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

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

Message-ID: <7f212c43-519a-ddcc-5cba-d23c487af321@linux.alibaba.com>
Date:   Fri, 12 Nov 2021 11:10:13 +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 11:02, Huang, Ying wrote:
> Baolin Wang <baolin.wang@...ux.alibaba.com> writes:
> 
>> 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.
> 
> Yes.  Please add my "reviewed-by" after changing this.

OK. Thanks for your reviewing.