Re: [PATCH] mm: compaction: skip memory compaction when there are not enough migratable pages

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 2025/1/6 16:49, Ge Yang wrote:
> 
> 
> 在 2025/1/6 16:12, Baolin Wang 写道:
>>
>>
>> On 2025/1/4 16:58, yangge1116@126.com wrote:
>>> From: yangge <yangge1116@126.com>
>>>
>>> There are 4 NUMA nodes on my machine, and each NUMA node has 32GB
>>> of memory. I have configured 16GB of CMA memory on each NUMA node,
>>> and starting a 32GB virtual machine with device passthrough is
>>> extremely slow, taking almost an hour.
>>>
>>> During the start-up of the virtual machine, it will call
>>> pin_user_pages_remote(..., FOLL_LONGTERM, ...) to allocate memory.
>>> Long term GUP cannot allocate memory from CMA area, so a maximum of
>>> 16 GB of no-CMA memory on a NUMA node can be used as virtual machine
>>> memory. There is 16GB of free CMA memory on a NUMA node, which is
>>> sufficient to pass the order-0 watermark check, causing the
>>> __compaction_suitable() function to  consistently return true.
>>> However, if there aren't enough migratable pages available, performing
>>> memory compaction is also meaningless. Besides checking whether
>>> the order-0 watermark is met, __compaction_suitable() also needs
>>> to determine whether there are sufficient migratable pages available
>>> for memory compaction.
>>>
>>> For costly allocations, because __compaction_suitable() always
>>> returns true, __alloc_pages_slowpath() can't exit at the appropriate
>>> place, resulting in excessively long virtual machine startup times.
>>> Call trace:
>>> __alloc_pages_slowpath
>>>      if (compact_result == COMPACT_SKIPPED ||
>>>          compact_result == COMPACT_DEFERRED)
>>>          goto nopage; // should exit __alloc_pages_slowpath() from here
>>>
>>> When the 16G of non-CMA memory on a single node is exhausted, we will
>>> fallback to allocating memory on other nodes. In order to quickly
>>> fallback to remote nodes, we should skip memory compaction when
>>> migratable pages are insufficient. After this fix, it only takes a
>>> few tens of seconds to start a 32GB virtual machine with device
>>> passthrough functionality.
>>>
>>> Signed-off-by: yangge <yangge1116@126.com>
>>> ---
>>>   mm/compaction.c | 19 +++++++++++++++++++
>>>   1 file changed, 19 insertions(+)
>>>
>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>> index 07bd227..1c469b3 100644
>>> --- a/mm/compaction.c
>>> +++ b/mm/compaction.c
>>> @@ -2383,7 +2383,26 @@ static bool __compaction_suitable(struct zone 
>>> *zone, int order,
>>>                     int highest_zoneidx,
>>>                     unsigned long wmark_target)
>>>   {
>>> +    pg_data_t *pgdat = zone->zone_pgdat;
>>> +    unsigned long sum, nr_pinned;
>>>       unsigned long watermark;
>>> +
>>> +    sum = node_page_state(pgdat, NR_INACTIVE_FILE) +
>>> +        node_page_state(pgdat, NR_INACTIVE_ANON) +
>>> +        node_page_state(pgdat, NR_ACTIVE_FILE) +
>>> +        node_page_state(pgdat, NR_ACTIVE_ANON);
>>> +
>>> +    nr_pinned = node_page_state(pgdat, NR_FOLL_PIN_ACQUIRED) -
>>> +        node_page_state(pgdat, NR_FOLL_PIN_RELEASED);
>>> +
>>> +    /*
>>> +     * Gup-pinned pages are non-migratable. After subtracting these 
>>> pages,
>>> +     * we need to check if the remaining pages are sufficient for 
>>> memory
>>> +     * compaction.
>>> +     */
>>> +    if ((sum - nr_pinned) < (1 << order))
>>> +        return false;
>>> +
>>
>> IMO, using the node's statistics to determine whether the zone is 
>> suitable for compaction doesn't make sense. It is possible that even 
>> though the normal zone has long-term pinned pages, the movable zone 
>> can still be suitable for compaction.
> If all the memory used on a node is pinned, then this memory cannot be 
> migrated anymore, and memory compaction operations would not succeed.
> I haven't used movable zone before, can you explain why memory 
> compaction is still necessary? Thank you.

Please consider unevictable folios that are not in the active/inactive 
file/anon LRU lists, yet can still be migrated.