buf_attr.region[0].size = length;
buf_attr.region[0].hopnum = mr->pbl_hop_num;
buf_attr.region_count = 1;
- buf_attr.fixed_page = true;
buf_attr.user_access = access;
/* fast MR's buffer is alloced before mapping, not at creation */
buf_attr.mtt_only = is_fast;
}
static int mtr_alloc_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
- struct hns_roce_buf_attr *buf_attr, bool is_direct,
+ struct hns_roce_buf_attr *buf_attr,
struct ib_udata *udata, unsigned long user_addr)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
- unsigned int best_pg_shift;
- int all_pg_count = 0;
size_t total_size;
- int ret;
total_size = mtr_bufs_size(buf_attr);
- if (total_size < 1) {
- ibdev_err(ibdev, "failed to check mtr size\n.");
- return -EINVAL;
- }
if (udata) {
- unsigned long pgsz_bitmap;
- unsigned long page_size;
-
mtr->kmem = NULL;
mtr->umem = ib_umem_get(ibdev, user_addr, total_size,
buf_attr->user_access);
PTR_ERR(mtr->umem));
return -ENOMEM;
}
- if (buf_attr->fixed_page)
- pgsz_bitmap = 1 << buf_attr->page_shift;
- else
- pgsz_bitmap = GENMASK(buf_attr->page_shift, PAGE_SHIFT);
-
- page_size = ib_umem_find_best_pgsz(mtr->umem, pgsz_bitmap,
- user_addr);
- if (!page_size)
- return -EINVAL;
- best_pg_shift = order_base_2(page_size);
- all_pg_count = ib_umem_num_dma_blocks(mtr->umem, page_size);
- ret = 0;
} else {
mtr->umem = NULL;
- mtr->kmem =
- hns_roce_buf_alloc(hr_dev, total_size,
- buf_attr->page_shift,
- is_direct ? HNS_ROCE_BUF_DIRECT : 0);
+ mtr->kmem = hns_roce_buf_alloc(hr_dev, total_size,
+ buf_attr->page_shift,
+ mtr->hem_cfg.is_direct ?
+ HNS_ROCE_BUF_DIRECT : 0);
if (IS_ERR(mtr->kmem)) {
ibdev_err(ibdev, "failed to alloc kmem, ret = %ld.\n",
PTR_ERR(mtr->kmem));
return PTR_ERR(mtr->kmem);
}
-
- best_pg_shift = buf_attr->page_shift;
- all_pg_count = mtr->kmem->npages;
}
- /* must bigger than minimum hardware page shift */
- if (best_pg_shift < HNS_HW_PAGE_SHIFT || all_pg_count < 1) {
- ret = -EINVAL;
- ibdev_err(ibdev,
- "failed to check mtr, page shift = %u count = %d.\n",
- best_pg_shift, all_pg_count);
- goto err_alloc_mem;
- }
-
- mtr->hem_cfg.buf_pg_shift = best_pg_shift;
- mtr->hem_cfg.buf_pg_count = all_pg_count;
-
return 0;
-err_alloc_mem:
- mtr_free_bufs(hr_dev, mtr);
- return ret;
}
-static int mtr_get_pages(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
- dma_addr_t *pages, int count, unsigned int page_shift)
+static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
+ int page_count, unsigned int page_shift)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
+ dma_addr_t *pages;
int npage;
- int err;
+ int ret;
+
+ /* alloc a tmp array to store buffer's dma address */
+ pages = kvcalloc(page_count, sizeof(dma_addr_t), GFP_KERNEL);
+ if (!pages)
+ return -ENOMEM;
if (mtr->umem)
- npage = hns_roce_get_umem_bufs(hr_dev, pages, count, 0,
+ npage = hns_roce_get_umem_bufs(hr_dev, pages, page_count, 0,
mtr->umem, page_shift);
else
- npage = hns_roce_get_kmem_bufs(hr_dev, pages, count, 0,
+ npage = hns_roce_get_kmem_bufs(hr_dev, pages, page_count, 0,
mtr->kmem);
+ if (npage != page_count) {
+ ibdev_err(ibdev, "failed to get mtr page %d != %d.\n", npage,
+ page_count);
+ ret = -ENOBUFS;
+ goto err_alloc_list;
+ }
+
if (mtr->hem_cfg.is_direct && npage > 1) {
- err = mtr_check_direct_pages(pages, npage, page_shift);
- if (err) {
- ibdev_err(ibdev, "Failed to check %s direct page-%d\n",
- mtr->umem ? "user" : "kernel", err);
- npage = err;
+ ret = mtr_check_direct_pages(pages, npage, page_shift);
+ if (ret) {
+ ibdev_err(ibdev, "failed to check %s mtr, idx = %d.\n",
+ mtr->umem ? "user" : "kernel", ret);
+ ret = -ENOBUFS;
+ goto err_alloc_list;
}
}
- return npage;
+ ret = hns_roce_mtr_map(hr_dev, mtr, pages, page_count);
+ if (ret)
+ ibdev_err(ibdev, "failed to map mtr pages, ret = %d.\n", ret);
+
+err_alloc_list:
+ kvfree(pages);
+
+ return ret;
}
int hns_roce_mtr_map(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
static int mtr_init_buf_cfg(struct hns_roce_dev *hr_dev,
struct hns_roce_buf_attr *attr,
struct hns_roce_hem_cfg *cfg,
- unsigned int *buf_page_shift)
+ unsigned int *buf_page_shift, int unalinged_size)
{
struct hns_roce_buf_region *r;
+ int first_region_padding;
+ int page_cnt, region_cnt;
unsigned int page_shift;
- int page_cnt = 0;
size_t buf_size;
- int region_cnt;
+ /* If mtt is disabled, all pages must be within a continuous range */
+ cfg->is_direct = !mtr_has_mtt(attr);
+ buf_size = mtr_bufs_size(attr);
if (cfg->is_direct) {
- buf_size = cfg->buf_pg_count << cfg->buf_pg_shift;
- page_cnt = DIV_ROUND_UP(buf_size, HNS_HW_PAGE_SIZE);
- /*
- * When HEM buffer use level-0 addressing, the page size equals
- * the buffer size, and the the page size = 4K * 2^N.
+ /* When HEM buffer uses 0-level addressing, the page size is
+ * equal to the whole buffer size, and we split the buffer into
+ * small pages which is used to check whether the adjacent
+ * units are in the continuous space and its size is fixed to
+ * 4K based on hns ROCEE's requirement.
*/
- cfg->buf_pg_shift = HNS_HW_PAGE_SHIFT + order_base_2(page_cnt);
- if (attr->region_count > 1) {
- cfg->buf_pg_count = page_cnt;
- page_shift = HNS_HW_PAGE_SHIFT;
- } else {
- cfg->buf_pg_count = 1;
- page_shift = cfg->buf_pg_shift;
- if (buf_size != 1 << page_shift) {
- ibdev_err(&hr_dev->ib_dev,
- "failed to check direct size %zu shift %d.\n",
- buf_size, page_shift);
- return -EINVAL;
- }
- }
+ page_shift = HNS_HW_PAGE_SHIFT;
+
+ /* The ROCEE requires the page size to be 4K * 2 ^ N. */
+ cfg->buf_pg_count = 1;
+ cfg->buf_pg_shift = HNS_HW_PAGE_SHIFT +
+ order_base_2(DIV_ROUND_UP(buf_size, HNS_HW_PAGE_SIZE));
+ first_region_padding = 0;
} else {
- page_shift = cfg->buf_pg_shift;
+ page_shift = attr->page_shift;
+ cfg->buf_pg_count = DIV_ROUND_UP(buf_size + unalinged_size,
+ 1 << page_shift);
+ cfg->buf_pg_shift = page_shift;
+ first_region_padding = unalinged_size;
}
- /* convert buffer size to page index and page count */
- for (page_cnt = 0, region_cnt = 0; page_cnt < cfg->buf_pg_count &&
- region_cnt < attr->region_count &&
+ /* Convert buffer size to page index and page count for each region and
+ * the buffer's offset needs to be appended to the first region.
+ */
+ for (page_cnt = 0, region_cnt = 0; region_cnt < attr->region_count &&
region_cnt < ARRAY_SIZE(cfg->region); region_cnt++) {
r = &cfg->region[region_cnt];
r->offset = page_cnt;
- buf_size = hr_hw_page_align(attr->region[region_cnt].size);
+ buf_size = hr_hw_page_align(attr->region[region_cnt].size +
+ first_region_padding);
r->count = DIV_ROUND_UP(buf_size, 1 << page_shift);
+ first_region_padding = 0;
page_cnt += r->count;
r->hopnum = to_hr_hem_hopnum(attr->region[region_cnt].hopnum,
r->count);
}
- if (region_cnt < 1) {
- ibdev_err(&hr_dev->ib_dev,
- "failed to check mtr region count, pages = %d.\n",
- cfg->buf_pg_count);
- return -ENOBUFS;
- }
-
cfg->region_count = region_cnt;
*buf_page_shift = page_shift;
return page_cnt;
}
+static int mtr_alloc_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
+ unsigned int ba_page_shift)
+{
+ struct hns_roce_hem_cfg *cfg = &mtr->hem_cfg;
+ int ret;
+
+ hns_roce_hem_list_init(&mtr->hem_list);
+ if (!cfg->is_direct) {
+ ret = hns_roce_hem_list_request(hr_dev, &mtr->hem_list,
+ cfg->region, cfg->region_count,
+ ba_page_shift);
+ if (ret)
+ return ret;
+ cfg->root_ba = mtr->hem_list.root_ba;
+ cfg->ba_pg_shift = ba_page_shift;
+ } else {
+ cfg->ba_pg_shift = cfg->buf_pg_shift;
+ }
+
+ return 0;
+}
+
+static void mtr_free_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr)
+{
+ hns_roce_hem_list_release(hr_dev, &mtr->hem_list);
+}
+
/**
* hns_roce_mtr_create - Create hns memory translate region.
*
unsigned int ba_page_shift, struct ib_udata *udata,
unsigned long user_addr)
{
- struct hns_roce_hem_cfg *cfg = &mtr->hem_cfg;
struct ib_device *ibdev = &hr_dev->ib_dev;
unsigned int buf_page_shift = 0;
- dma_addr_t *pages = NULL;
- int all_pg_cnt;
- int get_pg_cnt;
- int ret = 0;
-
- /* if disable mtt, all pages must in a continuous address range */
- cfg->is_direct = !mtr_has_mtt(buf_attr);
-
- /* if buffer only need mtt, just init the hem cfg */
- if (buf_attr->mtt_only) {
- cfg->buf_pg_shift = buf_attr->page_shift;
- cfg->buf_pg_count = mtr_bufs_size(buf_attr) >>
- buf_attr->page_shift;
- mtr->umem = NULL;
- mtr->kmem = NULL;
- } else {
- ret = mtr_alloc_bufs(hr_dev, mtr, buf_attr, cfg->is_direct,
- udata, user_addr);
- if (ret) {
- ibdev_err(ibdev,
- "failed to alloc mtr bufs, ret = %d.\n", ret);
- return ret;
- }
- }
+ int buf_page_cnt;
+ int ret;
- all_pg_cnt = mtr_init_buf_cfg(hr_dev, buf_attr, cfg, &buf_page_shift);
- if (all_pg_cnt < 1) {
- ret = -ENOBUFS;
- ibdev_err(ibdev, "failed to init mtr buf cfg.\n");
- goto err_alloc_bufs;
+ buf_page_cnt = mtr_init_buf_cfg(hr_dev, buf_attr, &mtr->hem_cfg,
+ &buf_page_shift,
+ udata ? user_addr & ~PAGE_MASK : 0);
+ if (buf_page_cnt < 1 || buf_page_shift < HNS_HW_PAGE_SHIFT) {
+ ibdev_err(ibdev, "failed to init mtr cfg, count %d shift %d.\n",
+ buf_page_cnt, buf_page_shift);
+ return -EINVAL;
}
- hns_roce_hem_list_init(&mtr->hem_list);
- if (!cfg->is_direct) {
- ret = hns_roce_hem_list_request(hr_dev, &mtr->hem_list,
- cfg->region, cfg->region_count,
- ba_page_shift);
- if (ret) {
- ibdev_err(ibdev, "failed to request mtr hem, ret = %d.\n",
- ret);
- goto err_alloc_bufs;
- }
- cfg->root_ba = mtr->hem_list.root_ba;
- cfg->ba_pg_shift = ba_page_shift;
- } else {
- cfg->ba_pg_shift = cfg->buf_pg_shift;
+ ret = mtr_alloc_mtt(hr_dev, mtr, ba_page_shift);
+ if (ret) {
+ ibdev_err(ibdev, "failed to alloc mtr mtt, ret = %d.\n", ret);
+ return ret;
}
- /* no buffer to map */
- if (buf_attr->mtt_only)
+ /* The caller has its own buffer list and invokes the hns_roce_mtr_map()
+ * to finish the MTT configuration.
+ */
+ if (buf_attr->mtt_only) {
+ mtr->umem = NULL;
+ mtr->kmem = NULL;
return 0;
-
- /* alloc a tmp array to store buffer's dma address */
- pages = kvcalloc(all_pg_cnt, sizeof(dma_addr_t), GFP_KERNEL);
- if (!pages) {
- ret = -ENOMEM;
- ibdev_err(ibdev, "failed to alloc mtr page list %d.\n",
- all_pg_cnt);
- goto err_alloc_hem_list;
- }
-
- get_pg_cnt = mtr_get_pages(hr_dev, mtr, pages, all_pg_cnt,
- buf_page_shift);
- if (get_pg_cnt != all_pg_cnt) {
- ibdev_err(ibdev, "failed to get mtr page %d != %d.\n",
- get_pg_cnt, all_pg_cnt);
- ret = -ENOBUFS;
- goto err_alloc_page_list;
}
- /* write buffer's dma address to BA table */
- ret = hns_roce_mtr_map(hr_dev, mtr, pages, all_pg_cnt);
+ ret = mtr_alloc_bufs(hr_dev, mtr, buf_attr, udata, user_addr);
if (ret) {
- ibdev_err(ibdev, "failed to map mtr pages, ret = %d.\n", ret);
- goto err_alloc_page_list;
+ ibdev_err(ibdev, "failed to alloc mtr bufs, ret = %d.\n", ret);
+ goto err_alloc_mtt;
}
- /* drop tmp array */
- kvfree(pages);
- return 0;
-err_alloc_page_list:
- kvfree(pages);
-err_alloc_hem_list:
- hns_roce_hem_list_release(hr_dev, &mtr->hem_list);
-err_alloc_bufs:
+ /* Write buffer's dma address to MTT */
+ ret = mtr_map_bufs(hr_dev, mtr, buf_page_cnt, buf_page_shift);
+ if (ret)
+ ibdev_err(ibdev, "failed to map mtr bufs, ret = %d.\n", ret);
+ else
+ return 0;
+
mtr_free_bufs(hr_dev, mtr);
+err_alloc_mtt:
+ mtr_free_mtt(hr_dev, mtr);
return ret;
}
projects / linux.git / commitdiff