raid5: set write hint for PPL
authorMariusz Dabrowski <mariusz.dabrowski@intel.com>
Mon, 18 Feb 2019 14:04:09 +0000 (15:04 +0100)
committerSong Liu <songliubraving@fb.com>
Tue, 12 Mar 2019 17:15:18 +0000 (10:15 -0700)
When the Partial Parity Log is enabled, circular buffer is used to store
PPL data. Each write to RAID device causes overwrite of data in this buffer
so some write_hint can be set to those request to help drives handle
garbage collection. This patch adds new sysfs attribute which can be used
to specify which write_hint should be assigned to PPL.

Acked-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Mariusz Dabrowski <mariusz.dabrowski@intel.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Documentation/admin-guide/md.rst
drivers/md/raid5-log.h
drivers/md/raid5-ppl.c
drivers/md/raid5.c

index 84de718..3c51084 100644 (file)
@@ -756,3 +756,6 @@ These currently include:
       The cache mode for raid5. raid5 could include an extra disk for
       caching. The mode can be "write-throuth" and "write-back". The
       default is "write-through".
+
+  ppl_write_hint
+      NVMe stream ID to be set for each PPL write request.
index bfb8114..43c714a 100644 (file)
@@ -45,6 +45,7 @@ extern void ppl_stripe_write_finished(struct stripe_head *sh);
 extern int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add);
 extern void ppl_quiesce(struct r5conf *conf, int quiesce);
 extern int ppl_handle_flush_request(struct r5l_log *log, struct bio *bio);
+extern struct md_sysfs_entry ppl_write_hint;
 
 static inline bool raid5_has_log(struct r5conf *conf)
 {
index 3a7c363..f2b3020 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/raid/md_p.h>
 #include "md.h"
 #include "raid5.h"
+#include "raid5-log.h"
 
 /*
  * PPL consists of a 4KB header (struct ppl_header) and at least 128KB for
@@ -116,6 +117,8 @@ struct ppl_conf {
        /* stripes to retry if failed to allocate io_unit */
        struct list_head no_mem_stripes;
        spinlock_t no_mem_stripes_lock;
+
+       unsigned short write_hint;
 };
 
 struct ppl_log {
@@ -476,6 +479,7 @@ static void ppl_submit_iounit(struct ppl_io_unit *io)
        bio_set_dev(bio, log->rdev->bdev);
        bio->bi_iter.bi_sector = log->next_io_sector;
        bio_add_page(bio, io->header_page, PAGE_SIZE, 0);
+       bio->bi_write_hint = ppl_conf->write_hint;
 
        pr_debug("%s: log->current_io_sector: %llu\n", __func__,
            (unsigned long long)log->next_io_sector);
@@ -505,6 +509,7 @@ static void ppl_submit_iounit(struct ppl_io_unit *io)
                        bio = bio_alloc_bioset(GFP_NOIO, BIO_MAX_PAGES,
                                               &ppl_conf->bs);
                        bio->bi_opf = prev->bi_opf;
+                       bio->bi_write_hint = prev->bi_write_hint;
                        bio_copy_dev(bio, prev);
                        bio->bi_iter.bi_sector = bio_end_sector(prev);
                        bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0);
@@ -1409,6 +1414,7 @@ int ppl_init_log(struct r5conf *conf)
        atomic64_set(&ppl_conf->seq, 0);
        INIT_LIST_HEAD(&ppl_conf->no_mem_stripes);
        spin_lock_init(&ppl_conf->no_mem_stripes_lock);
+       ppl_conf->write_hint = RWF_WRITE_LIFE_NOT_SET;
 
        if (!mddev->external) {
                ppl_conf->signature = ~crc32c_le(~0, mddev->uuid, sizeof(mddev->uuid));
@@ -1503,3 +1509,60 @@ int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add)
 
        return ret;
 }
+
+static ssize_t
+ppl_write_hint_show(struct mddev *mddev, char *buf)
+{
+       size_t ret = 0;
+       struct r5conf *conf;
+       struct ppl_conf *ppl_conf = NULL;
+
+       spin_lock(&mddev->lock);
+       conf = mddev->private;
+       if (conf && raid5_has_ppl(conf))
+               ppl_conf = conf->log_private;
+       ret = sprintf(buf, "%d\n", ppl_conf ? ppl_conf->write_hint : 0);
+       spin_unlock(&mddev->lock);
+
+       return ret;
+}
+
+static ssize_t
+ppl_write_hint_store(struct mddev *mddev, const char *page, size_t len)
+{
+       struct r5conf *conf;
+       struct ppl_conf *ppl_conf;
+       int err = 0;
+       unsigned short new;
+
+       if (len >= PAGE_SIZE)
+               return -EINVAL;
+       if (kstrtou16(page, 10, &new))
+               return -EINVAL;
+
+       err = mddev_lock(mddev);
+       if (err)
+               return err;
+
+       conf = mddev->private;
+       if (!conf) {
+               err = -ENODEV;
+       } else if (raid5_has_ppl(conf)) {
+               ppl_conf = conf->log_private;
+               if (!ppl_conf)
+                       err = -EINVAL;
+               else
+                       ppl_conf->write_hint = new;
+       } else {
+               err = -EINVAL;
+       }
+
+       mddev_unlock(mddev);
+
+       return err ?: len;
+}
+
+struct md_sysfs_entry
+ppl_write_hint = __ATTR(ppl_write_hint, S_IRUGO | S_IWUSR,
+                       ppl_write_hint_show,
+                       ppl_write_hint_store);
index cecea90..09562d7 100644 (file)
@@ -6660,6 +6660,7 @@ static struct attribute *raid5_attrs[] =  {
        &raid5_skip_copy.attr,
        &raid5_rmw_level.attr,
        &r5c_journal_mode.attr,
+       &ppl_write_hint.attr,
        NULL,
 };
 static struct attribute_group raid5_attrs_group = {