[PATCH for-next v2 0/2] RDMA/bnxt_re: Allow bigger user MRs

[PATCH for-next v2 0/2] RDMA/bnxt_re: Allow bigger user MRs

[Selvin Xavier]

Refactor user space MR code to handle bigger MRs. Removes couple of checks
that prevented the bigger MRs.

v1->v2:
 - Fix the build warning
   Reported-by: kernel test robot <lkp@intel.com>

Selvin Xavier (2):
  RDMA/bnxt_re: Code refactor while populating user MRs
  RDMA/bnxt_re: Allow bigger MR creation

 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 49 ++++----------------------------
 drivers/infiniband/hw/bnxt_re/qplib_sp.c | 29 +++++--------------
 drivers/infiniband/hw/bnxt_re/qplib_sp.h |  2 +-
 3 files changed, 14 insertions(+), 66 deletions(-)

-- 
2.5.5

[PATCH for-next v2 1/2] RDMA/bnxt_re: Code refactor while populating user MRs

[Selvin Xavier]

Refactor code that populates MR page buffer list. Instead of allocating
a pbl_tbl to hold the buffer list, pass the struct ib_umem directly
to bnxt_qplib_alloc_init_hwq as done for other user space memories.
Fix the PBL level to handle the above mentioned change.

Also, remove an unwanted flag from the input to bnxt_qplib_reg_mr
function.

Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 41 +++++---------------------------
 drivers/infiniband/hw/bnxt_re/qplib_sp.c | 17 +++++--------
 drivers/infiniband/hw/bnxt_re/qplib_sp.h |  2 +-
 3 files changed, 13 insertions(+), 47 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 401bdc9..00d2a5a 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -469,7 +469,6 @@ static int bnxt_re_create_fence_mr(struct bnxt_re_pd *pd)
 	struct bnxt_re_mr *mr = NULL;
 	dma_addr_t dma_addr = 0;
 	struct ib_mw *mw;
-	u64 pbl_tbl;
 	int rc;
 
 	dma_addr = dma_map_single(dev, fence->va, BNXT_RE_FENCE_BYTES,
@@ -504,9 +503,8 @@ static int bnxt_re_create_fence_mr(struct bnxt_re_pd *pd)
 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
 	mr->qplib_mr.va = (u64)(unsigned long)fence->va;
 	mr->qplib_mr.total_size = BNXT_RE_FENCE_BYTES;
-	pbl_tbl = dma_addr;
-	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, &pbl_tbl,
-			       BNXT_RE_FENCE_PBL_SIZE, false, PAGE_SIZE);
+	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, NULL,
+			       BNXT_RE_FENCE_PBL_SIZE, PAGE_SIZE);
 	if (rc) {
 		ibdev_err(&rdev->ibdev, "Failed to register fence-MR\n");
 		goto fail;
@@ -3589,7 +3587,6 @@ struct ib_mr *bnxt_re_get_dma_mr(struct ib_pd *ib_pd, int mr_access_flags)
 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
 	struct bnxt_re_dev *rdev = pd->rdev;
 	struct bnxt_re_mr *mr;
-	u64 pbl = 0;
 	int rc;
 
 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
@@ -3608,7 +3605,7 @@ struct ib_mr *bnxt_re_get_dma_mr(struct ib_pd *ib_pd, int mr_access_flags)
 
 	mr->qplib_mr.hwq.level = PBL_LVL_MAX;
 	mr->qplib_mr.total_size = -1; /* Infinte length */
-	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, &pbl, 0, false,
+	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, NULL, 0,
 			       PAGE_SIZE);
 	if (rc)
 		goto fail_mr;
@@ -3779,19 +3776,6 @@ int bnxt_re_dealloc_mw(struct ib_mw *ib_mw)
 	return rc;
 }
 
-static int fill_umem_pbl_tbl(struct ib_umem *umem, u64 *pbl_tbl_orig,
-			     int page_shift)
-{
-	u64 *pbl_tbl = pbl_tbl_orig;
-	u64 page_size =  BIT_ULL(page_shift);
-	struct ib_block_iter biter;
-
-	rdma_umem_for_each_dma_block(umem, &biter, page_size)
-		*pbl_tbl++ = rdma_block_iter_dma_address(&biter);
-
-	return pbl_tbl - pbl_tbl_orig;
-}
-
 /* uverbs */
 struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
 				  u64 virt_addr, int mr_access_flags,
@@ -3801,7 +3785,6 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
 	struct bnxt_re_dev *rdev = pd->rdev;
 	struct bnxt_re_mr *mr;
 	struct ib_umem *umem;
-	u64 *pbl_tbl = NULL;
 	unsigned long page_size;
 	int umem_pgs, rc;
 
@@ -3855,30 +3838,18 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
 	}
 
 	umem_pgs = ib_umem_num_dma_blocks(umem, page_size);
-	pbl_tbl = kcalloc(umem_pgs, sizeof(*pbl_tbl), GFP_KERNEL);
-	if (!pbl_tbl) {
-		rc = -ENOMEM;
-		goto free_umem;
-	}
-
-	/* Map umem buf ptrs to the PBL */
-	umem_pgs = fill_umem_pbl_tbl(umem, pbl_tbl, order_base_2(page_size));
-	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, pbl_tbl,
-			       umem_pgs, false, page_size);
+	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, umem,
+			       umem_pgs, page_size);
 	if (rc) {
 		ibdev_err(&rdev->ibdev, "Failed to register user MR");
-		goto fail;
+		goto free_umem;
 	}
 
-	kfree(pbl_tbl);
-
 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
 	mr->ib_mr.rkey = mr->qplib_mr.lkey;
 	atomic_inc(&rdev->mr_count);
 
 	return &mr->ib_mr;
-fail:
-	kfree(pbl_tbl);
 free_umem:
 	ib_umem_release(umem);
 free_mrw:
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index 6316179..22cb46a 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -650,16 +650,15 @@ int bnxt_qplib_dereg_mrw(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mrw,
 }
 
 int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
-		      u64 *pbl_tbl, int num_pbls, bool block, u32 buf_pg_size)
+		      struct ib_umem *umem, int num_pbls, u32 buf_pg_size)
 {
 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
 	struct bnxt_qplib_hwq_attr hwq_attr = {};
 	struct bnxt_qplib_sg_info sginfo = {};
 	struct creq_register_mr_resp resp;
 	struct cmdq_register_mr req;
-	int pg_ptrs, pages, i, rc;
 	u16 cmd_flags = 0, level;
-	dma_addr_t **pbl_ptr;
+	int pages, rc, pg_ptrs;
 	u32 pg_size;
 
 	if (num_pbls) {
@@ -683,9 +682,10 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 		/* Use system PAGE_SIZE */
 		hwq_attr.res = res;
 		hwq_attr.depth = pages;
-		hwq_attr.stride = PAGE_SIZE;
+		hwq_attr.stride = buf_pg_size;
 		hwq_attr.type = HWQ_TYPE_MR;
 		hwq_attr.sginfo = &sginfo;
+		hwq_attr.sginfo->umem = umem;
 		hwq_attr.sginfo->npages = pages;
 		hwq_attr.sginfo->pgsize = PAGE_SIZE;
 		hwq_attr.sginfo->pgshft = PAGE_SHIFT;
@@ -695,11 +695,6 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 				"SP: Reg MR memory allocation failed\n");
 			return -ENOMEM;
 		}
-		/* Write to the hwq */
-		pbl_ptr = (dma_addr_t **)mr->hwq.pbl_ptr;
-		for (i = 0; i < num_pbls; i++)
-			pbl_ptr[PTR_PG(i)][PTR_IDX(i)] =
-				(pbl_tbl[i] & PAGE_MASK) | PTU_PTE_VALID;
 	}
 
 	RCFW_CMD_PREP(req, REGISTER_MR, cmd_flags);
@@ -711,7 +706,7 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 		req.pbl = 0;
 		pg_size = PAGE_SIZE;
 	} else {
-		level = mr->hwq.level + 1;
+		level = mr->hwq.level;
 		req.pbl = cpu_to_le64(mr->hwq.pbl[PBL_LVL_0].pg_map_arr[0]);
 	}
 	pg_size = buf_pg_size ? buf_pg_size : PAGE_SIZE;
@@ -728,7 +723,7 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 	req.mr_size = cpu_to_le64(mr->total_size);
 
 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
-					  (void *)&resp, NULL, block);
+					  (void *)&resp, NULL, false);
 	if (rc)
 		goto fail;
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index 967890c..bc22834 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -254,7 +254,7 @@ int bnxt_qplib_alloc_mrw(struct bnxt_qplib_res *res,
 int bnxt_qplib_dereg_mrw(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mrw,
 			 bool block);
 int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
-		      u64 *pbl_tbl, int num_pbls, bool block, u32 buf_pg_size);
+		      struct ib_umem *umem, int num_pbls, u32 buf_pg_size);
 int bnxt_qplib_free_mrw(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr);
 int bnxt_qplib_alloc_fast_reg_mr(struct bnxt_qplib_res *res,
 				 struct bnxt_qplib_mrw *mr, int max);
-- 
2.5.5

[PATCH for-next v2 2/2] RDMA/bnxt_re: Allow bigger MR creation

[Selvin Xavier]

Allow users to create bigger MRs. Remove the check that
prevented creating MRs with number of pages more than 512.

Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c |  8 --------
 drivers/infiniband/hw/bnxt_re/qplib_sp.c | 14 ++------------
 2 files changed, 2 insertions(+), 20 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 00d2a5a..ba515ef 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -3829,14 +3829,6 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
 	}
 	mr->qplib_mr.total_size = length;
 
-	if (page_size == BNXT_RE_PAGE_SIZE_4K &&
-	    length > BNXT_RE_MAX_MR_SIZE_LOW) {
-		ibdev_err(&rdev->ibdev, "Requested MR Sz:%llu Max sup:%llu",
-			  length, (u64)BNXT_RE_MAX_MR_SIZE_LOW);
-		rc = -EINVAL;
-		goto free_umem;
-	}
-
 	umem_pgs = ib_umem_num_dma_blocks(umem, page_size);
 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, umem,
 			       umem_pgs, page_size);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index 22cb46a..049b357 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -658,24 +658,14 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 	struct creq_register_mr_resp resp;
 	struct cmdq_register_mr req;
 	u16 cmd_flags = 0, level;
-	int pages, rc, pg_ptrs;
+	int pages, rc;
 	u32 pg_size;
 
 	if (num_pbls) {
+		pages = roundup_pow_of_two(num_pbls);
 		/* Allocate memory for the non-leaf pages to store buf ptrs.
 		 * Non-leaf pages always uses system PAGE_SIZE
 		 */
-		pg_ptrs = roundup_pow_of_two(num_pbls);
-		pages = pg_ptrs >> MAX_PBL_LVL_1_PGS_SHIFT;
-		if (!pages)
-			pages++;
-
-		if (pages > MAX_PBL_LVL_1_PGS) {
-			dev_err(&res->pdev->dev,
-				"SP: Reg MR: pages requested (0x%x) exceeded max (0x%x)\n",
-				pages, MAX_PBL_LVL_1_PGS);
-			return -ENOMEM;
-		}
 		/* Free the hwq if it already exist, must be a rereg */
 		if (mr->hwq.max_elements)
 			bnxt_qplib_free_hwq(res, &mr->hwq);
-- 
2.5.5

Re: [PATCH for-next v2 0/2] RDMA/bnxt_re: Allow bigger user MRs

[Jason Gunthorpe]

On Thu, Jan 07, 2021 at 01:43:26AM -0800, Selvin Xavier wrote:
> Refactor user space MR code to handle bigger MRs. Removes couple of checks
> that prevented the bigger MRs.
> 
> v1->v2:
>  - Fix the build warning
>    Reported-by: kernel test robot <lkp@intel.com>
> 
> Selvin Xavier (2):
>   RDMA/bnxt_re: Code refactor while populating user MRs
>   RDMA/bnxt_re: Allow bigger MR creation

Applied to for-next, thanks

Jason