lkml.org 
[lkml]   [2012]   [Feb]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: MCE, AMD: Hide smp-only code around CONFIG_SMP
From
On 11 February 2012 10:07, Ingo Molnar <mingo@elte.hu> wrote:
>
> * Kevin Winchester <kjwinchester@gmail.com> wrote:
>
>> On 9 February 2012 04:06, Ingo Molnar <mingo@elte.hu> wrote:
>> >
>> > * Borislav Petkov <bp@alien8.de> wrote:
>> >
>> >> +++ b/arch/x86/include/asm/smp.h
>> >> @@ -33,8 +33,15 @@ static inline bool cpu_has_ht_siblings(void)
>> >>
>> >>  DECLARE_PER_CPU(cpumask_var_t, cpu_sibling_map);
>> >>  DECLARE_PER_CPU(cpumask_var_t, cpu_core_map);
>> >> -/* cpus sharing the last level cache: */
>> >> +
>> >> +#ifdef CONFIG_SMP
>> >> +/* CPUs sharing the last level cache: */
>> >>  DECLARE_PER_CPU(cpumask_var_t, cpu_llc_shared_map);
>> >> +#else
>> >> +static DECLARE_BITMAP(cpu_llc_shared_bits, NR_CPUS) __read_mostly = { [0] = 1UL };
>> >> +static struct cpumask *const cpu_llc_shared_map = to_cpumask(cpu_llc_shared_bits);
>> >> +#endif
>> >
>> > Why not just expose it like on SMP?
>> >
>> > We want to *reduce* the specialness of UP, not increase it - one
>> > more word of .data and .text does not matter much - UP is
>> > becoming more and more an oddball, rarely tested config. By the
>> > time these changes hit any real boxes it will be even more
>> > oddball.
>> >
>>
>> It seems that cpu_llc_shared_map is actually defined in
>> arch/x86/kernel/smpboot.c, which is not compiled/linked for UP
>> builds.
>>  Is there an equivalent file for UP that could be used
>> instead, or could the:
>>
>> DEFINE_PER_CPU(cpumask_var_t, cpu_llc_shared_map);
>>
>> be moved to some other file?
>
> Yes, it should be moved into struct cpuinfo_x86, and thus we'd
> remove cpu_llc_shared_map altogether, it would be named
> cpu->llc_shared_map or so - taking up a single bit (or maybe
> zero bits) on UP.

I just sent out a patch trying this out. I built and booted it on SMP
and UP and didn't see any issues, but I don't have more than my one PC
on which to test, so I hope I didn't miss anything. I did notice a
few other per cpu variables (e.g. cpu_llc_id) that perhaps could
perhaps use the same treatment. Where would we want to draw the line
here?
>
>> Generally, it sounds like you might approve of an eventual
>> merging of the boot paths for SMP and UP.  Is that true?  I
>> wonder how much work that would be.  That would really reduce
>> the specialness of UP.
>
> I generally approve just about any patch that works and reduces
> complexity! :-) The boot path is rather ambitious, but if you
> want to try, feel free ...
>

Yes, having walked through the boot path a little, it does seem rather
ambitious. The UP/SMP division seems to be coded right into
init/main.c, rather than being at the architecture level, so it would
be hard to make changes without disrupting other arches.

--
Kevin Winchester
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2012-02-12 01:33    [W:0.307 / U:0.252 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site