Re: [PATCH - resend] addition of loglevel for printf in Xen HV

[Steven Rostedt <srostedt@redhat.com>]

Keir Fraser wrote:

> 
> I think yet-another-little-tool in dom0 is the right way to go. We can add a
> sysctl to get/set the parameters. It can pretty-print the current parameters
> and also provide a way of changing them. I don't think there's any benefit
> to integration with xm -- it's not like this needs to be integrated with
> xend or anything like that.

That should be easy enough. What's the best way to communicate between 
dom0 and the HV?  Add a hypercall for this?

> 
> Yes, we should also have a way to set them as a boot parameter. I suggest we
> should only enact those explicit parameters when dom0 starts to boot though
> (no need to stop Xen being noisy in initial boot). We could have an extra
> parameter (e.g., 'quiet') to override that if someone really cares.

Yeah, but this can be handled after the initial work is done.

> 
> It'd be nice to have debug-key support for changing it too, but actually
> that's probably not so important if we have the dom0 tool.

My original patch included a debug key for log level.  I'm sure it won't 
be too hard to add something to change the thresholds too.  (for those 
that only communicate to the machine via a serial console).

> 
> Everything you said I agree with, so please do go hack it up. :-)

Thanks!  Will start hacking!

> 
> Another thought... By pushing the rate-limiting into printk() itself, we
> lose the ability to print all-or-nothing for related blocks of printk (e.g.,
> register dumps). We might want a 'same as previous line' specifier, which
> would print/not-print same as previous output line?
> 

Hmm, I wonder if we should have a XENLOG_CRASH level which is above 
XENLOG_ERR.  This will be when the system is actually taking a dump, and 
we want to show everything.  Or just simply force the thresholds to max 
just before showing the crash print outputs.  Or add a variable to know 
we are in the process of crashing, and have printk ignore the print rate 
limit if that's the case, similar to the oops_in_progress of Linux.

-- Steve