mthca_cq.c

h内核

/*
 * Copyright (c) 2004, 2005 Topspin Communications.  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.
 *
 * $Id: mthca_cq.c 1369 2004-12-20 16:17:07Z roland $
 */

#include <linux/init.h>

#include <ib_pack.h>

#include "mthca_dev.h"
#include "mthca_cmd.h"

enum {
	MTHCA_MAX_DIRECT_CQ_SIZE = 4 * PAGE_SIZE
};

enum {
	MTHCA_CQ_ENTRY_SIZE = 0x20
};

/*
 * Must be packed because start is 64 bits but only aligned to 32 bits.
 */
struct mthca_cq_context {
	u32 flags;
	u64 start;
	u32 logsize_usrpage;
	u32 error_eqn;
	u32 comp_eqn;
	u32 pd;
	u32 lkey;
	u32 last_notified_index;
	u32 solicit_producer_index;
	u32 consumer_index;
	u32 producer_index;
	u32 cqn;
	u32 reserved[3];
} __attribute__((packed));

#define MTHCA_CQ_STATUS_OK          ( 0 << 28)
#define MTHCA_CQ_STATUS_OVERFLOW    ( 9 << 28)
#define MTHCA_CQ_STATUS_WRITE_FAIL  (10 << 28)
#define MTHCA_CQ_FLAG_TR            ( 1 << 18)
#define MTHCA_CQ_FLAG_OI            ( 1 << 17)
#define MTHCA_CQ_STATE_DISARMED     ( 0 <<  8)
#define MTHCA_CQ_STATE_ARMED        ( 1 <<  8)
#define MTHCA_CQ_STATE_ARMED_SOL    ( 4 <<  8)
#define MTHCA_EQ_STATE_FIRED        (10 <<  8)

enum {
	MTHCA_ERROR_CQE_OPCODE_MASK = 0xfe
};

enum {
	SYNDROME_LOCAL_LENGTH_ERR 	 = 0x01,
	SYNDROME_LOCAL_QP_OP_ERR  	 = 0x02,
	SYNDROME_LOCAL_EEC_OP_ERR 	 = 0x03,
	SYNDROME_LOCAL_PROT_ERR   	 = 0x04,
	SYNDROME_WR_FLUSH_ERR     	 = 0x05,
	SYNDROME_MW_BIND_ERR      	 = 0x06,
	SYNDROME_BAD_RESP_ERR     	 = 0x10,
	SYNDROME_LOCAL_ACCESS_ERR 	 = 0x11,
	SYNDROME_REMOTE_INVAL_REQ_ERR 	 = 0x12,
	SYNDROME_REMOTE_ACCESS_ERR 	 = 0x13,
	SYNDROME_REMOTE_OP_ERR     	 = 0x14,
	SYNDROME_RETRY_EXC_ERR 		 = 0x15,
	SYNDROME_RNR_RETRY_EXC_ERR 	 = 0x16,
	SYNDROME_LOCAL_RDD_VIOL_ERR 	 = 0x20,
	SYNDROME_REMOTE_INVAL_RD_REQ_ERR = 0x21,
	SYNDROME_REMOTE_ABORTED_ERR 	 = 0x22,
	SYNDROME_INVAL_EECN_ERR 	 = 0x23,
	SYNDROME_INVAL_EEC_STATE_ERR 	 = 0x24
};

struct mthca_cqe {
	u32 my_qpn;
	u32 my_ee;
	u32 rqpn;
	u16 sl_g_mlpath;
	u16 rlid;
	u32 imm_etype_pkey_eec;
	u32 byte_cnt;
	u32 wqe;
	u8  opcode;
	u8  is_send;
	u8  reserved;
	u8  owner;
};

struct mthca_err_cqe {
	u32 my_qpn;
	u32 reserved1[3];
	u8  syndrome;
	u8  reserved2;
	u16 db_cnt;
	u32 reserved3;
	u32 wqe;
	u8  opcode;
	u8  reserved4[2];
	u8  owner;
};

#define MTHCA_CQ_ENTRY_OWNER_SW      (0 << 7)
#define MTHCA_CQ_ENTRY_OWNER_HW      (1 << 7)

#define MTHCA_CQ_DB_INC_CI       (1 << 24)
#define MTHCA_CQ_DB_REQ_NOT      (2 << 24)
#define MTHCA_CQ_DB_REQ_NOT_SOL  (3 << 24)
#define MTHCA_CQ_DB_SET_CI       (4 << 24)
#define MTHCA_CQ_DB_REQ_NOT_MULT (5 << 24)

static inline struct mthca_cqe *get_cqe(struct mthca_cq *cq, int entry)
{
	if (cq->is_direct)
		return cq->queue.direct.buf + (entry * MTHCA_CQ_ENTRY_SIZE);
	else
		return cq->queue.page_list[entry * MTHCA_CQ_ENTRY_SIZE / PAGE_SIZE].buf
			+ (entry * MTHCA_CQ_ENTRY_SIZE) % PAGE_SIZE;
}

static inline int cqe_sw(struct mthca_cq *cq, int i)
{
	return !(MTHCA_CQ_ENTRY_OWNER_HW &
		 get_cqe(cq, i)->owner);
}

static inline int next_cqe_sw(struct mthca_cq *cq)
{
	return cqe_sw(cq, cq->cons_index);
}

static inline void set_cqe_hw(struct mthca_cq *cq, int entry)
{
	get_cqe(cq, entry)->owner = MTHCA_CQ_ENTRY_OWNER_HW;
}

static inline void inc_cons_index(struct mthca_dev *dev, struct mthca_cq *cq,
				  int nent)
{
	u32 doorbell[2];

	doorbell[0] = cpu_to_be32(MTHCA_CQ_DB_INC_CI | cq->cqn);
	doorbell[1] = cpu_to_be32(nent - 1);

	mthca_write64(doorbell,
		      dev->kar + MTHCA_CQ_DOORBELL,
		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
}

void mthca_cq_event(struct mthca_dev *dev, u32 cqn)
{
	struct mthca_cq *cq;

	spin_lock(&dev->cq_table.lock);
	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
	if (cq)
		atomic_inc(&cq->refcount);
	spin_unlock(&dev->cq_table.lock);

	if (!cq) {
		mthca_warn(dev, "Completion event for bogus CQ %08x\n", cqn);
		return;
	}

	cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);

	if (atomic_dec_and_test(&cq->refcount))
		wake_up(&cq->wait);
}

void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn)
{
	struct mthca_cq *cq;
	struct mthca_cqe *cqe;
	int prod_index;
	int nfreed = 0;

	spin_lock_irq(&dev->cq_table.lock);
	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
	if (cq)
		atomic_inc(&cq->refcount);
	spin_unlock_irq(&dev->cq_table.lock);

	if (!cq)
		return;

	spin_lock_irq(&cq->lock);

	/*
	 * First we need to find the current producer index, so we
	 * know where to start cleaning from.  It doesn't matter if HW
	 * adds new entries after this loop -- the QP we're worried
	 * about is already in RESET, so the new entries won't come
	 * from our QP and therefore don't need to be checked.
	 */
	for (prod_index = cq->cons_index;
	     cqe_sw(cq, prod_index & cq->ibcq.cqe);
	     ++prod_index)
		if (prod_index == cq->cons_index + cq->ibcq.cqe)
			break;

	if (0)
		mthca_dbg(dev, "Cleaning QPN %06x from CQN %06x; ci %d, pi %d\n",
			  qpn, cqn, cq->cons_index, prod_index);

	/*
	 * Now sweep backwards through the CQ, removing CQ entries
	 * that match our QP by copying older entries on top of them.
	 */
	while (prod_index > cq->cons_index) {
		cqe = get_cqe(cq, (prod_index - 1) & cq->ibcq.cqe);
		if (cqe->my_qpn == cpu_to_be32(qpn))
			++nfreed;
		else if (nfreed)
			memcpy(get_cqe(cq, (prod_index - 1 + nfreed) &
				       cq->ibcq.cqe),
			       cqe,
			       MTHCA_CQ_ENTRY_SIZE);
		--prod_index;
	}

	if (nfreed) {
		wmb();
		inc_cons_index(dev, cq, nfreed);
		cq->cons_index = (cq->cons_index + nfreed) & cq->ibcq.cqe;
	}

	spin_unlock_irq(&cq->lock);
	if (atomic_dec_and_test(&cq->refcount))
		wake_up(&cq->wait);
}

static int handle_error_cqe(struct mthca_dev *dev, struct mthca_cq *cq,
			    struct mthca_qp *qp, int wqe_index, int is_send,
			    struct mthca_err_cqe *cqe,
			    struct ib_wc *entry, int *free_cqe)
{
	int err;
	int dbd;
	u32 new_wqe;

	if (1 && cqe->syndrome != SYNDROME_WR_FLUSH_ERR) {
		int j;

		mthca_dbg(dev, "%x/%d: error CQE -> QPN %06x, WQE @ %08x\n",
			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
			  be32_to_cpu(cqe->wqe));

		for (j = 0; j < 8; ++j)
			printk(KERN_DEBUG "  [%2x] %08x\n",
			       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
	}

	/*
	 * For completions in error, only work request ID, status (and
	 * freed resource count for RD) have to be set.
	 */
	switch (cqe->syndrome) {
	case SYNDROME_LOCAL_LENGTH_ERR:
		entry->status = IB_WC_LOC_LEN_ERR;
		break;
	case SYNDROME_LOCAL_QP_OP_ERR:
		entry->status = IB_WC_LOC_QP_OP_ERR;
		break;
	case SYNDROME_LOCAL_EEC_OP_ERR:
		entry->status = IB_WC_LOC_EEC_OP_ERR;
		break;
	case SYNDROME_LOCAL_PROT_ERR:
		entry->status = IB_WC_LOC_PROT_ERR;
		break;
	case SYNDROME_WR_FLUSH_ERR:
		entry->status = IB_WC_WR_FLUSH_ERR;
		break;
	case SYNDROME_MW_BIND_ERR:
		entry->status = IB_WC_MW_BIND_ERR;
		break;
	case SYNDROME_BAD_RESP_ERR:
		entry->status = IB_WC_BAD_RESP_ERR;
		break;
	case SYNDROME_LOCAL_ACCESS_ERR:
		entry->status = IB_WC_LOC_ACCESS_ERR;
		break;
	case SYNDROME_REMOTE_INVAL_REQ_ERR:
		entry->status = IB_WC_REM_INV_REQ_ERR;
		break;
	case SYNDROME_REMOTE_ACCESS_ERR:
		entry->status = IB_WC_REM_ACCESS_ERR;
		break;
	case SYNDROME_REMOTE_OP_ERR:
		entry->status = IB_WC_REM_OP_ERR;
		break;
	case SYNDROME_RETRY_EXC_ERR:
		entry->status = IB_WC_RETRY_EXC_ERR;
		break;
	case SYNDROME_RNR_RETRY_EXC_ERR:
		entry->status = IB_WC_RNR_RETRY_EXC_ERR;
		break;
	case SYNDROME_LOCAL_RDD_VIOL_ERR:
		entry->status = IB_WC_LOC_RDD_VIOL_ERR;
		break;
	case SYNDROME_REMOTE_INVAL_RD_REQ_ERR:
		entry->status = IB_WC_REM_INV_RD_REQ_ERR;
		break;
	case SYNDROME_REMOTE_ABORTED_ERR:
		entry->status = IB_WC_REM_ABORT_ERR;
		break;
	case SYNDROME_INVAL_EECN_ERR:
		entry->status = IB_WC_INV_EECN_ERR;
		break;
	case SYNDROME_INVAL_EEC_STATE_ERR:
		entry->status = IB_WC_INV_EEC_STATE_ERR;
		break;
	default:
		entry->status = IB_WC_GENERAL_ERR;
		break;
	}

	err = mthca_free_err_wqe(qp, is_send, wqe_index, &dbd, &new_wqe);
	if (err)
		return err;

	/*
	 * If we're at the end of the WQE chain, or we've used up our
	 * doorbell count, free the CQE.  Otherwise just update it for
	 * the next poll operation.
	 */
	if (!(new_wqe & cpu_to_be32(0x3f)) || (!cqe->db_cnt && dbd))
		return 0;

	cqe->db_cnt   = cpu_to_be16(be16_to_cpu(cqe->db_cnt) - dbd);
	cqe->wqe      = new_wqe;
	cqe->syndrome = SYNDROME_WR_FLUSH_ERR;

	*free_cqe = 0;

	return 0;
}

static void dump_cqe(struct mthca_cqe *cqe)
{
	int j;

	for (j = 0; j < 8; ++j)
		printk(KERN_DEBUG "  [%2x] %08x\n",
		       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
}

static inline int mthca_poll_one(struct mthca_dev *dev,
				 struct mthca_cq *cq,
				 struct mthca_qp **cur_qp,
				 int *freed,
				 struct ib_wc *entry)
{
	struct mthca_wq *wq;
	struct mthca_cqe *cqe;
	int wqe_index;
	int is_error = 0;
	int is_send;
	int free_cqe = 1;
	int err = 0;

	if (!next_cqe_sw(cq))
		return -EAGAIN;

	/*
	 * Make sure we read CQ entry contents after we've checked the
	 * ownership bit.
	 */
	rmb();

	cqe = get_cqe(cq, cq->cons_index);

	if (0) {
		mthca_dbg(dev, "%x/%d: CQE -> QPN %06x, WQE @ %08x\n",
			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
			  be32_to_cpu(cqe->wqe));

		dump_cqe(cqe);
	}

	if ((cqe->opcode & MTHCA_ERROR_CQE_OPCODE_MASK) ==
	    MTHCA_ERROR_CQE_OPCODE_MASK) {
		is_error = 1;
		is_send = cqe->opcode & 1;
	} else
		is_send = cqe->is_send & 0x80;

	if (!*cur_qp || be32_to_cpu(cqe->my_qpn) != (*cur_qp)->qpn) {
		if (*cur_qp) {
			if (*freed) {
				wmb();
				inc_cons_index(dev, cq, *freed);