[linux-2.6-block.git] / drivers / infiniband / sw / rxe / rxe_comp.c

/*
 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *	- Redistributions of source code must retain the above
 *	  copyright notice, this list of conditions and the following
 *	  disclaimer.
 *
 *	- Redistributions in binary form must reproduce the above
 *	  copyright notice, this list of conditions and the following
 *	  disclaimer in the documentation and/or other materials
 *	  provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/skbuff.h>

#include "rxe.h"
#include "rxe_loc.h"
#include "rxe_queue.h"
#include "rxe_task.h"

enum comp_state {
	COMPST_GET_ACK,
	COMPST_GET_WQE,
	COMPST_COMP_WQE,
	COMPST_COMP_ACK,
	COMPST_CHECK_PSN,
	COMPST_CHECK_ACK,
	COMPST_READ,
	COMPST_ATOMIC,
	COMPST_WRITE_SEND,
	COMPST_UPDATE_COMP,
	COMPST_ERROR_RETRY,
	COMPST_RNR_RETRY,
	COMPST_ERROR,
	COMPST_EXIT, /* We have an issue, and we want to rerun the completer */
	COMPST_DONE, /* The completer finished successflly */
};

static char *comp_state_name[] =  {
	[COMPST_GET_ACK]		= "GET ACK",
	[COMPST_GET_WQE]		= "GET WQE",
	[COMPST_COMP_WQE]		= "COMP WQE",
	[COMPST_COMP_ACK]		= "COMP ACK",
	[COMPST_CHECK_PSN]		= "CHECK PSN",
	[COMPST_CHECK_ACK]		= "CHECK ACK",
	[COMPST_READ]			= "READ",
	[COMPST_ATOMIC]			= "ATOMIC",
	[COMPST_WRITE_SEND]		= "WRITE/SEND",
	[COMPST_UPDATE_COMP]		= "UPDATE COMP",
	[COMPST_ERROR_RETRY]		= "ERROR RETRY",
	[COMPST_RNR_RETRY]		= "RNR RETRY",
	[COMPST_ERROR]			= "ERROR",
	[COMPST_EXIT]			= "EXIT",
	[COMPST_DONE]			= "DONE",
};

static unsigned long rnrnak_usec[32] = {
	[IB_RNR_TIMER_655_36] = 655360,
	[IB_RNR_TIMER_000_01] = 10,
	[IB_RNR_TIMER_000_02] = 20,
	[IB_RNR_TIMER_000_03] = 30,
	[IB_RNR_TIMER_000_04] = 40,
	[IB_RNR_TIMER_000_06] = 60,
	[IB_RNR_TIMER_000_08] = 80,
	[IB_RNR_TIMER_000_12] = 120,
	[IB_RNR_TIMER_000_16] = 160,
	[IB_RNR_TIMER_000_24] = 240,
	[IB_RNR_TIMER_000_32] = 320,
	[IB_RNR_TIMER_000_48] = 480,
	[IB_RNR_TIMER_000_64] = 640,
	[IB_RNR_TIMER_000_96] = 960,
	[IB_RNR_TIMER_001_28] = 1280,
	[IB_RNR_TIMER_001_92] = 1920,
	[IB_RNR_TIMER_002_56] = 2560,
	[IB_RNR_TIMER_003_84] = 3840,
	[IB_RNR_TIMER_005_12] = 5120,
	[IB_RNR_TIMER_007_68] = 7680,
	[IB_RNR_TIMER_010_24] = 10240,
	[IB_RNR_TIMER_015_36] = 15360,
	[IB_RNR_TIMER_020_48] = 20480,
	[IB_RNR_TIMER_030_72] = 30720,
	[IB_RNR_TIMER_040_96] = 40960,
	[IB_RNR_TIMER_061_44] = 61410,
	[IB_RNR_TIMER_081_92] = 81920,
	[IB_RNR_TIMER_122_88] = 122880,
	[IB_RNR_TIMER_163_84] = 163840,
	[IB_RNR_TIMER_245_76] = 245760,
	[IB_RNR_TIMER_327_68] = 327680,
	[IB_RNR_TIMER_491_52] = 491520,
};

static inline unsigned long rnrnak_jiffies(u8 timeout)
{
	return max_t(unsigned long,
		usecs_to_jiffies(rnrnak_usec[timeout]), 1);
}

static enum ib_wc_opcode wr_to_wc_opcode(enum ib_wr_opcode opcode)
{
	switch (opcode) {
	case IB_WR_RDMA_WRITE:			return IB_WC_RDMA_WRITE;
	case IB_WR_RDMA_WRITE_WITH_IMM:		return IB_WC_RDMA_WRITE;
	case IB_WR_SEND:			return IB_WC_SEND;
	case IB_WR_SEND_WITH_IMM:		return IB_WC_SEND;
	case IB_WR_RDMA_READ:			return IB_WC_RDMA_READ;
	case IB_WR_ATOMIC_CMP_AND_SWP:		return IB_WC_COMP_SWAP;
	case IB_WR_ATOMIC_FETCH_AND_ADD:	return IB_WC_FETCH_ADD;
	case IB_WR_LSO:				return IB_WC_LSO;
	case IB_WR_SEND_WITH_INV:		return IB_WC_SEND;
	case IB_WR_RDMA_READ_WITH_INV:		return IB_WC_RDMA_READ;
	case IB_WR_LOCAL_INV:			return IB_WC_LOCAL_INV;
	case IB_WR_REG_MR:			return IB_WC_REG_MR;

	default:
		return 0xff;
	}
}

void retransmit_timer(unsigned long data)
{
	struct rxe_qp *qp = (struct rxe_qp *)data;

	if (qp->valid) {
		qp->comp.timeout = 1;
		rxe_run_task(&qp->comp.task, 1);
	}
}

void rxe_comp_queue_pkt(struct rxe_dev *rxe, struct rxe_qp *qp,
			struct sk_buff *skb)
{
	int must_sched;

	skb_queue_tail(&qp->resp_pkts, skb);

	must_sched = skb_queue_len(&qp->resp_pkts) > 1;
	rxe_run_task(&qp->comp.task, must_sched);
}

static inline enum comp_state get_wqe(struct rxe_qp *qp,
				      struct rxe_pkt_info *pkt,
				      struct rxe_send_wqe **wqe_p)
{
	struct rxe_send_wqe *wqe;

	/* we come here whether or not we found a response packet to see if
	 * there are any posted WQEs
	 */
	wqe = queue_head(qp->sq.queue);
	*wqe_p = wqe;

	/* no WQE or requester has not started it yet */
	if (!wqe || wqe->state == wqe_state_posted)
		return pkt ? COMPST_DONE : COMPST_EXIT;

	/* WQE does not require an ack */
	if (wqe->state == wqe_state_done)
		return COMPST_COMP_WQE;

	/* WQE caused an error */
	if (wqe->state == wqe_state_error)
		return COMPST_ERROR;

	/* we have a WQE, if we also have an ack check its PSN */
	return pkt ? COMPST_CHECK_PSN : COMPST_EXIT;
}

static inline void reset_retry_counters(struct rxe_qp *qp)
{
	qp->comp.retry_cnt = qp->attr.retry_cnt;
	qp->comp.rnr_retry = qp->attr.rnr_retry;
}

static inline enum comp_state check_psn(struct rxe_qp *qp,
					struct rxe_pkt_info *pkt,
					struct rxe_send_wqe *wqe)
{
	s32 diff;

	/* check to see if response is past the oldest WQE. if it is, complete
	 * send/write or error read/atomic
	 */
	diff = psn_compare(pkt->psn, wqe->last_psn);
	if (diff > 0) {
		if (wqe->state == wqe_state_pending) {
			if (wqe->mask & WR_ATOMIC_OR_READ_MASK)
				return COMPST_ERROR_RETRY;

			reset_retry_counters(qp);
			return COMPST_COMP_WQE;
		} else {
			return COMPST_DONE;
		}
	}

	/* compare response packet to expected response */
	diff = psn_compare(pkt->psn, qp->comp.psn);
	if (diff < 0) {
		/* response is most likely a retried packet if it matches an
		 * uncompleted WQE go complete it else ignore it
		 */
		if (pkt->psn == wqe->last_psn)
			return COMPST_COMP_ACK;
		else
			return COMPST_DONE;
	} else if ((diff > 0) && (wqe->mask & WR_ATOMIC_OR_READ_MASK)) {
		return COMPST_DONE;
	} else {
		return COMPST_CHECK_ACK;
	}
}

static inline enum comp_state check_ack(struct rxe_qp *qp,
					struct rxe_pkt_info *pkt,
					struct rxe_send_wqe *wqe)
{
	unsigned int mask = pkt->mask;
	u8 syn;

	/* Check the sequence only */
	switch (qp->comp.opcode) {
	case -1:
		/* Will catch all *_ONLY cases. */
		if (!(mask & RXE_START_MASK))
			return COMPST_ERROR;

		break;

	case IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST:
	case IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE:
		if (pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE &&
		    pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_LAST) {
			return COMPST_ERROR;
		}
		break;
	default:
		WARN_ON(1);
	}

	/* Check operation validity. */
	switch (pkt->opcode) {
	case IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST:
	case IB_OPCODE_RC_RDMA_READ_RESPONSE_LAST:
	case IB_OPCODE_RC_RDMA_READ_RESPONSE_ONLY:
		syn = aeth_syn(pkt);

		if ((syn & AETH_TYPE_MASK) != AETH_ACK)
			return COMPST_ERROR;

		/* Fall through (IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE
		 * doesn't have an AETH)
		 */
	case IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE:
		if (wqe->wr.opcode != IB_WR_RDMA_READ &&
		    wqe->wr.opcode != IB_WR_RDMA_READ_WITH_INV) {
			return COMPST_ERROR;
		}
		reset_retry_counters(qp);
		return COMPST_READ;

	case IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE:
		syn = aeth_syn(pkt);

		if ((syn & AETH_TYPE_MASK) != AETH_ACK)
			return COMPST_ERROR;

		if (wqe->wr.opcode != IB_WR_ATOMIC_CMP_AND_SWP &&
		    wqe->wr.opcode != IB_WR_ATOMIC_FETCH_AND_ADD)
			return COMPST_ERROR;
		reset_retry_counters(qp);
		return COMPST_ATOMIC;

	case IB_OPCODE_RC_ACKNOWLEDGE:
		syn = aeth_syn(pkt);
		switch (syn & AETH_TYPE_MASK) {
		case AETH_ACK:
			reset_retry_counters(qp);
			return COMPST_WRITE_SEND;

		case AETH_RNR_NAK:
			return COMPST_RNR_RETRY;

		case AETH_NAK:
			switch (syn) {
			case AETH_NAK_PSN_SEQ_ERROR:
				/* a nak implicitly acks all packets with psns
				 * before
				 */
				if (psn_compare(pkt->psn, qp->comp.psn) > 0) {
					qp->comp.psn = pkt->psn;
					if (qp->req.wait_psn) {
						qp->req.wait_psn = 0;
						rxe_run_task(&qp->req.task, 1);
					}
				}
				return COMPST_ERROR_RETRY;

			case AETH_NAK_INVALID_REQ:
				wqe->status = IB_WC_REM_INV_REQ_ERR;
				return COMPST_ERROR;

			case AETH_NAK_REM_ACC_ERR:
				wqe->status = IB_WC_REM_ACCESS_ERR;
				return COMPST_ERROR;

			case AETH_NAK_REM_OP_ERR:
				wqe->status = IB_WC_REM_OP_ERR;
				return COMPST_ERROR;

			default:
				pr_warn("unexpected nak %x\n", syn);
				wqe->status = IB_WC_REM_OP_ERR;
				return COMPST_ERROR;
			}

		default:
			return COMPST_ERROR;
		}
		break;

	default:
		pr_warn("unexpected opcode\n");
	}

	return COMPST_ERROR;
}

static inline enum comp_state do_read(struct rxe_qp *qp,
				      struct rxe_pkt_info *pkt,
				      struct rxe_send_wqe *wqe)
{
	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
	int ret;

	ret = copy_data(rxe, qp->pd, IB_ACCESS_LOCAL_WRITE,
			&wqe->dma, payload_addr(pkt),
			payload_size(pkt), to_mem_obj, NULL);
	if (ret)
		return COMPST_ERROR;

	if (wqe->dma.resid == 0 && (pkt->mask & RXE_END_MASK))
		return COMPST_COMP_ACK;
	else
		return COMPST_UPDATE_COMP;
}

static inline enum comp_state do_atomic(struct rxe_qp *qp,
					struct rxe_pkt_info *pkt,
					struct rxe_send_wqe *wqe)
{
	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
	int ret;

	u64 atomic_orig = atmack_orig(pkt);

	ret = copy_data(rxe, qp->pd, IB_ACCESS_LOCAL_WRITE,
			&wqe->dma, &atomic_orig,
			sizeof(u64), to_mem_obj, NULL);
	if (ret)
		return COMPST_ERROR;
	else
		return COMPST_COMP_ACK;
}

static void make_send_cqe(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
			  struct rxe_cqe *cqe)
{
	memset(cqe, 0, sizeof(*cqe));

	if (!qp->is_user) {
		struct ib_wc		*wc	= &cqe->ibwc;

		wc->wr_id		= wqe->wr.wr_id;
		wc->status		= wqe->status;
		wc->opcode		= wr_to_wc_opcode(wqe->wr.opcode);
		if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
		    wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
			wc->wc_flags = IB_WC_WITH_IMM;
		wc->byte_len		= wqe->dma.length;
		wc->qp			= &qp->ibqp;
	} else {
		struct ib_uverbs_wc	*uwc	= &cqe->uibwc;

		uwc->wr_id		= wqe->wr.wr_id;
		uwc->status		= wqe->status;
		uwc->opcode		= wr_to_wc_opcode(wqe->wr.opcode);
		if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
		    wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
			uwc->wc_flags = IB_WC_WITH_IMM;
		uwc->byte_len		= wqe->dma.length;
		uwc->qp_num		= qp->ibqp.qp_num;
	}
}

static void do_complete(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
{
	struct rxe_cqe cqe;

	if ((qp->sq_sig_type == IB_SIGNAL_ALL_WR) ||
	    (wqe->wr.send_flags & IB_SEND_SIGNALED) ||
	    (qp->req.state == QP_STATE_ERROR)) {
		make_send_cqe(qp, wqe, &cqe);
		advance_consumer(qp->sq.queue);
		rxe_cq_post(qp->scq, &cqe, 0);
	} else {
		advance_consumer(qp->sq.queue);
	}

	/*
	 * we completed something so let req run again
	 * if it is trying to fence
	 */
	if (qp->req.wait_fence) {
		qp->req.wait_fence = 0;
		rxe_run_task(&qp->req.task, 1);
	}
}

static inline enum comp_state complete_ack(struct rxe_qp *qp,
					   struct rxe_pkt_info *pkt,
					   struct rxe_send_wqe *wqe)
{
	unsigned long flags;

	if (wqe->has_rd_atomic) {
		wqe->has_rd_atomic = 0;
		atomic_inc(&qp->req.rd_atomic);
		if (qp->req.need_rd_atomic) {
			qp->comp.timeout_retry = 0;
			qp->req.need_rd_atomic = 0;
			rxe_run_task(&qp->req.task, 1);
		}
	}

	if (unlikely(qp->req.state == QP_STATE_DRAIN)) {
		/* state_lock used by requester & completer */
		spin_lock_irqsave(&qp->state_lock, flags);
		if ((qp->req.state == QP_STATE_DRAIN) &&
		    (qp->comp.psn == qp->req.psn)) {
			qp->req.state = QP_STATE_DRAINED;
			spin_unlock_irqrestore(&qp->state_lock, flags);

			if (qp->ibqp.event_handler) {
				struct ib_event ev;

				ev.device = qp->ibqp.device;
				ev.element.qp = &qp->ibqp;
				ev.event = IB_EVENT_SQ_DRAINED;
				qp->ibqp.event_handler(&ev,
					qp->ibqp.qp_context);
			}
		} else {
			spin_unlock_irqrestore(&qp->state_lock, flags);
		}
	}

	do_complete(qp, wqe);

	if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
		return COMPST_UPDATE_COMP;
	else
		return COMPST_DONE;
}

static inline enum comp_state complete_wqe(struct rxe_qp *qp,
					   struct rxe_pkt_info *pkt,
					   struct rxe_send_wqe *wqe)
{
	qp->comp.opcode = -1;

	if (pkt) {
		if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
			qp->comp.psn = (pkt->psn + 1) & BTH_PSN_MASK;

		if (qp->req.wait_psn) {
			qp->req.wait_psn = 0;
			rxe_run_task(&qp->req.task, 1);
		}
	}

	do_complete(qp, wqe);

	return COMPST_GET_WQE;
}

int rxe_completer(void *arg)
{
	struct rxe_qp *qp = (struct rxe_qp *)arg;
	struct rxe_send_wqe *wqe = wqe;
	struct sk_buff *skb = NULL;
	struct rxe_pkt_info *pkt = NULL;
	enum comp_state state;

	rxe_add_ref(qp);

	if (!qp->valid) {
		while ((skb = skb_dequeue(&qp->resp_pkts))) {
			rxe_drop_ref(qp);
			kfree_skb(skb);
		}
		skb = NULL;
		pkt = NULL;

		while (queue_head(qp->sq.queue))
			advance_consumer(qp->sq.queue);

		goto exit;
	}

	if (qp->req.state == QP_STATE_ERROR) {
		while ((skb = skb_dequeue(&qp->resp_pkts))) {
			rxe_drop_ref(qp);
			kfree_skb(skb);
		}
		skb = NULL;
		pkt = NULL;

		while ((wqe = queue_head(qp->sq.queue))) {
			wqe->status = IB_WC_WR_FLUSH_ERR;
			do_complete(qp, wqe);
		}

		goto exit;
	}

	if (qp->req.state == QP_STATE_RESET) {
		while ((skb = skb_dequeue(&qp->resp_pkts))) {
			rxe_drop_ref(qp);
			kfree_skb(skb);
		}
		skb = NULL;
		pkt = NULL;

		while (queue_head(qp->sq.queue))
			advance_consumer(qp->sq.queue);

		goto exit;
	}

	if (qp->comp.timeout) {
		qp->comp.timeout_retry = 1;
		qp->comp.timeout = 0;
	} else {
		qp->comp.timeout_retry = 0;
	}

	if (qp->req.need_retry)
		goto exit;

	state = COMPST_GET_ACK;

	while (1) {
		pr_debug("qp#%d state = %s\n", qp_num(qp),
			 comp_state_name[state]);
		switch (state) {
		case COMPST_GET_ACK:
			skb = skb_dequeue(&qp->resp_pkts);
			if (skb) {
				pkt = SKB_TO_PKT(skb);
				qp->comp.timeout_retry = 0;
			}
			state = COMPST_GET_WQE;
			break;

		case COMPST_GET_WQE:
			state = get_wqe(qp, pkt, &wqe);
			break;

		case COMPST_CHECK_PSN:
			state = check_psn(qp, pkt, wqe);
			break;

		case COMPST_CHECK_ACK:
			state = check_ack(qp, pkt, wqe);
			break;

		case COMPST_READ:
			state = do_read(qp, pkt, wqe);
			break;

		case COMPST_ATOMIC:
			state = do_atomic(qp, pkt, wqe);
			break;

		case COMPST_WRITE_SEND:
			if (wqe->state == wqe_state_pending &&
			    wqe->last_psn == pkt->psn)
				state = COMPST_COMP_ACK;
			else
				state = COMPST_UPDATE_COMP;
			break;

		case COMPST_COMP_ACK:
			state = complete_ack(qp, pkt, wqe);
			break;

		case COMPST_COMP_WQE:
			state = complete_wqe(qp, pkt, wqe);
			break;

		case COMPST_UPDATE_COMP:
			if (pkt->mask & RXE_END_MASK)
				qp->comp.opcode = -1;
			else
				qp->comp.opcode = pkt->opcode;

			if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
				qp->comp.psn = (pkt->psn + 1) & BTH_PSN_MASK;

			if (qp->req.wait_psn) {
				qp->req.wait_psn = 0;
				rxe_run_task(&qp->req.task, 1);
			}

			state = COMPST_DONE;
			break;

		case COMPST_DONE:
			if (pkt) {
				rxe_drop_ref(pkt->qp);
				kfree_skb(skb);
			}
			goto done;

		case COMPST_EXIT:
			if (qp->comp.timeout_retry && wqe) {
				state = COMPST_ERROR_RETRY;
				break;
			}

			/* re reset the timeout counter if
			 * (1) QP is type RC
			 * (2) the QP is alive
			 * (3) there is a packet sent by the requester that
			 *     might be acked (we still might get spurious
			 *     timeouts but try to keep them as few as possible)
			 * (4) the timeout parameter is set
			 */
			if ((qp_type(qp) == IB_QPT_RC) &&
			    (qp->req.state == QP_STATE_READY) &&
			    (psn_compare(qp->req.psn, qp->comp.psn) > 0) &&
			    qp->qp_timeout_jiffies)
				mod_timer(&qp->retrans_timer,
					  jiffies + qp->qp_timeout_jiffies);
			goto exit;

		case COMPST_ERROR_RETRY:
			/* we come here if the retry timer fired and we did
			 * not receive a response packet. try to retry the send
			 * queue if that makes sense and the limits have not
			 * been exceeded. remember that some timeouts are
			 * spurious since we do not reset the timer but kick
			 * it down the road or let it expire
			 */

			/* there is nothing to retry in this case */
			if (!wqe || (wqe->state == wqe_state_posted))
				goto exit;

			if (qp->comp.retry_cnt > 0) {
				if (qp->comp.retry_cnt != 7)
					qp->comp.retry_cnt--;

				/* no point in retrying if we have already
				 * seen the last ack that the requester could
				 * have caused
				 */
				if (psn_compare(qp->req.psn,
						qp->comp.psn) > 0) {
					/* tell the requester to retry the
					 * send send queue next time around
					 */
					qp->req.need_retry = 1;
					rxe_run_task(&qp->req.task, 1);
				}

				if (pkt) {
					rxe_drop_ref(pkt->qp);
					kfree_skb(skb);
				}

				goto exit;

			} else {
				wqe->status = IB_WC_RETRY_EXC_ERR;
				state = COMPST_ERROR;
			}
			break;

		case COMPST_RNR_RETRY:
			if (qp->comp.rnr_retry > 0) {
				if (qp->comp.rnr_retry != 7)
					qp->comp.rnr_retry--;

				qp->req.need_retry = 1;
				pr_debug("qp#%d set rnr nak timer\n",
					 qp_num(qp));
				mod_timer(&qp->rnr_nak_timer,
					  jiffies + rnrnak_jiffies(aeth_syn(pkt)
						& ~AETH_TYPE_MASK));
				goto exit;
			} else {
				wqe->status = IB_WC_RNR_RETRY_EXC_ERR;
				state = COMPST_ERROR;
			}
			break;

		case COMPST_ERROR:
			do_complete(qp, wqe);
			rxe_qp_error(qp);

			if (pkt) {
				rxe_drop_ref(pkt->qp);
				kfree_skb(skb);
			}

			goto exit;
		}
	}

exit:
	/* we come here if we are done with processing and want the task to
	 * exit from the loop calling us
	 */
	rxe_drop_ref(qp);
	return -EAGAIN;

done:
	/* we come here if we have processed a packet we want the task to call
	 * us again to see if there is anything else to do
	 */
	rxe_drop_ref(qp);
	return 0;
}
Commit	Line	Data
8700e3e7 MS	1	/*
	2	* Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
	3	* Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
	4	*
	5	* This software is available to you under a choice of one of two
	6	* licenses. You may choose to be licensed under the terms of the GNU
	7	* General Public License (GPL) Version 2, available from the file
	8	* COPYING in the main directory of this source tree, or the
	9	* OpenIB.org BSD license below:
	10	*
	11	* Redistribution and use in source and binary forms, with or
	12	* without modification, are permitted provided that the following
	13	* conditions are met:
	14	*
	15	* - Redistributions of source code must retain the above
	16	* copyright notice, this list of conditions and the following
	17	* disclaimer.
	18	*
	19	* - Redistributions in binary form must reproduce the above
	20	* copyright notice, this list of conditions and the following
	21	* disclaimer in the documentation and/or other materials
	22	* provided with the distribution.
	23	*
	24	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	25	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	26	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	27	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	28	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	29	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	30	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	31	* SOFTWARE.
	32	*/
	33
	34	#include <linux/skbuff.h>
	35
	36	#include "rxe.h"
	37	#include "rxe_loc.h"
	38	#include "rxe_queue.h"
	39	#include "rxe_task.h"
	40
	41	enum comp_state {
	42	COMPST_GET_ACK,
	43	COMPST_GET_WQE,
	44	COMPST_COMP_WQE,
	45	COMPST_COMP_ACK,
	46	COMPST_CHECK_PSN,
	47	COMPST_CHECK_ACK,
	48	COMPST_READ,
	49	COMPST_ATOMIC,
	50	COMPST_WRITE_SEND,
	51	COMPST_UPDATE_COMP,
	52	COMPST_ERROR_RETRY,
	53	COMPST_RNR_RETRY,
	54	COMPST_ERROR,
	55	COMPST_EXIT, /* We have an issue, and we want to rerun the completer */
	56	COMPST_DONE, /* The completer finished successflly */
	57	};
	58
	59	static char *comp_state_name[] = {
	60	[COMPST_GET_ACK] = "GET ACK",
	61	[COMPST_GET_WQE] = "GET WQE",
	62	[COMPST_COMP_WQE] = "COMP WQE",
	63	[COMPST_COMP_ACK] = "COMP ACK",
	64	[COMPST_CHECK_PSN] = "CHECK PSN",
65	[COMPST_CHECK_ACK] = "CHECK ACK",
66	[COMPST_READ] = "READ",
67	[COMPST_ATOMIC] = "ATOMIC",
68	[COMPST_WRITE_SEND] = "WRITE/SEND",
69	[COMPST_UPDATE_COMP] = "UPDATE COMP",
70	[COMPST_ERROR_RETRY] = "ERROR RETRY",
71	[COMPST_RNR_RETRY] = "RNR RETRY",
72	[COMPST_ERROR] = "ERROR",
73	[COMPST_EXIT] = "EXIT",
74	[COMPST_DONE] = "DONE",
75	};
76
77	static unsigned long rnrnak_usec[32] = {
78	[IB_RNR_TIMER_655_36] = 655360,
79	[IB_RNR_TIMER_000_01] = 10,
80	[IB_RNR_TIMER_000_02] = 20,
81	[IB_RNR_TIMER_000_03] = 30,
82	[IB_RNR_TIMER_000_04] = 40,
83	[IB_RNR_TIMER_000_06] = 60,
84	[IB_RNR_TIMER_000_08] = 80,
85	[IB_RNR_TIMER_000_12] = 120,
86	[IB_RNR_TIMER_000_16] = 160,
87	[IB_RNR_TIMER_000_24] = 240,
88	[IB_RNR_TIMER_000_32] = 320,
89	[IB_RNR_TIMER_000_48] = 480,
90	[IB_RNR_TIMER_000_64] = 640,
91	[IB_RNR_TIMER_000_96] = 960,
92	[IB_RNR_TIMER_001_28] = 1280,
93	[IB_RNR_TIMER_001_92] = 1920,
94	[IB_RNR_TIMER_002_56] = 2560,
95	[IB_RNR_TIMER_003_84] = 3840,
96	[IB_RNR_TIMER_005_12] = 5120,
97	[IB_RNR_TIMER_007_68] = 7680,
98	[IB_RNR_TIMER_010_24] = 10240,
99	[IB_RNR_TIMER_015_36] = 15360,
100	[IB_RNR_TIMER_020_48] = 20480,
101	[IB_RNR_TIMER_030_72] = 30720,
102	[IB_RNR_TIMER_040_96] = 40960,
103	[IB_RNR_TIMER_061_44] = 61410,
104	[IB_RNR_TIMER_081_92] = 81920,
105	[IB_RNR_TIMER_122_88] = 122880,
106	[IB_RNR_TIMER_163_84] = 163840,
107	[IB_RNR_TIMER_245_76] = 245760,
108	[IB_RNR_TIMER_327_68] = 327680,
109	[IB_RNR_TIMER_491_52] = 491520,
110	};
111
112	static inline unsigned long rnrnak_jiffies(u8 timeout)
113	{
114	return max_t(unsigned long,
115	usecs_to_jiffies(rnrnak_usec[timeout]), 1);
116	}
117
118	static enum ib_wc_opcode wr_to_wc_opcode(enum ib_wr_opcode opcode)
119	{
120	switch (opcode) {
121	case IB_WR_RDMA_WRITE: return IB_WC_RDMA_WRITE;
122	case IB_WR_RDMA_WRITE_WITH_IMM: return IB_WC_RDMA_WRITE;
123	case IB_WR_SEND: return IB_WC_SEND;
124	case IB_WR_SEND_WITH_IMM: return IB_WC_SEND;
125	case IB_WR_RDMA_READ: return IB_WC_RDMA_READ;
126	case IB_WR_ATOMIC_CMP_AND_SWP: return IB_WC_COMP_SWAP;
127	case IB_WR_ATOMIC_FETCH_AND_ADD: return IB_WC_FETCH_ADD;
128	case IB_WR_LSO: return IB_WC_LSO;
129	case IB_WR_SEND_WITH_INV: return IB_WC_SEND;
130	case IB_WR_RDMA_READ_WITH_INV: return IB_WC_RDMA_READ;
131	case IB_WR_LOCAL_INV: return IB_WC_LOCAL_INV;
132	case IB_WR_REG_MR: return IB_WC_REG_MR;
133
134	default:
135	return 0xff;
136	}
137	}
138
139	void retransmit_timer(unsigned long data)
140	{
141	struct rxe_qp qp = (struct rxe_qp )data;
142
143	if (qp->valid) {
144	qp->comp.timeout = 1;
145	rxe_run_task(&qp->comp.task, 1);
146	}
147	}
148
149	void rxe_comp_queue_pkt(struct rxe_dev rxe, struct rxe_qp qp,
150	struct sk_buff *skb)
151	{
152	int must_sched;
153
154	skb_queue_tail(&qp->resp_pkts, skb);
155
156	must_sched = skb_queue_len(&qp->resp_pkts) > 1;
157	rxe_run_task(&qp->comp.task, must_sched);
158	}
159
160	static inline enum comp_state get_wqe(struct rxe_qp *qp,
161	struct rxe_pkt_info *pkt,
162	struct rxe_send_wqe **wqe_p)
163	{
164	struct rxe_send_wqe *wqe;
165
166	/* we come here whether or not we found a response packet to see if
167	* there are any posted WQEs
168	*/
169	wqe = queue_head(qp->sq.queue);
170	*wqe_p = wqe;
171
172	/* no WQE or requester has not started it yet */
173	if (!wqe \|\| wqe->state == wqe_state_posted)
174	return pkt ? COMPST_DONE : COMPST_EXIT;
175
176	/* WQE does not require an ack */
177	if (wqe->state == wqe_state_done)
178	return COMPST_COMP_WQE;
179
180	/* WQE caused an error */
181	if (wqe->state == wqe_state_error)
182	return COMPST_ERROR;
183
184	/* we have a WQE, if we also have an ack check its PSN */
185	return pkt ? COMPST_CHECK_PSN : COMPST_EXIT;
186	}
187
188	static inline void reset_retry_counters(struct rxe_qp *qp)
189	{
190	qp->comp.retry_cnt = qp->attr.retry_cnt;
191	qp->comp.rnr_retry = qp->attr.rnr_retry;
192	}
193
194	static inline enum comp_state check_psn(struct rxe_qp *qp,
195	struct rxe_pkt_info *pkt,
196	struct rxe_send_wqe *wqe)
197	{
198	s32 diff;
199
200	/* check to see if response is past the oldest WQE. if it is, complete
201	* send/write or error read/atomic
202	*/
203	diff = psn_compare(pkt->psn, wqe->last_psn);
204	if (diff > 0) {
205	if (wqe->state == wqe_state_pending) {
206	if (wqe->mask & WR_ATOMIC_OR_READ_MASK)
207	return COMPST_ERROR_RETRY;
208
209	reset_retry_counters(qp);
210	return COMPST_COMP_WQE;
211	} else {
212	return COMPST_DONE;
213	}
214	}
215
216	/* compare response packet to expected response */
217	diff = psn_compare(pkt->psn, qp->comp.psn);
218	if (diff < 0) {
219	/* response is most likely a retried packet if it matches an
220	* uncompleted WQE go complete it else ignore it
221	*/
222	if (pkt->psn == wqe->last_psn)
223	return COMPST_COMP_ACK;
224	else
225	return COMPST_DONE;
226	} else if ((diff > 0) && (wqe->mask & WR_ATOMIC_OR_READ_MASK)) {
cbf1f9a4	227	return COMPST_DONE;
8700e3e7 MS	228	} else {
	229	return COMPST_CHECK_ACK;
	230	}
	231	}
	232
	233	static inline enum comp_state check_ack(struct rxe_qp *qp,
	234	struct rxe_pkt_info *pkt,
	235	struct rxe_send_wqe *wqe)
	236	{
	237	unsigned int mask = pkt->mask;
	238	u8 syn;
	239
	240	/* Check the sequence only */
	241	switch (qp->comp.opcode) {
	242	case -1:
	243	/* Will catch all _ONLY cases. /
	244	if (!(mask & RXE_START_MASK))
	245	return COMPST_ERROR;
	246
	247	break;
	248
	249	case IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST:
	250	case IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE:
	251	if (pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE &&
	252	pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_LAST) {
	253	return COMPST_ERROR;
	254	}
	255	break;
	256	default:
	257	WARN_ON(1);
	258	}
	259
	260	/* Check operation validity. */
	261	switch (pkt->opcode) {
	262	case IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST:
	263	case IB_OPCODE_RC_RDMA_READ_RESPONSE_LAST:
	264	case IB_OPCODE_RC_RDMA_READ_RESPONSE_ONLY:
	265	syn = aeth_syn(pkt);
	266
	267	if ((syn & AETH_TYPE_MASK) != AETH_ACK)
	268	return COMPST_ERROR;
	269
	270	/* Fall through (IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE
	271	* doesn't have an AETH)
	272	*/
	273	case IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE:
	274	if (wqe->wr.opcode != IB_WR_RDMA_READ &&
	275	wqe->wr.opcode != IB_WR_RDMA_READ_WITH_INV) {
	276	return COMPST_ERROR;
	277	}
	278	reset_retry_counters(qp);
	279	return COMPST_READ;
	280
	281	case IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE:
	282	syn = aeth_syn(pkt);
	283
	284	if ((syn & AETH_TYPE_MASK) != AETH_ACK)
	285	return COMPST_ERROR;
	286
	287	if (wqe->wr.opcode != IB_WR_ATOMIC_CMP_AND_SWP &&
	288	wqe->wr.opcode != IB_WR_ATOMIC_FETCH_AND_ADD)
	289	return COMPST_ERROR;
	290	reset_retry_counters(qp);
	291	return COMPST_ATOMIC;
292
293	case IB_OPCODE_RC_ACKNOWLEDGE:
294	syn = aeth_syn(pkt);
295	switch (syn & AETH_TYPE_MASK) {
296	case AETH_ACK:
297	reset_retry_counters(qp);
298	return COMPST_WRITE_SEND;
299
300	case AETH_RNR_NAK:
301	return COMPST_RNR_RETRY;
302
303	case AETH_NAK:
304	switch (syn) {
305	case AETH_NAK_PSN_SEQ_ERROR:
306	/* a nak implicitly acks all packets with psns
307	* before
308	*/
309	if (psn_compare(pkt->psn, qp->comp.psn) > 0) {
310	qp->comp.psn = pkt->psn;
311	if (qp->req.wait_psn) {
312	qp->req.wait_psn = 0;
313	rxe_run_task(&qp->req.task, 1);
314	}
315	}
316	return COMPST_ERROR_RETRY;
317
318	case AETH_NAK_INVALID_REQ:
319	wqe->status = IB_WC_REM_INV_REQ_ERR;
320	return COMPST_ERROR;
321
322	case AETH_NAK_REM_ACC_ERR:
323	wqe->status = IB_WC_REM_ACCESS_ERR;
324	return COMPST_ERROR;
325
326	case AETH_NAK_REM_OP_ERR:
327	wqe->status = IB_WC_REM_OP_ERR;
328	return COMPST_ERROR;
329
330	default:
331	pr_warn("unexpected nak %x\n", syn);
332	wqe->status = IB_WC_REM_OP_ERR;
333	return COMPST_ERROR;
334	}
335
336	default:
337	return COMPST_ERROR;
338	}
339	break;
340
341	default:
342	pr_warn("unexpected opcode\n");
343	}
344
345	return COMPST_ERROR;
346	}
347
348	static inline enum comp_state do_read(struct rxe_qp *qp,
349	struct rxe_pkt_info *pkt,
350	struct rxe_send_wqe *wqe)
351	{
352	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
353	int ret;
354
355	ret = copy_data(rxe, qp->pd, IB_ACCESS_LOCAL_WRITE,
356	&wqe->dma, payload_addr(pkt),
357	payload_size(pkt), to_mem_obj, NULL);
358	if (ret)
359	return COMPST_ERROR;
360
361	if (wqe->dma.resid == 0 && (pkt->mask & RXE_END_MASK))
362	return COMPST_COMP_ACK;
363	else
364	return COMPST_UPDATE_COMP;
365	}
366
367	static inline enum comp_state do_atomic(struct rxe_qp *qp,
368	struct rxe_pkt_info *pkt,
369	struct rxe_send_wqe *wqe)
370	{
371	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
372	int ret;
373
374	u64 atomic_orig = atmack_orig(pkt);
375
376	ret = copy_data(rxe, qp->pd, IB_ACCESS_LOCAL_WRITE,
377	&wqe->dma, &atomic_orig,
378	sizeof(u64), to_mem_obj, NULL);
379	if (ret)
380	return COMPST_ERROR;
381	else
382	return COMPST_COMP_ACK;
383	}
384
385	static void make_send_cqe(struct rxe_qp qp, struct rxe_send_wqe wqe,
386	struct rxe_cqe *cqe)
387	{
388	memset(cqe, 0, sizeof(*cqe));
389
390	if (!qp->is_user) {
391	struct ib_wc *wc = &cqe->ibwc;
392
393	wc->wr_id = wqe->wr.wr_id;
394	wc->status = wqe->status;
395	wc->opcode = wr_to_wc_opcode(wqe->wr.opcode);
396	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM \|\|
397	wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
398	wc->wc_flags = IB_WC_WITH_IMM;
399	wc->byte_len = wqe->dma.length;
400	wc->qp = &qp->ibqp;
401	} else {
402	struct ib_uverbs_wc *uwc = &cqe->uibwc;
403
404	uwc->wr_id = wqe->wr.wr_id;
405	uwc->status = wqe->status;
406	uwc->opcode = wr_to_wc_opcode(wqe->wr.opcode);
407	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM \|\|
408	wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
409	uwc->wc_flags = IB_WC_WITH_IMM;
410	uwc->byte_len = wqe->dma.length;
411	uwc->qp_num = qp->ibqp.qp_num;
412	}
413	}
414
415	static void do_complete(struct rxe_qp qp, struct rxe_send_wqe wqe)
416	{
417	struct rxe_cqe cqe;
418
419	if ((qp->sq_sig_type == IB_SIGNAL_ALL_WR) \|\|
420	(wqe->wr.send_flags & IB_SEND_SIGNALED) \|\|
421	(qp->req.state == QP_STATE_ERROR)) {
422	make_send_cqe(qp, wqe, &cqe);
dd753d87	423	advance_consumer(qp->sq.queue);
8700e3e7	424	rxe_cq_post(qp->scq, &cqe, 0);
dd753d87 AB	425	} else {
dd753d87 AB	426	advance_consumer(qp->sq.queue);
8700e3e7 MS	427	}
8700e3e7 MS	428
8700e3e7 MS	429	/*
	430	* we completed something so let req run again
	431	* if it is trying to fence
	432	*/
	433	if (qp->req.wait_fence) {
	434	qp->req.wait_fence = 0;
	435	rxe_run_task(&qp->req.task, 1);
	436	}
	437	}
	438
	439	static inline enum comp_state complete_ack(struct rxe_qp *qp,
	440	struct rxe_pkt_info *pkt,
	441	struct rxe_send_wqe *wqe)
	442	{
	443	unsigned long flags;
	444
	445	if (wqe->has_rd_atomic) {
	446	wqe->has_rd_atomic = 0;
	447	atomic_inc(&qp->req.rd_atomic);
	448	if (qp->req.need_rd_atomic) {
	449	qp->comp.timeout_retry = 0;
	450	qp->req.need_rd_atomic = 0;
	451	rxe_run_task(&qp->req.task, 1);
	452	}
	453	}
	454
	455	if (unlikely(qp->req.state == QP_STATE_DRAIN)) {
	456	/* state_lock used by requester & completer */
	457	spin_lock_irqsave(&qp->state_lock, flags);
	458	if ((qp->req.state == QP_STATE_DRAIN) &&
	459	(qp->comp.psn == qp->req.psn)) {
	460	qp->req.state = QP_STATE_DRAINED;
	461	spin_unlock_irqrestore(&qp->state_lock, flags);
	462
	463	if (qp->ibqp.event_handler) {
	464	struct ib_event ev;
	465
	466	ev.device = qp->ibqp.device;
	467	ev.element.qp = &qp->ibqp;
	468	ev.event = IB_EVENT_SQ_DRAINED;
	469	qp->ibqp.event_handler(&ev,
	470	qp->ibqp.qp_context);
	471	}
	472	} else {
	473	spin_unlock_irqrestore(&qp->state_lock, flags);
	474	}
	475	}
	476
	477	do_complete(qp, wqe);
	478
	479	if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
	480	return COMPST_UPDATE_COMP;
	481	else
	482	return COMPST_DONE;
	483	}
	484
	485	static inline enum comp_state complete_wqe(struct rxe_qp *qp,
	486	struct rxe_pkt_info *pkt,
	487	struct rxe_send_wqe *wqe)
	488	{
	489	qp->comp.opcode = -1;
	490
	491	if (pkt) {
	492	if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
493	qp->comp.psn = (pkt->psn + 1) & BTH_PSN_MASK;
494
495	if (qp->req.wait_psn) {
496	qp->req.wait_psn = 0;
497	rxe_run_task(&qp->req.task, 1);
498	}
499	}
500
501	do_complete(qp, wqe);
502
503	return COMPST_GET_WQE;
504	}
505
506	int rxe_completer(void *arg)
507	{
508	struct rxe_qp qp = (struct rxe_qp )arg;
509	struct rxe_send_wqe *wqe = wqe;
510	struct sk_buff *skb = NULL;
511	struct rxe_pkt_info *pkt = NULL;
512	enum comp_state state;
513
37f69f43 AB	514	rxe_add_ref(qp);
37f69f43 AB	515
8700e3e7 MS	516	if (!qp->valid) {
	517	while ((skb = skb_dequeue(&qp->resp_pkts))) {
	518	rxe_drop_ref(qp);
	519	kfree_skb(skb);
	520	}
	521	skb = NULL;
	522	pkt = NULL;
	523
	524	while (queue_head(qp->sq.queue))
	525	advance_consumer(qp->sq.queue);
	526
	527	goto exit;
	528	}
	529
	530	if (qp->req.state == QP_STATE_ERROR) {
	531	while ((skb = skb_dequeue(&qp->resp_pkts))) {
	532	rxe_drop_ref(qp);
	533	kfree_skb(skb);
	534	}
	535	skb = NULL;
	536	pkt = NULL;
	537
	538	while ((wqe = queue_head(qp->sq.queue))) {
	539	wqe->status = IB_WC_WR_FLUSH_ERR;
	540	do_complete(qp, wqe);
	541	}
	542
	543	goto exit;
	544	}
	545
	546	if (qp->req.state == QP_STATE_RESET) {
	547	while ((skb = skb_dequeue(&qp->resp_pkts))) {
	548	rxe_drop_ref(qp);
	549	kfree_skb(skb);
	550	}
	551	skb = NULL;
	552	pkt = NULL;
	553
	554	while (queue_head(qp->sq.queue))
	555	advance_consumer(qp->sq.queue);
	556
	557	goto exit;
	558	}
	559
	560	if (qp->comp.timeout) {
	561	qp->comp.timeout_retry = 1;
	562	qp->comp.timeout = 0;
	563	} else {
	564	qp->comp.timeout_retry = 0;
	565	}
	566
	567	if (qp->req.need_retry)
	568	goto exit;
	569
	570	state = COMPST_GET_ACK;
	571
	572	while (1) {
e404f945 PP	573	pr_debug("qp#%d state = %s\n", qp_num(qp),
e404f945 PP	574	comp_state_name[state]);
8700e3e7 MS	575	switch (state) {
	576	case COMPST_GET_ACK:
	577	skb = skb_dequeue(&qp->resp_pkts);
	578	if (skb) {
	579	pkt = SKB_TO_PKT(skb);
	580	qp->comp.timeout_retry = 0;
	581	}
	582	state = COMPST_GET_WQE;
	583	break;
	584
	585	case COMPST_GET_WQE:
	586	state = get_wqe(qp, pkt, &wqe);
	587	break;
	588
	589	case COMPST_CHECK_PSN:
	590	state = check_psn(qp, pkt, wqe);
	591	break;
	592
	593	case COMPST_CHECK_ACK:
	594	state = check_ack(qp, pkt, wqe);
	595	break;
	596
	597	case COMPST_READ:
	598	state = do_read(qp, pkt, wqe);
	599	break;
	600
	601	case COMPST_ATOMIC:
	602	state = do_atomic(qp, pkt, wqe);
	603	break;
	604
	605	case COMPST_WRITE_SEND:
	606	if (wqe->state == wqe_state_pending &&
	607	wqe->last_psn == pkt->psn)
	608	state = COMPST_COMP_ACK;
	609	else
	610	state = COMPST_UPDATE_COMP;
	611	break;
	612
	613	case COMPST_COMP_ACK:
	614	state = complete_ack(qp, pkt, wqe);
	615	break;
	616
	617	case COMPST_COMP_WQE:
	618	state = complete_wqe(qp, pkt, wqe);
	619	break;
	620
	621	case COMPST_UPDATE_COMP:
	622	if (pkt->mask & RXE_END_MASK)
	623	qp->comp.opcode = -1;
	624	else
	625	qp->comp.opcode = pkt->opcode;
	626
	627	if (psn_compare(pkt->psn, qp->comp.psn) >= 0)
	628	qp->comp.psn = (pkt->psn + 1) & BTH_PSN_MASK;
	629
	630	if (qp->req.wait_psn) {
	631	qp->req.wait_psn = 0;
	632	rxe_run_task(&qp->req.task, 1);
	633	}
	634
	635	state = COMPST_DONE;
	636	break;
	637
	638	case COMPST_DONE:
639	if (pkt) {
640	rxe_drop_ref(pkt->qp);
641	kfree_skb(skb);
642	}
643	goto done;
644
645	case COMPST_EXIT:
646	if (qp->comp.timeout_retry && wqe) {
647	state = COMPST_ERROR_RETRY;
648	break;
649	}
650
651	/* re reset the timeout counter if
652	* (1) QP is type RC
653	* (2) the QP is alive
654	* (3) there is a packet sent by the requester that
655	* might be acked (we still might get spurious
656	* timeouts but try to keep them as few as possible)
657	* (4) the timeout parameter is set
658	*/
659	if ((qp_type(qp) == IB_QPT_RC) &&
660	(qp->req.state == QP_STATE_READY) &&
661	(psn_compare(qp->req.psn, qp->comp.psn) > 0) &&
662	qp->qp_timeout_jiffies)
663	mod_timer(&qp->retrans_timer,
664	jiffies + qp->qp_timeout_jiffies);
665	goto exit;
666
667	case COMPST_ERROR_RETRY:
668	/* we come here if the retry timer fired and we did
669	* not receive a response packet. try to retry the send
670	* queue if that makes sense and the limits have not
671	* been exceeded. remember that some timeouts are
672	* spurious since we do not reset the timer but kick
673	* it down the road or let it expire
674	*/
675
676	/* there is nothing to retry in this case */
677	if (!wqe \|\| (wqe->state == wqe_state_posted))
678	goto exit;
679
680	if (qp->comp.retry_cnt > 0) {
681	if (qp->comp.retry_cnt != 7)
682	qp->comp.retry_cnt--;
683
684	/* no point in retrying if we have already
685	* seen the last ack that the requester could
686	* have caused
687	*/
688	if (psn_compare(qp->req.psn,
689	qp->comp.psn) > 0) {
690	/* tell the requester to retry the
691	* send send queue next time around
692	*/
693	qp->req.need_retry = 1;
694	rxe_run_task(&qp->req.task, 1);
695	}
c1cc72cb YC	696
	697	if (pkt) {
	698	rxe_drop_ref(pkt->qp);
	699	kfree_skb(skb);
	700	}
	701
8700e3e7	702	goto exit;
c1cc72cb	703
8700e3e7 MS	704	} else {
	705	wqe->status = IB_WC_RETRY_EXC_ERR;
	706	state = COMPST_ERROR;
	707	}
	708	break;
	709
	710	case COMPST_RNR_RETRY:
	711	if (qp->comp.rnr_retry > 0) {
	712	if (qp->comp.rnr_retry != 7)
	713	qp->comp.rnr_retry--;
	714
	715	qp->req.need_retry = 1;
e404f945 PP	716	pr_debug("qp#%d set rnr nak timer\n",
e404f945 PP	717	qp_num(qp));
8700e3e7 MS	718	mod_timer(&qp->rnr_nak_timer,
	719	jiffies + rnrnak_jiffies(aeth_syn(pkt)
	720	& ~AETH_TYPE_MASK));
	721	goto exit;
	722	} else {
	723	wqe->status = IB_WC_RNR_RETRY_EXC_ERR;
	724	state = COMPST_ERROR;
	725	}
	726	break;
	727
	728	case COMPST_ERROR:
	729	do_complete(qp, wqe);
	730	rxe_qp_error(qp);
c1cc72cb YC	731
	732	if (pkt) {
	733	rxe_drop_ref(pkt->qp);
	734	kfree_skb(skb);
	735	}
	736
8700e3e7 MS	737	goto exit;
	738	}
	739	}
	740
	741	exit:
	742	/* we come here if we are done with processing and want the task to
	743	* exit from the loop calling us
	744	*/
37f69f43	745	rxe_drop_ref(qp);
8700e3e7 MS	746	return -EAGAIN;
	747
	748	done:
	749	/* we come here if we have processed a packet we want the task to call
	750	* us again to see if there is anything else to do
	751	*/
37f69f43	752	rxe_drop_ref(qp);
8700e3e7 MS	753	return 0;
8700e3e7 MS	754	}