Re: [LUG] Kernel drivers - printk()

[Jon Masters <jonathan@xxxxxxxxxxxxxx>]

John Daragon wrote:

> On Wednesday 11 August 2004 14:48, Andrew Rogers wrote:

> > I have a kernel module that must output a lot of info, a few megabytes.

Can you not buffer this data and make it available in /proc/ ?

> > As I can't write to a file from a kernel module I write the output to
> > /var/log/messages using printk().

Write it to relayfs - http://www.opersys.com/relayfs/ - or to the files 
directly but then you will have to handle everything yourself. The rule 
generally seems to be that this is a bad idea so nobody provides good 
examples of doing it.

> > The problem is that after I have called printk() a few thousand
> > times it stops printting to /var/log/messages and restarts
> > printting some time later. It is verylikely that I am
> > calling printk too often in a short time interval.

Indeed. printk() is not designed for this kind of application although 
the stopping effect observed is also affected by buffering in klogd.

> > How can I ensure that printk() has output to /var/log/messages before
> > calling printk() again?

> I don't think you can.

You cannot do that.

> I've just had a look at printk().  You'll be disappointed at the size of the 
> log buffer(s) :

printk() was not designed for this application though.

> #define LOG_BUF_LEN     (16384)              /* This must be a power of two */

> So - you can either :

> 1) write a new version of printk with sensible buffer handling :-),

It has sensible handling. printk() must never block and must be callable 
in every possible context - try improving upon it.

> (this is usually where someone comes along and offers a brilliantly simple 
> solution that I haven't thought of ...)

relayfs.

Jon.



--
The Mailing List for the Devon & Cornwall LUG
Mail majordomo@xxxxxxxxxxxx with "unsubscribe list" in the
message body to unsubscribe.