Re: [patch V2 09/29] mm/kasan: Simplify stacktrace handling
From: Andrey Ryabinin
Date: Thu Apr 18 2019 - 06:39:40 EST
On 4/18/19 11:41 AM, Thomas Gleixner wrote:
> Replace the indirection through struct stack_trace by using the storage
> array based interfaces.
>
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Acked-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> Cc: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
> Cc: Alexander Potapenko <glider@xxxxxxxxxx>
> Cc: kasan-dev@xxxxxxxxxxxxxxxx
> Cc: linux-mm@xxxxxxxxx
Acked-by: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
>
> static inline depot_stack_handle_t save_stack(gfp_t flags)
> {
> unsigned long entries[KASAN_STACK_DEPTH];
> - struct stack_trace trace = {
> - .nr_entries = 0,
> - .entries = entries,
> - .max_entries = KASAN_STACK_DEPTH,
> - .skip = 0
> - };
> + unsigned int nr_entries;
>
> - save_stack_trace(&trace);
> - filter_irq_stacks(&trace);
> -
> - return depot_save_stack(&trace, flags);
> + nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
> + nr_entries = filter_irq_stacks(entries, nr_entries);
> + return stack_depot_save(entries, nr_entries, flags);
Suggestion for further improvement:
stack_trace_save() shouldn't unwind beyond irq entry point so we wouldn't need filter_irq_stacks().
Probably all call sites doesn't care about random stack above irq entry point, so it doesn't
make sense to spend resources on unwinding non-irq stack from interrupt first an filtering out it later.
It would improve performance of stack_trace_save() called from interrupt and fix page_owner which feed unfiltered
stack to stack_depot_save(). Random non-irq part kills the benefit of using the stack_deopt_save().