xref: /freebsd/sys/dev/ntb/ntb_transport.c (revision fdafd315)

/*-
 * Copyright (c) 2016-2017 Alexander Motin <mav@FreeBSD.org>
 * Copyright (C) 2013 Intel Corporation
 * Copyright (C) 2015 EMC Corporation
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. 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.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * The Non-Transparent Bridge (NTB) is a device that allows you to connect
 * two or more systems using a PCI-e links, providing remote memory access.
 *
 * This module contains a transport for sending and receiving messages by
 * writing to remote memory window(s) provided by underlying NTB device.
 *
 * NOTE: Much of the code in this module is shared with Linux. Any patches may
 * be picked up and redistributed in Linux with a dual GPL/BSD license.
 */

#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/ktr.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/queue.h>
#include <sys/sbuf.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>

#include <vm/vm.h>
#include <vm/pmap.h>

#include <machine/bus.h>

#include "ntb.h"
#include "ntb_transport.h"

#define KTR_NTB KTR_SPARE3

#define NTB_TRANSPORT_VERSION	4

static SYSCTL_NODE(_hw, OID_AUTO, ntb_transport,
    CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
    "ntb_transport");

static unsigned g_ntb_transport_debug_level;
SYSCTL_UINT(_hw_ntb_transport, OID_AUTO, debug_level, CTLFLAG_RWTUN,
    &g_ntb_transport_debug_level, 0,
    "ntb_transport log level -- higher is more verbose");
#define ntb_printf(lvl, ...) do {			\
	if ((lvl) <= g_ntb_transport_debug_level) {	\
		printf(__VA_ARGS__);			\
	}						\
} while (0)

static unsigned transport_mtu = 0x10000;

static uint64_t max_mw_size = 256*1024*1024;
SYSCTL_UQUAD(_hw_ntb_transport, OID_AUTO, max_mw_size, CTLFLAG_RDTUN, &max_mw_size, 0,
    "If enabled (non-zero), limit the size of large memory windows. "
    "Both sides of the NTB MUST set the same value here.");

static unsigned enable_xeon_watchdog;
SYSCTL_UINT(_hw_ntb_transport, OID_AUTO, enable_xeon_watchdog, CTLFLAG_RDTUN,
    &enable_xeon_watchdog, 0, "If non-zero, write a register every second to "
    "keep a watchdog from tearing down the NTB link");

STAILQ_HEAD(ntb_queue_list, ntb_queue_entry);

typedef uint32_t ntb_q_idx_t;

struct ntb_queue_entry {
	/* ntb_queue list reference */
	STAILQ_ENTRY(ntb_queue_entry) entry;

	/* info on data to be transferred */
	void		*cb_data;
	void		*buf;
	uint32_t	len;
	uint32_t	flags;

	struct ntb_transport_qp		*qp;
	struct ntb_payload_header	*x_hdr;
	ntb_q_idx_t	index;
};

struct ntb_rx_info {
	ntb_q_idx_t	entry;
};

struct ntb_transport_qp {
	struct ntb_transport_ctx	*transport;
	device_t		 dev;

	void			*cb_data;

	bool			client_ready;
	volatile bool		link_is_up;
	uint8_t			qp_num;	/* Only 64 QPs are allowed.  0-63 */

	struct ntb_rx_info	*rx_info;
	struct ntb_rx_info	*remote_rx_info;

	void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data,
	    void *data, int len);
	struct ntb_queue_list	tx_free_q;
	struct mtx		ntb_tx_free_q_lock;
	caddr_t			tx_mw;
	bus_addr_t		tx_mw_phys;
	ntb_q_idx_t		tx_index;
	ntb_q_idx_t		tx_max_entry;
	uint64_t		tx_max_frame;

	void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data,
	    void *data, int len);
	struct ntb_queue_list	rx_post_q;
	struct ntb_queue_list	rx_pend_q;
	/* ntb_rx_q_lock: synchronize access to rx_XXXX_q */
	struct mtx		ntb_rx_q_lock;
	struct task		rxc_db_work;
	struct taskqueue	*rxc_tq;
	caddr_t			rx_buff;
	ntb_q_idx_t		rx_index;
	ntb_q_idx_t		rx_max_entry;
	uint64_t		rx_max_frame;

	void (*event_handler)(void *data, enum ntb_link_event status);
	struct callout		link_work;
	struct callout		rx_full;

	uint64_t		last_rx_no_buf;

	/* Stats */
	uint64_t		rx_bytes;
	uint64_t		rx_pkts;
	uint64_t		rx_ring_empty;
	uint64_t		rx_err_no_buf;
	uint64_t		rx_err_oflow;
	uint64_t		rx_err_ver;
	uint64_t		tx_bytes;
	uint64_t		tx_pkts;
	uint64_t		tx_ring_full;
	uint64_t		tx_err_no_buf;

	struct mtx		tx_lock;
};

struct ntb_transport_mw {
	vm_paddr_t	phys_addr;
	size_t		phys_size;
	size_t		xlat_align;
	size_t		xlat_align_size;
	bus_addr_t	addr_limit;
	/* Tx buff is vbase / phys_addr / tx_size */
	caddr_t		vbase;
	size_t		tx_size;
	/* Rx buff is virt_addr / dma_addr / rx_size */
	bus_dma_tag_t	dma_tag;
	bus_dmamap_t	dma_map;
	caddr_t		virt_addr;
	bus_addr_t	dma_addr;
	size_t		rx_size;
	/* rx_size increased to size alignment requirements of the hardware. */
	size_t		buff_size;
};

struct ntb_transport_child {
	device_t	dev;
	int		consumer;
	int		qpoff;
	int		qpcnt;
	struct ntb_transport_child *next;
};

struct ntb_transport_ctx {
	device_t		 dev;
	struct ntb_transport_child *child;
	struct ntb_transport_mw	*mw_vec;
	struct ntb_transport_qp	*qp_vec;
	int			compact;
	unsigned		mw_count;
	unsigned		qp_count;
	uint64_t		qp_bitmap;
	volatile bool		link_is_up;
	enum ntb_speed		link_speed;
	enum ntb_width		link_width;
	struct callout		link_work;
	struct callout		link_watchdog;
	struct task		link_cleanup;
};

enum {
	NTBT_DESC_DONE_FLAG = 1 << 0,
	NTBT_LINK_DOWN_FLAG = 1 << 1,
};

struct ntb_payload_header {
	ntb_q_idx_t ver;
	uint32_t len;
	uint32_t flags;
};

enum {
	/*
	 * The order of this enum is part of the remote protocol.  Do not
	 * reorder without bumping protocol version (and it's probably best
	 * to keep the protocol in lock-step with the Linux NTB driver.
	 */
	NTBT_VERSION = 0,
	NTBT_QP_LINKS,
	NTBT_NUM_QPS,
	NTBT_NUM_MWS,
	/*
	 * N.B.: transport_link_work assumes MW1 enums = MW0 + 2.
	 */
	NTBT_MW0_SZ_HIGH,
	NTBT_MW0_SZ_LOW,
	NTBT_MW1_SZ_HIGH,
	NTBT_MW1_SZ_LOW,

	/*
	 * Some NTB-using hardware have a watchdog to work around NTB hangs; if
	 * a register or doorbell isn't written every few seconds, the link is
	 * torn down.  Write an otherwise unused register every few seconds to
	 * work around this watchdog.
	 */
	NTBT_WATCHDOG_SPAD = 15
};

/*
 * Compart version of sratchpad protocol, using twice less registers.
 */
enum {
	NTBTC_PARAMS = 0,	/* NUM_QPS << 24 + NUM_MWS << 16 + VERSION */
	NTBTC_QP_LINKS,		/* QP links status */
	NTBTC_MW0_SZ,		/* MW size limited to 32 bits. */
};

#define QP_TO_MW(nt, qp)	((qp) % nt->mw_count)
#define NTB_QP_DEF_NUM_ENTRIES	100
#define NTB_LINK_DOWN_TIMEOUT	100

static int ntb_transport_probe(device_t dev);
static int ntb_transport_attach(device_t dev);
static int ntb_transport_detach(device_t dev);
static void ntb_transport_init_queue(struct ntb_transport_ctx *nt,
    unsigned int qp_num);
static int ntb_process_tx(struct ntb_transport_qp *qp,
    struct ntb_queue_entry *entry);
static void ntb_transport_rxc_db(void *arg, int pending);
static int ntb_process_rxc(struct ntb_transport_qp *qp);
static void ntb_memcpy_rx(struct ntb_transport_qp *qp,
    struct ntb_queue_entry *entry, void *offset);
static inline void ntb_rx_copy_callback(struct ntb_transport_qp *qp,
    void *data);
static void ntb_complete_rxc(struct ntb_transport_qp *qp);
static void ntb_transport_doorbell_callback(void *data, uint32_t vector);
static void ntb_transport_event_callback(void *data);
static void ntb_transport_link_work(void *arg);
static int ntb_set_mw(struct ntb_transport_ctx *, int num_mw, size_t size);
static void ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw);
static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt,
    unsigned int qp_num);
static void ntb_qp_link_work(void *arg);
static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt);
static void ntb_transport_link_cleanup_work(void *, int);
static void ntb_qp_link_down(struct ntb_transport_qp *qp);
static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp);
static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp);
static void ntb_send_link_down(struct ntb_transport_qp *qp);
static void ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
    struct ntb_queue_list *list);
static struct ntb_queue_entry *ntb_list_rm(struct mtx *lock,
    struct ntb_queue_list *list);
static struct ntb_queue_entry *ntb_list_mv(struct mtx *lock,
    struct ntb_queue_list *from, struct ntb_queue_list *to);
static void xeon_link_watchdog_hb(void *);

static const struct ntb_ctx_ops ntb_transport_ops = {
	.link_event = ntb_transport_event_callback,
	.db_event = ntb_transport_doorbell_callback,
};

MALLOC_DEFINE(M_NTB_T, "ntb_transport", "ntb transport driver");

static inline void
iowrite32(uint32_t val, void *addr)
{

	bus_space_write_4(X86_BUS_SPACE_MEM, 0/* HACK */, (uintptr_t)addr,
	    val);
}

/* Transport Init and teardown */

static void
xeon_link_watchdog_hb(void *arg)
{
	struct ntb_transport_ctx *nt;

	nt = arg;
	ntb_spad_write(nt->dev, NTBT_WATCHDOG_SPAD, 0);
	callout_reset(&nt->link_watchdog, 1 * hz, xeon_link_watchdog_hb, nt);
}

static int
ntb_transport_probe(device_t dev)
{

	device_set_desc(dev, "NTB Transport");
	return (0);
}

static int
ntb_transport_attach(device_t dev)
{
	struct ntb_transport_ctx *nt = device_get_softc(dev);
	struct ntb_transport_child **cpp = &nt->child;
	struct ntb_transport_child *nc;
	struct ntb_transport_mw *mw;
	uint64_t db_bitmap;
	int rc, i, db_count, spad_count, qp, qpu, qpo, qpt;
	char cfg[128] = "";
	char buf[32];
	char *n, *np, *c, *name;

	nt->dev = dev;
	nt->mw_count = ntb_mw_count(dev);
	spad_count = ntb_spad_count(dev);
	db_bitmap = ntb_db_valid_mask(dev);
	db_count = flsll(db_bitmap);
	KASSERT(db_bitmap == ((uint64_t)1 << db_count) - 1,
	    ("Doorbells are not sequential (%jx).\n", db_bitmap));

	if (nt->mw_count == 0) {
		device_printf(dev, "At least 1 memory window required.\n");
		return (ENXIO);
	}
	nt->compact = (spad_count < 4 + 2 * nt->mw_count);
	snprintf(buf, sizeof(buf), "hint.%s.%d.compact", device_get_name(dev),
	    device_get_unit(dev));
	TUNABLE_INT_FETCH(buf, &nt->compact);
	if (nt->compact) {
		if (spad_count < 3) {
			device_printf(dev, "At least 3 scratchpads required.\n");
			return (ENXIO);
		}
		if (spad_count < 2 + nt->mw_count) {
			nt->mw_count = spad_count - 2;
			device_printf(dev, "Scratchpads enough only for %d "
			    "memory windows.\n", nt->mw_count);
		}
	} else {
		if (spad_count < 6) {
			device_printf(dev, "At least 6 scratchpads required.\n");
			return (ENXIO);
		}
		if (spad_count < 4 + 2 * nt->mw_count) {
			nt->mw_count = (spad_count - 4) / 2;
			device_printf(dev, "Scratchpads enough only for %d "
			    "memory windows.\n", nt->mw_count);
		}
	}
	if (db_bitmap == 0) {
		device_printf(dev, "At least one doorbell required.\n");
		return (ENXIO);
	}

	nt->mw_vec = malloc(nt->mw_count * sizeof(*nt->mw_vec), M_NTB_T,
	    M_WAITOK | M_ZERO);
	for (i = 0; i < nt->mw_count; i++) {
		mw = &nt->mw_vec[i];

		rc = ntb_mw_get_range(dev, i, &mw->phys_addr, &mw->vbase,
		    &mw->phys_size, &mw->xlat_align, &mw->xlat_align_size,
		    &mw->addr_limit);
		if (rc != 0)
			goto err;

		mw->tx_size = mw->phys_size;
		if (max_mw_size != 0 && mw->tx_size > max_mw_size) {
			device_printf(dev, "Memory window %d limited from "
			    "%ju to %ju\n", i, (uintmax_t)mw->tx_size,
			    max_mw_size);
			mw->tx_size = max_mw_size;
		}
		if (nt->compact && mw->tx_size > UINT32_MAX) {
			device_printf(dev, "Memory window %d is too big "
			    "(%ju)\n", i, (uintmax_t)mw->tx_size);
			rc = ENXIO;
			goto err;
		}

		mw->rx_size = 0;
		mw->buff_size = 0;
		mw->virt_addr = NULL;
		mw->dma_addr = 0;

		rc = ntb_mw_set_wc(dev, i, VM_MEMATTR_WRITE_COMBINING);
		if (rc)
			ntb_printf(0, "Unable to set mw%d caching\n", i);

		/*
		 * Try to preallocate receive memory early, since there may
		 * be not enough contiguous memory later.  It is quite likely
		 * that NTB windows are symmetric and this allocation remain,
		 * but even if not, we will just reallocate it later.
		 */
		ntb_set_mw(nt, i, mw->tx_size);
	}

	qpu = 0;
	qpo = imin(db_count, nt->mw_count);
	qpt = db_count;

	snprintf(buf, sizeof(buf), "hint.%s.%d.config", device_get_name(dev),
	    device_get_unit(dev));
	TUNABLE_STR_FETCH(buf, cfg, sizeof(cfg));
	n = cfg;
	i = 0;
	while ((c = strsep(&n, ",")) != NULL) {
		np = c;
		name = strsep(&np, ":");
		if (name != NULL && name[0] == 0)
			name = NULL;
		qp = (np && np[0] != 0) ? strtol(np, NULL, 10) : qpo - qpu;
		if (qp <= 0)
			qp = 1;

		if (qp > qpt - qpu) {
			device_printf(dev, "Not enough resources for config\n");
			break;
		}

		nc = malloc(sizeof(*nc), M_DEVBUF, M_WAITOK | M_ZERO);
		nc->consumer = i;
		nc->qpoff = qpu;
		nc->qpcnt = qp;
		nc->dev = device_add_child(dev, name, -1);
		if (nc->dev == NULL) {
			device_printf(dev, "Can not add child.\n");
			break;
		}
		device_set_ivars(nc->dev, nc);
		*cpp = nc;
		cpp = &nc->next;

		if (bootverbose) {
			device_printf(dev, "%d \"%s\": queues %d",
			    i, name, qpu);
			if (qp > 1)
				printf("-%d", qpu + qp - 1);
			printf("\n");
		}

		qpu += qp;
		i++;
	}
	nt->qp_count = qpu;

	nt->qp_vec = malloc(nt->qp_count * sizeof(*nt->qp_vec), M_NTB_T,
	    M_WAITOK | M_ZERO);

	for (i = 0; i < nt->qp_count; i++)
		ntb_transport_init_queue(nt, i);

	callout_init(&nt->link_work, 1);
	callout_init(&nt->link_watchdog, 1);
	TASK_INIT(&nt->link_cleanup, 0, ntb_transport_link_cleanup_work, nt);
	nt->link_is_up = false;

	rc = ntb_set_ctx(dev, nt, &ntb_transport_ops);
	if (rc != 0)
		goto err;

	ntb_link_enable(dev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);

	for (i = 0; i < nt->mw_count; i++) {
		mw = &nt->mw_vec[i];
		rc = ntb_mw_set_trans(nt->dev, i, mw->dma_addr, mw->buff_size);
		if (rc != 0)
			ntb_printf(0, "load time mw%d xlat fails, rc %d\n", i, rc);
	}

	if (enable_xeon_watchdog != 0)
		callout_reset(&nt->link_watchdog, 0, xeon_link_watchdog_hb, nt);

	bus_generic_attach(dev);
	return (0);

err:
	free(nt->qp_vec, M_NTB_T);
	free(nt->mw_vec, M_NTB_T);
	return (rc);
}

static int
ntb_transport_detach(device_t dev)
{
	struct ntb_transport_ctx *nt = device_get_softc(dev);
	struct ntb_transport_child **cpp = &nt->child;
	struct ntb_transport_child *nc;
	int error = 0, i;

	while ((nc = *cpp) != NULL) {
		*cpp = (*cpp)->next;
		error = device_delete_child(dev, nc->dev);
		if (error)
			break;
		free(nc, M_DEVBUF);
	}
	KASSERT(nt->qp_bitmap == 0,
	    ("Some queues not freed on detach (%jx)", nt->qp_bitmap));

	ntb_transport_link_cleanup(nt);
	taskqueue_drain(taskqueue_swi, &nt->link_cleanup);
	callout_drain(&nt->link_work);
	callout_drain(&nt->link_watchdog);

	ntb_link_disable(dev);
	ntb_clear_ctx(dev);

	for (i = 0; i < nt->mw_count; i++)
		ntb_free_mw(nt, i);

	free(nt->qp_vec, M_NTB_T);
	free(nt->mw_vec, M_NTB_T);
	return (0);
}

static int
ntb_transport_print_child(device_t dev, device_t child)
{
	struct ntb_transport_child *nc = device_get_ivars(child);
	int retval;

	retval = bus_print_child_header(dev, child);
	if (nc->qpcnt > 0) {
		printf(" queue %d", nc->qpoff);
		if (nc->qpcnt > 1)
			printf("-%d", nc->qpoff + nc->qpcnt - 1);
	}
	retval += printf(" at consumer %d", nc->consumer);
	retval += bus_print_child_domain(dev, child);
	retval += bus_print_child_footer(dev, child);

	return (retval);
}

static int
ntb_transport_child_location(device_t dev, device_t child, struct sbuf *sb)
{
	struct ntb_transport_child *nc = device_get_ivars(child);

	sbuf_printf(sb, "consumer=%d", nc->consumer);
	return (0);
}

int
ntb_transport_queue_count(device_t dev)
{
	struct ntb_transport_child *nc = device_get_ivars(dev);

	return (nc->qpcnt);
}

static void
ntb_transport_init_queue(struct ntb_transport_ctx *nt, unsigned int qp_num)
{
	struct ntb_transport_mw *mw;
	struct ntb_transport_qp *qp;
	vm_paddr_t mw_base;
	uint64_t qp_offset;
	size_t tx_size;
	unsigned num_qps_mw, mw_num, mw_count;

	mw_count = nt->mw_count;
	mw_num = QP_TO_MW(nt, qp_num);
	mw = &nt->mw_vec[mw_num];

	qp = &nt->qp_vec[qp_num];
	qp->qp_num = qp_num;
	qp->transport = nt;
	qp->dev = nt->dev;
	qp->client_ready = false;
	qp->event_handler = NULL;
	ntb_qp_link_down_reset(qp);

	if (mw_num < nt->qp_count % mw_count)
		num_qps_mw = nt->qp_count / mw_count + 1;
	else
		num_qps_mw = nt->qp_count / mw_count;

	mw_base = mw->phys_addr;

	tx_size = mw->tx_size / num_qps_mw;
	qp_offset = tx_size * (qp_num / mw_count);

	qp->tx_mw = mw->vbase + qp_offset;
	KASSERT(qp->tx_mw != NULL, ("uh oh?"));

	/* XXX Assumes that a vm_paddr_t is equivalent to bus_addr_t */
	qp->tx_mw_phys = mw_base + qp_offset;
	KASSERT(qp->tx_mw_phys != 0, ("uh oh?"));

	tx_size -= sizeof(struct ntb_rx_info);
	qp->rx_info = (void *)(qp->tx_mw + tx_size);

	/* Due to house-keeping, there must be at least 2 buffs */
	qp->tx_max_frame = qmin(transport_mtu, tx_size / 2);
	qp->tx_max_entry = tx_size / qp->tx_max_frame;

	callout_init(&qp->link_work, 1);
	callout_init(&qp->rx_full, 1);

	mtx_init(&qp->ntb_rx_q_lock, "ntb rx q", NULL, MTX_SPIN);
	mtx_init(&qp->ntb_tx_free_q_lock, "ntb tx free q", NULL, MTX_SPIN);
	mtx_init(&qp->tx_lock, "ntb transport tx", NULL, MTX_DEF);
	TASK_INIT(&qp->rxc_db_work, 0, ntb_transport_rxc_db, qp);
	qp->rxc_tq = taskqueue_create("ntbt_rx", M_WAITOK,
	    taskqueue_thread_enqueue, &qp->rxc_tq);
	taskqueue_start_threads(&qp->rxc_tq, 1, PI_NET, "%s rx%d",
	    device_get_nameunit(nt->dev), qp_num);

	STAILQ_INIT(&qp->rx_post_q);
	STAILQ_INIT(&qp->rx_pend_q);
	STAILQ_INIT(&qp->tx_free_q);
}

void
ntb_transport_free_queue(struct ntb_transport_qp *qp)
{
	struct ntb_transport_ctx *nt = qp->transport;
	struct ntb_queue_entry *entry;

	callout_drain(&qp->link_work);

	ntb_db_set_mask(qp->dev, 1ull << qp->qp_num);
	taskqueue_drain_all(qp->rxc_tq);
	taskqueue_free(qp->rxc_tq);

	qp->cb_data = NULL;
	qp->rx_handler = NULL;
	qp->tx_handler = NULL;
	qp->event_handler = NULL;

	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q)))
		free(entry, M_NTB_T);

	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q)))
		free(entry, M_NTB_T);

	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
		free(entry, M_NTB_T);

	nt->qp_bitmap &= ~(1 << qp->qp_num);
}

/**
 * ntb_transport_create_queue - Create a new NTB transport layer queue
 * @rx_handler: receive callback function
 * @tx_handler: transmit callback function
 * @event_handler: event callback function
 *
 * Create a new NTB transport layer queue and provide the queue with a callback
 * routine for both transmit and receive.  The receive callback routine will be
 * used to pass up data when the transport has received it on the queue.   The
 * transmit callback routine will be called when the transport has completed the
 * transmission of the data on the queue and the data is ready to be freed.
 *
 * RETURNS: pointer to newly created ntb_queue, NULL on error.
 */
struct ntb_transport_qp *
ntb_transport_create_queue(device_t dev, int q,
    const struct ntb_queue_handlers *handlers, void *data)
{
	struct ntb_transport_child *nc = device_get_ivars(dev);
	struct ntb_transport_ctx *nt = device_get_softc(device_get_parent(dev));
	struct ntb_queue_entry *entry;
	struct ntb_transport_qp *qp;
	int i;

	if (q < 0 || q >= nc->qpcnt)
		return (NULL);

	qp = &nt->qp_vec[nc->qpoff + q];
	nt->qp_bitmap |= (1 << qp->qp_num);
	qp->cb_data = data;
	qp->rx_handler = handlers->rx_handler;
	qp->tx_handler = handlers->tx_handler;
	qp->event_handler = handlers->event_handler;

	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
		entry = malloc(sizeof(*entry), M_NTB_T, M_WAITOK | M_ZERO);
		entry->cb_data = data;
		entry->buf = NULL;
		entry->len = transport_mtu;
		entry->qp = qp;
		ntb_list_add(&qp->ntb_rx_q_lock, entry, &qp->rx_pend_q);
	}

	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
		entry = malloc(sizeof(*entry), M_NTB_T, M_WAITOK | M_ZERO);
		entry->qp = qp;
		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
	}

	ntb_db_clear(dev, 1ull << qp->qp_num);
	return (qp);
}

/**
 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
 * @qp: NTB transport layer queue to be enabled
 *
 * Notify NTB transport layer of client readiness to use queue
 */
void
ntb_transport_link_up(struct ntb_transport_qp *qp)
{
	struct ntb_transport_ctx *nt = qp->transport;

	qp->client_ready = true;

	ntb_printf(2, "qp %d client ready\n", qp->qp_num);

	if (nt->link_is_up)
		callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
}

/* Transport Tx */

/**
 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
 * @qp: NTB transport layer queue the entry is to be enqueued on
 * @cb: per buffer pointer for callback function to use
 * @data: pointer to data buffer that will be sent
 * @len: length of the data buffer
 *
 * Enqueue a new transmit buffer onto the transport queue from which a NTB
 * payload will be transmitted.  This assumes that a lock is being held to
 * serialize access to the qp.
 *
 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
 */
int
ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
    unsigned int len)
{
	struct ntb_queue_entry *entry;
	int rc;

	if (!qp->link_is_up || len == 0) {
		CTR0(KTR_NTB, "TX: link not up");
		return (EINVAL);
	}

	entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
	if (entry == NULL) {
		CTR0(KTR_NTB, "TX: could not get entry from tx_free_q");
		qp->tx_err_no_buf++;
		return (EBUSY);
	}
	CTR1(KTR_NTB, "TX: got entry %p from tx_free_q", entry);

	entry->cb_data = cb;
	entry->buf = data;
	entry->len = len;
	entry->flags = 0;

	mtx_lock(&qp->tx_lock);
	rc = ntb_process_tx(qp, entry);
	mtx_unlock(&qp->tx_lock);
	if (rc != 0) {
		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
		CTR1(KTR_NTB,
		    "TX: process_tx failed. Returning entry %p to tx_free_q",
		    entry);
	}
	return (rc);
}

static void
ntb_tx_copy_callback(void *data)
{
	struct ntb_queue_entry *entry = data;
	struct ntb_transport_qp *qp = entry->qp;
	struct ntb_payload_header *hdr = entry->x_hdr;

	iowrite32(entry->flags | NTBT_DESC_DONE_FLAG, &hdr->flags);
	CTR1(KTR_NTB, "TX: hdr %p set DESC_DONE", hdr);

	ntb_peer_db_set(qp->dev, 1ull << qp->qp_num);

	/*
	 * The entry length can only be zero if the packet is intended to be a
	 * "link down" or similar.  Since no payload is being sent in these
	 * cases, there is nothing to add to the completion queue.
	 */
	if (entry->len > 0) {
		qp->tx_bytes += entry->len;

		if (qp->tx_handler)
			qp->tx_handler(qp, qp->cb_data, entry->buf,
			    entry->len);
		else
			m_freem(entry->buf);
		entry->buf = NULL;
	}

	CTR3(KTR_NTB,
	    "TX: entry %p sent. hdr->ver = %u, hdr->flags = 0x%x, Returning "
	    "to tx_free_q", entry, hdr->ver, hdr->flags);
	ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
}

static void
ntb_memcpy_tx(struct ntb_queue_entry *entry, void *offset)
{

	CTR2(KTR_NTB, "TX: copying %d bytes to offset %p", entry->len, offset);
	if (entry->buf != NULL) {
		m_copydata((struct mbuf *)entry->buf, 0, entry->len, offset);

		/*
		 * Ensure that the data is fully copied before setting the
		 * flags
		 */
		wmb();
	}

	ntb_tx_copy_callback(entry);
}

static void
ntb_async_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry)
{
	struct ntb_payload_header *hdr;
	void *offset;

	offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index;
	hdr = (struct ntb_payload_header *)((char *)offset + qp->tx_max_frame -
	    sizeof(struct ntb_payload_header));
	entry->x_hdr = hdr;

	iowrite32(entry->len, &hdr->len);
	iowrite32(qp->tx_pkts, &hdr->ver);

	ntb_memcpy_tx(entry, offset);
}

static int
ntb_process_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry)
{

	CTR3(KTR_NTB,
	    "TX: process_tx: tx_pkts=%lu, tx_index=%u, remote entry=%u",
	    qp->tx_pkts, qp->tx_index, qp->remote_rx_info->entry);
	if (qp->tx_index == qp->remote_rx_info->entry) {
		CTR0(KTR_NTB, "TX: ring full");
		qp->tx_ring_full++;
		return (EAGAIN);
	}

	if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
		if (qp->tx_handler != NULL)
			qp->tx_handler(qp, qp->cb_data, entry->buf,
			    EIO);
		else
			m_freem(entry->buf);

		entry->buf = NULL;
		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
		CTR1(KTR_NTB,
		    "TX: frame too big. returning entry %p to tx_free_q",
		    entry);
		return (0);
	}
	CTR2(KTR_NTB, "TX: copying entry %p to index %u", entry, qp->tx_index);
	ntb_async_tx(qp, entry);

	qp->tx_index++;
	qp->tx_index %= qp->tx_max_entry;

	qp->tx_pkts++;

	return (0);
}

/* Transport Rx */
static void
ntb_transport_rxc_db(void *arg, int pending __unused)
{
	struct ntb_transport_qp *qp = arg;
	uint64_t qp_mask = 1ull << qp->qp_num;
	int rc;

	CTR0(KTR_NTB, "RX: transport_rx");
again:
	while ((rc = ntb_process_rxc(qp)) == 0)
		;
	CTR1(KTR_NTB, "RX: process_rxc returned %d", rc);

	if ((ntb_db_read(qp->dev) & qp_mask) != 0) {
		/* If db is set, clear it and check queue once more. */
		ntb_db_clear(qp->dev, qp_mask);
		goto again;
	}
	if (qp->link_is_up)
		ntb_db_clear_mask(qp->dev, qp_mask);
}

static int
ntb_process_rxc(struct ntb_transport_qp *qp)
{
	struct ntb_payload_header *hdr;
	struct ntb_queue_entry *entry;
	caddr_t offset;

	offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index;
	hdr = (void *)(offset + qp->rx_max_frame -
	    sizeof(struct ntb_payload_header));

	CTR1(KTR_NTB, "RX: process_rxc rx_index = %u", qp->rx_index);
	if ((hdr->flags & NTBT_DESC_DONE_FLAG) == 0) {
		CTR0(KTR_NTB, "RX: hdr not done");
		qp->rx_ring_empty++;
		return (EAGAIN);
	}

	if ((hdr->flags & NTBT_LINK_DOWN_FLAG) != 0) {
		CTR0(KTR_NTB, "RX: link down");
		ntb_qp_link_down(qp);
		hdr->flags = 0;
		return (EAGAIN);
	}

	if (hdr->ver != (uint32_t)qp->rx_pkts) {
		CTR2(KTR_NTB,"RX: ver != rx_pkts (%x != %lx). "
		    "Returning entry to rx_pend_q", hdr->ver, qp->rx_pkts);
		qp->rx_err_ver++;
		return (EIO);
	}

	entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q);
	if (entry == NULL) {
		qp->rx_err_no_buf++;
		CTR0(KTR_NTB, "RX: No entries in rx_pend_q");
		return (EAGAIN);
	}
	callout_stop(&qp->rx_full);
	CTR1(KTR_NTB, "RX: rx entry %p from rx_pend_q", entry);

	entry->x_hdr = hdr;
	entry->index = qp->rx_index;

	if (hdr->len > entry->len) {
		CTR2(KTR_NTB, "RX: len too long. Wanted %ju got %ju",
		    (uintmax_t)hdr->len, (uintmax_t)entry->len);
		qp->rx_err_oflow++;

		entry->len = -EIO;
		entry->flags |= NTBT_DESC_DONE_FLAG;

		ntb_complete_rxc(qp);
	} else {
		qp->rx_bytes += hdr->len;
		qp->rx_pkts++;

		CTR1(KTR_NTB, "RX: received %ld rx_pkts", qp->rx_pkts);

		entry->len = hdr->len;

		ntb_memcpy_rx(qp, entry, offset);
	}

	qp->rx_index++;
	qp->rx_index %= qp->rx_max_entry;
	return (0);
}

static void
ntb_memcpy_rx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry,
    void *offset)
{
	struct ifnet *ifp = entry->cb_data;
	unsigned int len = entry->len;

	CTR2(KTR_NTB, "RX: copying %d bytes from offset %p", len, offset);

	entry->buf = (void *)m_devget(offset, len, 0, ifp, NULL);
	if (entry->buf == NULL)
		entry->len = -ENOMEM;

	/* Ensure that the data is globally visible before clearing the flag */
	wmb();

	CTR2(KTR_NTB, "RX: copied entry %p to mbuf %p.", entry, entry->buf);
	ntb_rx_copy_callback(qp, entry);
}

static inline void
ntb_rx_copy_callback(struct ntb_transport_qp *qp, void *data)
{
	struct ntb_queue_entry *entry;

	entry = data;
	entry->flags |= NTBT_DESC_DONE_FLAG;
	ntb_complete_rxc(qp);
}

static void
ntb_complete_rxc(struct ntb_transport_qp *qp)
{
	struct ntb_queue_entry *entry;
	struct mbuf *m;
	unsigned len;

	CTR0(KTR_NTB, "RX: rx_completion_task");

	mtx_lock_spin(&qp->ntb_rx_q_lock);

	while (!STAILQ_EMPTY(&qp->rx_post_q)) {
		entry = STAILQ_FIRST(&qp->rx_post_q);
		if ((entry->flags & NTBT_DESC_DONE_FLAG) == 0)
			break;

		entry->x_hdr->flags = 0;
		iowrite32(entry->index, &qp->rx_info->entry);

		STAILQ_REMOVE_HEAD(&qp->rx_post_q, entry);

		len = entry->len;
		m = entry->buf;

		/*
		 * Re-initialize queue_entry for reuse; rx_handler takes
		 * ownership of the mbuf.
		 */
		entry->buf = NULL;
		entry->len = transport_mtu;
		entry->cb_data = qp->cb_data;

		STAILQ_INSERT_TAIL(&qp->rx_pend_q, entry, entry);

		mtx_unlock_spin(&qp->ntb_rx_q_lock);

		CTR2(KTR_NTB, "RX: completing entry %p, mbuf %p", entry, m);
		if (qp->rx_handler != NULL && qp->client_ready)
			qp->rx_handler(qp, qp->cb_data, m, len);
		else
			m_freem(m);

		mtx_lock_spin(&qp->ntb_rx_q_lock);
	}

	mtx_unlock_spin(&qp->ntb_rx_q_lock);
}

static void
ntb_transport_doorbell_callback(void *data, uint32_t vector)
{
	struct ntb_transport_ctx *nt = data;
	struct ntb_transport_qp *qp;
	uint64_t vec_mask;
	unsigned qp_num;

	vec_mask = ntb_db_vector_mask(nt->dev, vector);
	vec_mask &= nt->qp_bitmap;
	if ((vec_mask & (vec_mask - 1)) != 0)
		vec_mask &= ntb_db_read(nt->dev);
	if (vec_mask != 0) {
		ntb_db_set_mask(nt->dev, vec_mask);
		ntb_db_clear(nt->dev, vec_mask);
	}
	while (vec_mask != 0) {
		qp_num = ffsll(vec_mask) - 1;

		qp = &nt->qp_vec[qp_num];
		if (qp->link_is_up)
			taskqueue_enqueue(qp->rxc_tq, &qp->rxc_db_work);

		vec_mask &= ~(1ull << qp_num);
	}
}

/* Link Event handler */
static void
ntb_transport_event_callback(void *data)
{
	struct ntb_transport_ctx *nt = data;

	if (ntb_link_is_up(nt->dev, &nt->link_speed, &nt->link_width)) {
		ntb_printf(1, "HW link up\n");
		callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt);
	} else {
		ntb_printf(1, "HW link down\n");
		taskqueue_enqueue(taskqueue_swi, &nt->link_cleanup);
	}
}

/* Link bring up */
static void
ntb_transport_link_work(void *arg)
{
	struct ntb_transport_ctx *nt = arg;
	struct ntb_transport_mw *mw;
	device_t dev = nt->dev;
	struct ntb_transport_qp *qp;
	uint64_t val64, size;
	uint32_t val;
	unsigned i;
	int rc;

	/* send the local info, in the opposite order of the way we read it */
	if (nt->compact) {
		for (i = 0; i < nt->mw_count; i++) {
			size = nt->mw_vec[i].tx_size;
			KASSERT(size <= UINT32_MAX, ("size too big (%jx)", size));
			ntb_peer_spad_write(dev, NTBTC_MW0_SZ + i, size);
		}
		ntb_peer_spad_write(dev, NTBTC_QP_LINKS, 0);
		ntb_peer_spad_write(dev, NTBTC_PARAMS,
		    (nt->qp_count << 24) | (nt->mw_count << 16) |
		    NTB_TRANSPORT_VERSION);
	} else {
		for (i = 0; i < nt->mw_count; i++) {
			size = nt->mw_vec[i].tx_size;
			ntb_peer_spad_write(dev, NTBT_MW0_SZ_HIGH + (i * 2),
			    size >> 32);
			ntb_peer_spad_write(dev, NTBT_MW0_SZ_LOW + (i * 2), size);
		}
		ntb_peer_spad_write(dev, NTBT_NUM_MWS, nt->mw_count);
		ntb_peer_spad_write(dev, NTBT_NUM_QPS, nt->qp_count);
		ntb_peer_spad_write(dev, NTBT_QP_LINKS, 0);
		ntb_peer_spad_write(dev, NTBT_VERSION, NTB_TRANSPORT_VERSION);
	}

	/* Query the remote side for its info */
	val = 0;
	if (nt->compact) {
		ntb_spad_read(dev, NTBTC_PARAMS, &val);
		if (val != ((nt->qp_count << 24) | (nt->mw_count << 16) |
		    NTB_TRANSPORT_VERSION))
			goto out;
	} else {
		ntb_spad_read(dev, NTBT_VERSION, &val);
		if (val != NTB_TRANSPORT_VERSION)
			goto out;

		ntb_spad_read(dev, NTBT_NUM_QPS, &val);
		if (val != nt->qp_count)
			goto out;

		ntb_spad_read(dev, NTBT_NUM_MWS, &val);
		if (val != nt->mw_count)
			goto out;
	}

	for (i = 0; i < nt->mw_count; i++) {
		if (nt->compact) {
			ntb_spad_read(dev, NTBTC_MW0_SZ + i, &val);
			val64 = val;
		} else {
			ntb_spad_read(dev, NTBT_MW0_SZ_HIGH + (i * 2), &val);
			val64 = (uint64_t)val << 32;

			ntb_spad_read(dev, NTBT_MW0_SZ_LOW + (i * 2), &val);
			val64 |= val;
		}

		mw = &nt->mw_vec[i];
		mw->rx_size = val64;
		val64 = roundup(val64, mw->xlat_align_size);
		if (mw->buff_size != val64) {
			rc = ntb_set_mw(nt, i, val64);
			if (rc != 0) {
				ntb_printf(0, "link up set mw%d fails, rc %d\n",
				    i, rc);
				goto free_mws;
			}

			/* Notify HW the memory location of the receive buffer */
			rc = ntb_mw_set_trans(nt->dev, i, mw->dma_addr,
			    mw->buff_size);
			if (rc != 0) {
				ntb_printf(0, "link up mw%d xlat fails, rc %d\n",
				     i, rc);
				goto free_mws;
			}
		}
	}

	nt->link_is_up = true;
	ntb_printf(1, "transport link up\n");

	for (i = 0; i < nt->qp_count; i++) {
		qp = &nt->qp_vec[i];

		ntb_transport_setup_qp_mw(nt, i);

		if (qp->client_ready)
			callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
	}

	return;

free_mws:
	for (i = 0; i < nt->mw_count; i++)
		ntb_free_mw(nt, i);
out:
	if (ntb_link_is_up(dev, &nt->link_speed, &nt->link_width))
		callout_reset(&nt->link_work,
		    NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_transport_link_work, nt);
}

struct ntb_load_cb_args {
	bus_addr_t addr;
	int error;
};

static void
ntb_load_cb(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
{
	struct ntb_load_cb_args *cba = (struct ntb_load_cb_args *)xsc;

	if (!(cba->error = error))
		cba->addr = segs[0].ds_addr;
}

static int
ntb_set_mw(struct ntb_transport_ctx *nt, int num_mw, size_t size)
{
	struct ntb_transport_mw *mw = &nt->mw_vec[num_mw];
	struct ntb_load_cb_args cba;
	size_t buff_size;

	if (size == 0)
		return (EINVAL);

	buff_size = roundup(size, mw->xlat_align_size);

	/* No need to re-setup */
	if (mw->buff_size == buff_size)
		return (0);

	if (mw->buff_size != 0)
		ntb_free_mw(nt, num_mw);

	/* Alloc memory for receiving data.  Must be aligned */
	mw->buff_size = buff_size;

	if (bus_dma_tag_create(bus_get_dma_tag(nt->dev), mw->xlat_align, 0,
	    mw->addr_limit, BUS_SPACE_MAXADDR,
	    NULL, NULL, mw->buff_size, 1, mw->buff_size,
	    0, NULL, NULL, &mw->dma_tag)) {
		ntb_printf(0, "Unable to create MW tag of size %zu\n",
		    mw->buff_size);
		mw->buff_size = 0;
		return (ENOMEM);
	}
	if (bus_dmamem_alloc(mw->dma_tag, (void **)&mw->virt_addr,
	    BUS_DMA_WAITOK | BUS_DMA_ZERO, &mw->dma_map)) {
		bus_dma_tag_destroy(mw->dma_tag);
		ntb_printf(0, "Unable to allocate MW buffer of size %zu\n",
		    mw->buff_size);
		mw->buff_size = 0;
		return (ENOMEM);
	}
	if (bus_dmamap_load(mw->dma_tag, mw->dma_map, mw->virt_addr,
	    mw->buff_size, ntb_load_cb, &cba, BUS_DMA_NOWAIT) || cba.error) {
		bus_dmamem_free(mw->dma_tag, mw->virt_addr, mw->dma_map);
		bus_dma_tag_destroy(mw->dma_tag);
		ntb_printf(0, "Unable to load MW buffer of size %zu\n",
		    mw->buff_size);
		mw->buff_size = 0;
		return (ENOMEM);
	}
	mw->dma_addr = cba.addr;

	return (0);
}

static void
ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw)
{
	struct ntb_transport_mw *mw = &nt->mw_vec[num_mw];

	if (mw->virt_addr == NULL)
		return;

	ntb_mw_clear_trans(nt->dev, num_mw);
	bus_dmamap_unload(mw->dma_tag, mw->dma_map);
	bus_dmamem_free(mw->dma_tag, mw->virt_addr, mw->dma_map);
	bus_dma_tag_destroy(mw->dma_tag);
	mw->buff_size = 0;
	mw->virt_addr = NULL;
}

static int
ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt, unsigned int qp_num)
{
	struct ntb_transport_qp *qp = &nt->qp_vec[qp_num];
	struct ntb_transport_mw *mw;
	void *offset;
	ntb_q_idx_t i;
	size_t rx_size;
	unsigned num_qps_mw, mw_num, mw_count;

	mw_count = nt->mw_count;
	mw_num = QP_TO_MW(nt, qp_num);
	mw = &nt->mw_vec[mw_num];

	if (mw->virt_addr == NULL)
		return (ENOMEM);

	if (mw_num < nt->qp_count % mw_count)
		num_qps_mw = nt->qp_count / mw_count + 1;
	else
		num_qps_mw = nt->qp_count / mw_count;

	rx_size = mw->rx_size / num_qps_mw;
	qp->rx_buff = mw->virt_addr + rx_size * (qp_num / mw_count);
	rx_size -= sizeof(struct ntb_rx_info);

	qp->remote_rx_info = (void*)(qp->rx_buff + rx_size);

	/* Due to house-keeping, there must be at least 2 buffs */
	qp->rx_max_frame = qmin(transport_mtu, rx_size / 2);
	qp->rx_max_entry = rx_size / qp->rx_max_frame;
	qp->rx_index = 0;

	qp->remote_rx_info->entry = qp->rx_max_entry - 1;

	/* Set up the hdr offsets with 0s */
	for (i = 0; i < qp->rx_max_entry; i++) {
		offset = (void *)(qp->rx_buff + qp->rx_max_frame * (i + 1) -
		    sizeof(struct ntb_payload_header));
		memset(offset, 0, sizeof(struct ntb_payload_header));
	}

	qp->rx_pkts = 0;
	qp->tx_pkts = 0;
	qp->tx_index = 0;

	return (0);
}

static void
ntb_qp_link_work(void *arg)
{
	struct ntb_transport_qp *qp = arg;
	device_t dev = qp->dev;
	struct ntb_transport_ctx *nt = qp->transport;
	int i;
	uint32_t val;

	/* Report queues that are up on our side */
	for (i = 0, val = 0; i < nt->qp_count; i++) {
		if (nt->qp_vec[i].client_ready)
			val |= (1 << i);
	}
	ntb_peer_spad_write(dev, NTBT_QP_LINKS, val);

	/* See if the remote side is up */
	ntb_spad_read(dev, NTBT_QP_LINKS, &val);
	if ((val & (1ull << qp->qp_num)) != 0) {
		ntb_printf(2, "qp %d link up\n", qp->qp_num);
		qp->link_is_up = true;

		if (qp->event_handler != NULL)
			qp->event_handler(qp->cb_data, NTB_LINK_UP);

		ntb_db_clear_mask(dev, 1ull << qp->qp_num);
	} else if (nt->link_is_up)
		callout_reset(&qp->link_work,
		    NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp);
}

/* Link down event*/
static void
ntb_transport_link_cleanup(struct ntb_transport_ctx *nt)
{
	struct ntb_transport_qp *qp;
	int i;

	callout_drain(&nt->link_work);
	nt->link_is_up = 0;

	/* Pass along the info to any clients */
	for (i = 0; i < nt->qp_count; i++) {
		if ((nt->qp_bitmap & (1 << i)) != 0) {
			qp = &nt->qp_vec[i];
			ntb_qp_link_cleanup(qp);
			callout_drain(&qp->link_work);
		}
	}

	/*
	 * The scratchpad registers keep the values if the remote side
	 * goes down, blast them now to give them a sane value the next
	 * time they are accessed
	 */
	ntb_spad_clear(nt->dev);
}

static void
ntb_transport_link_cleanup_work(void *arg, int pending __unused)
{

	ntb_transport_link_cleanup(arg);
}

static void
ntb_qp_link_down(struct ntb_transport_qp *qp)
{

	ntb_qp_link_cleanup(qp);
}

static void
ntb_qp_link_down_reset(struct ntb_transport_qp *qp)
{

	qp->link_is_up = false;
	ntb_db_set_mask(qp->dev, 1ull << qp->qp_num);

	qp->tx_index = qp->rx_index = 0;
	qp->tx_bytes = qp->rx_bytes = 0;
	qp->tx_pkts = qp->rx_pkts = 0;

	qp->rx_ring_empty = 0;
	qp->tx_ring_full = 0;

	qp->rx_err_no_buf = qp->tx_err_no_buf = 0;
	qp->rx_err_oflow = qp->rx_err_ver = 0;
}

static void
ntb_qp_link_cleanup(struct ntb_transport_qp *qp)
{

	callout_drain(&qp->link_work);
	ntb_qp_link_down_reset(qp);

	if (qp->event_handler != NULL)
		qp->event_handler(qp->cb_data, NTB_LINK_DOWN);
}

/* Link commanded down */
/**
 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
 * @qp: NTB transport layer queue to be disabled
 *
 * Notify NTB transport layer of client's desire to no longer receive data on
 * transport queue specified.  It is the client's responsibility to ensure all
 * entries on queue are purged or otherwise handled appropriately.
 */
void
ntb_transport_link_down(struct ntb_transport_qp *qp)
{
	struct ntb_transport_ctx *nt = qp->transport;
	int i;
	uint32_t val;

	qp->client_ready = false;
	for (i = 0, val = 0; i < nt->qp_count; i++) {
		if (nt->qp_vec[i].client_ready)
			val |= (1 << i);
	}
	ntb_peer_spad_write(qp->dev, NTBT_QP_LINKS, val);

	if (qp->link_is_up)
		ntb_send_link_down(qp);
	else
		callout_drain(&qp->link_work);
}

/**
 * ntb_transport_link_query - Query transport link state
 * @qp: NTB transport layer queue to be queried
 *
 * Query connectivity to the remote system of the NTB transport queue
 *
 * RETURNS: true for link up or false for link down
 */
bool
ntb_transport_link_query(struct ntb_transport_qp *qp)
{

	return (qp->link_is_up);
}

/**
 * ntb_transport_link_speed - Query transport link speed
 * @qp: NTB transport layer queue to be queried
 *
 * Query connection speed to the remote system of the NTB transport queue
 *
 * RETURNS: link speed in bits per second
 */
uint64_t
ntb_transport_link_speed(struct ntb_transport_qp *qp)
{
	struct ntb_transport_ctx *nt = qp->transport;
	uint64_t rate;

	if (!nt->link_is_up)
		return (0);
	switch (nt->link_speed) {
	case NTB_SPEED_GEN1:
		rate = 2500000000 * 8 / 10;
		break;
	case NTB_SPEED_GEN2:
		rate = 5000000000 * 8 / 10;
		break;
	case NTB_SPEED_GEN3:
		rate = 8000000000 * 128 / 130;
		break;
	case NTB_SPEED_GEN4:
		rate = 16000000000 * 128 / 130;
		break;
	default:
		return (0);
	}
	if (nt->link_width <= 0)
		return (0);
	return (rate * nt->link_width);
}

static void
ntb_send_link_down(struct ntb_transport_qp *qp)
{
	struct ntb_queue_entry *entry;
	int i, rc;

	if (!qp->link_is_up)
		return;

	for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
		entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
		if (entry != NULL)
			break;
		pause("NTB Wait for link down", hz / 10);
	}

	if (entry == NULL)
		return;

	entry->cb_data = NULL;
	entry->buf = NULL;
	entry->len = 0;
	entry->flags = NTBT_LINK_DOWN_FLAG;

	mtx_lock(&qp->tx_lock);
	rc = ntb_process_tx(qp, entry);
	mtx_unlock(&qp->tx_lock);
	if (rc != 0)
		printf("ntb: Failed to send link down\n");

	ntb_qp_link_down_reset(qp);
}

/* List Management */

static void
ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
    struct ntb_queue_list *list)
{

	mtx_lock_spin(lock);
	STAILQ_INSERT_TAIL(list, entry, entry);
	mtx_unlock_spin(lock);
}

static struct ntb_queue_entry *
ntb_list_rm(struct mtx *lock, struct ntb_queue_list *list)
{
	struct ntb_queue_entry *entry;

	mtx_lock_spin(lock);
	if (STAILQ_EMPTY(list)) {
		entry = NULL;
		goto out;
	}
	entry = STAILQ_FIRST(list);
	STAILQ_REMOVE_HEAD(list, entry);
out:
	mtx_unlock_spin(lock);

	return (entry);
}

static struct ntb_queue_entry *
ntb_list_mv(struct mtx *lock, struct ntb_queue_list *from,
    struct ntb_queue_list *to)
{
	struct ntb_queue_entry *entry;

	mtx_lock_spin(lock);
	if (STAILQ_EMPTY(from)) {
		entry = NULL;
		goto out;
	}
	entry = STAILQ_FIRST(from);
	STAILQ_REMOVE_HEAD(from, entry);
	STAILQ_INSERT_TAIL(to, entry, entry);

out:
	mtx_unlock_spin(lock);
	return (entry);
}

/**
 * ntb_transport_qp_num - Query the qp number
 * @qp: NTB transport layer queue to be queried
 *
 * Query qp number of the NTB transport queue
 *
 * RETURNS: a zero based number specifying the qp number
 */
unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp)
{

	return (qp->qp_num);
}

/**
 * ntb_transport_max_size - Query the max payload size of a qp
 * @qp: NTB transport layer queue to be queried
 *
 * Query the maximum payload size permissible on the given qp
 *
 * RETURNS: the max payload size of a qp
 */
unsigned int
ntb_transport_max_size(struct ntb_transport_qp *qp)
{

	return (qp->tx_max_frame - sizeof(struct ntb_payload_header));
}

unsigned int
ntb_transport_tx_free_entry(struct ntb_transport_qp *qp)
{
	unsigned int head = qp->tx_index;
	unsigned int tail = qp->remote_rx_info->entry;

	return (tail >= head ? tail - head : qp->tx_max_entry + tail - head);
}

static device_method_t ntb_transport_methods[] = {
	/* Device interface */
	DEVMETHOD(device_probe,     ntb_transport_probe),
	DEVMETHOD(device_attach,    ntb_transport_attach),
	DEVMETHOD(device_detach,    ntb_transport_detach),
	/* Bus interface */
	DEVMETHOD(bus_child_location, ntb_transport_child_location),
	DEVMETHOD(bus_print_child,  ntb_transport_print_child),
	DEVMETHOD_END
};

static DEFINE_CLASS_0(ntb_transport, ntb_transport_driver,
    ntb_transport_methods, sizeof(struct ntb_transport_ctx));
DRIVER_MODULE(ntb_transport, ntb_hw, ntb_transport_driver, NULL, NULL);
MODULE_DEPEND(ntb_transport, ntb, 1, 1, 1);
MODULE_VERSION(ntb_transport, 1);