Re: [PATCH/RFC v5 4/5]: core: Add dump device to call on oopses and panics

[Artem Bityutskiy <dedekind1@gmail.com>]

On Mon, 2009-11-30 at 10:35 +0100, Jörn Engel wrote:
> On Mon, 30 November 2009 10:51:58 +0200, Artem Bityutskiy wrote:
> > 
> > How about this patch on top of the existing one (untested):
> > 
> > +	/*
> > +	 * Have we ever rotated around the circular buffer? If we never did,
> > +	 * we have to have zeroes at the end.
> > +	 */
> > +	if (log_buf[end]) {
> > +		s1 = log_buf + end;
> > +		l1 = log_buf_len - end;
> > +	} else {
> > +		s1 = "";
> > +		l1 = 0;
> 
> So now you are assuming that a) the buffer is initially zeroed and b)
> noone ever writes NUL to it.  Is that correct?

a) seems to be true because the buffer is either a static array or a
bootmem alloc, which seems to memzero the buffers it returns, at least
AFAICS. But I did not test this.

vs b). well, the printk ring buffer should contain ASCII, so I assumed
binary zeroes should not be possible there.

> I'm not sure whether those assumptions are valid.  If they are, then
> this will obviously work.  Otherwise we can just always assume the
> wrapped case.

Of course someone who has more knowlege about the printk buffer should
comment on this.

The other alternative I was thinking about was to introduce a boolean
flag, and set it to one as soon as 'lon_end' becomes larger than
'log_buf_len'.

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)