Message-ID: <930e111a-b13b-43d9-93f8-5bf28f343074@gmail.com>
Date: Mon, 5 Aug 2024 00:04:22 +0100
From: Usama Arif <usamaarif642@...il.com>
To: David Hildenbrand <david@...hat.com>, akpm@...ux-foundation.org,
 linux-mm@...ck.org
Cc: hannes@...xchg.org, riel@...riel.com, shakeel.butt@...ux.dev,
 roman.gushchin@...ux.dev, yuzhao@...gle.com, baohua@...nel.org,
 ryan.roberts@....com, rppt@...nel.org, willy@...radead.org,
 cerasuolodomenico@...il.com, corbet@....net, linux-kernel@...r.kernel.org,
 linux-doc@...r.kernel.org, kernel-team@...a.com
Subject: Re: [PATCH 0/6] mm: split underutilized THPs



On 01/08/2024 07:36, David Hildenbrand wrote:
>>> I just added a bunch of quick printfs to QEMU and ran a precopy+postcopy live migration. Looks like my assumption was right:
>>>
>>> On the destination:
>>>
>>> Writing received pages during precopy # ram_load_precopy()
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Writing received pages during precopy
>>> Disabling THP: MADV_NOHUGEPAGE # postcopy_ram_prepare_discard()
>>> Discarding pages # loadvm_postcopy_ram_handle_discard()
>>> Discarding pages
>>> Discarding pages
>>> Discarding pages
>>> Discarding pages
>>> Discarding pages
>>> Discarding pages
>>> Registering UFFD # postcopy_ram_incoming_setup()
>>>
>>
>> Thanks for this, yes it makes sense after you mentioned postcopy_ram_incoming_setup.
>> postcopy_ram_incoming_setup happens in the Listen phase, which is after the discard phase, so I was able to follow in code in qemu the same sequence of events that the above prints show.
> 
> 
> I just added another printf to postcopy_ram_supported_by_host(), where we temporarily do a UFFDIO_REGISTER on some test area.
> 
> Sensing UFFD support # postcopy_ram_supported_by_host()
> Sensing UFFD support
> Writing received pages during precopy # ram_load_precopy()
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Writing received pages during precopy
> Disabling THP: MADV_NOHUGEPAGE # postcopy_ram_prepare_discard()
> Discarding pages # loadvm_postcopy_ram_handle_discard()
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Discarding pages
> Registering UFFD # postcopy_ram_incoming_setup()
> 
> We could think about using this "ever user uffd" to avoid the shared zeropage in most processes.
> 
> Of course, there might be other applications where that wouldn't work, but I think this behavior (write to area before enabling uffd) might be fairly QEMU specific already.
> 
> Avoiding the shared zeropage has the benefit that a later write fault won't have to do a TLB flush and can simply install a fresh anon page.
> 

I checked CRIU and that does a check at the start as well before attempting to use uffd: https://github.com/checkpoint-restore/criu/blob/criu-dev/criu/kerndat.c#L1349

If writing to an area before enabling uffd is likely to be QEMU specific, then you make a good point to clear pte instead of using shared zeropage to avoid the TLB flush if uffd is ever used.

I think "ever used uffd" would need to be tracked using mm_struct. This also won't cause an issue if the check is done in a parent process and the actual use is in a forked process, as copy_mm should take care of it.
The possibilities would then be:
1) Have a new bit in mm->flags, set it in new_userfaultfd and test it in try_to_unmap_unused, but unfortunately all the bits in mm->flags are taken.
2) We could use mm->def_flags as it looks like there is an unused bit (0x800) just before VM_UFFD_WP. But that makes the code confusing as its used to initialize the default flags for VMAs and is not supposed to be used as a "mm flag".
3) Introducing mm->flags2 and set/test as 1. This would introduce a 8 byte overhead for all mm_structs.

I am not sure either 2 or 3 are acceptable upstream, unless there is a need for more flags in the near future and the 8 byte overhead starts to make sense. Maybe we go with shared zeropage?

Powered by blists - more mailing lists