rdma-core/0035-libhns-Use-shared-memory-to-sync-DCA-status.patch

From a5e62921afc2fcc152e8b0584f2d04d1a4db4f10 Mon Sep 17 00:00:00 2001
From: Chengchang Tang <tangchengchang@huawei.com>
Date: Tue, 29 Jun 2021 20:06:47 +0800
Subject: libhns: Use shared memory to sync DCA status

driver inclusion
category: feature
bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/I63L1M

----------------------------------------------------------

The user DCA needs to check the QP attaching state before filling wqe
buffer by the response from uverbs 'HNS_IB_METHOD_DCA_MEM_ATTACH', but
this will result in too much time being wasted on system calls, so use a
shared table between user driver and kernel driver to sync DCA status.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Reviewed-by: Yangyang Li <liyangyang20@huawei.com>
---
 providers/hns/hns_roce_u.c | 51 +++++++++++++++++++++++++++++++++++---
 providers/hns/hns_roce_u.h | 10 ++++++++
 2 files changed, 57 insertions(+), 4 deletions(-)

diff --git a/providers/hns/hns_roce_u.c b/providers/hns/hns_roce_u.c
index bd2b251..fe30cda 100644
--- a/providers/hns/hns_roce_u.c
+++ b/providers/hns/hns_roce_u.c
@@ -95,9 +95,33 @@ static const struct verbs_context_ops hns_common_ops = {
 	.alloc_parent_domain = hns_roce_u_alloc_pad,
 };
 
-static int init_dca_context(struct hns_roce_context *ctx, int page_size)
+static int mmap_dca(struct hns_roce_context *ctx, int cmd_fd,
+		    int page_size, size_t size, uint64_t mmap_key)
 {
 	struct hns_roce_dca_ctx *dca_ctx = &ctx->dca_ctx;
+	void *addr;
+
+	addr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, cmd_fd,
+		    mmap_key);
+	if (addr == MAP_FAILED) {
+		verbs_err(&ctx->ibv_ctx, "failed to mmap() dca prime qp.\n");
+		return -EINVAL;
+	}
+
+	dca_ctx->buf_status = addr;
+	dca_ctx->sync_status = addr + size / 2;
+
+	return 0;
+}
+
+static int init_dca_context(struct hns_roce_context *ctx, int cmd_fd,
+			    struct hns_roce_alloc_ucontext_resp *resp,
+			    int page_size)
+{
+	struct hns_roce_dca_ctx *dca_ctx = &ctx->dca_ctx;
+	uint64_t mmap_key = resp->dca_mmap_key;
+	int mmap_size = resp->dca_mmap_size;
+	int max_qps = resp->dca_qps;
 	int ret;
 
 	if (!(ctx->config & HNS_ROCE_UCTX_RSP_DCA_FLAGS))
@@ -112,6 +136,16 @@ static int init_dca_context(struct hns_roce_context *ctx, int page_size)
 	dca_ctx->max_size = HNS_DCA_MAX_MEM_SIZE;
 	dca_ctx->mem_cnt = 0;
 
+	if (mmap_key) {
+		const unsigned int bits_per_qp = 2 * HNS_DCA_BITS_PER_STATUS;
+
+		if (!mmap_dca(ctx, cmd_fd, page_size, mmap_size, mmap_key)) {
+			dca_ctx->status_size = mmap_size;
+			dca_ctx->max_qps = min_t(int, max_qps,
+						 mmap_size *  8 / bits_per_qp);
+		}
+	}
+
 	return 0;
 }
 
@@ -125,6 +159,8 @@ static void uninit_dca_context(struct hns_roce_context *ctx)
 	pthread_spin_lock(&dca_ctx->lock);
 	hns_roce_cleanup_dca_mem(ctx);
 	pthread_spin_unlock(&dca_ctx->lock);
+	if (dca_ctx->buf_status)
+		munmap(dca_ctx->buf_status, dca_ctx->status_size);
 
 	pthread_spin_destroy(&dca_ctx->lock);
 }
@@ -149,6 +185,14 @@ static uint32_t calc_table_shift(uint32_t entry_count, uint32_t size_shift)
 	return count_shift > size_shift ? count_shift - size_shift : 0;
 }
 
+static void ucontext_set_cmd(struct hns_roce_alloc_ucontext *cmd, int page_size)
+{
+	cmd->config |= HNS_ROCE_EXSGE_FLAGS | HNS_ROCE_RQ_INLINE_FLAGS |
+		       HNS_ROCE_CQE_INLINE_FLAGS | HNS_ROCE_UCTX_CONFIG_DCA;
+	cmd->comp = HNS_ROCE_ALLOC_UCTX_COMP_DCA_MAX_QPS;
+	cmd->dca_max_qps = page_size * 8 / 2 * HNS_DCA_BITS_PER_STATUS;
+}
+
 static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 						    int cmd_fd,
 						    void *private_data)
@@ -165,8 +209,7 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 	if (!context)
 		return NULL;
 
-	cmd.config |= HNS_ROCE_EXSGE_FLAGS | HNS_ROCE_RQ_INLINE_FLAGS |
-		      HNS_ROCE_CQE_INLINE_FLAGS | HNS_ROCE_UCTX_CONFIG_DCA;
+	ucontext_set_cmd(&cmd, hr_dev->page_size);
 	if (ibv_cmd_get_context(&context->ibv_ctx, &cmd.ibv_cmd, sizeof(cmd),
 				&resp.ibv_resp, sizeof(resp)))
 		goto err_free;
@@ -212,7 +255,7 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev,
 	context->max_srq_wr = dev_attrs.max_srq_wr;
 	context->max_srq_sge = dev_attrs.max_srq_sge;
 
-	if (init_dca_context(context, hr_dev->page_size))
+	if (init_dca_context(context, cmd_fd, &resp, hr_dev->page_size))
 		goto err_free;
 
 	if (hns_roce_mmap(hr_dev, context, cmd_fd))
diff --git a/providers/hns/hns_roce_u.h b/providers/hns/hns_roce_u.h
index 44a733f..a8f811e 100644
--- a/providers/hns/hns_roce_u.h
+++ b/providers/hns/hns_roce_u.h
@@ -35,6 +35,7 @@
 
 #include <stddef.h>
 #include <endian.h>
+#include <stdatomic.h>
 #include <util/compiler.h>
 
 #include <infiniband/driver.h>
@@ -44,6 +45,7 @@
 #include <ccan/array_size.h>
 #include <util/bitmap.h>
 #include <ccan/container_of.h>
+#include <ccan/minmax.h>
 #include <linux/if_ether.h>
 #include "hns_roce_u_abi.h"
 
@@ -52,6 +54,8 @@
 
 #define PFX				"hns: "
 
+typedef _Atomic(uint64_t) atomic_bitmap_t;
+
 /* The minimum page size is 4K for hardware */
 #define HNS_HW_PAGE_SHIFT 12
 #define HNS_HW_PAGE_SIZE (1 << HNS_HW_PAGE_SHIFT)
@@ -214,6 +218,12 @@ struct hns_roce_dca_ctx {
 	uint64_t max_size;
 	uint64_t min_size;
 	uint64_t curr_size;
+
+#define HNS_DCA_BITS_PER_STATUS 1
+	unsigned int max_qps;
+	unsigned int status_size;
+	atomic_bitmap_t *buf_status;
+	atomic_bitmap_t *sync_status;
 };
 
 struct hns_roce_context {
-- 
2.30.0