Re: [PATCH wq/for-3.19 3/3] workqueue: dump workqueues on sysrq-t

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 8 Dec 2014 13:40:35 -0500 Tejun Heo <tj@kernel.org> wrote:

> Hello, Andrew.
> 
> On Mon, Dec 08, 2014 at 10:06:13AM -0800, Andrew Morton wrote:
> > sysrq-t already produces thousands of lines of output.  Maybe create a
> > new keycode for this?
> 
> Believe it or not, we already used up all alphanumerics if we count in
> the arch-specific ones.  Given that the workqueue information would
> primarily be useful in tracking down hangs and we'd want to see the
> dump of tasks in that case anyway, sysrq-t isn't a bad fit for
> appending workqueue dump.  If anybody has a better idea, I'm all ears.

Really.  Upper case?

> ...
> > > +static void pr_cont_pool_info(struct worker_pool *pool)
> > > +{
> > > +	if (pool->cpu >= 0)
> > > +		pr_cont(" cpu=%d", pool->cpu);
> > > +	else if (pool->node != NUMA_NO_NODE)
> > > +		pr_cont(" node=%d", pool->node);
> > > +
> > > +	if (pool->cpu < 0) {
> > > +		static char cpus_buf[PAGE_SIZE];
> > 
> > Ouch.  This could be [NR_CPUS + epsilon]?
> 
> It's bitmap mask printing so each char can show four cpus.  PAGE_SIZE
> should be enough for now but I think we need cpumask_prcont().

I'm not concerned about it being too small ;) Not many people have 16k
CPUs - can it be shrunk?  It's particularly gross when CONFIG_SMP=n!