ARM64: configs: Make zsmalloc copy rather than mapping pages
VM mapping is ~5x slower than simply copying memory to access cross-page allocations on both our little and big cluster CPUs. zsmalloc copy zsmalloc vmap cpu0 1456637 255254 cpu1 1463270 243098 cpu2 1441192 242113 cpu3 1446096 269415 cpu4 1830798 486465 cpu5 2703679 641879 cpu6 2568903 642474 cpu7 2128928 480830 Little 10kmaps 145 25 (slowdown: 5.8x) Big 10kmaps 231 56 (slowdown: 4.1x)
This commit is contained in:
parent
8b6441f1f9
commit
28bdcbbab1
|
@ -491,7 +491,7 @@ CONFIG_CMA_AREAS=7
|
|||
CONFIG_ZPOOL=y
|
||||
# CONFIG_ZBUD is not set
|
||||
CONFIG_ZSMALLOC=y
|
||||
CONFIG_PGTABLE_MAPPING=y
|
||||
# CONFIG_PGTABLE_MAPPING is not set
|
||||
CONFIG_ZSMALLOC_STAT=y
|
||||
# CONFIG_DIRECT_RECLAIM_FILE_PAGES_ONLY is not set
|
||||
# CONFIG_INCREASE_MAXIMUM_SWAPPINESS is not set
|
||||
|
|
Loading…
Reference in New Issue
Block a user