diff options
Diffstat (limited to 'drivers/infiniband/hw/irdma/uk.c')
-rw-r--r-- | drivers/infiniband/hw/irdma/uk.c | 223 |
1 files changed, 103 insertions, 120 deletions
diff --git a/drivers/infiniband/hw/irdma/uk.c b/drivers/infiniband/hw/irdma/uk.c index dd428d915c17..d8285ca16293 100644 --- a/drivers/infiniband/hw/irdma/uk.c +++ b/drivers/infiniband/hw/irdma/uk.c @@ -127,10 +127,7 @@ void irdma_uk_qp_post_wr(struct irdma_qp_uk *qp) hw_sq_tail = (u32)FIELD_GET(IRDMA_QP_DBSA_HW_SQ_TAIL, temp); sw_sq_head = IRDMA_RING_CURRENT_HEAD(qp->sq_ring); if (sw_sq_head != qp->initial_ring.head) { - if (qp->push_dropped) { - writel(qp->qp_id, qp->wqe_alloc_db); - qp->push_dropped = false; - } else if (sw_sq_head != hw_sq_tail) { + if (sw_sq_head != hw_sq_tail) { if (sw_sq_head > qp->initial_ring.head) { if (hw_sq_tail >= qp->initial_ring.head && hw_sq_tail < sw_sq_head) @@ -147,38 +144,6 @@ void irdma_uk_qp_post_wr(struct irdma_qp_uk *qp) } /** - * irdma_qp_ring_push_db - ring qp doorbell - * @qp: hw qp ptr - * @wqe_idx: wqe index - */ -static void irdma_qp_ring_push_db(struct irdma_qp_uk *qp, u32 wqe_idx) -{ - set_32bit_val(qp->push_db, 0, - FIELD_PREP(IRDMA_WQEALLOC_WQE_DESC_INDEX, wqe_idx >> 3) | qp->qp_id); - qp->initial_ring.head = qp->sq_ring.head; - qp->push_mode = true; - qp->push_dropped = false; -} - -void irdma_qp_push_wqe(struct irdma_qp_uk *qp, __le64 *wqe, u16 quanta, - u32 wqe_idx, bool post_sq) -{ - __le64 *push; - - if (IRDMA_RING_CURRENT_HEAD(qp->initial_ring) != - IRDMA_RING_CURRENT_TAIL(qp->sq_ring) && - !qp->push_mode) { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } else { - push = (__le64 *)((uintptr_t)qp->push_wqe + - (wqe_idx & 0x7) * 0x20); - memcpy(push, wqe, quanta * IRDMA_QP_WQE_MIN_SIZE); - irdma_qp_ring_push_db(qp, wqe_idx); - } -} - -/** * irdma_qp_get_next_send_wqe - pad with NOP if needed, return where next WR should go * @qp: hw qp ptr * @wqe_idx: return wqe index @@ -192,7 +157,6 @@ __le64 *irdma_qp_get_next_send_wqe(struct irdma_qp_uk *qp, u32 *wqe_idx, { __le64 *wqe; __le64 *wqe_0 = NULL; - u32 nop_wqe_idx; u16 avail_quanta; u16 i; @@ -209,14 +173,10 @@ __le64 *irdma_qp_get_next_send_wqe(struct irdma_qp_uk *qp, u32 *wqe_idx, IRDMA_SQ_RING_FREE_QUANTA(qp->sq_ring)) return NULL; - nop_wqe_idx = IRDMA_RING_CURRENT_HEAD(qp->sq_ring); for (i = 0; i < avail_quanta; i++) { irdma_nop_1(qp); IRDMA_RING_MOVE_HEAD_NOCHECK(qp->sq_ring); } - if (qp->push_db && info->push_wqe) - irdma_qp_push_wqe(qp, qp->sq_base[nop_wqe_idx].elem, - avail_quanta, nop_wqe_idx, true); } *wqe_idx = IRDMA_RING_CURRENT_HEAD(qp->sq_ring); @@ -282,8 +242,6 @@ int irdma_uk_rdma_write(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, bool read_fence = false; u16 quanta; - info->push_wqe = qp->push_db ? true : false; - op_info = &info->op.rdma_write; if (op_info->num_lo_sges > qp->max_sq_frag_cnt) return -EINVAL; @@ -344,7 +302,6 @@ int irdma_uk_rdma_write(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, FIELD_PREP(IRDMAQPSQ_IMMDATAFLAG, info->imm_data_valid) | FIELD_PREP(IRDMAQPSQ_REPORTRTT, info->report_rtt) | FIELD_PREP(IRDMAQPSQ_ADDFRAGCNT, addl_frag_cnt) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe) | FIELD_PREP(IRDMAQPSQ_READFENCE, read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, info->local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -353,12 +310,9 @@ int irdma_uk_rdma_write(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, dma_wmb(); /* make sure WQE is populated before valid bit is set */ set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, quanta, wqe_idx, post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -383,8 +337,6 @@ int irdma_uk_rdma_read(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, u16 quanta; u64 hdr; - info->push_wqe = qp->push_db ? true : false; - op_info = &info->op.rdma_read; if (qp->max_sq_frag_cnt < op_info->num_lo_sges) return -EINVAL; @@ -431,7 +383,6 @@ int irdma_uk_rdma_read(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, FIELD_PREP(IRDMAQPSQ_ADDFRAGCNT, addl_frag_cnt) | FIELD_PREP(IRDMAQPSQ_OPCODE, (inv_stag ? IRDMAQP_OP_RDMA_READ_LOC_INV : IRDMAQP_OP_RDMA_READ)) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe) | FIELD_PREP(IRDMAQPSQ_READFENCE, info->read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -440,12 +391,9 @@ int irdma_uk_rdma_read(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, dma_wmb(); /* make sure WQE is populated before valid bit is set */ set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, quanta, wqe_idx, post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -468,8 +416,6 @@ int irdma_uk_send(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, bool read_fence = false; u16 quanta; - info->push_wqe = qp->push_db ? true : false; - op_info = &info->op.send; if (qp->max_sq_frag_cnt < op_info->num_sges) return -EINVAL; @@ -530,7 +476,6 @@ int irdma_uk_send(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, FIELD_PREP(IRDMAQPSQ_REPORTRTT, (info->report_rtt ? 1 : 0)) | FIELD_PREP(IRDMAQPSQ_OPCODE, info->op_type) | FIELD_PREP(IRDMAQPSQ_ADDFRAGCNT, addl_frag_cnt) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe) | FIELD_PREP(IRDMAQPSQ_READFENCE, read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, info->local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -541,12 +486,9 @@ int irdma_uk_send(struct irdma_qp_uk *qp, struct irdma_post_sq_info *info, dma_wmb(); /* make sure WQE is populated before valid bit is set */ set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, quanta, wqe_idx, post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -720,7 +662,6 @@ int irdma_uk_inline_rdma_write(struct irdma_qp_uk *qp, u32 i, total_size = 0; u16 quanta; - info->push_wqe = qp->push_db ? true : false; op_info = &info->op.rdma_write; if (unlikely(qp->max_sq_frag_cnt < op_info->num_lo_sges)) @@ -750,7 +691,6 @@ int irdma_uk_inline_rdma_write(struct irdma_qp_uk *qp, FIELD_PREP(IRDMAQPSQ_REPORTRTT, info->report_rtt ? 1 : 0) | FIELD_PREP(IRDMAQPSQ_INLINEDATAFLAG, 1) | FIELD_PREP(IRDMAQPSQ_IMMDATAFLAG, info->imm_data_valid ? 1 : 0) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe ? 1 : 0) | FIELD_PREP(IRDMAQPSQ_READFENCE, read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, info->local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -767,12 +707,8 @@ int irdma_uk_inline_rdma_write(struct irdma_qp_uk *qp, set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, quanta, wqe_idx, post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -794,7 +730,6 @@ int irdma_uk_inline_send(struct irdma_qp_uk *qp, u32 i, total_size = 0; u16 quanta; - info->push_wqe = qp->push_db ? true : false; op_info = &info->op.send; if (unlikely(qp->max_sq_frag_cnt < op_info->num_sges)) @@ -827,7 +762,6 @@ int irdma_uk_inline_send(struct irdma_qp_uk *qp, (info->imm_data_valid ? 1 : 0)) | FIELD_PREP(IRDMAQPSQ_REPORTRTT, (info->report_rtt ? 1 : 0)) | FIELD_PREP(IRDMAQPSQ_INLINEDATAFLAG, 1) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe) | FIELD_PREP(IRDMAQPSQ_READFENCE, read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, info->local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -845,12 +779,8 @@ int irdma_uk_inline_send(struct irdma_qp_uk *qp, set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, quanta, wqe_idx, post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -872,7 +802,6 @@ int irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp, bool local_fence = false; struct ib_sge sge = {}; - info->push_wqe = qp->push_db ? true : false; op_info = &info->op.inv_local_stag; local_fence = info->local_fence; @@ -889,7 +818,6 @@ int irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp, set_64bit_val(wqe, 16, 0); hdr = FIELD_PREP(IRDMAQPSQ_OPCODE, IRDMA_OP_TYPE_INV_STAG) | - FIELD_PREP(IRDMAQPSQ_PUSHWQE, info->push_wqe) | FIELD_PREP(IRDMAQPSQ_READFENCE, info->read_fence) | FIELD_PREP(IRDMAQPSQ_LOCALFENCE, local_fence) | FIELD_PREP(IRDMAQPSQ_SIGCOMPL, info->signaled) | @@ -899,13 +827,8 @@ int irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp, set_64bit_val(wqe, 24, hdr); - if (info->push_wqe) { - irdma_qp_push_wqe(qp, wqe, IRDMA_QP_WQE_MIN_QUANTA, wqe_idx, - post_sq); - } else { - if (post_sq) - irdma_uk_qp_post_wr(qp); - } + if (post_sq) + irdma_uk_qp_post_wr(qp); return 0; } @@ -1124,7 +1047,6 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq, info->q_type = (u8)FIELD_GET(IRDMA_CQ_SQ, qword3); info->error = (bool)FIELD_GET(IRDMA_CQ_ERROR, qword3); - info->push_dropped = (bool)FIELD_GET(IRDMACQ_PSHDROP, qword3); info->ipv4 = (bool)FIELD_GET(IRDMACQ_IPV4, qword3); if (info->error) { info->major_err = FIELD_GET(IRDMA_CQ_MAJERR, qword3); @@ -1161,7 +1083,7 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq, } wqe_idx = (u32)FIELD_GET(IRDMA_CQ_WQEIDX, qword3); info->qp_handle = (irdma_qp_handle)(unsigned long)qp; - info->op_type = (u8)FIELD_GET(IRDMA_CQ_SQ, qword3); + info->op_type = (u8)FIELD_GET(IRDMACQ_OP, qword3); if (info->q_type == IRDMA_CQE_QTYPE_RQ) { u32 array_idx; @@ -1213,11 +1135,6 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq, return irdma_uk_cq_poll_cmpl(cq, info); } } - /*cease posting push mode on push drop*/ - if (info->push_dropped) { - qp->push_mode = false; - qp->push_dropped = true; - } if (info->comp_status != IRDMA_COMPL_STATUS_FLUSHED) { info->wr_id = qp->sq_wrtrk_array[wqe_idx].wrid; if (!info->comp_status) @@ -1349,10 +1266,12 @@ void irdma_get_wqe_shift(struct irdma_uk_attrs *uk_attrs, u32 sge, int irdma_get_sqdepth(struct irdma_uk_attrs *uk_attrs, u32 sq_size, u8 shift, u32 *sqdepth) { + u32 min_size = (u32)uk_attrs->min_hw_wq_size << shift; + *sqdepth = irdma_qp_round_up((sq_size << shift) + IRDMA_SQ_RSVD); - if (*sqdepth < (IRDMA_QP_SW_MIN_WQSIZE << shift)) - *sqdepth = IRDMA_QP_SW_MIN_WQSIZE << shift; + if (*sqdepth < min_size) + *sqdepth = min_size; else if (*sqdepth > uk_attrs->max_hw_wq_quanta) return -EINVAL; @@ -1369,10 +1288,12 @@ int irdma_get_sqdepth(struct irdma_uk_attrs *uk_attrs, u32 sq_size, u8 shift, int irdma_get_rqdepth(struct irdma_uk_attrs *uk_attrs, u32 rq_size, u8 shift, u32 *rqdepth) { + u32 min_size = (u32)uk_attrs->min_hw_wq_size << shift; + *rqdepth = irdma_qp_round_up((rq_size << shift) + IRDMA_RQ_RSVD); - if (*rqdepth < (IRDMA_QP_SW_MIN_WQSIZE << shift)) - *rqdepth = IRDMA_QP_SW_MIN_WQSIZE << shift; + if (*rqdepth < min_size) + *rqdepth = min_size; else if (*rqdepth > uk_attrs->max_hw_rq_quanta) return -EINVAL; @@ -1415,6 +1336,78 @@ static void irdma_setup_connection_wqes(struct irdma_qp_uk *qp, } /** + * irdma_uk_calc_shift_wq - calculate WQE shift for both SQ and RQ + * @ukinfo: qp initialization info + * @sq_shift: Returns shift of SQ + * @rq_shift: Returns shift of RQ + */ +void irdma_uk_calc_shift_wq(struct irdma_qp_uk_init_info *ukinfo, u8 *sq_shift, + u8 *rq_shift) +{ + bool imm_support = ukinfo->uk_attrs->hw_rev >= IRDMA_GEN_2; + + irdma_get_wqe_shift(ukinfo->uk_attrs, + imm_support ? ukinfo->max_sq_frag_cnt + 1 : + ukinfo->max_sq_frag_cnt, + ukinfo->max_inline_data, sq_shift); + + irdma_get_wqe_shift(ukinfo->uk_attrs, ukinfo->max_rq_frag_cnt, 0, + rq_shift); + + if (ukinfo->uk_attrs->hw_rev == IRDMA_GEN_1) { + if (ukinfo->abi_ver > 4) + *rq_shift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1; + } +} + +/** + * irdma_uk_calc_depth_shift_sq - calculate depth and shift for SQ size. + * @ukinfo: qp initialization info + * @sq_depth: Returns depth of SQ + * @sq_shift: Returns shift of SQ + */ +int irdma_uk_calc_depth_shift_sq(struct irdma_qp_uk_init_info *ukinfo, + u32 *sq_depth, u8 *sq_shift) +{ + bool imm_support = ukinfo->uk_attrs->hw_rev >= IRDMA_GEN_2; + int status; + + irdma_get_wqe_shift(ukinfo->uk_attrs, + imm_support ? ukinfo->max_sq_frag_cnt + 1 : + ukinfo->max_sq_frag_cnt, + ukinfo->max_inline_data, sq_shift); + status = irdma_get_sqdepth(ukinfo->uk_attrs, ukinfo->sq_size, + *sq_shift, sq_depth); + + return status; +} + +/** + * irdma_uk_calc_depth_shift_rq - calculate depth and shift for RQ size. + * @ukinfo: qp initialization info + * @rq_depth: Returns depth of RQ + * @rq_shift: Returns shift of RQ + */ +int irdma_uk_calc_depth_shift_rq(struct irdma_qp_uk_init_info *ukinfo, + u32 *rq_depth, u8 *rq_shift) +{ + int status; + + irdma_get_wqe_shift(ukinfo->uk_attrs, ukinfo->max_rq_frag_cnt, 0, + rq_shift); + + if (ukinfo->uk_attrs->hw_rev == IRDMA_GEN_1) { + if (ukinfo->abi_ver > 4) + *rq_shift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1; + } + + status = irdma_get_rqdepth(ukinfo->uk_attrs, ukinfo->rq_size, + *rq_shift, rq_depth); + + return status; +} + +/** * irdma_uk_qp_init - initialize shared qp * @qp: hw qp (user and kernel) * @info: qp initialization info @@ -1428,23 +1421,12 @@ int irdma_uk_qp_init(struct irdma_qp_uk *qp, struct irdma_qp_uk_init_info *info) { int ret_code = 0; u32 sq_ring_size; - u8 sqshift, rqshift; qp->uk_attrs = info->uk_attrs; if (info->max_sq_frag_cnt > qp->uk_attrs->max_hw_wq_frags || info->max_rq_frag_cnt > qp->uk_attrs->max_hw_wq_frags) return -EINVAL; - irdma_get_wqe_shift(qp->uk_attrs, info->max_rq_frag_cnt, 0, &rqshift); - if (qp->uk_attrs->hw_rev == IRDMA_GEN_1) { - irdma_get_wqe_shift(qp->uk_attrs, info->max_sq_frag_cnt, - info->max_inline_data, &sqshift); - if (info->abi_ver > 4) - rqshift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1; - } else { - irdma_get_wqe_shift(qp->uk_attrs, info->max_sq_frag_cnt + 1, - info->max_inline_data, &sqshift); - } qp->qp_caps = info->qp_caps; qp->sq_base = info->sq; qp->rq_base = info->rq; @@ -1456,9 +1438,8 @@ int irdma_uk_qp_init(struct irdma_qp_uk *qp, struct irdma_qp_uk_init_info *info) qp->wqe_alloc_db = info->wqe_alloc_db; qp->qp_id = info->qp_id; qp->sq_size = info->sq_size; - qp->push_mode = false; qp->max_sq_frag_cnt = info->max_sq_frag_cnt; - sq_ring_size = qp->sq_size << sqshift; + sq_ring_size = qp->sq_size << info->sq_shift; IRDMA_RING_INIT(qp->sq_ring, sq_ring_size); IRDMA_RING_INIT(qp->initial_ring, sq_ring_size); if (info->first_sq_wq) { @@ -1473,9 +1454,9 @@ int irdma_uk_qp_init(struct irdma_qp_uk *qp, struct irdma_qp_uk_init_info *info) qp->rq_size = info->rq_size; qp->max_rq_frag_cnt = info->max_rq_frag_cnt; qp->max_inline_data = info->max_inline_data; - qp->rq_wqe_size = rqshift; + qp->rq_wqe_size = info->rq_shift; IRDMA_RING_INIT(qp->rq_ring, qp->rq_size); - qp->rq_wqe_size_multiplier = 1 << rqshift; + qp->rq_wqe_size_multiplier = 1 << info->rq_shift; if (qp->uk_attrs->hw_rev == IRDMA_GEN_1) qp->wqe_ops = iw_wqe_uk_ops_gen_1; else @@ -1527,6 +1508,9 @@ void irdma_uk_clean_cq(void *q, struct irdma_cq_uk *cq) if (polarity != temp) break; + /* Ensure CQE contents are read after valid bit is checked */ + dma_rmb(); + get_64bit_val(cqe, 8, &comp_ctx); if ((void *)(unsigned long)comp_ctx == q) set_64bit_val(cqe, 8, 0); @@ -1551,7 +1535,6 @@ int irdma_nop(struct irdma_qp_uk *qp, u64 wr_id, bool signaled, bool post_sq) u32 wqe_idx; struct irdma_post_sq_info info = {}; - info.push_wqe = false; info.wr_id = wr_id; wqe = irdma_qp_get_next_send_wqe(qp, &wqe_idx, IRDMA_QP_WQE_MIN_QUANTA, 0, &info); |