summaryrefslogtreecommitdiffstats
path: root/drivers/md
diff options
context:
space:
mode:
authorMike Christie <michael.christie@oracle.com>2020-10-23 02:27:50 +0200
committerMike Snitzer <snitzer@redhat.com>2020-12-05 00:04:35 +0100
commite4d2e82b2300b03f66b3ca8417590c86e661fab1 (patch)
tree6719538b312dae52e2952269816217e5cace4b69 /drivers/md
parentdm verity: Add support for signature verification with 2nd keyring (diff)
downloadlinux-e4d2e82b2300b03f66b3ca8417590c86e661fab1.tar.xz
linux-e4d2e82b2300b03f66b3ca8417590c86e661fab1.zip
dm mpath: add IO affinity path selector
This patch adds a path selector that selects paths based on a CPU to path mapping the user passes in and what CPU we are executing on. The primary user for this PS is where the app is optimized to use specific CPUs so other PSs undo the apps handy work, and the storage and it's transport are not a bottlneck. For these io-affinity PS setups a path's transport/interconnect perf is not going to flucuate a lot and there is no major differences between paths, so QL/HST smarts do not help and RR always messes up what the app is trying to do. On a system with 16 cores, where you have a job per CPU: fio --filename=/dev/dm-0 --direct=1 --rw=randrw --bs=4k \ --ioengine=libaio --iodepth=128 --numjobs=16 and a dm-multipath device setup where each CPU is mapped to one path: // When in mq mode I had to set dm_mq_nr_hw_queues=$NUM_PATHS. // Bio mode also showed similar results. 0 16777216 multipath 0 0 1 1 io-affinity 0 16 1 8:16 1 8:32 2 8:64 4 8:48 8 8:80 10 8:96 20 8:112 40 8:128 80 8:144 100 8:160 200 8:176 400 8:192 800 8:208 1000 8:224 2000 8:240 4000 65:0 8000 we can see a IOPs increase of 25%. The percent increase depends on the device and interconnect. For a slower/medium speed path/device that can do around 180K IOPs a path if you ran that fio command to it directly we saw a 25% increase like above. Slower path'd devices that could do around 90K per path showed maybe around a 2 - 5% increase. If you use something like null_blk or scsi_debug which can multi-million IOPs and hack it up so each device they export shows up as a path then you see 50%+ increases. Signed-off-by: Mike Christie <michael.christie@oracle.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Diffstat (limited to 'drivers/md')
-rw-r--r--drivers/md/Kconfig9
-rw-r--r--drivers/md/Makefile1
-rw-r--r--drivers/md/dm-io-affinity.c272
3 files changed, 282 insertions, 0 deletions
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 1d68935e45ef..b7e2d9666614 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -463,6 +463,15 @@ config DM_MULTIPATH_HST
If unsure, say N.
+config DM_MULTIPATH_IOA
+ tristate "I/O Path Selector based on CPU submission"
+ depends on DM_MULTIPATH
+ help
+ This path selector selects the path based on the CPU the IO is
+ executed on and the CPU to path mapping setup at path addition time.
+
+ If unsure, say N.
+
config DM_DELAY
tristate "I/O delaying target"
depends on BLK_DEV_DM
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 6d3e234dc46a..4f95f332d015 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -59,6 +59,7 @@ obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o
obj-$(CONFIG_DM_MULTIPATH_QL) += dm-queue-length.o
obj-$(CONFIG_DM_MULTIPATH_ST) += dm-service-time.o
obj-$(CONFIG_DM_MULTIPATH_HST) += dm-historical-service-time.o
+obj-$(CONFIG_DM_MULTIPATH_IOA) += dm-io-affinity.o
obj-$(CONFIG_DM_SWITCH) += dm-switch.o
obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
obj-$(CONFIG_DM_PERSISTENT_DATA) += persistent-data/
diff --git a/drivers/md/dm-io-affinity.c b/drivers/md/dm-io-affinity.c
new file mode 100644
index 000000000000..077655cd4fae
--- /dev/null
+++ b/drivers/md/dm-io-affinity.c
@@ -0,0 +1,272 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020 Oracle Corporation
+ *
+ * Module Author: Mike Christie
+ */
+#include "dm-path-selector.h"
+
+#include <linux/device-mapper.h>
+#include <linux/module.h>
+
+#define DM_MSG_PREFIX "multipath io-affinity"
+
+struct path_info {
+ struct dm_path *path;
+ cpumask_var_t cpumask;
+ refcount_t refcount;
+ bool failed;
+};
+
+struct selector {
+ struct path_info **path_map;
+ cpumask_var_t path_mask;
+ atomic_t map_misses;
+};
+
+static void ioa_free_path(struct selector *s, unsigned int cpu)
+{
+ struct path_info *pi = s->path_map[cpu];
+
+ if (!pi)
+ return;
+
+ if (refcount_dec_and_test(&pi->refcount)) {
+ cpumask_clear_cpu(cpu, s->path_mask);
+ free_cpumask_var(pi->cpumask);
+ kfree(pi);
+
+ s->path_map[cpu] = NULL;
+ }
+}
+
+static int ioa_add_path(struct path_selector *ps, struct dm_path *path,
+ int argc, char **argv, char **error)
+{
+ struct selector *s = ps->context;
+ struct path_info *pi = NULL;
+ unsigned int cpu;
+ int ret;
+
+ if (argc != 1) {
+ *error = "io-affinity ps: invalid number of arguments";
+ return -EINVAL;
+ }
+
+ pi = kzalloc(sizeof(*pi), GFP_KERNEL);
+ if (!pi) {
+ *error = "io-affinity ps: Error allocating path context";
+ return -ENOMEM;
+ }
+
+ pi->path = path;
+ path->pscontext = pi;
+ refcount_set(&pi->refcount, 1);
+
+ if (!zalloc_cpumask_var(&pi->cpumask, GFP_KERNEL)) {
+ *error = "io-affinity ps: Error allocating cpumask context";
+ ret = -ENOMEM;
+ goto free_pi;
+ }
+
+ ret = cpumask_parse(argv[0], pi->cpumask);
+ if (ret) {
+ *error = "io-affinity ps: invalid cpumask";
+ ret = -EINVAL;
+ goto free_mask;
+ }
+
+ for_each_cpu(cpu, pi->cpumask) {
+ if (cpu >= nr_cpu_ids) {
+ DMWARN_LIMIT("Ignoring mapping for CPU %u. Max CPU is %u",
+ cpu, nr_cpu_ids);
+ break;
+ }
+
+ if (s->path_map[cpu]) {
+ DMWARN("CPU mapping for %u exists. Ignoring.", cpu);
+ continue;
+ }
+
+ cpumask_set_cpu(cpu, s->path_mask);
+ s->path_map[cpu] = pi;
+ refcount_inc(&pi->refcount);
+ continue;
+ }
+
+ if (refcount_dec_and_test(&pi->refcount)) {
+ *error = "io-affinity ps: No new/valid CPU mapping found";
+ ret = -EINVAL;
+ goto free_mask;
+ }
+
+ return 0;
+
+free_mask:
+ free_cpumask_var(pi->cpumask);
+free_pi:
+ kfree(pi);
+ return ret;
+}
+
+static int ioa_create(struct path_selector *ps, unsigned argc, char **argv)
+{
+ struct selector *s;
+
+ s = kmalloc(sizeof(*s), GFP_KERNEL);
+ if (!s)
+ return -ENOMEM;
+
+ s->path_map = kzalloc(nr_cpu_ids * sizeof(struct path_info *),
+ GFP_KERNEL);
+ if (!s->path_map)
+ goto free_selector;
+
+ if (!zalloc_cpumask_var(&s->path_mask, GFP_KERNEL))
+ goto free_map;
+
+ atomic_set(&s->map_misses, 0);
+ ps->context = s;
+ return 0;
+
+free_map:
+ kfree(s->path_map);
+free_selector:
+ kfree(s);
+ return -ENOMEM;
+}
+
+static void ioa_destroy(struct path_selector *ps)
+{
+ struct selector *s = ps->context;
+ unsigned cpu;
+
+ for_each_cpu(cpu, s->path_mask)
+ ioa_free_path(s, cpu);
+
+ free_cpumask_var(s->path_mask);
+ kfree(s->path_map);
+ kfree(s);
+
+ ps->context = NULL;
+}
+
+static int ioa_status(struct path_selector *ps, struct dm_path *path,
+ status_type_t type, char *result, unsigned int maxlen)
+{
+ struct selector *s = ps->context;
+ struct path_info *pi;
+ int sz = 0;
+
+ if (!path) {
+ DMEMIT("0 ");
+ return sz;
+ }
+
+ switch(type) {
+ case STATUSTYPE_INFO:
+ DMEMIT("%d ", atomic_read(&s->map_misses));
+ break;
+ case STATUSTYPE_TABLE:
+ pi = path->pscontext;
+ DMEMIT("%*pb ", cpumask_pr_args(pi->cpumask));
+ break;
+ }
+
+ return sz;
+}
+
+static void ioa_fail_path(struct path_selector *ps, struct dm_path *p)
+{
+ struct path_info *pi = p->pscontext;
+
+ pi->failed = true;
+}
+
+static int ioa_reinstate_path(struct path_selector *ps, struct dm_path *p)
+{
+ struct path_info *pi = p->pscontext;
+
+ pi->failed = false;
+ return 0;
+}
+
+static struct dm_path *ioa_select_path(struct path_selector *ps,
+ size_t nr_bytes)
+{
+ unsigned int cpu, node;
+ struct selector *s = ps->context;
+ const struct cpumask *cpumask;
+ struct path_info *pi;
+ int i;
+
+ cpu = get_cpu();
+
+ pi = s->path_map[cpu];
+ if (pi && !pi->failed)
+ goto done;
+
+ /*
+ * Perf is not optimal, but we at least try the local node then just
+ * try not to fail.
+ */
+ if (!pi)
+ atomic_inc(&s->map_misses);
+
+ node = cpu_to_node(cpu);
+ cpumask = cpumask_of_node(node);
+ for_each_cpu(i, cpumask) {
+ pi = s->path_map[i];
+ if (pi && !pi->failed)
+ goto done;
+ }
+
+ for_each_cpu(i, s->path_mask) {
+ pi = s->path_map[i];
+ if (pi && !pi->failed)
+ goto done;
+ }
+ pi = NULL;
+
+done:
+ put_cpu();
+ return pi ? pi->path : NULL;
+}
+
+static struct path_selector_type ioa_ps = {
+ .name = "io-affinity",
+ .module = THIS_MODULE,
+ .table_args = 1,
+ .info_args = 1,
+ .create = ioa_create,
+ .destroy = ioa_destroy,
+ .status = ioa_status,
+ .add_path = ioa_add_path,
+ .fail_path = ioa_fail_path,
+ .reinstate_path = ioa_reinstate_path,
+ .select_path = ioa_select_path,
+};
+
+static int __init dm_ioa_init(void)
+{
+ int ret = dm_register_path_selector(&ioa_ps);
+
+ if (ret < 0)
+ DMERR("register failed %d", ret);
+ return ret;
+}
+
+static void __exit dm_ioa_exit(void)
+{
+ int ret = dm_unregister_path_selector(&ioa_ps);
+
+ if (ret < 0)
+ DMERR("unregister failed %d", ret);
+}
+
+module_init(dm_ioa_init);
+module_exit(dm_ioa_exit);
+
+MODULE_DESCRIPTION(DM_NAME " multipath path selector that selects paths based on the CPU IO is being executed on");
+MODULE_AUTHOR("Mike Christie <michael.christie@oracle.com>");
+MODULE_LICENSE("GPL");