Re: [PATCH bpf-next v6 0/6] bpf: Add support for sleepable tracepoint programs

[Mykyta Yatsenko <mykyta.yatsenko5@gmail.com>]

Kumar Kartikeya Dwivedi <memxor@gmail.com> writes:

> On Tue, 24 Mar 2026 at 20:03, Mykyta Yatsenko
> <mykyta.yatsenko5@gmail.com> wrote:
>>
>> This series adds support for sleepable BPF programs attached to raw
>> tracepoints (tp_btf), classic raw tracepoints (raw_tp), and classic
>> tracepoints (tp). The motivation is to allow BPF programs on syscall
>> tracepoints to use sleepable helpers such as bpf_copy_from_user(),
>> enabling reliable user memory reads that can page-fault.
>>
>> This series removes restriction for faultable tracepoints:
>>
>> Patch 1 modifies __bpf_trace_run() to support sleepable programs
>> following the uprobe_prog_run() pattern: use explicit
>> rcu_read_lock_trace() for sleepable programs and rcu_read_lock() for
>> non-sleepable programs. Also removes preempt_disable from the faultable
>> tracepoint BPF callback wrapper, since migration protection and RCU
>> locking are now managed per-program inside __bpf_trace_run().
>>
>> Patch 2 renames bpf_prog_run_array_uprobe() to
>> bpf_prog_run_array_sleepable() to support new usecase.
>>
>> Patch 3 adds sleepable support for classic tracepoints
>> (BPF_PROG_TYPE_TRACEPOINT) by introducing trace_call_bpf_faultable()
>> and restructuring perf_syscall_enter/exit() to run BPF programs in
>> faultable context before the preempt-disabled per-cpu buffer
>> allocation. trace_call_bpf_faultable() uses rcu_tasks_trace for
>> lifetime protection, following the uprobe pattern. This adds
>> rcu_tasks_trace overhead for all classic tracepoint BPF programs on
>> syscall tracepoints, not just sleepable ones.
>>
>> Patch 4 allows BPF_TRACE_RAW_TP, BPF_PROG_TYPE_RAW_TRACEPOINT, and
>> BPF_PROG_TYPE_TRACEPOINT programs to be loaded as sleepable, with
>> load-time and attach-time checks to reject sleepable programs on
>> non-faultable tracepoints.
>>
>> Patch 5 adds libbpf SEC_DEF handlers: tp_btf.s, raw_tp.s,
>> raw_tracepoint.s, tp.s, and tracepoint.s.
>>
>> Patch 6 adds selftests covering tp_btf.s, raw_tp.s, and tp.s positive
>> cases using bpf_copy_from_user() on sys_enter nanosleep, plus negative
>> tests for non-faultable tracepoints.
>>
>> Signed-off-by: Mykyta Yatsenko <yatsenko@meta.com>
>> ---
>> Changes in v6:
>> - Remove recursion check from trace_call_bpf_faultable(), sleepable
>> tracepoints are called from syscall enter/exit, no recursion is
>> possible.(Kumar)
>
> Is this true? Can't the same program reenter due to the same syscall
> from a different task which preempted the task running the previous
> program?
> Just asking because it wasn't clear to me, since we discussed
> bpf_prog_get_recursion_context() instead.
>
Good question, this is my understanding, though I'm not 100% confident:
There are 2 mechanisms we discussed:
 * bpf_prog_get_recursion_context() - per program recursion guard, I
 understand this is used to prevent self-recursion, where BPF program
 infinitely generates calls to itself.
 * bpf_prog_active - per CPU counter used to prevent deadlock between
 instrumentation BPF prog (tracepoint, kprobe) and map operation
 generating an event triggering those BPF progs when holding map bucket
 spinlock.

I suggest that none of these applies for the usecase, because we open
sleepable only for syscall enter/exit:
 * Deadlock on the map bucket spinlock is not possible (bpf_prog_active)
 * BPF program can't generate another syscall enter/exit - no infinite
 recursion (bpf_prog_get_recursion_context())

As for the situation that you describe: same program reenter due to the
same syscall from a different task which preempted the task running the
previous - this is normally supported, behaviour - similar to what
we have when NMI runs on the same CPU and calls common BPF subprog, or
two tasks schedule sleepable bpf_task_works callbacks, that can
interleave on the same CPU.

Let me know if my understanding is not correct. Thanks for checking this
series!
>> [...]