Message-ID: <SJ0PR11MB5678EE059638F55D0A664751C9752@SJ0PR11MB5678.namprd11.prod.outlook.com>
Date: Sun, 29 Sep 2024 21:24:04 +0000
From: "Sridhar, Kanchana P" <kanchana.p.sridhar@...el.com>
To: Yosry Ahmed <yosryahmed@...gle.com>
CC: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linux-mm@...ck.org" <linux-mm@...ck.org>, "hannes@...xchg.org"
	<hannes@...xchg.org>, "nphamcs@...il.com" <nphamcs@...il.com>,
	"chengming.zhou@...ux.dev" <chengming.zhou@...ux.dev>,
	"usamaarif642@...il.com" <usamaarif642@...il.com>, "shakeel.butt@...ux.dev"
	<shakeel.butt@...ux.dev>, "ryan.roberts@....com" <ryan.roberts@....com>,
	"Huang, Ying" <ying.huang@...el.com>, "21cnbao@...il.com"
	<21cnbao@...il.com>, "akpm@...ux-foundation.org" <akpm@...ux-foundation.org>,
	"Zou, Nanhai" <nanhai.zou@...el.com>, "Feghali, Wajdi K"
	<wajdi.k.feghali@...el.com>, "Gopal, Vinodh" <vinodh.gopal@...el.com>,
	"Sridhar, Kanchana P" <kanchana.p.sridhar@...el.com>
Subject: RE: [PATCH v8 6/8] mm: zswap: Support large folios in zswap_store().

> -----Original Message-----
> From: Yosry Ahmed <yosryahmed@...gle.com>
> Sent: Friday, September 27, 2024 8:42 PM
> To: Sridhar, Kanchana P <kanchana.p.sridhar@...el.com>
> Cc: linux-kernel@...r.kernel.org; linux-mm@...ck.org;
> hannes@...xchg.org; nphamcs@...il.com; chengming.zhou@...ux.dev;
> usamaarif642@...il.com; shakeel.butt@...ux.dev; ryan.roberts@....com;
> Huang, Ying <ying.huang@...el.com>; 21cnbao@...il.com; akpm@...ux-
> foundation.org; Zou, Nanhai <nanhai.zou@...el.com>; Feghali, Wajdi K
> <wajdi.k.feghali@...el.com>; Gopal, Vinodh <vinodh.gopal@...el.com>
> Subject: Re: [PATCH v8 6/8] mm: zswap: Support large folios in zswap_store().
> 
> On Fri, Sep 27, 2024 at 7:16 PM Kanchana P Sridhar
> <kanchana.p.sridhar@...el.com> wrote:
> >
> > zswap_store() will store large folios by compressing them page by page.
> >
> > This patch provides a sequential implementation of storing a large folio
> > in zswap_store() by iterating through each page in the folio to compress
> > and store it in the zswap zpool.
> >
> > Towards this goal, zswap_compress() is modified to take a page instead
> > of a folio as input.
> 
> It is already modified at this point, it's not part of this patch.
> 
> >
> > Each page's swap offset is stored as a separate zswap entry.
> 
> I think "Each page is compressed individually and stored as a separate
> zswap entry" is clearer. We do not store the offsets.
> 
> >
> > We check the cgroup zswap limit and the zpool utilization against
> > the zswap max/accept_threshold limits once, at the beginning of
> > zswap_store. We also obtain a percpu_ref_tryget() reference to the
> > current zswap_pool at the start of zswap_store to prevent it from
> > being deleted while the folio is being stored.
> 
> This can be clarified further:
> 
> We check the global and per-cgroup limits once at the beginning, and
> only check that the limit is not reached yet. This is racy and
> inaccurate, but it should be sufficient for now. We also obtain
> initial references to the relevant objcg and pool to guarantee that
> subsequent references can be acquired by zswap_store_page().
> 
> >
> > If these one-time checks pass, we compress the sub-pages of the folio,
> > while maintaining a running count of compressed bytes for all the folio's
> > sub-pages. If all pages are successfully compressed and stored, we do the
> > cgroup zswap charging with the total compressed bytes, and batch update
> > the zswap_stored_pages atomic/zswpout event stats with folio_nr_pages()
> > once, before returning from zswap_store.
> 
> Please consistently use "page" instead of "sub-page", and add "()" to
> the end of function names.
> 
> >
> > The patch adds a new zswap_pool_get() function that is called in the
> > sub-page level "zswap_store_page()" function.
> 
> "that is called by zswap_store_page(), which handles compressing and
> storing each page in the folio."
> 
> >
> > If an error is encountered during the store of any page in the folio,
> > all pages in that folio currently stored in zswap will be invalidated.
> > Thus, a folio is either entirely stored in zswap, or entirely not stored
> > in zswap.
> >
> > This patch forms the basis for building compress batching of pages in a
> > large folio in zswap_store by compressing up to say, 8 pages of the folio
> > in parallel in hardware using the Intel In-Memory Analytics Accelerator
> > (Intel IAA).
> 
> This patch also has its own merits, it batches some operations, and
> more importantly enables swapping out large folios to zswap without
> splitting them.

Thanks for the detailed commit log review comments! Sure, I will incorporate
all of these in v9. I will also modify the latency improvement calculation convention
as you've suggested in another comment on the cover letter. Will re-gather all
the "after" data once v9 is ready.

> 
> >
> > This change reuses and adapts the functionality in Ryan Roberts' RFC
> > patch [1]:
> >
> >   "[RFC,v1] mm: zswap: Store large folios without splitting"
> >
> >   [1] https://lore.kernel.org/linux-mm/20231019110543.3284654-1-
> ryan.roberts@....com/T/#u
> >
> > Also, addressed some of the RFC comments from the discussion in [1].
> >
> > Co-developed-by: Ryan Roberts
> > Signed-off-by:
> > Signed-off-by: Kanchana P Sridhar <kanchana.p.sridhar@...el.com>
> > ---
> >  mm/zswap.c | 227 ++++++++++++++++++++++++++++++++++++++----------
> -----
> >  1 file changed, 165 insertions(+), 62 deletions(-)
> >
> > diff --git a/mm/zswap.c b/mm/zswap.c
> > index 43e4e216db41..b8395ddf7b7c 100644
> > --- a/mm/zswap.c
> > +++ b/mm/zswap.c
> > @@ -411,6 +411,16 @@ static int __must_check zswap_pool_tryget(struct
> zswap_pool *pool)
> >         return percpu_ref_tryget(&pool->ref);
> >  }
> >
> > +/*
> > + * Note: zswap_pool_get() should only be called after zswap_pool_tryget()
> > + * returns success. zswap_pool_tryget() returns success only if the "pool" is
> > + * non-NULL and the "&pool->ref" is non-0.
> 
> Just use the same statement used by css_get():
> 
> "The caller must already have a reference."
> 
> > + */
> > +static void zswap_pool_get(struct zswap_pool *pool)
> > +{
> > +       percpu_ref_get(&pool->ref);
> > +}
> > +
> >  static void zswap_pool_put(struct zswap_pool *pool)
> >  {
> >         percpu_ref_put(&pool->ref);
> > @@ -1402,38 +1412,35 @@ static void shrink_worker(struct work_struct
> *w)
> >  /*********************************
> >  * main API
> >  **********************************/
> > -bool zswap_store(struct folio *folio)
> > +
> > +/*
> > + * Stores the page at specified "index" in a folio.
> > + *
> > + * @folio: The folio to store in zswap.
> > + * @index: Index into the page in the folio that this function will store.
> > + * @objcg: The folio's objcg.
> > + * @pool:  The zswap_pool to store the compressed data for the page.
> > + *         The caller should have obtained a reference to a valid
> > + *         zswap_pool by calling zswap_pool_tryget(), to pass as this
> > + *         argument.
> > + * @compressed_bytes: The compressed entry->length value is added
> > + *                    to this, so that the caller can get the total
> > + *                    compressed lengths of all sub-pages in a folio.
> > + */
> > +static bool zswap_store_page(struct folio *folio, long index,
> > +                            struct obj_cgroup *objcg,
> > +                            struct zswap_pool *pool,
> > +                            size_t *compressed_bytes)
> 
> Would it be clearer to just return the compressed size, and return -1
> upon failure?
> 
> >  {
> > +       struct page *page = folio_page(folio, index);
> >         swp_entry_t swp = folio->swap;
> > -       pgoff_t offset = swp_offset(swp);
> >         struct xarray *tree = swap_zswap_tree(swp);
> > +       pgoff_t offset = swp_offset(swp) + index;
> >         struct zswap_entry *entry, *old;
> > -       struct obj_cgroup *objcg = NULL;
> > -       struct mem_cgroup *memcg = NULL;
> > -
> > -       VM_WARN_ON_ONCE(!folio_test_locked(folio));
> > -       VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
> > +       int type = swp_type(swp);
> 
> Why do we need type? We use it when initializing entry->swpentry to
> reconstruct the swp_entry_t we already have.
> 
> >
> > -       /* Large folios aren't supported */
> > -       if (folio_test_large(folio))
> > -               return false;
> > -
> > -       if (!zswap_enabled)
> > -               goto check_old;
> > -
> > -       /* Check cgroup limits */
> > -       objcg = get_obj_cgroup_from_folio(folio);
> > -       if (objcg && !obj_cgroup_may_zswap(objcg)) {
> > -               memcg = get_mem_cgroup_from_objcg(objcg);
> > -               if (shrink_memcg(memcg)) {
> > -                       mem_cgroup_put(memcg);
> > -                       goto reject;
> > -               }
> > -               mem_cgroup_put(memcg);
> > -       }
> > -
> > -       if (zswap_check_limits())
> > -               goto reject;
> > +       if (objcg)
> > +               obj_cgroup_get(objcg);
> >
> >         /* allocate entry */
> >         entry = zswap_entry_cache_alloc(GFP_KERNEL, folio_nid(folio));
> > @@ -1442,24 +1449,21 @@ bool zswap_store(struct folio *folio)
> >                 goto reject;
> >         }
> >
> > -       /* if entry is successfully added, it keeps the reference */
> > -       entry->pool = zswap_pool_current_get();
> > -       if (!entry->pool)
> > -               goto freepage;
> > +       /*
> > +        * We get here only after the call to zswap_pool_tryget() in the
> > +        * caller, zswap_store(), has returned success. Hence it is safe
> > +        * to call zswap_pool_get().
> > +        *
> > +        * if entry is successfully added, it keeps the reference
> > +        */
> > +       zswap_pool_get(pool);
> 
> I would move this right above obj_cgroup_get() and have a single
> concise comment for both, e.g.:
> 
> /* zswap_store() already holds a ref on 'pool' and 'objcg' */
> 
> >
> > -       if (objcg) {
> > -               memcg = get_mem_cgroup_from_objcg(objcg);
> > -               if (memcg_list_lru_alloc(memcg, &zswap_list_lru, GFP_KERNEL)) {
> > -                       mem_cgroup_put(memcg);
> > -                       goto put_pool;
> > -               }
> > -               mem_cgroup_put(memcg);
> > -       }
> > +       entry->pool = pool;
> >
> > -       if (!zswap_compress(&folio->page, entry))
> > +       if (!zswap_compress(page, entry))
> >                 goto put_pool;
> >
> > -       entry->swpentry = swp;
> > +       entry->swpentry = swp_entry(type, offset);
> >         entry->objcg = objcg;
> >         entry->referenced = true;
> >
> > @@ -1480,11 +1484,6 @@ bool zswap_store(struct folio *folio)
> >         if (old)
> >                 zswap_entry_free(old);
> >
> > -       if (objcg) {
> > -               obj_cgroup_charge_zswap(objcg, entry->length);
> > -               count_objcg_event(objcg, ZSWPOUT);
> > -       }
> > -
> >         /*
> >          * We finish initializing the entry while it's already in xarray.
> >          * This is safe because:
> > @@ -1496,36 +1495,140 @@ bool zswap_store(struct folio *folio)
> >          *    an incoherent entry.
> >          */
> >         if (entry->length) {
> > +               *compressed_bytes += entry->length;
> >                 INIT_LIST_HEAD(&entry->lru);
> >                 zswap_lru_add(&zswap_list_lru, entry);
> >         }
> >
> > -       /* update stats */
> > -       atomic_long_inc(&zswap_stored_pages);
> > -       count_vm_event(ZSWPOUT);
> > -
> >         return true;
> >
> >  store_failed:
> >         zpool_free(entry->pool->zpool, entry->handle);
> >  put_pool:
> > -       zswap_pool_put(entry->pool);
> > -freepage:
> > +       zswap_pool_put(pool);
> >         zswap_entry_cache_free(entry);
> >  reject:
> 
> Please rename this to put_objcg, we are no longer "rejecting" here.

I have reworked getting the refs on pool and objcg within zswap_store_page(),
as per your suggestion. With these modifications, the "reject" label is only for
the case where we return "false" from zswap_store_page(), so it seems more
appropriate? Anyway, let me know once v9 is posted, and I can make any
necessary changes.

> 
> >         obj_cgroup_put(objcg);
> > -       if (zswap_pool_reached_full)
> > -               queue_work(shrink_wq, &zswap_shrink_work);
> > -check_old:
> > +       return false;
> > +}
> > +
> > +bool zswap_store(struct folio *folio)
> > +{
> > +       long nr_pages = folio_nr_pages(folio);
> > +       swp_entry_t swp = folio->swap;
> > +       struct xarray *tree = swap_zswap_tree(swp);
> > +       pgoff_t offset = swp_offset(swp);
> > +       struct obj_cgroup *objcg = NULL;
> > +       struct mem_cgroup *memcg = NULL;
> > +       struct zswap_pool *pool;
> > +       size_t compressed_bytes = 0;
> 
> Why size_t? entry->length is int.
> 
> > +       bool ret = false;
> > +       long index;
> > +
> > +       VM_WARN_ON_ONCE(!folio_test_locked(folio));
> > +       VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
> > +
> > +       if (!zswap_enabled)
> > +               goto reject;
> > +
> >         /*
> > -        * If the zswap store fails or zswap is disabled, we must invalidate the
> > -        * possibly stale entry which was previously stored at this offset.
> > -        * Otherwise, writeback could overwrite the new data in the swapfile.
> > +        * Check cgroup zswap limits:
> > +        *
> > +        * The cgroup zswap limit check is done once at the beginning of
> > +        * zswap_store(). The cgroup charging is done once, at the end
> > +        * of a successful folio store. What this means is, if the cgroup
> > +        * was within the zswap_max limit at the beginning of a large folio
> > +        * store, it could go over the limit by at most (HPAGE_PMD_NR - 1)
> > +        * pages.
> 
> Add ".. due to this store", as it can be much more over the limit when
> stores are racing. Also, this comment can be slightly generalized and
> the comment above zswap_check_limits() omitted.
> 
> >          */
> > -       entry = xa_erase(tree, offset);
> > -       if (entry)
> > -               zswap_entry_free(entry);
> > -       return false;
> > +       objcg = get_obj_cgroup_from_folio(folio);
> > +       if (objcg && !obj_cgroup_may_zswap(objcg)) {
> > +               memcg = get_mem_cgroup_from_objcg(objcg);
> > +               if (shrink_memcg(memcg)) {
> > +                       mem_cgroup_put(memcg);
> > +                       goto put_objcg;
> > +               }
> > +               mem_cgroup_put(memcg);
> > +       }
> > +
> > +       /*
> > +        * Check zpool utilization against zswap limits:
> > +        *
> > +        * The zswap zpool utilization is also checked against the limits
> > +        * just once, at the start of zswap_store(). If the check passes,
> > +        * any breaches of the limits set by zswap_max_pages() or
> > +        * zswap_accept_thr_pages() that may happen while storing this
> > +        * folio, will only be detected during the next call to
> > +        * zswap_store() by any process.
> > +        */
> > +       if (zswap_check_limits())
> > +               goto put_objcg;
> > +
> > +       pool = zswap_pool_current_get();
> > +       if (!pool)
> > +               goto put_objcg;
> > +
> > +       if (objcg) {
> > +               memcg = get_mem_cgroup_from_objcg(objcg);
> > +               if (memcg_list_lru_alloc(memcg, &zswap_list_lru, GFP_KERNEL)) {
> > +                       mem_cgroup_put(memcg);
> > +                       goto put_pool;
> > +               }
> > +               mem_cgroup_put(memcg);
> > +       }
> > +
> > +       /*
> > +        * Store each page of the folio as a separate entry. If we fail to
> > +        * store a page, unwind by deleting all the pages for this folio
> > +        * currently in zswap.
> > +        */
> > +       for (index = 0; index < nr_pages; ++index) {
> > +               if (!zswap_store_page(folio, index, objcg, pool,
> &compressed_bytes))
> > +                       goto put_pool;
> > +       }
> > +
> > +       /*
> > +        * All pages in the folio have been successfully stored.
> > +        * Batch update the cgroup zswap charging, zswap_stored_page
> atomic,
> > +        * and ZSWPOUT event stats.
> > +        */
> 
> This comment seems unnecessary.
> 
> > +       if (objcg) {
> > +               obj_cgroup_charge_zswap(objcg, compressed_bytes);
> > +               count_objcg_events(objcg, ZSWPOUT, nr_pages);
> > +       }
> > +
> > +       /* update stats */
> 
> Please delete this comment too.
> 
> > +       atomic_long_add(nr_pages, &zswap_stored_pages);
> > +       count_vm_events(ZSWPOUT, nr_pages);
> > +
> > +       ret = true;
> > +
> > +put_pool:
> > +       zswap_pool_put(pool);
> > +put_objcg:
> > +       obj_cgroup_put(objcg);
> > +reject:
> > +       /*
> > +        * If the zswap store fails or zswap is disabled, we must invalidate
> > +        * the possibly stale entries which were previously stored at the
> > +        * offsets corresponding to each page of the folio. Otherwise,
> > +        * writeback could overwrite the new data in the swapfile.
> > +        */
> > +       if (!ret) {
> > +               struct zswap_entry *entry;
> > +               long i;
> > +
> > +               for (i = 0; i < nr_pages; ++i) {
> 
> Just reuse 'index' here.
> 
> > +                       entry = xa_erase(tree, offset + i);
> > +                       if (entry)
> > +                               zswap_entry_free(entry);
> 
> I think we need a comment in zswap_store_page() that we shouldn't have
> any possibility of failure after the entry is added in the tree.
> Otherwise we may duplicate the cleanup work here (e.g. putting refs).
> This is already the case in zswap_store(), but it's now more subtle
> and easier to break when the code is split and we have two cleanup
> paths.

I have incorporated all these comments, thanks!

> 
> > +               }
> > +
> > +               if (zswap_pool_reached_full)
> > +                       queue_work(shrink_wq, &zswap_shrink_work);
> 
> We are now doing this check even when zswap is disabled. I think this
> is a problem.
> 
> If zswap gets disabled while 'zswap_pool_reached_full' is true, we
> will queue 'zswap_shrink_work' every time zswap_store() is called in
> the swapout path, but 'zswap_pool_reached_full' will never become
> false again become zswap_check_limits() will never be called.
> 
> I think we need to maintain two separate "reject" and "check_old"
> labels, and avoid this check when zswap is disabled.

Thanks Yosry, great catch! I have fixed this in the v9 that I am working on.

Thanks again,
Kanchana

> 
> > +       }
> > +
> > +       return ret;
> >  }
> >
> >  bool zswap_load(struct folio *folio)
> > --
> > 2.27.0
> >

Powered by blists - more mailing lists