Re: [PATCH v2 bpf-next 1/4] perf: export get/put_chain_entry()

From: Andrii Nakryiko
Date: Fri Jun 26 2020 - 16:06:15 EST


On Fri, Jun 26, 2020 at 5:10 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Thu, Jun 25, 2020 at 05:13:29PM -0700, Song Liu wrote:
> > This would be used by bpf stack mapo.
>
> Would it make sense to sanitize the API a little before exposing it?
>
> diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
> index 334d48b16c36..016894b0d2c2 100644
> --- a/kernel/events/callchain.c
> +++ b/kernel/events/callchain.c
> @@ -159,8 +159,10 @@ static struct perf_callchain_entry *get_callchain_entry(int *rctx)
> return NULL;
>
> entries = rcu_dereference(callchain_cpus_entries);
> - if (!entries)
> + if (!entries) {
> + put_recursion_context(this_cpu_ptr(callchain_recursion), rctx);
> return NULL;
> + }
>
> cpu = smp_processor_id();
>
> @@ -183,12 +185,9 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
> int rctx;
>
> entry = get_callchain_entry(&rctx);
> - if (rctx == -1)
> + if (!entry || rctx == -1)
> return NULL;
>

isn't rctx == -1 check here not necessary anymore? Seems like
get_callchain_entry() will always return NULL if rctx == -1?

> - if (!entry)
> - goto exit_put;
> -
> ctx.entry = entry;
> ctx.max_stack = max_stack;
> ctx.nr = entry->nr = init_nr;