diff options
Diffstat (limited to 'drivers/infiniband/hw')
24 files changed, 599 insertions, 480 deletions
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c index 04258676d..439d0c7c5 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c @@ -1013,8 +1013,7 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) hwq_attr.stride = sizeof(struct sq_sge); hwq_attr.depth = bnxt_qplib_get_depth(sq); hwq_attr.aux_stride = psn_sz; - hwq_attr.aux_depth = psn_sz ? bnxt_qplib_set_sq_size(sq, qp->wqe_mode) - : 0; + hwq_attr.aux_depth = bnxt_qplib_set_sq_size(sq, qp->wqe_mode); /* Update msn tbl size */ if (BNXT_RE_HW_RETX(qp->dev_cap_flags) && psn_sz) { hwq_attr.aux_depth = roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode)); diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h index 50cb2259b..fb8a0c248 100644 --- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h +++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h @@ -930,8 +930,6 @@ void c4iw_id_table_free(struct c4iw_id_table *alloc); typedef int (*c4iw_handler_func)(struct c4iw_dev *dev, struct sk_buff *skb); -int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, - struct l2t_entry *l2t); void c4iw_put_qpid(struct c4iw_rdev *rdev, u32 qpid, struct c4iw_dev_ucontext *uctx); u32 c4iw_get_resource(struct c4iw_id_table *id_table); diff --git a/drivers/infiniband/hw/efa/efa.h b/drivers/infiniband/hw/efa/efa.h index e2bdec32a..926f9ff1f 100644 --- a/drivers/infiniband/hw/efa/efa.h +++ b/drivers/infiniband/hw/efa/efa.h @@ -57,6 +57,7 @@ struct efa_dev { u64 db_bar_addr; u64 db_bar_len; + unsigned int num_irq_vectors; int admin_msix_vector_idx; struct efa_irq admin_irq; diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c index 7b1910a86..5fa3603c8 100644 --- a/drivers/infiniband/hw/efa/efa_main.c +++ b/drivers/infiniband/hw/efa/efa_main.c @@ -322,7 +322,9 @@ static int efa_create_eqs(struct efa_dev *dev) int err; int i; - neqs = min_t(unsigned int, neqs, num_online_cpus()); + neqs = min_t(unsigned int, neqs, + dev->num_irq_vectors - EFA_COMP_EQS_VEC_BASE); + dev->neqs = neqs; dev->eqs = kcalloc(neqs, sizeof(*dev->eqs), GFP_KERNEL); if (!dev->eqs) @@ -468,34 +470,30 @@ static void efa_disable_msix(struct efa_dev *dev) static int efa_enable_msix(struct efa_dev *dev) { - int msix_vecs, irq_num; + int max_vecs, num_vecs; /* * Reserve the max msix vectors we might need, one vector is reserved * for admin. */ - msix_vecs = min_t(int, pci_msix_vec_count(dev->pdev), - num_online_cpus() + 1); + max_vecs = min_t(int, pci_msix_vec_count(dev->pdev), + num_online_cpus() + 1); dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", - msix_vecs); + max_vecs); dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; - irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, - msix_vecs, PCI_IRQ_MSIX); + num_vecs = pci_alloc_irq_vectors(dev->pdev, 1, + max_vecs, PCI_IRQ_MSIX); - if (irq_num < 0) { - dev_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", - irq_num); + if (num_vecs < 0) { + dev_err(&dev->pdev->dev, "Failed to enable MSI-X. error %d\n", + num_vecs); return -ENOSPC; } - if (irq_num != msix_vecs) { - efa_disable_msix(dev); - dev_err(&dev->pdev->dev, - "Allocated %d MSI-X (out of %d requested)\n", - irq_num, msix_vecs); - return -ENOSPC; - } + dev_dbg(&dev->pdev->dev, "Allocated %d MSI-X vectors\n", num_vecs); + + dev->num_irq_vectors = num_vecs; return 0; } diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.c b/drivers/infiniband/hw/hfi1/tid_rdma.c index 18b05ffb4..c465966a1 100644 --- a/drivers/infiniband/hw/hfi1/tid_rdma.c +++ b/drivers/infiniband/hw/hfi1/tid_rdma.c @@ -315,7 +315,7 @@ int hfi1_kern_exp_rcv_init(struct hfi1_ctxtdata *rcd, int reinit) * This routine returns the receive context associated * with a a qp's qpn. * - * Returns the context. + * Return: the context. */ static struct hfi1_ctxtdata *qp_to_rcd(struct rvt_dev_info *rdi, struct rvt_qp *qp) @@ -710,7 +710,7 @@ void hfi1_tid_rdma_flush_wait(struct rvt_qp *qp) * The exp_lock must be held. * * Return: - * On success: a value postive value between 0 and RXE_NUM_TID_FLOWS - 1 + * On success: a value positive value between 0 and RXE_NUM_TID_FLOWS - 1 * On failure: -EAGAIN */ static int kern_reserve_flow(struct hfi1_ctxtdata *rcd, int last) @@ -1007,7 +1007,7 @@ static u32 tid_flush_pages(struct tid_rdma_pageset *list, * pages are tested two at a time, i, i + 1 for contiguous * pages and i - 1 and i contiguous pages. * - * If any condition is false, any accumlated pages are flushed and + * If any condition is false, any accumulated pages are flushed and * v0,v1 are emitted as separate PAGE_SIZE pagesets * * Otherwise, the current 8k is totaled for a future flush. @@ -1434,7 +1434,7 @@ static void kern_program_rcvarray(struct tid_rdma_flow *flow) * (5) computes a tidarray with formatted TID entries which can be sent * to the sender * (6) Reserves and programs HW flows. - * (7) It also manages queing the QP when TID/flow resources are not + * (7) It also manages queueing the QP when TID/flow resources are not * available. * * @req points to struct tid_rdma_request of which the segments are a part. The @@ -1604,7 +1604,7 @@ void hfi1_kern_exp_rcv_clear_all(struct tid_rdma_request *req) } /** - * hfi1_kern_exp_rcv_free_flows - free priviously allocated flow information + * hfi1_kern_exp_rcv_free_flows - free previously allocated flow information * @req: the tid rdma request to be cleaned */ static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req) @@ -2055,7 +2055,7 @@ static int tid_rdma_rcv_error(struct hfi1_packet *packet, * req->clear_tail is advanced). However, when an earlier * request is received, this request will not be complete any * more (qp->s_tail_ack_queue is moved back, see below). - * Consequently, we need to update the TID flow info everytime + * Consequently, we need to update the TID flow info every time * a duplicate request is received. */ bth0 = be32_to_cpu(ohdr->bth[0]); @@ -2219,7 +2219,7 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet) /* * 1. Verify TID RDMA READ REQ as per IB_OPCODE_RC_RDMA_READ * (see hfi1_rc_rcv()) - * 2. Put TID RDMA READ REQ into the response queueu (s_ack_queue) + * 2. Put TID RDMA READ REQ into the response queue (s_ack_queue) * - Setup struct tid_rdma_req with request info * - Initialize struct tid_rdma_flow info; * - Copy TID entries; @@ -2439,7 +2439,7 @@ find_tid_request(struct rvt_qp *qp, u32 psn, enum ib_wr_opcode opcode) void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet) { - /* HANDLER FOR TID RDMA READ RESPONSE packet (Requestor side */ + /* HANDLER FOR TID RDMA READ RESPONSE packet (Requester side) */ /* * 1. Find matching SWQE @@ -3649,7 +3649,7 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet) * 1. Verify TID RDMA WRITE REQ as per IB_OPCODE_RC_RDMA_WRITE_FIRST * (see hfi1_rc_rcv()) * - Don't allow 0-length requests. - * 2. Put TID RDMA WRITE REQ into the response queueu (s_ack_queue) + * 2. Put TID RDMA WRITE REQ into the response queue (s_ack_queue) * - Setup struct tid_rdma_req with request info * - Prepare struct tid_rdma_flow array? * 3. Set the qp->s_ack_state as state diagram in design doc. @@ -4026,7 +4026,7 @@ unlock_r_lock: void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet) { - /* HANDLER FOR TID RDMA WRITE RESPONSE packet (Requestor side */ + /* HANDLER FOR TID RDMA WRITE RESPONSE packet (Requester side) */ /* * 1. Find matching SWQE @@ -5440,8 +5440,9 @@ static bool _hfi1_schedule_tid_send(struct rvt_qp *qp) * the two state machines can step on each other with respect to the * RVT_S_BUSY flag. * Therefore, a modified test is used. - * @return true if the second leg is scheduled; - * false if the second leg is not scheduled. + * + * Return: %true if the second leg is scheduled; + * %false if the second leg is not scheduled. */ bool hfi1_schedule_tid_send(struct rvt_qp *qp) { diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.h b/drivers/infiniband/hw/hns/hns_roce_cmd.h index 052a3d609..11dbbabeb 100644 --- a/drivers/infiniband/hw/hns/hns_roce_cmd.h +++ b/drivers/infiniband/hw/hns/hns_roce_cmd.h @@ -108,6 +108,9 @@ enum { HNS_ROCE_CMD_QUERY_CEQC = 0x92, HNS_ROCE_CMD_DESTROY_CEQC = 0x93, + /* SCC CTX commands */ + HNS_ROCE_CMD_QUERY_SCCC = 0xa2, + /* SCC CTX BT commands */ HNS_ROCE_CMD_READ_SCCC_BT0 = 0xa4, HNS_ROCE_CMD_WRITE_SCCC_BT0 = 0xa5, diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c index 2517c972c..7250d0643 100644 --- a/drivers/infiniband/hw/hns/hns_roce_cq.c +++ b/drivers/infiniband/hw/hns/hns_roce_cq.c @@ -133,14 +133,12 @@ static int alloc_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq) struct hns_roce_cq_table *cq_table = &hr_dev->cq_table; struct ib_device *ibdev = &hr_dev->ib_dev; u64 mtts[MTT_MIN_COUNT] = {}; - dma_addr_t dma_handle; int ret; - ret = hns_roce_mtr_find(hr_dev, &hr_cq->mtr, 0, mtts, ARRAY_SIZE(mtts), - &dma_handle); - if (!ret) { + ret = hns_roce_mtr_find(hr_dev, &hr_cq->mtr, 0, mtts, ARRAY_SIZE(mtts)); + if (ret) { ibdev_err(ibdev, "failed to find CQ mtr, ret = %d.\n", ret); - return -EINVAL; + return ret; } /* Get CQC memory HEM(Hardware Entry Memory) table */ @@ -151,20 +149,21 @@ static int alloc_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq) return ret; } - ret = xa_err(xa_store_irq(&cq_table->array, hr_cq->cqn, hr_cq, GFP_KERNEL)); + ret = xa_err(xa_store(&cq_table->array, hr_cq->cqn, hr_cq, GFP_KERNEL)); if (ret) { ibdev_err(ibdev, "failed to xa_store CQ, ret = %d.\n", ret); goto err_put; } - ret = hns_roce_create_cqc(hr_dev, hr_cq, mtts, dma_handle); + ret = hns_roce_create_cqc(hr_dev, hr_cq, mtts, + hns_roce_get_mtr_ba(&hr_cq->mtr)); if (ret) goto err_xa; return 0; err_xa: - xa_erase_irq(&cq_table->array, hr_cq->cqn); + xa_erase(&cq_table->array, hr_cq->cqn); err_put: hns_roce_table_put(hr_dev, &cq_table->table, hr_cq->cqn); @@ -183,7 +182,7 @@ static void free_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq) dev_err(dev, "DESTROY_CQ failed (%d) for CQN %06lx\n", ret, hr_cq->cqn); - xa_erase_irq(&cq_table->array, hr_cq->cqn); + xa_erase(&cq_table->array, hr_cq->cqn); /* Waiting interrupt process procedure carried out */ synchronize_irq(hr_dev->eq_table.eq[hr_cq->vector].irq); @@ -477,6 +476,13 @@ void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type) struct ib_event event; struct ib_cq *ibcq; + hr_cq = xa_load(&hr_dev->cq_table.array, + cqn & (hr_dev->caps.num_cqs - 1)); + if (!hr_cq) { + dev_warn(dev, "async event for bogus CQ 0x%06x\n", cqn); + return; + } + if (event_type != HNS_ROCE_EVENT_TYPE_CQ_ID_INVALID && event_type != HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR && event_type != HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW) { @@ -485,16 +491,7 @@ void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type) return; } - xa_lock(&hr_dev->cq_table.array); - hr_cq = xa_load(&hr_dev->cq_table.array, - cqn & (hr_dev->caps.num_cqs - 1)); - if (hr_cq) - refcount_inc(&hr_cq->refcount); - xa_unlock(&hr_dev->cq_table.array); - if (!hr_cq) { - dev_warn(dev, "async event for bogus CQ 0x%06x\n", cqn); - return; - } + refcount_inc(&hr_cq->refcount); ibcq = &hr_cq->ib_cq; if (ibcq->event_handler) { diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h index 46f8a6310..c3cbd0a49 100644 --- a/drivers/infiniband/hw/hns/hns_roce_device.h +++ b/drivers/infiniband/hw/hns/hns_roce_device.h @@ -179,6 +179,7 @@ enum { #define HNS_ROCE_CMD_SUCCESS 1 +#define HNS_ROCE_MAX_HOP_NUM 3 /* The minimum page size is 4K for hardware */ #define HNS_HW_PAGE_SHIFT 12 #define HNS_HW_PAGE_SIZE (1 << HNS_HW_PAGE_SHIFT) @@ -269,6 +270,11 @@ struct hns_roce_hem_list { dma_addr_t root_ba; /* pointer to the root ba table */ }; +enum mtr_type { + MTR_DEFAULT = 0, + MTR_PBL, +}; + struct hns_roce_buf_attr { struct { size_t size; /* region size */ @@ -277,7 +283,10 @@ struct hns_roce_buf_attr { unsigned int region_count; /* valid region count */ unsigned int page_shift; /* buffer page shift */ unsigned int user_access; /* umem access flag */ + u64 iova; + enum mtr_type type; bool mtt_only; /* only alloc buffer-required MTT memory */ + bool adaptive; /* adaptive for page_shift and hopnum */ }; struct hns_roce_hem_cfg { @@ -836,7 +845,8 @@ struct hns_roce_caps { u16 default_aeq_period; u16 default_aeq_arm_st; u16 default_ceq_arm_st; - enum hns_roce_cong_type cong_type; + u8 cong_cap; + enum hns_roce_cong_type default_cong_type; }; enum hns_roce_device_state { @@ -937,6 +947,7 @@ struct hns_roce_hw { int (*query_qpc)(struct hns_roce_dev *hr_dev, u32 qpn, void *buffer); int (*query_mpt)(struct hns_roce_dev *hr_dev, u32 key, void *buffer); int (*query_srqc)(struct hns_roce_dev *hr_dev, u32 srqn, void *buffer); + int (*query_sccc)(struct hns_roce_dev *hr_dev, u32 qpn, void *buffer); int (*query_hw_counter)(struct hns_roce_dev *hr_dev, u64 *stats, u32 port, int *hw_counters); const struct ib_device_ops *hns_roce_dev_ops; @@ -1153,8 +1164,13 @@ void hns_roce_cmd_use_polling(struct hns_roce_dev *hr_dev); /* hns roce hw need current block and next block addr from mtt */ #define MTT_MIN_COUNT 2 +static inline dma_addr_t hns_roce_get_mtr_ba(struct hns_roce_mtr *mtr) +{ + return mtr->hem_cfg.root_ba; +} + int hns_roce_mtr_find(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, - u32 offset, u64 *mtt_buf, int mtt_max, u64 *base_addr); + u32 offset, u64 *mtt_buf, int mtt_max); int hns_roce_mtr_create(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, struct hns_roce_buf_attr *buf_attr, unsigned int page_shift, struct ib_udata *udata, diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c index c4ac06a33..a4b3f1916 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.c +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c @@ -249,61 +249,34 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev, } static struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, - int npages, unsigned long hem_alloc_size, gfp_t gfp_mask) { - struct hns_roce_hem_chunk *chunk = NULL; struct hns_roce_hem *hem; - struct scatterlist *mem; int order; void *buf; WARN_ON(gfp_mask & __GFP_HIGHMEM); + order = get_order(hem_alloc_size); + if (PAGE_SIZE << order != hem_alloc_size) { + dev_err(hr_dev->dev, "invalid hem_alloc_size: %lu!\n", + hem_alloc_size); + return NULL; + } + hem = kmalloc(sizeof(*hem), gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); if (!hem) return NULL; - INIT_LIST_HEAD(&hem->chunk_list); - - order = get_order(hem_alloc_size); - - while (npages > 0) { - if (!chunk) { - chunk = kmalloc(sizeof(*chunk), - gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); - if (!chunk) - goto fail; - - sg_init_table(chunk->mem, HNS_ROCE_HEM_CHUNK_LEN); - chunk->npages = 0; - chunk->nsg = 0; - memset(chunk->buf, 0, sizeof(chunk->buf)); - list_add_tail(&chunk->list, &hem->chunk_list); - } + buf = dma_alloc_coherent(hr_dev->dev, hem_alloc_size, + &hem->dma, gfp_mask); + if (!buf) + goto fail; - while (1 << order > npages) - --order; - - /* - * Alloc memory one time. If failed, don't alloc small block - * memory, directly return fail. - */ - mem = &chunk->mem[chunk->npages]; - buf = dma_alloc_coherent(hr_dev->dev, PAGE_SIZE << order, - &sg_dma_address(mem), gfp_mask); - if (!buf) - goto fail; - - chunk->buf[chunk->npages] = buf; - sg_dma_len(mem) = PAGE_SIZE << order; - - ++chunk->npages; - ++chunk->nsg; - npages -= 1 << order; - } + hem->buf = buf; + hem->size = hem_alloc_size; return hem; @@ -314,20 +287,10 @@ fail: void hns_roce_free_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem *hem) { - struct hns_roce_hem_chunk *chunk, *tmp; - int i; - if (!hem) return; - list_for_each_entry_safe(chunk, tmp, &hem->chunk_list, list) { - for (i = 0; i < chunk->npages; ++i) - dma_free_coherent(hr_dev->dev, - sg_dma_len(&chunk->mem[i]), - chunk->buf[i], - sg_dma_address(&chunk->mem[i])); - kfree(chunk); - } + dma_free_coherent(hr_dev->dev, hem->size, hem->buf, hem->dma); kfree(hem); } @@ -415,7 +378,6 @@ static int alloc_mhop_hem(struct hns_roce_dev *hr_dev, { u32 bt_size = mhop->bt_chunk_size; struct device *dev = hr_dev->dev; - struct hns_roce_hem_iter iter; gfp_t flag; u64 bt_ba; u32 size; @@ -456,16 +418,15 @@ static int alloc_mhop_hem(struct hns_roce_dev *hr_dev, */ size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size : bt_size; flag = GFP_KERNEL | __GFP_NOWARN; - table->hem[index->buf] = hns_roce_alloc_hem(hr_dev, size >> PAGE_SHIFT, - size, flag); + table->hem[index->buf] = hns_roce_alloc_hem(hr_dev, size, flag); if (!table->hem[index->buf]) { ret = -ENOMEM; goto err_alloc_hem; } index->inited |= HEM_INDEX_BUF; - hns_roce_hem_first(table->hem[index->buf], &iter); - bt_ba = hns_roce_hem_addr(&iter); + bt_ba = table->hem[index->buf]->dma; + if (table->type < HEM_TYPE_MTT) { if (mhop->hop_num == 2) *(table->bt_l1[index->l1] + mhop->l2_idx) = bt_ba; @@ -586,7 +547,6 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev, } table->hem[i] = hns_roce_alloc_hem(hr_dev, - table->table_chunk_size >> PAGE_SHIFT, table->table_chunk_size, GFP_KERNEL | __GFP_NOWARN); if (!table->hem[i]) { @@ -725,7 +685,6 @@ void *hns_roce_table_find(struct hns_roce_dev *hr_dev, struct hns_roce_hem_table *table, unsigned long obj, dma_addr_t *dma_handle) { - struct hns_roce_hem_chunk *chunk; struct hns_roce_hem_mhop mhop; struct hns_roce_hem *hem; unsigned long mhop_obj = obj; @@ -734,7 +693,6 @@ void *hns_roce_table_find(struct hns_roce_dev *hr_dev, int offset, dma_offset; void *addr = NULL; u32 hem_idx = 0; - int length; int i, j; mutex_lock(&table->mutex); @@ -767,23 +725,8 @@ void *hns_roce_table_find(struct hns_roce_dev *hr_dev, if (!hem) goto out; - list_for_each_entry(chunk, &hem->chunk_list, list) { - for (i = 0; i < chunk->npages; ++i) { - length = sg_dma_len(&chunk->mem[i]); - if (dma_handle && dma_offset >= 0) { - if (length > (u32)dma_offset) - *dma_handle = sg_dma_address( - &chunk->mem[i]) + dma_offset; - dma_offset -= length; - } - - if (length > (u32)offset) { - addr = chunk->buf[i] + offset; - goto out; - } - offset -= length; - } - } + *dma_handle = hem->dma + dma_offset; + addr = hem->buf + offset; out: mutex_unlock(&table->mutex); diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.h b/drivers/infiniband/hw/hns/hns_roce_hem.h index fea6d7d50..6fb51db96 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.h +++ b/drivers/infiniband/hw/hns/hns_roce_hem.h @@ -56,41 +56,25 @@ enum { HEM_TYPE_TRRL, }; -#define HNS_ROCE_HEM_CHUNK_LEN \ - ((256 - sizeof(struct list_head) - 2 * sizeof(int)) / \ - (sizeof(struct scatterlist) + sizeof(void *))) - #define check_whether_bt_num_3(type, hop_num) \ - ((type) < HEM_TYPE_MTT && (hop_num) == 2) + (type < HEM_TYPE_MTT && hop_num == 2) #define check_whether_bt_num_2(type, hop_num) \ - (((type) < HEM_TYPE_MTT && (hop_num) == 1) || \ - ((type) >= HEM_TYPE_MTT && (hop_num) == 2)) + ((type < HEM_TYPE_MTT && hop_num == 1) || \ + (type >= HEM_TYPE_MTT && hop_num == 2)) #define check_whether_bt_num_1(type, hop_num) \ - (((type) < HEM_TYPE_MTT && (hop_num) == HNS_ROCE_HOP_NUM_0) || \ - ((type) >= HEM_TYPE_MTT && (hop_num) == 1) || \ - ((type) >= HEM_TYPE_MTT && (hop_num) == HNS_ROCE_HOP_NUM_0)) - -struct hns_roce_hem_chunk { - struct list_head list; - int npages; - int nsg; - struct scatterlist mem[HNS_ROCE_HEM_CHUNK_LEN]; - void *buf[HNS_ROCE_HEM_CHUNK_LEN]; -}; + ((type < HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0) || \ + (type >= HEM_TYPE_MTT && hop_num == 1) || \ + (type >= HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0)) struct hns_roce_hem { - struct list_head chunk_list; + void *buf; + dma_addr_t dma; + unsigned long size; refcount_t refcount; }; -struct hns_roce_hem_iter { - struct hns_roce_hem *hem; - struct hns_roce_hem_chunk *chunk; - int page_idx; -}; - struct hns_roce_hem_mhop { u32 hop_num; u32 buf_chunk_size; @@ -133,38 +117,4 @@ void *hns_roce_hem_list_find_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_hem_list *hem_list, int offset, int *mtt_cnt); -static inline void hns_roce_hem_first(struct hns_roce_hem *hem, - struct hns_roce_hem_iter *iter) -{ - iter->hem = hem; - iter->chunk = list_empty(&hem->chunk_list) ? NULL : - list_entry(hem->chunk_list.next, - struct hns_roce_hem_chunk, list); - iter->page_idx = 0; -} - -static inline int hns_roce_hem_last(struct hns_roce_hem_iter *iter) -{ - return !iter->chunk; -} - -static inline void hns_roce_hem_next(struct hns_roce_hem_iter *iter) -{ - if (++iter->page_idx >= iter->chunk->nsg) { - if (iter->chunk->list.next == &iter->hem->chunk_list) { - iter->chunk = NULL; - return; - } - - iter->chunk = list_entry(iter->chunk->list.next, - struct hns_roce_hem_chunk, list); - iter->page_idx = 0; - } -} - -static inline dma_addr_t hns_roce_hem_addr(struct hns_roce_hem_iter *iter) -{ - return sg_dma_address(&iter->chunk->mem[iter->page_idx]); -} - #endif /* _HNS_ROCE_HEM_H */ diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index f95ec4618..ba7ae792d 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -2105,7 +2105,7 @@ static void apply_func_caps(struct hns_roce_dev *hr_dev) caps->gmv_bt_num * (HNS_HW_PAGE_SIZE / caps->gmv_entry_sz)); - caps->gmv_entry_num = caps->gmv_bt_num * (HNS_HW_PAGE_SIZE / + caps->gmv_entry_num = caps->gmv_bt_num * (PAGE_SIZE / caps->gmv_entry_sz); } else { u32 func_num = max_t(u32, 1, hr_dev->func_num); @@ -2209,11 +2209,12 @@ static int hns_roce_query_caps(struct hns_roce_dev *hr_dev) caps->max_wqes = 1 << le16_to_cpu(resp_c->sq_depth); caps->num_srqs = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_SRQS); - caps->cong_type = hr_reg_read(resp_d, PF_CAPS_D_CONG_TYPE); + caps->cong_cap = hr_reg_read(resp_d, PF_CAPS_D_CONG_CAP); caps->max_srq_wrs = 1 << le16_to_cpu(resp_d->srq_depth); caps->ceqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_CEQ_DEPTH); caps->num_comp_vectors = hr_reg_read(resp_d, PF_CAPS_D_NUM_CEQS); caps->aeqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_AEQ_DEPTH); + caps->default_cong_type = hr_reg_read(resp_d, PF_CAPS_D_DEFAULT_ALG); caps->reserved_pds = hr_reg_read(resp_d, PF_CAPS_D_RSV_PDS); caps->num_uars = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_UARS); caps->reserved_qps = hr_reg_read(resp_d, PF_CAPS_D_RSV_QPS); @@ -3195,21 +3196,22 @@ static int set_mtpt_pbl(struct hns_roce_dev *hr_dev, u64 pages[HNS_ROCE_V2_MAX_INNER_MTPT_NUM] = { 0 }; struct ib_device *ibdev = &hr_dev->ib_dev; dma_addr_t pbl_ba; - int i, count; + int ret; + int i; - count = hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, pages, - min_t(int, ARRAY_SIZE(pages), mr->npages), - &pbl_ba); - if (count < 1) { - ibdev_err(ibdev, "failed to find PBL mtr, count = %d.\n", - count); - return -ENOBUFS; + ret = hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, pages, + min_t(int, ARRAY_SIZE(pages), mr->npages)); + if (ret) { + ibdev_err(ibdev, "failed to find PBL mtr, ret = %d.\n", ret); + return ret; } /* Aligned to the hardware address access unit */ - for (i = 0; i < count; i++) + for (i = 0; i < ARRAY_SIZE(pages); i++) pages[i] >>= 6; + pbl_ba = hns_roce_get_mtr_ba(&mr->pbl_mtr); + mpt_entry->pbl_size = cpu_to_le32(mr->npages); mpt_entry->pbl_ba_l = cpu_to_le32(pbl_ba >> 3); hr_reg_write(mpt_entry, MPT_PBL_BA_H, upper_32_bits(pbl_ba >> 3)); @@ -3308,18 +3310,12 @@ static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev, static int hns_roce_v2_frmr_write_mtpt(struct hns_roce_dev *hr_dev, void *mb_buf, struct hns_roce_mr *mr) { - struct ib_device *ibdev = &hr_dev->ib_dev; + dma_addr_t pbl_ba = hns_roce_get_mtr_ba(&mr->pbl_mtr); struct hns_roce_v2_mpt_entry *mpt_entry; - dma_addr_t pbl_ba = 0; mpt_entry = mb_buf; memset(mpt_entry, 0, sizeof(*mpt_entry)); - if (hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, NULL, 0, &pbl_ba) < 0) { - ibdev_err(ibdev, "failed to find frmr mtr.\n"); - return -ENOBUFS; - } - hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_FREE); hr_reg_write(mpt_entry, MPT_PD, mr->pd); @@ -3715,9 +3711,8 @@ static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp, wc->status == IB_WC_WR_FLUSH_ERR)) return; - ibdev_err_ratelimited(&hr_dev->ib_dev, "error cqe status 0x%x:\n", - cqe_status); - print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_NONE, 16, 4, cqe, + ibdev_err(&hr_dev->ib_dev, "error cqe status 0x%x:\n", cqe_status); + print_hex_dump(KERN_ERR, "", DUMP_PREFIX_NONE, 16, 4, cqe, cq->cqe_size, false); wc->vendor_err = hr_reg_read(cqe, CQE_SUB_STATUS); @@ -4064,7 +4059,6 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem_table *table, int obj, u32 step_idx) { - struct hns_roce_hem_iter iter; struct hns_roce_hem_mhop mhop; struct hns_roce_hem *hem; unsigned long mhop_obj = obj; @@ -4101,12 +4095,8 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev, if (check_whether_last_step(hop_num, step_idx)) { hem = table->hem[hem_idx]; - for (hns_roce_hem_first(hem, &iter); - !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) { - bt_ba = hns_roce_hem_addr(&iter); - ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type, - step_idx); - } + + ret = set_hem_to_hw(hr_dev, obj, hem->dma, table->type, step_idx); } else { if (step_idx == 0) bt_ba = table->bt_l0_dma_addr[i]; @@ -4347,17 +4337,20 @@ static int config_qp_rq_buf(struct hns_roce_dev *hr_dev, { u64 mtts[MTT_MIN_COUNT] = { 0 }; u64 wqe_sge_ba; - int count; + int ret; /* Search qp buf's mtts */ - count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->rq.offset, mtts, - MTT_MIN_COUNT, &wqe_sge_ba); - if (hr_qp->rq.wqe_cnt && count < 1) { + ret = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->rq.offset, mtts, + MTT_MIN_COUNT); + if (hr_qp->rq.wqe_cnt && ret) { ibdev_err(&hr_dev->ib_dev, - "failed to find RQ WQE, QPN = 0x%lx.\n", hr_qp->qpn); - return -EINVAL; + "failed to find QP(0x%lx) RQ WQE buf, ret = %d.\n", + hr_qp->qpn, ret); + return ret; } + wqe_sge_ba = hns_roce_get_mtr_ba(&hr_qp->mtr); + context->wqe_sge_ba = cpu_to_le32(wqe_sge_ba >> 3); qpc_mask->wqe_sge_ba = 0; @@ -4419,23 +4412,23 @@ static int config_qp_sq_buf(struct hns_roce_dev *hr_dev, struct ib_device *ibdev = &hr_dev->ib_dev; u64 sge_cur_blk = 0; u64 sq_cur_blk = 0; - int count; + int ret; /* search qp buf's mtts */ - count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, 0, &sq_cur_blk, 1, NULL); - if (count < 1) { - ibdev_err(ibdev, "failed to find QP(0x%lx) SQ buf.\n", - hr_qp->qpn); - return -EINVAL; + ret = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->sq.offset, + &sq_cur_blk, 1); + if (ret) { + ibdev_err(ibdev, "failed to find QP(0x%lx) SQ WQE buf, ret = %d.\n", + hr_qp->qpn, ret); + return ret; } if (hr_qp->sge.sge_cnt > 0) { - count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, - hr_qp->sge.offset, - &sge_cur_blk, 1, NULL); - if (count < 1) { - ibdev_err(ibdev, "failed to find QP(0x%lx) SGE buf.\n", - hr_qp->qpn); - return -EINVAL; + ret = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, + hr_qp->sge.offset, &sge_cur_blk, 1); + if (ret) { + ibdev_err(ibdev, "failed to find QP(0x%lx) SGE buf, ret = %d.\n", + hr_qp->qpn, ret); + return ret; } } @@ -4745,14 +4738,8 @@ enum { static int check_cong_type(struct ib_qp *ibqp, struct hns_roce_congestion_algorithm *cong_alg) { - struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device); struct hns_roce_qp *hr_qp = to_hr_qp(ibqp); - if (ibqp->qp_type == IB_QPT_UD || ibqp->qp_type == IB_QPT_GSI) - hr_qp->cong_type = CONG_TYPE_DCQCN; - else - hr_qp->cong_type = hr_dev->caps.cong_type; - /* different congestion types match different configurations */ switch (hr_qp->cong_type) { case CONG_TYPE_DCQCN: @@ -4780,9 +4767,6 @@ static int check_cong_type(struct ib_qp *ibqp, cong_alg->wnd_mode_sel = WND_LIMIT; break; default: - ibdev_warn(&hr_dev->ib_dev, - "invalid type(%u) for congestion selection.\n", - hr_qp->cong_type); hr_qp->cong_type = CONG_TYPE_DCQCN; cong_alg->alg_sel = CONG_DCQCN; cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL; @@ -5333,6 +5317,30 @@ out: return ret; } +static int hns_roce_v2_query_sccc(struct hns_roce_dev *hr_dev, u32 qpn, + void *buffer) +{ + struct hns_roce_v2_scc_context *context; + struct hns_roce_cmd_mailbox *mailbox; + int ret; + + mailbox = hns_roce_alloc_cmd_mailbox(hr_dev); + if (IS_ERR(mailbox)) + return PTR_ERR(mailbox); + + ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_SCCC, + qpn); + if (ret) + goto out; + + context = mailbox->buf; + memcpy(buffer, context, sizeof(*context)); + +out: + hns_roce_free_cmd_mailbox(hr_dev, mailbox); + return ret; +} + static u8 get_qp_timeout_attr(struct hns_roce_dev *hr_dev, struct hns_roce_v2_qp_context *context) { @@ -5586,18 +5594,20 @@ static int hns_roce_v2_write_srqc_index_queue(struct hns_roce_srq *srq, struct ib_device *ibdev = srq->ibsrq.device; struct hns_roce_dev *hr_dev = to_hr_dev(ibdev); u64 mtts_idx[MTT_MIN_COUNT] = {}; - dma_addr_t dma_handle_idx = 0; + dma_addr_t dma_handle_idx; int ret; /* Get physical address of idx que buf */ ret = hns_roce_mtr_find(hr_dev, &idx_que->mtr, 0, mtts_idx, - ARRAY_SIZE(mtts_idx), &dma_handle_idx); - if (ret < 1) { + ARRAY_SIZE(mtts_idx)); + if (ret) { ibdev_err(ibdev, "failed to find mtr for SRQ idx, ret = %d.\n", ret); - return -ENOBUFS; + return ret; } + dma_handle_idx = hns_roce_get_mtr_ba(&idx_que->mtr); + hr_reg_write(ctx, SRQC_IDX_HOP_NUM, to_hr_hem_hopnum(hr_dev->caps.idx_hop_num, srq->wqe_cnt)); @@ -5629,20 +5639,22 @@ static int hns_roce_v2_write_srqc(struct hns_roce_srq *srq, void *mb_buf) struct hns_roce_dev *hr_dev = to_hr_dev(ibdev); struct hns_roce_srq_context *ctx = mb_buf; u64 mtts_wqe[MTT_MIN_COUNT] = {}; - dma_addr_t dma_handle_wqe = 0; + dma_addr_t dma_handle_wqe; int ret; memset(ctx, 0, sizeof(*ctx)); /* Get the physical address of srq buf */ ret = hns_roce_mtr_find(hr_dev, &srq->buf_mtr, 0, mtts_wqe, - ARRAY_SIZE(mtts_wqe), &dma_handle_wqe); - if (ret < 1) { + ARRAY_SIZE(mtts_wqe)); + if (ret) { ibdev_err(ibdev, "failed to find mtr for SRQ WQE, ret = %d.\n", ret); - return -ENOBUFS; + return ret; } + dma_handle_wqe = hns_roce_get_mtr_ba(&srq->buf_mtr); + hr_reg_write(ctx, SRQC_SRQ_ST, 1); hr_reg_write_bool(ctx, SRQC_SRQ_TYPE, srq->ibsrq.srq_type == IB_SRQT_XRC); @@ -6358,7 +6370,7 @@ static int config_eqc(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq, u64 eqe_ba[MTT_MIN_COUNT] = { 0 }; struct hns_roce_eq_context *eqc; u64 bt_ba = 0; - int count; + int ret; eqc = mb_buf; memset(eqc, 0, sizeof(struct hns_roce_eq_context)); @@ -6366,13 +6378,15 @@ static int config_eqc(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq, init_eq_config(hr_dev, eq); /* if not multi-hop, eqe buffer only use one trunk */ - count = hns_roce_mtr_find(hr_dev, &eq->mtr, 0, eqe_ba, MTT_MIN_COUNT, - &bt_ba); - if (count < 1) { - dev_err(hr_dev->dev, "failed to find EQE mtr\n"); - return -ENOBUFS; + ret = hns_roce_mtr_find(hr_dev, &eq->mtr, 0, eqe_ba, + ARRAY_SIZE(eqe_ba)); + if (ret) { + dev_err(hr_dev->dev, "failed to find EQE mtr, ret = %d\n", ret); + return ret; } + bt_ba = hns_roce_get_mtr_ba(&eq->mtr); + hr_reg_write(eqc, EQC_EQ_ST, HNS_ROCE_V2_EQ_STATE_VALID); hr_reg_write(eqc, EQC_EQE_HOP_NUM, eq->hop_num); hr_reg_write(eqc, EQC_OVER_IGNORE, eq->over_ignore); @@ -6719,6 +6733,7 @@ static const struct hns_roce_hw hns_roce_hw_v2 = { .query_qpc = hns_roce_v2_query_qpc, .query_mpt = hns_roce_v2_query_mpt, .query_srqc = hns_roce_v2_query_srqc, + .query_sccc = hns_roce_v2_query_sccc, .query_hw_counter = hns_roce_hw_v2_query_counter, .hns_roce_dev_ops = &hns_roce_v2_dev_ops, .hns_roce_dev_srq_ops = &hns_roce_v2_dev_srq_ops, diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h index cd97cbee6..df04bc8ed 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h @@ -646,6 +646,12 @@ struct hns_roce_v2_qp_context { #define QPCEX_SQ_RQ_NOT_FORBID_EN QPCEX_FIELD_LOC(23, 23) #define QPCEX_STASH QPCEX_FIELD_LOC(82, 82) +#define SCC_CONTEXT_SIZE 16 + +struct hns_roce_v2_scc_context { + __le32 data[SCC_CONTEXT_SIZE]; +}; + #define V2_QP_RWE_S 1 /* rdma write enable */ #define V2_QP_RRE_S 2 /* rdma read enable */ #define V2_QP_ATE_S 3 /* rdma atomic enable */ @@ -1214,12 +1220,13 @@ struct hns_roce_query_pf_caps_d { #define PF_CAPS_D_RQWQE_HOP_NUM PF_CAPS_D_FIELD_LOC(21, 20) #define PF_CAPS_D_EX_SGE_HOP_NUM PF_CAPS_D_FIELD_LOC(23, 22) #define PF_CAPS_D_SQWQE_HOP_NUM PF_CAPS_D_FIELD_LOC(25, 24) -#define PF_CAPS_D_CONG_TYPE PF_CAPS_D_FIELD_LOC(29, 26) +#define PF_CAPS_D_CONG_CAP PF_CAPS_D_FIELD_LOC(29, 26) #define PF_CAPS_D_CEQ_DEPTH PF_CAPS_D_FIELD_LOC(85, 64) #define PF_CAPS_D_NUM_CEQS PF_CAPS_D_FIELD_LOC(95, 86) #define PF_CAPS_D_AEQ_DEPTH PF_CAPS_D_FIELD_LOC(117, 96) #define PF_CAPS_D_AEQ_ARM_ST PF_CAPS_D_FIELD_LOC(119, 118) #define PF_CAPS_D_CEQ_ARM_ST PF_CAPS_D_FIELD_LOC(121, 120) +#define PF_CAPS_D_DEFAULT_ALG PF_CAPS_D_FIELD_LOC(127, 122) #define PF_CAPS_D_RSV_PDS PF_CAPS_D_FIELD_LOC(147, 128) #define PF_CAPS_D_NUM_UARS PF_CAPS_D_FIELD_LOC(155, 148) #define PF_CAPS_D_RSV_QPS PF_CAPS_D_FIELD_LOC(179, 160) diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c index a33d3cedb..1dc60c2b2 100644 --- a/drivers/infiniband/hw/hns/hns_roce_main.c +++ b/drivers/infiniband/hw/hns/hns_roce_main.c @@ -37,7 +37,6 @@ #include <rdma/ib_smi.h> #include <rdma/ib_user_verbs.h> #include <rdma/ib_cache.h> -#include "hnae3.h" #include "hns_roce_common.h" #include "hns_roce_device.h" #include "hns_roce_hem.h" @@ -395,6 +394,9 @@ static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx, resp.config |= HNS_ROCE_RSP_CQE_INLINE_FLAGS; } + if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) + resp.congest_type = hr_dev->caps.cong_cap; + ret = hns_roce_uar_alloc(hr_dev, &context->uar); if (ret) goto error_out; diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c index 0d42fd197..9e05b57a2 100644 --- a/drivers/infiniband/hw/hns/hns_roce_mr.c +++ b/drivers/infiniband/hw/hns/hns_roce_mr.c @@ -32,6 +32,7 @@ */ #include <linux/vmalloc.h> +#include <linux/count_zeros.h> #include <rdma/ib_umem.h> #include <linux/math.h> #include "hns_roce_device.h" @@ -103,14 +104,21 @@ static int alloc_mr_pbl(struct hns_roce_dev *hr_dev, struct hns_roce_mr *mr, buf_attr.user_access = mr->access; /* fast MR's buffer is alloced before mapping, not at creation */ buf_attr.mtt_only = is_fast; + buf_attr.iova = mr->iova; + /* pagesize and hopnum is fixed for fast MR */ + buf_attr.adaptive = !is_fast; + buf_attr.type = MTR_PBL; err = hns_roce_mtr_create(hr_dev, &mr->pbl_mtr, &buf_attr, hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT, udata, start); - if (err) + if (err) { ibdev_err(ibdev, "failed to alloc pbl mtr, ret = %d.\n", err); - else - mr->npages = mr->pbl_mtr.hem_cfg.buf_pg_count; + return err; + } + + mr->npages = mr->pbl_mtr.hem_cfg.buf_pg_count; + mr->pbl_hop_num = buf_attr.region[0].hopnum; return err; } @@ -433,18 +441,18 @@ int hns_roce_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, struct ib_device *ibdev = &hr_dev->ib_dev; struct hns_roce_mr *mr = to_hr_mr(ibmr); struct hns_roce_mtr *mtr = &mr->pbl_mtr; - int ret, sg_num = 0; + int ret = 0; mr->npages = 0; mr->page_list = kvcalloc(mr->pbl_mtr.hem_cfg.buf_pg_count, sizeof(dma_addr_t), GFP_KERNEL); if (!mr->page_list) - return sg_num; + return ret; - sg_num = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, hns_roce_set_page); - if (sg_num < 1) { + ret = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, hns_roce_set_page); + if (ret < 1) { ibdev_err(ibdev, "failed to store sg pages %u %u, cnt = %d.\n", - mr->npages, mr->pbl_mtr.hem_cfg.buf_pg_count, sg_num); + mr->npages, mr->pbl_mtr.hem_cfg.buf_pg_count, ret); goto err_page_list; } @@ -455,16 +463,17 @@ int hns_roce_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, ret = hns_roce_mtr_map(hr_dev, mtr, mr->page_list, mr->npages); if (ret) { ibdev_err(ibdev, "failed to map sg mtr, ret = %d.\n", ret); - sg_num = 0; + ret = 0; } else { mr->pbl_mtr.hem_cfg.buf_pg_shift = (u32)ilog2(ibmr->page_size); + ret = mr->npages; } err_page_list: kvfree(mr->page_list); mr->page_list = NULL; - return sg_num; + return ret; } static void hns_roce_mw_free(struct hns_roce_dev *hr_dev, @@ -694,7 +703,7 @@ static int mtr_alloc_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, mtr->umem = NULL; mtr->kmem = hns_roce_buf_alloc(hr_dev, total_size, buf_attr->page_shift, - mtr->hem_cfg.is_direct ? + !mtr_has_mtt(buf_attr) ? HNS_ROCE_BUF_DIRECT : 0); if (IS_ERR(mtr->kmem)) { ibdev_err(ibdev, "failed to alloc kmem, ret = %ld.\n", @@ -706,14 +715,41 @@ static int mtr_alloc_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, return 0; } -static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, - int page_count, unsigned int page_shift) +static int cal_mtr_pg_cnt(struct hns_roce_mtr *mtr) +{ + struct hns_roce_buf_region *region; + int page_cnt = 0; + int i; + + for (i = 0; i < mtr->hem_cfg.region_count; i++) { + region = &mtr->hem_cfg.region[i]; + page_cnt += region->count; + } + + return page_cnt; +} + +static bool need_split_huge_page(struct hns_roce_mtr *mtr) +{ + /* When HEM buffer uses 0-level addressing, the page size is + * equal to the whole buffer size. If the current MTR has multiple + * regions, we split the buffer into small pages(4k, required by hns + * ROCEE). These pages will be used in multiple regions. + */ + return mtr->hem_cfg.is_direct && mtr->hem_cfg.region_count > 1; +} + +static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr) { struct ib_device *ibdev = &hr_dev->ib_dev; + int page_count = cal_mtr_pg_cnt(mtr); + unsigned int page_shift; dma_addr_t *pages; int npage; int ret; + page_shift = need_split_huge_page(mtr) ? HNS_HW_PAGE_SHIFT : + mtr->hem_cfg.buf_pg_shift; /* alloc a tmp array to store buffer's dma address */ pages = kvcalloc(page_count, sizeof(dma_addr_t), GFP_KERNEL); if (!pages) @@ -733,7 +769,7 @@ static int mtr_map_bufs(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, goto err_alloc_list; } - if (mtr->hem_cfg.is_direct && npage > 1) { + if (need_split_huge_page(mtr) && npage > 1) { ret = mtr_check_direct_pages(pages, npage, page_shift); if (ret) { ibdev_err(ibdev, "failed to check %s page: %d / %d.\n", @@ -808,47 +844,53 @@ int hns_roce_mtr_map(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, return ret; } -int hns_roce_mtr_find(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, - u32 offset, u64 *mtt_buf, int mtt_max, u64 *base_addr) +static int hns_roce_get_direct_addr_mtt(struct hns_roce_hem_cfg *cfg, + u32 start_index, u64 *mtt_buf, + int mtt_cnt) { - struct hns_roce_hem_cfg *cfg = &mtr->hem_cfg; - int mtt_count, left; - u32 start_index; + int mtt_count; int total = 0; - __le64 *mtts; u32 npage; u64 addr; - if (!mtt_buf || mtt_max < 1) - goto done; - - /* no mtt memory in direct mode, so just return the buffer address */ - if (cfg->is_direct) { - start_index = offset >> HNS_HW_PAGE_SHIFT; - for (mtt_count = 0; mtt_count < cfg->region_count && - total < mtt_max; mtt_count++) { - npage = cfg->region[mtt_count].offset; - if (npage < start_index) - continue; + if (mtt_cnt > cfg->region_count) + return -EINVAL; - addr = cfg->root_ba + (npage << HNS_HW_PAGE_SHIFT); - mtt_buf[total] = addr; + for (mtt_count = 0; mtt_count < cfg->region_count && total < mtt_cnt; + mtt_count++) { + npage = cfg->region[mtt_count].offset; + if (npage < start_index) + continue; - total++; - } + addr = cfg->root_ba + (npage << HNS_HW_PAGE_SHIFT); + mtt_buf[total] = addr; - goto done; + total++; } - start_index = offset >> cfg->buf_pg_shift; - left = mtt_max; + if (!total) + return -ENOENT; + + return 0; +} + +static int hns_roce_get_mhop_mtt(struct hns_roce_dev *hr_dev, + struct hns_roce_mtr *mtr, u32 start_index, + u64 *mtt_buf, int mtt_cnt) +{ + int left = mtt_cnt; + int total = 0; + int mtt_count; + __le64 *mtts; + u32 npage; + while (left > 0) { mtt_count = 0; mtts = hns_roce_hem_list_find_mtt(hr_dev, &mtr->hem_list, start_index + total, &mtt_count); if (!mtts || !mtt_count) - goto done; + break; npage = min(mtt_count, left); left -= npage; @@ -856,69 +898,165 @@ int hns_roce_mtr_find(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, mtt_buf[total++] = le64_to_cpu(mtts[mtt_count]); } -done: - if (base_addr) - *base_addr = cfg->root_ba; + if (!total) + return -ENOENT; + + return 0; +} + +int hns_roce_mtr_find(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, + u32 offset, u64 *mtt_buf, int mtt_max) +{ + struct hns_roce_hem_cfg *cfg = &mtr->hem_cfg; + u32 start_index; + int ret; + + if (!mtt_buf || mtt_max < 1) + return -EINVAL; + + /* no mtt memory in direct mode, so just return the buffer address */ + if (cfg->is_direct) { + start_index = offset >> HNS_HW_PAGE_SHIFT; + ret = hns_roce_get_direct_addr_mtt(cfg, start_index, + mtt_buf, mtt_max); + } else { + start_index = offset >> cfg->buf_pg_shift; + ret = hns_roce_get_mhop_mtt(hr_dev, mtr, start_index, + mtt_buf, mtt_max); + } + return ret; +} + +static int get_best_page_shift(struct hns_roce_dev *hr_dev, + struct hns_roce_mtr *mtr, + struct hns_roce_buf_attr *buf_attr) +{ + unsigned int page_sz; + + if (!buf_attr->adaptive || buf_attr->type != MTR_PBL || !mtr->umem) + return 0; + + page_sz = ib_umem_find_best_pgsz(mtr->umem, + hr_dev->caps.page_size_cap, + buf_attr->iova); + if (!page_sz) + return -EINVAL; + + buf_attr->page_shift = order_base_2(page_sz); + return 0; +} + +static int get_best_hop_num(struct hns_roce_dev *hr_dev, + struct hns_roce_mtr *mtr, + struct hns_roce_buf_attr *buf_attr, + unsigned int ba_pg_shift) +{ +#define INVALID_HOPNUM -1 +#define MIN_BA_CNT 1 + size_t buf_pg_sz = 1 << buf_attr->page_shift; + struct ib_device *ibdev = &hr_dev->ib_dev; + size_t ba_pg_sz = 1 << ba_pg_shift; + int hop_num = INVALID_HOPNUM; + size_t unit = MIN_BA_CNT; + size_t ba_cnt; + int j; + + if (!buf_attr->adaptive || buf_attr->type != MTR_PBL) + return 0; + + /* Caculating the number of buf pages, each buf page need a BA */ + if (mtr->umem) + ba_cnt = ib_umem_num_dma_blocks(mtr->umem, buf_pg_sz); + else + ba_cnt = DIV_ROUND_UP(buf_attr->region[0].size, buf_pg_sz); + + for (j = 0; j <= HNS_ROCE_MAX_HOP_NUM; j++) { + if (ba_cnt <= unit) { + hop_num = j; + break; + } + /* Number of BAs can be represented at per hop */ + unit *= ba_pg_sz / BA_BYTE_LEN; + } + + if (hop_num < 0) { + ibdev_err(ibdev, + "failed to calculate a valid hopnum.\n"); + return -EINVAL; + } + + buf_attr->region[0].hopnum = hop_num; + + return 0; +} + +static bool is_buf_attr_valid(struct hns_roce_dev *hr_dev, + struct hns_roce_buf_attr *attr) +{ + struct ib_device *ibdev = &hr_dev->ib_dev; + + if (attr->region_count > ARRAY_SIZE(attr->region) || + attr->region_count < 1 || attr->page_shift < HNS_HW_PAGE_SHIFT) { + ibdev_err(ibdev, + "invalid buf attr, region count %d, page shift %u.\n", + attr->region_count, attr->page_shift); + return false; + } - return total; + return true; } static int mtr_init_buf_cfg(struct hns_roce_dev *hr_dev, - struct hns_roce_buf_attr *attr, - struct hns_roce_hem_cfg *cfg, - unsigned int *buf_page_shift, u64 unalinged_size) + struct hns_roce_mtr *mtr, + struct hns_roce_buf_attr *attr) { + struct hns_roce_hem_cfg *cfg = &mtr->hem_cfg; struct hns_roce_buf_region *r; - u64 first_region_padding; - int page_cnt, region_cnt; - unsigned int page_shift; + size_t buf_pg_sz; size_t buf_size; + int page_cnt, i; + u64 pgoff = 0; + + if (!is_buf_attr_valid(hr_dev, attr)) + return -EINVAL; /* If mtt is disabled, all pages must be within a continuous range */ cfg->is_direct = !mtr_has_mtt(attr); + cfg->region_count = attr->region_count; buf_size = mtr_bufs_size(attr); - if (cfg->is_direct) { - /* When HEM buffer uses 0-level addressing, the page size is - * equal to the whole buffer size, and we split the buffer into - * small pages which is used to check whether the adjacent - * units are in the continuous space and its size is fixed to - * 4K based on hns ROCEE's requirement. - */ - page_shift = HNS_HW_PAGE_SHIFT; - - /* The ROCEE requires the page size to be 4K * 2 ^ N. */ + if (need_split_huge_page(mtr)) { + buf_pg_sz = HNS_HW_PAGE_SIZE; cfg->buf_pg_count = 1; + /* The ROCEE requires the page size to be 4K * 2 ^ N. */ cfg->buf_pg_shift = HNS_HW_PAGE_SHIFT + order_base_2(DIV_ROUND_UP(buf_size, HNS_HW_PAGE_SIZE)); - first_region_padding = 0; } else { - page_shift = attr->page_shift; - cfg->buf_pg_count = DIV_ROUND_UP(buf_size + unalinged_size, - 1 << page_shift); - cfg->buf_pg_shift = page_shift; - first_region_padding = unalinged_size; + buf_pg_sz = 1 << attr->page_shift; + cfg->buf_pg_count = mtr->umem ? + ib_umem_num_dma_blocks(mtr->umem, buf_pg_sz) : + DIV_ROUND_UP(buf_size, buf_pg_sz); + cfg->buf_pg_shift = attr->page_shift; + pgoff = mtr->umem ? mtr->umem->address & ~PAGE_MASK : 0; } /* Convert buffer size to page index and page count for each region and * the buffer's offset needs to be appended to the first region. */ - for (page_cnt = 0, region_cnt = 0; region_cnt < attr->region_count && - region_cnt < ARRAY_SIZE(cfg->region); region_cnt++) { - r = &cfg->region[region_cnt]; + for (page_cnt = 0, i = 0; i < attr->region_count; i++) { + r = &cfg->region[i]; r->offset = page_cnt; - buf_size = hr_hw_page_align(attr->region[region_cnt].size + - first_region_padding); - r->count = DIV_ROUND_UP(buf_size, 1 << page_shift); - first_region_padding = 0; + buf_size = hr_hw_page_align(attr->region[i].size + pgoff); + if (attr->type == MTR_PBL && mtr->umem) + r->count = ib_umem_num_dma_blocks(mtr->umem, buf_pg_sz); + else + r->count = DIV_ROUND_UP(buf_size, buf_pg_sz); + + pgoff = 0; page_cnt += r->count; - r->hopnum = to_hr_hem_hopnum(attr->region[region_cnt].hopnum, - r->count); + r->hopnum = to_hr_hem_hopnum(attr->region[i].hopnum, r->count); } - cfg->region_count = region_cnt; - *buf_page_shift = page_shift; - - return page_cnt; + return 0; } static u64 cal_pages_per_l1ba(unsigned int ba_per_bt, unsigned int hopnum) @@ -1006,50 +1144,58 @@ int hns_roce_mtr_create(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr, unsigned long user_addr) { struct ib_device *ibdev = &hr_dev->ib_dev; - unsigned int buf_page_shift = 0; - int buf_page_cnt; int ret; - buf_page_cnt = mtr_init_buf_cfg(hr_dev, buf_attr, &mtr->hem_cfg, - &buf_page_shift, - udata ? user_addr & ~PAGE_MASK : 0); - if (buf_page_cnt < 1 || buf_page_shift < HNS_HW_PAGE_SHIFT) { - ibdev_err(ibdev, "failed to init mtr cfg, count %d shift %u.\n", - buf_page_cnt, buf_page_shift); - return -EINVAL; - } - - ret = mtr_alloc_mtt(hr_dev, mtr, ba_page_shift); - if (ret) { - ibdev_err(ibdev, "failed to alloc mtr mtt, ret = %d.\n", ret); - return ret; - } - /* The caller has its own buffer list and invokes the hns_roce_mtr_map() * to finish the MTT configuration. */ if (buf_attr->mtt_only) { mtr->umem = NULL; mtr->kmem = NULL; - return 0; + } else { + ret = mtr_alloc_bufs(hr_dev, mtr, buf_attr, udata, user_addr); + if (ret) { + ibdev_err(ibdev, + "failed to alloc mtr bufs, ret = %d.\n", ret); + return ret; + } + + ret = get_best_page_shift(hr_dev, mtr, buf_attr); + if (ret) + goto err_init_buf; + + ret = get_best_hop_num(hr_dev, mtr, buf_attr, ba_page_shift); + if (ret) + goto err_init_buf; } - ret = mtr_alloc_bufs(hr_dev, mtr, buf_attr, udata, user_addr); + ret = mtr_init_buf_cfg(hr_dev, mtr, buf_attr); + if (ret) + goto err_init_buf; + + ret = mtr_alloc_mtt(hr_dev, mtr, ba_page_shift); if (ret) { - ibdev_err(ibdev, "failed to alloc mtr bufs, ret = %d.\n", ret); - goto err_alloc_mtt; + ibdev_err(ibdev, "failed to alloc mtr mtt, ret = %d.\n", ret); + goto err_init_buf; } + if (buf_attr->mtt_only) + return 0; + /* Write buffer's dma address to MTT */ - ret = mtr_map_bufs(hr_dev, mtr, buf_page_cnt, buf_page_shift); - if (ret) + ret = mtr_map_bufs(hr_dev, mtr); + if (ret) { ibdev_err(ibdev, "failed to map mtr bufs, ret = %d.\n", ret); - else - return 0; + goto err_alloc_mtt; + } + + return 0; - mtr_free_bufs(hr_dev, mtr); err_alloc_mtt: mtr_free_mtt(hr_dev, mtr); +err_init_buf: + mtr_free_bufs(hr_dev, mtr); + return ret; } diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c index 31b147210..f35a66325 100644 --- a/drivers/infiniband/hw/hns/hns_roce_qp.c +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c @@ -1004,6 +1004,60 @@ static void free_kernel_wrid(struct hns_roce_qp *hr_qp) kfree(hr_qp->sq.wrid); } +static void default_congest_type(struct hns_roce_dev *hr_dev, + struct hns_roce_qp *hr_qp) +{ + if (hr_qp->ibqp.qp_type == IB_QPT_UD || + hr_qp->ibqp.qp_type == IB_QPT_GSI) + hr_qp->cong_type = CONG_TYPE_DCQCN; + else + hr_qp->cong_type = hr_dev->caps.default_cong_type; +} + +static int set_congest_type(struct hns_roce_qp *hr_qp, + struct hns_roce_ib_create_qp *ucmd) +{ + struct hns_roce_dev *hr_dev = to_hr_dev(hr_qp->ibqp.device); + + switch (ucmd->cong_type_flags) { + case HNS_ROCE_CREATE_QP_FLAGS_DCQCN: + hr_qp->cong_type = CONG_TYPE_DCQCN; + break; + case HNS_ROCE_CREATE_QP_FLAGS_LDCP: + hr_qp->cong_type = CONG_TYPE_LDCP; + break; + case HNS_ROCE_CREATE_QP_FLAGS_HC3: + hr_qp->cong_type = CONG_TYPE_HC3; + break; + case HNS_ROCE_CREATE_QP_FLAGS_DIP: + hr_qp->cong_type = CONG_TYPE_DIP; + break; + default: + return -EINVAL; + } + + if (!test_bit(hr_qp->cong_type, (unsigned long *)&hr_dev->caps.cong_cap)) + return -EOPNOTSUPP; + + if (hr_qp->ibqp.qp_type == IB_QPT_UD && + hr_qp->cong_type != CONG_TYPE_DCQCN) + return -EOPNOTSUPP; + + return 0; +} + +static int set_congest_param(struct hns_roce_dev *hr_dev, + struct hns_roce_qp *hr_qp, + struct hns_roce_ib_create_qp *ucmd) +{ + if (ucmd->comp_mask & HNS_ROCE_CREATE_QP_MASK_CONGEST_TYPE) + return set_congest_type(hr_qp, ucmd); + + default_congest_type(hr_dev, hr_qp); + + return 0; +} + static int set_qp_param(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, struct ib_qp_init_attr *init_attr, struct ib_udata *udata, @@ -1043,6 +1097,10 @@ static int set_qp_param(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, ibdev_err(ibdev, "failed to set user SQ size, ret = %d.\n", ret); + + ret = set_congest_param(hr_dev, hr_qp, ucmd); + if (ret) + return ret; } else { if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) hr_qp->config = HNS_ROCE_EXSGE_FLAGS; @@ -1051,6 +1109,8 @@ static int set_qp_param(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, ibdev_err(ibdev, "failed to set kernel SQ size, ret = %d.\n", ret); + + default_congest_type(hr_dev, hr_qp); } return ret; diff --git a/drivers/infiniband/hw/hns/hns_roce_restrack.c b/drivers/infiniband/hw/hns/hns_roce_restrack.c index f7f3c4cc7..356d98816 100644 --- a/drivers/infiniband/hw/hns/hns_roce_restrack.c +++ b/drivers/infiniband/hw/hns/hns_roce_restrack.c @@ -97,16 +97,33 @@ int hns_roce_fill_res_qp_entry_raw(struct sk_buff *msg, struct ib_qp *ib_qp) { struct hns_roce_dev *hr_dev = to_hr_dev(ib_qp->device); struct hns_roce_qp *hr_qp = to_hr_qp(ib_qp); - struct hns_roce_v2_qp_context context; + struct hns_roce_full_qp_ctx { + struct hns_roce_v2_qp_context qpc; + struct hns_roce_v2_scc_context sccc; + } context = {}; int ret; if (!hr_dev->hw->query_qpc) return -EINVAL; - ret = hr_dev->hw->query_qpc(hr_dev, hr_qp->qpn, &context); + ret = hr_dev->hw->query_qpc(hr_dev, hr_qp->qpn, &context.qpc); if (ret) - return -EINVAL; + return ret; + + /* If SCC is disabled or the query fails, the queried SCCC will + * be all 0. + */ + if (!(hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) || + !hr_dev->hw->query_sccc) + goto out; + + ret = hr_dev->hw->query_sccc(hr_dev, hr_qp->qpn, &context.sccc); + if (ret) + ibdev_warn_ratelimited(&hr_dev->ib_dev, + "failed to query SCCC, ret = %d.\n", + ret); +out: ret = nla_put(msg, RDMA_NLDEV_ATTR_RES_RAW, sizeof(context), &context); return ret; diff --git a/drivers/infiniband/hw/hns/hns_roce_srq.c b/drivers/infiniband/hw/hns/hns_roce_srq.c index 8f48c6723..4abae9477 100644 --- a/drivers/infiniband/hw/hns/hns_roce_srq.c +++ b/drivers/infiniband/hw/hns/hns_roce_srq.c @@ -123,7 +123,7 @@ static int alloc_srqc(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq) return ret; } - ret = xa_err(xa_store_irq(&srq_table->xa, srq->srqn, srq, GFP_KERNEL)); + ret = xa_err(xa_store(&srq_table->xa, srq->srqn, srq, GFP_KERNEL)); if (ret) { ibdev_err(ibdev, "failed to store SRQC, ret = %d.\n", ret); goto err_put; @@ -136,7 +136,7 @@ static int alloc_srqc(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq) return 0; err_xa: - xa_erase_irq(&srq_table->xa, srq->srqn); + xa_erase(&srq_table->xa, srq->srqn); err_put: hns_roce_table_put(hr_dev, &srq_table->table, srq->srqn); @@ -154,7 +154,7 @@ static void free_srqc(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq) dev_err(hr_dev->dev, "DESTROY_SRQ failed (%d) for SRQN %06lx\n", ret, srq->srqn); - xa_erase_irq(&srq_table->xa, srq->srqn); + xa_erase(&srq_table->xa, srq->srqn); if (refcount_dec_and_test(&srq->refcount)) complete(&srq->free); diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c index 89fcc09de..4a71e678d 100644 --- a/drivers/infiniband/hw/mana/cq.c +++ b/drivers/infiniband/hw/mana/cq.c @@ -39,13 +39,37 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, } cq->cqe = attr->cqe; - err = mana_ib_create_queue(mdev, ucmd.buf_addr, cq->cqe * COMP_ENTRY_SIZE, &cq->queue); - if (err) { - ibdev_dbg(ibdev, "Failed to create queue for create cq, %d\n", err); + cq->umem = ib_umem_get(ibdev, ucmd.buf_addr, cq->cqe * COMP_ENTRY_SIZE, + IB_ACCESS_LOCAL_WRITE); + if (IS_ERR(cq->umem)) { + err = PTR_ERR(cq->umem); + ibdev_dbg(ibdev, "Failed to get umem for create cq, err %d\n", + err); return err; } + err = mana_ib_create_zero_offset_dma_region(mdev, cq->umem, &cq->gdma_region); + if (err) { + ibdev_dbg(ibdev, + "Failed to create dma region for create cq, %d\n", + err); + goto err_release_umem; + } + + ibdev_dbg(ibdev, + "create_dma_region ret %d gdma_region 0x%llx\n", + err, cq->gdma_region); + + /* + * The CQ ID is not known at this time. The ID is generated at create_qp + */ + cq->id = INVALID_QUEUE_ID; + return 0; + +err_release_umem: + ib_umem_release(cq->umem); + return err; } int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata) @@ -54,16 +78,24 @@ int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata) struct ib_device *ibdev = ibcq->device; struct mana_ib_dev *mdev; struct gdma_context *gc; + int err; mdev = container_of(ibdev, struct mana_ib_dev, ib_dev); gc = mdev_to_gc(mdev); - if (cq->queue.id != INVALID_QUEUE_ID) { - kfree(gc->cq_table[cq->queue.id]); - gc->cq_table[cq->queue.id] = NULL; + err = mana_ib_gd_destroy_dma_region(mdev, cq->gdma_region); + if (err) { + ibdev_dbg(ibdev, + "Failed to destroy dma region, %d\n", err); + return err; } - mana_ib_destroy_queue(mdev, &cq->queue); + if (cq->id != INVALID_QUEUE_ID) { + kfree(gc->cq_table[cq->id]); + gc->cq_table[cq->id] = NULL; + } + + ib_umem_release(cq->umem); return 0; } @@ -81,10 +113,8 @@ int mana_ib_install_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq) struct gdma_context *gc = mdev_to_gc(mdev); struct gdma_queue *gdma_cq; - if (cq->queue.id >= gc->max_num_cqs) - return -EINVAL; /* Create CQ table entry */ - WARN_ON(gc->cq_table[cq->queue.id]); + WARN_ON(gc->cq_table[cq->id]); gdma_cq = kzalloc(sizeof(*gdma_cq), GFP_KERNEL); if (!gdma_cq) return -ENOMEM; @@ -92,7 +122,7 @@ int mana_ib_install_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq) gdma_cq->cq.context = cq; gdma_cq->type = GDMA_CQ; gdma_cq->cq.callback = mana_ib_cq_handler; - gdma_cq->id = cq->queue.id; - gc->cq_table[cq->queue.id] = gdma_cq; + gdma_cq->id = cq->id; + gc->cq_table[cq->id] = gdma_cq; return 0; } diff --git a/drivers/infiniband/hw/mana/main.c b/drivers/infiniband/hw/mana/main.c index 4524c6b80..71e33feee 100644 --- a/drivers/infiniband/hw/mana/main.c +++ b/drivers/infiniband/hw/mana/main.c @@ -237,49 +237,6 @@ void mana_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) ibdev_dbg(ibdev, "Failed to destroy doorbell page %d\n", ret); } -int mana_ib_create_queue(struct mana_ib_dev *mdev, u64 addr, u32 size, - struct mana_ib_queue *queue) -{ - struct ib_umem *umem; - int err; - - queue->umem = NULL; - queue->id = INVALID_QUEUE_ID; - queue->gdma_region = GDMA_INVALID_DMA_REGION; - - umem = ib_umem_get(&mdev->ib_dev, addr, size, IB_ACCESS_LOCAL_WRITE); - if (IS_ERR(umem)) { - err = PTR_ERR(umem); - ibdev_dbg(&mdev->ib_dev, "Failed to get umem, %d\n", err); - return err; - } - - err = mana_ib_create_zero_offset_dma_region(mdev, umem, &queue->gdma_region); - if (err) { - ibdev_dbg(&mdev->ib_dev, "Failed to create dma region, %d\n", err); - goto free_umem; - } - queue->umem = umem; - - ibdev_dbg(&mdev->ib_dev, - "create_dma_region ret %d gdma_region 0x%llx\n", - err, queue->gdma_region); - - return 0; -free_umem: - ib_umem_release(umem); - return err; -} - -void mana_ib_destroy_queue(struct mana_ib_dev *mdev, struct mana_ib_queue *queue) -{ - /* Ignore return code as there is not much we can do about it. - * The error message is printed inside. - */ - mana_ib_gd_destroy_dma_region(mdev, queue->gdma_region); - ib_umem_release(queue->umem); -} - static int mana_ib_gd_first_dma_region(struct mana_ib_dev *dev, struct gdma_context *gc, diff --git a/drivers/infiniband/hw/mana/mana_ib.h b/drivers/infiniband/hw/mana/mana_ib.h index 6acb5c281..f83390eeb 100644 --- a/drivers/infiniband/hw/mana/mana_ib.h +++ b/drivers/infiniband/hw/mana/mana_ib.h @@ -45,12 +45,6 @@ struct mana_ib_adapter_caps { u32 max_inline_data_size; }; -struct mana_ib_queue { - struct ib_umem *umem; - u64 gdma_region; - u64 id; -}; - struct mana_ib_dev { struct ib_device ib_dev; struct gdma_dev *gdma_dev; @@ -88,8 +82,10 @@ struct mana_ib_mr { struct mana_ib_cq { struct ib_cq ibcq; - struct mana_ib_queue queue; + struct ib_umem *umem; int cqe; + u64 gdma_region; + u64 id; u32 comp_vector; }; @@ -173,10 +169,6 @@ int mana_ib_create_dma_region(struct mana_ib_dev *dev, struct ib_umem *umem, int mana_ib_gd_destroy_dma_region(struct mana_ib_dev *dev, mana_handle_t gdma_region); -int mana_ib_create_queue(struct mana_ib_dev *mdev, u64 addr, u32 size, - struct mana_ib_queue *queue); -void mana_ib_destroy_queue(struct mana_ib_dev *mdev, struct mana_ib_queue *queue); - struct ib_wq *mana_ib_create_wq(struct ib_pd *pd, struct ib_wq_init_attr *init_attr, struct ib_udata *udata); diff --git a/drivers/infiniband/hw/mana/qp.c b/drivers/infiniband/hw/mana/qp.c index d7485ee6a..6e7627745 100644 --- a/drivers/infiniband/hw/mana/qp.c +++ b/drivers/infiniband/hw/mana/qp.c @@ -197,7 +197,7 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd, wq_spec.gdma_region = wq->gdma_region; wq_spec.queue_size = wq->wq_buf_size; - cq_spec.gdma_region = cq->queue.gdma_region; + cq_spec.gdma_region = cq->gdma_region; cq_spec.queue_size = cq->cqe * COMP_ENTRY_SIZE; cq_spec.modr_ctx_id = 0; eq = &mpc->ac->eqs[cq->comp_vector % gc->max_num_queues]; @@ -213,16 +213,16 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd, /* The GDMA regions are now owned by the WQ object */ wq->gdma_region = GDMA_INVALID_DMA_REGION; - cq->queue.gdma_region = GDMA_INVALID_DMA_REGION; + cq->gdma_region = GDMA_INVALID_DMA_REGION; wq->id = wq_spec.queue_index; - cq->queue.id = cq_spec.queue_index; + cq->id = cq_spec.queue_index; ibdev_dbg(&mdev->ib_dev, "ret %d rx_object 0x%llx wq id %llu cq id %llu\n", - ret, wq->rx_object, wq->id, cq->queue.id); + ret, wq->rx_object, wq->id, cq->id); - resp.entries[i].cqid = cq->queue.id; + resp.entries[i].cqid = cq->id; resp.entries[i].wqid = wq->id; mana_ind_table[i] = wq->rx_object; @@ -232,7 +232,7 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd, if (ret) goto fail; - gdma_cq_allocated[i] = gc->cq_table[cq->queue.id]; + gdma_cq_allocated[i] = gc->cq_table[cq->id]; } resp.num_entries = i; @@ -264,7 +264,7 @@ fail: wq = container_of(ibwq, struct mana_ib_wq, ibwq); cq = container_of(ibcq, struct mana_ib_cq, ibcq); - gc->cq_table[cq->queue.id] = NULL; + gc->cq_table[cq->id] = NULL; kfree(gdma_cq_allocated[i]); mana_destroy_wq_obj(mpc, GDMA_RQ, wq->rx_object); @@ -374,7 +374,7 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd, wq_spec.gdma_region = qp->sq_gdma_region; wq_spec.queue_size = ucmd.sq_buf_size; - cq_spec.gdma_region = send_cq->queue.gdma_region; + cq_spec.gdma_region = send_cq->gdma_region; cq_spec.queue_size = send_cq->cqe * COMP_ENTRY_SIZE; cq_spec.modr_ctx_id = 0; eq_vec = send_cq->comp_vector % gc->max_num_queues; @@ -392,10 +392,10 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd, /* The GDMA regions are now owned by the WQ object */ qp->sq_gdma_region = GDMA_INVALID_DMA_REGION; - send_cq->queue.gdma_region = GDMA_INVALID_DMA_REGION; + send_cq->gdma_region = GDMA_INVALID_DMA_REGION; qp->sq_id = wq_spec.queue_index; - send_cq->queue.id = cq_spec.queue_index; + send_cq->id = cq_spec.queue_index; /* Create CQ table entry */ err = mana_ib_install_cq_cb(mdev, send_cq); @@ -404,10 +404,10 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd, ibdev_dbg(&mdev->ib_dev, "ret %d qp->tx_object 0x%llx sq id %llu cq id %llu\n", err, - qp->tx_object, qp->sq_id, send_cq->queue.id); + qp->tx_object, qp->sq_id, send_cq->id); resp.sqid = qp->sq_id; - resp.cqid = send_cq->queue.id; + resp.cqid = send_cq->id; resp.tx_vp_offset = pd->tx_vp_offset; err = ib_copy_to_udata(udata, &resp, sizeof(resp)); @@ -422,7 +422,7 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd, err_release_gdma_cq: kfree(gdma_cq); - gc->cq_table[send_cq->queue.id] = NULL; + gc->cq_table[send_cq->id] = NULL; err_destroy_wq_obj: mana_destroy_wq_obj(mpc, GDMA_SQ, qp->tx_object); diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c index 5a22be14d..96ffbbaf0 100644 --- a/drivers/infiniband/hw/mlx5/mem.c +++ b/drivers/infiniband/hw/mlx5/mem.c @@ -30,7 +30,6 @@ * SOFTWARE. */ -#include <linux/io.h> #include <rdma/ib_umem_odp.h> #include "mlx5_ib.h" #include <linux/jiffies.h> @@ -109,6 +108,7 @@ static int post_send_nop(struct mlx5_ib_dev *dev, struct ib_qp *ibqp, u64 wr_id, __be32 mmio_wqe[16] = {}; unsigned long flags; unsigned int idx; + int i; if (unlikely(dev->mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)) return -EIO; @@ -148,8 +148,10 @@ static int post_send_nop(struct mlx5_ib_dev *dev, struct ib_qp *ibqp, u64 wr_id, * we hit doorbell */ wmb(); - __iowrite64_copy(bf->bfreg->map + bf->offset, mmio_wqe, - sizeof(mmio_wqe) / 8); + for (i = 0; i < 8; i++) + mlx5_write64(&mmio_wqe[i * 2], + bf->bfreg->map + bf->offset + i * 8); + io_stop_wc(); bf->offset ^= bf->buf_size; diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 79ebafecc..a8de35c07 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -643,10 +643,9 @@ struct mlx5_ib_mkey { unsigned int ndescs; struct wait_queue_head wait; refcount_t usecount; - /* Cacheable user Mkey must hold either a rb_key or a cache_ent. */ + /* User Mkey must hold either a rb_key or a cache_ent. */ struct mlx5r_cache_rb_key rb_key; struct mlx5_cache_ent *cache_ent; - u8 cacheable : 1; }; #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE) @@ -1378,7 +1377,6 @@ int mlx5_ib_query_port(struct ib_device *ibdev, u32 port, struct ib_port_attr *props); void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas, u64 access_flags); -void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); int mlx5_ib_get_cqe_size(struct ib_cq *ibcq); int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev); void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev); diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index ecc111ed5..a8ee2ca1f 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1158,7 +1158,6 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, if (IS_ERR(mr)) return mr; mr->mmkey.rb_key = rb_key; - mr->mmkey.cacheable = true; return mr; } @@ -1169,7 +1168,6 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, mr->ibmr.pd = pd; mr->umem = umem; mr->page_shift = order_base_2(page_size); - mr->mmkey.cacheable = true; set_mr_fields(dev, mr, umem->length, access_flags, iova); return mr; @@ -1572,8 +1570,7 @@ static bool can_use_umr_rereg_access(struct mlx5_ib_dev *dev, unsigned int diffs = current_access_flags ^ target_access_flags; if (diffs & ~(IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | - IB_ACCESS_REMOTE_READ | IB_ACCESS_RELAXED_ORDERING | - IB_ACCESS_REMOTE_ATOMIC)) + IB_ACCESS_REMOTE_READ | IB_ACCESS_RELAXED_ORDERING)) return false; return mlx5r_umr_can_reconfig(dev, current_access_flags, target_access_flags); @@ -1838,23 +1835,6 @@ end: return ret; } -static int mlx5_revoke_mr(struct mlx5_ib_mr *mr) -{ - struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); - struct mlx5_cache_ent *ent = mr->mmkey.cache_ent; - - if (mr->mmkey.cacheable && !mlx5r_umr_revoke_mr(mr) && !cache_ent_find_and_store(dev, mr)) - return 0; - - if (ent) { - spin_lock_irq(&ent->mkeys_queue.lock); - ent->in_use--; - mr->mmkey.cache_ent = NULL; - spin_unlock_irq(&ent->mkeys_queue.lock); - } - return destroy_mkey(dev, mr); -} - int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) { struct mlx5_ib_mr *mr = to_mmr(ibmr); @@ -1900,9 +1880,16 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) } /* Stop DMA */ - rc = mlx5_revoke_mr(mr); - if (rc) - return rc; + if (mr->umem && mlx5r_umr_can_load_pas(dev, mr->umem->length)) + if (mlx5r_umr_revoke_mr(mr) || + cache_ent_find_and_store(dev, mr)) + mr->mmkey.cache_ent = NULL; + + if (!mr->mmkey.cache_ent) { + rc = destroy_mkey(to_mdev(mr->ibmr.device), mr); + if (rc) + return rc; + } if (mr->umem) { bool is_odp = is_odp_mr(mr); |