mm: convert scan_control.priority int => byte
Subsystems (3)
PATCH 1/1 mm: convert scan_control.priority int => byte
Reclaim priorities range from 0..12(DEF_PRIORITY). scan_control.priority is a 4 byte int, which is overkill.
Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64 stack overflows are not an issue. But it's inefficient to use 4 bytes for priority.
Use s8 (signed byte) rather than u8 to allow for loops like:
do {
...
} while (--sc.priority >= 0);This reduces sizeof(struct scan_control) from 96 => 88 bytes (x86_64), which saves some stack.
scan_control.priority field order is changed to occupy otherwise unused padding.
--- mm/vmscan.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
mm/vmscan.c +3 -3
@@ -83,9 +83,6 @@ struct scan_control {*/ struct mem_cgroup *target_mem_cgroup; -/* Scan (total_size >> priority) pages at once */ -int priority; -/* The highest zone to isolate pages for reclaim from */ enum zone_type reclaim_idx;
@@ -111,6 +108,9 @@ struct scan_control {/* One of the zones is ready for compaction */ unsigned int compaction_ready:1; +/* Scan (total_size >> priority) pages at once */ +s8 priority; +/* Incremented by the number of inactive pages that were scanned */ unsigned long nr_scanned;
Signature · 1 line
2.17.0.921.gf22659ad46-goog
Quoted 6 lines · Greg Thelen
Reclaim priorities range from 0..12(DEF_PRIORITY). scan_control.priority is a 4 byte int, which is overkill.
Since commit
6538b8ea886e("x86_64: expand kernel stack to 16K")x86_64stack overflows are not an issue. But it's inefficient to use 4 bytes for priority.
If you're looking to shave a few more bytes, allocation order can fit in a u8 too (can't be more than 6 bits, and realistically won't be more than 4 bits). reclaim_idx likewise will fit in a u8, and actually won't be more than 3 bits.
I am sceptical that nr_to_reclaim should really be an unsigned long; I don't think we should be trying to free 4 billion pages in a single call. nr_scanned might be over 4 billion (!) but nr_reclaimed can probably shrink to unsigned int along with nr_to_reclaim.
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Greg Thelen
Reclaim priorities range from 0..12(DEF_PRIORITY). scan_control.priority is a 4 byte int, which is overkill.
Quoted 8 lines
Quoted 3 lines
Since commit
6538b8ea886e("x86_64: expand kernel stack to 16K")x86_64stack overflows are not an issue. But it's inefficient to use 4 bytes for priority.If you're looking to shave a few more bytes, allocation order can fit in a u8 too (can't be more than 6 bits, and realistically won't be more than 4 bits).
reclaim_idxlikewise will fit in a u8, and actually won't be more than 3 bits.
Nod. Good tip. Included in ("[PATCH v2] mm: condense scan_control").
Quoted 4 lines
I am sceptical that
nr_to_reclaimshould really be an unsigned long; I don't think we should be trying to free 4 billion pages in a single call.nr_scannedmight be over 4 billion (!) butnr_reclaimedcan probably shrink to unsigned int along withnr_to_reclaim.
Agreed. For patch simplicity, I'll pass on this for now.
PATCH v2 1/1 mm: condense scan_control
In reply to Matthew WilcoxUse smaller scan_control fields for order, priority, and reclaim_idx. Convert fields from int => s8. All easily fit within a byte:
* allocation order range: 0..MAX_ORDER(64?)
* priority range: 0..12(DEF_PRIORITY)
* reclaim_idx range: 0..6(__MAX_NR_ZONES)
Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64 stack overflows are not an issue. But it's inefficient to use ints.
Use s8 (signed byte) rather than u8 to allow for loops like:
do {
...
} while (--sc.priority >= 0);Add BUILD_BUG_ON to verify that s8 is capable of storing max values.
This reduces sizeof(struct scan_control):
* 96 => 80 bytes (x86_64)
* 68 => 56 bytes (i386)
scan_control structure field order is changed to utilize padding. After this patch there is 1 bit of scan_control padding.
--- mm/vmscan.c | 32 ++++++++++++++++++++------------ 1 file changed, 20 insertions(+), 12 deletions(-)
mm/vmscan.c +20 -12
@@ -65,12 +65,6 @@ struct scan_control {/* How many pages shrink_list() should reclaim */ unsigned long nr_to_reclaim; -/* This context's GFP mask */ -gfp_t gfp_mask; - -/* Allocation order */ -int order; -/* * Nodemask of nodes allowed by the caller. If NULL, all nodes * are scanned.
@@ -83,12 +77,6 @@ struct scan_control {*/ struct mem_cgroup *target_mem_cgroup; -/* Scan (total_size >> priority) pages at once */ -int priority; - -/* The highest zone to isolate pages for reclaim from */ -enum zone_type reclaim_idx; -/* Writepage batching in laptop mode; RECLAIM_WRITE */ unsigned int may_writepage:1;
@@ -111,6 +99,18 @@ struct scan_control {/* One of the zones is ready for compaction */ unsigned int compaction_ready:1; +/* Allocation order */ +s8 order; + +/* Scan (total_size >> priority) pages at once */ +s8 priority; + +/* The highest zone to isolate pages for reclaim from */ +s8 reclaim_idx; + +/* This context's GFP mask */ +gfp_t gfp_mask; +/* Incremented by the number of inactive pages that were scanned */ unsigned long nr_scanned;
@@ -3047,6 +3047,14 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order, .may_swap = 1, }; +/* +* scan_control uses s8 fields for order, priority, and reclaim_idx. +* Confirm they are large enough for max values. +*/ +BUILD_BUG_ON (MAX_ORDER >S8_MAX ); +BUILD_BUG_ON (DEF_PRIORITY >S8_MAX ); +BUILD_BUG_ON (MAX_NR_ZONES >S8_MAX ); +/* * Do not enter reclaim if fatal signal was delivered while throttled. * 1 is returned so that the page allocator does not OOM kill at this
Signature · 1 line
2.17.0.921.gf22659ad46-goog
Re: [PATCH v2] mm: condense scan_control
Quoted 21 lines · Greg Thelen
Use smaller
scan_controlfields for order, priority, andreclaim_idx. Convert fields from int => s8. All easily fit within a byte:
* allocation order range: 0..MAX_ORDER(64?)
* priority range: 0..12(DEF_PRIORITY)
*reclaim_idxrange: 0..6(__MAX_NR_ZONES)Since commit
6538b8ea886e("x86_64: expand kernel stack to 16K")x86_64stack overflows are not an issue. But it's inefficient to use ints.Use s8 (signed byte) rather than u8 to allow for loops like:
do { ... } while (--sc.priority >= 0);Add
BUILD_BUG_ONto verify that s8 is capable of storing max values.This reduces
sizeof(struct scan_control):
* 96 => 80 bytes (x86_64)
* 68 => 56 bytes (i386)
Quoted 5 lines
scan_controlstructure field order is changed to utilize padding. After this patch there is 1 bit ofscan_controlpadding.Signed-off-by Greg Thelen <[email protected]>Suggested-by Matthew Wilcox <[email protected]>
Is there any interest in this? Less stack usage could mean less dcache and dtlb pressure. But I understand if the complexity is distasteful.
Quoted 71 lines
--- mm/vmscan.c | 32 ++++++++++++++++++++------------ 1 file changed, 20 insertions(+), 12 deletions(-)mm/vmscan.c +20
-13@@ -65,12 +65,6 @@ struct scan_control {/* How many pages shrink_list() should reclaim */ unsigned long nr_to_reclaim; -/* This context's GFP mask */ -gfp_t gfp_mask; - -/* Allocation order */ -int order; -/* * Nodemask of nodes allowed by the caller. If NULL, all nodes * are scanned. @@ -83,12 +77,6 @@ struct scan_control {*/ struct mem_cgroup *target_mem_cgroup; -/* Scan (total_size >> priority) pages at once */ -int priority; - -/* The highest zone to isolate pages for reclaim from */ -enum zone_type reclaim_idx; -/* Writepage batching in laptop mode; RECLAIM_WRITE */ unsigned int may_writepage:1;@@ -111,6 +99,18 @@ struct scan_control {/* One of the zones is ready for compaction */ unsigned int compaction_ready:1; +/* Allocation order */ +s8 order; + +/* Scan (total_size >> priority) pages at once */ +s8 priority; + +/* The highest zone to isolate pages for reclaim from */ +s8 reclaim_idx; + +/* This context's GFP mask */ +gfp_t gfp_mask; +/* Incremented by the number of inactive pages that were scanned */ unsigned long nr_scanned;@@ -3047,6 +3047,14 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order, .may_swap = 1, }; +/* +* scan_control uses s8 fields for order, priority, and reclaim_idx. +* Confirm they are large enough for max values. +*/ +BUILD_BUG_ON (MAX_ORDER >S8_MAX ); +BUILD_BUG_ON (DEF_PRIORITY >S8_MAX ); +BUILD_BUG_ON (MAX_NR_ZONES >S8_MAX ); +/* * Do not enter reclaim if fatal signal was delivered while throttled. * 1 is returned so that the page allocator does not OOM kill at this -- 2.17.0.921.gf22659ad46-goog