aboutsummaryrefslogtreecommitdiffstats
path: root/net/rds/iw_send.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/rds/iw_send.c')
-rw-r--r--net/rds/iw_send.c93
1 files changed, 47 insertions, 46 deletions
diff --git a/net/rds/iw_send.c b/net/rds/iw_send.c
index 52182ff7519..6280ea020d4 100644
--- a/net/rds/iw_send.c
+++ b/net/rds/iw_send.c
@@ -36,7 +36,6 @@
36#include <linux/dmapool.h> 36#include <linux/dmapool.h>
37 37
38#include "rds.h" 38#include "rds.h"
39#include "rdma.h"
40#include "iw.h" 39#include "iw.h"
41 40
42static void rds_iw_send_rdma_complete(struct rds_message *rm, 41static void rds_iw_send_rdma_complete(struct rds_message *rm,
@@ -64,13 +63,13 @@ static void rds_iw_send_rdma_complete(struct rds_message *rm,
64} 63}
65 64
66static void rds_iw_send_unmap_rdma(struct rds_iw_connection *ic, 65static void rds_iw_send_unmap_rdma(struct rds_iw_connection *ic,
67 struct rds_rdma_op *op) 66 struct rm_rdma_op *op)
68{ 67{
69 if (op->r_mapped) { 68 if (op->op_mapped) {
70 ib_dma_unmap_sg(ic->i_cm_id->device, 69 ib_dma_unmap_sg(ic->i_cm_id->device,
71 op->r_sg, op->r_nents, 70 op->op_sg, op->op_nents,
72 op->r_write ? DMA_TO_DEVICE : DMA_FROM_DEVICE); 71 op->op_write ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
73 op->r_mapped = 0; 72 op->op_mapped = 0;
74 } 73 }
75} 74}
76 75
@@ -83,11 +82,11 @@ static void rds_iw_send_unmap_rm(struct rds_iw_connection *ic,
83 rdsdebug("ic %p send %p rm %p\n", ic, send, rm); 82 rdsdebug("ic %p send %p rm %p\n", ic, send, rm);
84 83
85 ib_dma_unmap_sg(ic->i_cm_id->device, 84 ib_dma_unmap_sg(ic->i_cm_id->device,
86 rm->m_sg, rm->m_nents, 85 rm->data.op_sg, rm->data.op_nents,
87 DMA_TO_DEVICE); 86 DMA_TO_DEVICE);
88 87
89 if (rm->m_rdma_op != NULL) { 88 if (rm->rdma.op_active) {
90 rds_iw_send_unmap_rdma(ic, rm->m_rdma_op); 89 rds_iw_send_unmap_rdma(ic, &rm->rdma);
91 90
92 /* If the user asked for a completion notification on this 91 /* If the user asked for a completion notification on this
93 * message, we can implement three different semantics: 92 * message, we can implement three different semantics:
@@ -111,10 +110,10 @@ static void rds_iw_send_unmap_rm(struct rds_iw_connection *ic,
111 */ 110 */
112 rds_iw_send_rdma_complete(rm, wc_status); 111 rds_iw_send_rdma_complete(rm, wc_status);
113 112
114 if (rm->m_rdma_op->r_write) 113 if (rm->rdma.op_write)
115 rds_stats_add(s_send_rdma_bytes, rm->m_rdma_op->r_bytes); 114 rds_stats_add(s_send_rdma_bytes, rm->rdma.op_bytes);
116 else 115 else
117 rds_stats_add(s_recv_rdma_bytes, rm->m_rdma_op->r_bytes); 116 rds_stats_add(s_recv_rdma_bytes, rm->rdma.op_bytes);
118 } 117 }
119 118
120 /* If anyone waited for this message to get flushed out, wake 119 /* If anyone waited for this message to get flushed out, wake
@@ -556,25 +555,27 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
556 } 555 }
557 556
558 /* map the message the first time we see it */ 557 /* map the message the first time we see it */
559 if (ic->i_rm == NULL) { 558 if (!ic->i_rm) {
560 /* 559 /*
561 printk(KERN_NOTICE "rds_iw_xmit prep msg dport=%u flags=0x%x len=%d\n", 560 printk(KERN_NOTICE "rds_iw_xmit prep msg dport=%u flags=0x%x len=%d\n",
562 be16_to_cpu(rm->m_inc.i_hdr.h_dport), 561 be16_to_cpu(rm->m_inc.i_hdr.h_dport),
563 rm->m_inc.i_hdr.h_flags, 562 rm->m_inc.i_hdr.h_flags,
564 be32_to_cpu(rm->m_inc.i_hdr.h_len)); 563 be32_to_cpu(rm->m_inc.i_hdr.h_len));
565 */ 564 */
566 if (rm->m_nents) { 565 if (rm->data.op_nents) {
567 rm->m_count = ib_dma_map_sg(dev, 566 rm->data.op_count = ib_dma_map_sg(dev,
568 rm->m_sg, rm->m_nents, DMA_TO_DEVICE); 567 rm->data.op_sg,
569 rdsdebug("ic %p mapping rm %p: %d\n", ic, rm, rm->m_count); 568 rm->data.op_nents,
570 if (rm->m_count == 0) { 569 DMA_TO_DEVICE);
570 rdsdebug("ic %p mapping rm %p: %d\n", ic, rm, rm->data.op_count);
571 if (rm->data.op_count == 0) {
571 rds_iw_stats_inc(s_iw_tx_sg_mapping_failure); 572 rds_iw_stats_inc(s_iw_tx_sg_mapping_failure);
572 rds_iw_ring_unalloc(&ic->i_send_ring, work_alloc); 573 rds_iw_ring_unalloc(&ic->i_send_ring, work_alloc);
573 ret = -ENOMEM; /* XXX ? */ 574 ret = -ENOMEM; /* XXX ? */
574 goto out; 575 goto out;
575 } 576 }
576 } else { 577 } else {
577 rm->m_count = 0; 578 rm->data.op_count = 0;
578 } 579 }
579 580
580 ic->i_unsignaled_wrs = rds_iw_sysctl_max_unsig_wrs; 581 ic->i_unsignaled_wrs = rds_iw_sysctl_max_unsig_wrs;
@@ -590,10 +591,10 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
590 591
591 /* If it has a RDMA op, tell the peer we did it. This is 592 /* If it has a RDMA op, tell the peer we did it. This is
592 * used by the peer to release use-once RDMA MRs. */ 593 * used by the peer to release use-once RDMA MRs. */
593 if (rm->m_rdma_op) { 594 if (rm->rdma.op_active) {
594 struct rds_ext_header_rdma ext_hdr; 595 struct rds_ext_header_rdma ext_hdr;
595 596
596 ext_hdr.h_rdma_rkey = cpu_to_be32(rm->m_rdma_op->r_key); 597 ext_hdr.h_rdma_rkey = cpu_to_be32(rm->rdma.op_rkey);
597 rds_message_add_extension(&rm->m_inc.i_hdr, 598 rds_message_add_extension(&rm->m_inc.i_hdr,
598 RDS_EXTHDR_RDMA, &ext_hdr, sizeof(ext_hdr)); 599 RDS_EXTHDR_RDMA, &ext_hdr, sizeof(ext_hdr));
599 } 600 }
@@ -621,7 +622,7 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
621 send = &ic->i_sends[pos]; 622 send = &ic->i_sends[pos];
622 first = send; 623 first = send;
623 prev = NULL; 624 prev = NULL;
624 scat = &rm->m_sg[sg]; 625 scat = &rm->data.op_sg[sg];
625 sent = 0; 626 sent = 0;
626 i = 0; 627 i = 0;
627 628
@@ -631,7 +632,7 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
631 * or when requested by the user. Right now, we let 632 * or when requested by the user. Right now, we let
632 * the application choose. 633 * the application choose.
633 */ 634 */
634 if (rm->m_rdma_op && rm->m_rdma_op->r_fence) 635 if (rm->rdma.op_active && rm->rdma.op_fence)
635 send_flags = IB_SEND_FENCE; 636 send_flags = IB_SEND_FENCE;
636 637
637 /* 638 /*
@@ -650,7 +651,7 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
650 } 651 }
651 652
652 /* if there's data reference it with a chain of work reqs */ 653 /* if there's data reference it with a chain of work reqs */
653 for (; i < work_alloc && scat != &rm->m_sg[rm->m_count]; i++) { 654 for (; i < work_alloc && scat != &rm->data.op_sg[rm->data.op_count]; i++) {
654 unsigned int len; 655 unsigned int len;
655 656
656 send = &ic->i_sends[pos]; 657 send = &ic->i_sends[pos];
@@ -728,7 +729,7 @@ add_header:
728 sent += sizeof(struct rds_header); 729 sent += sizeof(struct rds_header);
729 730
730 /* if we finished the message then send completion owns it */ 731 /* if we finished the message then send completion owns it */
731 if (scat == &rm->m_sg[rm->m_count]) { 732 if (scat == &rm->data.op_sg[rm->data.op_count]) {
732 prev->s_rm = ic->i_rm; 733 prev->s_rm = ic->i_rm;
733 prev->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED; 734 prev->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
734 ic->i_rm = NULL; 735 ic->i_rm = NULL;
@@ -784,7 +785,7 @@ static void rds_iw_build_send_fastreg(struct rds_iw_device *rds_iwdev, struct rd
784 ib_update_fast_reg_key(send->s_mr, send->s_remap_count++); 785 ib_update_fast_reg_key(send->s_mr, send->s_remap_count++);
785} 786}
786 787
787int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op) 788int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
788{ 789{
789 struct rds_iw_connection *ic = conn->c_transport_data; 790 struct rds_iw_connection *ic = conn->c_transport_data;
790 struct rds_iw_send_work *send = NULL; 791 struct rds_iw_send_work *send = NULL;
@@ -794,7 +795,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
794 struct rds_iw_device *rds_iwdev; 795 struct rds_iw_device *rds_iwdev;
795 struct scatterlist *scat; 796 struct scatterlist *scat;
796 unsigned long len; 797 unsigned long len;
797 u64 remote_addr = op->r_remote_addr; 798 u64 remote_addr = op->op_remote_addr;
798 u32 pos, fr_pos; 799 u32 pos, fr_pos;
799 u32 work_alloc; 800 u32 work_alloc;
800 u32 i; 801 u32 i;
@@ -806,21 +807,21 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
806 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client); 807 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
807 808
808 /* map the message the first time we see it */ 809 /* map the message the first time we see it */
809 if (!op->r_mapped) { 810 if (!op->op_mapped) {
810 op->r_count = ib_dma_map_sg(ic->i_cm_id->device, 811 op->op_count = ib_dma_map_sg(ic->i_cm_id->device,
811 op->r_sg, op->r_nents, (op->r_write) ? 812 op->op_sg, op->op_nents, (op->op_write) ?
812 DMA_TO_DEVICE : DMA_FROM_DEVICE); 813 DMA_TO_DEVICE : DMA_FROM_DEVICE);
813 rdsdebug("ic %p mapping op %p: %d\n", ic, op, op->r_count); 814 rdsdebug("ic %p mapping op %p: %d\n", ic, op, op->op_count);
814 if (op->r_count == 0) { 815 if (op->op_count == 0) {
815 rds_iw_stats_inc(s_iw_tx_sg_mapping_failure); 816 rds_iw_stats_inc(s_iw_tx_sg_mapping_failure);
816 ret = -ENOMEM; /* XXX ? */ 817 ret = -ENOMEM; /* XXX ? */
817 goto out; 818 goto out;
818 } 819 }
819 820
820 op->r_mapped = 1; 821 op->op_mapped = 1;
821 } 822 }
822 823
823 if (!op->r_write) { 824 if (!op->op_write) {
824 /* Alloc space on the send queue for the fastreg */ 825 /* Alloc space on the send queue for the fastreg */
825 work_alloc = rds_iw_ring_alloc(&ic->i_send_ring, 1, &fr_pos); 826 work_alloc = rds_iw_ring_alloc(&ic->i_send_ring, 1, &fr_pos);
826 if (work_alloc != 1) { 827 if (work_alloc != 1) {
@@ -835,7 +836,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
835 * Instead of knowing how to return a partial rdma read/write we insist that there 836 * Instead of knowing how to return a partial rdma read/write we insist that there
836 * be enough work requests to send the entire message. 837 * be enough work requests to send the entire message.
837 */ 838 */
838 i = ceil(op->r_count, rds_iwdev->max_sge); 839 i = ceil(op->op_count, rds_iwdev->max_sge);
839 840
840 work_alloc = rds_iw_ring_alloc(&ic->i_send_ring, i, &pos); 841 work_alloc = rds_iw_ring_alloc(&ic->i_send_ring, i, &pos);
841 if (work_alloc != i) { 842 if (work_alloc != i) {
@@ -846,17 +847,17 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
846 } 847 }
847 848
848 send = &ic->i_sends[pos]; 849 send = &ic->i_sends[pos];
849 if (!op->r_write) { 850 if (!op->op_write) {
850 first = prev = &ic->i_sends[fr_pos]; 851 first = prev = &ic->i_sends[fr_pos];
851 } else { 852 } else {
852 first = send; 853 first = send;
853 prev = NULL; 854 prev = NULL;
854 } 855 }
855 scat = &op->r_sg[0]; 856 scat = &op->op_sg[0];
856 sent = 0; 857 sent = 0;
857 num_sge = op->r_count; 858 num_sge = op->op_count;
858 859
859 for (i = 0; i < work_alloc && scat != &op->r_sg[op->r_count]; i++) { 860 for (i = 0; i < work_alloc && scat != &op->op_sg[op->op_count]; i++) {
860 send->s_wr.send_flags = 0; 861 send->s_wr.send_flags = 0;
861 send->s_queued = jiffies; 862 send->s_queued = jiffies;
862 863
@@ -873,13 +874,13 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
873 * for local access after RDS is finished with it, using 874 * for local access after RDS is finished with it, using
874 * IB_WR_RDMA_READ_WITH_INV will invalidate it after the read has completed. 875 * IB_WR_RDMA_READ_WITH_INV will invalidate it after the read has completed.
875 */ 876 */
876 if (op->r_write) 877 if (op->op_write)
877 send->s_wr.opcode = IB_WR_RDMA_WRITE; 878 send->s_wr.opcode = IB_WR_RDMA_WRITE;
878 else 879 else
879 send->s_wr.opcode = IB_WR_RDMA_READ_WITH_INV; 880 send->s_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
880 881
881 send->s_wr.wr.rdma.remote_addr = remote_addr; 882 send->s_wr.wr.rdma.remote_addr = remote_addr;
882 send->s_wr.wr.rdma.rkey = op->r_key; 883 send->s_wr.wr.rdma.rkey = op->op_rkey;
883 send->s_op = op; 884 send->s_op = op;
884 885
885 if (num_sge > rds_iwdev->max_sge) { 886 if (num_sge > rds_iwdev->max_sge) {
@@ -893,7 +894,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
893 if (prev) 894 if (prev)
894 prev->s_wr.next = &send->s_wr; 895 prev->s_wr.next = &send->s_wr;
895 896
896 for (j = 0; j < send->s_wr.num_sge && scat != &op->r_sg[op->r_count]; j++) { 897 for (j = 0; j < send->s_wr.num_sge && scat != &op->op_sg[op->op_count]; j++) {
897 len = ib_sg_dma_len(ic->i_cm_id->device, scat); 898 len = ib_sg_dma_len(ic->i_cm_id->device, scat);
898 899
899 if (send->s_wr.opcode == IB_WR_RDMA_READ_WITH_INV) 900 if (send->s_wr.opcode == IB_WR_RDMA_READ_WITH_INV)
@@ -927,7 +928,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
927 } 928 }
928 929
929 /* if we finished the message then send completion owns it */ 930 /* if we finished the message then send completion owns it */
930 if (scat == &op->r_sg[op->r_count]) 931 if (scat == &op->op_sg[op->op_count])
931 first->s_wr.send_flags = IB_SEND_SIGNALED; 932 first->s_wr.send_flags = IB_SEND_SIGNALED;
932 933
933 if (i < work_alloc) { 934 if (i < work_alloc) {
@@ -941,9 +942,9 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op)
941 * adapters do not allow using the lkey for this at all. To bypass this use a 942 * adapters do not allow using the lkey for this at all. To bypass this use a
942 * fastreg_mr (or possibly a dma_mr) 943 * fastreg_mr (or possibly a dma_mr)
943 */ 944 */
944 if (!op->r_write) { 945 if (!op->op_write) {
945 rds_iw_build_send_fastreg(rds_iwdev, ic, &ic->i_sends[fr_pos], 946 rds_iw_build_send_fastreg(rds_iwdev, ic, &ic->i_sends[fr_pos],
946 op->r_count, sent, conn->c_xmit_rm->m_rs->rs_user_addr); 947 op->op_count, sent, conn->c_xmit_rm->m_rs->rs_user_addr);
947 work_alloc++; 948 work_alloc++;
948 } 949 }
949 950