linux-kernel - Re: [PATCH v3 1/4] sched/fair: add SD

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <b2d1ceb8-4cce-4249-a282-619e6564fbf5@gmail.com>
Date: Wed, 7 Feb 2024 10:36:04 +0800
From: kuiliang Shi <seakeel@...il.com>
To: Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>,
 Valentin Schneider <vschneid@...hat.com>
Cc: alexs@...nel.org, Ingo Molnar <mingo@...hat.com>,
 Peter Zijlstra <peterz@...radead.org>, Juri Lelli <juri.lelli@...hat.com>,
 Vincent Guittot <vincent.guittot@...aro.org>,
 Dietmar Eggemann <dietmar.eggemann@....com>,
 Steven Rostedt <rostedt@...dmis.org>, Ben Segall <bsegall@...gle.com>,
 Daniel Bristot de Oliveira <bristot@...hat.com>,
 linux-kernel@...r.kernel.org, sshegde@...ux.ibm.com
Subject: Re: [PATCH v3 1/4] sched/fair: add SD_CLUSTER in comments



On 2/7/24 5:57 AM, Ricardo Neri wrote:
> On Tue, Feb 06, 2024 at 02:16:06PM +0100, Valentin Schneider wrote:
>> On 05/02/24 18:46, Ricardo Neri wrote:
>>> On Fri, Feb 02, 2024 at 03:27:32PM +0100, Valentin Schneider wrote:
>>>>
>>>> Subject nit: the prefix should be sched/topology
>>>>
>>>> On 01/02/24 19:54, alexs@...nel.org wrote:
>>>>> From: Alex Shi <alexs@...nel.org>
>>>>>
>>>>> The description of SD_CLUSTER is missing. Add it.
>>>>>
>>>>> Signed-off-by: Alex Shi <alexs@...nel.org>
>>>>> To: Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>
>>>>> To: Valentin Schneider <vschneid@...hat.com>
>>>>> To: Vincent Guittot <vincent.guittot@...aro.org>
>>>>> To: Juri Lelli <juri.lelli@...hat.com>
>>>>> To: Peter Zijlstra <peterz@...radead.org>
>>>>> To: Ingo Molnar <mingo@...hat.com>
>>>>> ---
>>>>>  kernel/sched/topology.c | 1 +
>>>>>  1 file changed, 1 insertion(+)
>>>>>
>>>>> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
>>>>> index 10d1391e7416..8b45f16a1890 100644
>>>>> --- a/kernel/sched/topology.c
>>>>> +++ b/kernel/sched/topology.c
>>>>> @@ -1554,6 +1554,7 @@ static struct cpumask		***sched_domains_numa_masks;
>>>>>   * function:
>>>>>   *
>>>>>   *   SD_SHARE_CPUCAPACITY   - describes SMT topologies
>>>>> + *   SD_CLUSTER             - describes CPU Cluster topologies
>>>>
>>>> So I know this is the naming we've gone for the "Cluster" naming, but this
>>>> comment isn't really explaining anything.
>>>>
>>>> include/linux/sched/sd_flags.h has a bit more info already:
>>>>  * Domain members share CPU cluster (LLC tags or L2 cache)
>>>
>>> I also thought of this, but I didn't want to suggest to repeat in topolog.c
>>> what is described in sd_flags.h.
>>>
>>> Maybe it is better to remove the descriptions of all flags here and instead
>>> direct the reader to sd_flags.h?
>>>
>>
>> Yeah I agree on less duplication.
>>
>>>>
>>>> I had to go through a bit of git history to remember what the CLUSTER thing
>>>> was about, how about this:
>>>>
>>>> * SD_CLUSTER             - describes shared shared caches, cache tags or busses
>>>
>>> AFAIK, this describes a subset of CPUs in the package that share a
>>> resource, likely L2 cache.
>>>
>>>> * SD_SHARE_PKG_RESOURCES - describes shared LLC cache
>>>>
>>>> And looking at this it would make sense to:
>>>>   rename SD_CLUSTER into SD_SHARE_PKG_RESOURCES
>>>
>>> but not all CPUs in the package share the resource
>>
>> But SD_CLUSTER never expands beyond the package, right?
> 
> Correct.
> 
>>
>> Regardless, my main point is that having both SD_CLUSTER and
>> SD_SHARE_PKG_RESOURCES is a source of confusion (at the very least for
>> myself),
> 
> Agreed!
> 
>> and given SD_SHARE_PKG_RESOURCES is really used to mean "shares
>> LLC" (see update_top_cache_domain()), we could make that flag more explicit
>> and lift some ambiguity with SD_CLUSTER.
> 
> As Yicong stated, cluster topology should mean CPUs beyond SMT that share
> some resource but not LLC.
> 
> It makes sense to me to keep SD_CLUSTER name as it is today and rename
> SD_SHARE_PKG_RESOURCES as SD_SHARE_LLC.

yes, agree with his.