Re: [PATCH v3 0/4] mm/vmscan: fix swappiness=max and clean up per-node proactive reclaim

[Ridong Chen <ridong.chen@linux.dev>]

On 8/14/2026 6:37 AM, Barry Song wrote:
> On Tue, Jul 28, 2026 at 4:34 PM Ridong Chen <ridong.chen@linux.dev> wrote:
>>
>>
>>
>> On 7/24/2026 7:12 PM, Barry Song wrote:
>>> On Fri, Jul 24, 2026 at 10:43 AM Ridong Chen <ridong.chen@linux.dev> wrote:
>>>>
>>>>
>>>>
>>>> On 7/24/2026 8:18 AM, Andrew Morton wrote:
>>>>> On Thu, 23 Jul 2026 12:57:14 +0800 Ridong <ridong.chen@linux.dev> wrote:
>>>>>
>>>>>> Fixes and one cleanup.
>>>>>>
>>>>>> Patch 1 fixes "swappiness=max": the anon-only test in get_scan_count()
>>>>>> sat after the "cannot reclaim anon" check, so when no anon was
>>>>>> reclaimable the request fell back to SCAN_FILE and evicted page cache
>>>>>> instead.
>>>>>>
>>>>>> Patch 2 fixes reclaim_store() collapsing every error into -EAGAIN, so
>>>>>> callers can no longer tell an invalid argument from a busy interface;
>>>>>> propagate the real error code, matching the memcg path.
>>>>>>
>>>>>> Patch 3 drops the now-unused gfp_mask parameter from __node_reclaim().
>>>>>>
>>>>>> Patch 4 fixes the same "swappiness=max" issue for MGLRU.
>>>>>
>>>>> Thanks.  AI review might have found several pre-existing issues which
>>>>> might interest the maintainers.  Also a suggestion against your [4/4]:
>>>>>
>>>>>         https://sashiko.dev/#/patchset/20260723045718.2052070-1-ridong.chen@linux.dev
>>>>
>>>> Hi all, sashiko review:
>>>>
>>>>      This is a pre-existing issue, but I noticed that per-node proactive reclaim
>>>>      fails to reclaim memory if the node's unmapped file or slab pages are below
>>>>      minimum thresholds, completely ignoring anonymous memory.
>>>>      In mm/vmscan.c:__node_reclaim():
>>>>           if (node_pagecache_reclaimable(pgdat) > pgdat->min_unmapped_pages ||
>>>>               node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) >
>>>> pgdat->min_slab_pages) {
>>>>      Does this condition incorrectly block proactive reclaim of anonymous memory
>>>>      by applying legacy zone-reclaim thresholds?
>>>>      It seems to bypass shrink_node() entirely if file/slab pages are low, even
>>>>      when sc->proactive is set, which could falsely return -EAGAIN to userspace
>>>>      for anon-heavy workloads.
>>>>
>>>> Should we add:
>>>>
>>>> if (sc->proactive ||
>>>>        node_pagecache_reclaimable(pgdat) > pgdat->min_unmapped_pages ||
>>>>        node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) > pgdat->min_slab_pages) {
>>>
>>
>> Hi Barry, sorry for the late reply.
>>
>>> Nop.
>>> I assume reclaiming file cache and slab becomes problematic when their
>>> amounts are already very limited, so we should still honor these two
>>> checks.
>>>
>>> Maybe we could relax them only when swappiness == 201
>>> (SWAPPINESS_ANON_ONLY)?
>>>
>>> BTW, for global proactive reclaim, when setting swappiness to 201, does
>>> it prevent slab shrinking? If not, it seems problematic when
>>> node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) <
>>> pgdat->min_slab_pages.
>>>
>>
>> in __node_reclaim, we will shrink the node when
>> node_pagecache_reclaimable(pgdat) > pgdat->min_unmapped_pages, even if
>> node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) <= pgdat->min_slab_pages.
>> This means slab shrinking can still occur even when below the limit (since
>> shrink_slab is called unconditionally after shrink_lruvec). This is not an issue
>> only for global proactive reclaim.
>>
>>
>>> Your recent patchset prevents all file reclamation when swappiness is
>>> set to 201, so we only need to check whether there could be a slab issue
>>> before allowing shrink_node() to continue in this case.
>>>
>>
>> So can we add just like?
>>
>> if ((sc->proactive && node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) >
>> pgdat->min_slab_pages) ||
>>          node_pagecache_reclaimable(pgdat) > pgdat->min_unmapped_pages ||
>>          node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) > pgdat->min_slab_pages) {
> 
> I feel both pgdat->min_unmapped_pages and
> pgdat->min_slab_pages are quite broken in mainline.
> 
> For example, even when the page cache is below
> min_unmapped_pages, it may still be reclaimed. Similarly, slab may
> still be reclaimed even when it is below min_slab_pages.
> 
> Also, when both the page cache and slab are below their respective
> thresholds, node_reclaim() may reclaim nothing even if we have
> plenty of anon folios available.
> 
>          if (node_pagecache_reclaimable(pgdat) <= pgdat->min_unmapped_pages &&
>              node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) <=
>              pgdat->min_slab_pages)
>                  return 0;
> 
> For example, if slab > min_slab_pages but the page cache is below
> min_unmapped_pages, we still reclaim file pages, even though the
> comment says we should not.
> 
> So we are not going to introduce another broken mechanism.
> Maybe we should start by fixing the existing broken protection
> against reclaiming slab and page cache?
> 
For example, Maybe we can skip shrink_slab when node_page_state_pages(pgdat, 
NR_SLAB_RECLAIMABLE_B) <= pgdat->min_slab_pages?

And similarly, in get_scan_count, we could avoid reclaiming file page cache if 
node_pagecache_reclaimable(pgdat) <= pgdat->min_unmapped_pages.

-- 
Best regards
Ridong