1 files changed, 104 insertions, 72 deletions

diff --git a/drivers/md/dm-kcopyd.c b/drivers/md/dm-kcopyd.c
index d8587bac5682..819e37eaaeba 100644
--- a/drivers/md/dm-kcopyd.c
+++ b/drivers/md/dm-kcopyd.c
@@ -27,15 +27,19 @@
 
 #include "dm.h"
 
+#define SUB_JOB_SIZE    128
+#define SPLIT_COUNT     8
+#define MIN_JOBS        8
+#define RESERVE_PAGES   (DIV_ROUND_UP(SUB_JOB_SIZE << SECTOR_SHIFT, PAGE_SIZE))
+
 /*-----------------------------------------------------------------
  * Each kcopyd client has its own little pool of preallocated
  * pages for kcopyd io.
  *---------------------------------------------------------------*/
 struct dm_kcopyd_client {
-        spinlock_t lock;
         struct page_list *pages;
-        unsigned int nr_pages;
-        unsigned int nr_free_pages;
+        unsigned nr_reserved_pages;
+        unsigned nr_free_pages;
 
         struct dm_io_client *io_client;
 
@@ -67,15 +71,18 @@ static void wake(struct dm_kcopyd_client *kc)
         queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
 }
 
-static struct page_list *alloc_pl(void)
+/*
+ * Obtain one page for the use of kcopyd.
+ */
+static struct page_list *alloc_pl(gfp_t gfp)
 {
         struct page_list *pl;
 
-        pl = kmalloc(sizeof(*pl), GFP_KERNEL);
+        pl = kmalloc(sizeof(*pl), gfp);
         if (!pl)
                 return NULL;
 
-        pl->page = alloc_page(GFP_KERNEL);
+        pl->page = alloc_page(gfp);
         if (!pl->page) {
                 kfree(pl);
                 return NULL;
@@ -90,41 +97,56 @@ static void free_pl(struct page_list *pl)
         kfree(pl);
 }
 
-static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
-                            unsigned int nr, struct page_list **pages)
+/*
+ * Add the provided pages to a client's free page list, releasing
+ * back to the system any beyond the reserved_pages limit.
+ */
+static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
 {
-        struct page_list *pl;
-
-        spin_lock(&kc->lock);
-        if (kc->nr_free_pages < nr) {
-                spin_unlock(&kc->lock);
-                return -ENOMEM;
-        }
-
-        kc->nr_free_pages -= nr;
-        for (*pages = pl = kc->pages; --nr; pl = pl->next)
-                ;
+        struct page_list *next;
 
-        kc->pages = pl->next;
-        pl->next = NULL;
+        do {
+                next = pl->next;
 
-        spin_unlock(&kc->lock);
+                if (kc->nr_free_pages >= kc->nr_reserved_pages)
+                        free_pl(pl);
+                else {
+                        pl->next = kc->pages;
+                        kc->pages = pl;
+                        kc->nr_free_pages++;
+                }
 
-        return 0;
+                pl = next;
+        } while (pl);
 }
 
-static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
+static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
+                            unsigned int nr, struct page_list **pages)
 {
-        struct page_list *cursor;
+        struct page_list *pl;
+
+        *pages = NULL;
+
+        do {
+                pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY);
+                if (unlikely(!pl)) {
+                        /* Use reserved pages */
+                        pl = kc->pages;
+                        if (unlikely(!pl))
+                                goto out_of_memory;
+                        kc->pages = pl->next;
+                        kc->nr_free_pages--;
+                }
+                pl->next = *pages;
+                *pages = pl;
+        } while (--nr);
 
-        spin_lock(&kc->lock);
-        for (cursor = pl; cursor->next; cursor = cursor->next)
-                kc->nr_free_pages++;
+        return 0;
 
-        kc->nr_free_pages++;
-        cursor->next = kc->pages;
-        kc->pages = pl;
-        spin_unlock(&kc->lock);
+out_of_memory:
+        if (*pages)
+                kcopyd_put_pages(kc, *pages);
+        return -ENOMEM;
 }
 
 /*
@@ -141,13 +163,16 @@ static void drop_pages(struct page_list *pl)
         }
 }
 
-static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
+/*
+ * Allocate and reserve nr_pages for the use of a specific client.
+ */
+static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
 {
-        unsigned int i;
+        unsigned i;
         struct page_list *pl = NULL, *next;
 
-        for (i = 0; i < nr; i++) {
-                next = alloc_pl();
+        for (i = 0; i < nr_pages; i++) {
+                next = alloc_pl(GFP_KERNEL);
                 if (!next) {
                         if (pl)
                                 drop_pages(pl);
@@ -157,17 +182,18 @@ static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
                 pl = next;
         }
 
+        kc->nr_reserved_pages += nr_pages;
         kcopyd_put_pages(kc, pl);
-        kc->nr_pages += nr;
+
         return 0;
 }
 
 static void client_free_pages(struct dm_kcopyd_client *kc)
 {
-        BUG_ON(kc->nr_free_pages != kc->nr_pages);
+        BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
         drop_pages(kc->pages);
         kc->pages = NULL;
-        kc->nr_free_pages = kc->nr_pages = 0;
+        kc->nr_free_pages = kc->nr_reserved_pages = 0;
 }
 
 /*-----------------------------------------------------------------
@@ -216,16 +242,17 @@ struct kcopyd_job {
         struct mutex lock;
         atomic_t sub_jobs;
         sector_t progress;
-};
 
-/* FIXME: this should scale with the number of pages */
-#define MIN_JOBS 512
+        struct kcopyd_job *master_job;
+};
 
 static struct kmem_cache *_job_cache;
 
 int __init dm_kcopyd_init(void)
 {
-        _job_cache = KMEM_CACHE(kcopyd_job, 0);
+        _job_cache = kmem_cache_create("kcopyd_job",
+                                sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
+                                __alignof__(struct kcopyd_job), 0, NULL);
         if (!_job_cache)
                 return -ENOMEM;
 
@@ -299,7 +326,12 @@ static int run_complete_job(struct kcopyd_job *job)
 
         if (job->pages)
                 kcopyd_put_pages(kc, job->pages);
-        mempool_free(job, kc->job_pool);
+        /*
+         * If this is the master job, the sub jobs have already
+         * completed so we can free everything.
+         */
+        if (job->master_job == job)
+                mempool_free(job, kc->job_pool);
         fn(read_err, write_err, context);
 
         if (atomic_dec_and_test(&kc->nr_jobs))
@@ -345,7 +377,7 @@ static int run_io_job(struct kcopyd_job *job)
 {
         int r;
         struct dm_io_request io_req = {
-                .bi_rw = job->rw | REQ_SYNC | REQ_UNPLUG,
+                .bi_rw = job->rw,
                 .mem.type = DM_IO_PAGE_LIST,
                 .mem.ptr.pl = job->pages,
                 .mem.offset = job->offset,
@@ -428,6 +460,7 @@ static void do_work(struct work_struct *work)
 {
         struct dm_kcopyd_client *kc = container_of(work,
                                         struct dm_kcopyd_client, kcopyd_work);
+        struct blk_plug plug;
 
         /*
          * The order that these are called is *very* important.
@@ -436,9 +469,11 @@ static void do_work(struct work_struct *work)
          * list.  io jobs call wake when they complete and it all
          * starts again.
          */
+        blk_start_plug(&plug);
         process_jobs(&kc->complete_jobs, kc, run_complete_job);
         process_jobs(&kc->pages_jobs, kc, run_pages_job);
         process_jobs(&kc->io_jobs, kc, run_io_job);
+        blk_finish_plug(&plug);
 }
 
 /*
@@ -457,14 +492,14 @@ static void dispatch_job(struct kcopyd_job *job)
         wake(kc);
 }
 
-#define SUB_JOB_SIZE 128
 static void segment_complete(int read_err, unsigned long write_err,
                              void *context)
 {
         /* FIXME: tidy this function */
         sector_t progress = 0;
         sector_t count = 0;
-        struct kcopyd_job *job = (struct kcopyd_job *) context;
+        struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
+        struct kcopyd_job *job = sub_job->master_job;
         struct dm_kcopyd_client *kc = job->kc;
 
         mutex_lock(&job->lock);
@@ -495,8 +530,6 @@ static void segment_complete(int read_err, unsigned long write_err,
 
         if (count) {
                 int i;
-                struct kcopyd_job *sub_job = mempool_alloc(kc->job_pool,
-                                                           GFP_NOIO);
 
                 *sub_job = *job;
                 sub_job->source.sector += progress;
@@ -508,7 +541,7 @@ static void segment_complete(int read_err, unsigned long write_err,
                 }
 
                 sub_job->fn = segment_complete;
-                sub_job->context = job;
+                sub_job->context = sub_job;
                 dispatch_job(sub_job);
 
         } else if (atomic_dec_and_test(&job->sub_jobs)) {
@@ -528,19 +561,19 @@ static void segment_complete(int read_err, unsigned long write_err,
 }
 
 /*
- * Create some little jobs that will do the move between
- * them.
+ * Create some sub jobs to share the work between them.
  */
-#define SPLIT_COUNT 8
-static void split_job(struct kcopyd_job *job)
+static void split_job(struct kcopyd_job *master_job)
 {
         int i;
 
-        atomic_inc(&job->kc->nr_jobs);
+        atomic_inc(&master_job->kc->nr_jobs);
 
-        atomic_set(&job->sub_jobs, SPLIT_COUNT);
-        for (i = 0; i < SPLIT_COUNT; i++)
-                segment_complete(0, 0u, job);
+        atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
+        for (i = 0; i < SPLIT_COUNT; i++) {
+                master_job[i + 1].master_job = master_job;
+                segment_complete(0, 0u, &master_job[i + 1]);
+        }
 }
 
 int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
@@ -550,7 +583,8 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
         struct kcopyd_job *job;
 
         /*
-         * Allocate a new job.
+         * Allocate an array of jobs consisting of one master job
+         * followed by SPLIT_COUNT sub jobs.
          */
         job = mempool_alloc(kc->job_pool, GFP_NOIO);
 
@@ -574,10 +608,10 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 
         job->fn = fn;
         job->context = context;
+        job->master_job = job;
 
-        if (job->source.count < SUB_JOB_SIZE)
+        if (job->source.count <= SUB_JOB_SIZE)
                 dispatch_job(job);
-
         else {
                 mutex_init(&job->lock);
                 job->progress = 0;
@@ -603,17 +637,15 @@ int kcopyd_cancel(struct kcopyd_job *job, int block)
 /*-----------------------------------------------------------------
  * Client setup
  *---------------------------------------------------------------*/
-int dm_kcopyd_client_create(unsigned int nr_pages,
-                            struct dm_kcopyd_client **result)
+struct dm_kcopyd_client *dm_kcopyd_client_create(void)
 {
         int r = -ENOMEM;
         struct dm_kcopyd_client *kc;
 
         kc = kmalloc(sizeof(*kc), GFP_KERNEL);
         if (!kc)
-                return -ENOMEM;
+                return ERR_PTR(-ENOMEM);
 
-        spin_lock_init(&kc->lock);
         spin_lock_init(&kc->job_lock);
         INIT_LIST_HEAD(&kc->complete_jobs);
         INIT_LIST_HEAD(&kc->io_jobs);
@@ -624,17 +656,18 @@ int dm_kcopyd_client_create(unsigned int nr_pages,
                 goto bad_slab;
 
         INIT_WORK(&kc->kcopyd_work, do_work);
-        kc->kcopyd_wq = create_singlethread_workqueue("kcopyd");
+        kc->kcopyd_wq = alloc_workqueue("kcopyd",
+                                        WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
         if (!kc->kcopyd_wq)
                 goto bad_workqueue;
 
         kc->pages = NULL;
-        kc->nr_pages = kc->nr_free_pages = 0;
-        r = client_alloc_pages(kc, nr_pages);
+        kc->nr_reserved_pages = kc->nr_free_pages = 0;
+        r = client_reserve_pages(kc, RESERVE_PAGES);
         if (r)
                 goto bad_client_pages;
 
-        kc->io_client = dm_io_client_create(nr_pages);
+        kc->io_client = dm_io_client_create();
         if (IS_ERR(kc->io_client)) {
                 r = PTR_ERR(kc->io_client);
                 goto bad_io_client;
@@ -643,8 +676,7 @@ int dm_kcopyd_client_create(unsigned int nr_pages,
         init_waitqueue_head(&kc->destroyq);
         atomic_set(&kc->nr_jobs, 0);
 
-        *result = kc;
-        return 0;
+        return kc;
 
 bad_io_client:
         client_free_pages(kc);
@@ -655,7 +687,7 @@ bad_workqueue:
 bad_slab:
         kfree(kc);
 
-        return r;
+        return ERR_PTR(r);
 }
 EXPORT_SYMBOL(dm_kcopyd_client_create);