RDMA/hns: Optimize base address table config flow for qp buffer

Currently, before the qp is created, a page size needs to be calculated
for the base address table to store all base addresses in the mtr. As a
result, the parameter configuration of the mtr is complex. So integrate
the process of calculating the base table page size into the hem related
interface to simplify the process of using mtr.

Link: https://lore.kernel.org/r/1583839084-31579-5-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index b6ae12dd402465335a8912f98a26ac0ffc6184ed..f6b3cf6b95d65f9ea0ee7800b5353835e9689e4c 100644 (file)
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -669,10 +669,6 @@ struct hns_roce_qp {
        struct ib_umem          *umem;
        struct hns_roce_mtt     mtt;
        struct hns_roce_mtr     mtr;
-
-       /* this define must less than HNS_ROCE_MAX_BT_REGION */
-#define HNS_ROCE_WQE_REGION_MAX         3
-       struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX];
        int                     wqe_bt_pg_shift;
 
        u32                     buff_size;

diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index e82215774032e9437d156995a62f6a73b12569c8..8380d71874946d9177ec3408fbb5bf15ee49ecdd 100644 (file)
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -1383,6 +1383,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
        void *cpu_base;
        u64 phy_base;
        int ret = 0;
+       int ba_num;
        int offset;
        int total;
        int step;
@@ -1393,12 +1394,16 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
        if (root_hem)
                return 0;
 
+       ba_num = hns_roce_hem_list_calc_root_ba(regions, region_cnt, unit);
+       if (ba_num < 1)
+               return -ENOMEM;
+
        INIT_LIST_HEAD(&temp_root);
-       total = r->offset;
+       offset = r->offset;
        /* indicate to last region */
        r = &regions[region_cnt - 1];
-       root_hem = hem_list_alloc_item(hr_dev, total, r->offset + r->count - 1,
-                                      unit, true, 0);
+       root_hem = hem_list_alloc_item(hr_dev, offset, r->offset + r->count - 1,
+                                      ba_num, true, 0);
        if (!root_hem)
                return -ENOMEM;
        list_add(&root_hem->list, &temp_root);
@@ -1410,7 +1415,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
                INIT_LIST_HEAD(&temp_list[i]);
 
        total = 0;
-       for (i = 0; i < region_cnt && total < unit; i++) {
+       for (i = 0; i < region_cnt && total < ba_num; i++) {
                r = &regions[i];
                if (!r->count)
                        continue;
@@ -1443,7 +1448,8 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
                        /* if exist mid bt, link L1 to L0 */
                        list_for_each_entry_safe(hem, temp_hem,
                                          &hem_list->mid_bt[i][1], list) {
-                               offset = hem->start / step * BA_BYTE_LEN;
+                               offset = (hem->start - r->offset) / step *
+                                         BA_BYTE_LEN;
                                hem_list_link_bt(hr_dev, cpu_base + offset,
                                                 hem->dma_addr);
                                total++;

diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index c2ea489b0df5ee8ddc7e288f99eca23d679fdb99..7ba3db5a6192d7fc14e2af9a183d1a14a4402299 100644 (file)
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -579,30 +579,6 @@ static int split_wqe_buf_region(struct hns_roce_dev *hr_dev,
        return region_cnt;
 }
 
-static int calc_wqe_bt_page_shift(struct hns_roce_dev *hr_dev,
-                                 struct hns_roce_buf_region *regions,
-                                 int region_cnt)
-{
-       int bt_pg_shift;
-       int ba_num;
-       int ret;
-
-       bt_pg_shift = PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz;
-
-       /* all root ba entries must in one bt page */
-       do {
-               ba_num = (1 << bt_pg_shift) / BA_BYTE_LEN;
-               ret = hns_roce_hem_list_calc_root_ba(regions, region_cnt,
-                                                    ba_num);
-               if (ret <= ba_num)
-                       break;
-
-               bt_pg_shift++;
-       } while (ret > ba_num);
-
-       return bt_pg_shift - PAGE_SHIFT;
-}
-
 static int set_extend_sge_param(struct hns_roce_dev *hr_dev,
                                struct hns_roce_qp *hr_qp)
 {
@@ -768,7 +744,10 @@ static void free_rq_inline_buf(struct hns_roce_qp *hr_qp)
 static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
                       u32 page_shift, bool is_user)
 {
-       dma_addr_t *buf_list[ARRAY_SIZE(hr_qp->regions)] = { NULL };
+/* WQE buffer include 3 parts: SQ, extend SGE and RQ. */
+#define HNS_ROCE_WQE_REGION_MAX         3
+       struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX] = {};
+       dma_addr_t *buf_list[HNS_ROCE_WQE_REGION_MAX] = {};
        struct ib_device *ibdev = &hr_dev->ib_dev;
        struct hns_roce_buf_region *r;
        int region_count;
@@ -776,18 +755,18 @@ static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
        int ret;
        int i;
 
-       region_count = split_wqe_buf_region(hr_dev, hr_qp, hr_qp->regions,
-                                       ARRAY_SIZE(hr_qp->regions), page_shift);
+       region_count = split_wqe_buf_region(hr_dev, hr_qp, regions,
+                                           ARRAY_SIZE(regions), page_shift);
 
        /* alloc a tmp list to store WQE buffers address */
-       ret = hns_roce_alloc_buf_list(hr_qp->regions, buf_list, region_count);
+       ret = hns_roce_alloc_buf_list(regions, buf_list, region_count);
        if (ret) {
                ibdev_err(ibdev, "Failed to alloc WQE buffer list\n");
                return ret;
        }
 
        for (i = 0; i < region_count; i++) {
-               r = &hr_qp->regions[i];
+               r = &regions[i];
                if (is_user)
                        buf_count = hns_roce_get_umem_bufs(hr_dev, buf_list[i],
                                        r->count, r->offset, hr_qp->umem,
@@ -805,11 +784,10 @@ static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
                }
        }
 
-       hr_qp->wqe_bt_pg_shift = calc_wqe_bt_page_shift(hr_dev, hr_qp->regions,
-                                                       region_count);
+       hr_qp->wqe_bt_pg_shift = hr_dev->caps.mtt_ba_pg_sz;
        hns_roce_mtr_init(&hr_qp->mtr, PAGE_SHIFT + hr_qp->wqe_bt_pg_shift,
                          page_shift);
-       ret = hns_roce_mtr_attach(hr_dev, &hr_qp->mtr, buf_list, hr_qp->regions,
+       ret = hns_roce_mtr_attach(hr_dev, &hr_qp->mtr, buf_list, regions,
                                  region_count);
        if (ret)
                ibdev_err(ibdev, "Failed to attach WQE's mtr\n");