]> git.proxmox.com Git - mirror_zfs.git/commitdiff
Fix free memory calculation on v3.14+
authorchrisrd <chris@onthe.net.au>
Fri, 23 Feb 2018 16:50:06 +0000 (03:50 +1100)
committerBrian Behlendorf <behlendorf1@llnl.gov>
Fri, 23 Feb 2018 16:50:06 +0000 (08:50 -0800)
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 <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Chris Dunlop <chris@onthe.net.au>
Closes #7170

config/kernel-global_page_state.m4 [new file with mode: 0644]
config/kernel-vm_node_stat.m4 [deleted file]
config/kernel.m4
include/linux/Makefile.am
include/linux/page_compat.h [new file with mode: 0644]
module/zfs/arc.c
scripts/Makefile.am
scripts/enum-extract.pl [new file with mode: 0755]

diff --git a/config/kernel-global_page_state.m4 b/config/kernel-global_page_state.m4
new file mode 100644 (file)
index 0000000..f4a4001
--- /dev/null
@@ -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 (file)
index 5dcd9d8..0000000
+++ /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)
-       ])
-])
index 9b11dc17980662c64bbdf2c973b73a9771c9c390..a1bbbc2de70ce71b022d8392af82291c8bc4574e 100644 (file)
@@ -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"], [
index 9bb0b3493e5d96e1c65c467943d751720783d553..89c2689f6f0c0b2006f89bf900b5d9bad6af7808 100644 (file)
@@ -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 (file)
index 0000000..95acb7d
--- /dev/null
@@ -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 */
index a78a029c21d1b3f76a1f6461d3430e0d0fca6ae8..c6ff6171c1612c6d37178ff30694fbf5c3ae0796 100644 (file)
 #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)
index 65d7d6662e09b174f24b48f3cb59124629f6bb37..57b9cfbf631db642950378e932f956a84c0b5fb9 100644 (file)
@@ -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 (executable)
index 0000000..5112cc8
--- /dev/null
@@ -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;