Re: [PATCH v2 3/3] printk/nmi: Prevent deadlock when accessing the main log buffer in NMI

[Sergey Senozhatsky]

On (06/27/18 16:20), Petr Mladek wrote:
> +/*
> + * Marks a code that might produce many messages in NMI context
> + * and the risk of losing them is more critical than eventual
> + * reordering.
> + *
> + * It has effect only when called in NMI context. Then printk()
> + * will try to store the messages into the main logbuf directly
> + * and use the per-CPU buffers only as a fallback when the lock
> + * is not available.
> + */
> +void printk_nmi_direct_enter(void)
> +{
> +	if (this_cpu_read(printk_context) & PRINTK_NMI_CONTEXT_MASK)
> +		this_cpu_or(printk_context, PRINTK_NMI_DIRECT_CONTEXT_MASK);
> +}

A side note: This nesting also handles recursive printk-s for us.

NMI:
	printk_nmi_enter
	ftrace_dump
	 printk_nmi_direct_enter
	  vprintk_func
	   spin_lock(logbuf_lock)
	    vprintk_store
	     vsprintf
	      WARN_ON
	       vprintk_func
	        vprintk_nmi

>  __printf(1, 0) int vprintk_func(const char *fmt, va_list args)
>  {
> +	/*
> +	 * Try to use the main logbuf even in NMI. But avoid calling console
> +	 * drivers that might have their own locks.
> +	 */
> +	if ((this_cpu_read(printk_context) & PRINTK_NMI_DIRECT_CONTEXT_MASK) &&
> +	    raw_spin_trylock(&logbuf_lock)) {
> +		int len;
> +
> +		len = vprintk_store(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
> +		raw_spin_unlock(&logbuf_lock);
> +		defer_console();
> +		return len;
> +	}

So, maybe, something a bit better than defer_console().

Otherwise,
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx>

	-ss