From patchwork Tue Jul 3 03:10:09 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Lijun Ou X-Patchwork-Id: 10502835 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork.web.codeaurora.org (Postfix) with ESMTP id 8DA926028F for ; Tue, 3 Jul 2018 02:38:47 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 7AE0128A21 for ; Tue, 3 Jul 2018 02:38:47 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 6E80328A27; Tue, 3 Jul 2018 02:38:47 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.9 required=2.0 tests=BAYES_00, MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 5691B28A21 for ; Tue, 3 Jul 2018 02:38:46 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932694AbeGCCio (ORCPT ); Mon, 2 Jul 2018 22:38:44 -0400 Received: from szxga07-in.huawei.com ([45.249.212.35]:49661 "EHLO huawei.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S932664AbeGCCil (ORCPT ); Mon, 2 Jul 2018 22:38:41 -0400 Received: from DGGEMS402-HUB.china.huawei.com (unknown [172.30.72.58]) by Forcepoint Email with ESMTP id 91BCC1287E6B7; Tue, 3 Jul 2018 10:38:37 +0800 (CST) Received: from linux-ioko.site (10.71.200.31) by DGGEMS402-HUB.china.huawei.com (10.3.19.202) with Microsoft SMTP Server id 14.3.382.0; Tue, 3 Jul 2018 10:38:33 +0800 From: Lijun Ou To: , CC: , Subject: [PATCH for-next 1/4] RDMA/hns: Add TSQ link table support Date: Tue, 3 Jul 2018 11:10:09 +0800 Message-ID: <1530587412-17866-2-git-send-email-oulijun@huawei.com> X-Mailer: git-send-email 1.9.1 In-Reply-To: <1530587412-17866-1-git-send-email-oulijun@huawei.com> References: <1530587412-17866-1-git-send-email-oulijun@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.71.200.31] X-CFilter-Loop: Reflected Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP In hip08, TSQ(Transport Service Queue) should be extended to host memory to store the doorbells. This patch adds the support of creating TSQ, and then configured to the hardware. Signed-off-by: Yixian Liu Signed-off-by: Lijun Ou Signed-off-by: Wei Hu (Xavier) --- drivers/infiniband/hw/hns/hns_roce_device.h | 2 + drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 217 +++++++++++++++++++++++++++- drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 85 ++++++++++- 3 files changed, 297 insertions(+), 7 deletions(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h index 65f7b68..d34ed42 100644 --- a/drivers/infiniband/hw/hns/hns_roce_device.h +++ b/drivers/infiniband/hw/hns/hns_roce_device.h @@ -720,6 +720,8 @@ struct hns_roce_caps { u32 eqe_ba_pg_sz; u32 eqe_buf_pg_sz; u32 eqe_hop_num; + u32 sl_num; + u32 tsq_buf_pg_sz; u32 chunk_sz; /* chunk size in non multihop mode*/ u64 flags; }; diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 454d391..c189626 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -925,7 +925,8 @@ static int hns_roce_config_global_param(struct hns_roce_dev *hr_dev) static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev) { struct hns_roce_cmq_desc desc[2]; - struct hns_roce_pf_res *res; + struct hns_roce_pf_res_a *req_a; + struct hns_roce_pf_res_b *req_b; int ret; int i; @@ -943,21 +944,26 @@ static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev) if (ret) return ret; - res = (struct hns_roce_pf_res *)desc[0].data; + req_a = (struct hns_roce_pf_res_a *)desc[0].data; + req_b = (struct hns_roce_pf_res_b *)desc[1].data; - hr_dev->caps.qpc_bt_num = roce_get_field(res->qpc_bt_idx_num, + hr_dev->caps.qpc_bt_num = roce_get_field(req_a->qpc_bt_idx_num, PF_RES_DATA_1_PF_QPC_BT_NUM_M, PF_RES_DATA_1_PF_QPC_BT_NUM_S); - hr_dev->caps.srqc_bt_num = roce_get_field(res->srqc_bt_idx_num, + hr_dev->caps.srqc_bt_num = roce_get_field(req_a->srqc_bt_idx_num, PF_RES_DATA_2_PF_SRQC_BT_NUM_M, PF_RES_DATA_2_PF_SRQC_BT_NUM_S); - hr_dev->caps.cqc_bt_num = roce_get_field(res->cqc_bt_idx_num, + hr_dev->caps.cqc_bt_num = roce_get_field(req_a->cqc_bt_idx_num, PF_RES_DATA_3_PF_CQC_BT_NUM_M, PF_RES_DATA_3_PF_CQC_BT_NUM_S); - hr_dev->caps.mpt_bt_num = roce_get_field(res->mpt_bt_idx_num, + hr_dev->caps.mpt_bt_num = roce_get_field(req_a->mpt_bt_idx_num, PF_RES_DATA_4_PF_MPT_BT_NUM_M, PF_RES_DATA_4_PF_MPT_BT_NUM_S); + hr_dev->caps.sl_num = roce_get_field(req_b->qid_idx_sl_num, + PF_RES_DATA_3_PF_SL_NUM_M, + PF_RES_DATA_3_PF_SL_NUM_S); + return 0; } @@ -1203,6 +1209,7 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) caps->eqe_ba_pg_sz = 0; caps->eqe_buf_pg_sz = 0; caps->eqe_hop_num = HNS_ROCE_EQE_HOP_NUM; + caps->tsq_buf_pg_sz = 0; caps->chunk_sz = HNS_ROCE_V2_TABLE_CHUNK_SIZE; caps->flags = HNS_ROCE_CAP_FLAG_REREG_MR | @@ -1224,6 +1231,202 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) return ret; } +static int hns_roce_config_link_table(struct hns_roce_dev *hr_dev, + enum hns_roce_link_table_type type) +{ + struct hns_roce_cmq_desc desc[2]; + struct hns_roce_cfg_llm_a *req_a = + (struct hns_roce_cfg_llm_a *)desc[0].data; + struct hns_roce_cfg_llm_b *req_b = + (struct hns_roce_cfg_llm_b *)desc[1].data; + struct hns_roce_v2_priv *priv = hr_dev->priv; + struct hns_roce_link_table *link_tbl; + struct hns_roce_link_table_entry *entry; + enum hns_roce_opcode_type opcode; + u32 page_num; + int i; + + switch (type) { + case TSQ_LINK_TABLE: + link_tbl = &priv->tsq; + opcode = HNS_ROCE_OPC_CFG_EXT_LLM; + break; + default: + return -EINVAL; + } + + page_num = link_tbl->npages; + entry = link_tbl->table.buf; + memset(req_a, 0, sizeof(*req_a)); + memset(req_b, 0, sizeof(*req_b)); + + for (i = 0; i < 2; i++) { + hns_roce_cmq_setup_basic_desc(&desc[i], opcode, false); + + if (i == 0) + desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT); + else + desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT); + + if (i == 0) { + req_a->base_addr_l = link_tbl->table.map & 0xffffffff; + req_a->base_addr_h = (link_tbl->table.map >> 32) & + 0xffffffff; + roce_set_field(req_a->depth_pgsz_init_en, + CFG_LLM_QUE_DEPTH_M, + CFG_LLM_QUE_DEPTH_S, + link_tbl->npages); + roce_set_field(req_a->depth_pgsz_init_en, + CFG_LLM_QUE_PGSZ_M, + CFG_LLM_QUE_PGSZ_S, + link_tbl->pg_sz); + req_a->head_ba_l = entry[0].blk_ba0; + req_a->head_ba_h_nxtptr = entry[0].blk_ba1_nxt_ptr; + roce_set_field(req_a->head_ptr, + CFG_LLM_HEAD_PTR_M, + CFG_LLM_HEAD_PTR_S, 0); + } else { + req_b->tail_ba_l = entry[page_num - 1].blk_ba0; + roce_set_field(req_b->tail_ba_h, + CFG_LLM_TAIL_BA_H_M, + CFG_LLM_TAIL_BA_H_S, + entry[page_num - 1].blk_ba1_nxt_ptr & + HNS_ROCE_LINK_TABLE_BA1_M); + roce_set_field(req_b->tail_ptr, + CFG_LLM_TAIL_PTR_M, + CFG_LLM_TAIL_PTR_S, + (entry[page_num - 2].blk_ba1_nxt_ptr & + HNS_ROCE_LINK_TABLE_NXT_PTR_M) >> + HNS_ROCE_LINK_TABLE_NXT_PTR_S); + } + } + roce_set_field(req_a->depth_pgsz_init_en, + CFG_LLM_INIT_EN_M, CFG_LLM_INIT_EN_S, 1); + + return hns_roce_cmq_send(hr_dev, desc, 2); +} + +static int hns_roce_init_link_table(struct hns_roce_dev *hr_dev, + enum hns_roce_link_table_type type) +{ + struct hns_roce_v2_priv *priv = hr_dev->priv; + struct hns_roce_link_table *link_tbl; + struct hns_roce_link_table_entry *entry; + struct device *dev = hr_dev->dev; + u32 buf_chk_sz; + dma_addr_t t; + int pg_num_a; + int pg_num_b; + int pg_num; + int size; + int i; + + switch (type) { + case TSQ_LINK_TABLE: + link_tbl = &priv->tsq; + buf_chk_sz = 1 << (hr_dev->caps.tsq_buf_pg_sz + PAGE_SHIFT); + pg_num_a = hr_dev->caps.num_qps * 8 / buf_chk_sz; + pg_num_b = hr_dev->caps.sl_num * 4 + 2; + break; + default: + return -EINVAL; + } + + pg_num = max(pg_num_a, pg_num_b); + size = pg_num * sizeof(struct hns_roce_link_table_entry); + + link_tbl->table.buf = dma_alloc_coherent(dev, size, + &link_tbl->table.map, + GFP_KERNEL); + if (!link_tbl->table.buf) + goto out; + + link_tbl->pg_list = kcalloc(pg_num, sizeof(*link_tbl->pg_list), + GFP_KERNEL); + if (!link_tbl->pg_list) + goto err_kcalloc_failed; + + entry = link_tbl->table.buf; + for (i = 0; i < pg_num; ++i) { + link_tbl->pg_list[i].buf = dma_alloc_coherent(dev, buf_chk_sz, + &t, GFP_KERNEL); + if (!link_tbl->pg_list[i].buf) + goto err_alloc_buf_failed; + + link_tbl->pg_list[i].map = t; + memset(link_tbl->pg_list[i].buf, 0, buf_chk_sz); + + entry[i].blk_ba0 = (t >> 12) & 0xffffffff; + roce_set_field(entry[i].blk_ba1_nxt_ptr, + HNS_ROCE_LINK_TABLE_BA1_M, + HNS_ROCE_LINK_TABLE_BA1_S, + t >> 44); + + if (i < (pg_num - 1)) + roce_set_field(entry[i].blk_ba1_nxt_ptr, + HNS_ROCE_LINK_TABLE_NXT_PTR_M, + HNS_ROCE_LINK_TABLE_NXT_PTR_S, + i + 1); + } + link_tbl->npages = pg_num; + link_tbl->pg_sz = buf_chk_sz; + + return hns_roce_config_link_table(hr_dev, type); + +err_alloc_buf_failed: + for (i -= 1; i >= 0; i--) + dma_free_coherent(dev, buf_chk_sz, + link_tbl->pg_list[i].buf, + link_tbl->pg_list[i].map); + kfree(link_tbl->pg_list); + +err_kcalloc_failed: + dma_free_coherent(dev, size, link_tbl->table.buf, + link_tbl->table.map); + +out: + return -ENOMEM; +} + +static void hns_roce_free_link_table(struct hns_roce_dev *hr_dev, + struct hns_roce_link_table *link_tbl) +{ + struct device *dev = hr_dev->dev; + int size; + int i; + + size = link_tbl->npages * sizeof(struct hns_roce_link_table_entry); + + for (i = 0; i < link_tbl->npages; ++i) + if (link_tbl->pg_list[i].buf) + dma_free_coherent(dev, link_tbl->pg_sz, + link_tbl->pg_list[i].buf, + link_tbl->pg_list[i].map); + kfree(link_tbl->pg_list); + + dma_free_coherent(dev, size, link_tbl->table.buf, + link_tbl->table.map); +} + +static int hns_roce_v2_init(struct hns_roce_dev *hr_dev) +{ + int ret; + + /* TSQ includes SQ doorbell and ack doorbell */ + ret = hns_roce_init_link_table(hr_dev, TSQ_LINK_TABLE); + if (ret) + dev_err(hr_dev->dev, "TSQ init failed, ret = %d.\n", ret); + + return ret; +} + +static void hns_roce_v2_exit(struct hns_roce_dev *hr_dev) +{ + struct hns_roce_v2_priv *priv = hr_dev->priv; + + hns_roce_free_link_table(hr_dev, &priv->tsq); +} + static int hns_roce_v2_cmd_pending(struct hns_roce_dev *hr_dev) { u32 status = readl(hr_dev->reg_base + ROCEE_VF_MB_STATUS_REG); @@ -4725,6 +4928,8 @@ static void hns_roce_v2_cleanup_eq_table(struct hns_roce_dev *hr_dev) .cmq_init = hns_roce_v2_cmq_init, .cmq_exit = hns_roce_v2_cmq_exit, .hw_profile = hns_roce_v2_profile, + .hw_init = hns_roce_v2_init, + .hw_exit = hns_roce_v2_exit, .post_mbox = hns_roce_v2_post_mbox, .chk_mbox = hns_roce_v2_chk_mbox, .set_gid = hns_roce_v2_set_gid, diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h index d47675f..18626d3 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h @@ -203,6 +203,7 @@ enum hns_roce_opcode_type { HNS_ROCE_OPC_ALLOC_PF_RES = 0x8004, HNS_ROCE_OPC_QUERY_PF_RES = 0x8400, HNS_ROCE_OPC_ALLOC_VF_RES = 0x8401, + HNS_ROCE_OPC_CFG_EXT_LLM = 0x8403, HNS_ROCE_OPC_CFG_BT_ATTR = 0x8506, }; @@ -1061,6 +1062,40 @@ struct hns_roce_query_version { __le32 rsv[5]; }; +struct hns_roce_cfg_llm_a { + __le32 base_addr_l; + __le32 base_addr_h; + __le32 depth_pgsz_init_en; + __le32 head_ba_l; + __le32 head_ba_h_nxtptr; + __le32 head_ptr; +}; + +#define CFG_LLM_QUE_DEPTH_S 0 +#define CFG_LLM_QUE_DEPTH_M GENMASK(12, 0) + +#define CFG_LLM_QUE_PGSZ_S 16 +#define CFG_LLM_QUE_PGSZ_M GENMASK(19, 16) + +#define CFG_LLM_INIT_EN_S 20 +#define CFG_LLM_INIT_EN_M GENMASK(20, 20) + +#define CFG_LLM_HEAD_PTR_S 0 +#define CFG_LLM_HEAD_PTR_M GENMASK(11, 0) + +struct hns_roce_cfg_llm_b { + __le32 tail_ba_l; + __le32 tail_ba_h; + __le32 tail_ptr; + __le32 rsv[3]; +}; + +#define CFG_LLM_TAIL_BA_H_S 0 +#define CFG_LLM_TAIL_BA_H_M GENMASK(19, 0) + +#define CFG_LLM_TAIL_PTR_S 0 +#define CFG_LLM_TAIL_PTR_M GENMASK(11, 0) + struct hns_roce_cfg_global_param { __le32 time_cfg_udp_port; __le32 rsv[5]; @@ -1072,7 +1107,7 @@ struct hns_roce_cfg_global_param { #define CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_S 16 #define CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_M GENMASK(31, 16) -struct hns_roce_pf_res { +struct hns_roce_pf_res_a { __le32 rsv; __le32 qpc_bt_idx_num; __le32 srqc_bt_idx_num; @@ -1111,6 +1146,32 @@ struct hns_roce_pf_res { #define PF_RES_DATA_5_PF_EQC_BT_NUM_S 16 #define PF_RES_DATA_5_PF_EQC_BT_NUM_M GENMASK(25, 16) +struct hns_roce_pf_res_b { + __le32 rsv0; + __le32 smac_idx_num; + __le32 sgid_idx_num; + __le32 qid_idx_sl_num; + __le32 rsv[2]; +}; + +#define PF_RES_DATA_1_PF_SMAC_IDX_S 0 +#define PF_RES_DATA_1_PF_SMAC_IDX_M GENMASK(7, 0) + +#define PF_RES_DATA_1_PF_SMAC_NUM_S 8 +#define PF_RES_DATA_1_PF_SMAC_NUM_M GENMASK(16, 8) + +#define PF_RES_DATA_2_PF_SGID_IDX_S 0 +#define PF_RES_DATA_2_PF_SGID_IDX_M GENMASK(7, 0) + +#define PF_RES_DATA_2_PF_SGID_NUM_S 8 +#define PF_RES_DATA_2_PF_SGID_NUM_M GENMASK(16, 8) + +#define PF_RES_DATA_3_PF_QID_IDX_S 0 +#define PF_RES_DATA_3_PF_QID_IDX_M GENMASK(9, 0) + +#define PF_RES_DATA_3_PF_SL_NUM_S 16 +#define PF_RES_DATA_3_PF_SL_NUM_M GENMASK(26, 16) + struct hns_roce_vf_res_a { __le32 vf_id; __le32 vf_qpc_bt_idx_num; @@ -1276,8 +1337,30 @@ struct hns_roce_v2_cmq { u16 last_status; }; +enum hns_roce_link_table_type { + TSQ_LINK_TABLE, +}; + +struct hns_roce_link_table { + struct hns_roce_buf_list table; + struct hns_roce_buf_list *pg_list; + u32 npages; + u32 pg_sz; +}; + +struct hns_roce_link_table_entry { + u32 blk_ba0; + u32 blk_ba1_nxt_ptr; +}; +#define HNS_ROCE_LINK_TABLE_BA1_S 0 +#define HNS_ROCE_LINK_TABLE_BA1_M GENMASK(19, 0) + +#define HNS_ROCE_LINK_TABLE_NXT_PTR_S 20 +#define HNS_ROCE_LINK_TABLE_NXT_PTR_M GENMASK(31, 20) + struct hns_roce_v2_priv { struct hns_roce_v2_cmq cmq; + struct hns_roce_link_table tsq; }; struct hns_roce_eq_context {