From 4ab13d3b5c2d0724632dd29c0970ff7df00985f5 Mon Sep 17 00:00:00 2001 From: Brian Behlendorf Date: Thu, 5 Feb 2009 12:26:34 -0800 Subject: [PATCH] Additional Linux VM integration Added support for Solaris swapfs_minfree, and swapfs_reserve tunables. In additional availrmem is now available and return a reasonable value which is reasonably analogous to the Solaris meaning. On linux we return the sun of free and inactive pages since these are all easily reclaimable. All tunables are available in /proc/sys/kernel/spl/vm/* and they may need a little adjusting once we observe the real behavior. Some of the defaults are mapped to similar linux counterparts, others are straight from the OpenSolaris defaults. --- config/spl-build.m4 | 18 +++++++-- configure | 39 ++++++++++++++++++++ configure.ac | 1 + include/sys/vmsystm.h | 20 +++------- module/spl/spl-kmem.c | 86 ++++++++++++++++++++++++++++++++++++++----- module/spl/spl-proc.c | 86 ++++++++++++++++++++++++++++++++++++------- spl_config.h.in | 3 ++ 7 files changed, 214 insertions(+), 39 deletions(-) diff --git a/config/spl-build.m4 b/config/spl-build.m4 index b7aa024..5ee615b 100644 --- a/config/spl-build.m4 +++ b/config/spl-build.m4 @@ -707,7 +707,7 @@ AC_DEFUN([SPL_AC_3ARGS_ON_EACH_CPU], [ ]) dnl # -dnl # Distro specific first_online_pgdat symbol export. +dnl # Distro specific first_online_pgdat() symbol export. dnl # AC_DEFUN([SPL_AC_FIRST_ONLINE_PGDAT], [ SPL_CHECK_SYMBOL_EXPORT( @@ -719,7 +719,7 @@ AC_DEFUN([SPL_AC_FIRST_ONLINE_PGDAT], [ ]) dnl # -dnl # Distro specific next_online_pgdat symbol export. +dnl # Distro specific next_online_pgdat() symbol export. dnl # AC_DEFUN([SPL_AC_NEXT_ONLINE_PGDAT], [ SPL_CHECK_SYMBOL_EXPORT( @@ -731,7 +731,7 @@ AC_DEFUN([SPL_AC_NEXT_ONLINE_PGDAT], [ ]) dnl # -dnl # Distro specific next_zone symbol export. +dnl # Distro specific next_zone() symbol export. dnl # AC_DEFUN([SPL_AC_NEXT_ZONE], [ SPL_CHECK_SYMBOL_EXPORT( @@ -741,3 +741,15 @@ AC_DEFUN([SPL_AC_NEXT_ZONE], [ [next_zone() is available])], []) ]) + +dnl # +dnl # Distro specific get_zone_counts() symbol export. +dnl # +AC_DEFUN([SPL_AC_GET_ZONE_COUNTS], [ + SPL_CHECK_SYMBOL_EXPORT( + [get_zone_counts], + [], + [AC_DEFINE(HAVE_GET_ZONE_COUNTS, 1, + [get_zone_counts() is available])], + []) +]) diff --git a/configure b/configure index 7e59d67..21e222d 100755 --- a/configure +++ b/configure @@ -20598,6 +20598,45 @@ _ACEOF + echo "$as_me:$LINENO: checking whether symbol get_zone_counts is exported" >&5 +echo $ECHO_N "checking whether symbol get_zone_counts is exported... $ECHO_C" >&6 + grep -q -E '[[:space:]]get_zone_counts[[:space:]]' $LINUX/Module.symvers 2>/dev/null + rc=$? + if test $rc -ne 0; then + export=0 + for file in ; do + grep -q -E "EXPORT_SYMBOL.*(get_zone_counts)" "$LINUX/$file" 2>/dev/null + rc=$? + if test $rc -eq 0; then + export=1 + break; + fi + done + if test $export -eq 0; then + echo "$as_me:$LINENO: result: no" >&5 +echo "${ECHO_T}no" >&6 + + else + echo "$as_me:$LINENO: result: yes" >&5 +echo "${ECHO_T}yes" >&6 + +cat >>confdefs.h <<\_ACEOF +#define HAVE_GET_ZONE_COUNTS 1 +_ACEOF + + fi + else + echo "$as_me:$LINENO: result: yes" >&5 +echo "${ECHO_T}yes" >&6 + +cat >>confdefs.h <<\_ACEOF +#define HAVE_GET_ZONE_COUNTS 1 +_ACEOF + + fi + + + TOPDIR=`/bin/pwd` # Add "V=1" to KERNELMAKE_PARAMS to enable verbose module build diff --git a/configure.ac b/configure.ac index bd4f699..259c4e8 100644 --- a/configure.ac +++ b/configure.ac @@ -72,6 +72,7 @@ SPL_AC_3ARGS_ON_EACH_CPU SPL_AC_FIRST_ONLINE_PGDAT SPL_AC_NEXT_ONLINE_PGDAT SPL_AC_NEXT_ZONE +SPL_AC_GET_ZONE_COUNTS TOPDIR=`/bin/pwd` diff --git a/include/sys/vmsystm.h b/include/sys/vmsystm.h index a6e9e7d..75ae8a9 100644 --- a/include/sys/vmsystm.h +++ b/include/sys/vmsystm.h @@ -45,15 +45,14 @@ #define physmem num_physpages #define freemem nr_free_pages() +#define availrmem spl_kmem_availrmem() extern pgcnt_t minfree; /* Sum of zone->pages_min */ extern pgcnt_t desfree; /* Sum of zone->pages_low */ extern pgcnt_t lotsfree; /* Sum of zone->pages_high */ -extern pgcnt_t needfree; /* Always 0 */ -extern pgcnt_t swapfs_minfree; -extern pgcnt_t swapfs_desfree; -extern pgcnt_t swapfs_reserve; -extern pgcnt_t availrmem; +extern pgcnt_t needfree; /* Always 0 unused in new Solaris */ +extern pgcnt_t swapfs_minfree; /* Solaris default value */ +extern pgcnt_t swapfs_reserve; /* Solaris default value */ extern vmem_t *heap_arena; /* primary kernel heap arena */ extern vmem_t *zio_alloc_arena; /* arena for zio caches */ @@ -62,15 +61,8 @@ extern vmem_t *zio_arena; /* arena for allocating zio memory */ #define VMEM_ALLOC 0x01 #define VMEM_FREE 0x02 -static __inline__ size_t -vmem_size(vmem_t *vmp, int typemask) -{ - /* Arena's unsupported */ - ASSERT(vmp == NULL); - ASSERT(typemask & (VMEM_ALLOC | VMEM_FREE)); - - return 0; -} +extern pgcnt_t spl_kmem_availrmem(void); +extern size_t vmem_size(vmem_t *vmp, int typemask); #define xcopyin(from, to, size) copy_from_user(to, from, size) #define xcopyout(from, to, size) copy_to_user(to, from, size) diff --git a/module/spl/spl-kmem.c b/module/spl/spl-kmem.c index c39636e..b5cd9fb 100644 --- a/module/spl/spl-kmem.c +++ b/module/spl/spl-kmem.c @@ -64,18 +64,12 @@ EXPORT_SYMBOL(lotsfree); pgcnt_t needfree = 0; EXPORT_SYMBOL(needfree); -pgcnt_t swapfs_desfree = 0; -EXPORT_SYMBOL(swapfs_desfree); - pgcnt_t swapfs_minfree = 0; EXPORT_SYMBOL(swapfs_minfree); pgcnt_t swapfs_reserve = 0; EXPORT_SYMBOL(swapfs_reserve); -pgcnt_t availrmem = 0; -EXPORT_SYMBOL(availrmem); - vmem_t *heap_arena = NULL; EXPORT_SYMBOL(heap_arena); @@ -86,14 +80,17 @@ vmem_t *zio_arena = NULL; EXPORT_SYMBOL(zio_arena); #ifndef HAVE_FIRST_ONLINE_PGDAT -struct pglist_data *first_online_pgdat(void) +struct pglist_data * +first_online_pgdat(void) { return NODE_DATA(first_online_node); } +EXPORT_SYMBOL(first_online_pgdat); #endif /* HAVE_FIRST_ONLINE_PGDAT */ #ifndef HAVE_NEXT_ONLINE_PGDAT -struct pglist_data *next_online_pgdat(struct pglist_data *pgdat) +struct pglist_data * +next_online_pgdat(struct pglist_data *pgdat) { int nid = next_online_node(pgdat->node_id); @@ -102,10 +99,12 @@ struct pglist_data *next_online_pgdat(struct pglist_data *pgdat) return NODE_DATA(nid); } +EXPORT_SYMBOL(next_online_pgdat); #endif /* HAVE_NEXT_ONLINE_PGDAT */ #ifndef HAVE_NEXT_ZONE -struct zone *next_zone(struct zone *zone) +struct zone * +next_zone(struct zone *zone) { pg_data_t *pgdat = zone->zone_pgdat; @@ -120,8 +119,73 @@ struct zone *next_zone(struct zone *zone) } return zone; } +EXPORT_SYMBOL(next_zone); #endif /* HAVE_NEXT_ZONE */ +#ifndef HAVE_GET_ZONE_COUNTS +void +__get_zone_counts(unsigned long *active, unsigned long *inactive, + unsigned long *free, struct pglist_data *pgdat) +{ + struct zone *zones = pgdat->node_zones; + int i; + + *active = 0; + *inactive = 0; + *free = 0; + for (i = 0; i < MAX_NR_ZONES; i++) { + *active += zones[i].nr_active; + *inactive += zones[i].nr_inactive; + *free += zones[i].free_pages; + } +} + +void +get_zone_counts(unsigned long *active, unsigned long *inactive, + unsigned long *free) +{ + struct pglist_data *pgdat; + + *active = 0; + *inactive = 0; + *free = 0; + for_each_online_pgdat(pgdat) { + unsigned long l, m, n; + __get_zone_counts(&l, &m, &n, pgdat); + *active += l; + *inactive += m; + *free += n; + } +} +EXPORT_SYMBOL(get_zone_counts); +#endif /* HAVE_GET_ZONE_COUNTS */ + +pgcnt_t +spl_kmem_availrmem(void) +{ + unsigned long active; + unsigned long inactive; + unsigned long free; + + get_zone_counts(&active, &inactive, &free); + + /* The amount of easily available memory */ + return free + inactive; +} +EXPORT_SYMBOL(spl_kmem_availrmem); + +size_t +vmem_size(vmem_t *vmp, int typemask) +{ + /* Arena's unsupported */ + ASSERT(vmp == NULL); + ASSERT(typemask & (VMEM_ALLOC | VMEM_FREE)); + + return 0; +} +EXPORT_SYMBOL(vmem_size); + + /* * Memory allocation interfaces and debugging for basic kmem_* * and vmem_* style memory allocation. When DEBUG_KMEM is enable @@ -1707,6 +1771,10 @@ spl_kmem_init_globals(void) desfree += zone->pages_low; lotsfree += zone->pages_high; } + + /* Solaris default values */ + swapfs_minfree = MAX(2*1024*1024 / PAGE_SIZE, physmem / 8); + swapfs_reserve = MIN(4*1024*1024 / PAGE_SIZE, physmem / 16); } int diff --git a/module/spl/spl-proc.c b/module/spl/spl-proc.c index 024118a..1ae1c12 100644 --- a/module/spl/spl-proc.c +++ b/module/spl/spl-proc.c @@ -91,9 +91,10 @@ struct proc_dir_entry *proc_spl_kstat = NULL; #define CTL_VM_LOTSFREE CTL_UNNUMBERED /* Lots of free memory */ #define CTL_VM_NEEDFREE CTL_UNNUMBERED /* Need free memory */ #define CTL_VM_SWAPFS_MINFREE CTL_UNNUMBERED /* Minimum swapfs memory */ -#define CTL_VM_SWAPFS_DESFREE CTL_UNNUMBERED /* Desired swapfs memory */ #define CTL_VM_SWAPFS_RESERVE CTL_UNNUMBERED /* Reserved swapfs memory */ -#define CTL_VM_AVAILRMEM CTL_UNNUMBERED /* Available reserved memory */ +#define CTL_VM_AVAILRMEM CTL_UNNUMBERED /* Easily available memory */ +#define CTL_VM_FREEMEM CTL_UNNUMBERED /* Free memory */ +#define CTL_VM_PHYSMEM CTL_UNNUMBERED /* Total physical memory */ #ifdef DEBUG_KMEM #define CTL_KMEM_KMEMUSED CTL_UNNUMBERED /* Alloc'd kmem bytes */ @@ -145,9 +146,10 @@ enum { CTL_VM_LOTSFREE, /* Lots of free memory threshold */ CTL_VM_NEEDFREE, /* Need free memory deficit */ CTL_VM_SWAPFS_MINFREE, /* Minimum swapfs memory */ - CTL_VM_SWAPFS_DESFREE, /* Desired swapfs memory */ CTL_VM_SWAPFS_RESERVE, /* Reserved swapfs memory */ - CTL_VM_AVAILRMEM, /* Available reserved memory */ + CTL_VM_AVAILRMEM, /* Easily available memory */ + CTL_VM_FREEMEM, /* Free memory */ + CTL_VM_PHYSMEM, /* Total physical memory */ #ifdef DEBUG_KMEM CTL_KMEM_KMEMUSED, /* Alloc'd kmem bytes */ @@ -486,6 +488,58 @@ proc_dohostid(struct ctl_table *table, int write, struct file *filp, RETURN(rc); } +static int +proc_doavailrmem(struct ctl_table *table, int write, struct file *filp, + void __user *buffer, size_t *lenp, loff_t *ppos) +{ + int len, rc = 0; + char str[32]; + ENTRY; + + if (write) { + *ppos += *lenp; + } else { + len = snprintf(str, sizeof(str), "%lu", (unsigned long)availrmem); + if (*ppos >= len) + rc = 0; + else + rc = proc_copyout_string(buffer, *lenp, str + *ppos, "\n"); + + if (rc >= 0) { + *lenp = rc; + *ppos += rc; + } + } + + RETURN(rc); +} + +static int +proc_dofreemem(struct ctl_table *table, int write, struct file *filp, + void __user *buffer, size_t *lenp, loff_t *ppos) +{ + int len, rc = 0; + char str[32]; + ENTRY; + + if (write) { + *ppos += *lenp; + } else { + len = snprintf(str, sizeof(str), "%lu", (unsigned long)freemem); + if (*ppos >= len) + rc = 0; + else + rc = proc_copyout_string(buffer, *lenp, str + *ppos, "\n"); + + if (rc >= 0) { + *lenp = rc; + *ppos += rc; + } + } + + RETURN(rc); +} + #ifdef DEBUG_MUTEX static void mutex_seq_show_headers(struct seq_file *f) @@ -832,14 +886,6 @@ static struct ctl_table spl_vm_table[] = { .mode = 0644, .proc_handler = &proc_dointvec, }, - { - .ctl_name = CTL_VM_SWAPFS_DESFREE, - .procname = "swapfs_desfree", - .data = &swapfs_desfree, - .maxlen = sizeof(int), - .mode = 0644, - .proc_handler = &proc_dointvec, - }, { .ctl_name = CTL_VM_SWAPFS_RESERVE, .procname = "swapfs_reserve", @@ -851,7 +897,21 @@ static struct ctl_table spl_vm_table[] = { { .ctl_name = CTL_VM_AVAILRMEM, .procname = "availrmem", - .data = &availrmem, + .mode = 0444, + .proc_handler = &proc_doavailrmem, + }, + { + .ctl_name = CTL_VM_FREEMEM, + .procname = "freemem", + .data = (void *)2, + .maxlen = sizeof(int), + .mode = 0444, + .proc_handler = &proc_dofreemem, + }, + { + .ctl_name = CTL_VM_PHYSMEM, + .procname = "physmem", + .data = &physmem, .maxlen = sizeof(int), .mode = 0444, .proc_handler = &proc_dointvec, diff --git a/spl_config.h.in b/spl_config.h.in index 2bfb5c4..1b7c8ab 100644 --- a/spl_config.h.in +++ b/spl_config.h.in @@ -51,6 +51,9 @@ /* fls64() is available */ #undef HAVE_FLS64 +/* get_zone_counts() is available */ +#undef HAVE_GET_ZONE_COUNTS + /* init_utsname() is available */ #undef HAVE_INIT_UTSNAME -- 2.39.2