Re: [PATCH] mm: release the spinlock on zap_pte_range

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Jul 30, 2019 at 02:57:51PM +0200, Michal Hocko wrote:
> [Cc Nick - the email thread starts http://lkml.kernel.org/r/20190729071037.241581-1-minchan@xxxxxxxxxx
>  A very brief summary is that mark_page_accessed seems to be quite
>  expensive and the question is whether we still need it and why
>  SetPageReferenced cannot be used instead. More below.]
> 
> On Tue 30-07-19 21:39:35, Minchan Kim wrote:
> > On Tue, Jul 30, 2019 at 02:32:37PM +0200, Michal Hocko wrote:
> > > On Tue 30-07-19 21:11:10, Minchan Kim wrote:
> > > > On Mon, Jul 29, 2019 at 10:35:15AM +0200, Michal Hocko wrote:
> > > > > On Mon 29-07-19 17:20:52, Minchan Kim wrote:
> > > > > > On Mon, Jul 29, 2019 at 09:45:23AM +0200, Michal Hocko wrote:
> > > > > > > On Mon 29-07-19 16:10:37, Minchan Kim wrote:
> > > > > > > > In our testing(carmera recording), Miguel and Wei found unmap_page_range
> > > > > > > > takes above 6ms with preemption disabled easily. When I see that, the
> > > > > > > > reason is it holds page table spinlock during entire 512 page operation
> > > > > > > > in a PMD. 6.2ms is never trivial for user experince if RT task couldn't
> > > > > > > > run in the time because it could make frame drop or glitch audio problem.
> > > > > > > 
> > > > > > > Where is the time spent during the tear down? 512 pages doesn't sound
> > > > > > > like a lot to tear down. Is it the TLB flushing?
> > > > > > 
> > > > > > Miguel confirmed there is no such big latency without mark_page_accessed
> > > > > > in zap_pte_range so I guess it's the contention of LRU lock as well as
> > > > > > heavy activate_page overhead which is not trivial, either.
> > > > > 
> > > > > Please give us more details ideally with some numbers.
> > > > 
> > > > I had a time to benchmark it via adding some trace_printk hooks between
> > > > pte_offset_map_lock and pte_unmap_unlock in zap_pte_range. The testing
> > > > device is 2018 premium mobile device.
> > > > 
> > > > I can get 2ms delay rather easily to release 2M(ie, 512 pages) when the
> > > > task runs on little core even though it doesn't have any IPI and LRU
> > > > lock contention. It's already too heavy.
> > > > 
> > > > If I remove activate_page, 35-40% overhead of zap_pte_range is gone
> > > > so most of overhead(about 0.7ms) comes from activate_page via
> > > > mark_page_accessed. Thus, if there are LRU contention, that 0.7ms could
> > > > accumulate up to several ms.
> > > 
> > > Thanks for this information. This is something that should be a part of
> > > the changelog. I am sorry to still poke into this because I still do not
> > 
> > I will include it.
> > 
> > > have a full understanding of what is going on and while I do not object
> > > to drop the spinlock I still suspect this is papering over a deeper
> > > problem.
> > 
> > I couldn't come up with better solution. Feel free to suggest it.
> > 
> > > 
> > > If mark_page_accessed is really expensive then why do we even bother to
> > > do it in the tear down path in the first place? Why don't we simply set
> > > a referenced bit on the page to reflect the young pte bit? I might be
> > > missing something here of course.
> > 
> > commit bf3f3bc5e73
> > Author: Nick Piggin <npiggin@xxxxxxx>
> > Date:   Tue Jan 6 14:38:55 2009 -0800
> > 
> >     mm: don't mark_page_accessed in fault path
> > 
> >     Doing a mark_page_accessed at fault-time, then doing SetPageReferenced at
> >     unmap-time if the pte is young has a number of problems.
> > 
> >     mark_page_accessed is supposed to be roughly the equivalent of a young pte
> >     for unmapped references. Unfortunately it doesn't come with any context:
> >     after being called, reclaim doesn't know who or why the page was touched.
> > 
> >     So calling mark_page_accessed not only adds extra lru or PG_referenced
> >     manipulations for pages that are already going to have pte_young ptes anyway,
> >     but it also adds these references which are difficult to work with from the
> >     context of vma specific references (eg. MADV_SEQUENTIAL pte_young may not
> >     wish to contribute to the page being referenced).
> > 
> >     Then, simply doing SetPageReferenced when zapping a pte and finding it is
> >     young, is not a really good solution either. SetPageReferenced does not
> >     correctly promote the page to the active list for example. So after removing
> >     mark_page_accessed from the fault path, several mmap()+touch+munmap() would
> >     have a very different result from several read(2) calls for example, which
> >     is not really desirable.
> 
> Well, I have to say that this is rather vague to me. Nick, could you be
> more specific about which workloads do benefit from this change? Let's
> say that the zapped pte is the only referenced one and then reclaim
> finds the page on inactive list. We would go and reclaim it. But does
> that matter so much? Hot pages would be referenced from multiple ptes
> very likely, no?

As Nick mentioned in the description, without mark_page_accessed in
zapping part, repeated mmap + touch + munmap never acticated the page
while several read(2) calls easily promote it.




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux