]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
null_blk: allow byte aligned memory offsets
authorKeith Busch <kbusch@kernel.org>
Thu, 6 Nov 2025 01:54:47 +0000 (17:54 -0800)
committerJens Axboe <axboe@kernel.dk>
Thu, 6 Nov 2025 23:28:55 +0000 (16:28 -0700)
Allowing byte aligned memory provides a nice testing ground for
direct-io.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Tested-by: Hans Holmberg <hans.holmberg@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/block/null_blk/main.c
drivers/block/null_blk/zoned.c

index ea3fc4241f82cd11b01ec3adf6b7e8108d23bfd0..f1e67962ecaeb9afc408609de3c0932dad486cef 100644 (file)
@@ -1130,25 +1130,27 @@ again:
 }
 
 static blk_status_t copy_to_nullb(struct nullb *nullb, void *source,
-                                 sector_t sector, size_t n, bool is_fua)
+                                 loff_t pos, size_t n, bool is_fua)
 {
        size_t temp, count = 0;
-       unsigned int offset;
        struct nullb_page *t_page;
+       sector_t sector;
 
        while (count < n) {
-               temp = min_t(size_t, nullb->dev->blocksize, n - count);
+               temp = min3(nullb->dev->blocksize, n - count,
+                           PAGE_SIZE - offset_in_page(pos));
+               sector = pos >> SECTOR_SHIFT;
 
                if (null_cache_active(nullb) && !is_fua)
                        null_make_cache_space(nullb, PAGE_SIZE);
 
-               offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
                t_page = null_insert_page(nullb, sector,
                        !null_cache_active(nullb) || is_fua);
                if (!t_page)
                        return BLK_STS_NOSPC;
 
-               memcpy_to_page(t_page->page, offset, source + count, temp);
+               memcpy_to_page(t_page->page, offset_in_page(pos),
+                              source + count, temp);
 
                __set_bit(sector & SECTOR_MASK, t_page->bitmap);
 
@@ -1156,33 +1158,33 @@ static blk_status_t copy_to_nullb(struct nullb *nullb, void *source,
                        null_free_sector(nullb, sector, true);
 
                count += temp;
-               sector += temp >> SECTOR_SHIFT;
+               pos += temp;
        }
        return BLK_STS_OK;
 }
 
-static void copy_from_nullb(struct nullb *nullb, void *dest, sector_t sector,
+static void copy_from_nullb(struct nullb *nullb, void *dest, loff_t pos,
                            size_t n)
 {
        size_t temp, count = 0;
-       unsigned int offset;
        struct nullb_page *t_page;
+       sector_t sector;
 
        while (count < n) {
-               temp = min_t(size_t, nullb->dev->blocksize, n - count);
+               temp = min3(nullb->dev->blocksize, n - count,
+                           PAGE_SIZE - offset_in_page(pos));
+               sector = pos >> SECTOR_SHIFT;
 
-               offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
                t_page = null_lookup_page(nullb, sector, false,
                        !null_cache_active(nullb));
-
                if (t_page)
-                       memcpy_from_page(dest + count, t_page->page, offset,
-                                        temp);
+                       memcpy_from_page(dest + count, t_page->page,
+                                        offset_in_page(pos), temp);
                else
                        memset(dest + count, 0, temp);
 
                count += temp;
-               sector += temp >> SECTOR_SHIFT;
+               pos += temp;
        }
 }
 
@@ -1228,7 +1230,7 @@ static blk_status_t null_handle_flush(struct nullb *nullb)
 }
 
 static blk_status_t null_transfer(struct nullb *nullb, struct page *page,
-       unsigned int len, unsigned int off, bool is_write, sector_t sector,
+       unsigned int len, unsigned int off, bool is_write, loff_t pos,
        bool is_fua)
 {
        struct nullb_device *dev = nullb->dev;
@@ -1240,10 +1242,10 @@ static blk_status_t null_transfer(struct nullb *nullb, struct page *page,
        if (!is_write) {
                if (dev->zoned)
                        valid_len = null_zone_valid_read_len(nullb,
-                               sector, len);
+                               pos >> SECTOR_SHIFT, len);
 
                if (valid_len) {
-                       copy_from_nullb(nullb, p, sector, valid_len);
+                       copy_from_nullb(nullb, p, pos, valid_len);
                        off += valid_len;
                        len -= valid_len;
                }
@@ -1253,7 +1255,7 @@ static blk_status_t null_transfer(struct nullb *nullb, struct page *page,
                flush_dcache_page(page);
        } else {
                flush_dcache_page(page);
-               err = copy_to_nullb(nullb, p, sector, len, is_fua);
+               err = copy_to_nullb(nullb, p, pos, len, is_fua);
        }
 
        kunmap_local(p);
@@ -1271,7 +1273,7 @@ static blk_status_t null_handle_data_transfer(struct nullb_cmd *cmd,
        struct nullb *nullb = cmd->nq->dev->nullb;
        blk_status_t err = BLK_STS_OK;
        unsigned int len;
-       sector_t sector = blk_rq_pos(rq);
+       loff_t pos = blk_rq_pos(rq) << SECTOR_SHIFT;
        unsigned int max_bytes = nr_sectors << SECTOR_SHIFT;
        unsigned int transferred_bytes = 0;
        struct req_iterator iter;
@@ -1283,11 +1285,11 @@ static blk_status_t null_handle_data_transfer(struct nullb_cmd *cmd,
                if (transferred_bytes + len > max_bytes)
                        len = max_bytes - transferred_bytes;
                err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
-                                    op_is_write(req_op(rq)), sector,
+                                    op_is_write(req_op(rq)), pos,
                                     rq->cmd_flags & REQ_FUA);
                if (err)
                        break;
-               sector += len >> SECTOR_SHIFT;
+               pos += len;
                transferred_bytes += len;
                if (transferred_bytes >= max_bytes)
                        break;
@@ -1944,6 +1946,7 @@ static int null_add_dev(struct nullb_device *dev)
                .logical_block_size     = dev->blocksize,
                .physical_block_size    = dev->blocksize,
                .max_hw_sectors         = dev->max_sectors,
+               .dma_alignment          = 1,
        };
 
        struct nullb *nullb;
index 6a93b12a06ff774c1209ffaf2b979b215fb40e27..dbf292a8eae96920ed6cae5eccadf8fba676bf99 100644 (file)
@@ -242,7 +242,7 @@ size_t null_zone_valid_read_len(struct nullb *nullb,
 {
        struct nullb_device *dev = nullb->dev;
        struct nullb_zone *zone = &dev->zones[null_zone_no(dev, sector)];
-       unsigned int nr_sectors = len >> SECTOR_SHIFT;
+       unsigned int nr_sectors = DIV_ROUND_UP(len, SECTOR_SHIFT);
 
        /* Read must be below the write pointer position */
        if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL ||