mm: optimize shadow entries removal
Subsystems (2)
v2 cover mm: optimize shadow entries removal
Some of our production workloads which processes a large amount of data spends considerable amount of CPUs on truncation and invalidation of large sized files (100s of GiBs of size). Tracing the operations showed that most of the time is in shadow entries removal. This patch series optimizes the truncation and invalidation operations.
Shakeel Butt (2):
mm: optimize truncation of shadow entries
mm: optimize invalidation of shadow entries
mm/truncate.c | 99 +++++++++++++++++++++++---------------------------- 1 file changed, 44 insertions(+), 55 deletions(-)
Signature · 1 line
2.43.5
PATCH v2 1/2 mm: optimize truncation of shadow entries
The kernel truncates the page cache in batches of PAGEVEC_SIZE. For each batch, it traverses the page cache tree and collects the entries (folio and shadow entries) in the struct folio_batch. For the shadow entries present in the folio_batch, it has to traverse the page cache tree for each individual entry to remove them. This patch optimize this by removing them in a single tree traversal.
On large machines in our production which run workloads manipulating large amount of data, we have observed that a large amount of CPUs are spent on truncation of very large files (100s of GiBs file sizes). More specifically most of time was spent on shadow entries cleanup, so optimizing the shadow entries cleanup, even a little bit, has good impact.
To evaluate the changes, we created 200GiB file on a fuse fs and in a memcg. We created the shadow entries by triggering reclaim through memory.reclaim in that specific memcg and measure the simple truncation operation.
# time truncate -s 0 file
time (sec)
Without 5.164 +- 0.059
With-patch 4.21 +- 0.066 (18.47% decrease)
---
Changes since v1:
- Added a comment on the assumption of indices array (Johannes)
mm/truncate.c | 53 +++++++++++++++++++++++++-------------------------- 1 file changed, 26 insertions(+), 27 deletions(-)
mm/truncate.c +26 -27
@@ -68,54 +68,53 @@ static void clear_shadow_entries(struct address_space *mapping,* Unconditionally remove exceptional entries. Usually called from truncate * path. Note that the folio_batch may be altered by this function by removing * exceptional entries similar to what folio_batch_remove_exceptionals() does. +* Please note that indices[] has entries in ascending order as guaranteed by +* either find_get_entries() or find_lock_entries(). */ static void truncate_folio_batch_exceptionals (struct address_space *mapping,struct folio_batch *fbatch,pgoff_t *indices) { +XA_STATE (xas, &mapping->i_pages, indices[0 ]); +int nr =folio_batch_count (fbatch); +struct folio *folio;int i, j; -bool dax;/* Handled by shmem itself */ if (shmem_mapping (mapping))return ; -for (j =0 ; j <folio_batch_count (fbatch); j++) +for (j =0 ; j < nr; j++)if (xa_is_value (fbatch->folios[j]))break ; -if (j ==folio_batch_count (fbatch)) +if (j == nr)return ; - dax =dax_mapping (mapping); -if (!dax) { -spin_lock (&mapping->host->i_lock); -xa_lock_irq (&mapping->i_pages); +if (dax_mapping (mapping)) { +for (i = j; i < nr; i++) { +if (xa_is_value (fbatch->folios[i])) +dax_delete_mapping_entry (mapping, indices[i]); + } + goto out; } -for (i = j; i <folio_batch_count (fbatch); i++) { -struct folio *folio = fbatch->folios[i]; -pgoff_t index = indices[i]; - -if (!xa_is_value (folio)) { - fbatch->folios[j++] = folio; -continue ; - } +xas_set (&xas, indices[j]); +xas_set_update (&xas, workingset_update_node); -if (unlikely (dax)) { -dax_delete_mapping_entry (mapping, index); -continue ; - } +spin_lock (&mapping->host->i_lock); +xas_lock_irq (&xas); -__clear_shadow_entry (mapping, index, folio); +xas_for_each (&xas, folio, indices[nr-1 ]) { +if (xa_is_value (folio)) +xas_store (&xas,NULL ); } -if (!dax) { -xa_unlock_irq (&mapping->i_pages); -if (mapping_shrinkable (mapping)) -inode_add_lru (mapping->host); -spin_unlock (&mapping->host->i_lock); - } - fbatch->nr = j; +xas_unlock_irq (&xas); +if (mapping_shrinkable (mapping)) +inode_add_lru (mapping->host); +spin_unlock (&mapping->host->i_lock); +out: +folio_batch_remove_exceptionals (fbatch); }/**
Signature · 1 line
2.43.5
PATCH v2 2/2 mm: optimize invalidation of shadow entries
In reply to Shakeel ButtThe kernel invalidates the page cache in batches of PAGEVEC_SIZE. For each batch, it traverses the page cache tree and collects the entries (folio and shadow entries) in the struct folio_batch. For the shadow entries present in the folio_batch, it has to traverse the page cache tree for each individual entry to remove them. This patch optimize this by removing them in a single tree traversal.
To evaluate the changes, we created 200GiB file on a fuse fs and in a memcg. We created the shadow entries by triggering reclaim through memory.reclaim in that specific memcg and measure the simple fadvise(DONTNEED) operation.
# time xfs_io -c 'fadvise -d 0 ${file_size}' file time (sec)
Without 5.12 +- 0.061
With-patch 4.19 +- 0.086 (18.16% decrease)
---
Changes since v1:
- N/A
mm/truncate.c | 46 ++++++++++++++++++---------------------------- 1 file changed, 18 insertions(+), 28 deletions(-)
mm/truncate.c +18 -28
@@ -23,42 +23,28 @@#include <linux/rmap.h> #include "internal.h" -/* -* Regular page slots are stabilized by the page lock even without the tree -* itself locked. These unlocked entries need verification under the tree -* lock. -*/ -static inline void __clear_shadow_entry (struct address_space *mapping, -pgoff_t index,void *entry) -{ -XA_STATE (xas, &mapping->i_pages, index); - -xas_set_update (&xas, workingset_update_node); -if (xas_load (&xas) != entry) -return ; -xas_store (&xas,NULL ); -} -static void clear_shadow_entries (struct address_space *mapping, -struct folio_batch *fbatch,pgoff_t *indices) +unsigned long start,unsigned long max) { -int i; +XA_STATE (xas, &mapping->i_pages, start); +struct folio *folio;/* Handled by shmem itself, or for DAX we do nothing. */ if (shmem_mapping (mapping) ||dax_mapping (mapping))return ; -spin_lock (&mapping->host->i_lock); -xa_lock_irq (&mapping->i_pages); +xas_set_update (&xas, workingset_update_node); -for (i =0 ; i <folio_batch_count (fbatch); i++) { -struct folio *folio = fbatch->folios[i]; +spin_lock (&mapping->host->i_lock); +xas_lock_irq (&xas); +/* Clear all shadow entries from start to max */ +xas_for_each (&xas, folio, max) {if (xa_is_value (folio)) -__clear_shadow_entry (mapping, indices[i], folio); +xas_store (&xas,NULL ); } -xa_unlock_irq (&mapping->i_pages); +xas_unlock_irq (&xas);if (mapping_shrinkable (mapping))inode_add_lru (mapping->host);spin_unlock (&mapping->host->i_lock);
@@ -481,7 +467,9 @@ unsigned long mapping_try_invalidate(struct address_space *mapping,folio_batch_init (&fbatch);while (find_lock_entries (mapping, &index, end, &fbatch, indices)) { -for (i =0 ; i <folio_batch_count (&fbatch); i++) { +int nr =folio_batch_count (&fbatch); + +for (i =0 ; i < nr; i++) {struct folio *folio = fbatch.folios[i];/* We rely upon deletion not changing folio->index */
@@ -508,7 +496,7 @@ unsigned long mapping_try_invalidate(struct address_space *mapping, }if (xa_has_values) -clear_shadow_entries (mapping, &fbatch, indices); +clear_shadow_entries (mapping, indices[0 ], indices[nr-1 ]);folio_batch_remove_exceptionals (&fbatch);folio_batch_release (&fbatch);
@@ -612,7 +600,9 @@ int invalidate_inode_pages2_range(struct address_space *mapping,folio_batch_init (&fbatch); index = start;while (find_get_entries (mapping, &index, end, &fbatch, indices)) { -for (i =0 ; i <folio_batch_count (&fbatch); i++) { +int nr =folio_batch_count (&fbatch); + +for (i =0 ; i < nr; i++) {struct folio *folio = fbatch.folios[i];/* We rely upon deletion not changing folio->index */
@@ -658,7 +648,7 @@ int invalidate_inode_pages2_range(struct address_space *mapping, }if (xa_has_values) -clear_shadow_entries (mapping, &fbatch, indices); +clear_shadow_entries (mapping, indices[0 ], indices[nr-1 ]);folio_batch_remove_exceptionals (&fbatch);folio_batch_release (&fbatch);
Signature · 1 line
2.43.5