/proc/cpuinfo versus /proc/softirqs: how to properly count CPUs?

[Robert P. J. Day]

On Mon, 22 Oct 2012, Vijay Chauhan wrote:

> On Mon, Oct 22, 2012 at 7:30 PM, Robert P. J. Day <rpjday at crashcourse.ca> wrote:
> >
> >   poking around under /proc on my quad-core asus laptop and just
> > noticed that while /proc/cpuinfo (properly) lists my 8 processors,
> > /proc/softirqs instead lists 16, with the last 8 having values of all
> > zeroes (not surprisingly).  with the middle columns snipped for
> > brevity, my /proc/softirqs shows:
> >
> >                     CPU0       CPU1 ...     CPU13      CPU14      CPU15
> >           HI:          0          0 ...         0          0          0
> >        TIMER:    1018887      45313 ...         0          0          0
> >       NET_TX:       1182       1813 ...         0          0          0
> >       NET_RX:        991        719 ...         0          0          0
> >        BLOCK:      44559         13 ...         0          0          0
> > BLOCK_IOPOLL:          0          0 ...         0          0          0
> >      TASKLET:      41582        419 ...         0          0          0
> >        SCHED:      76438      28658 ...         0          0          0
> >      HRTIMER:        738        709 ...         0          0          0
> >          RCU:     136797      82392 ...         0          0          0
> >
> > with all of the zero values from CPU8-CPU15.  so the obvious question
> > is -- why?  why the difference in the way those two proc files count
> > the "number" of CPUs on my system?
> >
> >   for /proc/cpuinfo, the logic is in arch/x86/kernel/cpu/proc.c, and
> > the way the seq_file is implemented:
> >
> >   static void *c_start(struct seq_file *m, loff_t *pos)
> >   {
> >         *pos = cpumask_next(*pos - 1, cpu_online_mask);
> >         if ((*pos) < nr_cpu_ids)
> >                 return &cpu_data(*pos);
> >         return NULL;
> >   }
> >
> > so that loop clearly iterates through the "online" CPUs, which would
> > appear to be the correct loop criteria.
> >
> >   the code for softirqs, however, is in fs/proc/softirqs.c, and is
> > much simpler:
> >
> >   static int show_softirqs(struct seq_file *p, void *v)
> >   {
> >         int i, j;
> >
> >         seq_puts(p, "                    ");
> >         for_each_possible_cpu(i)
> >                 seq_printf(p, "CPU%-8d", i);
> >         seq_putc(p, '\n');
> >
> >         for (i = 0; i < NR_SOFTIRQS; i++) {
> >                 seq_printf(p, "%12s:", softirq_to_name[i]);
> >                 for_each_possible_cpu(j)
> >                         seq_printf(p, " %10u", kstat_softirqs_cpu(i, j));
> >                 seq_putc(p, '\n');
> >         }
> >         return 0;
> >   }
> >
> > note that that code uses the macro "for_each_possible_cpu()" rather
> > than examining only *online* CPUs.  in dmesg, i see the lines:
> >
> > [    0.000000] smpboot: Allowing 16 CPUs, 8 hotplug CPUs
> > [    0.000000] setup_percpu: NR_CPUS:256 nr_cpumask_bits:256 nr_cpu_ids:16 nr_node_ids:1
> >
> > and the code for all this counting is in arch/x86/kernel/smpboot.c,
> > which you can check out for yourself, but this brings me back to the
> > basic question -- why is the code for softirqs iterating through all
> > *possible* CPUs (in my case, apparently, 16), when i have only 8
> > *online* CPUs?
>
> Sorry for jumping in but i have basic question here.
> What is the difference between *online* CPUs and *possible* CPUs?

  that is precisely the confusion i'm having here, so i'm pawing thru
the code as we speak to figure that out.

rday

-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================