^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4) * This software is available to you under a choice of one of two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) * licenses. You may choose to be licensed under the terms of the GNU
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) * General Public License (GPL) Version 2, available from the file
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) * COPYING in the main directory of this source tree, or the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) * OpenIB.org BSD license below:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) * Redistribution and use in source and binary forms, with or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11) * without modification, are permitted provided that the following
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) * conditions are met:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) * - Redistributions of source code must retain the above
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) * copyright notice, this list of conditions and the following
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) * disclaimer.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18) * - Redistributions in binary form must reproduce the above
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) * copyright notice, this list of conditions and the following
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20) * disclaimer in the documentation and/or other materials
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) * provided with the distribution.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23) * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26) * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27) * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29) * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) * SOFTWARE.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) #include <linux/kernel.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34) #include <linux/in.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) #include <linux/device.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36) #include <linux/dmapool.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) #include <linux/ratelimit.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39) #include "rds_single_path.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40) #include "rds.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) #include "ib.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) #include "ib_mr.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45) * Convert IB-specific error message to RDS error message and call core
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46) * completion handler.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48) static void rds_ib_send_complete(struct rds_message *rm,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49) int wc_status,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) void (*complete)(struct rds_message *rm, int status))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52) int notify_status;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) switch (wc_status) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) case IB_WC_WR_FLUSH_ERR:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) case IB_WC_SUCCESS:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59) notify_status = RDS_RDMA_SUCCESS;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62) case IB_WC_REM_ACCESS_ERR:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63) notify_status = RDS_RDMA_REMOTE_ERROR;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66) default:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) notify_status = RDS_RDMA_OTHER_ERROR;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70) complete(rm, notify_status);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73) static void rds_ib_send_unmap_data(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) struct rm_data_op *op,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) int wc_status)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77) if (op->op_nents)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) ib_dma_unmap_sg(ic->i_cm_id->device,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) op->op_sg, op->op_nents,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80) DMA_TO_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) static void rds_ib_send_unmap_rdma(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84) struct rm_rdma_op *op,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) int wc_status)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) if (op->op_mapped) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) ib_dma_unmap_sg(ic->i_cm_id->device,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) op->op_sg, op->op_nents,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90) op->op_write ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) op->op_mapped = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) /* If the user asked for a completion notification on this
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) * message, we can implement three different semantics:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) * 1. Notify when we received the ACK on the RDS message
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97) * that was queued with the RDMA. This provides reliable
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) * notification of RDMA status at the expense of a one-way
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99) * packet delay.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) * 2. Notify when the IB stack gives us the completion event for
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) * the RDMA operation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) * 3. Notify when the IB stack gives us the completion event for
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) * the accompanying RDS messages.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104) * Here, we implement approach #3. To implement approach #2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) * we would need to take an event for the rdma WR. To implement #1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) * don't call rds_rdma_send_complete at all, and fall back to the notify
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) * handling in the ACK processing code.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) * Note: There's no need to explicitly sync any RDMA buffers using
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) * ib_dma_sync_sg_for_cpu - the completion for the RDMA
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111) * operation itself unmapped the RDMA buffers, which takes care
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) * of synching.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) rds_ib_send_complete(container_of(op, struct rds_message, rdma),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) wc_status, rds_rdma_send_complete);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117) if (op->op_write)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) rds_stats_add(s_send_rdma_bytes, op->op_bytes);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) rds_stats_add(s_recv_rdma_bytes, op->op_bytes);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) static void rds_ib_send_unmap_atomic(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) struct rm_atomic_op *op,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) int wc_status)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) /* unmap atomic recvbuf */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) if (op->op_mapped) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) ib_dma_unmap_sg(ic->i_cm_id->device, op->op_sg, 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) DMA_FROM_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) op->op_mapped = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) rds_ib_send_complete(container_of(op, struct rds_message, atomic),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) wc_status, rds_atomic_send_complete);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) if (op->op_type == RDS_ATOMIC_TYPE_CSWP)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) rds_ib_stats_inc(s_ib_atomic_cswp);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) rds_ib_stats_inc(s_ib_atomic_fadd);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) * Unmap the resources associated with a struct send_work.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) * Returns the rm for no good reason other than it is unobtainable
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) * other than by switching on wr.opcode, currently, and the caller,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148) * the event handler, needs it.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) static struct rds_message *rds_ib_send_unmap_op(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) struct rds_ib_send_work *send,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) int wc_status)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) struct rds_message *rm = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) /* In the error case, wc.opcode sometimes contains garbage */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) switch (send->s_wr.opcode) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) case IB_WR_SEND:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) if (send->s_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) rm = container_of(send->s_op, struct rds_message, data);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) rds_ib_send_unmap_data(ic, send->s_op, wc_status);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) case IB_WR_RDMA_WRITE:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) case IB_WR_RDMA_READ:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) if (send->s_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) rm = container_of(send->s_op, struct rds_message, rdma);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) rds_ib_send_unmap_rdma(ic, send->s_op, wc_status);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) case IB_WR_ATOMIC_FETCH_AND_ADD:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) case IB_WR_ATOMIC_CMP_AND_SWP:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) if (send->s_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) rm = container_of(send->s_op, struct rds_message, atomic);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) rds_ib_send_unmap_atomic(ic, send->s_op, wc_status);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) default:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) printk_ratelimited(KERN_NOTICE
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) "RDS/IB: %s: unexpected opcode 0x%x in WR!\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) __func__, send->s_wr.opcode);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) break;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) send->s_wr.opcode = 0xdead;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) return rm;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) void rds_ib_send_init_ring(struct rds_ib_connection *ic)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) struct rds_ib_send_work *send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) u32 i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) for (i = 0, send = ic->i_sends; i < ic->i_send_ring.w_nr; i++, send++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) struct ib_sge *sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198) send->s_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) send->s_wr.wr_id = i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201) send->s_wr.sg_list = send->s_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) send->s_wr.ex.imm_data = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204) sge = &send->s_sge[0];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) sge->addr = ic->i_send_hdrs_dma[i];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) sge->length = sizeof(struct rds_header);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) sge->lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) send->s_sge[1].lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) void rds_ib_send_clear_ring(struct rds_ib_connection *ic)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) struct rds_ib_send_work *send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) u32 i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) for (i = 0, send = ic->i_sends; i < ic->i_send_ring.w_nr; i++, send++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220) if (send->s_op && send->s_wr.opcode != 0xdead)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) rds_ib_send_unmap_op(ic, send, IB_WC_WR_FLUSH_ERR);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) * The only fast path caller always has a non-zero nr, so we don't
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227) * bother testing nr before performing the atomic sub.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) static void rds_ib_sub_signaled(struct rds_ib_connection *ic, int nr)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) if ((atomic_sub_return(nr, &ic->i_signaled_sends) == 0) &&
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) waitqueue_active(&rds_ib_ring_empty_wait))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) wake_up(&rds_ib_ring_empty_wait);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) BUG_ON(atomic_read(&ic->i_signaled_sends) < 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) * The _oldest/_free ring operations here race cleanly with the alloc/unalloc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) * operations performed in the send path. As the sender allocs and potentially
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) * unallocs the next free entry in the ring it doesn't alter which is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) * the next to be freed, which is what this is concerned with.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) void rds_ib_send_cqe_handler(struct rds_ib_connection *ic, struct ib_wc *wc)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) struct rds_message *rm = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246) struct rds_connection *conn = ic->conn;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) struct rds_ib_send_work *send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) u32 completed;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) u32 oldest;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) u32 i = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) int nr_sig = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254) rdsdebug("wc wr_id 0x%llx status %u (%s) byte_len %u imm_data %u\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) (unsigned long long)wc->wr_id, wc->status,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) ib_wc_status_msg(wc->status), wc->byte_len,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) be32_to_cpu(wc->ex.imm_data));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) rds_ib_stats_inc(s_ib_tx_cq_event);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) if (wc->wr_id == RDS_IB_ACK_WR_ID) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) if (time_after(jiffies, ic->i_ack_queued + HZ / 2))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) rds_ib_stats_inc(s_ib_tx_stalled);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263) rds_ib_ack_send_complete(ic);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) oldest = rds_ib_ring_oldest(&ic->i_send_ring);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) completed = rds_ib_ring_completed(&ic->i_send_ring, wc->wr_id, oldest);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) for (i = 0; i < completed; i++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) send = &ic->i_sends[oldest];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273) if (send->s_wr.send_flags & IB_SEND_SIGNALED)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) nr_sig++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) rm = rds_ib_send_unmap_op(ic, send, wc->status);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278) if (time_after(jiffies, send->s_queued + HZ / 2))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) rds_ib_stats_inc(s_ib_tx_stalled);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) if (send->s_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282) if (send->s_op == rm->m_final_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) /* If anyone waited for this message to get
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) * flushed out, wake them up now
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) rds_message_unmapped(rm);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288) rds_message_put(rm);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) send->s_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) oldest = (oldest + 1) % ic->i_send_ring.w_nr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) rds_ib_ring_free(&ic->i_send_ring, completed);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) rds_ib_sub_signaled(ic, nr_sig);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) nr_sig = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) if (test_and_clear_bit(RDS_LL_SEND_FULL, &conn->c_flags) ||
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) test_bit(0, &conn->c_map_queued))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) queue_delayed_work(rds_wq, &conn->c_send_w, 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) /* We expect errors as the qp is drained during shutdown */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) if (wc->status != IB_WC_SUCCESS && rds_conn_up(conn)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305) rds_ib_conn_error(conn, "send completion on <%pI6c,%pI6c,%d> had status %u (%s), vendor err 0x%x, disconnecting and reconnecting\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) &conn->c_laddr, &conn->c_faddr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307) conn->c_tos, wc->status,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) ib_wc_status_msg(wc->status), wc->vendor_err);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) * This is the main function for allocating credits when sending
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) * messages.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316) * Conceptually, we have two counters:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) * - send credits: this tells us how many WRs we're allowed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) * to submit without overruning the receiver's queue. For
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319) * each SEND WR we post, we decrement this by one.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) * - posted credits: this tells us how many WRs we recently
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322) * posted to the receive queue. This value is transferred
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) * to the peer as a "credit update" in a RDS header field.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324) * Every time we transmit credits to the peer, we subtract
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) * the amount of transferred credits from this counter.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) * It is essential that we avoid situations where both sides have
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) * exhausted their send credits, and are unable to send new credits
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) * to the peer. We achieve this by requiring that we send at least
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330) * one credit update to the peer before exhausting our credits.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331) * When new credits arrive, we subtract one credit that is withheld
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) * until we've posted new buffers and are ready to transmit these
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) * credits (see rds_ib_send_add_credits below).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335) * The RDS send code is essentially single-threaded; rds_send_xmit
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336) * sets RDS_IN_XMIT to ensure exclusive access to the send ring.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) * However, the ACK sending code is independent and can race with
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338) * message SENDs.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340) * In the send path, we need to update the counters for send credits
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) * and the counter of posted buffers atomically - when we use the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) * last available credit, we cannot allow another thread to race us
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) * and grab the posted credits counter. Hence, we have to use a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) * spinlock to protect the credit counter, or use atomics.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346) * Spinlocks shared between the send and the receive path are bad,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347) * because they create unnecessary delays. An early implementation
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348) * using a spinlock showed a 5% degradation in throughput at some
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) * loads.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351) * This implementation avoids spinlocks completely, putting both
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352) * counters into a single atomic, and updating that atomic using
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) * atomic_add (in the receive path, when receiving fresh credits),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354) * and using atomic_cmpxchg when updating the two counters.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) int rds_ib_send_grab_credits(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) u32 wanted, u32 *adv_credits, int need_posted, int max_posted)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) unsigned int avail, posted, got = 0, advertise;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) long oldval, newval;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) *adv_credits = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) if (!ic->i_flowctl)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364) return wanted;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) try_again:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) advertise = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) oldval = newval = atomic_read(&ic->i_credits);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369) posted = IB_GET_POST_CREDITS(oldval);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) avail = IB_GET_SEND_CREDITS(oldval);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372) rdsdebug("wanted=%u credits=%u posted=%u\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) wanted, avail, posted);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375) /* The last credit must be used to send a credit update. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376) if (avail && !posted)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377) avail--;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) if (avail < wanted) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380) struct rds_connection *conn = ic->i_cm_id->context;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) /* Oops, there aren't that many credits left! */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) got = avail;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386) /* Sometimes you get what you want, lalala. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) got = wanted;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389) newval -= IB_SET_SEND_CREDITS(got);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392) * If need_posted is non-zero, then the caller wants
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393) * the posted regardless of whether any send credits are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) * available.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) if (posted && (got || need_posted)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) advertise = min_t(unsigned int, posted, max_posted);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398) newval -= IB_SET_POST_CREDITS(advertise);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401) /* Finally bill everything */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402) if (atomic_cmpxchg(&ic->i_credits, oldval, newval) != oldval)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) goto try_again;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405) *adv_credits = advertise;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406) return got;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409) void rds_ib_send_add_credits(struct rds_connection *conn, unsigned int credits)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413) if (credits == 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416) rdsdebug("credits=%u current=%u%s\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) credits,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418) IB_GET_SEND_CREDITS(atomic_read(&ic->i_credits)),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419) test_bit(RDS_LL_SEND_FULL, &conn->c_flags) ? ", ll_send_full" : "");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) atomic_add(IB_SET_SEND_CREDITS(credits), &ic->i_credits);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422) if (test_and_clear_bit(RDS_LL_SEND_FULL, &conn->c_flags))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) queue_delayed_work(rds_wq, &conn->c_send_w, 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425) WARN_ON(IB_GET_SEND_CREDITS(credits) >= 16384);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) rds_ib_stats_inc(s_ib_rx_credit_updates);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) void rds_ib_advertise_credits(struct rds_connection *conn, unsigned int posted)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434) if (posted == 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437) atomic_add(IB_SET_POST_CREDITS(posted), &ic->i_credits);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) /* Decide whether to send an update to the peer now.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440) * If we would send a credit update for every single buffer we
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441) * post, we would end up with an ACK storm (ACK arrives,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442) * consumes buffer, we refill the ring, send ACK to remote
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443) * advertising the newly posted buffer... ad inf)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445) * Performance pretty much depends on how often we send
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) * credit updates - too frequent updates mean lots of ACKs.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) * Too infrequent updates, and the peer will run out of
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) * credits and has to throttle.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) * For the time being, 16 seems to be a good compromise.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451) if (IB_GET_POST_CREDITS(atomic_read(&ic->i_credits)) >= 16)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) static inline int rds_ib_set_wr_signal_state(struct rds_ib_connection *ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456) struct rds_ib_send_work *send,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457) bool notify)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460) * We want to delay signaling completions just enough to get
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) * the batching benefits but not so much that we create dead time
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462) * on the wire.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464) if (ic->i_unsignaled_wrs-- == 0 || notify) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465) ic->i_unsignaled_wrs = rds_ib_sysctl_max_unsig_wrs;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) send->s_wr.send_flags |= IB_SEND_SIGNALED;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) return 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 468) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 469) return 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 470) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 471)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 472) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 473) * This can be called multiple times for a given message. The first time
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 474) * we see a message we map its scatterlist into the IB device so that
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 475) * we can provide that mapped address to the IB scatter gather entries
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 476) * in the IB work requests. We translate the scatterlist into a series
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 477) * of work requests that fragment the message. These work requests complete
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 478) * in order so we pass ownership of the message to the completion handler
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 479) * once we send the final fragment.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 480) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 481) * The RDS core uses the c_send_lock to only enter this function once
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 482) * per connection. This makes sure that the tx ring alloc/unalloc pairs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 483) * don't get out of sync and confuse the ring.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 484) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 485) int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 486) unsigned int hdr_off, unsigned int sg, unsigned int off)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 487) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 488) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 489) struct ib_device *dev = ic->i_cm_id->device;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 490) struct rds_ib_send_work *send = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 491) struct rds_ib_send_work *first;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 492) struct rds_ib_send_work *prev;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 493) const struct ib_send_wr *failed_wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 494) struct scatterlist *scat;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 495) u32 pos;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 496) u32 i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 497) u32 work_alloc;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 498) u32 credit_alloc = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 499) u32 posted;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 500) u32 adv_credits = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 501) int send_flags = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 502) int bytes_sent = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 503) int ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 504) int flow_controlled = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 505) int nr_sig = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 506)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 507) BUG_ON(off % RDS_FRAG_SIZE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 508) BUG_ON(hdr_off != 0 && hdr_off != sizeof(struct rds_header));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 509)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 510) /* Do not send cong updates to IB loopback */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 511) if (conn->c_loopback
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 512) && rm->m_inc.i_hdr.h_flags & RDS_FLAG_CONG_BITMAP) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 513) rds_cong_map_updated(conn->c_fcong, ~(u64) 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 514) scat = &rm->data.op_sg[sg];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 515) ret = max_t(int, RDS_CONG_MAP_BYTES, scat->length);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 516) return sizeof(struct rds_header) + ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 517) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 518)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 519) /* FIXME we may overallocate here */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 520) if (be32_to_cpu(rm->m_inc.i_hdr.h_len) == 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 521) i = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 522) else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 523) i = DIV_ROUND_UP(be32_to_cpu(rm->m_inc.i_hdr.h_len), RDS_FRAG_SIZE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 524)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 525) work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, i, &pos);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 526) if (work_alloc == 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 527) set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 528) rds_ib_stats_inc(s_ib_tx_ring_full);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 529) ret = -ENOMEM;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 530) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 531) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 532)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 533) if (ic->i_flowctl) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 534) credit_alloc = rds_ib_send_grab_credits(ic, work_alloc, &posted, 0, RDS_MAX_ADV_CREDIT);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 535) adv_credits += posted;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 536) if (credit_alloc < work_alloc) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 537) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - credit_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 538) work_alloc = credit_alloc;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 539) flow_controlled = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 540) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 541) if (work_alloc == 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 542) set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 543) rds_ib_stats_inc(s_ib_tx_throttle);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 544) ret = -ENOMEM;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 545) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 546) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 547) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 548)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 549) /* map the message the first time we see it */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 550) if (!ic->i_data_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 551) if (rm->data.op_nents) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 552) rm->data.op_count = ib_dma_map_sg(dev,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 553) rm->data.op_sg,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 554) rm->data.op_nents,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 555) DMA_TO_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 556) rdsdebug("ic %p mapping rm %p: %d\n", ic, rm, rm->data.op_count);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 557) if (rm->data.op_count == 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 558) rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 559) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 560) ret = -ENOMEM; /* XXX ? */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 561) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 562) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 563) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 564) rm->data.op_count = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 565) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 566)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 567) rds_message_addref(rm);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 568) rm->data.op_dmasg = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 569) rm->data.op_dmaoff = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 570) ic->i_data_op = &rm->data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 571)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 572) /* Finalize the header */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 573) if (test_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 574) rm->m_inc.i_hdr.h_flags |= RDS_FLAG_ACK_REQUIRED;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 575) if (test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 576) rm->m_inc.i_hdr.h_flags |= RDS_FLAG_RETRANSMITTED;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 577)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 578) /* If it has a RDMA op, tell the peer we did it. This is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 579) * used by the peer to release use-once RDMA MRs. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 580) if (rm->rdma.op_active) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 581) struct rds_ext_header_rdma ext_hdr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 582)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 583) ext_hdr.h_rdma_rkey = cpu_to_be32(rm->rdma.op_rkey);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 584) rds_message_add_extension(&rm->m_inc.i_hdr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 585) RDS_EXTHDR_RDMA, &ext_hdr, sizeof(ext_hdr));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 586) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 587) if (rm->m_rdma_cookie) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 588) rds_message_add_rdma_dest_extension(&rm->m_inc.i_hdr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 589) rds_rdma_cookie_key(rm->m_rdma_cookie),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 590) rds_rdma_cookie_offset(rm->m_rdma_cookie));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 591) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 592)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 593) /* Note - rds_ib_piggyb_ack clears the ACK_REQUIRED bit, so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 594) * we should not do this unless we have a chance of at least
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 595) * sticking the header into the send ring. Which is why we
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 596) * should call rds_ib_ring_alloc first. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 597) rm->m_inc.i_hdr.h_ack = cpu_to_be64(rds_ib_piggyb_ack(ic));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 598) rds_message_make_checksum(&rm->m_inc.i_hdr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 599)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 600) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 601) * Update adv_credits since we reset the ACK_REQUIRED bit.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 602) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 603) if (ic->i_flowctl) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 604) rds_ib_send_grab_credits(ic, 0, &posted, 1, RDS_MAX_ADV_CREDIT - adv_credits);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 605) adv_credits += posted;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 606) BUG_ON(adv_credits > 255);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 607) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 608) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 609)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 610) /* Sometimes you want to put a fence between an RDMA
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 611) * READ and the following SEND.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 612) * We could either do this all the time
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 613) * or when requested by the user. Right now, we let
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 614) * the application choose.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 615) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 616) if (rm->rdma.op_active && rm->rdma.op_fence)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 617) send_flags = IB_SEND_FENCE;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 618)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 619) /* Each frag gets a header. Msgs may be 0 bytes */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 620) send = &ic->i_sends[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 621) first = send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 622) prev = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 623) scat = &ic->i_data_op->op_sg[rm->data.op_dmasg];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 624) i = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 625) do {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 626) unsigned int len = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 627)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 628) /* Set up the header */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 629) send->s_wr.send_flags = send_flags;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 630) send->s_wr.opcode = IB_WR_SEND;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 631) send->s_wr.num_sge = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 632) send->s_wr.next = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 633) send->s_queued = jiffies;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 634) send->s_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 635)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 636) send->s_sge[0].addr = ic->i_send_hdrs_dma[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 637)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 638) send->s_sge[0].length = sizeof(struct rds_header);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 639) send->s_sge[0].lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 640)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 641) ib_dma_sync_single_for_cpu(ic->rds_ibdev->dev,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 642) ic->i_send_hdrs_dma[pos],
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 643) sizeof(struct rds_header),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 644) DMA_TO_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 645) memcpy(ic->i_send_hdrs[pos], &rm->m_inc.i_hdr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 646) sizeof(struct rds_header));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 647)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 648)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 649) /* Set up the data, if present */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 650) if (i < work_alloc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 651) && scat != &rm->data.op_sg[rm->data.op_count]) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 652) len = min(RDS_FRAG_SIZE,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 653) sg_dma_len(scat) - rm->data.op_dmaoff);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 654) send->s_wr.num_sge = 2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 655)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 656) send->s_sge[1].addr = sg_dma_address(scat);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 657) send->s_sge[1].addr += rm->data.op_dmaoff;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 658) send->s_sge[1].length = len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 659) send->s_sge[1].lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 660)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 661) bytes_sent += len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 662) rm->data.op_dmaoff += len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 663) if (rm->data.op_dmaoff == sg_dma_len(scat)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 664) scat++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 665) rm->data.op_dmasg++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 666) rm->data.op_dmaoff = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 667) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 668) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 669)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 670) rds_ib_set_wr_signal_state(ic, send, false);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 671)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 672) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 673) * Always signal the last one if we're stopping due to flow control.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 674) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 675) if (ic->i_flowctl && flow_controlled && i == (work_alloc - 1)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 676) rds_ib_set_wr_signal_state(ic, send, true);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 677) send->s_wr.send_flags |= IB_SEND_SOLICITED;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 678) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 679)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 680) if (send->s_wr.send_flags & IB_SEND_SIGNALED)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 681) nr_sig++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 682)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 683) rdsdebug("send %p wr %p num_sge %u next %p\n", send,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 684) &send->s_wr, send->s_wr.num_sge, send->s_wr.next);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 685)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 686) if (ic->i_flowctl && adv_credits) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 687) struct rds_header *hdr = ic->i_send_hdrs[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 688)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 689) /* add credit and redo the header checksum */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 690) hdr->h_credit = adv_credits;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 691) rds_message_make_checksum(hdr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 692) adv_credits = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 693) rds_ib_stats_inc(s_ib_tx_credit_updates);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 694) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 695) ib_dma_sync_single_for_device(ic->rds_ibdev->dev,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 696) ic->i_send_hdrs_dma[pos],
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 697) sizeof(struct rds_header),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 698) DMA_TO_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 699)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 700) if (prev)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 701) prev->s_wr.next = &send->s_wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 702) prev = send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 703)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 704) pos = (pos + 1) % ic->i_send_ring.w_nr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 705) send = &ic->i_sends[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 706) i++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 707)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 708) } while (i < work_alloc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 709) && scat != &rm->data.op_sg[rm->data.op_count]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 710)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 711) /* Account the RDS header in the number of bytes we sent, but just once.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 712) * The caller has no concept of fragmentation. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 713) if (hdr_off == 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 714) bytes_sent += sizeof(struct rds_header);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 715)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 716) /* if we finished the message then send completion owns it */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 717) if (scat == &rm->data.op_sg[rm->data.op_count]) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 718) prev->s_op = ic->i_data_op;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 719) prev->s_wr.send_flags |= IB_SEND_SOLICITED;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 720) if (!(prev->s_wr.send_flags & IB_SEND_SIGNALED))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 721) nr_sig += rds_ib_set_wr_signal_state(ic, prev, true);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 722) ic->i_data_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 723) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 724)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 725) /* Put back wrs & credits we didn't use */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 726) if (i < work_alloc) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 727) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - i);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 728) work_alloc = i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 729) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 730) if (ic->i_flowctl && i < credit_alloc)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 731) rds_ib_send_add_credits(conn, credit_alloc - i);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 732)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 733) if (nr_sig)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 734) atomic_add(nr_sig, &ic->i_signaled_sends);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 735)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 736) /* XXX need to worry about failed_wr and partial sends. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 737) failed_wr = &first->s_wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 738) ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 739) rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 740) first, &first->s_wr, ret, failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 741) BUG_ON(failed_wr != &first->s_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 742) if (ret) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 743) printk(KERN_WARNING "RDS/IB: ib_post_send to %pI6c "
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 744) "returned %d\n", &conn->c_faddr, ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 745) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 746) rds_ib_sub_signaled(ic, nr_sig);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 747) if (prev->s_op) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 748) ic->i_data_op = prev->s_op;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 749) prev->s_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 750) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 751)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 752) rds_ib_conn_error(ic->conn, "ib_post_send failed\n");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 753) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 754) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 755)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 756) ret = bytes_sent;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 757) out:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 758) BUG_ON(adv_credits);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 759) return ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 760) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 761)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 762) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 763) * Issue atomic operation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 764) * A simplified version of the rdma case, we always map 1 SG, and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 765) * only 8 bytes, for the return value from the atomic operation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 766) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 767) int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 768) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 769) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 770) struct rds_ib_send_work *send = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 771) const struct ib_send_wr *failed_wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 772) u32 pos;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 773) u32 work_alloc;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 774) int ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 775) int nr_sig = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 776)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 777) work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, 1, &pos);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 778) if (work_alloc != 1) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 779) rds_ib_stats_inc(s_ib_tx_ring_full);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 780) ret = -ENOMEM;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 781) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 782) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 783)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 784) /* address of send request in ring */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 785) send = &ic->i_sends[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 786) send->s_queued = jiffies;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 787)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 788) if (op->op_type == RDS_ATOMIC_TYPE_CSWP) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 789) send->s_atomic_wr.wr.opcode = IB_WR_MASKED_ATOMIC_CMP_AND_SWP;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 790) send->s_atomic_wr.compare_add = op->op_m_cswp.compare;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 791) send->s_atomic_wr.swap = op->op_m_cswp.swap;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 792) send->s_atomic_wr.compare_add_mask = op->op_m_cswp.compare_mask;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 793) send->s_atomic_wr.swap_mask = op->op_m_cswp.swap_mask;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 794) } else { /* FADD */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 795) send->s_atomic_wr.wr.opcode = IB_WR_MASKED_ATOMIC_FETCH_AND_ADD;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 796) send->s_atomic_wr.compare_add = op->op_m_fadd.add;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 797) send->s_atomic_wr.swap = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 798) send->s_atomic_wr.compare_add_mask = op->op_m_fadd.nocarry_mask;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 799) send->s_atomic_wr.swap_mask = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 800) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 801) send->s_wr.send_flags = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 802) nr_sig = rds_ib_set_wr_signal_state(ic, send, op->op_notify);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 803) send->s_atomic_wr.wr.num_sge = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 804) send->s_atomic_wr.wr.next = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 805) send->s_atomic_wr.remote_addr = op->op_remote_addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 806) send->s_atomic_wr.rkey = op->op_rkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 807) send->s_op = op;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 808) rds_message_addref(container_of(send->s_op, struct rds_message, atomic));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 809)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 810) /* map 8 byte retval buffer to the device */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 811) ret = ib_dma_map_sg(ic->i_cm_id->device, op->op_sg, 1, DMA_FROM_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 812) rdsdebug("ic %p mapping atomic op %p. mapped %d pg\n", ic, op, ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 813) if (ret != 1) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 814) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 815) rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 816) ret = -ENOMEM; /* XXX ? */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 817) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 818) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 819)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 820) /* Convert our struct scatterlist to struct ib_sge */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 821) send->s_sge[0].addr = sg_dma_address(op->op_sg);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 822) send->s_sge[0].length = sg_dma_len(op->op_sg);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 823) send->s_sge[0].lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 824)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 825) rdsdebug("rva %Lx rpa %Lx len %u\n", op->op_remote_addr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 826) send->s_sge[0].addr, send->s_sge[0].length);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 827)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 828) if (nr_sig)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 829) atomic_add(nr_sig, &ic->i_signaled_sends);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 830)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 831) failed_wr = &send->s_atomic_wr.wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 832) ret = ib_post_send(ic->i_cm_id->qp, &send->s_atomic_wr.wr, &failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 833) rdsdebug("ic %p send %p (wr %p) ret %d wr %p\n", ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 834) send, &send->s_atomic_wr, ret, failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 835) BUG_ON(failed_wr != &send->s_atomic_wr.wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 836) if (ret) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 837) printk(KERN_WARNING "RDS/IB: atomic ib_post_send to %pI6c "
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 838) "returned %d\n", &conn->c_faddr, ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 839) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 840) rds_ib_sub_signaled(ic, nr_sig);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 841) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 842) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 843)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 844) if (unlikely(failed_wr != &send->s_atomic_wr.wr)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 845) printk(KERN_WARNING "RDS/IB: atomic ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 846) BUG_ON(failed_wr != &send->s_atomic_wr.wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 847) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 848)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 849) out:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 850) return ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 851) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 852)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 853) int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 854) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 855) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 856) struct rds_ib_send_work *send = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 857) struct rds_ib_send_work *first;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 858) struct rds_ib_send_work *prev;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 859) const struct ib_send_wr *failed_wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 860) struct scatterlist *scat;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 861) unsigned long len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 862) u64 remote_addr = op->op_remote_addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 863) u32 max_sge = ic->rds_ibdev->max_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 864) u32 pos;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 865) u32 work_alloc;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 866) u32 i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 867) u32 j;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 868) int sent;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 869) int ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 870) int num_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 871) int nr_sig = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 872) u64 odp_addr = op->op_odp_addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 873) u32 odp_lkey = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 874)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 875) /* map the op the first time we see it */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 876) if (!op->op_odp_mr) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 877) if (!op->op_mapped) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 878) op->op_count =
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 879) ib_dma_map_sg(ic->i_cm_id->device, op->op_sg,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 880) op->op_nents,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 881) (op->op_write) ? DMA_TO_DEVICE :
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 882) DMA_FROM_DEVICE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 883) rdsdebug("ic %p mapping op %p: %d\n", ic, op,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 884) op->op_count);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 885) if (op->op_count == 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 886) rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 887) ret = -ENOMEM; /* XXX ? */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 888) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 889) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 890) op->op_mapped = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 891) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 892) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 893) op->op_count = op->op_nents;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 894) odp_lkey = rds_ib_get_lkey(op->op_odp_mr->r_trans_private);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 895) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 896)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 897) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 898) * Instead of knowing how to return a partial rdma read/write we insist that there
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 899) * be enough work requests to send the entire message.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 900) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 901) i = DIV_ROUND_UP(op->op_count, max_sge);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 902)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 903) work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, i, &pos);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 904) if (work_alloc != i) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 905) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 906) rds_ib_stats_inc(s_ib_tx_ring_full);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 907) ret = -ENOMEM;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 908) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 909) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 910)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 911) send = &ic->i_sends[pos];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 912) first = send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 913) prev = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 914) scat = &op->op_sg[0];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 915) sent = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 916) num_sge = op->op_count;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 917)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 918) for (i = 0; i < work_alloc && scat != &op->op_sg[op->op_count]; i++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 919) send->s_wr.send_flags = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 920) send->s_queued = jiffies;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 921) send->s_op = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 922)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 923) if (!op->op_notify)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 924) nr_sig += rds_ib_set_wr_signal_state(ic, send,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 925) op->op_notify);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 926)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 927) send->s_wr.opcode = op->op_write ? IB_WR_RDMA_WRITE : IB_WR_RDMA_READ;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 928) send->s_rdma_wr.remote_addr = remote_addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 929) send->s_rdma_wr.rkey = op->op_rkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 930)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 931) if (num_sge > max_sge) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 932) send->s_rdma_wr.wr.num_sge = max_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 933) num_sge -= max_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 934) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 935) send->s_rdma_wr.wr.num_sge = num_sge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 936) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 937)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 938) send->s_rdma_wr.wr.next = NULL;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 939)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 940) if (prev)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 941) prev->s_rdma_wr.wr.next = &send->s_rdma_wr.wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 942)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 943) for (j = 0; j < send->s_rdma_wr.wr.num_sge &&
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 944) scat != &op->op_sg[op->op_count]; j++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 945) len = sg_dma_len(scat);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 946) if (!op->op_odp_mr) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 947) send->s_sge[j].addr = sg_dma_address(scat);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 948) send->s_sge[j].lkey = ic->i_pd->local_dma_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 949) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 950) send->s_sge[j].addr = odp_addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 951) send->s_sge[j].lkey = odp_lkey;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 952) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 953) send->s_sge[j].length = len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 954)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 955) sent += len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 956) rdsdebug("ic %p sent %d remote_addr %llu\n", ic, sent, remote_addr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 957)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 958) remote_addr += len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 959) odp_addr += len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 960) scat++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 961) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 962)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 963) rdsdebug("send %p wr %p num_sge %u next %p\n", send,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 964) &send->s_rdma_wr.wr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 965) send->s_rdma_wr.wr.num_sge,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 966) send->s_rdma_wr.wr.next);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 967)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 968) prev = send;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 969) if (++send == &ic->i_sends[ic->i_send_ring.w_nr])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 970) send = ic->i_sends;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 971) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 972)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 973) /* give a reference to the last op */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 974) if (scat == &op->op_sg[op->op_count]) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 975) prev->s_op = op;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 976) rds_message_addref(container_of(op, struct rds_message, rdma));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 977) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 978)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 979) if (i < work_alloc) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 980) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - i);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 981) work_alloc = i;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 982) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 983)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 984) if (nr_sig)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 985) atomic_add(nr_sig, &ic->i_signaled_sends);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 986)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 987) failed_wr = &first->s_rdma_wr.wr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 988) ret = ib_post_send(ic->i_cm_id->qp, &first->s_rdma_wr.wr, &failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 989) rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 990) first, &first->s_rdma_wr.wr, ret, failed_wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 991) BUG_ON(failed_wr != &first->s_rdma_wr.wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 992) if (ret) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 993) printk(KERN_WARNING "RDS/IB: rdma ib_post_send to %pI6c "
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 994) "returned %d\n", &conn->c_faddr, ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 995) rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 996) rds_ib_sub_signaled(ic, nr_sig);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 997) goto out;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 998) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 999)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1000) if (unlikely(failed_wr != &first->s_rdma_wr.wr)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1001) printk(KERN_WARNING "RDS/IB: ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1002) BUG_ON(failed_wr != &first->s_rdma_wr.wr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1003) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1004)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1005)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1006) out:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1007) return ret;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1008) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1009)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1010) void rds_ib_xmit_path_complete(struct rds_conn_path *cp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1011) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1012) struct rds_connection *conn = cp->cp_conn;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1013) struct rds_ib_connection *ic = conn->c_transport_data;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1014)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1015) /* We may have a pending ACK or window update we were unable
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1016) * to send previously (due to flow control). Try again. */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1017) rds_ib_attempt_ack(ic);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1018) }
Orange Pi5 kernel
Deprecated Linux kernel 5.10.110 for OrangePi 5/5B/5+ boards
3 Commits
0 Branches
0 Tags