Re: [PATCH next v1 1/2] dump_stack: move cpu lock to printk.c

[John Ogness]

On 2021-06-01, Petr Mladek <pmladek@xxxxxxxx> wrote:
>> +void printk_cpu_lock(unsigned int *cpu_store, unsigned long *flags)
>
> I think about calling this printk_cpu_lock_irqsave() to make it clear
> that it disables interrupts.

Agreed.

> Strictly speaking, it should be enough to disable preemption. If it is
> safe when interrupted by NMI, it must be safe also when interrupted
> by a normal interrupt.
>
> I guess that the interrupts are disabled because it reduces the risk
> of nested (messed) backtraces.

If it was just about synchronizing output triggered by sysreq, then it
probably would be acceptable to leave interrupts active. But when atomic
consoles are involved, we are talking about a crashing machine that is
trying to get log messages out. Any interrupt is a risk that the machine
may not survive long enough to return from that interruption.

John Ogness