Re: [PATCH v4 3/9] mm/lru: replace pgdat lru_lock with lruvec lock

[Alex Shi <alex.shi@linux.alibaba.com>]

在 2019/11/23 上午12:16, Johannes Weiner 写道:
> On Fri, Nov 22, 2019 at 10:36:32AM +0800, Alex Shi wrote:
>> 在 2019/11/22 上午6:06, Johannes Weiner 写道:
>>> If we could restrict lock_page_lruvec() to working only on PageLRU
>>> pages, we could fix the problem with memory barriers. But this won't
>>> work for split_huge_page(), which is AFAICT the only user that needs
>>> to freeze the lru state of a page that could be isolated elsewhere.
>>>
>>> So AFAICS the only option is to lock out mem_cgroup_move_account()
>>> entirely when the lru_lock is held. Which I guess should be fine.
>>
>> I guess we can try from lock_page_memcg, is that a good start?
> 
> Yes.
> 
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 7e6387ad01f0..f4bbbf72c5b8 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -1224,7 +1224,7 @@ struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgd
>>                 goto out;
>>         }
>>
>> -       memcg = page->mem_cgroup;
>> +       memcg = lock_page_memcg(page);
>>         /*
>>          * Swapcache readahead pages are added to the LRU - and
>>          * possibly migrated - before they are charged.
> 
> test_clear_page_writeback() calls this function with that lock already
> held so that would deadlock. Let's keep locking in lock_page_lruvec().
> 
> lock_page_lruvec():
> 
> 	memcg = lock_page_memcg(page);
> 	lruvec = mem_cgroup_lruvec(page_pgdat(page), memcg);
> 
> 	spin_lock_irqsave(&lruvec->lru_lock, *flags);
> 	return lruvec;
> 
> unlock_lruvec();
> 
> 	spin_unlock_irqrestore(&lruvec->lru_lock);
> 	__unlock_page_memcg(lruvec_memcg(lruvec));
> 
> The lock ordering should be fine as well. But it might be a good idea
> to stick a might_lock(&memcg->move_lock) in lock_page_memcg() before
> that atomic_read() and test with lockdep enabled.

Hi Johannes,

Thanks a lot for the suggestion. I will look into this and try.

> 
> 
> But that leaves me with one more worry: compaction. We locked out
> charge moving now, so between that and knowing that the page is alive,
> we have page->mem_cgroup stable. But compaction doesn't know whether
> the page is alive - it comes from a pfn and finds out using PageLRU.
> 
> In the current code, pgdat->lru_lock remains the same before and after
> the page is charged to a cgroup, so once compaction has that locked
> and it observes PageLRU, it can go ahead and isolate the page.
> 
> But lruvec->lru_lock changes during charging, and then compaction may
> hold the wrong lock during isolation:
> 
> compaction:				generic_file_buffered_read:
> 
> 					page_cache_alloc()
> 
> !PageBuddy()
> 
> lock_page_lruvec(page)
>   lruvec = mem_cgroup_page_lruvec()
>   spin_lock(&lruvec->lru_lock)
>   if lruvec != mem_cgroup_page_lruvec()
>     goto again
> 
> 					add_to_page_cache_lru()
> 					  mem_cgroup_commit_charge()
> 					    page->mem_cgroup = foo
> 					  lru_cache_add()
> 					    __pagevec_lru_add()
> 					      SetPageLRU()
> 
> if PageLRU(page):
>   __isolate_lru_page()
> 
> I don't see what prevents the lruvec from changing under compaction,
> neither in your patches nor in Hugh's. Maybe I'm missing something?

Yes, it's a problem. 
Guess we could move the lruvec recheck after PageLRU() test in compaction. Then it could be safe, and add a bit more burden on compaction should be fine.  at last we have no disturb to file read.

Thanks
Alex