From: Gautham R Shenoy <ego@linux.vnet.ibm.com>
To: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Nathan Lynch <nathanl@linux.ibm.com>,
Gautham R Shenoy <ego@linux.vnet.ibm.com>,
Oliver OHalloran <oliveroh@au1.ibm.com>,
Michael Neuling <mikey@linux.ibm.com>,
Michael Ellerman <michaele@au1.ibm.com>,
Anton Blanchard <anton@au1.ibm.com>,
linuxppc-dev <linuxppc-dev@lists.ozlabs.org>,
Nick Piggin <npiggin@au1.ibm.com>
Subject: Re: [PATCH 05/11] powerpc/smp: Dont assume l2-cache to be superset of sibling
Date: Mon, 20 Jul 2020 14:28:12 +0530 [thread overview]
Message-ID: <20200720085812.GA6680@in.ibm.com> (raw)
In-Reply-To: <20200720064504.GD21103@linux.vnet.ibm.com>
Hi Srikar,
On Mon, Jul 20, 2020 at 12:15:04PM +0530, Srikar Dronamraju wrote:
> * Gautham R Shenoy <ego@linux.vnet.ibm.com> [2020-07-17 11:30:11]:
>
> > Hi Srikar,
> >
> > On Tue, Jul 14, 2020 at 10:06:18AM +0530, Srikar Dronamraju wrote:
> > > Current code assumes that cpumask of cpus sharing a l2-cache mask will
> > > always be a superset of cpu_sibling_mask.
> > >
> > > Lets stop that assumption.
> > >
> > > Cc: linuxppc-dev <linuxppc-dev@lists.ozlabs.org>
> > > Cc: Michael Ellerman <michaele@au1.ibm.com>
> > > Cc: Nick Piggin <npiggin@au1.ibm.com>
> > > Cc: Oliver OHalloran <oliveroh@au1.ibm.com>
> > > Cc: Nathan Lynch <nathanl@linux.ibm.com>
> > > Cc: Michael Neuling <mikey@linux.ibm.com>
> > > Cc: Anton Blanchard <anton@au1.ibm.com>
> > > Cc: Gautham R Shenoy <ego@linux.vnet.ibm.com>
> > > Cc: Vaidyanathan Srinivasan <svaidy@linux.ibm.com>
> > > Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
> > > ---
> > > arch/powerpc/kernel/smp.c | 28 +++++++++++++++-------------
> > > 1 file changed, 15 insertions(+), 13 deletions(-)
> > >
> > > diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> > > index 7d430fc536cc..875f57e41355 100644
> > > --- a/arch/powerpc/kernel/smp.c
> > > +++ b/arch/powerpc/kernel/smp.c
> > > @@ -1198,6 +1198,7 @@ static bool update_mask_by_l2(int cpu, struct cpumask *(*mask_fn)(int))
> > > struct device_node *l2_cache, *np;
> > > int i;
> > >
> > > + cpumask_set_cpu(cpu, mask_fn(cpu));
> >
> > It would be good to comment why do we need to do set the CPU in the
> > l2-mask if we don't have a l2cache domain.
> >
>
> Good Catch,
> We should move this after the cpu_to_l2cache.
>
> > > l2_cache = cpu_to_l2cache(cpu);
> > > if (!l2_cache)
> > > return false;
> > > @@ -1284,29 +1285,30 @@ static void add_cpu_to_masks(int cpu)
> > > * add it to it's own thread sibling mask.
> > > */
> > > cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
> > > + cpumask_set_cpu(cpu, cpu_core_mask(cpu));
> > >
> > > for (i = first_thread; i < first_thread + threads_per_core; i++)
> > > if (cpu_online(i))
> > > set_cpus_related(i, cpu, cpu_sibling_mask);
> > >
> > > add_cpu_to_smallcore_masks(cpu);
> > > - /*
> > > - * Copy the thread sibling mask into the cache sibling mask
> > > - * and mark any CPUs that share an L2 with this CPU.
> > > - */
> > > - for_each_cpu(i, cpu_sibling_mask(cpu))
> > > - set_cpus_related(cpu, i, cpu_l2_cache_mask);
> > > update_mask_by_l2(cpu, cpu_l2_cache_mask);
> > >
> > > - /*
> > > - * Copy the cache sibling mask into core sibling mask and mark
> > > - * any CPUs on the same chip as this CPU.
> > > - */
> > > - for_each_cpu(i, cpu_l2_cache_mask(cpu))
> > > - set_cpus_related(cpu, i, cpu_core_mask);
> > > + if (pkg_id == -1) {
> > > + struct cpumask *(*mask)(int) = cpu_sibling_mask;
> > > +
> > > + /*
> > > + * Copy the sibling mask into core sibling mask and
> > > + * mark any CPUs on the same chip as this CPU.
> > > + */
> > > + if (shared_caches)
> > > + mask = cpu_l2_cache_mask;
> > > +
> >
> >
> > Now that we decoupling the containment relationship between
> > sibling_mask and l2-cache mask, should we set all the CPUs that are
> > both in cpu_sibling_mask(cpu) as well as cpu_l2_mask(cpu) in
> > cpu_core_mask ?
> >
>
> Are you saying instead of setting this cpu in this cpu_core_mask, can we set
> all the cpus in the mask in cpu_core_mask?
No. What I am referring to is in the for-loop below, you are setting
the CPUs that are set in mask(cpu) in the cpu_core_mask.
Now, the above code sets
mask(cpu) == cpu_sibling_mask(cpu) in the absence of shared_caches, and
== cpu_l2_cache_mask(cpu) in the presence of shared_cache.
Since we have decoupled the assumption that cpu_sibling_mask(cpu) may not
be contained within cpu_l2_cache_mask(cpu), in the presence of a
shared-cache, why are we only picking the CPUs in
cpu_l2_cache_mask(cpu) to be set in cpu_core_maks(cpu) ? It should
ideally be the superset whose CPUs should be set in
cpu_core_mask(cpu). And the correct cpuset is
cpumask_or(cpu_sibling_mask(cpu), cpu_l2_cache_mask(cpu))
> Currently we dont know if any of the cpus of the mask were already set or
> not. Plus we need to anyway update cpumask of all other cpus to says they
> are related. So setting a mask instead of cpu at a time will not change
> anything for our side.
>
> > > + for_each_cpu(i, mask(cpu))
> > > + set_cpus_related(cpu, i, cpu_core_mask);
> > >
> > > - if (pkg_id == -1)
> > > return;
> > > + }
> > >
> > > for_each_cpu(i, cpu_online_mask)
> > > if (get_physical_package_id(i) == pkg_id)
> > > --
> > > 2.17.1
> > >
> > --
> > Thanks and Regards
> > gautham.
>
> --
> Thanks and Regards
> Srikar Dronamraju
next prev parent reply other threads:[~2020-07-20 9:00 UTC|newest]
Thread overview: 42+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-14 4:36 [PATCH 00/11] Support for grouping cores Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 01/11] powerpc/smp: Cache node for reuse Srikar Dronamraju
2020-07-17 4:51 ` Gautham R Shenoy
2020-07-14 4:36 ` [PATCH 02/11] powerpc/smp: Merge Power9 topology with Power topology Srikar Dronamraju
2020-07-17 5:44 ` Gautham R Shenoy
2020-07-20 8:10 ` Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 03/11] powerpc/smp: Move powerpc_topology above Srikar Dronamraju
2020-07-17 5:45 ` Gautham R Shenoy
2020-07-14 4:36 ` [PATCH 04/11] powerpc/smp: Enable small core scheduling sooner Srikar Dronamraju
2020-07-17 5:48 ` Gautham R Shenoy
2020-07-20 7:20 ` Srikar Dronamraju
2020-07-20 7:47 ` Jordan Niethe
2020-07-20 8:52 ` Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 05/11] powerpc/smp: Dont assume l2-cache to be superset of sibling Srikar Dronamraju
2020-07-14 5:40 ` Oliver O'Halloran
2020-07-14 6:30 ` Srikar Dronamraju
2020-07-17 6:00 ` Gautham R Shenoy
2020-07-20 6:45 ` Srikar Dronamraju
2020-07-20 8:58 ` Gautham R Shenoy [this message]
2020-07-14 4:36 ` [PATCH 06/11] powerpc/smp: Generalize 2nd sched domain Srikar Dronamraju
2020-07-17 6:37 ` Gautham R Shenoy
2020-07-20 6:19 ` Srikar Dronamraju
2020-07-20 9:07 ` Gautham R Shenoy
2020-07-14 4:36 ` [PATCH 07/11] Powerpc/numa: Detect support for coregroup Srikar Dronamraju
2020-07-17 8:08 ` Gautham R Shenoy
2020-07-20 13:56 ` Michael Ellerman
2020-07-21 2:57 ` Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 08/11] powerpc/smp: Allocate cpumask only after searching thread group Srikar Dronamraju
2020-07-17 8:08 ` Gautham R Shenoy
2020-07-14 4:36 ` [PATCH 09/11] Powerpc/smp: Create coregroup domain Srikar Dronamraju
2020-07-17 8:19 ` Gautham R Shenoy
2020-07-17 8:23 ` Gautham R Shenoy
2020-07-20 6:02 ` Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 10/11] powerpc/smp: Implement cpu_to_coregroup_id Srikar Dronamraju
2020-07-17 8:26 ` Gautham R Shenoy
2020-07-20 5:48 ` Srikar Dronamraju
2020-07-20 9:10 ` Gautham R Shenoy
2020-07-20 10:26 ` Srikar Dronamraju
2020-07-14 4:36 ` [PATCH 11/11] powerpc/smp: Provide an ability to disable coregroup Srikar Dronamraju
2020-07-17 8:28 ` Gautham R Shenoy
2020-07-20 13:57 ` Michael Ellerman
2020-07-14 5:06 ` [PATCH 00/11] Support for grouping cores Srikar Dronamraju
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200720085812.GA6680@in.ibm.com \
--to=ego@linux.vnet.ibm.com \
--cc=anton@au1.ibm.com \
--cc=linuxppc-dev@lists.ozlabs.org \
--cc=michaele@au1.ibm.com \
--cc=mikey@linux.ibm.com \
--cc=nathanl@linux.ibm.com \
--cc=npiggin@au1.ibm.com \
--cc=oliveroh@au1.ibm.com \
--cc=srikar@linux.vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).