From: Weihang Li <liweihang@huawei.com>
To: <dledford@redhat.com>, <jgg@nvidia.com>
Cc: <leon@kernel.org>, <linux-rdma@vger.kernel.org>,
<linuxarm@huawei.com>, "Xi Wang" <wangxi11@huawei.com>,
Weihang Li <liweihang@huawei.com>
Subject: [PATCH for-next 1/5] RDMA/hns: Optimize the base address table config for MTR
Date: Fri, 21 May 2021 17:29:51 +0800 [thread overview]
Message-ID: <1621589395-2435-2-git-send-email-liweihang@huawei.com> (raw)
In-Reply-To: <1621589395-2435-1-git-send-email-liweihang@huawei.com>
From: Xi Wang <wangxi11@huawei.com>
The base address table is allocated by dma allocator, and the size is
always aligned to PAGE_SIZE. If a fixed size is used to allocate the table,
the number of base address entries stored in the table will be smaller than
that can actually stored.
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_alloc.c | 51 ++++++++++++-----------------
drivers/infiniband/hw/hns/hns_roce_cq.c | 4 +--
drivers/infiniband/hw/hns/hns_roce_device.h | 15 ++++++---
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 26 ++++++++++++---
drivers/infiniband/hw/hns/hns_roce_mr.c | 14 ++++----
drivers/infiniband/hw/hns/hns_roce_qp.c | 2 +-
drivers/infiniband/hw/hns/hns_roce_srq.c | 10 +++---
7 files changed, 68 insertions(+), 54 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_alloc.c b/drivers/infiniband/hw/hns/hns_roce_alloc.c
index 5d389ed..51374b68 100644
--- a/drivers/infiniband/hw/hns/hns_roce_alloc.c
+++ b/drivers/infiniband/hw/hns/hns_roce_alloc.c
@@ -208,10 +208,10 @@ struct hns_roce_buf *hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size,
/* Calc the trunk size and num by required size and page_shift */
if (flags & HNS_ROCE_BUF_DIRECT) {
- buf->trunk_shift = ilog2(ALIGN(size, PAGE_SIZE));
+ buf->trunk_shift = order_base_2(ALIGN(size, PAGE_SIZE));
ntrunk = 1;
} else {
- buf->trunk_shift = ilog2(ALIGN(page_size, PAGE_SIZE));
+ buf->trunk_shift = order_base_2(ALIGN(page_size, PAGE_SIZE));
ntrunk = DIV_ROUND_UP(size, 1 << buf->trunk_shift);
}
@@ -252,50 +252,41 @@ struct hns_roce_buf *hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size,
}
int hns_roce_get_kmem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs,
- int buf_cnt, int start, struct hns_roce_buf *buf)
+ int buf_cnt, struct hns_roce_buf *buf,
+ unsigned int page_shift)
{
- int i, end;
- int total;
-
- end = start + buf_cnt;
- if (end > buf->npages) {
- dev_err(hr_dev->dev,
- "failed to check kmem bufs, end %d + %d total %u!\n",
- start, buf_cnt, buf->npages);
+ unsigned int offset, max_size;
+ int total = 0;
+ int i;
+
+ if (page_shift > buf->trunk_shift) {
+ dev_err(hr_dev->dev, "failed to check kmem buf shift %u > %u\n",
+ page_shift, buf->trunk_shift);
return -EINVAL;
}
- total = 0;
- for (i = start; i < end; i++)
- bufs[total++] = hns_roce_buf_page(buf, i);
+ offset = 0;
+ max_size = buf->ntrunks << buf->trunk_shift;
+ for (i = 0; i < buf_cnt && offset < max_size; i++) {
+ bufs[total++] = hns_roce_buf_dma_addr(buf, offset);
+ offset += (1 << page_shift);
+ }
return total;
}
int hns_roce_get_umem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs,
- int buf_cnt, int start, struct ib_umem *umem,
+ int buf_cnt, struct ib_umem *umem,
unsigned int page_shift)
{
struct ib_block_iter biter;
int total = 0;
- int idx = 0;
- u64 addr;
-
- if (page_shift < HNS_HW_PAGE_SHIFT) {
- dev_err(hr_dev->dev, "failed to check umem page shift %u!\n",
- page_shift);
- return -EINVAL;
- }
/* convert system page cnt to hw page cnt */
rdma_umem_for_each_dma_block(umem, &biter, 1 << page_shift) {
- addr = rdma_block_iter_dma_address(&biter);
- if (idx >= start) {
- bufs[total++] = addr;
- if (total >= buf_cnt)
- goto done;
- }
- idx++;
+ bufs[total++] = rdma_block_iter_dma_address(&biter);
+ if (total >= buf_cnt)
+ goto done;
}
done:
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 488d86b..a5a74b6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -202,13 +202,13 @@ static int alloc_cq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq,
struct hns_roce_buf_attr buf_attr = {};
int ret;
- buf_attr.page_shift = hr_dev->caps.cqe_buf_pg_sz + HNS_HW_PAGE_SHIFT;
+ buf_attr.page_shift = hr_dev->caps.cqe_buf_pg_sz + PAGE_SHIFT;
buf_attr.region[0].size = hr_cq->cq_depth * hr_cq->cqe_size;
buf_attr.region[0].hopnum = hr_dev->caps.cqe_hop_num;
buf_attr.region_count = 1;
ret = hns_roce_mtr_create(hr_dev, &hr_cq->mtr, &buf_attr,
- hr_dev->caps.cqe_ba_pg_sz + HNS_HW_PAGE_SHIFT,
+ hr_dev->caps.cqe_ba_pg_sz + PAGE_SHIFT,
udata, addr);
if (ret)
ibdev_err(ibdev, "failed to alloc CQ mtr, ret = %d.\n", ret);
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 111cab5..02fffb7 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -1060,14 +1060,18 @@ static inline void *hns_roce_buf_offset(struct hns_roce_buf *buf,
(offset & ((1 << buf->trunk_shift) - 1));
}
-static inline dma_addr_t hns_roce_buf_page(struct hns_roce_buf *buf, u32 idx)
+static inline dma_addr_t hns_roce_buf_dma_addr(struct hns_roce_buf *buf,
+ unsigned int offset)
{
- unsigned int offset = idx << buf->page_shift;
-
return buf->trunk_list[offset >> buf->trunk_shift].map +
(offset & ((1 << buf->trunk_shift) - 1));
}
+static inline dma_addr_t hns_roce_buf_page(struct hns_roce_buf *buf, u32 idx)
+{
+ return hns_roce_buf_dma_addr(buf, idx << buf->page_shift);
+}
+
#define hr_hw_page_align(x) ALIGN(x, 1 << HNS_HW_PAGE_SHIFT)
static inline u64 to_hr_hw_page_addr(u64 addr)
@@ -1204,9 +1208,10 @@ struct hns_roce_buf *hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size,
u32 page_shift, u32 flags);
int hns_roce_get_kmem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs,
- int buf_cnt, int start, struct hns_roce_buf *buf);
+ int buf_cnt, struct hns_roce_buf *buf,
+ unsigned int page_shift);
int hns_roce_get_umem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs,
- int buf_cnt, int start, struct ib_umem *umem,
+ int buf_cnt, struct ib_umem *umem,
unsigned int page_shift);
int hns_roce_create_srq(struct ib_srq *srq,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index d5e72e5..0c12d87 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -2018,6 +2018,8 @@ static void set_hem_page_size(struct hns_roce_dev *hr_dev)
caps->llm_buf_pg_sz = 0;
/* MR */
+ caps->mpt_ba_pg_sz = 0;
+ caps->mpt_buf_pg_sz = 0;
caps->pbl_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_16K;
caps->pbl_buf_pg_sz = 0;
calc_pg_sz(caps->num_mtpts, caps->mtpt_entry_sz, caps->mpt_hop_num,
@@ -2025,8 +2027,12 @@ static void set_hem_page_size(struct hns_roce_dev *hr_dev)
HEM_TYPE_MTPT);
/* QP */
- caps->qpc_timer_ba_pg_sz = 0;
+ caps->qpc_ba_pg_sz = 0;
+ caps->qpc_buf_pg_sz = 0;
+ caps->qpc_timer_ba_pg_sz = 0;
caps->qpc_timer_buf_pg_sz = 0;
+ caps->sccc_ba_pg_sz = 0;
+ caps->sccc_buf_pg_sz = 0;
caps->mtt_ba_pg_sz = 0;
caps->mtt_buf_pg_sz = 0;
calc_pg_sz(caps->num_qps, caps->qpc_sz, caps->qpc_hop_num,
@@ -2039,6 +2045,12 @@ static void set_hem_page_size(struct hns_roce_dev *hr_dev)
&caps->sccc_ba_pg_sz, HEM_TYPE_SCCC);
/* CQ */
+ caps->cqc_ba_pg_sz = 0;
+ caps->cqc_buf_pg_sz = 0;
+ caps->cqc_timer_ba_pg_sz = 0;
+ caps->cqc_timer_buf_pg_sz = 0;
+ caps->cqe_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_256K;
+ caps->cqe_buf_pg_sz = 0;
calc_pg_sz(caps->num_cqs, caps->cqc_entry_sz, caps->cqc_hop_num,
caps->cqc_bt_num, &caps->cqc_buf_pg_sz, &caps->cqc_ba_pg_sz,
HEM_TYPE_CQC);
@@ -2053,6 +2065,12 @@ static void set_hem_page_size(struct hns_roce_dev *hr_dev)
/* SRQ */
if (caps->flags & HNS_ROCE_CAP_FLAG_SRQ) {
+ caps->srqc_ba_pg_sz = 0;
+ caps->srqc_buf_pg_sz = 0;
+ caps->srqwqe_ba_pg_sz = 0;
+ caps->srqwqe_buf_pg_sz = 0;
+ caps->idx_ba_pg_sz = 0;
+ caps->idx_buf_pg_sz = 0;
calc_pg_sz(caps->num_srqs, caps->srqc_entry_sz,
caps->srqc_hop_num, caps->srqc_bt_num,
&caps->srqc_buf_pg_sz, &caps->srqc_ba_pg_sz,
@@ -6161,14 +6179,14 @@ static int alloc_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
else
eq->hop_num = hr_dev->caps.eqe_hop_num;
- buf_attr.page_shift = hr_dev->caps.eqe_buf_pg_sz + HNS_HW_PAGE_SHIFT;
+ buf_attr.page_shift = hr_dev->caps.eqe_buf_pg_sz + PAGE_SHIFT;
buf_attr.region[0].size = eq->entries * eq->eqe_size;
buf_attr.region[0].hopnum = eq->hop_num;
buf_attr.region_count = 1;
err = hns_roce_mtr_create(hr_dev, &eq->mtr, &buf_attr,
- hr_dev->caps.eqe_ba_pg_sz +
- HNS_HW_PAGE_SHIFT, NULL, 0);
+ hr_dev->caps.eqe_ba_pg_sz + PAGE_SHIFT, NULL,
+ 0);
if (err)
dev_err(hr_dev->dev, "Failed to alloc EQE mtr, err %d\n", err);
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index 79b3c30..8e6b1ae 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -122,7 +122,7 @@ static int alloc_mr_pbl(struct hns_roce_dev *hr_dev, struct hns_roce_mr *mr,
buf_attr.mtt_only = is_fast;
err = hns_roce_mtr_create(hr_dev, &mr->pbl_mtr, &buf_attr,
- hr_dev->caps.pbl_ba_pg_sz + HNS_HW_PAGE_SHIFT,
+ hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT,
udata, start);
if (err)
ibdev_err(ibdev, "failed to alloc pbl mtr, ret = %d.\n", err);
@@ -737,11 +737,11 @@ static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
return -ENOMEM;
if (mtr->umem)
- npage = hns_roce_get_umem_bufs(hr_dev, pages, page_count, 0,
+ npage = hns_roce_get_umem_bufs(hr_dev, pages, page_count,
mtr->umem, page_shift);
else
- npage = hns_roce_get_kmem_bufs(hr_dev, pages, page_count, 0,
- mtr->kmem);
+ npage = hns_roce_get_kmem_bufs(hr_dev, pages, page_count,
+ mtr->kmem, page_shift);
if (npage != page_count) {
ibdev_err(ibdev, "failed to get mtr page %d != %d.\n", npage,
@@ -753,8 +753,8 @@ static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
if (mtr->hem_cfg.is_direct && npage > 1) {
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);
+ ibdev_err(ibdev, "failed to check %s page: %d / %d.\n",
+ mtr->umem ? "umtr" : "kmtr", ret, npage);
ret = -ENOBUFS;
goto err_alloc_list;
}
@@ -799,7 +799,7 @@ int hns_roce_mtr_map(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
if (r->offset + r->count > page_cnt) {
ret = -EINVAL;
ibdev_err(ibdev,
- "failed to check mtr%u end %u + %u, max %u.\n",
+ "failed to check mtr%u count %u + %u > %u.\n",
i, r->offset, r->count, page_cnt);
return ret;
}
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index c6e120e..9203cf1 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -761,7 +761,7 @@ static int alloc_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
goto err_inline;
}
ret = hns_roce_mtr_create(hr_dev, &hr_qp->mtr, &buf_attr,
- HNS_HW_PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz,
+ PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz,
udata, addr);
if (ret) {
ibdev_err(ibdev, "failed to create WQE mtr, ret = %d.\n", ret);
diff --git a/drivers/infiniband/hw/hns/hns_roce_srq.c b/drivers/infiniband/hw/hns/hns_roce_srq.c
index 546d182..c842210 100644
--- a/drivers/infiniband/hw/hns/hns_roce_srq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_srq.c
@@ -167,14 +167,14 @@ static int alloc_srq_idx(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq,
srq->idx_que.entry_shift = ilog2(HNS_ROCE_IDX_QUE_ENTRY_SZ);
- buf_attr.page_shift = hr_dev->caps.idx_buf_pg_sz + HNS_HW_PAGE_SHIFT;
+ buf_attr.page_shift = hr_dev->caps.idx_buf_pg_sz + PAGE_SHIFT;
buf_attr.region[0].size = to_hr_hem_entries_size(srq->wqe_cnt,
srq->idx_que.entry_shift);
buf_attr.region[0].hopnum = hr_dev->caps.idx_hop_num;
buf_attr.region_count = 1;
ret = hns_roce_mtr_create(hr_dev, &idx_que->mtr, &buf_attr,
- hr_dev->caps.idx_ba_pg_sz + HNS_HW_PAGE_SHIFT,
+ hr_dev->caps.idx_ba_pg_sz + PAGE_SHIFT,
udata, addr);
if (ret) {
ibdev_err(ibdev,
@@ -222,15 +222,15 @@ static int alloc_srq_wqe_buf(struct hns_roce_dev *hr_dev,
HNS_ROCE_SGE_SIZE *
srq->max_gs)));
- buf_attr.page_shift = hr_dev->caps.srqwqe_buf_pg_sz + HNS_HW_PAGE_SHIFT;
+ buf_attr.page_shift = hr_dev->caps.srqwqe_buf_pg_sz + PAGE_SHIFT;
buf_attr.region[0].size = to_hr_hem_entries_size(srq->wqe_cnt,
srq->wqe_shift);
buf_attr.region[0].hopnum = hr_dev->caps.srqwqe_hop_num;
buf_attr.region_count = 1;
ret = hns_roce_mtr_create(hr_dev, &srq->buf_mtr, &buf_attr,
- hr_dev->caps.srqwqe_ba_pg_sz +
- HNS_HW_PAGE_SHIFT, udata, addr);
+ hr_dev->caps.srqwqe_ba_pg_sz + PAGE_SHIFT,
+ udata, addr);
if (ret)
ibdev_err(ibdev,
"failed to alloc SRQ buf mtr, ret = %d.\n", ret);
--
2.7.4
next prev parent reply other threads:[~2021-05-21 9:30 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-21 9:29 [PATCH for-next 0/5] RDMA/hns: Changes about memory allocation and configuration Weihang Li
2021-05-21 9:29 ` Weihang Li [this message]
2021-05-21 9:29 ` [PATCH for-next 2/5] RDMA/hns: Refactor root BT allocation for MTR Weihang Li
2021-05-21 9:29 ` [PATCH for-next 3/5] RDMA/hns: Fix wrong timer context buffer page size Weihang Li
2021-05-21 9:29 ` [PATCH for-next 4/5] RDMA/hns: Use refcount_t APIs for HEM Weihang Li
2021-05-21 9:29 ` [PATCH for-next 5/5] RDMA/hns: Clean the hardware related code " Weihang Li
2021-05-28 23:20 ` [PATCH for-next 0/5] RDMA/hns: Changes about memory allocation and configuration Jason Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1621589395-2435-2-git-send-email-liweihang@huawei.com \
--to=liweihang@huawei.com \
--cc=dledford@redhat.com \
--cc=jgg@nvidia.com \
--cc=leon@kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=linuxarm@huawei.com \
--cc=wangxi11@huawei.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.