Skip to contentSkip to conversation

libhns: Add support for Dynamic Context Attachment

PATCH Superseded Weihang Li 2 participants 9 messages May 10, 2021 – Jun 5, 2021 InfiniBand

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Weihang Li

author
#0 lore raw

cover libhns: Add support for Dynamic Context Attachment

The HIP09 introduces the DCA(Dynamic Context Attachment) feature which supports many RC QPs to share the WQE buffer in a memory pool. If a QP enables DCA feature, the WQE's buffer will not be allocated when creating but when the users start to post WRs. This will reduce the memory consumption when there are too many QPs are inactive.

For more detailed information, please refer to the man pages provided by the last patch of this series.

This series is associated with the kernel one "RDMA/hns: Add support for Dynamic Context Attachment", and two RFC versions of this series has been sent before.

No changes since RFC v2.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Changes since RFC v1:
* Add direct verbs to set the parameters about size that used to
configuring DCA.
* Add man pages to explain what is DCA, how does it works and how to use
it.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Weihang Li (1):
Update kernel headers

Xi Wang (5):
libhns: Introduce DCA for RC QP
libhns: Add support for shrinking DCA memory pool
libhns: Add support for attaching QP's WQE buffer
libhns: Add direct verbs support to config DCA
libhns: Add man pages to introduce DCA feature

 CMakeLists.txt                             |   1 +
 debian/control                             |   2 +-
 debian/ibverbs-providers.install           |   1 +
 debian/ibverbs-providers.lintian-overrides |   4 +-
 debian/ibverbs-providers.symbols           |   6 +
 debian/libibverbs-dev.install              |   6 +
 kernel-headers/rdma/hns-abi.h              |  64 +++++
 providers/hns/CMakeLists.txt               |   9 +-
 providers/hns/hns_roce_u.c                 | 100 ++++++++
 providers/hns/hns_roce_u.h                 |  44 ++++
 providers/hns/hns_roce_u_abi.h             |   1 +
 providers/hns/hns_roce_u_buf.c             | 387 +++++++++++++++++++++++++++++
 providers/hns/hns_roce_u_hw_v2.c           | 130 ++++++++--
 providers/hns/hns_roce_u_hw_v2.h           |   7 +
 providers/hns/hns_roce_u_verbs.c           |  67 ++++-
 providers/hns/hnsdv.h                      |  61 +++++
 providers/hns/libhns.map                   |   9 +
 providers/hns/man/CMakeLists.txt           |   7 +
 providers/hns/man/hns_dca.7.md             |  35 +++
 providers/hns/man/hnsdv.7.md               |  34 +++
 providers/hns/man/hnsdv_create_qp.3.md     |  69 +++++
 providers/hns/man/hnsdv_is_supported.3.md  |  39 +++
 providers/hns/man/hnsdv_open_device.3.md   |  70 ++++++
 redhat/rdma-core.spec                      |   7 +-
 suse/rdma-core.spec                        |  21 +-
 25 files changed, 1149 insertions(+), 32 deletions(-)
 create mode 100644 providers/hns/hnsdv.h
 create mode 100644 providers/hns/libhns.map
 create mode 100644 providers/hns/man/CMakeLists.txt
 create mode 100644 providers/hns/man/hns_dca.7.md
 create mode 100644 providers/hns/man/hnsdv.7.md
 create mode 100644 providers/hns/man/hnsdv_create_qp.3.md
 create mode 100644 providers/hns/man/hnsdv_is_supported.3.md
 create mode 100644 providers/hns/man/hnsdv_open_device.3.md
Signature · 1 line
2.7.4
author
#1 lore raw

PATCH 1/6 Update kernel headers

To commit ?? ("RDMA/hns: Add method to query WQE buffer's address").

---
 kernel-headers/rdma/hns-abi.h | 64 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 64 insertions(+)
kernel-headers/rdma/hns-abi.h +64 -0
index 42b1776..edcecc1 100644
@@ -77,21 +77,85 @@ enum hns_roce_qp_cap_flags {
 	HNS_ROCE_QP_CAP_RQ_RECORD_DB = 1 << 0,
 	HNS_ROCE_QP_CAP_SQ_RECORD_DB = 1 << 1,
 	HNS_ROCE_QP_CAP_OWNER_DB = 1 << 2,
+	HNS_ROCE_QP_CAP_DCA = 1 << 4,
 };
 
 struct hns_roce_ib_create_qp_resp {
 	__aligned_u64 cap_flags;
 };
 
+enum {
+	HNS_ROCE_CAP_FLAG_DCA_MODE = 1 << 15,
+};
+
 struct hns_roce_ib_alloc_ucontext_resp {
 	__u32	qp_tab_size;
 	__u32	cqe_size;
 	__u32	srq_tab_size;
 	__u32	reserved;
+	__aligned_u64 cap_flags;
 };
 
 struct hns_roce_ib_alloc_pd_resp {
 	__u32 pdn;
 };
 
+#define UVERBS_ID_NS_MASK 0xF000
+#define UVERBS_ID_NS_SHIFT 12
+
+enum hns_ib_objects {
+	HNS_IB_OBJECT_DCA_MEM = (1U << UVERBS_ID_NS_SHIFT),
+};
+
+enum hns_ib_dca_mem_methods {
+	HNS_IB_METHOD_DCA_MEM_REG = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_METHOD_DCA_MEM_DEREG,
+	HNS_IB_METHOD_DCA_MEM_SHRINK,
+	HNS_IB_METHOD_DCA_MEM_ATTACH,
+	HNS_IB_METHOD_DCA_MEM_DETACH,
+	HNS_IB_METHOD_DCA_MEM_QUERY,
+};
+
+enum hns_ib_dca_mem_reg_attrs {
+	HNS_IB_ATTR_DCA_MEM_REG_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_ATTR_DCA_MEM_REG_LEN,
+	HNS_IB_ATTR_DCA_MEM_REG_ADDR,
+	HNS_IB_ATTR_DCA_MEM_REG_KEY,
+};
+
+enum hns_ib_dca_mem_dereg_attrs {
+	HNS_IB_ATTR_DCA_MEM_DEREG_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+};
+
+enum hns_ib_dca_mem_shrink_attrs {
+	HNS_IB_ATTR_DCA_MEM_SHRINK_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_ATTR_DCA_MEM_SHRINK_RESERVED_SIZE,
+	HNS_IB_ATTR_DCA_MEM_SHRINK_OUT_FREE_KEY,
+	HNS_IB_ATTR_DCA_MEM_SHRINK_OUT_FREE_MEMS,
+};
+
+#define HNS_IB_ATTACH_FLAGS_NEW_BUFFER 1U
+
+enum hns_ib_dca_mem_attach_attrs {
+	HNS_IB_ATTR_DCA_MEM_ATTACH_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_ATTR_DCA_MEM_ATTACH_SQ_OFFSET,
+	HNS_IB_ATTR_DCA_MEM_ATTACH_SGE_OFFSET,
+	HNS_IB_ATTR_DCA_MEM_ATTACH_RQ_OFFSET,
+	HNS_IB_ATTR_DCA_MEM_ATTACH_OUT_ALLOC_FLAGS,
+	HNS_IB_ATTR_DCA_MEM_ATTACH_OUT_ALLOC_PAGES,
+};
+
+enum hns_ib_dca_mem_detach_attrs {
+	HNS_IB_ATTR_DCA_MEM_DETACH_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_ATTR_DCA_MEM_DETACH_SQ_INDEX,
+};
+
+enum hns_ib_dca_mem_query_attrs {
+	HNS_IB_ATTR_DCA_MEM_QUERY_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	HNS_IB_ATTR_DCA_MEM_QUERY_PAGE_INDEX,
+	HNS_IB_ATTR_DCA_MEM_QUERY_OUT_KEY,
+	HNS_IB_ATTR_DCA_MEM_QUERY_OUT_OFFSET,
+	HNS_IB_ATTR_DCA_MEM_QUERY_OUT_PAGE_COUNT,
+};
+
 #endif /* HNS_ABI_USER_H */
Signature · 1 line
2.7.4
author
#2 lore raw

PATCH 2/6 libhns: Introduce DCA for RC QP

In reply to Weihang Li

From: Xi Wang <[email protected]>

The HIP09 introduces the DCA(Dynamic context attachment) feature which supports many RC QPs to share the WQE buffer in a memory pool, this will reduce the memory consumption when there are too many QPs inactive.

This patch wraps two functions for adding buffers to memory pool and removing buffers from memory pool by calling ib cmd implemented in hns kernel driver.

If a QP enables DCA feature, the WQE's buffer will be attached to the memory pool when the users start to post WRs and be detached when all CQEs has been polled.

---
 providers/hns/hns_roce_u.c     |  45 ++++++++++++++
 providers/hns/hns_roce_u.h     |  18 ++++++
 providers/hns/hns_roce_u_buf.c | 137 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 200 insertions(+)
providers/hns/hns_roce_u.c +45 -0
index 3b31ad3..a4e0997 100644
@@ -95,6 +95,40 @@ static const struct verbs_context_ops hns_common_ops = {
 	.get_srq_num = hns_roce_u_get_srq_num,
 };
 
+static int init_dca_context(struct hns_roce_context *ctx, int page_size)
+{
+	struct hns_roce_dca_ctx *dca_ctx = &ctx->dca_ctx;
+	int ret;
+
+	if (!(ctx->cap_flags & HNS_ROCE_CAP_FLAG_DCA_MODE))
+		return 0;
+
+	list_head_init(&dca_ctx->mem_list);
+	ret = pthread_spin_init(&dca_ctx->lock, PTHREAD_PROCESS_PRIVATE);
+	if (ret)
+		return ret;
+
+	dca_ctx->unit_size = page_size * HNS_DCA_DEFAULT_UNIT_PAGES;
+	dca_ctx->max_size = HNS_DCA_MAX_MEM_SIZE;
+	dca_ctx->mem_cnt = 0;
+
+	return 0;
+}
+
+static void uninit_dca_context(struct hns_roce_context *ctx)
+{
+	struct hns_roce_dca_ctx *dca_ctx = &ctx->dca_ctx;
+
+	if (!(ctx->cap_flags & HNS_ROCE_CAP_FLAG_DCA_MODE))
+		return;
+
+	pthread_spin_lock(&dca_ctx->lock);
+	hns_roce_cleanup_dca_mem(ctx);
+	pthread_spin_unlock(&dca_ctx->lock);
+
+	pthread_spin_destroy(&dca_ctx->lock);
+}
+
 static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 						    int cmd_fd,
 						    void *private_data)
@@ -123,6 +157,8 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 	else
 		context->cqe_size = HNS_ROCE_V3_CQE_SIZE;
 
+	context->cap_flags = resp.cap_flags;
+
 	context->num_qps = resp.qp_tab_size;
 	context->num_srqs = resp.srq_tab_size;
 
@@ -178,8 +214,15 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 	verbs_set_ops(&context->ibv_ctx, &hns_common_ops);
 	verbs_set_ops(&context->ibv_ctx, &hr_dev->u_hw->hw_ops);
 
+	if (init_dca_context(context, hr_dev->page_size))
+		goto tptr_free;
+
 	return &context->ibv_ctx;
 
+tptr_free:
+	if (hr_dev->hw_version == HNS_ROCE_HW_VER1)
+		munmap(context->cq_tptr_base, HNS_ROCE_CQ_DB_BUF_SIZE);
+
 db_free:
 	munmap(context->uar, hr_dev->page_size);
 	context->uar = NULL;
@@ -199,6 +242,8 @@ static void hns_roce_free_context(struct ibv_context *ibctx)
 	if (hr_dev->hw_version == HNS_ROCE_HW_VER1)
 		munmap(context->cq_tptr_base, HNS_ROCE_CQ_DB_BUF_SIZE);
 
+	uninit_dca_context(context);
+
 	verbs_uninit_context(&context->ibv_ctx);
 	free(context);
 }
providers/hns/hns_roce_u.h +18 -0
index 8f805dd..d26ef9c 100644
@@ -145,8 +145,21 @@ struct hns_roce_db_page {
 	bitmap			*bitmap;
 };
 
+#define HNS_DCA_MAX_MEM_SIZE ~0UL
+#define HNS_DCA_DEFAULT_UNIT_PAGES 16
+
+struct hns_roce_dca_ctx {
+	struct list_head mem_list;
+	pthread_spinlock_t lock;
+	uint64_t max_size;
+	uint64_t curr_size;
+	int mem_cnt;
+	unsigned int unit_size;
+};
+
 struct hns_roce_context {
 	struct verbs_context		ibv_ctx;
+	uint32_t			cap_flags;
 	void				*uar;
 	pthread_spinlock_t		uar_lock;
 
@@ -179,6 +192,8 @@ struct hns_roce_context {
 	unsigned int			max_srq_sge;
 	int				max_cqe;
 	unsigned int			cqe_size;
+
+	struct hns_roce_dca_ctx		dca_ctx;
 };
 
 struct hns_roce_pd {
@@ -424,6 +439,9 @@ void hns_roce_free_buf(struct hns_roce_buf *buf);
 
 void hns_roce_free_qp_buf(struct hns_roce_qp *qp, struct hns_roce_context *ctx);
 
+void hns_roce_cleanup_dca_mem(struct hns_roce_context *ctx);
+int hns_roce_add_dca_mem(struct hns_roce_context *ctx, uint32_t size);
+
 void hns_roce_init_qp_indices(struct hns_roce_qp *qp);
 
 extern const struct hns_roce_u_hw hns_roce_u_hw_v1;
providers/hns/hns_roce_u_buf.c +137 -0

140 lines not on this page · raw message

Signature · 1 line
2.7.4
author
#3 lore raw

PATCH 3/6 libhns: Add support for shrinking DCA memory pool

In reply to Weihang Li

From: Xi Wang <[email protected]>

The QP's WQE buffer may be detached after QP is modified or CQE is polled, and the state of DCA mem object may be changed as clean for no QP is using it. So shrink the clean DCA mem from the memory pool and destroy the DCA mem's buffer to reduce the memory consumption.

---
 providers/hns/hns_roce_u.h       |  2 +
 providers/hns/hns_roce_u_buf.c   | 96 ++++++++++++++++++++++++++++++++++++++++
 providers/hns/hns_roce_u_hw_v2.c |  7 +++
 3 files changed, 105 insertions(+)
providers/hns/hns_roce_u.h +2 -0
index d26ef9c..a8a007e 100644
@@ -152,6 +152,7 @@ struct hns_roce_dca_ctx {
 	struct list_head mem_list;
 	pthread_spinlock_t lock;
 	uint64_t max_size;
+	uint64_t min_size;
 	uint64_t curr_size;
 	int mem_cnt;
 	unsigned int unit_size;
@@ -439,6 +440,7 @@ void hns_roce_free_buf(struct hns_roce_buf *buf);
 
 void hns_roce_free_qp_buf(struct hns_roce_qp *qp, struct hns_roce_context *ctx);
 
+void hns_roce_shrink_dca_mem(struct hns_roce_context *ctx);
 void hns_roce_cleanup_dca_mem(struct hns_roce_context *ctx);
 int hns_roce_add_dca_mem(struct hns_roce_context *ctx, uint32_t size);
 
providers/hns/hns_roce_u_buf.c +96 -0

111 lines not on this page · raw message

providers/hns/hns_roce_u_hw_v2.c +7 -0
index 4988943..16f033c 100644
@@ -654,6 +654,10 @@ static int hns_roce_u_v2_poll_cq(struct ibv_cq *ibvcq, int ne,
 
 	pthread_spin_unlock(&cq->lock);
 
+	/* Try to shrink the DCA mem */
+	if (ctx->dca_ctx.mem_cnt > 0)
+		hns_roce_shrink_dca_mem(ctx);
+
 	return err == V2_CQ_POLL_ERR ? err : npolled;
 }
 
@@ -1524,6 +1528,9 @@ static int hns_roce_u_v2_destroy_qp(struct ibv_qp *ibqp)
 
 	free(qp);
 
+	if (ctx->dca_ctx.mem_cnt > 0)
+		hns_roce_shrink_dca_mem(ctx);
+
 	return ret;
 }
 
Signature · 1 line
2.7.4
author
#4 lore raw

PATCH 4/6 libhns: Add support for attaching QP's WQE buffer

In reply to Weihang Li

From: Xi Wang <[email protected]>

If a uQP works in DCA mode, the WQE's buffer will be split as many blocks and be stored into a list. The blocks are allocated from the DCA's memory pool before posting WRs and are dropped when the QP's CI is equal to PI after polling CQ.

---
 providers/hns/hns_roce_u.h       |  24 +++++-
 providers/hns/hns_roce_u_buf.c   | 156 ++++++++++++++++++++++++++++++++++++++-
 providers/hns/hns_roce_u_hw_v2.c | 123 ++++++++++++++++++++++++++----
 providers/hns/hns_roce_u_hw_v2.h |   7 ++
 providers/hns/hns_roce_u_verbs.c |  32 ++++++--
 5 files changed, 318 insertions(+), 24 deletions(-)
providers/hns/hns_roce_u.h +23 -1
index a8a007e..a488694 100644
@@ -122,6 +122,12 @@ struct hns_roce_device {
 	int				hw_version;
 };
 
+struct hns_roce_buf_list {
+	void **bufs;
+	unsigned int max_cnt;
+	unsigned int shift;
+};
+
 struct hns_roce_buf {
 	void				*buf;
 	unsigned int			length;
@@ -282,9 +288,20 @@ struct hns_roce_rinl_buf {
 	unsigned int			wqe_cnt;
 };
 
+struct hns_roce_dca_attach_attr {
+	uint32_t sq_offset;
+	uint32_t sge_offset;
+	uint32_t rq_offset;
+};
+
+struct hns_roce_dca_detach_attr {
+	uint32_t sq_index;
+};
+
 struct hns_roce_qp {
 	struct verbs_qp			verbs_qp;
 	struct hns_roce_buf		buf;
+	struct hns_roce_buf_list	page_list;
 	int				max_inline_data;
 	int				buf_size;
 	unsigned int			sq_signal_bits;
@@ -331,6 +348,7 @@ struct hns_roce_u_hw {
  * minimum page size.
  */
 #define hr_hw_page_align(x) align(x, HNS_HW_PAGE_SIZE)
+#define hr_hw_page_count(x) (hr_hw_page_align(x) / HNS_HW_PAGE_SIZE)
 
 static inline unsigned int to_hr_hem_entries_size(int count, int buf_shift)
 {
@@ -440,9 +458,13 @@ void hns_roce_free_buf(struct hns_roce_buf *buf);
 
 void hns_roce_free_qp_buf(struct hns_roce_qp *qp, struct hns_roce_context *ctx);
 
+int hns_roce_attach_dca_mem(struct hns_roce_context *ctx, uint32_t handle,
+			    struct hns_roce_dca_attach_attr *attr,
+			    uint32_t size, struct hns_roce_buf_list *buf_list);
+void hns_roce_detach_dca_mem(struct hns_roce_context *ctx, uint32_t handle,
+			     struct hns_roce_dca_detach_attr *attr);
 void hns_roce_shrink_dca_mem(struct hns_roce_context *ctx);
 void hns_roce_cleanup_dca_mem(struct hns_roce_context *ctx);
-int hns_roce_add_dca_mem(struct hns_roce_context *ctx, uint32_t size);
 
 void hns_roce_init_qp_indices(struct hns_roce_qp *qp);
 
providers/hns/hns_roce_u_buf.c +155 -1

171 lines not on this page · raw message

providers/hns/hns_roce_u_hw_v2.c +107 -16

197 lines not on this page · raw message

providers/hns/hns_roce_u_hw_v2.h +7 -0
index c13d82e..be6be73 100644
@@ -239,6 +239,13 @@ struct hns_roce_rc_sq_wqe {
 
 #define RC_SQ_WQE_BYTE_4_RDMA_WRITE_S 22
 
+#define RC_SQ_WQE_BYTE_4_SQPN_L_W 2
+#define RC_SQ_WQE_BYTE_4_SQPN_L_S 5
+#define RC_SQ_WQE_BYTE_4_SQPN_L_M GENMASK(6, 5)
+
+#define RC_SQ_WQE_BYTE_4_SQPN_H_S 13
+#define RC_SQ_WQE_BYTE_4_SQPN_H_M GENMASK(30, 13)
+
 #define RC_SQ_WQE_BYTE_16_XRC_SRQN_S 0
 #define RC_SQ_WQE_BYTE_16_XRC_SRQN_M \
 	(((1UL << 24) - 1) << RC_SQ_WQE_BYTE_16_XRC_SRQN_S)
providers/hns/hns_roce_u_verbs.c +26 -6
index abff092..21e295a 100644
@@ -870,6 +870,14 @@ static int calc_qp_buff_size(struct hns_roce_device *hr_dev,
 	return 0;
 }
 
+static inline bool check_qp_support_dca(bool pool_en, enum ibv_qp_type qp_type)
+{
+	if (pool_en && (qp_type == IBV_QPT_RC || qp_type == IBV_QPT_XRC_SEND))
+		return true;
+
+	return false;
+}
+
 static void qp_free_wqe(struct hns_roce_qp *qp)
 {
 	qp_free_recv_inl_buf(qp);
@@ -881,8 +889,8 @@ static void qp_free_wqe(struct hns_roce_qp *qp)
 	hns_roce_free_buf(&qp->buf);
 }
 
-static int qp_alloc_wqe(struct ibv_qp_cap *cap, struct hns_roce_qp *qp,
-			struct hns_roce_context *ctx)
+static int qp_alloc_wqe(struct ibv_qp_init_attr_ex *attr,
+			struct hns_roce_qp *qp, struct hns_roce_context *ctx)
 {
 	struct hns_roce_device *hr_dev = to_hr_dev(ctx->ibv_ctx.context.device);
 
@@ -900,12 +908,24 @@ static int qp_alloc_wqe(struct ibv_qp_cap *cap, struct hns_roce_qp *qp,
 	}
 
 	if (qp->rq_rinl_buf.wqe_cnt) {
-		if (qp_alloc_recv_inl_buf(cap, qp))
+		if (qp_alloc_recv_inl_buf(&attr->cap, qp))
 			goto err_alloc;
 	}
 
-	if (hns_roce_alloc_buf(&qp->buf, qp->buf_size, HNS_HW_PAGE_SIZE))
-		goto err_alloc;
+	if (check_qp_support_dca(ctx->dca_ctx.max_size != 0, attr->qp_type)) {
+		/* when DCA is enabled, use a buffer list to store page addr */
+		qp->buf.buf = NULL;
+		qp->page_list.max_cnt = hr_hw_page_count(qp->buf_size);
+		qp->page_list.shift = HNS_HW_PAGE_SHIFT;
+		qp->page_list.bufs = calloc(qp->page_list.max_cnt,
+					    sizeof(void *));
+		if (!qp->page_list.bufs)
+			goto err_alloc;
+	} else {
+		if (hns_roce_alloc_buf(&qp->buf, qp->buf_size,
+				       HNS_HW_PAGE_SIZE))
+			goto err_alloc;
+	}
 
 	return 0;
 
@@ -1123,7 +1143,7 @@ static int hns_roce_alloc_qp_buf(struct ibv_qp_init_attr_ex *attr,
 	    pthread_spin_init(&qp->rq.lock, PTHREAD_PROCESS_PRIVATE))
 		return -ENOMEM;
 
-	ret = qp_alloc_wqe(&attr->cap, qp, ctx);
+	ret = qp_alloc_wqe(attr, qp, ctx);
 	if (ret)
 		return ret;
 
Signature · 1 line
2.7.4
author
#5 lore raw

PATCH 5/6 libhns: Add direct verbs support to config DCA

In reply to Weihang Li

From: Xi Wang <[email protected]>

Add two direct verbs to config DCA:
1. hnsdv_open_device() is used to config DCA memory pool.
2. hnsdv_create_qp() is used to create a DCA QP.

---
 debian/control                             |  2 +-
 debian/ibverbs-providers.install           |  1 +
 debian/ibverbs-providers.lintian-overrides |  4 +-
 debian/ibverbs-providers.symbols           |  6 +++
 debian/libibverbs-dev.install              |  4 ++
 providers/hns/CMakeLists.txt               |  9 +++-
 providers/hns/hns_roce_u.c                 | 71 ++++++++++++++++++++++++++----
 providers/hns/hns_roce_u.h                 |  4 +-
 providers/hns/hns_roce_u_abi.h             |  1 +
 providers/hns/hns_roce_u_verbs.c           | 43 ++++++++++++++----
 providers/hns/hnsdv.h                      | 61 +++++++++++++++++++++++++
 providers/hns/libhns.map                   |  9 ++++
 redhat/rdma-core.spec                      |  5 ++-
 suse/rdma-core.spec                        | 21 ++++++++-
 14 files changed, 218 insertions(+), 23 deletions(-)
 create mode 100644 providers/hns/hnsdv.h
 create mode 100644 providers/hns/libhns.map
debian/control +1 -1
index 8cbab0b..1ccb7f4 100644
@@ -94,7 +94,7 @@ Description: User space provider drivers for libibverbs
   - cxgb4: Chelsio T4 iWARP HCAs
   - efa: Amazon Elastic Fabric Adapter
   - hfi1verbs: Intel Omni-Path HFI
-  - hns: HiSilicon Hip06 SoC
+  - hns: HiSilicon Hip06+ SoC
   - i40iw: Intel Ethernet Connection X722 RDMA
   - ipathverbs: QLogic InfiniPath HCAs
   - mlx4: Mellanox ConnectX-3 InfiniBand HCAs
debian/ibverbs-providers.install +1 -0
index 4f971fb..c6ecbbc 100644
@@ -1,5 +1,6 @@
 etc/libibverbs.d/
 usr/lib/*/libefa.so.*
 usr/lib/*/libibverbs/lib*-rdmav*.so
+usr/lib/*/libhns.so.*
 usr/lib/*/libmlx4.so.*
 usr/lib/*/libmlx5.so.*
debian/ibverbs-providers.lintian-overrides +2 -2
index 8a44d54..f6afb70 100644
@@ -1,2 +1,2 @@
-# libefa, libmlx4 and libmlx5 are ibverbs provider that provides more functions.
-ibverbs-providers: package-name-doesnt-match-sonames libefa1 libmlx4-1 libmlx5-1
+# libefa, libhns, libmlx4 and libmlx5 are ibverbs provider that provides more functions.
+ibverbs-providers: package-name-doesnt-match-sonames libefa1 libhns-1 libmlx4-1 libmlx5-1
debian/ibverbs-providers.symbols +6 -0
index 3c75ecc..34bc9d5 100644
@@ -136,3 +136,9 @@ libefa.so.1 ibverbs-providers #MINVER#
  efadv_create_qp_ex@EFA_1.1 26
  efadv_query_device@EFA_1.1 26
  efadv_query_ah@EFA_1.1 26
+libhns.so.1 ibverbs-providers #MINVER#
+* Build-Depends-Package: libibverbs-dev
+ HNS_1.0@HNS_1.0 34
+ hnsdv_is_supported@HNS_1.0 34
+ hnsdv_open_device@HNS_1.0 34
+ hnsdv_create_qp@HNS_1.0 34
debian/libibverbs-dev.install +4 -0
index bc8caa5..7d6e6a2 100644
@@ -1,5 +1,6 @@
 usr/include/infiniband/arch.h
 usr/include/infiniband/efadv.h
+usr/include/infiniband/hnsdv.h
 usr/include/infiniband/ib_user_ioctl_verbs.h
 usr/include/infiniband/mlx4dv.h
 usr/include/infiniband/mlx5_api.h
@@ -14,6 +15,8 @@ usr/include/infiniband/verbs_api.h
 usr/lib/*/lib*-rdmav*.a
 usr/lib/*/libefa.a
 usr/lib/*/libefa.so
+usr/lib/*/libhns.a
+usr/lib/*/libhns.so
 usr/lib/*/libibverbs*.so
 usr/lib/*/libibverbs.a
 usr/lib/*/libmlx4.a
@@ -21,6 +24,7 @@ usr/lib/*/libmlx4.so
 usr/lib/*/libmlx5.a
 usr/lib/*/libmlx5.so
 usr/lib/*/pkgconfig/libefa.pc
+usr/lib/*/pkgconfig/libhns.pc
 usr/lib/*/pkgconfig/libibverbs.pc
 usr/lib/*/pkgconfig/libmlx4.pc
 usr/lib/*/pkgconfig/libmlx5.pc
providers/hns/CMakeLists.txt +8 -1
index 697dbd7..6e602f6 100644
@@ -1,4 +1,5 @@
-rdma_provider(hns
+rdma_shared_provider(hns libhns.map
+  1 1.0.${PACKAGE_VERSION}
   hns_roce_u.c
   hns_roce_u_buf.c
   hns_roce_u_db.c
@@ -6,3 +7,9 @@ rdma_provider(hns
   hns_roce_u_hw_v2.c
   hns_roce_u_verbs.c
 )
+
+publish_headers(infiniband
+	hnsdv.h
+)
+
+rdma_pkg_config("hns" "libibverbs" "${CMAKE_THREAD_LIBS_INIT}")
providers/hns/hns_roce_u.c +63 -8

102 lines not on this page · raw message

providers/hns/hns_roce_u.h +3 -1
index a488694..5c8427a 100644
@@ -157,11 +157,11 @@ struct hns_roce_db_page {
 struct hns_roce_dca_ctx {
 	struct list_head mem_list;
 	pthread_spinlock_t lock;
+	uint32_t unit_size;
 	uint64_t max_size;
 	uint64_t min_size;
 	uint64_t curr_size;
 	int mem_cnt;
-	unsigned int unit_size;
 };
 
 struct hns_roce_context {
@@ -391,6 +391,8 @@ static inline struct hns_roce_ah *to_hr_ah(struct ibv_ah *ibv_ah)
 	return container_of(ibv_ah, struct hns_roce_ah, ibv_ah);
 }
 
+bool is_hns_dev(struct ibv_device *device);
+
 int hns_roce_u_query_device(struct ibv_context *context,
 			    const struct ibv_query_device_ex_input *input,
 			    struct ibv_device_attr_ex *attr, size_t attr_size);
providers/hns/hns_roce_u_abi.h +1 -0
index e56f9d3..92404bc 100644
@@ -36,6 +36,7 @@
 #include <infiniband/kern-abi.h>
 #include <rdma/hns-abi.h>
 #include <kernel-abi/hns-abi.h>
+#include "hnsdv.h"
 
 DECLARE_DRV_CMD(hns_roce_alloc_pd, IB_USER_VERBS_CMD_ALLOC_PD,
 		empty, hns_roce_ib_alloc_pd_resp);
providers/hns/hns_roce_u_verbs.c +35 -8

98 lines not on this page · raw message

providers/hns/hnsdv.h new file +61 -0

61 lines not on this page · raw message

providers/hns/libhns.map new file +9 -0
new file mode 100644
index 0000000..aed491c
@@ -0,0 +1,9 @@
+/* Export symbols should be added below according to
+   Documentation/versioning.md document. */
+HNS_1.0 {
+	global:
+		hnsdv_is_supported;
+		hnsdv_open_device;
+		hnsdv_create_qp;
+	local: *;
+};
redhat/rdma-core.spec +4 -1
index 207859d..e1dda8f 100644
@@ -151,6 +151,8 @@ Provides: libefa = %{version}-%{release}
 Obsoletes: libefa < %{version}-%{release}
 Provides: libhfi1 = %{version}-%{release}
 Obsoletes: libhfi1 < %{version}-%{release}
+Provides: libhns = %{version}-%{release}
+Obsoletes: libhns < %{version}-%{release}
 Provides: libi40iw = %{version}-%{release}
 Obsoletes: libi40iw < %{version}-%{release}
 Provides: libipathverbs = %{version}-%{release}
@@ -178,7 +180,7 @@ Device-specific plug-in ibverbs userspace drivers are included:
 - libcxgb4: Chelsio T4 iWARP HCA
 - libefa: Amazon Elastic Fabric Adapter
 - libhfi1: Intel Omni-Path HFI
-- libhns: HiSilicon Hip06 SoC
+- libhns: HiSilicon Hip06+ SoC
 - libi40iw: Intel Ethernet Connection X722 RDMA
 - libipathverbs: QLogic InfiniPath HCA
 - libmlx4: Mellanox ConnectX-3 InfiniBand HCA
@@ -563,6 +565,7 @@ fi
 %dir %{_sysconfdir}/libibverbs.d
 %dir %{_libdir}/libibverbs
 %{_libdir}/libefa.so.*
+%{_libdir}/libhns.so.*
 %{_libdir}/libibverbs*.so.*
 %{_libdir}/libibverbs/*.so
 %{_libdir}/libmlx5.so.*
suse/rdma-core.spec +20 -1
index db6a361..1c14773 100644
@@ -30,6 +30,7 @@ License:        GPL-2.0-only OR BSD-2-Clause
 Group:          Productivity/Networking/Other
 
 %define efa_so_major    1
+%define hns_so_major    1
 %define verbs_so_major  1
 %define rdmacm_so_major 1
 %define umad_so_major   3
@@ -39,6 +40,7 @@ Group:          Productivity/Networking/Other
 %define mad_major       5
 
 %define  efa_lname    libefa%{efa_so_major}
+%define  hns_lname    libhns%{hns_so_major}
 %define  verbs_lname  libibverbs%{verbs_so_major}
 %define  rdmacm_lname librdmacm%{rdmacm_so_major}
 %define  umad_lname   libibumad%{umad_so_major}
@@ -145,6 +147,7 @@ Requires:       %{umad_lname} = %{version}-%{release}
 Requires:       %{verbs_lname} = %{version}-%{release}
 %if 0%{?dma_coherent}
 Requires:       %{efa_lname} = %{version}-%{release}
+Requires:       %{hns_lname} = %{version}-%{release}
 Requires:       %{mlx4_lname} = %{version}-%{release}
 Requires:       %{mlx5_lname} = %{version}-%{release}
 %endif
@@ -185,6 +188,7 @@ Requires:       %{name}%{?_isa} = %{version}-%{release}
 Obsoletes:      libcxgb4-rdmav2 < %{version}-%{release}
 Obsoletes:      libefa-rdmav2 < %{version}-%{release}
 Obsoletes:      libhfi1verbs-rdmav2 < %{version}-%{release}
+Obsoletes:      libhns-rdmav2 < %{version}-%{release}
 Obsoletes:      libi40iw-rdmav2 < %{version}-%{release}
 Obsoletes:      libipathverbs-rdmav2 < %{version}-%{release}
 Obsoletes:      libmlx4-rdmav2 < %{version}-%{release}
@@ -194,6 +198,7 @@ Obsoletes:      libocrdma-rdmav2 < %{version}-%{release}
 Obsoletes:      librxe-rdmav2 < %{version}-%{release}
 %if 0%{?dma_coherent}
 Requires:       %{efa_lname} = %{version}-%{release}
+Requires:       %{hns_lname} = %{version}-%{release}
 Requires:       %{mlx4_lname} = %{version}-%{release}
 Requires:       %{mlx5_lname} = %{version}-%{release}
 %endif
@@ -212,7 +217,7 @@ Device-specific plug-in ibverbs userspace drivers are included:
 - libcxgb4: Chelsio T4 iWARP HCA
 - libefa: Amazon Elastic Fabric Adapter
 - libhfi1: Intel Omni-Path HFI
-- libhns: HiSilicon Hip06 SoC
+- libhns: HiSilicon Hip06+ SoC
 - libi40iw: Intel Ethernet Connection X722 RDMA
 - libipathverbs: QLogic InfiniPath HCA
 - libmlx4: Mellanox ConnectX-3 InfiniBand HCA
@@ -239,6 +244,13 @@ Group:          System/Libraries
 %description -n %efa_lname
 This package contains the efa runtime library.
 
+%package -n %hns_lname
+Summary:        HNS runtime library
+Group:          System/Libraries
+
+%description -n %hns_lname
+This package contains the hns runtime library.
+
 %package -n %mlx4_lname
 Summary:        MLX4 runtime library
 Group:          System/Libraries
@@ -482,6 +494,9 @@ rm -rf %{buildroot}/%{_sbindir}/srp_daemon.sh
 %post -n %efa_lname -p /sbin/ldconfig
 %postun -n %efa_lname -p /sbin/ldconfig
 
+%post -n %hns_lname -p /sbin/ldconfig
+%postun -n %hns_lname -p /sbin/ldconfig
+
 %post -n %mlx4_lname -p /sbin/ldconfig
 %postun -n %mlx4_lname -p /sbin/ldconfig
 
@@ -664,6 +679,10 @@ rm -rf %{buildroot}/%{_sbindir}/srp_daemon.sh
 %defattr(-,root,root)
 %{_libdir}/libefa*.so.*
 
+%files -n %hns_lname
+%defattr(-,root,root)
+%{_libdir}/libhns*.so.*
+
 %files -n %mlx4_lname
 %defattr(-,root,root)
 %{_libdir}/libmlx4*.so.*
Signature · 1 line
2.7.4
author
#6 lore raw

PATCH 6/6 libhns: Add man pages to introduce DCA feature

In reply to Weihang Li

From: Xi Wang <[email protected]>

Document hns DCA feature and related direct verbs.

---
 CMakeLists.txt                            |  1 +
 debian/ibverbs-providers.install          |  2 +-
 debian/libibverbs-dev.install             |  2 +
 providers/hns/man/CMakeLists.txt          |  7 ++++
 providers/hns/man/hns_dca.7.md            | 35 ++++++++++++++++
 providers/hns/man/hnsdv.7.md              | 34 +++++++++++++++
 providers/hns/man/hnsdv_create_qp.3.md    | 69 ++++++++++++++++++++++++++++++
 providers/hns/man/hnsdv_is_supported.3.md | 39 +++++++++++++++++
 providers/hns/man/hnsdv_open_device.3.md  | 70 +++++++++++++++++++++++++++++++
 redhat/rdma-core.spec                     |  2 +
 10 files changed, 260 insertions(+), 1 deletion(-)
 create mode 100644 providers/hns/man/CMakeLists.txt
 create mode 100644 providers/hns/man/hns_dca.7.md
 create mode 100644 providers/hns/man/hnsdv.7.md
 create mode 100644 providers/hns/man/hnsdv_create_qp.3.md
 create mode 100644 providers/hns/man/hnsdv_is_supported.3.md
 create mode 100644 providers/hns/man/hnsdv_open_device.3.md
CMakeLists.txt +1 -0
index 74293bf..744179d 100644
@@ -669,6 +669,7 @@ add_subdirectory(providers/cxgb4) # NO SPARSE
 add_subdirectory(providers/efa)
 add_subdirectory(providers/efa/man)
 add_subdirectory(providers/hns)
+add_subdirectory(providers/hns/man)
 add_subdirectory(providers/i40iw) # NO SPARSE
 add_subdirectory(providers/mlx4)
 add_subdirectory(providers/mlx4/man)
debian/ibverbs-providers.install +1 -1
index c6ecbbc..c4c4c11 100644
@@ -1,6 +1,6 @@
 etc/libibverbs.d/
 usr/lib/*/libefa.so.*
-usr/lib/*/libibverbs/lib*-rdmav*.so
 usr/lib/*/libhns.so.*
+usr/lib/*/libibverbs/lib*-rdmav*.so
 usr/lib/*/libmlx4.so.*
 usr/lib/*/libmlx5.so.*
debian/libibverbs-dev.install +2 -0
index 7d6e6a2..89a02a8 100644
@@ -29,11 +29,13 @@ usr/lib/*/pkgconfig/libibverbs.pc
 usr/lib/*/pkgconfig/libmlx4.pc
 usr/lib/*/pkgconfig/libmlx5.pc
 usr/share/man/man3/efadv_*.3
+usr/share/man/man3/hns*.3
 usr/share/man/man3/ibv_*
 usr/share/man/man3/mbps_to_ibv_rate.3
 usr/share/man/man3/mlx4dv_*.3
 usr/share/man/man3/mlx5dv_*.3
 usr/share/man/man3/mult_to_ibv_rate.3
 usr/share/man/man7/efadv.7
+usr/share/man/man3/hns*.7
 usr/share/man/man7/mlx4dv.7
 usr/share/man/man7/mlx5dv.7
providers/hns/man/CMakeLists.txt new file +7 -0
new file mode 100644
index 0000000..b375a65
@@ -0,0 +1,7 @@
+rdma_man_pages(
+  hnsdv_is_supported.3.md
+  hnsdv_open_device.3.md
+  hnsdv_create_qp.3.md
+  hnsdv.7
+  hns_dca.7
+)
providers/hns/man/hns_dca.7.md new file +35 -0
new file mode 100644
index 0000000..de26d07
@@ -0,0 +1,35 @@
+---
+layout: page
+title: DCA
+section: 7
+tagline: DCA
+date: 2021-03-03
+header: "HNS DCA Manual"
+footer: hns
+---
+
+# NAME
+
+DCA - Dynamic Context Attachment
+
+This allows all WQEs to share a memory pool that belongs to the user context.
+
+# DESCRIPTION
+
+The DCA feature aims to reduce memory consumption by sharing WQE memory for QPs working in sparse traffic scenarios.
+
+The DCA memory pool consists of multiple umem objects. Each umem object is a buffer allocated in user driver and register into kernel driver. The ULP need to setup the memory pool's parameter by calling hnsdv_open_device() and the driver will expand or shrink the memory pool based on this parameter.
+
+When a QP's DCA was enabled by setting create flags through ibv_create_qp_ex(), the WQE buffer will not be allocated directly until the ULP invokes the ibv_post_xxx(). If the memory in the pool is insufficient and the capacity expansion conditions are met, the driver will add new umem objects to the pool.
+
+When all WQEs of a QP are not used by the ROCEE after ibv_poll_cq() or ibv_modify_qp() are invoked, the WQE buffer will be reclaimed to the DCA memory pool. If the free memory in the pool meets the shrink conditions, the driver will delete the unused umem object.
+
+# SEE ALSO
+
+*hnsdv_open_device(3)*, *hnsdv_create_qp(3)*
+
+# AUTHORS
+
+Xi Wang <[email protected]>
+
+Weihang Li <[email protected]>
providers/hns/man/hnsdv.7.md new file +34 -0
new file mode 100644
index 0000000..ada73ec
@@ -0,0 +1,34 @@
+---
+layout: page
+title: HNSDV
+section: 7
+tagline: Verbs
+date: 2021-03-03
+header: "HNS Direct Verbs Manual"
+footer: hns
+---
+
+# NAME
+
+hnsdv - Direct verbs for hns devices
+
+This provides low level access to hns devices to perform direct operations,
+without general branching performed by libibverbs.
+
+# DESCRIPTION
+
+The libibverbs API is an abstract one. It is agnostic to any underlying provider specific implementation. While this abstraction has the advantage of user applications portability, it has a performance penalty. For some applications optimizing performance is more important than portability.
+
+The hns direct verbs API is intended for such applications. It exposes hns specific low level operations, allowing the application to bypass the libibverbs API.
+
+The direct include of hnsdv.h together with linkage to hns library will allow usage of this new interface.
+
+# SEE ALSO
+
+**verbs**(7)
+
+# AUTHORS
+
+Xi Wang <[email protected]>
+
+Weihang Li <[email protected]>
providers/hns/man/hnsdv_create_qp.3.md new file +69 -0

69 lines not on this page · raw message

providers/hns/man/hnsdv_is_supported.3.md new file +39 -0
new file mode 100644
index 0000000..b5f00bd
@@ -0,0 +1,39 @@
+---
+layout: page
+title: hnsdv_is_supported
+section: 3
+tagline: Verbs
+---
+
+# NAME
+
+hnsdv_is_supported - Check whether an RDMA device implemented by the hns provider
+
+# SYNOPSIS
+
+```c
+#include <infiniband/hnsdv.h>
+
+bool hnsdv_is_supported(struct ibv_device *device);
+```
+
+# DESCRIPTION
+
+hnsdv functions may be called only if this function returns true for the RDMA device.
+
+# ARGUMENTS
+
+*device*
+:	RDMA device to check.
+
+# RETURN VALUE
+
+Returns true if device is implemented by hns provider.
+
+# SEE ALSO
+
+*hnsdv(7)*
+
+# AUTHOR
+
+Xi Wang <[email protected]>
providers/hns/man/hnsdv_open_device.3.md new file +70 -0

70 lines not on this page · raw message

redhat/rdma-core.spec +2 -0
index e1dda8f..fcbec50 100644
@@ -440,6 +440,7 @@ fi
 %{_libdir}/lib*.so
 %{_libdir}/pkgconfig/*.pc
 %{_mandir}/man3/efadv*
+%{_mandir}/man3/hns*
 %{_mandir}/man3/ibv_*
 %{_mandir}/man3/rdma*
 %{_mandir}/man3/umad*
@@ -448,6 +449,7 @@ fi
 %{_mandir}/man3/mlx5dv*
 %{_mandir}/man3/mlx4dv*
 %{_mandir}/man7/efadv*
+%{_mandir}/man7/hns*
 %{_mandir}/man7/mlx5dv*
 %{_mandir}/man7/mlx4dv*
 %{_mandir}/man3/ibnd_*
Signature · 1 line
2.7.4
maintainer
#7 lore raw
In reply to Weihang Li
Quoted 26 lines · Weihang Li

The HIP09 introduces the DCA(Dynamic Context Attachment) feature which supports many RC QPs to share the WQE buffer in a memory pool. If a QP enables DCA feature, the WQE's buffer will not be allocated when creating but when the users start to post WRs. This will reduce the memory consumption when there are too many QPs are inactive.

For more detailed information, please refer to the man pages provided by the last patch of this series.

This series is associated with the kernel one "RDMA/hns: Add support for Dynamic Context Attachment", and two RFC versions of this series has been sent before.

No changes since RFC v2.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Changes since RFC v1:
* Add direct verbs to set the parameters about size that used to
configuring DCA.
* Add man pages to explain what is DCA, how does it works and how to use
it.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Weihang Li (1):
Update kernel headers

Quoted 6 lines

Xi Wang (5):
libhns: Introduce DCA for RC QP
libhns: Add support for shrinking DCA memory pool
libhns: Add support for attaching QP's WQE buffer
libhns: Add direct verbs support to config DCA
libhns: Add man pages to introduce DCA feature

This should be sent on GitHub

Jason

author
#8 lore raw
Quoted 34 lines · Jason Gunthorpe
Quoted 26 lines · Weihang Li

The HIP09 introduces the DCA(Dynamic Context Attachment) feature which supports many RC QPs to share the WQE buffer in a memory pool. If a QP enables DCA feature, the WQE's buffer will not be allocated when creating but when the users start to post WRs. This will reduce the memory consumption when there are too many QPs are inactive.

For more detailed information, please refer to the man pages provided by the last patch of this series.

This series is associated with the kernel one "RDMA/hns: Add support for Dynamic Context Attachment", and two RFC versions of this series has been sent before.

No changes since RFC v2.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Changes since RFC v1:
* Add direct verbs to set the parameters about size that used to
configuring DCA.
* Add man pages to explain what is DCA, how does it works and how to use
it.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[email protected]/

Weihang Li (1):
Update kernel headers

Quoted 6 lines

Xi Wang (5):
libhns: Introduce DCA for RC QP
libhns: Add support for shrinking DCA memory pool
libhns: Add support for attaching QP's WQE buffer
libhns: Add direct verbs support to config DCA
libhns: Add man pages to introduce DCA feature

Quoted 4 lines

This should be sent on GitHub

Jason

The implementation needs some modification, so I haven't sent it to github yet. The next version will be sent to github at the same time, thank you.

Weihang