LoongArch: Fix errors about "cannot find unwind pc"

[PATCH v2 2/4] LoongArch: Handle percpu handler address for ORC unwinder

Posted by Tiezhu Yang 2 weeks, 5 days ago

After commit 4cd641a79e69 ("LoongArch: Remove unnecessary checks for
ORC unwinder"), the system can not boot normally under some configs,
there are many error messages "cannot find unwind pc".

The kernel boots normally with the defconfig, so no problem found out
at the first time. Here is one way to reproduce:

  cd linux
  make mrproper defconfig -j"$(nproc)"
  scripts/config -e KASAN
  make olddefconfig all -j"$(nproc)"
  sudo make modules_install
  sudo make install
  sudo reboot

The address that can not unwind is not a valid kernel address which is
between "pcpu_handlers[cpu]" and "pcpu_handlers[cpu] + vec_sz" due to
the code of eentry was copied to the new area of pcpu_handlers[cpu] in
setup_tlb_handler(), handle this special case to get the valid address
to unwind normally.

Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
---
 arch/loongarch/include/asm/setup.h |  2 ++
 arch/loongarch/kernel/unwind_orc.c | 15 +++++++++++++++
 2 files changed, 17 insertions(+)

diff --git a/arch/loongarch/include/asm/setup.h b/arch/loongarch/include/asm/setup.h
index 3c2fb16b11b6..79eb63e749f9 100644
--- a/arch/loongarch/include/asm/setup.h
+++ b/arch/loongarch/include/asm/setup.h
@@ -14,6 +14,8 @@
 
 extern unsigned long eentry;
 extern unsigned long tlbrentry;
+extern unsigned long pcpu_handlers[NR_CPUS];
+extern long exception_handlers[VECSIZE * 128 / sizeof(long)];
 extern char init_command_line[COMMAND_LINE_SIZE];
 extern void tlb_init(int cpu);
 extern void cpu_cache_init(void);
diff --git a/arch/loongarch/kernel/unwind_orc.c b/arch/loongarch/kernel/unwind_orc.c
index d6b3688a1ce9..e0c0dc22eb08 100644
--- a/arch/loongarch/kernel/unwind_orc.c
+++ b/arch/loongarch/kernel/unwind_orc.c
@@ -351,6 +351,21 @@ EXPORT_SYMBOL_GPL(unwind_start);
 static inline unsigned long bt_address(unsigned long ra)
 {
 	extern unsigned long eentry;
+#if defined(CONFIG_NUMA) && !defined(CONFIG_PREEMPT_RT)
+	int cpu;
+	int vec_sz = sizeof(exception_handlers);
+
+	for_each_possible_cpu(cpu) {
+		if (!pcpu_handlers[cpu])
+			continue;
+
+		if (ra >= pcpu_handlers[cpu] &&
+		    ra < pcpu_handlers[cpu] + vec_sz) {
+			ra = eentry + (ra - pcpu_handlers[cpu]);
+			break;
+		}
+	}
+#endif
 
 	if (ra >= eentry && ra < eentry +  EXCCODE_INT_END * VECSIZE) {
 		unsigned long func;
-- 
2.42.0

Re: [PATCH v2 2/4] LoongArch: Handle percpu handler address for ORC unwinder

Posted by Huacai Chen 1 week, 5 days ago

Hi, Tiezhu,

On Wed, Jan 21, 2026 at 5:59 PM Tiezhu Yang <yangtiezhu@loongson.cn> wrote:
>
> After commit 4cd641a79e69 ("LoongArch: Remove unnecessary checks for
> ORC unwinder"), the system can not boot normally under some configs,
> there are many error messages "cannot find unwind pc".
Unfortunately, after this series sometimes there are still error
messages, such as:
cannot find unwind pc at ffff80000231fe1c

Obviously, this is no longer related to per-cpu handler, and from the
call stack it seems related to BPF:

[   75.025389] CPU: 0 UID: 0 PID: 175 Comm: systemd-udevd Not tainted
6.19-rc7+ #721
[   75.025409] Hardware name: Loongson Loongson-3A6000-7A2000-EVB/EVB,
BIOS Loongson-FwSdk-V5.0.0343-stable2511 01/13/26 20:10:42
[   75.025422] Stack : 0000000000000000 0000000000000000
9000000000223c20 9000000116460000
[   75.025455]         900000010009f5e0 900000010009f5e8
0000000000000000 900000010009f728
[   75.025487]         900000010009f720 900000010009f720
900000010009f490 0000000000000001
[   75.025519]         0000000000000001 900000010009f5e8
9e6a35e67549a6db 000000fffffffffe
[   75.025551]         0000010000000000 0000000000000022
0000000fffffffff ffff800000000000
[   75.025582]         0000800000000000 9000000000000000
00000003f3fe4000 900000010009f9e0
[   75.025615]         0000000000000000 9000000002fd9000
0000000000000000 9000000002c15220
[   75.025646]         9000000002ea5820 900000010009f848
90000000040c1320 ffff80000231fe1c
[   75.025679]         90000000003a8f20 0000000000000000
9000000000223c3c 000055557726cb04
[   75.025710]         00000000000000b0 0000000000000004
0000000000000000 0000000000071c1d
[   75.025739]         ...
[   75.025751] Call Trace:
[   75.025756] [<9000000000223c3c>] show_stack+0x5c/0x190
[   75.025782] [<9000000001f2cc34>] dump_stack_lvl+0x6c/0x9c
[   75.025807] [<9000000000236714>] unwind_next_frame+0x764/0xa70
[   75.025837] [<9000000000231e14>] arch_stack_walk+0xd4/0x1b0
[   75.025863] [<900000000038aa04>] stack_trace_save+0x54/0x90
[   75.025892] [<9000000000612f80>] kasan_save_stack+0x30/0x60
[   75.025914] [<9000000000615a68>] __kasan_record_aux_stack+0x108/0x130
[   75.025941] [<90000000002b1c40>] task_work_add+0x100/0x300
[   75.025968] [<90000000002d3cd8>] sched_tick+0x158/0x3e0
[   75.025987] [<900000000038fa40>] update_process_times+0xf0/0x160
[   75.026011] [<90000000003a9018>] tick_nohz_handler+0xf8/0x260
[   75.026038] [<90000000003908dc>] __hrtimer_run_queues+0x20c/0x330
[   75.026062] [<9000000000392190>] hrtimer_interrupt+0x1a0/0x3f0
[   75.026086] [<90000000002285ac>] constant_timer_interrupt+0x5c/0x80
[   75.026110] [<900000000034358c>] __handle_irq_event_percpu+0x6c/0x190
[   75.026139] [<90000000003436c4>] handle_irq_event_percpu+0x14/0xa0
[   75.026167] [<900000000034d66c>] handle_percpu_irq+0x8c/0xe0
[   75.026189] [<9000000000342670>] generic_handle_domain_irq+0x50/0xc0
[   75.026217] [<9000000000ffaad4>] handle_cpu_irq+0x54/0x90
[   75.026245] [<9000000001f2d128>] handle_loongarch_irq+0x28/0x40
[   75.026269] [<9000000001f2d1bc>] do_vint+0x7c/0xf0
[   75.026289] [<900000000023fa20>] mem_to_shadow+0x0/0x2b0
[   75.026313] [<9000000000615af4>] kasan_check_range+0x64/0x2c0
[   75.026336] [<9000000001b2b7a0>] bpf_skb_load_helper_32+0x40/0x220





Huacai

>
> The kernel boots normally with the defconfig, so no problem found out
> at the first time. Here is one way to reproduce:
>
>   cd linux
>   make mrproper defconfig -j"$(nproc)"
>   scripts/config -e KASAN
>   make olddefconfig all -j"$(nproc)"
>   sudo make modules_install
>   sudo make install
>   sudo reboot
>
> The address that can not unwind is not a valid kernel address which is
> between "pcpu_handlers[cpu]" and "pcpu_handlers[cpu] + vec_sz" due to
> the code of eentry was copied to the new area of pcpu_handlers[cpu] in
> setup_tlb_handler(), handle this special case to get the valid address
> to unwind normally.
>
> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> ---
>  arch/loongarch/include/asm/setup.h |  2 ++
>  arch/loongarch/kernel/unwind_orc.c | 15 +++++++++++++++
>  2 files changed, 17 insertions(+)
>
> diff --git a/arch/loongarch/include/asm/setup.h b/arch/loongarch/include/asm/setup.h
> index 3c2fb16b11b6..79eb63e749f9 100644
> --- a/arch/loongarch/include/asm/setup.h
> +++ b/arch/loongarch/include/asm/setup.h
> @@ -14,6 +14,8 @@
>
>  extern unsigned long eentry;
>  extern unsigned long tlbrentry;
> +extern unsigned long pcpu_handlers[NR_CPUS];
> +extern long exception_handlers[VECSIZE * 128 / sizeof(long)];
>  extern char init_command_line[COMMAND_LINE_SIZE];
>  extern void tlb_init(int cpu);
>  extern void cpu_cache_init(void);
> diff --git a/arch/loongarch/kernel/unwind_orc.c b/arch/loongarch/kernel/unwind_orc.c
> index d6b3688a1ce9..e0c0dc22eb08 100644
> --- a/arch/loongarch/kernel/unwind_orc.c
> +++ b/arch/loongarch/kernel/unwind_orc.c
> @@ -351,6 +351,21 @@ EXPORT_SYMBOL_GPL(unwind_start);
>  static inline unsigned long bt_address(unsigned long ra)
>  {
>         extern unsigned long eentry;
> +#if defined(CONFIG_NUMA) && !defined(CONFIG_PREEMPT_RT)
> +       int cpu;
> +       int vec_sz = sizeof(exception_handlers);
> +
> +       for_each_possible_cpu(cpu) {
> +               if (!pcpu_handlers[cpu])
> +                       continue;
> +
> +               if (ra >= pcpu_handlers[cpu] &&
> +                   ra < pcpu_handlers[cpu] + vec_sz) {
> +                       ra = eentry + (ra - pcpu_handlers[cpu]);
> +                       break;
> +               }
> +       }
> +#endif
>
>         if (ra >= eentry && ra < eentry +  EXCCODE_INT_END * VECSIZE) {
>                 unsigned long func;
> --
> 2.42.0
>

[PATCH v2 1/4] LoongArch: Use %px to print unmodified unwinding address
[PATCH v2 2/4] LoongArch: Handle percpu handler address for ORC unwinder
[PATCH v2 3/4] LoongArch: Guard percpu handler under !CONFIG_PREEMPT_RT
[PATCH v2 4/4] LoongArch: Remove some extern variables in source files