Re: [PATCH RESEND 1/3] printk: convert byte-buffer to variable-length record buffer

[Anton Vorontsov <anton.vorontsov@linaro.org>]

Hello Kay,

On Thu, May 03, 2012 at 02:29:13AM +0200, Kay Sievers wrote:
> From: Kay Sievers <kay@vrfy.org>
> Subject: printk: convert byte-buffer to variable-length record buffer

Bad news, this patch breaks all kmsg_dump users.

Per current printk.c:

 * The 'struct log' buffer header must never be directly exported to
 * userspace, it is a kernel-private implementation detail that might
 * need to be changed in the future, when the requirements change.

But kmsg_dump users get the raw log, not text. Some users may even
use strlen() on the buffers, e.g. ramoops module was using it just
recently, before persistent_ram rework.

So, starting from this patch every kmsg_dump user is getting the
binary stuff, which is barely usable, and breaks a longstanding
user-visible ABIs (e.g. mtdoops, oops dumping into nvram on PowerPC,
and ramoops of course).

I'm not sure how we can fix this without allocating a huge buffer
(and thus wasting memory), or reworking the whole kmsg_dump concept
and its users.

Ideas?

p.s. And, btw, during HW bring-up, when there's even no serial
console usable, I somewhat liked to directly print log_buf memory
from a JTAG debugger, and thus see the kernel logs. Now that'll
look ugly. Heh. :-)

-- 
Anton Vorontsov
Email: cbouatmailru@gmail.com