Re: [PATCH] sched/x86: construct all sibling maps if smt

[Ingo Molnar <mingo@kernel.org>]

* Peter Zijlstra <peterz@infradead.org> wrote:

> On Wed, May 29, 2013 at 12:26:01PM +0200, Andrew Jones wrote:
> > On Mon, May 27, 2013 at 07:09:00PM +0200, Andrew Jones wrote:
> > > Commit 316ad248307fb ("sched/x86: Rewrite set_cpu_sibling_map()") broke
> > > the construction of sibling maps, which also broke the booted_cores
> > > accounting.
> > > 
> > > Before the rewrite, if smt was present, then each map was updated for
> > > each smt sibling. After the rewrite only cpu_sibling_mask gets updated,
> > > as the llc and core maps depend on 'has_mc = x86_max_cores > 1' instead.
> > > This leads to problems with topologies like the following
> > > 
> > > (qemu -smp sockets=2,cores=1,threads=2)
> > > 
> > > processor	: 0
> > > physical id	: 0
> > > siblings	: 1    <= should be 2
> > > core id		: 0
> > > cpu cores	: 1
> > > 
> > > processor	: 1
> > > physical id	: 0
> > > siblings	: 1    <= should be 2
> > > core id		: 0
> > > cpu cores	: 0    <= should be 1
> > > 
> > > processor	: 2
> > > physical id	: 1
> > > siblings	: 1    <= should be 2
> > > core id		: 0
> > > cpu cores	: 1
> > > 
> > > processor	: 3
> > > physical id	: 1
> > > siblings	: 1    <= should be 2
> > > core id		: 0
> > > cpu cores	: 0    <= should be 1
> > > 
> > > This patch restores the former construction by defining has_mc as
> > > (has_smt || x86_max_cores > 1). This should be fine as there were no
> > > (has_smt && !has_mc) conditions in the context.
> > > 
> > > Signed-off-by: Andrew Jones <drjones@redhat.com>
> > > ---
> > >  arch/x86/kernel/smpboot.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
> > > index 9c73b51817e47..886a3234eaff3 100644
> > > --- a/arch/x86/kernel/smpboot.c
> > > +++ b/arch/x86/kernel/smpboot.c
> > > @@ -372,15 +372,15 @@ static bool __cpuinit match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
> > >  
> > >  void __cpuinit set_cpu_sibling_map(int cpu)
> > >  {
> > > -	bool has_mc = boot_cpu_data.x86_max_cores > 1;
> > >  	bool has_smt = smp_num_siblings > 1;
> > > +	bool has_mc = has_smt || boot_cpu_data.x86_max_cores > 1;
> > >  	struct cpuinfo_x86 *c = &cpu_data(cpu);
> > >  	struct cpuinfo_x86 *o;
> > >  	int i;
> > >  
> > >  	cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
> > >  
> > > -	if (!has_smt && !has_mc) {
> > > +	if (!has_mc) {
> > >  		cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
> > >  		cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
> > >  		cpumask_set_cpu(cpu, cpu_core_mask(cpu));
> > > -- 
> > > 1.8.1.4
> > >
> > 
> > Any acks? This patch fixes a regression. Also, in case anybody is
> > wondering, this is not the same regression as was already fixed with
> > 
> > ceb1cbac8eda6 sched/x86: Calculate booted cores after construction of sibling_mask
> > 
> > (Hmm, I probably should have renamed has_mc to has_mp, as the redefinition
> > expands its scope. I'm not sure if that deserves a v2 though.)
> 
> Right, took me a while to bend my brain around that code again -- I
> obviously don't have the best track record since this is the second bug
> in it since I rewrote the thing (with the intent of making it 'easier'
> to read ha!).
> 
> Yes, I think your patch is correct, and your suggestion of doing
> s/has_mc/has_mp/ seems a sensible one too.
> 
> Thanks!

Andrew, it would be nice to have a -v2 with that rename and with Peter's 
Acked-by included.

Thanks,

	Ingo