From: Chuanhua Han <hanchuanhua@oppo.com>
Currently, we have mTHP features, but unfortunately, without support for
large folio swap-ins, once these large folios are swapped out, they are
lost because mTHP swap is a one-way process. The lack of mTHP swap-in
functionality prevents mTHP from being used on devices like Android that
heavily rely on swap.
This patch introduces mTHP swap-in support. It starts from sync devices
such as zRAM. This is probably the simplest and most common use case,
benefiting billions of Android phones and similar devices with minimal
implementation cost. In this straightforward scenario, large folios are
always exclusive, eliminating the need to handle complex rmap and
swapcache issues.
It offers several benefits:
1. Enables bidirectional mTHP swapping, allowing retrieval of mTHP after
swap-out and swap-in. Large folios in the buddy system are also
preserved as much as possible, rather than being fragmented due
to swap-in.
2. Eliminates fragmentation in swap slots and supports successful
THP_SWPOUT.
w/o this patch (Refer to the data from Chris's and Kairui's latest
swap allocator optimization while running ./thp_swap_allocator_test
w/o "-a" option [1]):
./thp_swap_allocator_test
Iteration 1: swpout inc: 233, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 2: swpout inc: 131, swpout fallback inc: 101, Fallback percentage: 43.53%
Iteration 3: swpout inc: 71, swpout fallback inc: 155, Fallback percentage: 68.58%
Iteration 4: swpout inc: 55, swpout fallback inc: 168, Fallback percentage: 75.34%
Iteration 5: swpout inc: 35, swpout fallback inc: 191, Fallback percentage: 84.51%
Iteration 6: swpout inc: 25, swpout fallback inc: 199, Fallback percentage: 88.84%
Iteration 7: swpout inc: 23, swpout fallback inc: 205, Fallback percentage: 89.91%
Iteration 8: swpout inc: 9, swpout fallback inc: 219, Fallback percentage: 96.05%
Iteration 9: swpout inc: 13, swpout fallback inc: 213, Fallback percentage: 94.25%
Iteration 10: swpout inc: 12, swpout fallback inc: 216, Fallback percentage: 94.74%
Iteration 11: swpout inc: 16, swpout fallback inc: 213, Fallback percentage: 93.01%
Iteration 12: swpout inc: 10, swpout fallback inc: 210, Fallback percentage: 95.45%
Iteration 13: swpout inc: 16, swpout fallback inc: 212, Fallback percentage: 92.98%
Iteration 14: swpout inc: 12, swpout fallback inc: 212, Fallback percentage: 94.64%
Iteration 15: swpout inc: 15, swpout fallback inc: 211, Fallback percentage: 93.36%
Iteration 16: swpout inc: 15, swpout fallback inc: 200, Fallback percentage: 93.02%
Iteration 17: swpout inc: 9, swpout fallback inc: 220, Fallback percentage: 96.07%
w/ this patch (always 0%):
Iteration 1: swpout inc: 948, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 2: swpout inc: 953, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 3: swpout inc: 950, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 4: swpout inc: 952, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 5: swpout inc: 950, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 6: swpout inc: 950, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 7: swpout inc: 947, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 8: swpout inc: 950, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 9: swpout inc: 950, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 10: swpout inc: 945, swpout fallback inc: 0, Fallback percentage: 0.00%
Iteration 11: swpout inc: 947, swpout fallback inc: 0, Fallback percentage: 0.00%
...
3. With both mTHP swap-out and swap-in supported, we offer the option to enable
zsmalloc compression/decompression with larger granularity[2]. The upcoming
optimization in zsmalloc will significantly increase swap speed and improve
compression efficiency. Tested by running 100 iterations of swapping 100MiB
of anon memory, the swap speed improved dramatically:
time consumption of swapin(ms) time consumption of swapout(ms)
lz4 4k 45274 90540
lz4 64k 22942 55667
zstdn 4k 85035 186585
zstdn 64k 46558 118533
The compression ratio also improved, as evaluated with 1 GiB of data:
granularity orig_data_size compr_data_size
4KiB-zstd 1048576000 246876055
64KiB-zstd 1048576000 199763892
Without mTHP swap-in, the potential optimizations in zsmalloc cannot be
realized.
4. Even mTHP swap-in itself can reduce swap-in page faults by a factor
of nr_pages. Swapping in content filled with the same data 0x11, w/o
and w/ the patch for five rounds (Since the content is the same,
decompression will be very fast. This primarily assesses the impact of
reduced page faults):
swp in bandwidth(bytes/ms) w/o w/
round1 624152 1127501
round2 631672 1127501
round3 620459 1139756
round4 606113 1139756
round5 624152 1152281
avg 621310 1137359 +83%
5. With both mTHP swap-out and swap-in supported, we offer the option to enable
hardware accelerators(Intel IAA) to do parallel decompression with which
Kanchana reported 7X improvement on zRAM read latency[3].
[1] https://lore.kernel.org/all/20240730-swap-allocator-v5-0-cb9c148b9297@kernel.org/
[2] https://lore.kernel.org/all/20240327214816.31191-1-21cnbao@gmail.com/
[3] https://lore.kernel.org/all/cover.1714581792.git.andre.glover@linux.intel.com/
Signed-off-by: Chuanhua Han <hanchuanhua@oppo.com>
Co-developed-by: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gao Xiang <xiang@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Cc: Usama Arif <usamaarif642@gmail.com>
Cc: Kanchana P Sridhar <kanchana.p.sridhar@intel.com>
---
mm/memory.c | 261 ++++++++++++++++++++++++++++++++++++++++++++++------
1 file changed, 234 insertions(+), 27 deletions(-)
diff --git a/mm/memory.c b/mm/memory.c
index cdf03b39a92c..d35dd8d99c8a 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3985,6 +3985,194 @@ static vm_fault_t handle_pte_marker(struct vm_fault *vmf)
return VM_FAULT_SIGBUS;
}
+static struct folio *__alloc_swap_folio(struct vm_fault *vmf)
+{
+ struct vm_area_struct *vma = vmf->vma;
+ struct folio *folio;
+ swp_entry_t entry;
+
+ folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma,
+ vmf->address, false);
+ if (!folio)
+ return NULL;
+
+ entry = pte_to_swp_entry(vmf->orig_pte);
+ if (mem_cgroup_swapin_charge_folio(folio, vma->vm_mm,
+ GFP_KERNEL, entry)) {
+ folio_put(folio);
+ return NULL;
+ }
+
+ return folio;
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static inline int non_swapcache_batch(swp_entry_t entry, int max_nr)
+{
+ struct swap_info_struct *si = swp_swap_info(entry);
+ pgoff_t offset = swp_offset(entry);
+ int i;
+
+ /*
+ * While allocating a large folio and doing swap_read_folio, which is
+ * the case the being faulted pte doesn't have swapcache. We need to
+ * ensure all PTEs have no cache as well, otherwise, we might go to
+ * swap devices while the content is in swapcache.
+ */
+ for (i = 0; i < max_nr; i++) {
+ if ((si->swap_map[offset + i] & SWAP_HAS_CACHE))
+ return i;
+ }
+
+ return i;
+}
+
+/*
+ * Check if the PTEs within a range are contiguous swap entries
+ * and have consistent swapcache, zeromap.
+ */
+static bool can_swapin_thp(struct vm_fault *vmf, pte_t *ptep, int nr_pages)
+{
+ unsigned long addr;
+ swp_entry_t entry;
+ int idx;
+ pte_t pte;
+
+ addr = ALIGN_DOWN(vmf->address, nr_pages * PAGE_SIZE);
+ idx = (vmf->address - addr) / PAGE_SIZE;
+ pte = ptep_get(ptep);
+
+ if (!pte_same(pte, pte_move_swp_offset(vmf->orig_pte, -idx)))
+ return false;
+ entry = pte_to_swp_entry(pte);
+ if (swap_pte_batch(ptep, nr_pages, pte) != nr_pages)
+ return false;
+
+ /*
+ * swap_read_folio() can't handle the case a large folio is hybridly
+ * from different backends. And they are likely corner cases. Similar
+ * things might be added once zswap support large folios.
+ */
+ if (unlikely(swap_zeromap_batch(entry, nr_pages, NULL) != nr_pages))
+ return false;
+ if (unlikely(non_swapcache_batch(entry, nr_pages) != nr_pages))
+ return false;
+
+ return true;
+}
+
+static inline unsigned long thp_swap_suitable_orders(pgoff_t swp_offset,
+ unsigned long addr,
+ unsigned long orders)
+{
+ int order, nr;
+
+ order = highest_order(orders);
+
+ /*
+ * To swap in a THP with nr pages, we require that its first swap_offset
+ * is aligned with that number, as it was when the THP was swapped out.
+ * This helps filter out most invalid entries.
+ */
+ while (orders) {
+ nr = 1 << order;
+ if ((addr >> PAGE_SHIFT) % nr == swp_offset % nr)
+ break;
+ order = next_order(&orders, order);
+ }
+
+ return orders;
+}
+
+static struct folio *alloc_swap_folio(struct vm_fault *vmf)
+{
+ struct vm_area_struct *vma = vmf->vma;
+ unsigned long orders;
+ struct folio *folio;
+ unsigned long addr;
+ swp_entry_t entry;
+ spinlock_t *ptl;
+ pte_t *pte;
+ gfp_t gfp;
+ int order;
+
+ /*
+ * If uffd is active for the vma we need per-page fault fidelity to
+ * maintain the uffd semantics.
+ */
+ if (unlikely(userfaultfd_armed(vma)))
+ goto fallback;
+
+ /*
+ * A large swapped out folio could be partially or fully in zswap. We
+ * lack handling for such cases, so fallback to swapping in order-0
+ * folio.
+ */
+ if (!zswap_never_enabled())
+ goto fallback;
+
+ entry = pte_to_swp_entry(vmf->orig_pte);
+ /*
+ * Get a list of all the (large) orders below PMD_ORDER that are enabled
+ * and suitable for swapping THP.
+ */
+ orders = thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_IN_PF | TVA_ENFORCE_SYSFS, BIT(PMD_ORDER) - 1);
+ orders = thp_vma_suitable_orders(vma, vmf->address, orders);
+ orders = thp_swap_suitable_orders(swp_offset(entry),
+ vmf->address, orders);
+
+ if (!orders)
+ goto fallback;
+
+ pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
+ vmf->address & PMD_MASK, &ptl);
+ if (unlikely(!pte))
+ goto fallback;
+
+ /*
+ * For do_swap_page, find the highest order where the aligned range is
+ * completely swap entries with contiguous swap offsets.
+ */
+ order = highest_order(orders);
+ while (orders) {
+ addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order);
+ if (can_swapin_thp(vmf, pte + pte_index(addr), 1 << order))
+ break;
+ order = next_order(&orders, order);
+ }
+
+ pte_unmap_unlock(pte, ptl);
+
+ /* Try allocating the highest of the remaining orders. */
+ gfp = vma_thp_gfp_mask(vma);
+ while (orders) {
+ addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order);
+ folio = vma_alloc_folio(gfp, order, vma, addr, true);
+ if (folio) {
+ if (!mem_cgroup_swapin_charge_folio(folio, vma->vm_mm,
+ gfp, entry))
+ return folio;
+ folio_put(folio);
+ }
+ order = next_order(&orders, order);
+ }
+
+fallback:
+ return __alloc_swap_folio(vmf);
+}
+#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
+static inline bool can_swapin_thp(struct vm_fault *vmf, pte_t *ptep, int nr_pages)
+{
+ return false;
+}
+
+static struct folio *alloc_swap_folio(struct vm_fault *vmf)
+{
+ return __alloc_swap_folio(vmf);
+}
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
/*
* We enter with non-exclusive mmap_lock (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
@@ -4073,34 +4261,34 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
if (!folio) {
if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
__swap_count(entry) == 1) {
- /*
- * Prevent parallel swapin from proceeding with
- * the cache flag. Otherwise, another thread may
- * finish swapin first, free the entry, and swapout
- * reusing the same entry. It's undetectable as
- * pte_same() returns true due to entry reuse.
- */
- if (swapcache_prepare(entry, 1)) {
- /* Relax a bit to prevent rapid repeated page faults */
- schedule_timeout_uninterruptible(1);
- goto out;
- }
- need_clear_cache = true;
-
/* skip swapcache */
- folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
- vma, vmf->address, false);
+ folio = alloc_swap_folio(vmf);
if (folio) {
__folio_set_locked(folio);
__folio_set_swapbacked(folio);
- if (mem_cgroup_swapin_charge_folio(folio,
- vma->vm_mm, GFP_KERNEL,
- entry)) {
- ret = VM_FAULT_OOM;
+ nr_pages = folio_nr_pages(folio);
+ if (folio_test_large(folio))
+ entry.val = ALIGN_DOWN(entry.val, nr_pages);
+ /*
+ * Prevent parallel swapin from proceeding with
+ * the cache flag. Otherwise, another thread
+ * may finish swapin first, free the entry, and
+ * swapout reusing the same entry. It's
+ * undetectable as pte_same() returns true due
+ * to entry reuse.
+ */
+ if (swapcache_prepare(entry, nr_pages)) {
+ /*
+ * Relax a bit to prevent rapid
+ * repeated page faults.
+ */
+ schedule_timeout_uninterruptible(1);
goto out_page;
}
- mem_cgroup_swapin_uncharge_swap(entry, 1);
+ need_clear_cache = true;
+
+ mem_cgroup_swapin_uncharge_swap(entry, nr_pages);
shadow = get_shadow_from_swap_cache(entry);
if (shadow)
@@ -4206,6 +4394,24 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
goto out_nomap;
}
+ /* allocated large folios for SWP_SYNCHRONOUS_IO */
+ if (folio_test_large(folio) && !folio_test_swapcache(folio)) {
+ unsigned long nr = folio_nr_pages(folio);
+ unsigned long folio_start = ALIGN_DOWN(vmf->address, nr * PAGE_SIZE);
+ unsigned long idx = (vmf->address - folio_start) / PAGE_SIZE;
+ pte_t *folio_ptep = vmf->pte - idx;
+ pte_t folio_pte = ptep_get(folio_ptep);
+
+ if (!pte_same(folio_pte, pte_move_swp_offset(vmf->orig_pte, -idx)) ||
+ swap_pte_batch(folio_ptep, nr, folio_pte) != nr)
+ goto out_nomap;
+
+ page_idx = idx;
+ address = folio_start;
+ ptep = folio_ptep;
+ goto check_folio;
+ }
+
nr_pages = 1;
page_idx = 0;
address = vmf->address;
@@ -4337,11 +4543,12 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
folio_add_lru_vma(folio, vma);
} else if (!folio_test_anon(folio)) {
/*
- * We currently only expect small !anon folios, which are either
- * fully exclusive or fully shared. If we ever get large folios
- * here, we have to be careful.
+ * We currently only expect small !anon folios which are either
+ * fully exclusive or fully shared, or new allocated large
+ * folios which are fully exclusive. If we ever get large
+ * folios within swapcache here, we have to be careful.
*/
- VM_WARN_ON_ONCE(folio_test_large(folio));
+ VM_WARN_ON_ONCE(folio_test_large(folio) && folio_test_swapcache(folio));
VM_WARN_ON_FOLIO(!folio_test_locked(folio), folio);
folio_add_new_anon_rmap(folio, vma, address, rmap_flags);
} else {
@@ -4384,7 +4591,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
out:
/* Clear the swap cache pin for direct swapin after PTL unlock */
if (need_clear_cache)
- swapcache_clear(si, entry, 1);
+ swapcache_clear(si, entry, nr_pages);
if (si)
put_swap_device(si);
return ret;
@@ -4400,7 +4607,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
folio_put(swapcache);
}
if (need_clear_cache)
- swapcache_clear(si, entry, 1);
+ swapcache_clear(si, entry, nr_pages);
if (si)
put_swap_device(si);
return ret;
--
2.34.1© 2016 - 2024 Red Hat, Inc.