freebsd-dev/lib/libmemstat/memstat_uma.c
Mark Johnston 08cfa56ea3 Extend uma_reclaim() to permit different reclamation targets.
The page daemon periodically invokes uma_reclaim() to reclaim cached
items from each zone when the system is under memory pressure.  This
is important since the size of these caches is unbounded by default.
However it also results in bursts of high latency when allocating from
heavily used zones as threads miss in the per-CPU caches and must
access the keg in order to allocate new items.

With r340405 we maintain an estimate of each zone's usage of its
(per-NUMA domain) cache of full buckets.  Start making use of this
estimate to avoid reclaiming the entire cache when under memory
pressure.  In particular, introduce TRIM, DRAIN and DRAIN_CPU
verbs for uma_reclaim() and uma_zone_reclaim().  When trimming, only
items in excess of the estimate are reclaimed.  Draining a zone
reclaims all of the cached full buckets (the previous behaviour of
uma_reclaim()), and may further drain the per-CPU caches in extreme
cases.

Now, when under memory pressure, the page daemon will trim zones
rather than draining them.  As a result, heavily used zones do not incur
bursts of bucket cache misses following reclamation, but large, unused
caches will be reclaimed as before.

Reviewed by:	jeff
Tested by:	pho (an earlier version)
MFC after:	2 months
Sponsored by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D16667
2019-09-01 22:22:43 +00:00

496 lines
14 KiB
C

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2005-2006 Robert N. M. Watson
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/counter.h>
#include <sys/cpuset.h>
#include <sys/sysctl.h>
#include <vm/uma.h>
#include <vm/uma_int.h>
#include <err.h>
#include <errno.h>
#include <kvm.h>
#include <nlist.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "memstat.h"
#include "memstat_internal.h"
static struct nlist namelist[] = {
#define X_UMA_KEGS 0
{ .n_name = "_uma_kegs" },
#define X_MP_MAXID 1
{ .n_name = "_mp_maxid" },
#define X_ALL_CPUS 2
{ .n_name = "_all_cpus" },
#define X_VM_NDOMAINS 3
{ .n_name = "_vm_ndomains" },
{ .n_name = "" },
};
/*
* Extract uma(9) statistics from the running kernel, and store all memory
* type information in the passed list. For each type, check the list for an
* existing entry with the right name/allocator -- if present, update that
* entry. Otherwise, add a new entry. On error, the entire list will be
* cleared, as entries will be in an inconsistent state.
*
* To reduce the level of work for a list that starts empty, we keep around a
* hint as to whether it was empty when we began, so we can avoid searching
* the list for entries to update. Updates are O(n^2) due to searching for
* each entry before adding it.
*/
int
memstat_sysctl_uma(struct memory_type_list *list, int flags)
{
struct uma_stream_header *ushp;
struct uma_type_header *uthp;
struct uma_percpu_stat *upsp;
struct memory_type *mtp;
int count, hint_dontsearch, i, j, maxcpus, maxid;
char *buffer, *p;
size_t size;
hint_dontsearch = LIST_EMPTY(&list->mtl_list);
/*
* Query the number of CPUs, number of malloc types so that we can
* guess an initial buffer size. We loop until we succeed or really
* fail. Note that the value of maxcpus we query using sysctl is not
* the version we use when processing the real data -- that is read
* from the header.
*/
retry:
size = sizeof(maxid);
if (sysctlbyname("kern.smp.maxid", &maxid, &size, NULL, 0) < 0) {
if (errno == EACCES || errno == EPERM)
list->mtl_error = MEMSTAT_ERROR_PERMISSION;
else
list->mtl_error = MEMSTAT_ERROR_DATAERROR;
return (-1);
}
if (size != sizeof(maxid)) {
list->mtl_error = MEMSTAT_ERROR_DATAERROR;
return (-1);
}
size = sizeof(count);
if (sysctlbyname("vm.zone_count", &count, &size, NULL, 0) < 0) {
if (errno == EACCES || errno == EPERM)
list->mtl_error = MEMSTAT_ERROR_PERMISSION;
else
list->mtl_error = MEMSTAT_ERROR_VERSION;
return (-1);
}
if (size != sizeof(count)) {
list->mtl_error = MEMSTAT_ERROR_DATAERROR;
return (-1);
}
size = sizeof(*uthp) + count * (sizeof(*uthp) + sizeof(*upsp) *
(maxid + 1));
buffer = malloc(size);
if (buffer == NULL) {
list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
return (-1);
}
if (sysctlbyname("vm.zone_stats", buffer, &size, NULL, 0) < 0) {
/*
* XXXRW: ENOMEM is an ambiguous return, we should bound the
* number of loops, perhaps.
*/
if (errno == ENOMEM) {
free(buffer);
goto retry;
}
if (errno == EACCES || errno == EPERM)
list->mtl_error = MEMSTAT_ERROR_PERMISSION;
else
list->mtl_error = MEMSTAT_ERROR_VERSION;
free(buffer);
return (-1);
}
if (size == 0) {
free(buffer);
return (0);
}
if (size < sizeof(*ushp)) {
list->mtl_error = MEMSTAT_ERROR_VERSION;
free(buffer);
return (-1);
}
p = buffer;
ushp = (struct uma_stream_header *)p;
p += sizeof(*ushp);
if (ushp->ush_version != UMA_STREAM_VERSION) {
list->mtl_error = MEMSTAT_ERROR_VERSION;
free(buffer);
return (-1);
}
/*
* For the remainder of this function, we are quite trusting about
* the layout of structures and sizes, since we've determined we have
* a matching version and acceptable CPU count.
*/
maxcpus = ushp->ush_maxcpus;
count = ushp->ush_count;
for (i = 0; i < count; i++) {
uthp = (struct uma_type_header *)p;
p += sizeof(*uthp);
if (hint_dontsearch == 0) {
mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
uthp->uth_name);
} else
mtp = NULL;
if (mtp == NULL)
mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
uthp->uth_name, maxid + 1);
if (mtp == NULL) {
_memstat_mtl_empty(list);
free(buffer);
list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
return (-1);
}
/*
* Reset the statistics on a current node.
*/
_memstat_mt_reset_stats(mtp, maxid + 1);
mtp->mt_numallocs = uthp->uth_allocs;
mtp->mt_numfrees = uthp->uth_frees;
mtp->mt_failures = uthp->uth_fails;
mtp->mt_sleeps = uthp->uth_sleeps;
mtp->mt_xdomain = uthp->uth_xdomain;
for (j = 0; j < maxcpus; j++) {
upsp = (struct uma_percpu_stat *)p;
p += sizeof(*upsp);
mtp->mt_percpu_cache[j].mtp_free =
upsp->ups_cache_free;
mtp->mt_free += upsp->ups_cache_free;
mtp->mt_numallocs += upsp->ups_allocs;
mtp->mt_numfrees += upsp->ups_frees;
}
/*
* Values for uth_allocs and uth_frees frees are snap.
* It may happen that kernel reports that number of frees
* is greater than number of allocs. See counter(9) for
* details.
*/
if (mtp->mt_numallocs < mtp->mt_numfrees)
mtp->mt_numallocs = mtp->mt_numfrees;
mtp->mt_size = uthp->uth_size;
mtp->mt_rsize = uthp->uth_rsize;
mtp->mt_memalloced = mtp->mt_numallocs * uthp->uth_size;
mtp->mt_memfreed = mtp->mt_numfrees * uthp->uth_size;
mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
mtp->mt_countlimit = uthp->uth_limit;
mtp->mt_byteslimit = uthp->uth_limit * uthp->uth_size;
mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
mtp->mt_zonefree = uthp->uth_zone_free;
/*
* UMA secondary zones share a keg with the primary zone. To
* avoid double-reporting of free items, report keg free
* items only in the primary zone.
*/
if (!(uthp->uth_zone_flags & UTH_ZONE_SECONDARY)) {
mtp->mt_kegfree = uthp->uth_keg_free;
mtp->mt_free += mtp->mt_kegfree;
}
mtp->mt_free += mtp->mt_zonefree;
}
free(buffer);
return (0);
}
static int
kread(kvm_t *kvm, void *kvm_pointer, void *address, size_t size,
size_t offset)
{
ssize_t ret;
ret = kvm_read(kvm, (unsigned long)kvm_pointer + offset, address,
size);
if (ret < 0)
return (MEMSTAT_ERROR_KVM);
if ((size_t)ret != size)
return (MEMSTAT_ERROR_KVM_SHORTREAD);
return (0);
}
static int
kread_string(kvm_t *kvm, const void *kvm_pointer, char *buffer, int buflen)
{
ssize_t ret;
int i;
for (i = 0; i < buflen; i++) {
ret = kvm_read(kvm, (unsigned long)kvm_pointer + i,
&(buffer[i]), sizeof(char));
if (ret < 0)
return (MEMSTAT_ERROR_KVM);
if ((size_t)ret != sizeof(char))
return (MEMSTAT_ERROR_KVM_SHORTREAD);
if (buffer[i] == '\0')
return (0);
}
/* Truncate. */
buffer[i-1] = '\0';
return (0);
}
static int
kread_symbol(kvm_t *kvm, int index, void *address, size_t size,
size_t offset)
{
ssize_t ret;
ret = kvm_read(kvm, namelist[index].n_value + offset, address, size);
if (ret < 0)
return (MEMSTAT_ERROR_KVM);
if ((size_t)ret != size)
return (MEMSTAT_ERROR_KVM_SHORTREAD);
return (0);
}
/*
* memstat_kvm_uma() is similar to memstat_sysctl_uma(), only it extracts
* UMA(9) statistics from a kernel core/memory file.
*/
int
memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
{
LIST_HEAD(, uma_keg) uma_kegs;
struct memory_type *mtp;
struct uma_zone_domain uzd;
struct uma_bucket *ubp, ub;
struct uma_cache *ucp, *ucp_array;
struct uma_zone *uzp, uz;
struct uma_keg *kzp, kz;
int hint_dontsearch, i, mp_maxid, ndomains, ret;
char name[MEMTYPE_MAXNAME];
cpuset_t all_cpus;
long cpusetsize;
kvm_t *kvm;
kvm = (kvm_t *)kvm_handle;
hint_dontsearch = LIST_EMPTY(&list->mtl_list);
if (kvm_nlist(kvm, namelist) != 0) {
list->mtl_error = MEMSTAT_ERROR_KVM;
return (-1);
}
if (namelist[X_UMA_KEGS].n_type == 0 ||
namelist[X_UMA_KEGS].n_value == 0) {
list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
return (-1);
}
ret = kread_symbol(kvm, X_MP_MAXID, &mp_maxid, sizeof(mp_maxid), 0);
if (ret != 0) {
list->mtl_error = ret;
return (-1);
}
ret = kread_symbol(kvm, X_VM_NDOMAINS, &ndomains,
sizeof(ndomains), 0);
if (ret != 0) {
list->mtl_error = ret;
return (-1);
}
ret = kread_symbol(kvm, X_UMA_KEGS, &uma_kegs, sizeof(uma_kegs), 0);
if (ret != 0) {
list->mtl_error = ret;
return (-1);
}
cpusetsize = sysconf(_SC_CPUSET_SIZE);
if (cpusetsize == -1 || (u_long)cpusetsize > sizeof(cpuset_t)) {
list->mtl_error = MEMSTAT_ERROR_KVM_NOSYMBOL;
return (-1);
}
CPU_ZERO(&all_cpus);
ret = kread_symbol(kvm, X_ALL_CPUS, &all_cpus, cpusetsize, 0);
if (ret != 0) {
list->mtl_error = ret;
return (-1);
}
ucp_array = malloc(sizeof(struct uma_cache) * (mp_maxid + 1));
if (ucp_array == NULL) {
list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
return (-1);
}
for (kzp = LIST_FIRST(&uma_kegs); kzp != NULL; kzp =
LIST_NEXT(&kz, uk_link)) {
ret = kread(kvm, kzp, &kz, sizeof(kz), 0);
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
for (uzp = LIST_FIRST(&kz.uk_zones); uzp != NULL; uzp =
LIST_NEXT(&uz, uz_link)) {
ret = kread(kvm, uzp, &uz, sizeof(uz), 0);
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
ret = kread(kvm, uzp, ucp_array,
sizeof(struct uma_cache) * (mp_maxid + 1),
offsetof(struct uma_zone, uz_cpu[0]));
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
ret = kread_string(kvm, uz.uz_name, name,
MEMTYPE_MAXNAME);
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
if (hint_dontsearch == 0) {
mtp = memstat_mtl_find(list, ALLOCATOR_UMA,
name);
} else
mtp = NULL;
if (mtp == NULL)
mtp = _memstat_mt_allocate(list, ALLOCATOR_UMA,
name, mp_maxid + 1);
if (mtp == NULL) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = MEMSTAT_ERROR_NOMEMORY;
return (-1);
}
/*
* Reset the statistics on a current node.
*/
_memstat_mt_reset_stats(mtp, mp_maxid + 1);
mtp->mt_numallocs = kvm_counter_u64_fetch(kvm,
(unsigned long )uz.uz_allocs);
mtp->mt_numfrees = kvm_counter_u64_fetch(kvm,
(unsigned long )uz.uz_frees);
mtp->mt_failures = kvm_counter_u64_fetch(kvm,
(unsigned long )uz.uz_fails);
mtp->mt_sleeps = uz.uz_sleeps;
/* See comment above in memstat_sysctl_uma(). */
if (mtp->mt_numallocs < mtp->mt_numfrees)
mtp->mt_numallocs = mtp->mt_numfrees;
mtp->mt_xdomain = uz.uz_xdomain;
if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
goto skip_percpu;
for (i = 0; i < mp_maxid + 1; i++) {
if (!CPU_ISSET(i, &all_cpus))
continue;
ucp = &ucp_array[i];
mtp->mt_numallocs += ucp->uc_allocs;
mtp->mt_numfrees += ucp->uc_frees;
if (ucp->uc_allocbucket != NULL) {
ret = kread(kvm, ucp->uc_allocbucket,
&ub, sizeof(ub), 0);
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
mtp->mt_free += ub.ub_cnt;
}
if (ucp->uc_freebucket != NULL) {
ret = kread(kvm, ucp->uc_freebucket,
&ub, sizeof(ub), 0);
if (ret != 0) {
free(ucp_array);
_memstat_mtl_empty(list);
list->mtl_error = ret;
return (-1);
}
mtp->mt_free += ub.ub_cnt;
}
}
skip_percpu:
mtp->mt_size = kz.uk_size;
mtp->mt_rsize = kz.uk_rsize;
mtp->mt_memalloced = mtp->mt_numallocs * mtp->mt_size;
mtp->mt_memfreed = mtp->mt_numfrees * mtp->mt_size;
mtp->mt_bytes = mtp->mt_memalloced - mtp->mt_memfreed;
mtp->mt_countlimit = uz.uz_max_items;
mtp->mt_byteslimit = mtp->mt_countlimit * mtp->mt_size;
mtp->mt_count = mtp->mt_numallocs - mtp->mt_numfrees;
for (i = 0; i < ndomains; i++) {
ret = kread(kvm, &uz.uz_domain[i], &uzd,
sizeof(uzd), 0);
for (ubp =
TAILQ_FIRST(&uzd.uzd_buckets);
ubp != NULL;
ubp = TAILQ_NEXT(&ub, ub_link)) {
ret = kread(kvm, ubp, &ub,
sizeof(ub), 0);
mtp->mt_zonefree += ub.ub_cnt;
}
}
if (!((kz.uk_flags & UMA_ZONE_SECONDARY) &&
LIST_FIRST(&kz.uk_zones) != uzp)) {
mtp->mt_kegfree = kz.uk_free;
mtp->mt_free += mtp->mt_kegfree;
}
mtp->mt_free += mtp->mt_zonefree;
}
}
free(ucp_array);
return (0);
}