From: Somnath Kotur <somnath.kotur@broadcom.com>
Date: Thu, 11 Jan 2018 11:52:09 -0500
Subject: RDMA/bnxt_re: Add support for MRs with Huge pages
Patch-mainline: v4.16-rc1
Git-commit: 872f3578241d7e648b3bfcf6451a55faf97ce2e9
References: bsc#1050244 FATE#322915
Depending on the OS page-table configurations, applications
may request MRs which has page size alignment other than 4K
Underlying provider driver needs to adjust its PBL boundaries
according to the incoming page boundaries in the PA list.
Adding a capability to register MRs having pages-sizes other
than 4K (Hugepages).
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
drivers/infiniband/hw/bnxt_re/bnxt_re.h | 22 +++--
drivers/infiniband/hw/bnxt_re/ib_verbs.c | 129 ++++++++++++++++++++-----------
drivers/infiniband/hw/bnxt_re/qplib_sp.c | 12 ++
drivers/infiniband/hw/bnxt_re/qplib_sp.h | 2
drivers/infiniband/hw/bnxt_re/roce_hsi.h | 28 ++++++
5 files changed, 138 insertions(+), 55 deletions(-)
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -43,15 +43,23 @@
#define ROCE_DRV_MODULE_VERSION "1.0.0"
#define BNXT_RE_DESC "Broadcom NetXtreme-C/E RoCE Driver"
+#define BNXT_RE_PAGE_SHIFT_4K (12)
+#define BNXT_RE_PAGE_SHIFT_8K (13)
+#define BNXT_RE_PAGE_SHIFT_64K (16)
+#define BNXT_RE_PAGE_SHIFT_2M (21)
+#define BNXT_RE_PAGE_SHIFT_8M (23)
+#define BNXT_RE_PAGE_SHIFT_1G (30)
-#define BNXT_RE_PAGE_SIZE_4K BIT(12)
-#define BNXT_RE_PAGE_SIZE_8K BIT(13)
-#define BNXT_RE_PAGE_SIZE_64K BIT(16)
-#define BNXT_RE_PAGE_SIZE_2M BIT(21)
-#define BNXT_RE_PAGE_SIZE_8M BIT(23)
-#define BNXT_RE_PAGE_SIZE_1G BIT(30)
+#define BNXT_RE_PAGE_SIZE_4K BIT(BNXT_RE_PAGE_SHIFT_4K)
+#define BNXT_RE_PAGE_SIZE_8K BIT(BNXT_RE_PAGE_SHIFT_8K)
+#define BNXT_RE_PAGE_SIZE_64K BIT(BNXT_RE_PAGE_SHIFT_64K)
+#define BNXT_RE_PAGE_SIZE_2M BIT(BNXT_RE_PAGE_SHIFT_2M)
+#define BNXT_RE_PAGE_SIZE_8M BIT(BNXT_RE_PAGE_SHIFT_8M)
+#define BNXT_RE_PAGE_SIZE_1G BIT(BNXT_RE_PAGE_SHIFT_1G)
-#define BNXT_RE_MAX_MR_SIZE BIT(30)
+#define BNXT_RE_MAX_MR_SIZE_LOW BIT(BNXT_RE_PAGE_SHIFT_1G)
+#define BNXT_RE_MAX_MR_SIZE_HIGH BIT(39)
+#define BNXT_RE_MAX_MR_SIZE BNXT_RE_MAX_MR_SIZE_HIGH
#define BNXT_RE_MAX_QPC_COUNT (64 * 1024)
#define BNXT_RE_MAX_MRW_COUNT (64 * 1024)
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -147,7 +147,7 @@ int bnxt_re_query_device(struct ib_devic
bnxt_qplib_get_guid(rdev->netdev->dev_addr,
(u8 *)&ib_attr->sys_image_guid);
ib_attr->max_mr_size = BNXT_RE_MAX_MR_SIZE;
- ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_4K;
+ ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_4K | BNXT_RE_PAGE_SIZE_2M;
ib_attr->vendor_id = rdev->en_dev->pdev->vendor;
ib_attr->vendor_part_id = rdev->en_dev->pdev->device;
@@ -248,8 +248,7 @@ int bnxt_re_query_port(struct ib_device
IB_PORT_VENDOR_CLASS_SUP |
IB_PORT_IP_BASED_GIDS;
- /* Max MSG size set to 2G for now */
- port_attr->max_msg_sz = 0x80000000;
+ port_attr->max_msg_sz = (u32)BNXT_RE_MAX_MR_SIZE_LOW;
port_attr->bad_pkey_cntr = 0;
port_attr->qkey_viol_cntr = 0;
port_attr->pkey_tbl_len = dev_attr->max_pkey;
@@ -542,7 +541,7 @@ static int bnxt_re_create_fence_mr(struc
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);
+ BNXT_RE_FENCE_PBL_SIZE, false, PAGE_SIZE);
if (rc) {
dev_err(rdev_to_dev(rdev), "Failed to register fence-MR\n");
goto fail;
@@ -3072,7 +3071,8 @@ struct ib_mr *bnxt_re_get_dma_mr(struct
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, &pbl, 0, false,
+ PAGE_SIZE);
if (rc)
goto fail_mr;
@@ -3098,10 +3098,8 @@ int bnxt_re_dereg_mr(struct ib_mr *ib_mr
int rc;
rc = bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
- if (rc) {
+ if (rc)
dev_err(rdev_to_dev(rdev), "Dereg MR failed: %#x\n", rc);
- return rc;
- }
if (mr->pages) {
rc = bnxt_qplib_free_fast_reg_page_list(&rdev->qplib_res,
@@ -3164,7 +3162,7 @@ struct ib_mr *bnxt_re_alloc_mr(struct ib
rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
if (rc)
- goto fail;
+ goto bail;
mr->ib_mr.lkey = mr->qplib_mr.lkey;
mr->ib_mr.rkey = mr->ib_mr.lkey;
@@ -3186,9 +3184,10 @@ struct ib_mr *bnxt_re_alloc_mr(struct ib
return &mr->ib_mr;
fail_mr:
- bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
-fail:
kfree(mr->pages);
+fail:
+ bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
+bail:
kfree(mr);
return ERR_PTR(rc);
}
@@ -3242,6 +3241,46 @@ int bnxt_re_dealloc_mw(struct ib_mw *ib_
return rc;
}
+static int bnxt_re_page_size_ok(int page_shift)
+{
+ switch (page_shift) {
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_4K:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_8K:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_64K:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_2M:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_256K:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_1M:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_4M:
+ case CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_1G:
+ return 1;
+ default:
+ return 0;
+ }
+}
+
+static int fill_umem_pbl_tbl(struct ib_umem *umem, u64 *pbl_tbl_orig,
+ int page_shift)
+{
+ u64 *pbl_tbl = pbl_tbl_orig;
+ u64 paddr;
+ u64 page_mask = (1ULL << page_shift) - 1;
+ int i, pages;
+ struct scatterlist *sg;
+ int entry;
+
+ for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+ pages = sg_dma_len(sg) >> PAGE_SHIFT;
+ for (i = 0; i < pages; i++) {
+ paddr = sg_dma_address(sg) + (i << PAGE_SHIFT);
+ if (pbl_tbl == pbl_tbl_orig)
+ *pbl_tbl++ = paddr & ~page_mask;
+ else if ((paddr & page_mask) == 0)
+ *pbl_tbl++ = paddr;
+ }
+ }
+ 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,
@@ -3251,10 +3290,8 @@ struct ib_mr *bnxt_re_reg_user_mr(struct
struct bnxt_re_dev *rdev = pd->rdev;
struct bnxt_re_mr *mr;
struct ib_umem *umem;
- u64 *pbl_tbl, *pbl_tbl_orig;
- int i, umem_pgs, pages, rc;
- struct scatterlist *sg;
- int entry;
+ u64 *pbl_tbl = NULL;
+ int umem_pgs, page_shift, rc;
if (length > BNXT_RE_MAX_MR_SIZE) {
dev_err(rdev_to_dev(rdev), "MR Size: %lld > Max supported:%ld\n",
@@ -3271,64 +3308,68 @@ struct ib_mr *bnxt_re_reg_user_mr(struct
mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_MR;
+ rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
+ if (rc) {
+ dev_err(rdev_to_dev(rdev), "Failed to allocate MR");
+ goto free_mr;
+ }
+ /* The fixed portion of the rkey is the same as the lkey */
+ mr->ib_mr.rkey = mr->qplib_mr.rkey;
+
umem = ib_umem_get(ib_pd->uobject->context, start, length,
mr_access_flags, 0);
if (IS_ERR(umem)) {
dev_err(rdev_to_dev(rdev), "Failed to get umem");
rc = -EFAULT;
- goto free_mr;
+ goto free_mrw;
}
mr->ib_umem = umem;
- rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
- if (rc) {
- dev_err(rdev_to_dev(rdev), "Failed to allocate MR");
- goto release_umem;
- }
- /* The fixed portion of the rkey is the same as the lkey */
- mr->ib_mr.rkey = mr->qplib_mr.rkey;
-
mr->qplib_mr.va = virt_addr;
umem_pgs = ib_umem_page_count(umem);
if (!umem_pgs) {
dev_err(rdev_to_dev(rdev), "umem is invalid!");
rc = -EINVAL;
- goto free_mrw;
+ goto free_umem;
}
mr->qplib_mr.total_size = length;
pbl_tbl = kcalloc(umem_pgs, sizeof(u64 *), GFP_KERNEL);
if (!pbl_tbl) {
- rc = -EINVAL;
- goto free_mrw;
+ rc = -ENOMEM;
+ goto free_umem;
}
- pbl_tbl_orig = pbl_tbl;
- if (umem->hugetlb) {
- dev_err(rdev_to_dev(rdev), "umem hugetlb not supported!");
+ page_shift = umem->page_shift;
+
+ if (!bnxt_re_page_size_ok(page_shift)) {
+ dev_err(rdev_to_dev(rdev), "umem page size unsupported!");
rc = -EFAULT;
goto fail;
}
- if (umem->page_shift != PAGE_SHIFT) {
- dev_err(rdev_to_dev(rdev), "umem page shift unsupported!");
- rc = -EFAULT;
+ if (!umem->hugetlb && length > BNXT_RE_MAX_MR_SIZE_LOW) {
+ dev_err(rdev_to_dev(rdev), "Requested MR Sz:%llu Max sup:%llu",
+ length, (u64)BNXT_RE_MAX_MR_SIZE_LOW);
+ rc = -EINVAL;
goto fail;
}
- /* Map umem buf ptrs to the PBL */
- for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
- pages = sg_dma_len(sg) >> umem->page_shift;
- for (i = 0; i < pages; i++, pbl_tbl++)
- *pbl_tbl = sg_dma_address(sg) + (i << umem->page_shift);
+ if (umem->hugetlb && length > BNXT_RE_PAGE_SIZE_2M) {
+ page_shift = BNXT_RE_PAGE_SHIFT_2M;
+ dev_warn(rdev_to_dev(rdev), "umem hugetlb set page_size %x",
+ 1 << page_shift);
}
- rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, pbl_tbl_orig,
- umem_pgs, false);
+
+ /* Map umem buf ptrs to the PBL */
+ umem_pgs = fill_umem_pbl_tbl(umem, pbl_tbl, page_shift);
+ rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, pbl_tbl,
+ umem_pgs, false, 1 << page_shift);
if (rc) {
dev_err(rdev_to_dev(rdev), "Failed to register user MR");
goto fail;
}
- kfree(pbl_tbl_orig);
+ kfree(pbl_tbl);
mr->ib_mr.lkey = mr->qplib_mr.lkey;
mr->ib_mr.rkey = mr->qplib_mr.lkey;
@@ -3336,11 +3377,11 @@ struct ib_mr *bnxt_re_reg_user_mr(struct
return &mr->ib_mr;
fail:
- kfree(pbl_tbl_orig);
+ kfree(pbl_tbl);
+free_umem:
+ ib_umem_release(umem);
free_mrw:
bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
-release_umem:
- ib_umem_release(umem);
free_mr:
kfree(mr);
return ERR_PTR(rc);
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -657,7 +657,7 @@ int bnxt_qplib_dereg_mrw(struct bnxt_qpl
}
int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr,
- u64 *pbl_tbl, int num_pbls, bool block)
+ u64 *pbl_tbl, int num_pbls, bool block, u32 buf_pg_size)
{
struct bnxt_qplib_rcfw *rcfw = res->rcfw;
struct cmdq_register_mr req;
@@ -668,6 +668,9 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_
u32 pg_size;
if (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)
@@ -685,6 +688,7 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_
bnxt_qplib_free_hwq(res->pdev, &mr->hwq);
mr->hwq.max_elements = pages;
+ /* Use system PAGE_SIZE */
rc = bnxt_qplib_alloc_init_hwq(res->pdev, &mr->hwq, NULL, 0,
&mr->hwq.max_elements,
PAGE_SIZE, 0, PAGE_SIZE,
@@ -705,18 +709,22 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_
/* Configure the request */
if (mr->hwq.level == PBL_LVL_MAX) {
+ /* No PBL provided, just use system PAGE_SIZE */
level = 0;
req.pbl = 0;
pg_size = PAGE_SIZE;
} else {
level = mr->hwq.level + 1;
req.pbl = cpu_to_le64(mr->hwq.pbl[PBL_LVL_0].pg_map_arr[0]);
- pg_size = mr->hwq.pbl[PBL_LVL_0].pg_size;
}
+ pg_size = buf_pg_size ? buf_pg_size : PAGE_SIZE;
req.log2_pg_size_lvl = (level << CMDQ_REGISTER_MR_LVL_SFT) |
((ilog2(pg_size) <<
CMDQ_REGISTER_MR_LOG2_PG_SIZE_SFT) &
CMDQ_REGISTER_MR_LOG2_PG_SIZE_MASK);
+ req.log2_pbl_pg_size = cpu_to_le16(((ilog2(PAGE_SIZE) <<
+ CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_SFT) &
+ CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_MASK));
req.access = (mr->flags & 0xFFFF);
req.va = cpu_to_le64(mr->va);
req.key = cpu_to_le32(mr->lkey);
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -159,7 +159,7 @@ int bnxt_qplib_alloc_mrw(struct bnxt_qpl
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);
+ u64 *pbl_tbl, int num_pbls, bool block, 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);
--- a/drivers/infiniband/hw/bnxt_re/roce_hsi.h
+++ b/drivers/infiniband/hw/bnxt_re/roce_hsi.h
@@ -1383,8 +1383,20 @@ struct cmdq_register_mr {
#define CMDQ_REGISTER_MR_LVL_LVL_0 0x0UL
#define CMDQ_REGISTER_MR_LVL_LVL_1 0x1UL
#define CMDQ_REGISTER_MR_LVL_LVL_2 0x2UL
+ #define CMDQ_REGISTER_MR_LVL_LAST CMDQ_REGISTER_MR_LVL_LVL_2
#define CMDQ_REGISTER_MR_LOG2_PG_SIZE_MASK 0x7cUL
#define CMDQ_REGISTER_MR_LOG2_PG_SIZE_SFT 2
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_4K (0xcUL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_8K (0xdUL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_64K (0x10UL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_256K (0x12UL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_1M (0x14UL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_2M (0x15UL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_4M (0x16UL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_1G (0x1eUL << 2)
+ #define CMDQ_REGISTER_MR_LOG2_PG_SIZE_LAST \
+ CMDQ_REGISTER_MR_LOG2_PG_SIZE_PG_1G
+ #define CMDQ_REGISTER_MR_UNUSED1 0x80UL
u8 access;
#define CMDQ_REGISTER_MR_ACCESS_LOCAL_WRITE 0x1UL
#define CMDQ_REGISTER_MR_ACCESS_REMOTE_READ 0x2UL
@@ -1392,7 +1404,21 @@ struct cmdq_register_mr {
#define CMDQ_REGISTER_MR_ACCESS_REMOTE_ATOMIC 0x8UL
#define CMDQ_REGISTER_MR_ACCESS_MW_BIND 0x10UL
#define CMDQ_REGISTER_MR_ACCESS_ZERO_BASED 0x20UL
- __le16 unused_1;
+ __le16 log2_pbl_pg_size;
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_MASK 0x1fUL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_SFT 0
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_4K 0xcUL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_8K 0xdUL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_64K 0x10UL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_256K 0x12UL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_1M 0x14UL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_2M 0x15UL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_4M 0x16UL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_1G 0x1eUL
+ #define CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_LAST \
+ CMDQ_REGISTER_MR_LOG2_PBL_PG_SIZE_PG_1G
+ #define CMDQ_REGISTER_MR_UNUSED11_MASK 0xffe0UL
+ #define CMDQ_REGISTER_MR_UNUSED11_SFT 5
__le32 key;
__le64 pbl;
__le64 va;