Re: [PATCH v6 13/18] tracing/remotes: Add dump_on_panic tracefs file

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
- [High] The panic handler unconditionally attempts to acquire a per-CPU reader spinlock that may be held by a CPU halted by smp_send_stop(), leading to a permanent deadlock during kernel panic.
- [High] Missing touch_nmi_watchdog() in the panic dump loop can cause a secondary panic or hard reset.
- [Medium] Mixing goto-based error handling with __free() scope-based cleanup in trace_remote_iter().
- [Low] Missing READ_ONCE() when reading remote->panic_on in dump_on_panic_show().
--

--- Patch [13]: [PATCH v6 13/18] tracing/remotes: Add dump_on_panic tracefs file ---
commit 33a5774bde5e05c9ff209cc5026a9b962c22df16
Author: Vincent Donnefort <[email protected]>

tracing/remotes: Add dump_on_panic tracefs file

This commit adds a dump_on_panic tracefs file. When enabled, it dumps the
content of the trace remote buffer to the console if the system panics.

> diff --git a/kernel/trace/trace_remote.c b/kernel/trace/trace_remote.c
> index 9e78267d71f8b..5536e0df43795 100644
> --- a/kernel/trace/trace_remote.c
> +++ b/kernel/trace/trace_remote.c

[ ... ]

> @@ -434,58 +455,68 @@ static void trace_remote_dec_poll(struct trace_remote *remote)
>  static struct trace_remote_iterator
>  *trace_remote_iter(struct trace_remote *remote, int cpu, enum tri_type type)
>  {
> -	struct trace_remote_iterator *iter = NULL;
> +	struct trace_remote_iterator *iter __free(kfree) = kzalloc_obj(*iter);

[Severity: Medium]
This isn't a bug, but this function mixes scope-based cleanup (__free) with
a goto-based error handling pattern. According to the cleanup subsystem
guidelines, mixing goto-based cleanup labels and __free() declarations in
the same function is forbidden because it creates confusing ownership
semantics.

Should this function either use goto-based cleanup exclusively, or be
fully converted to scope-based cleanup?

[ ... ]

> @@ -989,6 +1024,117 @@ static int dmesg_show(struct seq_file *s, void *unused)
>  }
>  DEFINE_TRACE_REMOTE_ATTRIBUTE(dmesg);
>  
> +static int trace_remote_panic_handler(struct notifier_block *self, unsigned long ev, void *v)
> +{
> +	struct trace_remote *remote = container_of(self, struct trace_remote, panic_notifier);
> +	struct trace_remote_iterator *iter = smp_load_acquire(&remote->panic_iter);
> +	int cpu;
> +
> +	if (!iter) {
> +		pr_warn("Unexpected error: no panic iterator for the trace remote\n");
> +		return NOTIFY_DONE;
> +	}
> +
> +	for_each_possible_cpu(cpu) {
> +		if (iter->rb_iters[cpu]) {
> +			/* No RING_BUFFER_ALL_CPUS to avoid taking cpu_read_lock() */
> +			ring_buffer_read_remote_meta_page(remote->trace_buffer, cpu);
> +			ring_buffer_iter_reset(iter->rb_iters[cpu]);
> +		}
> +	}

[Severity: High]
Can this loop deadlock indefinitely during a panic?

If another CPU is actively reading the remote trace buffer and holds its
reader_lock, the panic() function will stop that CPU via smp_send_stop().

When the panicking CPU executes this loop,
ring_buffer_read_remote_meta_page() calls rb_reader_lock(). If in_nmi() is
false (which happens on a normal oops or SysRq-C), rb_reader_lock() will
unconditionally attempt to acquire the lock held by the stopped CPU rather
than using a trylock.

Will this cause the panic recovery process to hang forever?

> +
> +	while (trace_remote_iter_read_event(iter)) {
> +		trace_seq_init(&iter->seq);
> +
> +		trace_remote_iter_print_event(iter);
> +		trace_seq_putc(&iter->seq, 0);
> +		pr_emerg("%s", seq_buf_str(&iter->seq.seq));
> +
> +		trace_remote_iter_move(iter);
> +	}

[Severity: High]
Since serial consoles can be slow and remote trace buffers can contain
many events, could the time spent in this printing loop trigger the NMI
watchdog and cause a hard reset before the panic dump completes?

Does this loop need a call to touch_nmi_watchdog()?

> +
> +	return NOTIFY_DONE;
> +}

[ ... ]

> +static int dump_on_panic_show(struct seq_file *s, void *unused)
> +{
> +	struct trace_remote *remote = s->private;
> +
> +	seq_printf(s, "%d\n", remote->panic_on);

[Severity: Low]
Does this unprotected read of remote->panic_on need a READ_ONCE()?

Since dump_on_panic_write() modifies this flag concurrently under
remote->lock, this is formally a data race under the Linux kernel memory
model and might be flagged by KCSAN.

> +
> +	return 0;
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=13