OSDN Git Service

md/raid5: add a new member of offset into r5dev
authorYufen Yu <yuyufen@huawei.com>
Thu, 20 Aug 2020 13:22:06 +0000 (09:22 -0400)
committerSong Liu <songliubraving@fb.com>
Thu, 24 Sep 2020 23:44:44 +0000 (16:44 -0700)
Add a new member of offset into struct r5dev. It indicates the
offset of related dev[i].page. For now, since each device have a
privated page, the value is always 0. Thus, we set offset as 0
when allcate page in grow_buffers() and resize_stripes().

To support following different page offset, we try to use the page
offset rather than '0' directly for async_memcpy() and ops_run_io().

We try to support different page offset for xor compution functions
in the following. To avoid repeatly allocate a new array each time,
we add a memory region into scribble buffer to record offset.

No functional change.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
drivers/md/raid5.c
drivers/md/raid5.h

index d589d26..b64f56c 100644 (file)
@@ -477,6 +477,7 @@ static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
                }
                sh->dev[i].page = page;
                sh->dev[i].orig_page = page;
+               sh->dev[i].offset = 0;
        }
 
        return 0;
@@ -1130,7 +1131,7 @@ again:
                                sh->dev[i].vec.bv_page = sh->dev[i].page;
                        bi->bi_vcnt = 1;
                        bi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf);
-                       bi->bi_io_vec[0].bv_offset = 0;
+                       bi->bi_io_vec[0].bv_offset = sh->dev[i].offset;
                        bi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf);
                        bi->bi_write_hint = sh->dev[i].write_hint;
                        if (!rrdev)
@@ -1184,7 +1185,7 @@ again:
                        sh->dev[i].rvec.bv_page = sh->dev[i].page;
                        rbi->bi_vcnt = 1;
                        rbi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf);
-                       rbi->bi_io_vec[0].bv_offset = 0;
+                       rbi->bi_io_vec[0].bv_offset = sh->dev[i].offset;
                        rbi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf);
                        rbi->bi_write_hint = sh->dev[i].write_hint;
                        sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET;
@@ -1404,14 +1405,25 @@ static addr_conv_t *to_addr_conv(struct stripe_head *sh,
        return (void *) (to_addr_page(percpu, i) + sh->disks + 2);
 }
 
+/*
+ * Return a pointer to record offset address.
+ */
+static unsigned int *
+to_addr_offs(struct stripe_head *sh, struct raid5_percpu *percpu)
+{
+       return (unsigned int *) (to_addr_conv(sh, percpu, 0) + sh->disks + 2);
+}
+
 static struct dma_async_tx_descriptor *
 ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
 {
        int disks = sh->disks;
        struct page **xor_srcs = to_addr_page(percpu, 0);
+       unsigned int *off_srcs = to_addr_offs(sh, percpu);
        int target = sh->ops.target;
        struct r5dev *tgt = &sh->dev[target];
        struct page *xor_dest = tgt->page;
+       unsigned int off_dest = tgt->offset;
        int count = 0;
        struct dma_async_tx_descriptor *tx;
        struct async_submit_ctl submit;
@@ -1423,16 +1435,19 @@ ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
                __func__, (unsigned long long)sh->sector, target);
        BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
 
-       for (i = disks; i--; )
-               if (i != target)
+       for (i = disks; i--; ) {
+               if (i != target) {
+                       off_srcs[count] = sh->dev[i].offset;
                        xor_srcs[count++] = sh->dev[i].page;
+               }
+       }
 
        atomic_inc(&sh->count);
 
        init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
                          ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
        if (unlikely(count == 1))
-               tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0,
+               tx = async_memcpy(xor_dest, xor_srcs[0], off_dest, off_srcs[0],
                                RAID5_STRIPE_SIZE(sh->raid_conf), &submit);
        else
                tx = async_xor(xor_dest, xor_srcs, 0, count,
@@ -1854,9 +1869,11 @@ ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
 {
        int disks = sh->disks;
        struct page **xor_srcs;
+       unsigned int *off_srcs;
        struct async_submit_ctl submit;
        int count, pd_idx = sh->pd_idx, i;
        struct page *xor_dest;
+       unsigned int off_dest;
        int prexor = 0;
        unsigned long flags;
        int j = 0;
@@ -1881,24 +1898,31 @@ ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
 again:
        count = 0;
        xor_srcs = to_addr_page(percpu, j);
+       off_srcs = to_addr_offs(sh, percpu);
        /* check if prexor is active which means only process blocks
         * that are part of a read-modify-write (written)
         */
        if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
                prexor = 1;
+               off_dest = off_srcs[count] = sh->dev[pd_idx].offset;
                xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
                for (i = disks; i--; ) {
                        struct r5dev *dev = &sh->dev[i];
                        if (head_sh->dev[i].written ||
-                           test_bit(R5_InJournal, &head_sh->dev[i].flags))
+                           test_bit(R5_InJournal, &head_sh->dev[i].flags)) {
+                               off_srcs[count] = dev->offset;
                                xor_srcs[count++] = dev->page;
+                       }
                }
        } else {
                xor_dest = sh->dev[pd_idx].page;
+               off_dest = sh->dev[pd_idx].offset;
                for (i = disks; i--; ) {
                        struct r5dev *dev = &sh->dev[i];
-                       if (i != pd_idx)
+                       if (i != pd_idx) {
+                               off_srcs[count] = dev->offset;
                                xor_srcs[count++] = dev->page;
+                       }
                }
        }
 
@@ -1924,7 +1948,7 @@ again:
        }
 
        if (unlikely(count == 1))
-               tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0,
+               tx = async_memcpy(xor_dest, xor_srcs[0], off_dest, off_srcs[0],
                                RAID5_STRIPE_SIZE(sh->raid_conf), &submit);
        else
                tx = async_xor(xor_dest, xor_srcs, 0, count,
@@ -2253,8 +2277,9 @@ static int scribble_alloc(struct raid5_percpu *percpu,
                          int num, int cnt)
 {
        size_t obj_size =
-               sizeof(struct page *) * (num+2) +
-               sizeof(addr_conv_t) * (num+2);
+               sizeof(struct page *) * (num + 2) +
+               sizeof(addr_conv_t) * (num + 2) +
+               sizeof(unsigned int) * (num + 2);
        void *scribble;
 
        /*
@@ -2389,6 +2414,7 @@ static int resize_stripes(struct r5conf *conf, int newsize)
                for(i=0; i<conf->pool_size; i++) {
                        nsh->dev[i].page = osh->dev[i].page;
                        nsh->dev[i].orig_page = osh->dev[i].page;
+                       nsh->dev[i].offset = osh->dev[i].offset;
                }
                nsh->hash_lock_index = hash;
                free_stripe(conf->slab_cache, osh);
@@ -2444,6 +2470,7 @@ static int resize_stripes(struct r5conf *conf, int newsize)
                                struct page *p = alloc_page(GFP_NOIO);
                                nsh->dev[i].page = p;
                                nsh->dev[i].orig_page = p;
+                               nsh->dev[i].offset = 0;
                                if (!p)
                                        err = -ENOMEM;
                        }
@@ -4369,7 +4396,8 @@ static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
                        /* place all the copies on one channel */
                        init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
                        tx = async_memcpy(sh2->dev[dd_idx].page,
-                                         sh->dev[i].page, 0, 0, RAID5_STRIPE_SIZE(conf),
+                                         sh->dev[i].page, sh2->dev[dd_idx].offset,
+                                         sh->dev[i].offset, RAID5_STRIPE_SIZE(conf),
                                          &submit);
 
                        set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
index 16fc294..6afdd3a 100644 (file)
@@ -253,6 +253,7 @@ struct stripe_head {
                struct bio      req, rreq;
                struct bio_vec  vec, rvec;
                struct page     *page, *orig_page;
+               unsigned int    offset;     /* offset of the page */
                struct bio      *toread, *read, *towrite, *written;
                sector_t        sector;                 /* sector of this page */
                unsigned long   flags;