Skip to contentSkip to conversation

RDMA/bnxt_re: Fix the page_size used during the MR creation

PATCH Applied Selvin Xavier 2 participants 2 messages May 7, 2023 – May 12, 2023

Subsystems (2)
lore
Share & tools
lore

Selvin Xavier

author maintainer
#0 lore raw

PATCH 1/1 RDMA/bnxt_re: Fix the page_size used during the MR creation

Driver populates the list of pages used for Memory region wrongly when page size is more than system page size. This is causing a failure when some of the applications that creates MR with page size as 2M. Since HW can support multiple page sizes, pass the correct page size while creating the MR.

Also, driver need not adjust the number of pages when HW Queues are created with user memory. It should work with the number of dma blocks returned by ib_umem_num_dma_blocks. Fix this calculation also.

Fixes 0c4dcd602817 ("RDMA/bnxt_re: Refactor hardware queue memory allocation")
Fixes f6919d56388c ("RDMA/bnxt_re: Code refactor while populating user MRs")
---
 drivers/infiniband/hw/bnxt_re/qplib_res.c | 12 ++----------
 drivers/infiniband/hw/bnxt_re/qplib_sp.c  |  7 +++----
 2 files changed, 5 insertions(+), 14 deletions(-)
drivers/infiniband/hw/bnxt_re/qplib_res.c +2 -10
index 126d4f2..81b0c5e 100644
@@ -215,17 +215,9 @@ int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
 			return -EINVAL;
 		hwq_attr->sginfo->npages = npages;
 	} else {
-		unsigned long sginfo_num_pages = ib_umem_num_dma_blocks(
-			hwq_attr->sginfo->umem, hwq_attr->sginfo->pgsize);
-
+		npages = ib_umem_num_dma_blocks(hwq_attr->sginfo->umem,
+						hwq_attr->sginfo->pgsize);
 		hwq->is_user = true;
-		npages = sginfo_num_pages;
-		npages = (npages * PAGE_SIZE) /
-			  BIT_ULL(hwq_attr->sginfo->pgshft);
-		if ((sginfo_num_pages * PAGE_SIZE) %
-		     BIT_ULL(hwq_attr->sginfo->pgshft))
-			if (!npages)
-				npages++;
 	}
 
 	if (npages == MAX_PBL_LVL_0_PGS && !hwq_attr->sginfo->nopte) {
drivers/infiniband/hw/bnxt_re/qplib_sp.c +3 -4
index b802981..bae7d89 100644
@@ -584,16 +584,15 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
 		/* Free the hwq if it already exist, must be a rereg */
 		if (mr->hwq.max_elements)
 			bnxt_qplib_free_hwq(res, &mr->hwq);
-		/* Use system PAGE_SIZE */
 		hwq_attr.res = res;
 		hwq_attr.depth = pages;
-		hwq_attr.stride = buf_pg_size;
+		hwq_attr.stride = sizeof(dma_addr_t);
 		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;
+		hwq_attr.sginfo->pgsize = buf_pg_size;
+		hwq_attr.sginfo->pgshft = ilog2(buf_pg_size);
 		rc = bnxt_qplib_alloc_init_hwq(&mr->hwq, &hwq_attr);
 		if (rc) {
 			dev_err(&res->pdev->dev,
Signature · 1 line
2.5.5
maintainer Applied
#1 lore raw
Quoted 15 lines · Selvin Xavier

Driver populates the list of pages used for Memory region wrongly when page size is more than system page size. This is causing a failure when some of the applications that creates MR with page size as 2M. Since HW can support multiple page sizes, pass the correct page size while creating the MR.

Also, driver need not adjust the number of pages when HW Queues are created with user memory. It should work with the number of dma blocks returned by ib_umem_num_dma_blocks. Fix this calculation also.

Fixes 0c4dcd602817 ("RDMA/bnxt_re: Refactor hardware queue memory allocation")
Fixes f6919d56388c ("RDMA/bnxt_re: Code refactor while populating user MRs")
Quoted 4 lines
---
 drivers/infiniband/hw/bnxt_re/qplib_res.c | 12 ++----------
 drivers/infiniband/hw/bnxt_re/qplib_sp.c  |  7 +++----
 2 files changed, 5 insertions(+), 14 deletions(-)

Applied to for-rc, thanks

Jason