Received-SPF: pass (google.com: domain of linux-kernel+bounces-72467-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
From: "Huang, Ying" <ying.huang@intel.com>
To: "Aneesh Kumar K.V" <aneesh.kumar@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>,  Donet Tom
 <donettom@linux.ibm.com>,  linux-mm@kvack.org,
  linux-kernel@vger.kernel.org,  Dave Hansen <dave.hansen@linux.intel.com>,
  Mel Gorman <mgorman@suse.de>,  Ben Widawsky <ben.widawsky@intel.com>,
  Feng Tang <feng.tang@intel.com>,  Michal Hocko <mhocko@kernel.org>,
  Andrea Arcangeli <aarcange@redhat.com>,  Peter Zijlstra
 <peterz@infradead.org>,  Ingo Molnar <mingo@redhat.com>,  Rik van Riel
 <riel@surriel.com>,  Johannes Weiner <hannes@cmpxchg.org>,  Matthew Wilcox
 <willy@infradead.org>,  Mike Kravetz <mike.kravetz@oracle.com>,  Vlastimil
 Babka <vbabka@suse.cz>,  Dan Williams <dan.j.williams@intel.com>,  Hugh
 Dickins <hughd@google.com>,  Kefeng Wang <wangkefeng.wang@huawei.com>,
  Suren Baghdasaryan <surenb@google.com>
Subject: Re: [PATCH 1/3] mm/mempolicy: Use the already fetched local variable
In-Reply-To: <b599bfe5-1c4d-4750-b0d6-a086e1c8a34c@kernel.org> (Aneesh Kumar
	K. V.'s message of "Tue, 20 Feb 2024 12:33:51 +0530")
References: <9c3f7b743477560d1c5b12b8c111a584a2cc92ee.1708097962.git.donettom@linux.ibm.com>
	<20240218133851.22c22b55460e866a099be5ce@linux-foundation.org>
	<63a0f7c4-3c3f-4097-9a24-d1e3fc7b6030@linux.ibm.com>
	<20240219172130.82a16c1ebecbf8ba86a8987d@linux-foundation.org>
	<21f343fa-84a7-4539-91e2-6fc963dbfb62@kernel.org>
	<87frxnps8w.fsf@yhuang6-desk2.ccr.corp.intel.com>
	<7097ff95-6077-4744-a770-b90d224c0c9b@kernel.org>
	<b599bfe5-1c4d-4750-b0d6-a086e1c8a34c@kernel.org>
Date: Tue, 20 Feb 2024 15:22:07 +0800
Message-ID: <8734tnppls.fsf@yhuang6-desk2.ccr.corp.intel.com>
User-Agent: Gnus/5.13 (Gnus v5.13)
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=ascii

"Aneesh Kumar K.V" <aneesh.kumar@kernel.org> writes:

> On 2/20/24 12:02 PM, Aneesh Kumar K.V wrote:
>> On 2/20/24 11:55 AM, Huang, Ying wrote:
>>> "Aneesh Kumar K.V" <aneesh.kumar@kernel.org> writes:
>>>
>>>> On 2/20/24 6:51 AM, Andrew Morton wrote:
>>>>> On Mon, 19 Feb 2024 14:04:23 +0530 Donet Tom <donettom@linux.ibm.com> wrote:
>>>>>
>>>>>>>> --- a/mm/mempolicy.c
>>>>>>>> +++ b/mm/mempolicy.c
>>>>>>>> @@ -2526,7 +2526,7 @@ int mpol_misplaced(struct folio *folio, struct vm_area_struct *vma,
>>>>>>>>   		if (node_isset(curnid, pol->nodes))
>>>>>>>>   			goto out;
>>>>>>>>   		z = first_zones_zonelist(
>>>>>>>> -				node_zonelist(numa_node_id(), GFP_HIGHUSER),
>>>>>>>> +				node_zonelist(thisnid, GFP_HIGHUSER),
>>>>>>>>   				gfp_zone(GFP_HIGHUSER),
>>>>>>>>   				&pol->nodes);
>>>>>>>>   		polnid = zone_to_nid(z->zone);
>>>>>>> 	int thisnid = cpu_to_node(thiscpu);
>>>>>>>
>>>>>>> Is there any dofference between numa_node_id() and
>>>>>>> cpu_to_node(raw_smp_processor_id())?  And it it explicable that we're
>>>>>>> using one here and not the other?
>>>>>>
>>>>>> Hi Andrew
>>>>>>
>>>>>> Both numa_node_id() and cpu_to_node(raw_smp_processor_id()) return the current execution node id,
>>>>>> Since the current execution node is already fetched at the beginning (thisnid) we can reuse it instead of getting it again.
>>>>>
>>>>> Sure, but mine was a broader thought: why do we have both?  Is one
>>>>> preferable and if so why?
>>>>
>>>> IIUC these are two helpers to fetch current numa node id. and either of them can be used based on need. The default implementation shows the details.
>>>> (One small difference is numa_node_id() can use optimized per cpu reader because it is fetching the per cpu variable of the currently running cpu.)
>>>>
>>>> #ifndef numa_node_id
>>>> /* Returns the number of the current Node. */
>>>> static inline int numa_node_id(void)
>>>> {
>>>> 	return raw_cpu_read(numa_node);
>>>> }
>>>> #endif
>>>>
>>>> #ifndef cpu_to_node
>>>> static inline int cpu_to_node(int cpu)
>>>> {
>>>> 	return per_cpu(numa_node, cpu);
>>>> }
>>>> #endif
>>>>
>>>> In mpol_misplaced function, we need the cpu details because we are using that in other place (should_numa_migreate_memory()). So it makes it easy
>>>> to use cpu_to_node(thiscpu) instead of numa_node_id(). 
>>>
>>> IIUC, numa_node_id() is faster than cpu_to_node(thiscpu), even if we
>>> have thiscpu already.  cpu_to_node() is mainly used to get the node of
>>> NOT current CPU.  So, IMHO, we should only use numa_node_id() in this
>>> function.
>>>
>> 
>> This change?
>> 
>> modified   mm/mempolicy.c
>> @@ -2502,8 +2502,7 @@ int mpol_misplaced(struct folio *folio, struct vm_area_struct *vma,
>>  	pgoff_t ilx;
>>  	struct zoneref *z;
>>  	int curnid = folio_nid(folio);
>> -	int thiscpu = raw_smp_processor_id();
>> -	int thisnid = cpu_to_node(thiscpu);
>> +	int thisnid = numa_node_id();
>>  	int polnid = NUMA_NO_NODE;
>>  	int ret = NUMA_NO_NODE;
>>  
>> @@ -2573,7 +2572,7 @@ int mpol_misplaced(struct folio *folio, struct vm_area_struct *vma,
>>  		polnid = thisnid;
>>  
>>  		if (!should_numa_migrate_memory(current, folio, curnid,
>> -						thiscpu))
>> +						raw_smp_processor_id()))
>>  			goto out;
>>  	}
>>  
>> 
>
> One of the problem with the above change will be the need to make sure smp processor id remaining stable, which
> I am not sure we want in this function. With that we can end up with processor id not related to the numa node id
> we are using. 

This isn't an issue now, because mpol_misplaced() are always called with
PTL held.  And, we can still keep thiscpu local variable.

--
Best Regards,
Huang, Ying