g_eli: better handling of absent/disabled CPUs

[mhorne]

Summary

Checking hlt_cpus_mask is a no-op, and the mask will be removed in the
next commit. However, we can use the more recent CPU_ABSENT() macro to
check the status.

[kib]

sys/geom/eli/g_eli.c

1087

Why not use CPU_FOREACH() and break out of the loop if the counter >= threads?

[olce]

sys/geom/eli/g_eli.c

1087

I concur. In any case, the debug message does not seem useful and I'd remove it.

[mhorne]

sys/geom/eli/g_eli.c

1087

I hoped for this, but it is difficult to handle both cases because CPU_FOREACH() embeds a CPU_ABSENT() check.

Let's take an example scenario:

• g_eli_threads is set to some explicit value (!= mp_ncpus), e.g. 6.
• The CPU map contains an absent CPU at index 2.

What will happen is that it will skip thread creation on i == 2, and break on loop i == 5, resulting in only 5 threads created.

So this ends up being the simplest change that does not further obfuscate things.

[kib]

sys/geom/eli/g_eli.c

1087

for (i = 0;;) {
    CPU_FOREACH(k) {
          do_something_allocate_thread_on(k);
          if (++i >= threads)
             goto done;
     }
}
done:;

[mhorne]

sys/geom/eli/g_eli.c

1087

I do not think this is better.

Let me propose a follow-up which disentangles the sc_cpubind and !sc_cpubind cases.