[PATCH v2 3/3] mm: swap: treat SWP_SYNCHRONOUS_IO devices as synchronous in readahead

[Alexandre Ghiti]

A readahead neighbour on an SWP_SYNCHRONOUS_IO device (zram) has no
asynchronous I/O to overlap either, exactly like a zswap-resident one.
Extend swap_entry_synchronous() to take the swap device and also return
true for such an entry, so it is deferred to the second pass too. A
same-type neighbour shares the target's device, which cannot be
SWP_SYNCHRONOUS_IO here (do_swap_page() routes synchronous-device
targets through swapin_sync()), so @si is only non-NULL for cross-device
neighbours, which is exactly where it matters.

This helps a mixed zram+disk workload when the faulting target is on
disk: that fault now overlaps its zram neighbours' reads behind its own
disk I/O, exactly as patch 1 does for zswap neighbours.

Results [vm-scalability mix-rw-seq, 1G zram (SWP_SYNCHRONOUS_IO, high
prio) + NVMe disk swap (~76% of swap-in from disk), zram=zstd, zswap
off, 4G memcg / 8G region, unthrottled NVMe, n=3]. With zswap off
patches 1-2 are inert, so the delta isolates this patch:

  metric                  mm-new    +patch    delta
  swap-in tput (KB/s)    171226    182046     +6.3%
  major faults           431904    432657     +0.2%
  swap-in pswpin (pages) 1778579   1778473     ~0%
  swap_ra (pages)        1349299   1348382     -0.1%

Signed-off-by: Alexandre Ghiti <alex@xxxxxxxx>
---
 include/linux/swap.h |  4 ++--
 mm/swap_state.c      | 10 +++++-----
 mm/swapfile.c        |  6 ++++--
 3 files changed, 11 insertions(+), 9 deletions(-)

diff --git a/include/linux/swap.h b/include/linux/swap.h
index 6bb35c093d77..e6e0d07a4f19 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -441,7 +441,7 @@ extern unsigned int count_swap_pages(int, int);
 extern sector_t swapdev_block(int, pgoff_t);
 extern int __swap_count(swp_entry_t entry);
 extern bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t entry);
-extern bool swap_entry_synchronous(swp_entry_t entry);
+extern bool swap_entry_synchronous(struct swap_info_struct *si, swp_entry_t entry);
 extern int swp_swapcount(swp_entry_t entry);
 struct backing_dev_info;
 extern struct swap_info_struct *get_swap_device(swp_entry_t entry);
@@ -520,7 +520,7 @@ static inline bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t e
 	return false;
 }
 
-static inline bool swap_entry_synchronous(swp_entry_t entry)
+static inline bool swap_entry_synchronous(struct swap_info_struct *si, swp_entry_t entry)
 {
 	return false;
 }
diff --git a/mm/swap_state.c b/mm/swap_state.c
index 23137714d7cb..12fe86218557 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -829,7 +829,7 @@ struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 	blk_start_plug(&plug);
 	for (offset = start_offset; offset <= end_offset ; offset++) {
 		ra_entry = swp_entry(swp_type(entry), offset);
-		if (swap_entry_synchronous(ra_entry)) {
+		if (swap_entry_synchronous(si, ra_entry)) {
 			has_sync = true;
 			continue;
 		}
@@ -844,13 +844,13 @@ struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 	blk_finish_plug(&plug);
 	swap_read_unplug(splug);
 
-	if (!has_sync || swap_entry_synchronous(entry))
+	if (!has_sync || swap_entry_synchronous(si, entry))
 		goto drain;
 
 	/* Overlap the in flight IOs with the synchronous neighbours reads */
 	for (offset = start_offset; offset <= end_offset; offset++) {
 		ra_entry = swp_entry(swp_type(entry), offset);
-		if (!swap_entry_synchronous(ra_entry))
+		if (!swap_entry_synchronous(si, ra_entry))
 			continue;
 
 		folio = swap_cache_read_folio(ra_entry, gfp_mask, mpol, ilx,
@@ -968,7 +968,7 @@ static struct folio *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask,
 				continue;
 		}
 
-		if (swap_entry_synchronous(entry)) {
+		if (swap_entry_synchronous(si, entry)) {
 			sync_entries[nr_sync] = entry;
 			sync_ilx[nr_sync] = ilx;
 			nr_sync++;
@@ -990,7 +990,7 @@ static struct folio *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask,
 	blk_finish_plug(&plug);
 	swap_read_unplug(splug);
 
-	if (!nr_sync || swap_entry_synchronous(targ_entry))
+	if (!nr_sync || swap_entry_synchronous(NULL, targ_entry))
 		goto drain;
 
 	/* Overlap the in flight IOs with the synchronous neighbours reads */
diff --git a/mm/swapfile.c b/mm/swapfile.c
index ee2bbfdfc988..e8f88156b767 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1985,11 +1985,13 @@ bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t entry)
 
 /**
  * swap_entry_synchronous - Check if @entry is served by a synchronous source.
+ * @si: the swap device of @entry, or NULL if it is not needed/held.
  * @entry: the swap entry.
  */
-bool swap_entry_synchronous(swp_entry_t entry)
+bool swap_entry_synchronous(struct swap_info_struct *si, swp_entry_t entry)
 {
-	return zswap_is_present(entry, 1);
+	return zswap_is_present(entry, 1) ||
+	       (si && data_race(si->flags & SWP_SYNCHRONOUS_IO));
 }
 
 /*
-- 
2.53.0-Meta