_set_irq_stack_size(ms); > >> > + if (res > 0){ > >> > + ms->irq_stack_size = res; > >> > + } else { > >> > + ms->irq_stack_size = > ARM64_IRQ_STACK_SIZE; > >> > + } > >> > + } > >> > > >> > machdep->flags |= IRQ_STACKS; > >> > > >> > @@ -4921,6 +4928,44 @@ static void arm64_calc_KERNELPACMASK(void) > >> > } > >> > } > >> > > >> > +static ulong arm64_set_irq_stack_size(struct machine_specific *ms) > >> > +{ > >> > + char *string; > >> > + int ret; > >> > + int KASAN_THREAD_SHIFT = 0; > >> > + int MIN_THREAD_SHIFT; > >> > + ulong ARM64_PAGE_SHIFT; > >> > + ulong THREAD_SHIFT = 0; > >> > + ulong THREAD_SIZE; > >> > >> I guess the upper case of variable names is not encouraged, though it > >> is the variable that comes from kernel config file. > >> > >> > + if (kt->ikconfig_flags & IKCONFIG_AVAIL) { > >> > + if ((ret = get_kernel_config("CONFIG_KASAN_GENERIC", > NULL) == IKCONFIG_Y) || > >> > + (ret = > get_kernel_config("CONFIG_KASAN_SW_TAGS", NULL) == IKCONFIG_Y)) { > >> > + KASAN_THREAD_SHIFT = 1; > >> > + } > >> > + } > >> > + MIN_THREAD_SHIFT = 14 + KASAN_THREAD_SHIFT; > >> > + > >> > + if (kt->ikconfig_flags & IKCONFIG_AVAIL) { > >> > >> Could the if condition be merged with the prior one? > >> > >> > + if ((ret = get_kernel_config("CONFIG_VMAP_STACK", > NULL)) == IKCONFIG_Y){ > >> > + if ((ret = > get_kernel_config("CONFIG_ARM64_PAGE_SHIFT", &string)) == IKCONFIG_STR){ > >> > + ARM64_PAGE_SHIFT = atol(string); > >> > + } > >> > + if (MIN_THREAD_SHIFT < ARM64_PAGE_SHIFT){ > >> > + THREAD_SHIFT = ARM64_PAGE_SHIFT; > >> > + } else { > >> > + THREAD_SHIFT = MIN_THREAD_SHIFT; > >> > + } > >> > + } > >> > + } > >> > + > >> > + if (THREAD_SHIFT == 0) { > >> > + return -1; > >> > + } > >> > + > >> > + THREAD_SIZE = ((1UL) << THREAD_SHIFT); > >> > + return THREAD_SIZE; > >> > +} > >> > >> I'm OK with the approach above, since it directly came from the kernel > >> source. However I'm not a fan of checking kernel configs, there might > >> be kernels which are compiled without CONFIG_IKCONFIG. > >> > >> Could we add an approach here, to get the value from disassembly when > >> CONFIG_IKCONFIG is negative? > >> > >> kernel source: arch/arm64/kernel/entry.S: > >> > >> .macro kernel_ventry, el:req, ht:req, regsize:req, label:req > >> .... > >> add sp, sp, x0 // sp' = sp + x0 > >> sub x0, sp, x0 // x0' = sp' - x0 = (sp + x0) - x0 = sp > >> tbnz x0, #THREAD_SHIFT, 0f <<<<<<<< > >> > >> $ objdump -d vmlinux > >> ... > >> ffff800080010800 <vectors>: > >> ffff800080010800: d10543ff sub sp, sp, #0x150 > >> ffff800080010804: 8b2063ff add sp, sp, x0 > >> ffff800080010808: cb2063e0 sub x0, sp, x0 > >> ffff80008001080c: 37800080 tbnz w0, #16, > >> ffff80008001081c <vectors+0x1c> <<<<<<<<<< > >> > >> It is easy to get the THREAD_SHIFT value by disassembling the tbnz > >> instruction. What do you think @Lianbo Jiang > >> > >> Thanks, > >> Tao Liu > >> > >> > + > >> > #endif /* ARM64 */ > >> > > >> > > >> > -- > >> > 2.25.1 > >> > -- > >> > Crash-utility mailing list -- devel@lists.crash-utility.osci.io > >> > To unsubscribe send an email to > devel-le...@lists.crash-utility.osci.io > >> > https://${domain_name}/admin/lists/devel.lists.crash-utility.osci.io/ > >> > Contribution Guidelines: https://github.com/crash-utility/crash/wiki > >> > >
-- Crash-utility mailing list -- devel@lists.crash-utility.osci.io To unsubscribe send an email to devel-le...@lists.crash-utility.osci.io https://${domain_name}/admin/lists/devel.lists.crash-utility.osci.io/ Contribution Guidelines: https://github.com/crash-utility/crash/wiki