Re: [RFC PATCH v1 10/25] printk: redirect emit/store to new ringbuffer

From: Petr Mladek <pmladek@suse.com>
To: John Ogness <john.ogness@linutronix.de>
Cc: linux-kernel@vger.kernel.org,
	Peter Zijlstra <peterz@infradead.org>,
	Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Daniel Wang <wonderfly@google.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Alan Cox <gnomes@lxorguk.ukuu.org.uk>,
	Jiri Slaby <jslaby@suse.com>, Peter Feiner <pfeiner@google.com>,
	linux-serial@vger.kernel.org,
	Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Subject: Re: [RFC PATCH v1 10/25] printk: redirect emit/store to new ringbuffer
Date: Tue, 26 Feb 2019 10:45:37 +0100	[thread overview]
Message-ID: <20190226094537.46xfhv5rr5m4c33s@pathway.suse.cz> (raw)
In-Reply-To: <875zt7bz1t.fsf@linutronix.de>

On Mon 2019-02-25 17:41:50, John Ogness wrote:
> On 2019-02-25, Petr Mladek <pmladek@suse.com> wrote:
> >> >> vprintk_emit and vprintk_store are the main functions that all printk
> >> >> variants eventually go through. Change these to store the message in
> >> >> the new printk ring buffer that the printk kthread is reading.
> >> >
> >> > Please, are there any candidates or plans to reuse the new ring
> >> > buffer implementation?
> >> 
> >> As you pointed out below, this patch already uses the ring buffer
> >> implementation for a totally different purpose: NMI safe dynamic memory
> >> allocation.
> >
> > I have found an alternative solution. We could calculate the length
> > of the formatted string without any buffer:
> >
> > 	va_list args_copy;
> >
> > 	va_copy(args_copy, args);
> > 	len = vscprintf(NULL, fmt, args_copy);
> > 	va_end(args_copy);
> >
> > This vsprintf() mode was implemented for exactly this purpose.
> 
> For vprintk_emit() that would work. As you will see in later (patch 23),
> the sprint_rb ringbuffer is used for dynamic memory allocation for
> kmsg_dump functions as well.

It looks dangerous to share a limited buffer between core kernel
functionality and user-space triggered one. I mean that an unlimited
number of devkmsg operations must not cause loosing printk() messages.

> The current printk implementation allows readers to read directly from
> the ringbuffer. The proposed ringbuffer requires the reader (printk) to
> have its own buffers.
> 
> We may be able to find an alternate solution here as well if that is
> desired.

I hope that we will be able to find one. The previous implementation
needed some buffers as well. We should be able to use the same
approach.

I guess that one problem is that the new ringbuffer API is not
able to copy just the text directly into the user-provided buffer.
It might get solved by extending the API.

Anyway, I still have to look at the remaining patches.

Best Regards,
Petr