Re: [PATCH v4 6/9] hugetlb: disable region_add file_region coalescing

[Mike Kravetz <mike.kravetz@oracle.com>]

On 9/10/19 4:31 PM, Mina Almasry wrote:
> A follow up patch in this series adds hugetlb cgroup uncharge info the
> file_region entries in resv->regions. The cgroup uncharge info may
> differ for different regions, so they can no longer be coalesced at
> region_add time. So, disable region coalescing in region_add in this
> patch.
> 
> Behavior change:
> 
> Say a resv_map exists like this [0->1], [2->3], and [5->6].
> 
> Then a region_chg/add call comes in region_chg/add(f=0, t=5).
> 
> Old code would generate resv->regions: [0->5], [5->6].
> New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5],
> [5->6].
> 
> Special care needs to be taken to handle the resv->adds_in_progress
> variable correctly. In the past, only 1 region would be added for every
> region_chg and region_add call. But now, each call may add multiple
> regions, so we can no longer increment adds_in_progress by 1 in region_chg,
> or decrement adds_in_progress by 1 after region_add or region_abort. Instead,
> region_chg calls add_reservation_in_range() to count the number of regions
> needed and allocates those, and that info is passed to region_add and
> region_abort to decrement adds_in_progress correctly.

Hate to throw more theoretical examples at you but ...

Consider an existing reserv_map like [3-10]
Then a region_chg/add call comes in region_chg/add(f=0, t=10).
The region_chg is going to return 3 (additional reservations needed), and
also out_regions_needed = 1 as it would want to create a region [0-3].
Correct?
But, there is nothing to prevent another thread from doing a region_del [5-7]
after the region_chg and before region_add.  Correct?
If so, it seems the region_add would need to create two regions, but there
is only one in the cache and we would BUG in get_file_region_entry_from_cache.
Am I reading the code correctly?

The existing code wants to make sure region_add called after region_chg will
never return error.  This is why all needed allocations were done in the
region_chg call, and it was relatively easy to do in existing code when
region_chg would only need one additional region at most.

I'm thinking that we may have to make region_chg allocate the worst case
number of regions (t - f)/2, OR change to the code such that region_add
could return an error.
-- 
Mike Kravetz