IB/iser: Reuse ib_sg_to_pages
authorSagi Grimberg <sagig@mellanox.com>
Wed, 9 Dec 2015 12:12:00 +0000 (14:12 +0200)
committerDoug Ledford <dledford@redhat.com>
Thu, 24 Dec 2015 05:17:34 +0000 (00:17 -0500)
We have in iser iser_sg_to_page_vec which has exactly
the same role as ib_sg_to_pages. Customize the page_vec
to hold a fake MR so we can reuse ib_sg_to_pages.

Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/infiniband/ulp/iser/iscsi_iser.h
drivers/infiniband/ulp/iser/iser_memory.c

index aeaa758..7b0fb9d 100644 (file)
@@ -585,9 +585,8 @@ struct iscsi_iser_task {
 
 struct iser_page_vec {
        u64 *pages;
-       int length;
-       int offset;
-       int data_size;
+       int npages;
+       struct ib_mr fake_mr;
 };
 
 /**
index bcec0a4..e3d9d13 100644 (file)
@@ -135,67 +135,6 @@ iser_reg_desc_put_fmr(struct ib_conn *ib_conn,
 {
 }
 
-#define IS_4K_ALIGNED(addr)    ((((unsigned long)addr) & ~MASK_4K) == 0)
-
-/**
- * iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
- * and returns the length of resulting physical address array (may be less than
- * the original due to possible compaction).
- *
- * we build a "page vec" under the assumption that the SG meets the RDMA
- * alignment requirements. Other then the first and last SG elements, all
- * the "internal" elements can be compacted into a list whose elements are
- * dma addresses of physical pages. The code supports also the weird case
- * where --few fragments of the same page-- are present in the SG as
- * consecutive elements. Also, it handles one entry SG.
- */
-
-static int iser_sg_to_page_vec(struct iser_data_buf *data,
-                              struct ib_device *ibdev, u64 *pages,
-                              int *offset, int *data_size)
-{
-       struct scatterlist *sg, *sgl = data->sg;
-       u64 start_addr, end_addr, page, chunk_start = 0;
-       unsigned long total_sz = 0;
-       unsigned int dma_len;
-       int i, new_chunk, cur_page, last_ent = data->dma_nents - 1;
-
-       /* compute the offset of first element */
-       *offset = (u64) sgl[0].offset & ~MASK_4K;
-
-       new_chunk = 1;
-       cur_page  = 0;
-       for_each_sg(sgl, sg, data->dma_nents, i) {
-               start_addr = ib_sg_dma_address(ibdev, sg);
-               if (new_chunk)
-                       chunk_start = start_addr;
-               dma_len = ib_sg_dma_len(ibdev, sg);
-               end_addr = start_addr + dma_len;
-               total_sz += dma_len;
-
-               /* collect page fragments until aligned or end of SG list */
-               if (!IS_4K_ALIGNED(end_addr) && i < last_ent) {
-                       new_chunk = 0;
-                       continue;
-               }
-               new_chunk = 1;
-
-               /* address of the first page in the contiguous chunk;
-                  masking relevant for the very first SG entry,
-                  which might be unaligned */
-               page = chunk_start & MASK_4K;
-               do {
-                       pages[cur_page++] = page;
-                       page += SIZE_4K;
-               } while (page < end_addr);
-       }
-
-       *data_size = total_sz;
-       iser_dbg("page_vec->data_size:%d cur_page %d\n",
-                *data_size, cur_page);
-       return cur_page;
-}
-
 static void iser_data_buf_dump(struct iser_data_buf *data,
                               struct ib_device *ibdev)
 {
@@ -214,10 +153,10 @@ static void iser_dump_page_vec(struct iser_page_vec *page_vec)
 {
        int i;
 
-       iser_err("page vec length %d data size %d\n",
-                page_vec->length, page_vec->data_size);
-       for (i = 0; i < page_vec->length; i++)
-               iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
+       iser_err("page vec npages %d data length %d\n",
+                page_vec->npages, page_vec->fake_mr.length);
+       for (i = 0; i < page_vec->npages; i++)
+               iser_err("vec[%d]: %llx\n", i, page_vec->pages[i]);
 }
 
 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
@@ -266,11 +205,16 @@ iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
        return 0;
 }
 
-/**
- * iser_reg_page_vec - Register physical memory
- *
- * returns: 0 on success, errno code on failure
- */
+static int iser_set_page(struct ib_mr *mr, u64 addr)
+{
+       struct iser_page_vec *page_vec =
+               container_of(mr, struct iser_page_vec, fake_mr);
+
+       page_vec->pages[page_vec->npages++] = addr;
+
+       return 0;
+}
+
 static
 int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
                      struct iser_data_buf *mem,
@@ -284,22 +228,19 @@ int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
        struct ib_pool_fmr *fmr;
        int ret, plen;
 
-       plen = iser_sg_to_page_vec(mem, device->ib_device,
-                                  page_vec->pages,
-                                  &page_vec->offset,
-                                  &page_vec->data_size);
-       page_vec->length = plen;
-       if (plen * SIZE_4K < page_vec->data_size) {
+       page_vec->npages = 0;
+       page_vec->fake_mr.page_size = SIZE_4K;
+       plen = ib_sg_to_pages(&page_vec->fake_mr, mem->sg,
+                             mem->size, iser_set_page);
+       if (unlikely(plen < mem->size)) {
                iser_err("page vec too short to hold this SG\n");
                iser_data_buf_dump(mem, device->ib_device);
                iser_dump_page_vec(page_vec);
                return -EINVAL;
        }
 
-       fmr  = ib_fmr_pool_map_phys(fmr_pool,
-                                   page_vec->pages,
-                                   page_vec->length,
-                                   page_vec->pages[0]);
+       fmr  = ib_fmr_pool_map_phys(fmr_pool, page_vec->pages,
+                                   page_vec->npages, page_vec->pages[0]);
        if (IS_ERR(fmr)) {
                ret = PTR_ERR(fmr);
                iser_err("ib_fmr_pool_map_phys failed: %d\n", ret);
@@ -308,8 +249,8 @@ int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
 
        reg->sge.lkey = fmr->fmr->lkey;
        reg->rkey = fmr->fmr->rkey;
-       reg->sge.addr = page_vec->pages[0] + page_vec->offset;
-       reg->sge.length = page_vec->data_size;
+       reg->sge.addr = page_vec->fake_mr.iova;
+       reg->sge.length = page_vec->fake_mr.length;
        reg->mem_h = fmr;
 
        iser_dbg("fmr reg: lkey=0x%x, rkey=0x%x, addr=0x%llx,"