linux-kernel - Re: [PATCH v3] mm: migrate: Support multiple target nodes demotion

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 for Android: free password hash cracker in your pocket

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

Date:   Fri, 12 Nov 2021 10:58:42 +0800
From:   Baolin Wang <baolin.wang@...ux.alibaba.com>
To:     "Huang, Ying" <ying.huang@...el.com>
Cc:     akpm@...ux-foundation.org, dave.hansen@...ux.intel.com,
        ziy@...dia.com, osalvador@...e.de, shy828301@...il.com,
        zhongjiang-ali@...ux.alibaba.com, xlpang@...ux.alibaba.com,
        linux-mm@...ck.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3] mm: migrate: Support multiple target nodes demotion



On 2021/11/12 10:44, Huang, Ying wrote:
> Baolin Wang <baolin.wang@...ux.alibaba.com> writes:
> 
>> We have some machines with multiple memory types like below, which
>> have one fast (DRAM) memory node and two slow (persistent memory) memory
>> nodes. According to current node demotion policy, if node 0 fills up,
>> its memory should be migrated to node 1, when node 1 fills up, its
>> memory will be migrated to node 2: node 0 -> node 1 -> node 2 ->stop.
>>
>> But this is not efficient and suitbale memory migration route
>> for our machine with multiple slow memory nodes. Since the distance
>> between node 0 to node 1 and node 0 to node 2 is equal, and memory
>> migration between slow memory nodes will increase persistent memory
>> bandwidth greatly, which will hurt the whole system's performance.
>>
>> Thus for this case, we can treat the slow memory node 1 and node 2
>> as a whole slow memory region, and we should migrate memory from
>> node 0 to node 1 and node 2 if node 0 fills up.
>>
>> This patch changes the node_demotion data structure to support multiple
>> target nodes, and establishes the migration path to support multiple
>> target nodes with validating if the node distance is the best or not.
>>
>> available: 3 nodes (0-2)
>> node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
>> node 0 size: 62153 MB
>> node 0 free: 55135 MB
>> node 1 cpus:
>> node 1 size: 127007 MB
>> node 1 free: 126930 MB
>> node 2 cpus:
>> node 2 size: 126968 MB
>> node 2 free: 126878 MB
>> node distances:
>> node   0   1   2
>>    0:  10  20  20
>>    1:  20  10  20
>>    2:  20  20  10
>>
>> Signed-off-by: Baolin Wang <baolin.wang@...ux.alibaba.com>

snip

>>   	/*
>>   	 * 'next_pass' contains nodes which became migration
>> @@ -3192,6 +3281,14 @@ static int __init migrate_on_reclaim_init(void)
>>   {
>>   	int ret;
>>   
>> +	/*
>> +	 * Ignore allocation failure, if this kmalloc fails
>> +	 * at boot time, we are likely in bigger trouble.
>> +	 */
>> +	node_demotion = kmalloc_array(nr_node_ids,
>> +				      sizeof(struct demotion_nodes),
>> +				      GFP_KERNEL);
>> +
> 
> I think we should WARN_ON() here.

In this unlikey case, I think the mm core will print more information, 
IMHO WARN_ON() will help little. Anyway no strong opinion on this. Other 
than that, can I get your reviewed-by tag with this nit fixed? Thanks.