Re: v5.19-rc2-rt3: nouveau might sleep splat

[John Ogness <john.ogness@linutronix.de>]

On 2022-06-24, Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote:
>> (gdb) list *vprintk_store+0x3cc
>> 0xffffffff8114283c is in vprintk_store (kernel/printk/printk.c:2280).
>> warning: Source file is more recent than executable.
>> 2275            u32 caller_id;
>> 2276            u16 text_len;
>> 2277            int ret = 0;
>> 2278            u64 ts_nsec;
>> 2279
>> 2280            if (!printk_enter_irqsave(recursion_ptr, irqflags))
>> 2281                    return 0;
>> 2282
>> 2283            /*
>> 2284             * Since the duration of printk() can vary depending on the message
>> (gdb)
>
> Why do we need to keep interrupts disabled for vsnprintf()?

vsnprintf is printing the current CPU local clock and the current CPU
ID. Are you recommending we split printk_enter_irqsave() into various
preempt_disable() and local_irq_save() calls?

As to this the actual problem reported in this thread, I believe this
patch solves it:

https://lore.kernel.org/lkml/20220609121709.12939-1-Jason@zx2c4.com

John Ogness