Message-ID: <015855b3-ced3-8d84-e21d-cc6ce112b556@suse.cz>
Date:   Mon, 20 Mar 2023 09:05:57 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.8.0
Subject: Re: [PATCH] mm/slub: Reduce memory consumption in extreme scenarios
Content-Language: en-US
To:     "chenjun (AM)" <chenjun102@huawei.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        "cl@linux.com" <cl@linux.com>,
        "penberg@kernel.org" <penberg@kernel.org>,
        "rientjes@google.com" <rientjes@google.com>,
        "iamjoonsoo.kim@lge.com" <iamjoonsoo.kim@lge.com>,
        "akpm@linux-foundation.org" <akpm@linux-foundation.org>,
        Hyeonggon Yoo <42.hyeyoo@gmail.com>
Cc:     "xuqiang (M)" <xuqiang36@huawei.com>,
        "Wangkefeng (OS Kernel Lab)" <wangkefeng.wang@huawei.com>,
        Michal Hocko <mhocko@kernel.org>,
        Mike Rapoport <rppt@kernel.org>,
        Mel Gorman <mgorman@techsingularity.net>
References: <20230314123403.100158-1-chenjun102@huawei.com>
 <0cad1ff3-8339-a3eb-fc36-c8bda1392451@suse.cz>
 <344c7521d72e4107b451c19b329e9864@huawei.com>
 <8c700468-245d-72e9-99e7-b99d4547e6d8@suse.cz>
 <aeb2bd3990004b9eb4f151aa833ddcf2@huawei.com>
From:   Vlastimil Babka <vbabka@suse.cz>
In-Reply-To: <aeb2bd3990004b9eb4f151aa833ddcf2@huawei.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 3/19/23 08:22, chenjun (AM) wrote:
> 在 2023/3/17 20:06, Vlastimil Babka 写道:
>> On 3/17/23 12:32, chenjun (AM) wrote:
>>> 在 2023/3/14 22:41, Vlastimil Babka 写道:
>>>>>    	pc.flags = gfpflags;
>>>>> +
>>>>> +	/*
>>>>> +	 * when (node != NUMA_NO_NODE) && (gfpflags & __GFP_THISNODE)
>>>>> +	 * 1) try to get a partial slab from target node with __GFP_THISNODE.
>>>>> +	 * 2) if 1) failed, try to allocate a new slab from target node with
>>>>> +	 *    __GFP_THISNODE.
>>>>> +	 * 3) if 2) failed, retry 1) and 2) without __GFP_THISNODE constraint.
>>>>> +	 */
>>>>> +	if (node != NUMA_NO_NODE && !(gfpflags & __GFP_THISNODE) && try_thisnode)
>>>>> +			pc.flags |= __GFP_THISNODE;
>>>>
>>>> Hmm I'm thinking we should also perhaps remove direct reclaim possibilities
>>>> from the attempt 2). In your qemu test it should make no difference, as it
>>>> fills everything with kernel memory that is not reclaimable. But in practice
>>>> the target node might be filled with user memory, and I think it's better to
>>>> quickly allocate on a different node than spend time in direct reclaim. So
>>>> the following should work I think?
>>>>
>>>> pc.flags = GFP_NOWAIT | __GFP_NOWARN |__GFP_THISNODE
>>>>
>>>
>>> Hmm, Should it be that:
>>>
>>> pc.flags |= GFP_NOWAIT | __GFP_NOWARN |__GFP_THISNODE
>> 
>> No, we need to ignore the other reclaim-related flags that the caller
>> passed, or it wouldn't work as intended.
>> The danger is that we ignore some flag that would be necessary to pass, but
>> I don't think there's any?
>> 
>> 
> 
> If we ignore __GFP_ZERO passed by kzalloc， kzalloc will not work.
> Could we just unmask __GFP_RECLAIMABLE | __GFP_RECLAIM?
> 
> pc.flags &= ~(__GFP_RECLAIMABLE | __GFP_RECLAIM)
> pc.flags |= __GFP_THISNODE

__GFP_RECLAIMABLE would be wrong, but also ignored as new_slab() does:
	flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK)

which would filter out __GFP_ZERO as well. That's not a problem as kzalloc()
will zero out the individual allocated objects, so it doesn't matter if we
don't zero out the whole slab page.

But I wonder, if we're not past due time for a helper e.g.
gfp_opportunistic(flags) that would turn any allocation flags to a
GFP_NOWAIT while keeping the rest of relevant flags intact, and thus there
would be one canonical way to do it - I'm sure there's a number of places
with their own variants now?
With such helper we'd just add __GFP_THISNODE to the result here as that's
specific to this particular opportunistic allocation.