/*
*
* This should be called with the QP r_lock held.
*
* The s_lock will be acquired around the qib_migrate_qp() call.
*/
int qib_ruc_check_hdr(struct qib_ibport *ibp, struct ib_header *hdr,
int has_grh, struct rvt_qp *qp, u32 bth0)
{
__be64 guid;
unsigned long flags;
if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
if (!has_grh) {
if (rdma_ah_get_ah_flags(&qp->alt_ah_attr) &
IB_AH_GRH)
goto err;
} else {
const struct ib_global_route *grh;
if (!(rdma_ah_get_ah_flags(&qp->alt_ah_attr) &
IB_AH_GRH))
goto err;
grh = rdma_ah_read_grh(&qp->alt_ah_attr);
guid = get_sguid(ibp, grh->sgid_index);
if (!gid_ok(&hdr->u.l.grh.dgid,
ibp->rvp.gid_prefix, guid))
goto err;
if (!gid_ok(&hdr->u.l.grh.sgid,
grh->dgid.global.subnet_prefix,
grh->dgid.global.interface_id))
goto err;
}
if (!qib_pkey_ok((u16)bth0,
qib_get_pkey(ibp, qp->s_alt_pkey_index))) {
qib_bad_pkey(ibp,
(u16)bth0,
(be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
0, qp->ibqp.qp_num,
hdr->lrh[3], hdr->lrh[1]);
goto err;
}
/* Validate the SLID. See Ch. 9.6.1.5 and 17.2.8 */
if ((be16_to_cpu(hdr->lrh[3]) !=
rdma_ah_get_dlid(&qp->alt_ah_attr)) ||
ppd_from_ibp(ibp)->port !=
rdma_ah_get_port_num(&qp->alt_ah_attr))
goto err;
spin_lock_irqsave(&qp->s_lock, flags);
qib_migrate_qp(qp);
spin_unlock_irqrestore(&qp->s_lock, flags);
} else {
static struct ib_ah *create_ib_ah(struct mlx5_ib_dev *dev,
struct mlx5_ib_ah *ah,
struct rdma_ah_attr *ah_attr)
{
enum ib_gid_type gid_type;
int err;
if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
const struct ib_global_route *grh = rdma_ah_read_grh(ah_attr);
memcpy(ah->av.rgid, &grh->dgid, 16);
ah->av.grh_gid_fl = cpu_to_be32(grh->flow_label |
(1 << 30) |
grh->sgid_index << 20);
ah->av.hop_limit = grh->hop_limit;
ah->av.tclass = grh->traffic_class;
}
ah->av.stat_rate_sl = (rdma_ah_get_static_rate(ah_attr) << 4);
if (ah_attr->type == RDMA_AH_ATTR_TYPE_ROCE) {
err = mlx5_get_roce_gid_type(dev, ah_attr->port_num,
ah_attr->grh.sgid_index,
&gid_type);
if (err)
return ERR_PTR(err);
memcpy(ah->av.rmac, ah_attr->roce.dmac,
sizeof(ah_attr->roce.dmac));
ah->av.udp_sport =
mlx5_get_roce_udp_sport(dev,
rdma_ah_get_port_num(ah_attr),
rdma_ah_read_grh(ah_attr)->sgid_index);
ah->av.stat_rate_sl |= (rdma_ah_get_sl(ah_attr) & 0x7) << 1;
if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
#define MLX5_ECN_ENABLED BIT(1)
ah->av.tclass |= MLX5_ECN_ENABLED;
} else {
ah->av.rlid = cpu_to_be16(rdma_ah_get_dlid(ah_attr));
ah->av.fl_mlid = rdma_ah_get_path_bits(ah_attr) & 0x7f;
ah->av.stat_rate_sl |= (rdma_ah_get_sl(ah_attr) & 0xf);
}
return &ah->ibah;
}
/**
* hfi1_make_uc_req - construct a request packet (SEND, RDMA write)
* @qp: a pointer to the QP
*
* Assume s_lock is held.
*
* Return 1 if constructed; otherwise, return 0.
*/
int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
{
struct hfi1_qp_priv *priv = qp->priv;
struct ib_other_headers *ohdr;
struct rvt_swqe *wqe;
u32 hwords;
u32 bth0 = 0;
u32 len;
u32 pmtu = qp->pmtu;
int middle = 0;
ps->s_txreq = get_txreq(ps->dev, qp);
if (IS_ERR(ps->s_txreq))
goto bail_no_tx;
if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK)) {
if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
goto bail;
/* We are in the error state, flush the work request. */
if (qp->s_last == READ_ONCE(qp->s_head))
goto bail;
/* If DMAs are in progress, we can't flush immediately. */
if (iowait_sdma_pending(&priv->s_iowait)) {
qp->s_flags |= RVT_S_WAIT_DMA;
goto bail;
}
clear_ahg(qp);
wqe = rvt_get_swqe_ptr(qp, qp->s_last);
hfi1_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
goto done_free_tx;
}
if (priv->hdr_type == HFI1_PKT_TYPE_9B) {
/* header size in 32-bit words LRH+BTH = (8+12)/4. */
hwords = 5;
if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
else
ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;
} else {
/* header size in 32-bit words 16B LRH+BTH = (16+12)/4. */
hwords = 7;
if ((rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
(hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))))
ohdr = &ps->s_txreq->phdr.hdr.opah.u.l.oth;
else
ohdr = &ps->s_txreq->phdr.hdr.opah.u.oth;
}
/* Get the next send request. */
wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
qp->s_wqe = NULL;
switch (qp->s_state) {
default:
if (!(ib_rvt_state_ops[qp->state] &
RVT_PROCESS_NEXT_SEND_OK))
goto bail;
/* Check if send work queue is empty. */
if (qp->s_cur == READ_ONCE(qp->s_head)) {
clear_ahg(qp);
goto bail;
}
/*
* Local operations are processed immediately
* after all prior requests have completed.
*/
if (wqe->wr.opcode == IB_WR_REG_MR ||
wqe->wr.opcode == IB_WR_LOCAL_INV) {
int local_ops = 0;
int err = 0;
if (qp->s_last != qp->s_cur)
goto bail;
if (++qp->s_cur == qp->s_size)
qp->s_cur = 0;
if (!(wqe->wr.send_flags & RVT_SEND_COMPLETION_ONLY)) {
err = rvt_invalidate_rkey(
qp, wqe->wr.ex.invalidate_rkey);
local_ops = 1;
}
hfi1_send_complete(qp, wqe, err ? IB_WC_LOC_PROT_ERR
: IB_WC_SUCCESS);
if (local_ops)
atomic_dec(&qp->local_ops_pending);
goto done_free_tx;
}
/*
* Start a new request.
*/
qp->s_psn = wqe->psn;
qp->s_sge.sge = wqe->sg_list[0];
qp->s_sge.sg_list = wqe->sg_list + 1;
qp->s_sge.num_sge = wqe->wr.num_sge;
qp->s_sge.total_len = wqe->length;
len = wqe->length;
qp->s_len = len;
switch (wqe->wr.opcode) {
case IB_WR_SEND:
case IB_WR_SEND_WITH_IMM:
if (len > pmtu) {
qp->s_state = OP(SEND_FIRST);
len = pmtu;
break;
}
if (wqe->wr.opcode == IB_WR_SEND) {
qp->s_state = OP(SEND_ONLY);
} else {
qp->s_state =
OP(SEND_ONLY_WITH_IMMEDIATE);
/* Immediate data comes after the BTH */
ohdr->u.imm_data = wqe->wr.ex.imm_data;
hwords += 1;
}
if (wqe->wr.send_flags & IB_SEND_SOLICITED)
bth0 |= IB_BTH_SOLICITED;
qp->s_wqe = wqe;
if (++qp->s_cur >= qp->s_size)
qp->s_cur = 0;
break;
case IB_WR_RDMA_WRITE:
case IB_WR_RDMA_WRITE_WITH_IMM:
ohdr->u.rc.reth.vaddr =
cpu_to_be64(wqe->rdma_wr.remote_addr);
ohdr->u.rc.reth.rkey =
cpu_to_be32(wqe->rdma_wr.rkey);
ohdr->u.rc.reth.length = cpu_to_be32(len);
hwords += sizeof(struct ib_reth) / 4;
if (len > pmtu) {
qp->s_state = OP(RDMA_WRITE_FIRST);
len = pmtu;
break;
}
if (wqe->wr.opcode == IB_WR_RDMA_WRITE) {
qp->s_state = OP(RDMA_WRITE_ONLY);
} else {
qp->s_state =
OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
/* Immediate data comes after the RETH */
ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
hwords += 1;
if (wqe->wr.send_flags & IB_SEND_SOLICITED)
bth0 |= IB_BTH_SOLICITED;
}
qp->s_wqe = wqe;
if (++qp->s_cur >= qp->s_size)
qp->s_cur = 0;
break;
default:
goto bail;
}
break;
case OP(SEND_FIRST):
qp->s_state = OP(SEND_MIDDLE);
/* FALLTHROUGH */
case OP(SEND_MIDDLE):
len = qp->s_len;
if (len > pmtu) {
len = pmtu;
middle = HFI1_CAP_IS_KSET(SDMA_AHG);
break;
}
if (wqe->wr.opcode == IB_WR_SEND) {
qp->s_state = OP(SEND_LAST);
} else {
qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
/* Immediate data comes after the BTH */
ohdr->u.imm_data = wqe->wr.ex.imm_data;
hwords += 1;
}
if (wqe->wr.send_flags & IB_SEND_SOLICITED)
bth0 |= IB_BTH_SOLICITED;
qp->s_wqe = wqe;
if (++qp->s_cur >= qp->s_size)
qp->s_cur = 0;
break;
case OP(RDMA_WRITE_FIRST):
qp->s_state = OP(RDMA_WRITE_MIDDLE);
/* FALLTHROUGH */
case OP(RDMA_WRITE_MIDDLE):
len = qp->s_len;
if (len > pmtu) {
len = pmtu;
middle = HFI1_CAP_IS_KSET(SDMA_AHG);
break;
}
if (wqe->wr.opcode == IB_WR_RDMA_WRITE) {
qp->s_state = OP(RDMA_WRITE_LAST);
} else {
qp->s_state =
OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
/* Immediate data comes after the BTH */
ohdr->u.imm_data = wqe->wr.ex.imm_data;
hwords += 1;
if (wqe->wr.send_flags & IB_SEND_SOLICITED)
bth0 |= IB_BTH_SOLICITED;
}
qp->s_wqe = wqe;
if (++qp->s_cur >= qp->s_size)
qp->s_cur = 0;
break;
}
qp->s_len -= len;
ps->s_txreq->hdr_dwords = hwords;
ps->s_txreq->sde = priv->s_sde;
ps->s_txreq->ss = &qp->s_sge;
ps->s_txreq->s_cur_size = len;
hfi1_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
mask_psn(qp->s_psn++), middle, ps);
return 1;
done_free_tx:
hfi1_put_txreq(ps->s_txreq);
ps->s_txreq = NULL;
return 1;
bail:
hfi1_put_txreq(ps->s_txreq);
bail_no_tx:
ps->s_txreq = NULL;
qp->s_flags &= ~RVT_S_BUSY;
return 0;
}
/**
* hfi1_uc_rcv - handle an incoming UC packet
* @ibp: the port the packet came in on
* @hdr: the header of the packet
* @rcv_flags: flags relevant to rcv processing
* @data: the packet data
* @tlen: the length of the packet
* @qp: the QP for this packet.
*
* This is called from qp_rcv() to process an incoming UC packet
* for the given QP.
* Called at interrupt level.
*/
void hfi1_uc_rcv(struct hfi1_packet *packet)
{
struct hfi1_ibport *ibp = rcd_to_iport(packet->rcd);
void *data = packet->payload;
u32 tlen = packet->tlen;
struct rvt_qp *qp = packet->qp;
struct ib_other_headers *ohdr = packet->ohdr;
u32 opcode = packet->opcode;
u32 hdrsize = packet->hlen;
u32 psn;
u32 pad = packet->pad;
struct ib_wc wc;
u32 pmtu = qp->pmtu;
struct ib_reth *reth;
int ret;
u8 extra_bytes = pad + packet->extra_byte + (SIZE_OF_CRC << 2);
if (hfi1_ruc_check_hdr(ibp, packet))
return;
process_ecn(qp, packet, true);
psn = ib_bth_get_psn(ohdr);
/* Compare the PSN verses the expected PSN. */
if (unlikely(cmp_psn(psn, qp->r_psn) != 0)) {
/*
* Handle a sequence error.
* Silently drop any current message.
*/
qp->r_psn = psn;
inv:
if (qp->r_state == OP(SEND_FIRST) ||
qp->r_state == OP(SEND_MIDDLE)) {
set_bit(RVT_R_REWIND_SGE, &qp->r_aflags);
qp->r_sge.num_sge = 0;
} else {
rvt_put_ss(&qp->r_sge);
}
qp->r_state = OP(SEND_LAST);
switch (opcode) {
case OP(SEND_FIRST):
case OP(SEND_ONLY):
case OP(SEND_ONLY_WITH_IMMEDIATE):
goto send_first;
case OP(RDMA_WRITE_FIRST):
case OP(RDMA_WRITE_ONLY):
case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
goto rdma_first;
default:
goto drop;
}
}
/* Check for opcode sequence errors. */
switch (qp->r_state) {
case OP(SEND_FIRST):
case OP(SEND_MIDDLE):
if (opcode == OP(SEND_MIDDLE) ||
opcode == OP(SEND_LAST) ||
opcode == OP(SEND_LAST_WITH_IMMEDIATE))
break;
goto inv;
case OP(RDMA_WRITE_FIRST):
case OP(RDMA_WRITE_MIDDLE):
if (opcode == OP(RDMA_WRITE_MIDDLE) ||
opcode == OP(RDMA_WRITE_LAST) ||
opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
break;
goto inv;
default:
if (opcode == OP(SEND_FIRST) ||
opcode == OP(SEND_ONLY) ||
opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
opcode == OP(RDMA_WRITE_FIRST) ||
opcode == OP(RDMA_WRITE_ONLY) ||
opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
break;
goto inv;
}
if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST))
rvt_comm_est(qp);
/* OK, process the packet. */
switch (opcode) {
case OP(SEND_FIRST):
case OP(SEND_ONLY):
case OP(SEND_ONLY_WITH_IMMEDIATE):
send_first:
if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags)) {
qp->r_sge = qp->s_rdma_read_sge;
} else {
ret = hfi1_rvt_get_rwqe(qp, 0);
if (ret < 0)
goto op_err;
if (!ret)
goto drop;
/*
* qp->s_rdma_read_sge will be the owner
* of the mr references.
*/
qp->s_rdma_read_sge = qp->r_sge;
}
qp->r_rcv_len = 0;
if (opcode == OP(SEND_ONLY))
goto no_immediate_data;
else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
goto send_last_imm;
/* FALLTHROUGH */
case OP(SEND_MIDDLE):
/* Check for invalid length PMTU or posted rwqe len. */
/*
* There will be no padding for 9B packet but 16B packets
* will come in with some padding since we always add
* CRC and LT bytes which will need to be flit aligned
*/
if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
goto rewind;
qp->r_rcv_len += pmtu;
if (unlikely(qp->r_rcv_len > qp->r_len))
goto rewind;
hfi1_copy_sge(&qp->r_sge, data, pmtu, false, false);
break;
case OP(SEND_LAST_WITH_IMMEDIATE):
send_last_imm:
wc.ex.imm_data = ohdr->u.imm_data;
wc.wc_flags = IB_WC_WITH_IMM;
goto send_last;
case OP(SEND_LAST):
no_immediate_data:
wc.ex.imm_data = 0;
wc.wc_flags = 0;
send_last:
/* Check for invalid length. */
/* LAST len should be >= 1 */
if (unlikely(tlen < (hdrsize + extra_bytes)))
goto rewind;
/* Don't count the CRC. */
tlen -= (hdrsize + extra_bytes);
wc.byte_len = tlen + qp->r_rcv_len;
if (unlikely(wc.byte_len > qp->r_len))
goto rewind;
wc.opcode = IB_WC_RECV;
hfi1_copy_sge(&qp->r_sge, data, tlen, false, false);
rvt_put_ss(&qp->s_rdma_read_sge);
last_imm:
wc.wr_id = qp->r_wr_id;
wc.status = IB_WC_SUCCESS;
wc.qp = &qp->ibqp;
wc.src_qp = qp->remote_qpn;
wc.slid = rdma_ah_get_dlid(&qp->remote_ah_attr) & U16_MAX;
/*
* It seems that IB mandates the presence of an SL in a
* work completion only for the UD transport (see section
* 11.4.2 of IBTA Vol. 1).
*
* However, the way the SL is chosen below is consistent
* with the way that IB/qib works and is trying avoid
* introducing incompatibilities.
*
* See also OPA Vol. 1, section 9.7.6, and table 9-17.
*/
wc.sl = rdma_ah_get_sl(&qp->remote_ah_attr);
/* zero fields that are N/A */
wc.vendor_err = 0;
wc.pkey_index = 0;
wc.dlid_path_bits = 0;
wc.port_num = 0;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
ib_bth_is_solicited(ohdr));
break;
case OP(RDMA_WRITE_FIRST):
case OP(RDMA_WRITE_ONLY):
case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
rdma_first:
if (unlikely(!(qp->qp_access_flags &
IB_ACCESS_REMOTE_WRITE))) {
goto drop;
}
reth = &ohdr->u.rc.reth;
qp->r_len = be32_to_cpu(reth->length);
qp->r_rcv_len = 0;
qp->r_sge.sg_list = NULL;
if (qp->r_len != 0) {
u32 rkey = be32_to_cpu(reth->rkey);
u64 vaddr = be64_to_cpu(reth->vaddr);
int ok;
/* Check rkey */
ok = rvt_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
if (unlikely(!ok))
goto drop;
qp->r_sge.num_sge = 1;
} else {
qp->r_sge.num_sge = 0;
qp->r_sge.sge.mr = NULL;
qp->r_sge.sge.vaddr = NULL;
qp->r_sge.sge.length = 0;
qp->r_sge.sge.sge_length = 0;
}
if (opcode == OP(RDMA_WRITE_ONLY)) {
goto rdma_last;
} else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE)) {
wc.ex.imm_data = ohdr->u.rc.imm_data;
goto rdma_last_imm;
}
/* FALLTHROUGH */
case OP(RDMA_WRITE_MIDDLE):
/* Check for invalid length PMTU or posted rwqe len. */
if (unlikely(tlen != (hdrsize + pmtu + 4)))
goto drop;
qp->r_rcv_len += pmtu;
if (unlikely(qp->r_rcv_len > qp->r_len))
goto drop;
hfi1_copy_sge(&qp->r_sge, data, pmtu, true, false);
break;
case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
wc.ex.imm_data = ohdr->u.imm_data;
rdma_last_imm:
wc.wc_flags = IB_WC_WITH_IMM;
/* Check for invalid length. */
/* LAST len should be >= 1 */
if (unlikely(tlen < (hdrsize + pad + 4)))
goto drop;
/* Don't count the CRC. */
tlen -= (hdrsize + extra_bytes);
if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
goto drop;
if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags)) {
rvt_put_ss(&qp->s_rdma_read_sge);
} else {
ret = hfi1_rvt_get_rwqe(qp, 1);
if (ret < 0)
goto op_err;
if (!ret)
goto drop;
}
wc.byte_len = qp->r_len;
wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
hfi1_copy_sge(&qp->r_sge, data, tlen, true, false);
rvt_put_ss(&qp->r_sge);
goto last_imm;
case OP(RDMA_WRITE_LAST):
rdma_last:
/* Check for invalid length. */
/* LAST len should be >= 1 */
if (unlikely(tlen < (hdrsize + pad + 4)))
goto drop;
/* Don't count the CRC. */
tlen -= (hdrsize + extra_bytes);
if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
goto drop;
hfi1_copy_sge(&qp->r_sge, data, tlen, true, false);
rvt_put_ss(&qp->r_sge);
break;
default:
/* Drop packet for unknown opcodes. */
goto drop;
}
qp->r_psn++;
qp->r_state = opcode;
return;
rewind:
set_bit(RVT_R_REWIND_SGE, &qp->r_aflags);
qp->r_sge.num_sge = 0;
drop:
ibp->rvp.n_pkt_drops++;
return;
op_err:
rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
}
