zsmalloc: reduce size_class memory usage
authorSergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Sat, 7 Nov 2015 00:29:38 +0000 (16:29 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 7 Nov 2015 01:50:42 +0000 (17:50 -0800)
commit6fe5186f0c7c18a8beb6d96c21e2390df7a12375
tree2a8a65b4a48252c2fa9701653e4ef229cd01089a
parent6f0b22760b7d8317569252cc7c36cbed22ebe401
zsmalloc: reduce size_class memory usage

Each `struct size_class' contains `struct zs_size_stat': an array of
NR_ZS_STAT_TYPE `unsigned long'.  For zsmalloc built with no
CONFIG_ZSMALLOC_STAT this results in a waste of `2 * sizeof(unsigned
long)' per-class.

The patch removes unneeded `struct zs_size_stat' members by redefining
NR_ZS_STAT_TYPE (max stat idx in array).

Since both NR_ZS_STAT_TYPE and zs_stat_type are compile time constants,
GCC can eliminate zs_stat_inc()/zs_stat_dec() calls that use zs_stat_type
larger than NR_ZS_STAT_TYPE: CLASS_ALMOST_EMPTY and CLASS_ALMOST_FULL at
the moment.

./scripts/bloat-o-meter mm/zsmalloc.o.old mm/zsmalloc.o.new
add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-39 (-39)
function                                     old     new   delta
fix_fullness_group                            97      94      -3
insert_zspage                                100      86     -14
remove_zspage                                141     119     -22

To summarize:
a) each class now uses less memory
b) we avoid a number of dec/inc stats (a minor optimization,
   but still).

The gain will increase once we introduce additional stats.

A simple IO test.

iozone -t 4 -R -r 32K -s 60M -I +Z
                        patched                 base
"  Initial write "       4145599.06              4127509.75
"        Rewrite "       4146225.94              4223618.50
"           Read "      17157606.00             17211329.50
"        Re-read "      17380428.00             17267650.50
"   Reverse Read "      16742768.00             16162732.75
"    Stride read "      16586245.75             16073934.25
"    Random read "      16349587.50             15799401.75
" Mixed workload "      10344230.62              9775551.50
"   Random write "       4277700.62              4260019.69
"         Pwrite "       4302049.12              4313703.88
"          Pread "       6164463.16              6126536.72
"         Fwrite "       7131195.00              6952586.00
"          Fread "      12682602.25             12619207.50

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/zsmalloc.c