[PATCH 1/1] arm64: Print slab alloc and free paths for addresses in registers

[Mark Rutland]

On Thu, Nov 27, 2025 at 11:32:27AM +0530, Onkarnath wrote:
> When debugging use-after-free kernel oopses, knowing the allocation and
> freeing paths of an object is crucial. Like arm this patch enhances arm64
> debugging by checking if register addresses belong to a slab and printing
> their corresponding alloc and free paths.
> 
> For example x21 prints alloc and free path:
> 
> pc : crash_init+0x44/0x64 [crash]
> lr : crash_init+0x34/0x64 [crash]
> .....

These dots are hiding *tonnes* of lines.

Please see my response from the last time this was proposed:
  
  https://lore.kernel.org/linux-arm-kernel/ZcDa2RXC6z7XuwAD@FVFF77S0Q05N/

At a high level, I still don't think this is a good idea.

> Register x21 information: slab task_struct start ffff0000c3cc7000 data offset 64 pointer offset 0 size 3904 allocated at copy_process+0x1ac/0x14a4
>     kmem_cache_alloc_node_noprof+0x208/0x4a8
>     copy_process+0x1ac/0x14a4
>     kernel_clone+0x70/0x380
>     __arm64_sys_fork+0x40/0x7c
>     invoke_syscall+0x48/0x104
>     el0_svc_common.constprop.0+0x40/0xe0
>     do_el0_svc_compat+0x1c/0x34
>     el0_svc_compat+0x2c/0x90
>     el0t_32_sync_handler+0x88/0xac
>     el0t_32_sync+0x19c/0x1a0
>  Free path:
>     kmem_cache_free+0x3c0/0x430
>     free_task+0x54/0x80
>     __put_task_struct+0x100/0x15c
>     __put_task_struct_rcu_cb+0x14/0x20
>     rcu_core+0x264/0x680
>     rcu_core_si+0x10/0x1c
>     handle_softirqs+0x100/0x244
>     __do_softirq+0x14/0x20
> 
> Co-developed-by: Sarvesh Kadam <s.kadam at samsung.com>
> Signed-off-by: Sarvesh Kadam <s.kadam at samsung.com>
> Signed-off-by: Onkarnath <onkarnath.1 at samsung.com>
> ---
>  arch/arm64/include/asm/system_misc.h |  1 +
>  arch/arm64/kernel/process.c          | 11 +++++++++++
>  arch/arm64/kernel/traps.c            |  2 +-
>  3 files changed, 13 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/include/asm/system_misc.h b/arch/arm64/include/asm/system_misc.h
> index d316a804eb38..9cb9749d8853 100644
> --- a/arch/arm64/include/asm/system_misc.h
> +++ b/arch/arm64/include/asm/system_misc.h
> @@ -27,6 +27,7 @@ void arm64_notify_die(const char *str, struct pt_regs *regs,
>  
>  struct mm_struct;
>  extern void __show_regs(struct pt_regs *);
> +extern void __show_regs_alloc_free(struct pt_regs *regs);
>  
>  #endif	/* __ASSEMBLER__ */
>  
> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> index fba7ca102a8c..7738ec8e5cd5 100644
> --- a/arch/arm64/kernel/process.c
> +++ b/arch/arm64/kernel/process.c
> @@ -199,6 +199,17 @@ static void print_pstate(struct pt_regs *regs)
>  	}
>  }
>  
> +void __show_regs_alloc_free(struct pt_regs *regs)
> +{
> +	int i;
> +
> +	/* check for x0 - x31 only */
> +	for (i = 0; i < 31; i++) {
> +		pr_alert("Register x%d information:", i);
> +		mem_dump_obj((void *)regs->regs[i]);
> +	}
> +}

The comment should say 'x31' rather than 'x30'.

> +
>  void __show_regs(struct pt_regs *regs)
>  {
>  	int i, top_reg;
> diff --git a/arch/arm64/kernel/traps.c b/arch/arm64/kernel/traps.c
> index 914282016069..3b01379b8880 100644
> --- a/arch/arm64/kernel/traps.c
> +++ b/arch/arm64/kernel/traps.c
> @@ -189,7 +189,7 @@ static int __die(const char *str, long err, struct pt_regs *regs)
>  
>  	print_modules();
>  	show_regs(regs);
> -
> +	__show_regs_alloc_free(regs);
>  	if (user_mode(regs))
>  		return ret;

If the regs are user regs, then the registers do not contain kernel
addresses. We shouldn't interpret the registers in that case.

We use die() and __die() for many exceptions that are entirely unrelated
to use-after-free (e.g. BTI exceptions), so this is going to be noisy
for no benefit.

Mark.