diff options
author | chrisrd <[email protected]> | 2018-02-24 03:50:06 +1100 |
---|---|---|
committer | Brian Behlendorf <[email protected]> | 2018-02-23 08:50:06 -0800 |
commit | e9a77290081b578144e9a911ad734f67274df82f (patch) | |
tree | 29036268b80f1a2695867d079e2a4facf916a616 | |
parent | 7088545d0166aa05b2c783f18aa821d95a1f023d (diff) |
Fix free memory calculation on v3.14+
Provide infrastructure to auto-configure to enum and API changes in the
global page stats used for our free memory calculations.
arc_free_memory has been broken since an API change in Linux v3.14:
2016-07-28 v4.8 599d0c95 mm, vmscan: move LRU lists to node
2016-07-28 v4.8 75ef7184 mm, vmstat: add infrastructure for per-node
vmstats
These commits moved some of global_page_state() into
global_node_page_state(). The API change was particularly egregious as,
instead of breaking the old code, it silently did the wrong thing and we
continued using global_page_state() where we should have been using
global_node_page_state(), thus indexing into the wrong array via
NR_SLAB_RECLAIMABLE et al.
There have been further API changes along the way:
2017-07-06 v4.13 385386cf mm: vmstat: move slab statistics from zone to
node counters
2017-09-06 v4.14 c41f012a mm: rename global_page_state to
global_zone_page_state
...and various (incomplete, as it turns out) attempts to accomodate
these changes in ZoL:
2017-08-24 2209e409 Linux 4.8+ compatibility fix for vm stats
2017-09-16 787acae0 Linux 3.14 compat: IO acct, global_page_state, etc
2017-09-19 661907e6 Linux 4.14 compat: IO acct, global_page_state, etc
The config infrastructure provided here resolves these issues going back
to the original API change in v3.14 and is robust against further Linux
changes in this area.
Reviewed-by: Giuseppe Di Natale <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: George Melikov <[email protected]>
Signed-off-by: Chris Dunlop <[email protected]>
Closes #7170
-rw-r--r-- | config/kernel-global_page_state.m4 | 109 | ||||
-rw-r--r-- | config/kernel-vm_node_stat.m4 | 22 | ||||
-rw-r--r-- | config/kernel.m4 | 2 | ||||
-rw-r--r-- | include/linux/Makefile.am | 3 | ||||
-rw-r--r-- | include/linux/page_compat.h | 78 | ||||
-rw-r--r-- | module/zfs/arc.c | 23 | ||||
-rw-r--r-- | scripts/Makefile.am | 1 | ||||
-rwxr-xr-x | scripts/enum-extract.pl | 58 |
8 files changed, 255 insertions, 41 deletions
diff --git a/config/kernel-global_page_state.m4 b/config/kernel-global_page_state.m4 new file mode 100644 index 000000000..f4a40011f --- /dev/null +++ b/config/kernel-global_page_state.m4 @@ -0,0 +1,109 @@ +dnl # +dnl # 4.8 API change +dnl # +dnl # 75ef71840539 mm, vmstat: add infrastructure for per-node vmstats +dnl # 599d0c954f91 mm, vmscan: move LRU lists to node +dnl # +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_NODE_PAGE_STATE], [ + AC_MSG_CHECKING([whether global_node_page_state() exists]) + ZFS_LINUX_TRY_COMPILE([ + #include <linux/mm.h> + #include <linux/vmstat.h> + ],[ + (void) global_node_page_state(0); + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(ZFS_GLOBAL_NODE_PAGE_STATE, 1, [global_node_page_state() exists]) + ],[ + AC_MSG_RESULT(no) + ]) +]) + +dnl # +dnl # 4.14 API change +dnl # +dnl # c41f012ade0b mm: rename global_page_state to global_zone_page_state +dnl # +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_ZONE_PAGE_STATE], [ + AC_MSG_CHECKING([whether global_zone_page_state() exists]) + ZFS_LINUX_TRY_COMPILE([ + #include <linux/mm.h> + #include <linux/vmstat.h> + ],[ + (void) global_zone_page_state(0); + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(ZFS_GLOBAL_ZONE_PAGE_STATE, 1, [global_zone_page_state() exists]) + ],[ + AC_MSG_RESULT(no) + ]) +]) + +dnl # +dnl # Create a define and autoconf variable for an enum member +dnl # +AC_DEFUN([ZFS_AC_KERNEL_ENUM_MEMBER], [ + AC_MSG_CHECKING([whether enum $2 contains $1]) + AS_IF([AC_TRY_COMMAND("${srcdir}/scripts/enum-extract.pl" "$2" "$3" | egrep -qx $1)],[ + AC_MSG_RESULT([yes]) + AC_DEFINE(m4_join([_], [ZFS_ENUM], m4_toupper($2), $1), 1, [enum $2 contains $1]) + m4_join([_], [ZFS_ENUM], m4_toupper($2), $1)=1 + ],[ + AC_MSG_RESULT([no]) + ]) +]) + +dnl # +dnl # Sanity check helpers +dnl # +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_ERROR],[ + AC_MSG_RESULT(no) + AC_MSG_RESULT([$1 in either node_stat_item or zone_stat_item: $2]) + AC_MSG_RESULT([configure needs updating, see: config/kernel-global_page_state.m4]) + AC_MSG_FAILURE([SHUT 'ER DOWN CLANCY, SHE'S PUMPIN' MUD!]) +]) + +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_CHECK], [ + enum_check_a="m4_join([_], [$ZFS_ENUM_NODE_STAT_ITEM], $1)" + enum_check_b="m4_join([_], [$ZFS_ENUM_ZONE_STAT_ITEM], $1)" + AS_IF([test -n "$enum_check_a" -a -n "$enum_check_b"],[ + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_ERROR([$1], [DUPLICATE]) + ]) + AS_IF([test -z "$enum_check_a" -a -z "$enum_check_b"],[ + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_ERROR([$1], [NOT FOUND]) + ]) +]) + +dnl # +dnl # Ensure the config tests are finding one and only one of each enum of interest +dnl # +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_ZONE_PAGE_STATE_SANITY], [ + AC_MSG_CHECKING([global_page_state enums are sane]) + + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_CHECK([NR_FILE_PAGES]) + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_CHECK([NR_INACTIVE_ANON]) + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_CHECK([NR_INACTIVE_FILE]) + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE_ENUM_CHECK([NR_SLAB_RECLAIMABLE]) + + AC_MSG_RESULT(yes) +]) + +dnl # +dnl # enum members in which we're interested +dnl # +AC_DEFUN([ZFS_AC_KERNEL_GLOBAL_PAGE_STATE], [ + ZFS_AC_KERNEL_GLOBAL_NODE_PAGE_STATE + ZFS_AC_KERNEL_GLOBAL_ZONE_PAGE_STATE + + ZFS_AC_KERNEL_ENUM_MEMBER([NR_FILE_PAGES], [node_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_INACTIVE_ANON], [node_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_INACTIVE_FILE], [node_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_SLAB_RECLAIMABLE], [node_stat_item], [$LINUX/include/linux/mmzone.h]) + + ZFS_AC_KERNEL_ENUM_MEMBER([NR_FILE_PAGES], [zone_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_INACTIVE_ANON], [zone_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_INACTIVE_FILE], [zone_stat_item], [$LINUX/include/linux/mmzone.h]) + ZFS_AC_KERNEL_ENUM_MEMBER([NR_SLAB_RECLAIMABLE], [zone_stat_item], [$LINUX/include/linux/mmzone.h]) + + ZFS_AC_KERNEL_GLOBAL_ZONE_PAGE_STATE_SANITY +]) diff --git a/config/kernel-vm_node_stat.m4 b/config/kernel-vm_node_stat.m4 deleted file mode 100644 index 5dcd9d827..000000000 --- a/config/kernel-vm_node_stat.m4 +++ /dev/null @@ -1,22 +0,0 @@ -dnl # -dnl # 4.8 API change -dnl # kernel vm counters change -dnl # -AC_DEFUN([ZFS_AC_KERNEL_VM_NODE_STAT], [ - AC_MSG_CHECKING([whether to use vm_node_stat based fn's]) - ZFS_LINUX_TRY_COMPILE([ - #include <linux/mm.h> - #include <linux/vmstat.h> - ],[ - int a __attribute__ ((unused)) = NR_VM_NODE_STAT_ITEMS; - long x __attribute__ ((unused)) = - atomic_long_read(&vm_node_stat[0]); - (void) global_node_page_state(0); - ],[ - AC_MSG_RESULT(yes) - AC_DEFINE(ZFS_GLOBAL_NODE_PAGE_STATE, 1, - [using global_node_page_state()]) - ],[ - AC_MSG_RESULT(no) - ]) -]) diff --git a/config/kernel.m4 b/config/kernel.m4 index 9b11dc179..a1bbbc2de 100644 --- a/config/kernel.m4 +++ b/config/kernel.m4 @@ -123,7 +123,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_KERNEL_RENAME_WANTS_FLAGS ZFS_AC_KERNEL_HAVE_GENERIC_SETXATTR ZFS_AC_KERNEL_CURRENT_TIME - ZFS_AC_KERNEL_VM_NODE_STAT + ZFS_AC_KERNEL_GLOBAL_PAGE_STATE ZFS_AC_KERNEL_ACL_HAS_REFCOUNT AS_IF([test "$LINUX_OBJ" != "$LINUX"], [ diff --git a/include/linux/Makefile.am b/include/linux/Makefile.am index 9bb0b3493..89c2689f6 100644 --- a/include/linux/Makefile.am +++ b/include/linux/Makefile.am @@ -9,7 +9,8 @@ KERNEL_H = \ $(top_srcdir)/include/linux/kmap_compat.h \ $(top_srcdir)/include/linux/simd_x86.h \ $(top_srcdir)/include/linux/simd_aarch64.h \ - $(top_srcdir)/include/linux/mod_compat.h + $(top_srcdir)/include/linux/mod_compat.h \ + $(top_srcdir)/include/linux/page_compat.h USER_H = diff --git a/include/linux/page_compat.h b/include/linux/page_compat.h new file mode 100644 index 000000000..95acb7d53 --- /dev/null +++ b/include/linux/page_compat.h @@ -0,0 +1,78 @@ +#ifndef _ZFS_PAGE_COMPAT_H +#define _ZFS_PAGE_COMPAT_H + +/* + * We have various enum members moving between two separate enum types, + * and accessed by different functions at various times. Centralise the + * insanity. + * + * < v4.8: all enums in zone_stat_item, via global_page_state() + * v4.8: some enums moved to node_stat_item, global_node_page_state() introduced + * v4.13: some enums moved from zone_stat_item to node_state_item + * v4.14: global_page_state() rename to global_zone_page_state() + * + * The defines used here are created by config/kernel-global_page_state.m4 + */ + +/* + * Create our own accessor functions to follow the Linux API changes + */ +#if defined(ZFS_GLOBAL_ZONE_PAGE_STATE) + +/* global_zone_page_state() introduced */ +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_FILE_PAGES) +#define nr_file_pages() global_node_page_state(NR_FILE_PAGES) +#else +#define nr_file_pages() global_zone_page_state(NR_FILE_PAGES) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_INACTIVE_ANON) +#define nr_inactive_anon_pages() global_node_page_state(NR_INACTIVE_ANON) +#else +#define nr_inactive_anon_pages() global_zone_page_state(NR_INACTIVE_ANON) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_INACTIVE_FILE) +#define nr_inactive_file_pages() global_node_page_state(NR_INACTIVE_FILE) +#else +#define nr_inactive_file_pages() global_zone_page_state(NR_INACTIVE_FILE) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_SLAB_RECLAIMABLE) +#define nr_slab_reclaimable_pages() global_node_page_state(NR_SLAB_RECLAIMABLE) +#else +#define nr_slab_reclaimable_pages() global_zone_page_state(NR_SLAB_RECLAIMABLE) +#endif + +#elif defined(ZFS_GLOBAL_NODE_PAGE_STATE) + +/* global_node_page_state() introduced */ +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_FILE_PAGES) +#define nr_file_pages() global_node_page_state(NR_FILE_PAGES) +#else +#define nr_file_pages() global_page_state(NR_FILE_PAGES) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_INACTIVE_ANON) +#define nr_inactive_anon_pages() global_node_page_state(NR_INACTIVE_ANON) +#else +#define nr_inactive_anon_pages() global_page_state(NR_INACTIVE_ANON) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_INACTIVE_FILE) +#define nr_inactive_file_pages() global_node_page_state(NR_INACTIVE_FILE) +#else +#define nr_inactive_file_pages() global_page_state(NR_INACTIVE_FILE) +#endif +#if defined(ZFS_ENUM_NODE_STAT_ITEM_NR_SLAB_RECLAIMABLE) +#define nr_slab_reclaimable_pages() global_node_page_state(NR_SLAB_RECLAIMABLE) +#else +#define nr_slab_reclaimable_pages() global_page_state(NR_SLAB_RECLAIMABLE) +#endif + +#else + +/* global_page_state() only */ +#define nr_file_pages() global_page_state(NR_FILE_PAGES) +#define nr_inactive_anon_pages() global_page_state(NR_INACTIVE_ANON) +#define nr_inactive_file_pages() global_page_state(NR_INACTIVE_FILE) +#define nr_slab_reclaimable_pages() global_page_state(NR_SLAB_RECLAIMABLE) + +#endif /* ZFS_GLOBAL_ZONE_PAGE_STATE */ + +#endif /* _ZFS_PAGE_COMPAT_H */ diff --git a/module/zfs/arc.c b/module/zfs/arc.c index a78a029c2..c6ff6171c 100644 --- a/module/zfs/arc.c +++ b/module/zfs/arc.c @@ -297,6 +297,7 @@ #include <sys/fs/swapnode.h> #include <sys/zpl.h> #include <linux/mm_compat.h> +#include <linux/page_compat.h> #endif #include <sys/callb.h> #include <sys/kstat.h> @@ -4699,17 +4700,11 @@ arc_free_memory(void) si_meminfo(&si); return (ptob(si.freeram - si.freehigh)); #else -#ifdef ZFS_GLOBAL_NODE_PAGE_STATE return (ptob(nr_free_pages() + - global_node_page_state(NR_INACTIVE_FILE) + - global_node_page_state(NR_INACTIVE_ANON) + - global_node_page_state(NR_SLAB_RECLAIMABLE))); -#else - return (ptob(nr_free_pages() + - global_page_state(NR_INACTIVE_FILE) + - global_page_state(NR_INACTIVE_ANON) + - global_page_state(NR_SLAB_RECLAIMABLE))); -#endif /* ZFS_GLOBAL_NODE_PAGE_STATE */ + nr_inactive_file_pages() + + nr_inactive_anon_pages() + + nr_slab_reclaimable_pages())); + #endif /* CONFIG_HIGHMEM */ #else return (spa_get_random(arc_all_memory() * 20 / 100)); @@ -5121,13 +5116,7 @@ arc_evictable_memory(void) * Scale reported evictable memory in proportion to page cache, cap * at specified min/max. */ -#ifdef ZFS_GLOBAL_NODE_PAGE_STATE - uint64_t min = (ptob(global_node_page_state(NR_FILE_PAGES)) / 100) * - zfs_arc_pc_percent; -#else - uint64_t min = (ptob(global_page_state(NR_FILE_PAGES)) / 100) * - zfs_arc_pc_percent; -#endif + uint64_t min = (ptob(nr_file_pages()) / 100) * zfs_arc_pc_percent; min = MAX(arc_c_min, MIN(arc_c_max, min)); if (arc_dirty >= min) diff --git a/scripts/Makefile.am b/scripts/Makefile.am index 65d7d6662..57b9cfbf6 100644 --- a/scripts/Makefile.am +++ b/scripts/Makefile.am @@ -12,6 +12,7 @@ EXTRA_DIST = \ commitcheck.sh \ dkms.mkconf \ dkms.postbuild \ + enum-extract.pl \ kmodtool \ paxcheck.sh \ zfs2zol-patch.sed \ diff --git a/scripts/enum-extract.pl b/scripts/enum-extract.pl new file mode 100755 index 000000000..5112cc807 --- /dev/null +++ b/scripts/enum-extract.pl @@ -0,0 +1,58 @@ +#!/usr/bin/perl -w + +my $usage = <<EOT; +usage: config-enum enum [file ...] + +Returns the elements from an enum declaration. + +"Best effort": we're not building an entire C interpreter here! +EOT + +use warnings; +use strict; +use Getopt::Std; + +my %opts; + +if (!getopts("", \%opts) || @ARGV < 1) { + print $usage; + exit 2; +} + +my $enum = shift; + +my $in_enum = 0; + +while (<>) { + # comments + s/\/\*.*\*\///; + if (m/\/\*/) { + while ($_ .= <>) { + last if s/\/\*.*\*\///s; + } + } + + # preprocessor stuff + next if /^#/; + + # find our enum + $in_enum = 1 if s/^\s*enum\s+${enum}(?:\s|$)//; + next unless $in_enum; + + # remove explicit values + s/\s*=[^,]+,/,/g; + + # extract each identifier + while (m/\b([a-z_][a-z0-9_]*)\b/ig) { + print $1, "\n"; + } + + # + # don't exit: there may be multiple versions of the same enum, e.g. + # inside different #ifdef blocks. Let's explicitly return all of + # them and let external tooling deal with it. + # + $in_enum = 0 if m/}\s*;/; +} + +exit 0; |