Re: [PATCH 01/14] kmemleak: Add the base support

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

On Tue, 30 Dec 2008 08:52:02 +0100 Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > On Tue, 30 Dec 2008 08:38:07 +0100 Ingo Molnar <mingo@elte.hu> wrote:
> > 
> > > * Andrew Morton <akpm@linux-foundation.org> wrote:
> > > 
> > > > > +/*
> > > > > + * Stop the automatic memory scanning thread. This function must be called
> > > > > + * with the kmemleak_mutex held.
> > > > > + */
> > > > > +void stop_scan_thread(void)
> > > > > +{
> > > > > +	if (scan_thread) {
> > > > > +		kthread_stop(scan_thread);
> > > > > +		scan_thread = NULL;
> > > > > +	}
> > > > > +}
> > > > 
> > > > so... why do we need a kernel thread?
> > > > 
> > > > We could have (for the sake of argument) a sys_kmemleak_scan() which 
> > > > does a single scan then returns.  Or something like that.  That way, 
> > > > userspace directly gets to set the scanning frequency, thread priority, 
> > > > etc.
> > > 
> > > thread priority of a kernel thread can be set anyway. Kernel threads tend 
> > > to be better for such simple things because we can control all aspects, 
> > > start them automatically so that test setups catch it (without needing any 
> > > userspace component), etc.
> > > 
> > 
> > yeah yeah, userspace is too hard for kernel programmers, so we put our 
> > applications, English-only pretty-printers etc into the kernel.  It's a 
> > broken record.
> 
> above a certain threshold i think we need to start thinking about merging 
> klibc, and moving some key system applications into the kernel source 
> proper (those which closely depend on the kernel version anyway and need 
> to be updated together).
> 

Sure, but I don't think we'd need klibc for this.  We already ship
quite a lot of ad-hoc userspace in Documentation/, and Sam's recently
(and hopefully temporarily) lost patch which moves all that stuff into
./tests/ (should have been ./userspace/tests/) seemed to work OK.