Commit 36d1c647 authored by Dan Williams's avatar Dan Williams

md/raid6: move the spare page to a percpu allocation

In preparation for asynchronous handling of raid6 operations move the
spare page to a percpu allocation to allow multiple simultaneous
synchronous raid6 recovery operations.

Make this allocation cpu hotplug aware to maximize allocation
efficiency.
Signed-off-by: default avatarDan Williams <dan.j.williams@intel.com>

parent a11034b4
......@@ -48,6 +48,7 @@
#include <linux/raid/pq.h>
#include <linux/async_tx.h>
#include <linux/seq_file.h>
#include <linux/cpu.h>
#include "md.h"
#include "raid5.h"
#include "bitmap.h"
......@@ -2566,13 +2567,14 @@ static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
struct stripe_head_state *s,
struct r6_state *r6s, struct page *tmp_page,
int disks)
struct r6_state *r6s, int disks)
{
int update_p = 0, update_q = 0;
struct r5dev *dev;
int pd_idx = sh->pd_idx;
int qd_idx = sh->qd_idx;
unsigned long cpu;
struct page *tmp_page;
set_bit(STRIPE_HANDLE, &sh->state);
......@@ -2583,12 +2585,8 @@ static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
* case we can only check one of them, possibly using the
* other to generate missing data
*/
/* If !tmp_page, we cannot do the calculations,
* but as we have set STRIPE_HANDLE, we will soon be called
* by stripe_handle with a tmp_page - just wait until then.
*/
if (tmp_page) {
cpu = get_cpu();
tmp_page = per_cpu_ptr(conf->percpu, cpu)->spare_page;
if (s->failed == r6s->q_failed) {
/* The only possible failed device holds 'Q', so it
* makes sense to check P (If anything else were failed,
......@@ -2615,6 +2613,8 @@ static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
update_q = 1;
}
}
put_cpu();
if (update_p || update_q) {
conf->mddev->resync_mismatches += STRIPE_SECTORS;
if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
......@@ -2654,7 +2654,6 @@ static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
clear_bit(STRIPE_DEGRADED, &sh->state);
set_bit(STRIPE_INSYNC, &sh->state);
}
}
static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
......@@ -3009,7 +3008,7 @@ static bool handle_stripe5(struct stripe_head *sh)
return blocked_rdev == NULL;
}
static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
static bool handle_stripe6(struct stripe_head *sh)
{
raid5_conf_t *conf = sh->raid_conf;
int disks = sh->disks;
......@@ -3164,7 +3163,7 @@ static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
* data is available
*/
if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state))
handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks);
handle_parity_checks6(conf, sh, &s, &r6s, disks);
if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
md_done_sync(conf->mddev, STRIPE_SECTORS,1);
......@@ -3247,16 +3246,14 @@ static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
}
/* returns true if the stripe was handled */
static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page)
static bool handle_stripe(struct stripe_head *sh)
{
if (sh->raid_conf->level == 6)
return handle_stripe6(sh, tmp_page);
return handle_stripe6(sh);
else
return handle_stripe5(sh);
}
static void raid5_activate_delayed(raid5_conf_t *conf)
{
if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
......@@ -4047,7 +4044,7 @@ static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *ski
spin_unlock(&sh->lock);
/* wait for any blocked device to be handled */
while(unlikely(!handle_stripe(sh, NULL)))
while (unlikely(!handle_stripe(sh)))
;
release_stripe(sh);
......@@ -4104,7 +4101,7 @@ static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
return handled;
}
handle_stripe(sh, NULL);
handle_stripe(sh);
release_stripe(sh);
handled++;
}
......@@ -4168,7 +4165,7 @@ static void raid5d(mddev_t *mddev)
spin_unlock_irq(&conf->device_lock);
handled++;
handle_stripe(sh, conf->spare_page);
handle_stripe(sh);
release_stripe(sh);
spin_lock_irq(&conf->device_lock);
......@@ -4309,15 +4306,104 @@ raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
return sectors * (raid_disks - conf->max_degraded);
}
static void raid5_free_percpu(raid5_conf_t *conf)
{
struct raid5_percpu *percpu;
unsigned long cpu;
if (!conf->percpu)
return;
get_online_cpus();
for_each_possible_cpu(cpu) {
percpu = per_cpu_ptr(conf->percpu, cpu);
safe_put_page(percpu->spare_page);
}
#ifdef CONFIG_HOTPLUG_CPU
unregister_cpu_notifier(&conf->cpu_notify);
#endif
put_online_cpus();
free_percpu(conf->percpu);
}
static void free_conf(raid5_conf_t *conf)
{
shrink_stripes(conf);
safe_put_page(conf->spare_page);
raid5_free_percpu(conf);
kfree(conf->disks);
kfree(conf->stripe_hashtbl);
kfree(conf);
}
#ifdef CONFIG_HOTPLUG_CPU
static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
void *hcpu)
{
raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
long cpu = (long)hcpu;
struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
switch (action) {
case CPU_UP_PREPARE:
case CPU_UP_PREPARE_FROZEN:
if (!percpu->spare_page)
percpu->spare_page = alloc_page(GFP_KERNEL);
if (!percpu->spare_page) {
pr_err("%s: failed memory allocation for cpu%ld\n",
__func__, cpu);
return NOTIFY_BAD;
}
break;
case CPU_DEAD:
case CPU_DEAD_FROZEN:
safe_put_page(percpu->spare_page);
percpu->spare_page = NULL;
break;
default:
break;
}
return NOTIFY_OK;
}
#endif
static int raid5_alloc_percpu(raid5_conf_t *conf)
{
unsigned long cpu;
struct page *spare_page;
struct raid5_percpu *allcpus;
int err;
/* the only percpu data is the raid6 spare page */
if (conf->level != 6)
return 0;
allcpus = alloc_percpu(struct raid5_percpu);
if (!allcpus)
return -ENOMEM;
conf->percpu = allcpus;
get_online_cpus();
err = 0;
for_each_present_cpu(cpu) {
spare_page = alloc_page(GFP_KERNEL);
if (!spare_page) {
err = -ENOMEM;
break;
}
per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
}
#ifdef CONFIG_HOTPLUG_CPU
conf->cpu_notify.notifier_call = raid456_cpu_notify;
conf->cpu_notify.priority = 0;
if (err == 0)
err = register_cpu_notifier(&conf->cpu_notify);
#endif
put_online_cpus();
return err;
}
static raid5_conf_t *setup_conf(mddev_t *mddev)
{
raid5_conf_t *conf;
......@@ -4372,11 +4458,10 @@ static raid5_conf_t *setup_conf(mddev_t *mddev)
if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
goto abort;
if (mddev->new_level == 6) {
conf->spare_page = alloc_page(GFP_KERNEL);
if (!conf->spare_page)
conf->level = mddev->new_level;
if (raid5_alloc_percpu(conf) != 0)
goto abort;
}
spin_lock_init(&conf->device_lock);
init_waitqueue_head(&conf->wait_for_stripe);
init_waitqueue_head(&conf->wait_for_overlap);
......@@ -4412,7 +4497,6 @@ static raid5_conf_t *setup_conf(mddev_t *mddev)
}
conf->chunk_size = mddev->new_chunk;
conf->level = mddev->new_level;
if (conf->level == 6)
conf->max_degraded = 2;
else
......
......@@ -383,8 +383,13 @@ struct raid5_private_data {
* (fresh device added).
* Cleared when a sync completes.
*/
/* per cpu variables */
struct raid5_percpu {
struct page *spare_page; /* Used when checking P/Q in raid6 */
} *percpu;
#ifdef CONFIG_HOTPLUG_CPU
struct notifier_block cpu_notify;
#endif
/*
* Free stripes pool
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment