On Mon, Oct 21, 2019 at 12:02 PM Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
>
> On 10/12/19 5:30 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.
> >
> > Signed-off-by: Mina Almasry <almasrymina@xxxxxxxxxx>
> >
> > ---
> >
> > Changes in v6:
> > - Fix bug in number of region_caches allocated by region_chg
> >
> > ---
> > mm/hugetlb.c | 256 +++++++++++++++++++++++++++++----------------------
> > 1 file changed, 147 insertions(+), 109 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 4a60d7d44b4c3..f9c1947925bb9 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> <snip>
> > -static long region_chg(struct resv_map *resv, long f, long t)
> > +static long region_chg(struct resv_map *resv, long f, long t,
> > + long *out_regions_needed)
> > {
> > + struct file_region *trg = NULL;
> > long chg = 0;
> >
> > + /* Allocate the maximum number of regions we're going to need for this
> > + * reservation. The maximum number of regions we're going to need is
> > + * (t - f) / 2 + 1, which corresponds to a region with alternating
> > + * reserved and unreserved pages.
> > + */
> > + *out_regions_needed = (t - f) / 2 + 1;
> > +
> > spin_lock(&resv->lock);
> > -retry_locked:
> > - resv->adds_in_progress++;
> > +
> > + resv->adds_in_progress += *out_regions_needed;
> >
> > /*
> > * Check for sufficient descriptors in the cache to accommodate
> > * the number of in progress add operations.
> > */
> > - if (resv->adds_in_progress > resv->region_cache_count) {
> > - struct file_region *trg;
> > -
> > - VM_BUG_ON(resv->adds_in_progress - resv->region_cache_count > 1);
> > + while (resv->region_cache_count < resv->adds_in_progress) {
> > /* Must drop lock to allocate a new descriptor. */
> > - resv->adds_in_progress--;
> > spin_unlock(&resv->lock);
> > -
> > trg = kmalloc(sizeof(*trg), GFP_KERNEL);
> > if (!trg)
> > return -ENOMEM;
> > @@ -393,9 +395,9 @@ static long region_chg(struct resv_map *resv, long f, long t)
> > spin_lock(&resv->lock);
> > list_add(&trg->link, &resv->region_cache);
> > resv->region_cache_count++;
> > - goto retry_locked;
> > }
>
>
> I know that I suggested allocating the worst case number of entries, but this
> is going to be too much of a hit for existing hugetlbfs users. It is not
> uncommon for DBs to have a shared areas in excess of 1TB mapped by hugetlbfs.
> With this new scheme, the above while loop will allocate over a half million
> file region entries and end up only using one.
>
> I think we need to step back and come up with a different approach. Let me
> give it some more thought before throwing out ideas that may waste more of
> your time. Sorry.
No problem at all. The other more reasonable option is to have it such
that region_add allocates its own cache entries if it needs to, and
the effect of that is that region_add may fail, so the callers must
handle that possibility. Doesn't seem too difficult to handle.
> --
> Mike Kravetz
