Re: [PATCH v6 02/17] arm64: stacktrace: Factor out on_accessible_stack_common()

[Marc Zyngier]

On Tue, 26 Jul 2022 08:37:35 +0100,
Kalesh Singh <kaleshsingh@xxxxxxxxxx> wrote:
> 
> Move common on_accessible_stack checks to stacktrace/common.h. This is
> used in the implementation of the nVHE hypervisor unwinder later in
> this series.
> 
> Signed-off-by: Kalesh Singh <kaleshsingh@xxxxxxxxxx>
> Reviewed-by: Fuad Tabba <tabba@xxxxxxxxxx>
> Reviewed-by: Mark Brown <broonie@xxxxxxxxxx>
> Tested-by: Fuad Tabba <tabba@xxxxxxxxxx>
> ---
> 
> Changes in v6:
>   - Add Fuad's Tested-by tag
> 
> Changes in v5:
>   - Add Reviewed-by tags from Mark Brown and Fuad
>   - Remove random whitespace change, per Mark Brown
> 
>  arch/arm64/include/asm/stacktrace.h        |  6 ++----
>  arch/arm64/include/asm/stacktrace/common.h | 18 ++++++++++++++++++
>  2 files changed, 20 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/stacktrace.h b/arch/arm64/include/asm/stacktrace.h
> index 79f455b37c84..43f4b4a6d383 100644
> --- a/arch/arm64/include/asm/stacktrace.h
> +++ b/arch/arm64/include/asm/stacktrace.h
> @@ -65,8 +65,8 @@ static inline bool on_accessible_stack(const struct task_struct *tsk,
>  				       unsigned long sp, unsigned long size,
>  				       struct stack_info *info)
>  {
> -	if (info)
> -		info->type = STACK_TYPE_UNKNOWN;
> +	if (on_accessible_stack_common(tsk, sp, size, info))
> +		return true;
>  
>  	if (on_task_stack(tsk, sp, size, info))
>  		return true;
> @@ -74,8 +74,6 @@ static inline bool on_accessible_stack(const struct task_struct *tsk,
>  		return false;
>  	if (on_irq_stack(sp, size, info))
>  		return true;
> -	if (on_overflow_stack(sp, size, info))
> -		return true;
>  	if (on_sdei_stack(sp, size, info))
>  		return true;
>  
> diff --git a/arch/arm64/include/asm/stacktrace/common.h b/arch/arm64/include/asm/stacktrace/common.h
> index 64ae4f6b06fe..f58b786460d3 100644
> --- a/arch/arm64/include/asm/stacktrace/common.h
> +++ b/arch/arm64/include/asm/stacktrace/common.h
> @@ -62,6 +62,9 @@ struct unwind_state {
>  	struct task_struct *task;
>  };
>  
> +static inline bool on_overflow_stack(unsigned long sp, unsigned long size,
> +				     struct stack_info *info);
> +
>  static inline bool on_stack(unsigned long sp, unsigned long size,
>  			    unsigned long low, unsigned long high,
>  			    enum stack_type type, struct stack_info *info)
> @@ -80,6 +83,21 @@ static inline bool on_stack(unsigned long sp, unsigned long size,
>  	return true;
>  }
>  
> +static inline bool on_accessible_stack_common(const struct task_struct *tsk,
> +					      unsigned long sp,
> +					      unsigned long size,
> +					      struct stack_info *info)
> +{
> +	if (info)
> +		info->type = STACK_TYPE_UNKNOWN;
> +
> +	/*
> +	 * Both the kernel and nvhe hypervisor make use of
> +	 * an overflow_stack
> +	 */
> +	return on_overflow_stack(sp, size, info);
> +}

on_accessible_stack has the following comment:

/*
 * We can only safely access per-cpu stacks from current in a non-preemptible
 * context.
 */

With this change, I don't think we satisfy this requirement anymore,
as we're checking the overflow stack *before* the preemptible check,
which is a big change in behaviour.

The hypervisor doesn't have this requirement: the unwinding is either
done out of context (nVHE, where EL1 unwinds EL2) or in a
non-preemptible section (pKVM, where the whole thing is
non-preemptible). But the kernel is usually preemptible, so this patch
needs fixing.

I'll see if I can address it locally (I'm currently moving things
around, stay tuned).

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.