| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAAhV-H6pQytDg_fKE7rpC2OogWYGEsHUxUMKLqsvtfSPWMWvaA@mail.gmail.com>
Date: Wed, 26 Nov 2025 17:27:13 +0800
From: Huacai Chen <chenhuacai@...nel.org>
To: Tiezhu Yang <yangtiezhu@...ngson.cn>
Cc: Jinyang He <hejinyang@...ngson.cn>, Xi Zhang <zhangxi@...inos.cn>,
Xianglai Li <lixianglai@...ngson.cn>, loongarch@...ts.linux.dev,
linux-kernel@...r.kernel.org, Bibo Mao <maobibo@...ngson.cn>
Subject: Re: [PATCH v1] LoongArch: Handle special PC in unwind_next_frame()
On Wed, Nov 26, 2025 at 5:09 PM Tiezhu Yang <yangtiezhu@...ngson.cn> wrote:
>
> Cc: Bibo Mao <maobibo@...ngson.cn>
>
> On 2025/11/26 下午2:28, Tiezhu Yang wrote:
> > On 2025/11/26 下午2:08, Tiezhu Yang wrote:
> >> On 2025/11/25 下午6:52, Jinyang He wrote:
> >>> On 2025-11-25 14:33, Tiezhu Yang wrote:
> >>>
> >>>> When running virtual machine before testing the kernel live patching
> >>>> with
> >>>> "modprobe livepatch-sample", there is a timeout over 15 seconds, the
> >>>> dmesg
> >>>> command shows "unreliable stack" for user tasks in debug mode.
> >>>>
> >>>> The "unreliable stack" is because it can not unwind from
> >>>> kvm_handle_exit()
> >>>> to its previous frame kvm_exc_entry() due to the PC is not a valid
> >>>> kernel
> >>>> address, the root cause is that the code of kvm_exc_entry() was
> >>>> copied to
> >>>> the DMW area in kvm_loongarch_env_init(), so it should check the PC
> >>>> range
> >>>> and then finish unwinding for this special case.
> >>>
> >>> Hi, Tiezhu,
> >>>
> >>> Since kvm_loongarch_env_init copies kvm_exc_entry which similar to how
> >>> trap_init copies the exception handler. We should apply the same
> >>> offset-based adjustment to compute the shadow PC address:
> >>> shadow_pc = kvm_exc_entry + (cur_pc − copied_base_address)
> >>> This allows the ORC unwinder to correctly locate the corresponding
> >>> unwind info.
> >>
> >> Thank you. I did some tests, it can unwind from kvm_handle_exit()
> >> to its previous frame kvm_exc_entry() only when CONFIG_KVM=y,
> >> but it can not unwind from kvm_handle_exit() to its previous frame
> >> kvm_exc_entry() when CONFIG_KVM=m because we can not get the two
> >> kvm address (unsigned long)kvm_exc_entry and
> >> (unsigned long)kvm_loongarch_ops->exc_entry
> >> in arch/loongarch/kernel/unwind_orc.c.
> >
> > Use function pointer can get get the two kvm address
> > (unsigned long)kvm_exc_entry and
> > (unsigned long)kvm_loongarch_ops->exc_entry
> > in arch/loongarch/kernel/unwind_orc.c.
> >
> > I will do more test and send v2 later.
>
> Here are the draft changes:
>
> ----->8-----
> diff --git a/arch/loongarch/kernel/unwind_guess.c
> b/arch/loongarch/kernel/unwind_guess.c
> index 08d7951b2f60..64c50c483cbb 100644
> --- a/arch/loongarch/kernel/unwind_guess.c
> +++ b/arch/loongarch/kernel/unwind_guess.c
> @@ -5,6 +5,11 @@
> #include <asm/unwind.h>
> #include <linux/export.h>
>
> +#if IS_ENABLED(CONFIG_KVM)
> +void (*get_kvm_entry_info)(unsigned long *old, unsigned long *new,
> unsigned long *size);
> +EXPORT_SYMBOL_GPL(get_kvm_entry_info);
> +#endif
Define in unwind.c then you don't need to duplicate it.
> +
> unsigned long unwind_get_return_address(struct unwind_state *state)
> {
> return __unwind_get_return_address(state);
> diff --git a/arch/loongarch/kernel/unwind_orc.c
> b/arch/loongarch/kernel/unwind_orc.c
> index 1d60a593479a..ef8bb6cd7af8 100644
> --- a/arch/loongarch/kernel/unwind_orc.c
> +++ b/arch/loongarch/kernel/unwind_orc.c
> @@ -356,6 +356,11 @@ static bool is_entry_func(unsigned long addr)
> return addr >= (unsigned long)&kernel_entry && addr < (unsigned
> long)&kernel_entry_end;
> }
>
> +#if IS_ENABLED(CONFIG_KVM)
> +void (*get_kvm_entry_info)(unsigned long *old, unsigned long *new,
> unsigned long *size);
> +EXPORT_SYMBOL_GPL(get_kvm_entry_info);
> +#endif
> +
> static inline unsigned long bt_address(unsigned long ra)
> {
> extern unsigned long eentry;
> @@ -386,6 +391,16 @@ static inline unsigned long bt_address(unsigned
> long ra)
> return func + offset;
> }
>
> +#if IS_ENABLED(CONFIG_KVM)
> + unsigned long old, new, size;
Define them in the if block, and you can consider removing all "#if
IS_ENABLED(CONFIG_KVM)".
Huacai
> +
> + if (get_kvm_entry_info) {
> + get_kvm_entry_info(&old, &new, &size);
> + if (ra >= new && ra < new + size)
> + return old + (ra - new);
> + }
> +#endif
> +
> return ra;
> }
>
> diff --git a/arch/loongarch/kernel/unwind_prologue.c
> b/arch/loongarch/kernel/unwind_prologue.c
> index 729e775bd40d..42c210b0a378 100644
> --- a/arch/loongarch/kernel/unwind_prologue.c
> +++ b/arch/loongarch/kernel/unwind_prologue.c
> @@ -28,6 +28,11 @@ extern unsigned long eentry;
> extern unsigned long pcpu_handlers[NR_CPUS];
> #endif
>
> +#if IS_ENABLED(CONFIG_KVM)
> +void (*get_kvm_entry_info)(unsigned long *old, unsigned long *new,
> unsigned long *size);
> +EXPORT_SYMBOL_GPL(get_kvm_entry_info);
> +#endif
> +
> static inline bool scan_handlers(unsigned long entry_offset)
> {
> int idx, offset;
> diff --git a/arch/loongarch/kvm/main.c b/arch/loongarch/kvm/main.c
> index 80ea63d465b8..3606554e43e0 100644
> --- a/arch/loongarch/kvm/main.c
> +++ b/arch/loongarch/kvm/main.c
> @@ -338,6 +338,13 @@ void kvm_arch_disable_virtualization_cpu(void)
> kvm_flush_tlb_all();
> }
>
> +static void kvm_entry_info(unsigned long *old, unsigned long *new,
> unsigned long *size)
> +{
> + *old = (unsigned long)kvm_exc_entry;
> + *new = (unsigned long)kvm_loongarch_ops->exc_entry;
> + *size = kvm_exception_size;
> +}
> +
> static int kvm_loongarch_env_init(void)
> {
> int cpu, order, ret;
> @@ -430,6 +437,7 @@ static void kvm_loongarch_env_exit(void)
> kvm_unregister_perf_callbacks();
> }
>
> +extern void (*get_kvm_entry_info)(unsigned long *old, unsigned long
> *new, unsigned long *size);
> static int kvm_loongarch_init(void)
> {
> int r;
> @@ -442,6 +450,8 @@ static int kvm_loongarch_init(void)
> if (r)
> return r;
>
> + get_kvm_entry_info = kvm_entry_info;
> +
> return kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
> }
>
> diff --git a/arch/loongarch/kvm/switch.S b/arch/loongarch/kvm/switch.S
> index f1768b7a6194..9eaad5dbed91 100644
> --- a/arch/loongarch/kvm/switch.S
> +++ b/arch/loongarch/kvm/switch.S
> @@ -104,7 +104,7 @@
> .text
> .cfi_sections .debug_frame
> SYM_CODE_START(kvm_exc_entry)
> - UNWIND_HINT_UNDEFINED
> + UNWIND_HINT_END_OF_STACK
> csrwr a2, KVM_TEMP_KS
> csrrd a2, KVM_VCPU_KS
> addi.d a2, a2, KVM_VCPU_ARCH
> ----->8-----
>
> I tested with config UNWINDER_ORC, it can unwind from kvm_handle_exit()
> to its previous frame kvm_exc_entry() which is the end of stack, there
> is no "unreliable stack" in debug mode and also no timeout for kernel
> livepatching.
>
> I also tested with config UNWINDER_PROLOGUE and config UNWINDER_GUESS,
> no build errors and the virtual machine can boot normally.
>
> I would like to receive comments for the draft changes first and then
> send the formal v2, the function and variable name may be not proper,
> so any comments are welcome.
>
> Since the merge window is coming soon and I am busy with the other
> higher priority stuff, so maybe I will send v2 after the merge window.
>
> Thanks,
> Tiezhu
>
Powered by blists - more mailing lists