LCOV - code coverage report
Current view: top level - block - mq-deadline.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 2 260 0.8 %
Date: 2023-07-19 18:55:55 Functions: 1 37 2.7 %

          Line data    Source code
       1             : // SPDX-License-Identifier: GPL-2.0
       2             : /*
       3             :  *  MQ Deadline i/o scheduler - adaptation of the legacy deadline scheduler,
       4             :  *  for the blk-mq scheduling framework
       5             :  *
       6             :  *  Copyright (C) 2016 Jens Axboe <axboe@kernel.dk>
       7             :  */
       8             : #include <linux/kernel.h>
       9             : #include <linux/fs.h>
      10             : #include <linux/blkdev.h>
      11             : #include <linux/bio.h>
      12             : #include <linux/module.h>
      13             : #include <linux/slab.h>
      14             : #include <linux/init.h>
      15             : #include <linux/compiler.h>
      16             : #include <linux/rbtree.h>
      17             : #include <linux/sbitmap.h>
      18             : 
      19             : #include <trace/events/block.h>
      20             : 
      21             : #include "elevator.h"
      22             : #include "blk.h"
      23             : #include "blk-mq.h"
      24             : #include "blk-mq-debugfs.h"
      25             : #include "blk-mq-sched.h"
      26             : 
      27             : /*
      28             :  * See Documentation/block/deadline-iosched.rst
      29             :  */
      30             : static const int read_expire = HZ / 2;  /* max time before a read is submitted. */
      31             : static const int write_expire = 5 * HZ; /* ditto for writes, these limits are SOFT! */
      32             : /*
      33             :  * Time after which to dispatch lower priority requests even if higher
      34             :  * priority requests are pending.
      35             :  */
      36             : static const int prio_aging_expire = 10 * HZ;
      37             : static const int writes_starved = 2;    /* max times reads can starve a write */
      38             : static const int fifo_batch = 16;       /* # of sequential requests treated as one
      39             :                                      by the above parameters. For throughput. */
      40             : 
      41             : enum dd_data_dir {
      42             :         DD_READ         = READ,
      43             :         DD_WRITE        = WRITE,
      44             : };
      45             : 
      46             : enum { DD_DIR_COUNT = 2 };
      47             : 
      48             : enum dd_prio {
      49             :         DD_RT_PRIO      = 0,
      50             :         DD_BE_PRIO      = 1,
      51             :         DD_IDLE_PRIO    = 2,
      52             :         DD_PRIO_MAX     = 2,
      53             : };
      54             : 
      55             : enum { DD_PRIO_COUNT = 3 };
      56             : 
      57             : /*
      58             :  * I/O statistics per I/O priority. It is fine if these counters overflow.
      59             :  * What matters is that these counters are at least as wide as
      60             :  * log2(max_outstanding_requests).
      61             :  */
      62             : struct io_stats_per_prio {
      63             :         uint32_t inserted;
      64             :         uint32_t merged;
      65             :         uint32_t dispatched;
      66             :         atomic_t completed;
      67             : };
      68             : 
      69             : /*
      70             :  * Deadline scheduler data per I/O priority (enum dd_prio). Requests are
      71             :  * present on both sort_list[] and fifo_list[].
      72             :  */
      73             : struct dd_per_prio {
      74             :         struct list_head dispatch;
      75             :         struct rb_root sort_list[DD_DIR_COUNT];
      76             :         struct list_head fifo_list[DD_DIR_COUNT];
      77             :         /* Next request in FIFO order. Read, write or both are NULL. */
      78             :         struct request *next_rq[DD_DIR_COUNT];
      79             :         struct io_stats_per_prio stats;
      80             : };
      81             : 
      82             : struct deadline_data {
      83             :         /*
      84             :          * run time data
      85             :          */
      86             : 
      87             :         struct dd_per_prio per_prio[DD_PRIO_COUNT];
      88             : 
      89             :         /* Data direction of latest dispatched request. */
      90             :         enum dd_data_dir last_dir;
      91             :         unsigned int batching;          /* number of sequential requests made */
      92             :         unsigned int starved;           /* times reads have starved writes */
      93             : 
      94             :         /*
      95             :          * settings that change how the i/o scheduler behaves
      96             :          */
      97             :         int fifo_expire[DD_DIR_COUNT];
      98             :         int fifo_batch;
      99             :         int writes_starved;
     100             :         int front_merges;
     101             :         u32 async_depth;
     102             :         int prio_aging_expire;
     103             : 
     104             :         spinlock_t lock;
     105             :         spinlock_t zone_lock;
     106             : };
     107             : 
     108             : /* Maps an I/O priority class to a deadline scheduler priority. */
     109             : static const enum dd_prio ioprio_class_to_prio[] = {
     110             :         [IOPRIO_CLASS_NONE]     = DD_BE_PRIO,
     111             :         [IOPRIO_CLASS_RT]       = DD_RT_PRIO,
     112             :         [IOPRIO_CLASS_BE]       = DD_BE_PRIO,
     113             :         [IOPRIO_CLASS_IDLE]     = DD_IDLE_PRIO,
     114             : };
     115             : 
     116             : static inline struct rb_root *
     117             : deadline_rb_root(struct dd_per_prio *per_prio, struct request *rq)
     118             : {
     119           0 :         return &per_prio->sort_list[rq_data_dir(rq)];
     120             : }
     121             : 
     122             : /*
     123             :  * Returns the I/O priority class (IOPRIO_CLASS_*) that has been assigned to a
     124             :  * request.
     125             :  */
     126             : static u8 dd_rq_ioclass(struct request *rq)
     127             : {
     128           0 :         return IOPRIO_PRIO_CLASS(req_get_ioprio(rq));
     129             : }
     130             : 
     131             : /*
     132             :  * get the request before `rq' in sector-sorted order
     133             :  */
     134             : static inline struct request *
     135             : deadline_earlier_request(struct request *rq)
     136             : {
     137             :         struct rb_node *node = rb_prev(&rq->rb_node);
     138             : 
     139             :         if (node)
     140             :                 return rb_entry_rq(node);
     141             : 
     142             :         return NULL;
     143             : }
     144             : 
     145             : /*
     146             :  * get the request after `rq' in sector-sorted order
     147             :  */
     148             : static inline struct request *
     149             : deadline_latter_request(struct request *rq)
     150             : {
     151           0 :         struct rb_node *node = rb_next(&rq->rb_node);
     152             : 
     153           0 :         if (node)
     154           0 :                 return rb_entry_rq(node);
     155             : 
     156             :         return NULL;
     157             : }
     158             : 
     159             : static void
     160             : deadline_add_rq_rb(struct dd_per_prio *per_prio, struct request *rq)
     161             : {
     162           0 :         struct rb_root *root = deadline_rb_root(per_prio, rq);
     163             : 
     164           0 :         elv_rb_add(root, rq);
     165             : }
     166             : 
     167             : static inline void
     168           0 : deadline_del_rq_rb(struct dd_per_prio *per_prio, struct request *rq)
     169             : {
     170           0 :         const enum dd_data_dir data_dir = rq_data_dir(rq);
     171             : 
     172           0 :         if (per_prio->next_rq[data_dir] == rq)
     173           0 :                 per_prio->next_rq[data_dir] = deadline_latter_request(rq);
     174             : 
     175           0 :         elv_rb_del(deadline_rb_root(per_prio, rq), rq);
     176           0 : }
     177             : 
     178             : /*
     179             :  * remove rq from rbtree and fifo.
     180             :  */
     181           0 : static void deadline_remove_request(struct request_queue *q,
     182             :                                     struct dd_per_prio *per_prio,
     183             :                                     struct request *rq)
     184             : {
     185           0 :         list_del_init(&rq->queuelist);
     186             : 
     187             :         /*
     188             :          * We might not be on the rbtree, if we are doing an insert merge
     189             :          */
     190           0 :         if (!RB_EMPTY_NODE(&rq->rb_node))
     191           0 :                 deadline_del_rq_rb(per_prio, rq);
     192             : 
     193           0 :         elv_rqhash_del(q, rq);
     194           0 :         if (q->last_merge == rq)
     195           0 :                 q->last_merge = NULL;
     196           0 : }
     197             : 
     198           0 : static void dd_request_merged(struct request_queue *q, struct request *req,
     199             :                               enum elv_merge type)
     200             : {
     201           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     202           0 :         const u8 ioprio_class = dd_rq_ioclass(req);
     203           0 :         const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
     204           0 :         struct dd_per_prio *per_prio = &dd->per_prio[prio];
     205             : 
     206             :         /*
     207             :          * if the merge was a front merge, we need to reposition request
     208             :          */
     209           0 :         if (type == ELEVATOR_FRONT_MERGE) {
     210           0 :                 elv_rb_del(deadline_rb_root(per_prio, req), req);
     211             :                 deadline_add_rq_rb(per_prio, req);
     212             :         }
     213           0 : }
     214             : 
     215             : /*
     216             :  * Callback function that is invoked after @next has been merged into @req.
     217             :  */
     218           0 : static void dd_merged_requests(struct request_queue *q, struct request *req,
     219             :                                struct request *next)
     220             : {
     221           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     222           0 :         const u8 ioprio_class = dd_rq_ioclass(next);
     223           0 :         const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
     224             : 
     225             :         lockdep_assert_held(&dd->lock);
     226             : 
     227           0 :         dd->per_prio[prio].stats.merged++;
     228             : 
     229             :         /*
     230             :          * if next expires before rq, assign its expire time to rq
     231             :          * and move into next position (next will be deleted) in fifo
     232             :          */
     233           0 :         if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) {
     234           0 :                 if (time_before((unsigned long)next->fifo_time,
     235             :                                 (unsigned long)req->fifo_time)) {
     236           0 :                         list_move(&req->queuelist, &next->queuelist);
     237           0 :                         req->fifo_time = next->fifo_time;
     238             :                 }
     239             :         }
     240             : 
     241             :         /*
     242             :          * kill knowledge of next, this one is a goner
     243             :          */
     244           0 :         deadline_remove_request(q, &dd->per_prio[prio], next);
     245           0 : }
     246             : 
     247             : /*
     248             :  * move an entry to dispatch queue
     249             :  */
     250             : static void
     251           0 : deadline_move_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
     252             :                       struct request *rq)
     253             : {
     254           0 :         const enum dd_data_dir data_dir = rq_data_dir(rq);
     255             : 
     256           0 :         per_prio->next_rq[data_dir] = deadline_latter_request(rq);
     257             : 
     258             :         /*
     259             :          * take it off the sort and fifo list
     260             :          */
     261           0 :         deadline_remove_request(rq->q, per_prio, rq);
     262           0 : }
     263             : 
     264             : /* Number of requests queued for a given priority level. */
     265             : static u32 dd_queued(struct deadline_data *dd, enum dd_prio prio)
     266             : {
     267           0 :         const struct io_stats_per_prio *stats = &dd->per_prio[prio].stats;
     268             : 
     269             :         lockdep_assert_held(&dd->lock);
     270             : 
     271           0 :         return stats->inserted - atomic_read(&stats->completed);
     272             : }
     273             : 
     274             : /*
     275             :  * deadline_check_fifo returns 0 if there are no expired requests on the fifo,
     276             :  * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir])
     277             :  */
     278             : static inline int deadline_check_fifo(struct dd_per_prio *per_prio,
     279             :                                       enum dd_data_dir data_dir)
     280             : {
     281           0 :         struct request *rq = rq_entry_fifo(per_prio->fifo_list[data_dir].next);
     282             : 
     283             :         /*
     284             :          * rq is expired!
     285             :          */
     286           0 :         if (time_after_eq(jiffies, (unsigned long)rq->fifo_time))
     287             :                 return 1;
     288             : 
     289             :         return 0;
     290             : }
     291             : 
     292             : /*
     293             :  * Check if rq has a sequential request preceding it.
     294             :  */
     295             : static bool deadline_is_seq_write(struct deadline_data *dd, struct request *rq)
     296             : {
     297             :         struct request *prev = deadline_earlier_request(rq);
     298             : 
     299             :         if (!prev)
     300             :                 return false;
     301             : 
     302             :         return blk_rq_pos(prev) + blk_rq_sectors(prev) == blk_rq_pos(rq);
     303             : }
     304             : 
     305             : /*
     306             :  * Skip all write requests that are sequential from @rq, even if we cross
     307             :  * a zone boundary.
     308             :  */
     309             : static struct request *deadline_skip_seq_writes(struct deadline_data *dd,
     310             :                                                 struct request *rq)
     311             : {
     312             :         sector_t pos = blk_rq_pos(rq);
     313             :         sector_t skipped_sectors = 0;
     314             : 
     315             :         while (rq) {
     316             :                 if (blk_rq_pos(rq) != pos + skipped_sectors)
     317             :                         break;
     318             :                 skipped_sectors += blk_rq_sectors(rq);
     319             :                 rq = deadline_latter_request(rq);
     320             :         }
     321             : 
     322             :         return rq;
     323             : }
     324             : 
     325             : /*
     326             :  * For the specified data direction, return the next request to
     327             :  * dispatch using arrival ordered lists.
     328             :  */
     329             : static struct request *
     330             : deadline_fifo_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
     331             :                       enum dd_data_dir data_dir)
     332             : {
     333             :         struct request *rq;
     334             :         unsigned long flags;
     335             : 
     336           0 :         if (list_empty(&per_prio->fifo_list[data_dir]))
     337             :                 return NULL;
     338             : 
     339           0 :         rq = rq_entry_fifo(per_prio->fifo_list[data_dir].next);
     340             :         if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
     341             :                 return rq;
     342             : 
     343             :         /*
     344             :          * Look for a write request that can be dispatched, that is one with
     345             :          * an unlocked target zone. For some HDDs, breaking a sequential
     346             :          * write stream can lead to lower throughput, so make sure to preserve
     347             :          * sequential write streams, even if that stream crosses into the next
     348             :          * zones and these zones are unlocked.
     349             :          */
     350             :         spin_lock_irqsave(&dd->zone_lock, flags);
     351             :         list_for_each_entry(rq, &per_prio->fifo_list[DD_WRITE], queuelist) {
     352             :                 if (blk_req_can_dispatch_to_zone(rq) &&
     353             :                     (blk_queue_nonrot(rq->q) ||
     354             :                      !deadline_is_seq_write(dd, rq)))
     355             :                         goto out;
     356             :         }
     357             :         rq = NULL;
     358             : out:
     359             :         spin_unlock_irqrestore(&dd->zone_lock, flags);
     360             : 
     361             :         return rq;
     362             : }
     363             : 
     364             : /*
     365             :  * For the specified data direction, return the next request to
     366             :  * dispatch using sector position sorted lists.
     367             :  */
     368             : static struct request *
     369             : deadline_next_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
     370             :                       enum dd_data_dir data_dir)
     371             : {
     372             :         struct request *rq;
     373             :         unsigned long flags;
     374             : 
     375           0 :         rq = per_prio->next_rq[data_dir];
     376           0 :         if (!rq)
     377             :                 return NULL;
     378             : 
     379             :         if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
     380             :                 return rq;
     381             : 
     382             :         /*
     383             :          * Look for a write request that can be dispatched, that is one with
     384             :          * an unlocked target zone. For some HDDs, breaking a sequential
     385             :          * write stream can lead to lower throughput, so make sure to preserve
     386             :          * sequential write streams, even if that stream crosses into the next
     387             :          * zones and these zones are unlocked.
     388             :          */
     389             :         spin_lock_irqsave(&dd->zone_lock, flags);
     390             :         while (rq) {
     391             :                 if (blk_req_can_dispatch_to_zone(rq))
     392             :                         break;
     393             :                 if (blk_queue_nonrot(rq->q))
     394             :                         rq = deadline_latter_request(rq);
     395             :                 else
     396             :                         rq = deadline_skip_seq_writes(dd, rq);
     397             :         }
     398             :         spin_unlock_irqrestore(&dd->zone_lock, flags);
     399             : 
     400             :         return rq;
     401             : }
     402             : 
     403             : /*
     404             :  * Returns true if and only if @rq started after @latest_start where
     405             :  * @latest_start is in jiffies.
     406             :  */
     407             : static bool started_after(struct deadline_data *dd, struct request *rq,
     408             :                           unsigned long latest_start)
     409             : {
     410           0 :         unsigned long start_time = (unsigned long)rq->fifo_time;
     411             : 
     412           0 :         start_time -= dd->fifo_expire[rq_data_dir(rq)];
     413             : 
     414           0 :         return time_after(start_time, latest_start);
     415             : }
     416             : 
     417             : /*
     418             :  * deadline_dispatch_requests selects the best request according to
     419             :  * read/write expire, fifo_batch, etc and with a start time <= @latest_start.
     420             :  */
     421           0 : static struct request *__dd_dispatch_request(struct deadline_data *dd,
     422             :                                              struct dd_per_prio *per_prio,
     423             :                                              unsigned long latest_start)
     424             : {
     425             :         struct request *rq, *next_rq;
     426             :         enum dd_data_dir data_dir;
     427             :         enum dd_prio prio;
     428             :         u8 ioprio_class;
     429             : 
     430             :         lockdep_assert_held(&dd->lock);
     431             : 
     432           0 :         if (!list_empty(&per_prio->dispatch)) {
     433           0 :                 rq = list_first_entry(&per_prio->dispatch, struct request,
     434             :                                       queuelist);
     435           0 :                 if (started_after(dd, rq, latest_start))
     436             :                         return NULL;
     437           0 :                 list_del_init(&rq->queuelist);
     438             :                 goto done;
     439             :         }
     440             : 
     441             :         /*
     442             :          * batches are currently reads XOR writes
     443             :          */
     444           0 :         rq = deadline_next_request(dd, per_prio, dd->last_dir);
     445           0 :         if (rq && dd->batching < dd->fifo_batch)
     446             :                 /* we have a next request are still entitled to batch */
     447             :                 goto dispatch_request;
     448             : 
     449             :         /*
     450             :          * at this point we are not running a batch. select the appropriate
     451             :          * data direction (read / write)
     452             :          */
     453             : 
     454           0 :         if (!list_empty(&per_prio->fifo_list[DD_READ])) {
     455           0 :                 BUG_ON(RB_EMPTY_ROOT(&per_prio->sort_list[DD_READ]));
     456             : 
     457           0 :                 if (deadline_fifo_request(dd, per_prio, DD_WRITE) &&
     458           0 :                     (dd->starved++ >= dd->writes_starved))
     459             :                         goto dispatch_writes;
     460             : 
     461             :                 data_dir = DD_READ;
     462             : 
     463             :                 goto dispatch_find_request;
     464             :         }
     465             : 
     466             :         /*
     467             :          * there are either no reads or writes have been starved
     468             :          */
     469             : 
     470           0 :         if (!list_empty(&per_prio->fifo_list[DD_WRITE])) {
     471             : dispatch_writes:
     472           0 :                 BUG_ON(RB_EMPTY_ROOT(&per_prio->sort_list[DD_WRITE]));
     473             : 
     474           0 :                 dd->starved = 0;
     475             : 
     476           0 :                 data_dir = DD_WRITE;
     477             : 
     478           0 :                 goto dispatch_find_request;
     479             :         }
     480             : 
     481             :         return NULL;
     482             : 
     483             : dispatch_find_request:
     484             :         /*
     485             :          * we are not running a batch, find best request for selected data_dir
     486             :          */
     487           0 :         next_rq = deadline_next_request(dd, per_prio, data_dir);
     488           0 :         if (deadline_check_fifo(per_prio, data_dir) || !next_rq) {
     489             :                 /*
     490             :                  * A deadline has expired, the last request was in the other
     491             :                  * direction, or we have run out of higher-sectored requests.
     492             :                  * Start again from the request with the earliest expiry time.
     493             :                  */
     494           0 :                 rq = deadline_fifo_request(dd, per_prio, data_dir);
     495             :         } else {
     496             :                 /*
     497             :                  * The last req was the same dir and we have a next request in
     498             :                  * sort order. No expired requests so continue on from here.
     499             :                  */
     500             :                 rq = next_rq;
     501             :         }
     502             : 
     503             :         /*
     504             :          * For a zoned block device, if we only have writes queued and none of
     505             :          * them can be dispatched, rq will be NULL.
     506             :          */
     507           0 :         if (!rq)
     508             :                 return NULL;
     509             : 
     510           0 :         dd->last_dir = data_dir;
     511           0 :         dd->batching = 0;
     512             : 
     513             : dispatch_request:
     514           0 :         if (started_after(dd, rq, latest_start))
     515             :                 return NULL;
     516             : 
     517             :         /*
     518             :          * rq is the selected appropriate request.
     519             :          */
     520           0 :         dd->batching++;
     521           0 :         deadline_move_request(dd, per_prio, rq);
     522             : done:
     523           0 :         ioprio_class = dd_rq_ioclass(rq);
     524           0 :         prio = ioprio_class_to_prio[ioprio_class];
     525           0 :         dd->per_prio[prio].stats.dispatched++;
     526             :         /*
     527             :          * If the request needs its target zone locked, do it.
     528             :          */
     529           0 :         blk_req_zone_write_lock(rq);
     530           0 :         rq->rq_flags |= RQF_STARTED;
     531           0 :         return rq;
     532             : }
     533             : 
     534             : /*
     535             :  * Check whether there are any requests with priority other than DD_RT_PRIO
     536             :  * that were inserted more than prio_aging_expire jiffies ago.
     537             :  */
     538           0 : static struct request *dd_dispatch_prio_aged_requests(struct deadline_data *dd,
     539             :                                                       unsigned long now)
     540             : {
     541             :         struct request *rq;
     542             :         enum dd_prio prio;
     543             :         int prio_cnt;
     544             : 
     545             :         lockdep_assert_held(&dd->lock);
     546             : 
     547           0 :         prio_cnt = !!dd_queued(dd, DD_RT_PRIO) + !!dd_queued(dd, DD_BE_PRIO) +
     548           0 :                    !!dd_queued(dd, DD_IDLE_PRIO);
     549           0 :         if (prio_cnt < 2)
     550             :                 return NULL;
     551             : 
     552           0 :         for (prio = DD_BE_PRIO; prio <= DD_PRIO_MAX; prio++) {
     553           0 :                 rq = __dd_dispatch_request(dd, &dd->per_prio[prio],
     554           0 :                                            now - dd->prio_aging_expire);
     555           0 :                 if (rq)
     556             :                         return rq;
     557             :         }
     558             : 
     559             :         return NULL;
     560             : }
     561             : 
     562             : /*
     563             :  * Called from blk_mq_run_hw_queue() -> __blk_mq_sched_dispatch_requests().
     564             :  *
     565             :  * One confusing aspect here is that we get called for a specific
     566             :  * hardware queue, but we may return a request that is for a
     567             :  * different hardware queue. This is because mq-deadline has shared
     568             :  * state for all hardware queues, in terms of sorting, FIFOs, etc.
     569             :  */
     570           0 : static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
     571             : {
     572           0 :         struct deadline_data *dd = hctx->queue->elevator->elevator_data;
     573           0 :         const unsigned long now = jiffies;
     574             :         struct request *rq;
     575             :         enum dd_prio prio;
     576             : 
     577           0 :         spin_lock(&dd->lock);
     578           0 :         rq = dd_dispatch_prio_aged_requests(dd, now);
     579           0 :         if (rq)
     580             :                 goto unlock;
     581             : 
     582             :         /*
     583             :          * Next, dispatch requests in priority order. Ignore lower priority
     584             :          * requests if any higher priority requests are pending.
     585             :          */
     586           0 :         for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
     587           0 :                 rq = __dd_dispatch_request(dd, &dd->per_prio[prio], now);
     588           0 :                 if (rq || dd_queued(dd, prio))
     589             :                         break;
     590             :         }
     591             : 
     592             : unlock:
     593           0 :         spin_unlock(&dd->lock);
     594             : 
     595           0 :         return rq;
     596             : }
     597             : 
     598             : /*
     599             :  * Called by __blk_mq_alloc_request(). The shallow_depth value set by this
     600             :  * function is used by __blk_mq_get_tag().
     601             :  */
     602           0 : static void dd_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
     603             : {
     604           0 :         struct deadline_data *dd = data->q->elevator->elevator_data;
     605             : 
     606             :         /* Do not throttle synchronous reads. */
     607           0 :         if (op_is_sync(opf) && !op_is_write(opf))
     608             :                 return;
     609             : 
     610             :         /*
     611             :          * Throttle asynchronous requests and writes such that these requests
     612             :          * do not block the allocation of synchronous requests.
     613             :          */
     614           0 :         data->shallow_depth = dd->async_depth;
     615             : }
     616             : 
     617             : /* Called by blk_mq_update_nr_requests(). */
     618           0 : static void dd_depth_updated(struct blk_mq_hw_ctx *hctx)
     619             : {
     620           0 :         struct request_queue *q = hctx->queue;
     621           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     622           0 :         struct blk_mq_tags *tags = hctx->sched_tags;
     623             : 
     624           0 :         dd->async_depth = max(1UL, 3 * q->nr_requests / 4);
     625             : 
     626           0 :         sbitmap_queue_min_shallow_depth(&tags->bitmap_tags, dd->async_depth);
     627           0 : }
     628             : 
     629             : /* Called by blk_mq_init_hctx() and blk_mq_init_sched(). */
     630           0 : static int dd_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
     631             : {
     632           0 :         dd_depth_updated(hctx);
     633           0 :         return 0;
     634             : }
     635             : 
     636           0 : static void dd_exit_sched(struct elevator_queue *e)
     637             : {
     638           0 :         struct deadline_data *dd = e->elevator_data;
     639             :         enum dd_prio prio;
     640             : 
     641           0 :         for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
     642           0 :                 struct dd_per_prio *per_prio = &dd->per_prio[prio];
     643           0 :                 const struct io_stats_per_prio *stats = &per_prio->stats;
     644             :                 uint32_t queued;
     645             : 
     646           0 :                 WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_READ]));
     647           0 :                 WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_WRITE]));
     648             : 
     649           0 :                 spin_lock(&dd->lock);
     650           0 :                 queued = dd_queued(dd, prio);
     651           0 :                 spin_unlock(&dd->lock);
     652             : 
     653           0 :                 WARN_ONCE(queued != 0,
     654             :                           "statistics for priority %d: i %u m %u d %u c %u\n",
     655             :                           prio, stats->inserted, stats->merged,
     656             :                           stats->dispatched, atomic_read(&stats->completed));
     657             :         }
     658             : 
     659           0 :         kfree(dd);
     660           0 : }
     661             : 
     662             : /*
     663             :  * initialize elevator private data (deadline_data).
     664             :  */
     665           0 : static int dd_init_sched(struct request_queue *q, struct elevator_type *e)
     666             : {
     667             :         struct deadline_data *dd;
     668             :         struct elevator_queue *eq;
     669             :         enum dd_prio prio;
     670           0 :         int ret = -ENOMEM;
     671             : 
     672           0 :         eq = elevator_alloc(q, e);
     673           0 :         if (!eq)
     674             :                 return ret;
     675             : 
     676           0 :         dd = kzalloc_node(sizeof(*dd), GFP_KERNEL, q->node);
     677           0 :         if (!dd)
     678             :                 goto put_eq;
     679             : 
     680           0 :         eq->elevator_data = dd;
     681             : 
     682           0 :         for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
     683           0 :                 struct dd_per_prio *per_prio = &dd->per_prio[prio];
     684             : 
     685           0 :                 INIT_LIST_HEAD(&per_prio->dispatch);
     686           0 :                 INIT_LIST_HEAD(&per_prio->fifo_list[DD_READ]);
     687           0 :                 INIT_LIST_HEAD(&per_prio->fifo_list[DD_WRITE]);
     688           0 :                 per_prio->sort_list[DD_READ] = RB_ROOT;
     689           0 :                 per_prio->sort_list[DD_WRITE] = RB_ROOT;
     690             :         }
     691           0 :         dd->fifo_expire[DD_READ] = read_expire;
     692           0 :         dd->fifo_expire[DD_WRITE] = write_expire;
     693           0 :         dd->writes_starved = writes_starved;
     694           0 :         dd->front_merges = 1;
     695           0 :         dd->last_dir = DD_WRITE;
     696           0 :         dd->fifo_batch = fifo_batch;
     697           0 :         dd->prio_aging_expire = prio_aging_expire;
     698           0 :         spin_lock_init(&dd->lock);
     699           0 :         spin_lock_init(&dd->zone_lock);
     700             : 
     701             :         /* We dispatch from request queue wide instead of hw queue */
     702           0 :         blk_queue_flag_set(QUEUE_FLAG_SQ_SCHED, q);
     703             : 
     704           0 :         q->elevator = eq;
     705           0 :         return 0;
     706             : 
     707             : put_eq:
     708           0 :         kobject_put(&eq->kobj);
     709           0 :         return ret;
     710             : }
     711             : 
     712             : /*
     713             :  * Try to merge @bio into an existing request. If @bio has been merged into
     714             :  * an existing request, store the pointer to that request into *@rq.
     715             :  */
     716           0 : static int dd_request_merge(struct request_queue *q, struct request **rq,
     717             :                             struct bio *bio)
     718             : {
     719           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     720           0 :         const u8 ioprio_class = IOPRIO_PRIO_CLASS(bio->bi_ioprio);
     721           0 :         const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
     722           0 :         struct dd_per_prio *per_prio = &dd->per_prio[prio];
     723           0 :         sector_t sector = bio_end_sector(bio);
     724             :         struct request *__rq;
     725             : 
     726           0 :         if (!dd->front_merges)
     727             :                 return ELEVATOR_NO_MERGE;
     728             : 
     729           0 :         __rq = elv_rb_find(&per_prio->sort_list[bio_data_dir(bio)], sector);
     730           0 :         if (__rq) {
     731           0 :                 BUG_ON(sector != blk_rq_pos(__rq));
     732             : 
     733           0 :                 if (elv_bio_merge_ok(__rq, bio)) {
     734           0 :                         *rq = __rq;
     735           0 :                         if (blk_discard_mergable(__rq))
     736             :                                 return ELEVATOR_DISCARD_MERGE;
     737           0 :                         return ELEVATOR_FRONT_MERGE;
     738             :                 }
     739             :         }
     740             : 
     741             :         return ELEVATOR_NO_MERGE;
     742             : }
     743             : 
     744             : /*
     745             :  * Attempt to merge a bio into an existing request. This function is called
     746             :  * before @bio is associated with a request.
     747             :  */
     748           0 : static bool dd_bio_merge(struct request_queue *q, struct bio *bio,
     749             :                 unsigned int nr_segs)
     750             : {
     751           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     752           0 :         struct request *free = NULL;
     753             :         bool ret;
     754             : 
     755           0 :         spin_lock(&dd->lock);
     756           0 :         ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
     757           0 :         spin_unlock(&dd->lock);
     758             : 
     759           0 :         if (free)
     760           0 :                 blk_mq_free_request(free);
     761             : 
     762           0 :         return ret;
     763             : }
     764             : 
     765             : /*
     766             :  * add rq to rbtree and fifo
     767             :  */
     768           0 : static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
     769             :                               blk_insert_t flags)
     770             : {
     771           0 :         struct request_queue *q = hctx->queue;
     772           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     773           0 :         const enum dd_data_dir data_dir = rq_data_dir(rq);
     774           0 :         u16 ioprio = req_get_ioprio(rq);
     775           0 :         u8 ioprio_class = IOPRIO_PRIO_CLASS(ioprio);
     776             :         struct dd_per_prio *per_prio;
     777             :         enum dd_prio prio;
     778           0 :         LIST_HEAD(free);
     779             : 
     780             :         lockdep_assert_held(&dd->lock);
     781             : 
     782             :         /*
     783             :          * This may be a requeue of a write request that has locked its
     784             :          * target zone. If it is the case, this releases the zone lock.
     785             :          */
     786           0 :         blk_req_zone_write_unlock(rq);
     787             : 
     788           0 :         prio = ioprio_class_to_prio[ioprio_class];
     789           0 :         per_prio = &dd->per_prio[prio];
     790           0 :         if (!rq->elv.priv[0]) {
     791           0 :                 per_prio->stats.inserted++;
     792           0 :                 rq->elv.priv[0] = (void *)(uintptr_t)1;
     793             :         }
     794             : 
     795           0 :         if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
     796           0 :                 blk_mq_free_requests(&free);
     797           0 :                 return;
     798             :         }
     799             : 
     800           0 :         trace_block_rq_insert(rq);
     801             : 
     802           0 :         if (flags & BLK_MQ_INSERT_AT_HEAD) {
     803           0 :                 list_add(&rq->queuelist, &per_prio->dispatch);
     804           0 :                 rq->fifo_time = jiffies;
     805             :         } else {
     806           0 :                 deadline_add_rq_rb(per_prio, rq);
     807             : 
     808           0 :                 if (rq_mergeable(rq)) {
     809           0 :                         elv_rqhash_add(q, rq);
     810           0 :                         if (!q->last_merge)
     811           0 :                                 q->last_merge = rq;
     812             :                 }
     813             : 
     814             :                 /*
     815             :                  * set expire time and add to fifo list
     816             :                  */
     817           0 :                 rq->fifo_time = jiffies + dd->fifo_expire[data_dir];
     818           0 :                 list_add_tail(&rq->queuelist, &per_prio->fifo_list[data_dir]);
     819             :         }
     820             : }
     821             : 
     822             : /*
     823             :  * Called from blk_mq_insert_request() or blk_mq_dispatch_plug_list().
     824             :  */
     825           0 : static void dd_insert_requests(struct blk_mq_hw_ctx *hctx,
     826             :                                struct list_head *list,
     827             :                                blk_insert_t flags)
     828             : {
     829           0 :         struct request_queue *q = hctx->queue;
     830           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     831             : 
     832           0 :         spin_lock(&dd->lock);
     833           0 :         while (!list_empty(list)) {
     834             :                 struct request *rq;
     835             : 
     836           0 :                 rq = list_first_entry(list, struct request, queuelist);
     837           0 :                 list_del_init(&rq->queuelist);
     838           0 :                 dd_insert_request(hctx, rq, flags);
     839             :         }
     840           0 :         spin_unlock(&dd->lock);
     841           0 : }
     842             : 
     843             : /* Callback from inside blk_mq_rq_ctx_init(). */
     844           0 : static void dd_prepare_request(struct request *rq)
     845             : {
     846           0 :         rq->elv.priv[0] = NULL;
     847           0 : }
     848             : 
     849             : static bool dd_has_write_work(struct blk_mq_hw_ctx *hctx)
     850             : {
     851             :         struct deadline_data *dd = hctx->queue->elevator->elevator_data;
     852             :         enum dd_prio p;
     853             : 
     854             :         for (p = 0; p <= DD_PRIO_MAX; p++)
     855             :                 if (!list_empty_careful(&dd->per_prio[p].fifo_list[DD_WRITE]))
     856             :                         return true;
     857             : 
     858             :         return false;
     859             : }
     860             : 
     861             : /*
     862             :  * Callback from inside blk_mq_free_request().
     863             :  *
     864             :  * For zoned block devices, write unlock the target zone of
     865             :  * completed write requests. Do this while holding the zone lock
     866             :  * spinlock so that the zone is never unlocked while deadline_fifo_request()
     867             :  * or deadline_next_request() are executing. This function is called for
     868             :  * all requests, whether or not these requests complete successfully.
     869             :  *
     870             :  * For a zoned block device, __dd_dispatch_request() may have stopped
     871             :  * dispatching requests if all the queued requests are write requests directed
     872             :  * at zones that are already locked due to on-going write requests. To ensure
     873             :  * write request dispatch progress in this case, mark the queue as needing a
     874             :  * restart to ensure that the queue is run again after completion of the
     875             :  * request and zones being unlocked.
     876             :  */
     877           0 : static void dd_finish_request(struct request *rq)
     878             : {
     879           0 :         struct request_queue *q = rq->q;
     880           0 :         struct deadline_data *dd = q->elevator->elevator_data;
     881           0 :         const u8 ioprio_class = dd_rq_ioclass(rq);
     882           0 :         const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
     883           0 :         struct dd_per_prio *per_prio = &dd->per_prio[prio];
     884             : 
     885             :         /*
     886             :          * The block layer core may call dd_finish_request() without having
     887             :          * called dd_insert_requests(). Skip requests that bypassed I/O
     888             :          * scheduling. See also blk_mq_request_bypass_insert().
     889             :          */
     890           0 :         if (!rq->elv.priv[0])
     891             :                 return;
     892             : 
     893           0 :         atomic_inc(&per_prio->stats.completed);
     894             : 
     895           0 :         if (blk_queue_is_zoned(q)) {
     896             :                 unsigned long flags;
     897             : 
     898             :                 spin_lock_irqsave(&dd->zone_lock, flags);
     899             :                 blk_req_zone_write_unlock(rq);
     900             :                 spin_unlock_irqrestore(&dd->zone_lock, flags);
     901             : 
     902             :                 if (dd_has_write_work(rq->mq_hctx))
     903             :                         blk_mq_sched_mark_restart_hctx(rq->mq_hctx);
     904             :         }
     905             : }
     906             : 
     907           0 : static bool dd_has_work_for_prio(struct dd_per_prio *per_prio)
     908             : {
     909           0 :         return !list_empty_careful(&per_prio->dispatch) ||
     910           0 :                 !list_empty_careful(&per_prio->fifo_list[DD_READ]) ||
     911           0 :                 !list_empty_careful(&per_prio->fifo_list[DD_WRITE]);
     912             : }
     913             : 
     914           0 : static bool dd_has_work(struct blk_mq_hw_ctx *hctx)
     915             : {
     916           0 :         struct deadline_data *dd = hctx->queue->elevator->elevator_data;
     917             :         enum dd_prio prio;
     918             : 
     919           0 :         for (prio = 0; prio <= DD_PRIO_MAX; prio++)
     920           0 :                 if (dd_has_work_for_prio(&dd->per_prio[prio]))
     921             :                         return true;
     922             : 
     923             :         return false;
     924             : }
     925             : 
     926             : /*
     927             :  * sysfs parts below
     928             :  */
     929             : #define SHOW_INT(__FUNC, __VAR)                                         \
     930             : static ssize_t __FUNC(struct elevator_queue *e, char *page)             \
     931             : {                                                                       \
     932             :         struct deadline_data *dd = e->elevator_data;                 \
     933             :                                                                         \
     934             :         return sysfs_emit(page, "%d\n", __VAR);                               \
     935             : }
     936             : #define SHOW_JIFFIES(__FUNC, __VAR) SHOW_INT(__FUNC, jiffies_to_msecs(__VAR))
     937           0 : SHOW_JIFFIES(deadline_read_expire_show, dd->fifo_expire[DD_READ]);
     938           0 : SHOW_JIFFIES(deadline_write_expire_show, dd->fifo_expire[DD_WRITE]);
     939           0 : SHOW_JIFFIES(deadline_prio_aging_expire_show, dd->prio_aging_expire);
     940           0 : SHOW_INT(deadline_writes_starved_show, dd->writes_starved);
     941           0 : SHOW_INT(deadline_front_merges_show, dd->front_merges);
     942           0 : SHOW_INT(deadline_async_depth_show, dd->async_depth);
     943           0 : SHOW_INT(deadline_fifo_batch_show, dd->fifo_batch);
     944             : #undef SHOW_INT
     945             : #undef SHOW_JIFFIES
     946             : 
     947             : #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
     948             : static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
     949             : {                                                                       \
     950             :         struct deadline_data *dd = e->elevator_data;                 \
     951             :         int __data, __ret;                                              \
     952             :                                                                         \
     953             :         __ret = kstrtoint(page, 0, &__data);                                \
     954             :         if (__ret < 0)                                                       \
     955             :                 return __ret;                                           \
     956             :         if (__data < (MIN))                                          \
     957             :                 __data = (MIN);                                         \
     958             :         else if (__data > (MAX))                                     \
     959             :                 __data = (MAX);                                         \
     960             :         *(__PTR) = __CONV(__data);                                      \
     961             :         return count;                                                   \
     962             : }
     963             : #define STORE_INT(__FUNC, __PTR, MIN, MAX)                              \
     964             :         STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, )
     965             : #define STORE_JIFFIES(__FUNC, __PTR, MIN, MAX)                          \
     966             :         STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, msecs_to_jiffies)
     967           0 : STORE_JIFFIES(deadline_read_expire_store, &dd->fifo_expire[DD_READ], 0, INT_MAX);
     968           0 : STORE_JIFFIES(deadline_write_expire_store, &dd->fifo_expire[DD_WRITE], 0, INT_MAX);
     969           0 : STORE_JIFFIES(deadline_prio_aging_expire_store, &dd->prio_aging_expire, 0, INT_MAX);
     970           0 : STORE_INT(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX);
     971           0 : STORE_INT(deadline_front_merges_store, &dd->front_merges, 0, 1);
     972           0 : STORE_INT(deadline_async_depth_store, &dd->async_depth, 1, INT_MAX);
     973           0 : STORE_INT(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX);
     974             : #undef STORE_FUNCTION
     975             : #undef STORE_INT
     976             : #undef STORE_JIFFIES
     977             : 
     978             : #define DD_ATTR(name) \
     979             :         __ATTR(name, 0644, deadline_##name##_show, deadline_##name##_store)
     980             : 
     981             : static struct elv_fs_entry deadline_attrs[] = {
     982             :         DD_ATTR(read_expire),
     983             :         DD_ATTR(write_expire),
     984             :         DD_ATTR(writes_starved),
     985             :         DD_ATTR(front_merges),
     986             :         DD_ATTR(async_depth),
     987             :         DD_ATTR(fifo_batch),
     988             :         DD_ATTR(prio_aging_expire),
     989             :         __ATTR_NULL
     990             : };
     991             : 
     992             : #ifdef CONFIG_BLK_DEBUG_FS
     993             : #define DEADLINE_DEBUGFS_DDIR_ATTRS(prio, data_dir, name)               \
     994             : static void *deadline_##name##_fifo_start(struct seq_file *m,           \
     995             :                                           loff_t *pos)                  \
     996             :         __acquires(&dd->lock)                                            \
     997             : {                                                                       \
     998             :         struct request_queue *q = m->private;                                \
     999             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1000             :         struct dd_per_prio *per_prio = &dd->per_prio[prio];              \
    1001             :                                                                         \
    1002             :         spin_lock(&dd->lock);                                            \
    1003             :         return seq_list_start(&per_prio->fifo_list[data_dir], *pos);     \
    1004             : }                                                                       \
    1005             :                                                                         \
    1006             : static void *deadline_##name##_fifo_next(struct seq_file *m, void *v,   \
    1007             :                                          loff_t *pos)                   \
    1008             : {                                                                       \
    1009             :         struct request_queue *q = m->private;                                \
    1010             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1011             :         struct dd_per_prio *per_prio = &dd->per_prio[prio];              \
    1012             :                                                                         \
    1013             :         return seq_list_next(v, &per_prio->fifo_list[data_dir], pos);    \
    1014             : }                                                                       \
    1015             :                                                                         \
    1016             : static void deadline_##name##_fifo_stop(struct seq_file *m, void *v)    \
    1017             :         __releases(&dd->lock)                                            \
    1018             : {                                                                       \
    1019             :         struct request_queue *q = m->private;                                \
    1020             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1021             :                                                                         \
    1022             :         spin_unlock(&dd->lock);                                          \
    1023             : }                                                                       \
    1024             :                                                                         \
    1025             : static const struct seq_operations deadline_##name##_fifo_seq_ops = {   \
    1026             :         .start  = deadline_##name##_fifo_start,                         \
    1027             :         .next   = deadline_##name##_fifo_next,                          \
    1028             :         .stop   = deadline_##name##_fifo_stop,                          \
    1029             :         .show   = blk_mq_debugfs_rq_show,                               \
    1030             : };                                                                      \
    1031             :                                                                         \
    1032             : static int deadline_##name##_next_rq_show(void *data,                   \
    1033             :                                           struct seq_file *m)           \
    1034             : {                                                                       \
    1035             :         struct request_queue *q = data;                                 \
    1036             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1037             :         struct dd_per_prio *per_prio = &dd->per_prio[prio];              \
    1038             :         struct request *rq = per_prio->next_rq[data_dir];            \
    1039             :                                                                         \
    1040             :         if (rq)                                                         \
    1041             :                 __blk_mq_debugfs_rq_show(m, rq);                        \
    1042             :         return 0;                                                       \
    1043             : }
    1044             : 
    1045             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_RT_PRIO, DD_READ, read0);
    1046             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_RT_PRIO, DD_WRITE, write0);
    1047             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_BE_PRIO, DD_READ, read1);
    1048             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_BE_PRIO, DD_WRITE, write1);
    1049             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_IDLE_PRIO, DD_READ, read2);
    1050             : DEADLINE_DEBUGFS_DDIR_ATTRS(DD_IDLE_PRIO, DD_WRITE, write2);
    1051             : #undef DEADLINE_DEBUGFS_DDIR_ATTRS
    1052             : 
    1053             : static int deadline_batching_show(void *data, struct seq_file *m)
    1054             : {
    1055             :         struct request_queue *q = data;
    1056             :         struct deadline_data *dd = q->elevator->elevator_data;
    1057             : 
    1058             :         seq_printf(m, "%u\n", dd->batching);
    1059             :         return 0;
    1060             : }
    1061             : 
    1062             : static int deadline_starved_show(void *data, struct seq_file *m)
    1063             : {
    1064             :         struct request_queue *q = data;
    1065             :         struct deadline_data *dd = q->elevator->elevator_data;
    1066             : 
    1067             :         seq_printf(m, "%u\n", dd->starved);
    1068             :         return 0;
    1069             : }
    1070             : 
    1071             : static int dd_async_depth_show(void *data, struct seq_file *m)
    1072             : {
    1073             :         struct request_queue *q = data;
    1074             :         struct deadline_data *dd = q->elevator->elevator_data;
    1075             : 
    1076             :         seq_printf(m, "%u\n", dd->async_depth);
    1077             :         return 0;
    1078             : }
    1079             : 
    1080             : static int dd_queued_show(void *data, struct seq_file *m)
    1081             : {
    1082             :         struct request_queue *q = data;
    1083             :         struct deadline_data *dd = q->elevator->elevator_data;
    1084             :         u32 rt, be, idle;
    1085             : 
    1086             :         spin_lock(&dd->lock);
    1087             :         rt = dd_queued(dd, DD_RT_PRIO);
    1088             :         be = dd_queued(dd, DD_BE_PRIO);
    1089             :         idle = dd_queued(dd, DD_IDLE_PRIO);
    1090             :         spin_unlock(&dd->lock);
    1091             : 
    1092             :         seq_printf(m, "%u %u %u\n", rt, be, idle);
    1093             : 
    1094             :         return 0;
    1095             : }
    1096             : 
    1097             : /* Number of requests owned by the block driver for a given priority. */
    1098             : static u32 dd_owned_by_driver(struct deadline_data *dd, enum dd_prio prio)
    1099             : {
    1100             :         const struct io_stats_per_prio *stats = &dd->per_prio[prio].stats;
    1101             : 
    1102             :         lockdep_assert_held(&dd->lock);
    1103             : 
    1104             :         return stats->dispatched + stats->merged -
    1105             :                 atomic_read(&stats->completed);
    1106             : }
    1107             : 
    1108             : static int dd_owned_by_driver_show(void *data, struct seq_file *m)
    1109             : {
    1110             :         struct request_queue *q = data;
    1111             :         struct deadline_data *dd = q->elevator->elevator_data;
    1112             :         u32 rt, be, idle;
    1113             : 
    1114             :         spin_lock(&dd->lock);
    1115             :         rt = dd_owned_by_driver(dd, DD_RT_PRIO);
    1116             :         be = dd_owned_by_driver(dd, DD_BE_PRIO);
    1117             :         idle = dd_owned_by_driver(dd, DD_IDLE_PRIO);
    1118             :         spin_unlock(&dd->lock);
    1119             : 
    1120             :         seq_printf(m, "%u %u %u\n", rt, be, idle);
    1121             : 
    1122             :         return 0;
    1123             : }
    1124             : 
    1125             : #define DEADLINE_DISPATCH_ATTR(prio)                                    \
    1126             : static void *deadline_dispatch##prio##_start(struct seq_file *m,        \
    1127             :                                              loff_t *pos)               \
    1128             :         __acquires(&dd->lock)                                            \
    1129             : {                                                                       \
    1130             :         struct request_queue *q = m->private;                                \
    1131             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1132             :         struct dd_per_prio *per_prio = &dd->per_prio[prio];              \
    1133             :                                                                         \
    1134             :         spin_lock(&dd->lock);                                            \
    1135             :         return seq_list_start(&per_prio->dispatch, *pos);                \
    1136             : }                                                                       \
    1137             :                                                                         \
    1138             : static void *deadline_dispatch##prio##_next(struct seq_file *m,         \
    1139             :                                             void *v, loff_t *pos)       \
    1140             : {                                                                       \
    1141             :         struct request_queue *q = m->private;                                \
    1142             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1143             :         struct dd_per_prio *per_prio = &dd->per_prio[prio];              \
    1144             :                                                                         \
    1145             :         return seq_list_next(v, &per_prio->dispatch, pos);               \
    1146             : }                                                                       \
    1147             :                                                                         \
    1148             : static void deadline_dispatch##prio##_stop(struct seq_file *m, void *v) \
    1149             :         __releases(&dd->lock)                                            \
    1150             : {                                                                       \
    1151             :         struct request_queue *q = m->private;                                \
    1152             :         struct deadline_data *dd = q->elevator->elevator_data;            \
    1153             :                                                                         \
    1154             :         spin_unlock(&dd->lock);                                          \
    1155             : }                                                                       \
    1156             :                                                                         \
    1157             : static const struct seq_operations deadline_dispatch##prio##_seq_ops = { \
    1158             :         .start  = deadline_dispatch##prio##_start,                      \
    1159             :         .next   = deadline_dispatch##prio##_next,                       \
    1160             :         .stop   = deadline_dispatch##prio##_stop,                       \
    1161             :         .show   = blk_mq_debugfs_rq_show,                               \
    1162             : }
    1163             : 
    1164             : DEADLINE_DISPATCH_ATTR(0);
    1165             : DEADLINE_DISPATCH_ATTR(1);
    1166             : DEADLINE_DISPATCH_ATTR(2);
    1167             : #undef DEADLINE_DISPATCH_ATTR
    1168             : 
    1169             : #define DEADLINE_QUEUE_DDIR_ATTRS(name)                                 \
    1170             :         {#name "_fifo_list", 0400,                                    \
    1171             :                         .seq_ops = &deadline_##name##_fifo_seq_ops}
    1172             : #define DEADLINE_NEXT_RQ_ATTR(name)                                     \
    1173             :         {#name "_next_rq", 0400, deadline_##name##_next_rq_show}
    1174             : static const struct blk_mq_debugfs_attr deadline_queue_debugfs_attrs[] = {
    1175             :         DEADLINE_QUEUE_DDIR_ATTRS(read0),
    1176             :         DEADLINE_QUEUE_DDIR_ATTRS(write0),
    1177             :         DEADLINE_QUEUE_DDIR_ATTRS(read1),
    1178             :         DEADLINE_QUEUE_DDIR_ATTRS(write1),
    1179             :         DEADLINE_QUEUE_DDIR_ATTRS(read2),
    1180             :         DEADLINE_QUEUE_DDIR_ATTRS(write2),
    1181             :         DEADLINE_NEXT_RQ_ATTR(read0),
    1182             :         DEADLINE_NEXT_RQ_ATTR(write0),
    1183             :         DEADLINE_NEXT_RQ_ATTR(read1),
    1184             :         DEADLINE_NEXT_RQ_ATTR(write1),
    1185             :         DEADLINE_NEXT_RQ_ATTR(read2),
    1186             :         DEADLINE_NEXT_RQ_ATTR(write2),
    1187             :         {"batching", 0400, deadline_batching_show},
    1188             :         {"starved", 0400, deadline_starved_show},
    1189             :         {"async_depth", 0400, dd_async_depth_show},
    1190             :         {"dispatch0", 0400, .seq_ops = &deadline_dispatch0_seq_ops},
    1191             :         {"dispatch1", 0400, .seq_ops = &deadline_dispatch1_seq_ops},
    1192             :         {"dispatch2", 0400, .seq_ops = &deadline_dispatch2_seq_ops},
    1193             :         {"owned_by_driver", 0400, dd_owned_by_driver_show},
    1194             :         {"queued", 0400, dd_queued_show},
    1195             :         {},
    1196             : };
    1197             : #undef DEADLINE_QUEUE_DDIR_ATTRS
    1198             : #endif
    1199             : 
    1200             : static struct elevator_type mq_deadline = {
    1201             :         .ops = {
    1202             :                 .depth_updated          = dd_depth_updated,
    1203             :                 .limit_depth            = dd_limit_depth,
    1204             :                 .insert_requests        = dd_insert_requests,
    1205             :                 .dispatch_request       = dd_dispatch_request,
    1206             :                 .prepare_request        = dd_prepare_request,
    1207             :                 .finish_request         = dd_finish_request,
    1208             :                 .next_request           = elv_rb_latter_request,
    1209             :                 .former_request         = elv_rb_former_request,
    1210             :                 .bio_merge              = dd_bio_merge,
    1211             :                 .request_merge          = dd_request_merge,
    1212             :                 .requests_merged        = dd_merged_requests,
    1213             :                 .request_merged         = dd_request_merged,
    1214             :                 .has_work               = dd_has_work,
    1215             :                 .init_sched             = dd_init_sched,
    1216             :                 .exit_sched             = dd_exit_sched,
    1217             :                 .init_hctx              = dd_init_hctx,
    1218             :         },
    1219             : 
    1220             : #ifdef CONFIG_BLK_DEBUG_FS
    1221             :         .queue_debugfs_attrs = deadline_queue_debugfs_attrs,
    1222             : #endif
    1223             :         .elevator_attrs = deadline_attrs,
    1224             :         .elevator_name = "mq-deadline",
    1225             :         .elevator_alias = "deadline",
    1226             :         .elevator_features = ELEVATOR_F_ZBD_SEQ_WRITE,
    1227             :         .elevator_owner = THIS_MODULE,
    1228             : };
    1229             : MODULE_ALIAS("mq-deadline-iosched");
    1230             : 
    1231           1 : static int __init deadline_init(void)
    1232             : {
    1233           1 :         return elv_register(&mq_deadline);
    1234             : }
    1235             : 
    1236           0 : static void __exit deadline_exit(void)
    1237             : {
    1238           0 :         elv_unregister(&mq_deadline);
    1239           0 : }
    1240             : 
    1241             : module_init(deadline_init);
    1242             : module_exit(deadline_exit);
    1243             : 
    1244             : MODULE_AUTHOR("Jens Axboe, Damien Le Moal and Bart Van Assche");
    1245             : MODULE_LICENSE("GPL");
    1246             : MODULE_DESCRIPTION("MQ deadline IO scheduler");

Generated by: LCOV version 1.14