Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Copyright (C) 2017 Western Digital Corporation or its affiliates.
0004  *
0005  * This file is released under the GPL.
0006  */
0007 
0008 #include "dm-zoned.h"
0009 
0010 #include <linux/module.h>
0011 
0012 #define DM_MSG_PREFIX       "zoned reclaim"
0013 
0014 struct dmz_reclaim {
0015     struct dmz_metadata     *metadata;
0016 
0017     struct delayed_work work;
0018     struct workqueue_struct *wq;
0019 
0020     struct dm_kcopyd_client *kc;
0021     struct dm_kcopyd_throttle kc_throttle;
0022     int         kc_err;
0023 
0024     int         dev_idx;
0025 
0026     unsigned long       flags;
0027 
0028     /* Last target access time */
0029     unsigned long       atime;
0030 };
0031 
0032 /*
0033  * Reclaim state flags.
0034  */
0035 enum {
0036     DMZ_RECLAIM_KCOPY,
0037 };
0038 
0039 /*
0040  * Number of seconds of target BIO inactivity to consider the target idle.
0041  */
0042 #define DMZ_IDLE_PERIOD         (10UL * HZ)
0043 
0044 /*
0045  * Percentage of unmapped (free) random zones below which reclaim starts
0046  * even if the target is busy.
0047  */
0048 #define DMZ_RECLAIM_LOW_UNMAP_ZONES 30
0049 
0050 /*
0051  * Percentage of unmapped (free) random zones above which reclaim will
0052  * stop if the target is busy.
0053  */
0054 #define DMZ_RECLAIM_HIGH_UNMAP_ZONES    50
0055 
0056 /*
0057  * Align a sequential zone write pointer to chunk_block.
0058  */
0059 static int dmz_reclaim_align_wp(struct dmz_reclaim *zrc, struct dm_zone *zone,
0060                 sector_t block)
0061 {
0062     struct dmz_metadata *zmd = zrc->metadata;
0063     struct dmz_dev *dev = zone->dev;
0064     sector_t wp_block = zone->wp_block;
0065     unsigned int nr_blocks;
0066     int ret;
0067 
0068     if (wp_block == block)
0069         return 0;
0070 
0071     if (wp_block > block)
0072         return -EIO;
0073 
0074     /*
0075      * Zeroout the space between the write
0076      * pointer and the requested position.
0077      */
0078     nr_blocks = block - wp_block;
0079     ret = blkdev_issue_zeroout(dev->bdev,
0080                    dmz_start_sect(zmd, zone) + dmz_blk2sect(wp_block),
0081                    dmz_blk2sect(nr_blocks), GFP_NOIO, 0);
0082     if (ret) {
0083         dmz_dev_err(dev,
0084                 "Align zone %u wp %llu to %llu (wp+%u) blocks failed %d",
0085                 zone->id, (unsigned long long)wp_block,
0086                 (unsigned long long)block, nr_blocks, ret);
0087         dmz_check_bdev(dev);
0088         return ret;
0089     }
0090 
0091     zone->wp_block = block;
0092 
0093     return 0;
0094 }
0095 
0096 /*
0097  * dm_kcopyd_copy end notification.
0098  */
0099 static void dmz_reclaim_kcopy_end(int read_err, unsigned long write_err,
0100                   void *context)
0101 {
0102     struct dmz_reclaim *zrc = context;
0103 
0104     if (read_err || write_err)
0105         zrc->kc_err = -EIO;
0106     else
0107         zrc->kc_err = 0;
0108 
0109     clear_bit_unlock(DMZ_RECLAIM_KCOPY, &zrc->flags);
0110     smp_mb__after_atomic();
0111     wake_up_bit(&zrc->flags, DMZ_RECLAIM_KCOPY);
0112 }
0113 
0114 /*
0115  * Copy valid blocks of src_zone into dst_zone.
0116  */
0117 static int dmz_reclaim_copy(struct dmz_reclaim *zrc,
0118                 struct dm_zone *src_zone, struct dm_zone *dst_zone)
0119 {
0120     struct dmz_metadata *zmd = zrc->metadata;
0121     struct dm_io_region src, dst;
0122     sector_t block = 0, end_block;
0123     sector_t nr_blocks;
0124     sector_t src_zone_block;
0125     sector_t dst_zone_block;
0126     unsigned long flags = 0;
0127     int ret;
0128 
0129     if (dmz_is_seq(src_zone))
0130         end_block = src_zone->wp_block;
0131     else
0132         end_block = dmz_zone_nr_blocks(zmd);
0133     src_zone_block = dmz_start_block(zmd, src_zone);
0134     dst_zone_block = dmz_start_block(zmd, dst_zone);
0135 
0136     if (dmz_is_seq(dst_zone))
0137         flags |= BIT(DM_KCOPYD_WRITE_SEQ);
0138 
0139     while (block < end_block) {
0140         if (src_zone->dev->flags & DMZ_BDEV_DYING)
0141             return -EIO;
0142         if (dst_zone->dev->flags & DMZ_BDEV_DYING)
0143             return -EIO;
0144 
0145         if (dmz_reclaim_should_terminate(src_zone))
0146             return -EINTR;
0147 
0148         /* Get a valid region from the source zone */
0149         ret = dmz_first_valid_block(zmd, src_zone, &block);
0150         if (ret <= 0)
0151             return ret;
0152         nr_blocks = ret;
0153 
0154         /*
0155          * If we are writing in a sequential zone, we must make sure
0156          * that writes are sequential. So Zeroout any eventual hole
0157          * between writes.
0158          */
0159         if (dmz_is_seq(dst_zone)) {
0160             ret = dmz_reclaim_align_wp(zrc, dst_zone, block);
0161             if (ret)
0162                 return ret;
0163         }
0164 
0165         src.bdev = src_zone->dev->bdev;
0166         src.sector = dmz_blk2sect(src_zone_block + block);
0167         src.count = dmz_blk2sect(nr_blocks);
0168 
0169         dst.bdev = dst_zone->dev->bdev;
0170         dst.sector = dmz_blk2sect(dst_zone_block + block);
0171         dst.count = src.count;
0172 
0173         /* Copy the valid region */
0174         set_bit(DMZ_RECLAIM_KCOPY, &zrc->flags);
0175         dm_kcopyd_copy(zrc->kc, &src, 1, &dst, flags,
0176                    dmz_reclaim_kcopy_end, zrc);
0177 
0178         /* Wait for copy to complete */
0179         wait_on_bit_io(&zrc->flags, DMZ_RECLAIM_KCOPY,
0180                    TASK_UNINTERRUPTIBLE);
0181         if (zrc->kc_err)
0182             return zrc->kc_err;
0183 
0184         block += nr_blocks;
0185         if (dmz_is_seq(dst_zone))
0186             dst_zone->wp_block = block;
0187     }
0188 
0189     return 0;
0190 }
0191 
0192 /*
0193  * Move valid blocks of dzone buffer zone into dzone (after its write pointer)
0194  * and free the buffer zone.
0195  */
0196 static int dmz_reclaim_buf(struct dmz_reclaim *zrc, struct dm_zone *dzone)
0197 {
0198     struct dm_zone *bzone = dzone->bzone;
0199     sector_t chunk_block = dzone->wp_block;
0200     struct dmz_metadata *zmd = zrc->metadata;
0201     int ret;
0202 
0203     DMDEBUG("(%s/%u): Chunk %u, move buf zone %u (weight %u) to data zone %u (weight %u)",
0204         dmz_metadata_label(zmd), zrc->dev_idx,
0205         dzone->chunk, bzone->id, dmz_weight(bzone),
0206         dzone->id, dmz_weight(dzone));
0207 
0208     /* Flush data zone into the buffer zone */
0209     ret = dmz_reclaim_copy(zrc, bzone, dzone);
0210     if (ret < 0)
0211         return ret;
0212 
0213     dmz_lock_flush(zmd);
0214 
0215     /* Validate copied blocks */
0216     ret = dmz_merge_valid_blocks(zmd, bzone, dzone, chunk_block);
0217     if (ret == 0) {
0218         /* Free the buffer zone */
0219         dmz_invalidate_blocks(zmd, bzone, 0, dmz_zone_nr_blocks(zmd));
0220         dmz_lock_map(zmd);
0221         dmz_unmap_zone(zmd, bzone);
0222         dmz_unlock_zone_reclaim(dzone);
0223         dmz_free_zone(zmd, bzone);
0224         dmz_unlock_map(zmd);
0225     }
0226 
0227     dmz_unlock_flush(zmd);
0228 
0229     return ret;
0230 }
0231 
0232 /*
0233  * Merge valid blocks of dzone into its buffer zone and free dzone.
0234  */
0235 static int dmz_reclaim_seq_data(struct dmz_reclaim *zrc, struct dm_zone *dzone)
0236 {
0237     unsigned int chunk = dzone->chunk;
0238     struct dm_zone *bzone = dzone->bzone;
0239     struct dmz_metadata *zmd = zrc->metadata;
0240     int ret = 0;
0241 
0242     DMDEBUG("(%s/%u): Chunk %u, move data zone %u (weight %u) to buf zone %u (weight %u)",
0243         dmz_metadata_label(zmd), zrc->dev_idx,
0244         chunk, dzone->id, dmz_weight(dzone),
0245         bzone->id, dmz_weight(bzone));
0246 
0247     /* Flush data zone into the buffer zone */
0248     ret = dmz_reclaim_copy(zrc, dzone, bzone);
0249     if (ret < 0)
0250         return ret;
0251 
0252     dmz_lock_flush(zmd);
0253 
0254     /* Validate copied blocks */
0255     ret = dmz_merge_valid_blocks(zmd, dzone, bzone, 0);
0256     if (ret == 0) {
0257         /*
0258          * Free the data zone and remap the chunk to
0259          * the buffer zone.
0260          */
0261         dmz_invalidate_blocks(zmd, dzone, 0, dmz_zone_nr_blocks(zmd));
0262         dmz_lock_map(zmd);
0263         dmz_unmap_zone(zmd, bzone);
0264         dmz_unmap_zone(zmd, dzone);
0265         dmz_unlock_zone_reclaim(dzone);
0266         dmz_free_zone(zmd, dzone);
0267         dmz_map_zone(zmd, bzone, chunk);
0268         dmz_unlock_map(zmd);
0269     }
0270 
0271     dmz_unlock_flush(zmd);
0272 
0273     return ret;
0274 }
0275 
0276 /*
0277  * Move valid blocks of the random data zone dzone into a free sequential zone.
0278  * Once blocks are moved, remap the zone chunk to the sequential zone.
0279  */
0280 static int dmz_reclaim_rnd_data(struct dmz_reclaim *zrc, struct dm_zone *dzone)
0281 {
0282     unsigned int chunk = dzone->chunk;
0283     struct dm_zone *szone = NULL;
0284     struct dmz_metadata *zmd = zrc->metadata;
0285     int ret;
0286     int alloc_flags = DMZ_ALLOC_SEQ;
0287 
0288     /* Get a free random or sequential zone */
0289     dmz_lock_map(zmd);
0290 again:
0291     szone = dmz_alloc_zone(zmd, zrc->dev_idx,
0292                    alloc_flags | DMZ_ALLOC_RECLAIM);
0293     if (!szone && alloc_flags == DMZ_ALLOC_SEQ && dmz_nr_cache_zones(zmd)) {
0294         alloc_flags = DMZ_ALLOC_RND;
0295         goto again;
0296     }
0297     dmz_unlock_map(zmd);
0298     if (!szone)
0299         return -ENOSPC;
0300 
0301     DMDEBUG("(%s/%u): Chunk %u, move %s zone %u (weight %u) to %s zone %u",
0302         dmz_metadata_label(zmd), zrc->dev_idx, chunk,
0303         dmz_is_cache(dzone) ? "cache" : "rnd",
0304         dzone->id, dmz_weight(dzone),
0305         dmz_is_rnd(szone) ? "rnd" : "seq", szone->id);
0306 
0307     /* Flush the random data zone into the sequential zone */
0308     ret = dmz_reclaim_copy(zrc, dzone, szone);
0309 
0310     dmz_lock_flush(zmd);
0311 
0312     if (ret == 0) {
0313         /* Validate copied blocks */
0314         ret = dmz_copy_valid_blocks(zmd, dzone, szone);
0315     }
0316     if (ret) {
0317         /* Free the sequential zone */
0318         dmz_lock_map(zmd);
0319         dmz_free_zone(zmd, szone);
0320         dmz_unlock_map(zmd);
0321     } else {
0322         /* Free the data zone and remap the chunk */
0323         dmz_invalidate_blocks(zmd, dzone, 0, dmz_zone_nr_blocks(zmd));
0324         dmz_lock_map(zmd);
0325         dmz_unmap_zone(zmd, dzone);
0326         dmz_unlock_zone_reclaim(dzone);
0327         dmz_free_zone(zmd, dzone);
0328         dmz_map_zone(zmd, szone, chunk);
0329         dmz_unlock_map(zmd);
0330     }
0331 
0332     dmz_unlock_flush(zmd);
0333 
0334     return ret;
0335 }
0336 
0337 /*
0338  * Reclaim an empty zone.
0339  */
0340 static void dmz_reclaim_empty(struct dmz_reclaim *zrc, struct dm_zone *dzone)
0341 {
0342     struct dmz_metadata *zmd = zrc->metadata;
0343 
0344     dmz_lock_flush(zmd);
0345     dmz_lock_map(zmd);
0346     dmz_unmap_zone(zmd, dzone);
0347     dmz_unlock_zone_reclaim(dzone);
0348     dmz_free_zone(zmd, dzone);
0349     dmz_unlock_map(zmd);
0350     dmz_unlock_flush(zmd);
0351 }
0352 
0353 /*
0354  * Test if the target device is idle.
0355  */
0356 static inline int dmz_target_idle(struct dmz_reclaim *zrc)
0357 {
0358     return time_is_before_jiffies(zrc->atime + DMZ_IDLE_PERIOD);
0359 }
0360 
0361 /*
0362  * Find a candidate zone for reclaim and process it.
0363  */
0364 static int dmz_do_reclaim(struct dmz_reclaim *zrc)
0365 {
0366     struct dmz_metadata *zmd = zrc->metadata;
0367     struct dm_zone *dzone;
0368     struct dm_zone *rzone;
0369     unsigned long start;
0370     int ret;
0371 
0372     /* Get a data zone */
0373     dzone = dmz_get_zone_for_reclaim(zmd, zrc->dev_idx,
0374                      dmz_target_idle(zrc));
0375     if (!dzone) {
0376         DMDEBUG("(%s/%u): No zone found to reclaim",
0377             dmz_metadata_label(zmd), zrc->dev_idx);
0378         return -EBUSY;
0379     }
0380     rzone = dzone;
0381 
0382     start = jiffies;
0383     if (dmz_is_cache(dzone) || dmz_is_rnd(dzone)) {
0384         if (!dmz_weight(dzone)) {
0385             /* Empty zone */
0386             dmz_reclaim_empty(zrc, dzone);
0387             ret = 0;
0388         } else {
0389             /*
0390              * Reclaim the random data zone by moving its
0391              * valid data blocks to a free sequential zone.
0392              */
0393             ret = dmz_reclaim_rnd_data(zrc, dzone);
0394         }
0395     } else {
0396         struct dm_zone *bzone = dzone->bzone;
0397         sector_t chunk_block = 0;
0398 
0399         ret = dmz_first_valid_block(zmd, bzone, &chunk_block);
0400         if (ret < 0)
0401             goto out;
0402 
0403         if (ret == 0 || chunk_block >= dzone->wp_block) {
0404             /*
0405              * The buffer zone is empty or its valid blocks are
0406              * after the data zone write pointer.
0407              */
0408             ret = dmz_reclaim_buf(zrc, dzone);
0409             rzone = bzone;
0410         } else {
0411             /*
0412              * Reclaim the data zone by merging it into the
0413              * buffer zone so that the buffer zone itself can
0414              * be later reclaimed.
0415              */
0416             ret = dmz_reclaim_seq_data(zrc, dzone);
0417         }
0418     }
0419 out:
0420     if (ret) {
0421         if (ret == -EINTR)
0422             DMDEBUG("(%s/%u): reclaim zone %u interrupted",
0423                 dmz_metadata_label(zmd), zrc->dev_idx,
0424                 rzone->id);
0425         else
0426             DMDEBUG("(%s/%u): Failed to reclaim zone %u, err %d",
0427                 dmz_metadata_label(zmd), zrc->dev_idx,
0428                 rzone->id, ret);
0429         dmz_unlock_zone_reclaim(dzone);
0430         return ret;
0431     }
0432 
0433     ret = dmz_flush_metadata(zrc->metadata);
0434     if (ret) {
0435         DMDEBUG("(%s/%u): Metadata flush for zone %u failed, err %d",
0436             dmz_metadata_label(zmd), zrc->dev_idx, rzone->id, ret);
0437         return ret;
0438     }
0439 
0440     DMDEBUG("(%s/%u): Reclaimed zone %u in %u ms",
0441         dmz_metadata_label(zmd), zrc->dev_idx,
0442         rzone->id, jiffies_to_msecs(jiffies - start));
0443     return 0;
0444 }
0445 
0446 static unsigned int dmz_reclaim_percentage(struct dmz_reclaim *zrc)
0447 {
0448     struct dmz_metadata *zmd = zrc->metadata;
0449     unsigned int nr_cache = dmz_nr_cache_zones(zmd);
0450     unsigned int nr_unmap, nr_zones;
0451 
0452     if (nr_cache) {
0453         nr_zones = nr_cache;
0454         nr_unmap = dmz_nr_unmap_cache_zones(zmd);
0455     } else {
0456         nr_zones = dmz_nr_rnd_zones(zmd, zrc->dev_idx);
0457         nr_unmap = dmz_nr_unmap_rnd_zones(zmd, zrc->dev_idx);
0458     }
0459     if (nr_unmap <= 1)
0460         return 0;
0461     return nr_unmap * 100 / nr_zones;
0462 }
0463 
0464 /*
0465  * Test if reclaim is necessary.
0466  */
0467 static bool dmz_should_reclaim(struct dmz_reclaim *zrc, unsigned int p_unmap)
0468 {
0469     unsigned int nr_reclaim;
0470 
0471     nr_reclaim = dmz_nr_rnd_zones(zrc->metadata, zrc->dev_idx);
0472 
0473     if (dmz_nr_cache_zones(zrc->metadata)) {
0474         /*
0475          * The first device in a multi-device
0476          * setup only contains cache zones, so
0477          * never start reclaim there.
0478          */
0479         if (zrc->dev_idx == 0)
0480             return false;
0481         nr_reclaim += dmz_nr_cache_zones(zrc->metadata);
0482     }
0483 
0484     /* Reclaim when idle */
0485     if (dmz_target_idle(zrc) && nr_reclaim)
0486         return true;
0487 
0488     /* If there are still plenty of cache zones, do not reclaim */
0489     if (p_unmap >= DMZ_RECLAIM_HIGH_UNMAP_ZONES)
0490         return false;
0491 
0492     /*
0493      * If the percentage of unmapped cache zones is low,
0494      * reclaim even if the target is busy.
0495      */
0496     return p_unmap <= DMZ_RECLAIM_LOW_UNMAP_ZONES;
0497 }
0498 
0499 /*
0500  * Reclaim work function.
0501  */
0502 static void dmz_reclaim_work(struct work_struct *work)
0503 {
0504     struct dmz_reclaim *zrc = container_of(work, struct dmz_reclaim, work.work);
0505     struct dmz_metadata *zmd = zrc->metadata;
0506     unsigned int p_unmap;
0507     int ret;
0508 
0509     if (dmz_dev_is_dying(zmd))
0510         return;
0511 
0512     p_unmap = dmz_reclaim_percentage(zrc);
0513     if (!dmz_should_reclaim(zrc, p_unmap)) {
0514         mod_delayed_work(zrc->wq, &zrc->work, DMZ_IDLE_PERIOD);
0515         return;
0516     }
0517 
0518     /*
0519      * We need to start reclaiming random zones: set up zone copy
0520      * throttling to either go fast if we are very low on random zones
0521      * and slower if there are still some free random zones to avoid
0522      * as much as possible to negatively impact the user workload.
0523      */
0524     if (dmz_target_idle(zrc) || p_unmap < DMZ_RECLAIM_LOW_UNMAP_ZONES / 2) {
0525         /* Idle or very low percentage: go fast */
0526         zrc->kc_throttle.throttle = 100;
0527     } else {
0528         /* Busy but we still have some random zone: throttle */
0529         zrc->kc_throttle.throttle = min(75U, 100U - p_unmap / 2);
0530     }
0531 
0532     DMDEBUG("(%s/%u): Reclaim (%u): %s, %u%% free zones (%u/%u cache %u/%u random)",
0533         dmz_metadata_label(zmd), zrc->dev_idx,
0534         zrc->kc_throttle.throttle,
0535         (dmz_target_idle(zrc) ? "Idle" : "Busy"),
0536         p_unmap, dmz_nr_unmap_cache_zones(zmd),
0537         dmz_nr_cache_zones(zmd),
0538         dmz_nr_unmap_rnd_zones(zmd, zrc->dev_idx),
0539         dmz_nr_rnd_zones(zmd, zrc->dev_idx));
0540 
0541     ret = dmz_do_reclaim(zrc);
0542     if (ret && ret != -EINTR) {
0543         if (!dmz_check_dev(zmd))
0544             return;
0545     }
0546 
0547     dmz_schedule_reclaim(zrc);
0548 }
0549 
0550 /*
0551  * Initialize reclaim.
0552  */
0553 int dmz_ctr_reclaim(struct dmz_metadata *zmd,
0554             struct dmz_reclaim **reclaim, int idx)
0555 {
0556     struct dmz_reclaim *zrc;
0557     int ret;
0558 
0559     zrc = kzalloc(sizeof(struct dmz_reclaim), GFP_KERNEL);
0560     if (!zrc)
0561         return -ENOMEM;
0562 
0563     zrc->metadata = zmd;
0564     zrc->atime = jiffies;
0565     zrc->dev_idx = idx;
0566 
0567     /* Reclaim kcopyd client */
0568     zrc->kc = dm_kcopyd_client_create(&zrc->kc_throttle);
0569     if (IS_ERR(zrc->kc)) {
0570         ret = PTR_ERR(zrc->kc);
0571         zrc->kc = NULL;
0572         goto err;
0573     }
0574 
0575     /* Reclaim work */
0576     INIT_DELAYED_WORK(&zrc->work, dmz_reclaim_work);
0577     zrc->wq = alloc_ordered_workqueue("dmz_rwq_%s_%d", WQ_MEM_RECLAIM,
0578                       dmz_metadata_label(zmd), idx);
0579     if (!zrc->wq) {
0580         ret = -ENOMEM;
0581         goto err;
0582     }
0583 
0584     *reclaim = zrc;
0585     queue_delayed_work(zrc->wq, &zrc->work, 0);
0586 
0587     return 0;
0588 err:
0589     if (zrc->kc)
0590         dm_kcopyd_client_destroy(zrc->kc);
0591     kfree(zrc);
0592 
0593     return ret;
0594 }
0595 
0596 /*
0597  * Terminate reclaim.
0598  */
0599 void dmz_dtr_reclaim(struct dmz_reclaim *zrc)
0600 {
0601     cancel_delayed_work_sync(&zrc->work);
0602     destroy_workqueue(zrc->wq);
0603     dm_kcopyd_client_destroy(zrc->kc);
0604     kfree(zrc);
0605 }
0606 
0607 /*
0608  * Suspend reclaim.
0609  */
0610 void dmz_suspend_reclaim(struct dmz_reclaim *zrc)
0611 {
0612     cancel_delayed_work_sync(&zrc->work);
0613 }
0614 
0615 /*
0616  * Resume reclaim.
0617  */
0618 void dmz_resume_reclaim(struct dmz_reclaim *zrc)
0619 {
0620     queue_delayed_work(zrc->wq, &zrc->work, DMZ_IDLE_PERIOD);
0621 }
0622 
0623 /*
0624  * BIO accounting.
0625  */
0626 void dmz_reclaim_bio_acc(struct dmz_reclaim *zrc)
0627 {
0628     zrc->atime = jiffies;
0629 }
0630 
0631 /*
0632  * Start reclaim if necessary.
0633  */
0634 void dmz_schedule_reclaim(struct dmz_reclaim *zrc)
0635 {
0636     unsigned int p_unmap = dmz_reclaim_percentage(zrc);
0637 
0638     if (dmz_should_reclaim(zrc, p_unmap))
0639         mod_delayed_work(zrc->wq, &zrc->work, 0);
0640 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!