linux-kernel - Re: [PATCH 4/5] irq_work: Handle some irq_work in SOFTIRQ on PREEMPT

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20210930095348.tud6jdcenfkfzugz@linutronix.de>
Date:   Thu, 30 Sep 2021 11:53:48 +0200
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     linux-kernel@...r.kernel.org, Thomas Gleixner <tglx@...utronix.de>,
        Jiri Olsa <jolsa@...nel.org>
Subject: Re: [PATCH 4/5] irq_work: Handle some irq_work in SOFTIRQ on
 PREEMPT_RT

On 2021-09-30 11:07:18 [+0200], Peter Zijlstra wrote:
> 
> IIRC we have existing problems in -RT due to this irq_work softirq muck.

We have existing problems in -RT due irq_work being used without knowing
the consequences.

> I think the problem was something Jolsa found a while ago, where perf
> defers to an irq_work (from NMI context) and that irq_work wants to
> deliver signals, which it can't on -RT, so the whole thing gets punted
> to softirq. With the end-result that if you self-profile RT tasks,
> things come apart or something.

For signals (at least on x86) we this ARCH_RT_DELAYS_SIGNAL_SEND thingy
where the signal is delayed until exit_to_user_mode_loop().

perf_pending_event() is the only non-HARD on RT (on the perf side). I
think that is due to perf_event_wakeup() where we have wake_up_all() and
read_lock_irqsave().

> There might have been others as well, I don't know. But generally I
> think we want *less* softirq, not more.

I agree. The anonymous softirqs concept brings problems of its own.
But what should I do with things like that:
- kernel/trace/ring_buffer.c rb_wake_up_waiters()
  kernel/bpf/ringbuf.c bpf_ringbuf_notify()
   wake_up_all()

- drivers/acpi/apei/ghes.c ghes_proc_in_irq()
  spinlock_t

- drivers/dma-buf/dma-fence-array.c irq_dma_fence_array_work()
  spinlock_t, callbacks, potential kfree().

- drivers/gpu/drm/i915/gt/intel_breadcrumbs.c signal_irq_work()
  spinlock_t, rpm_put() -> wake_up_var(), callbacks, kref_put() like
  constructs which may free memory.

I didn't look at _all_ of them but just a few. And the only one I looked
at and didn't add to the list was
    drivers/edac/igen6_edac.c ecclog_irq_work_cb()

which simply reads PCI registers (which acquires raw_spinlock_t only
(however only on x86 are those raw_spinlock_t now that looked around))
and does schedule_work(). All harmless.

I *think* the irq_work in printk is going to leave once John is done
with it. But there are way more of these things in the kernel now
compared to when I first pushed them into softirq because they were
causing trouble.

Sebastian