[linux-block.git] / net / smc / smc_ib.c

// SPDX-License-Identifier: GPL-2.0
/*
 *  Shared Memory Communications over RDMA (SMC-R) and RoCE
 *
 *  IB infrastructure:
 *  Establish SMC-R as an Infiniband Client to be notified about added and
 *  removed IB devices of type RDMA.
 *  Determine device and port characteristics for these IB devices.
 *
 *  Copyright IBM Corp. 2016
 *
 *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
 */

#include <linux/random.h>
#include <linux/workqueue.h>
#include <linux/scatterlist.h>
#include <linux/wait.h>
#include <rdma/ib_verbs.h>
#include <rdma/ib_cache.h>

#include "smc_pnet.h"
#include "smc_ib.h"
#include "smc_core.h"
#include "smc_wr.h"
#include "smc.h"

#define SMC_MAX_CQE 32766	/* max. # of completion queue elements */

#define SMC_QP_MIN_RNR_TIMER		5
#define SMC_QP_TIMEOUT			15 /* 4096 * 2 ** timeout usec */
#define SMC_QP_RETRY_CNT			7 /* 7: infinite */
#define SMC_QP_RNR_RETRY			7 /* 7: infinite */

struct smc_ib_devices smc_ib_devices = {	/* smc-registered ib devices */
	.lock = __SPIN_LOCK_UNLOCKED(smc_ib_devices.lock),
	.list = LIST_HEAD_INIT(smc_ib_devices.list),
};

u8 local_systemid[SMC_SYSTEMID_LEN];		/* unique system identifier */

static int smc_ib_modify_qp_init(struct smc_link *lnk)
{
	struct ib_qp_attr qp_attr;

	memset(&qp_attr, 0, sizeof(qp_attr));
	qp_attr.qp_state = IB_QPS_INIT;
	qp_attr.pkey_index = 0;
	qp_attr.port_num = lnk->ibport;
	qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE
				| IB_ACCESS_REMOTE_WRITE;
	return ib_modify_qp(lnk->roce_qp, &qp_attr,
			    IB_QP_STATE | IB_QP_PKEY_INDEX |
			    IB_QP_ACCESS_FLAGS | IB_QP_PORT);
}

static int smc_ib_modify_qp_rtr(struct smc_link *lnk)
{
	enum ib_qp_attr_mask qp_attr_mask =
		IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN |
		IB_QP_RQ_PSN | IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER;
	struct ib_qp_attr qp_attr;

	memset(&qp_attr, 0, sizeof(qp_attr));
	qp_attr.qp_state = IB_QPS_RTR;
	qp_attr.path_mtu = min(lnk->path_mtu, lnk->peer_mtu);
	qp_attr.ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
	rdma_ah_set_port_num(&qp_attr.ah_attr, lnk->ibport);
	rdma_ah_set_grh(&qp_attr.ah_attr, NULL, 0, lnk->sgid_index, 1, 0);
	rdma_ah_set_dgid_raw(&qp_attr.ah_attr, lnk->peer_gid);
	memcpy(&qp_attr.ah_attr.roce.dmac, lnk->peer_mac,
	       sizeof(lnk->peer_mac));
	qp_attr.dest_qp_num = lnk->peer_qpn;
	qp_attr.rq_psn = lnk->peer_psn; /* starting receive packet seq # */
	qp_attr.max_dest_rd_atomic = 1; /* max # of resources for incoming
					 * requests
					 */
	qp_attr.min_rnr_timer = SMC_QP_MIN_RNR_TIMER;

	return ib_modify_qp(lnk->roce_qp, &qp_attr, qp_attr_mask);
}

int smc_ib_modify_qp_rts(struct smc_link *lnk)
{
	struct ib_qp_attr qp_attr;

	memset(&qp_attr, 0, sizeof(qp_attr));
	qp_attr.qp_state = IB_QPS_RTS;
	qp_attr.timeout = SMC_QP_TIMEOUT;	/* local ack timeout */
	qp_attr.retry_cnt = SMC_QP_RETRY_CNT;	/* retry count */
	qp_attr.rnr_retry = SMC_QP_RNR_RETRY;	/* RNR retries, 7=infinite */
	qp_attr.sq_psn = lnk->psn_initial;	/* starting send packet seq # */
	qp_attr.max_rd_atomic = 1;	/* # of outstanding RDMA reads and
					 * atomic ops allowed
					 */
	return ib_modify_qp(lnk->roce_qp, &qp_attr,
			    IB_QP_STATE | IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
			    IB_QP_SQ_PSN | IB_QP_RNR_RETRY |
			    IB_QP_MAX_QP_RD_ATOMIC);
}

int smc_ib_modify_qp_reset(struct smc_link *lnk)
{
	struct ib_qp_attr qp_attr;

	memset(&qp_attr, 0, sizeof(qp_attr));
	qp_attr.qp_state = IB_QPS_RESET;
	return ib_modify_qp(lnk->roce_qp, &qp_attr, IB_QP_STATE);
}

int smc_ib_ready_link(struct smc_link *lnk)
{
	struct smc_link_group *lgr = smc_get_lgr(lnk);
	int rc = 0;

	rc = smc_ib_modify_qp_init(lnk);
	if (rc)
		goto out;

	rc = smc_ib_modify_qp_rtr(lnk);
	if (rc)
		goto out;
	smc_wr_remember_qp_attr(lnk);
	rc = ib_req_notify_cq(lnk->smcibdev->roce_cq_recv,
			      IB_CQ_SOLICITED_MASK);
	if (rc)
		goto out;
	rc = smc_wr_rx_post_init(lnk);
	if (rc)
		goto out;
	smc_wr_remember_qp_attr(lnk);

	if (lgr->role == SMC_SERV) {
		rc = smc_ib_modify_qp_rts(lnk);
		if (rc)
			goto out;
		smc_wr_remember_qp_attr(lnk);
	}
out:
	return rc;
}

static int smc_ib_fill_mac(struct smc_ib_device *smcibdev, u8 ibport)
{
	const struct ib_gid_attr *attr;
	int rc;

	attr = rdma_get_gid_attr(smcibdev->ibdev, ibport, 0);
	if (IS_ERR(attr))
		return -ENODEV;

	rc = rdma_read_gid_l2_fields(attr, NULL, smcibdev->mac[ibport - 1]);
	rdma_put_gid_attr(attr);
	return rc;
}

/* Create an identifier unique for this instance of SMC-R.
 * The MAC-address of the first active registered IB device
 * plus a random 2-byte number is used to create this identifier.
 * This name is delivered to the peer during connection initialization.
 */
static inline void smc_ib_define_local_systemid(struct smc_ib_device *smcibdev,
						u8 ibport)
{
	memcpy(&local_systemid[2], &smcibdev->mac[ibport - 1],
	       sizeof(smcibdev->mac[ibport - 1]));
}

bool smc_ib_is_valid_local_systemid(void)
{
	return !is_zero_ether_addr(&local_systemid[2]);
}

static void smc_ib_init_local_systemid(void)
{
	get_random_bytes(&local_systemid[0], 2);
}

bool smc_ib_port_active(struct smc_ib_device *smcibdev, u8 ibport)
{
	return smcibdev->pattr[ibport - 1].state == IB_PORT_ACTIVE;
}

/* determine the gid for an ib-device port and vlan id */
int smc_ib_determine_gid(struct smc_ib_device *smcibdev, u8 ibport,
			 unsigned short vlan_id, u8 gid[], u8 *sgid_index)
{
	const struct ib_gid_attr *attr;
	const struct net_device *ndev;
	int i;

	for (i = 0; i < smcibdev->pattr[ibport - 1].gid_tbl_len; i++) {
		attr = rdma_get_gid_attr(smcibdev->ibdev, ibport, i);
		if (IS_ERR(attr))
			continue;

		rcu_read_lock();
		ndev = rdma_read_gid_attr_ndev_rcu(attr);
		if (!IS_ERR(ndev) &&
		    ((!vlan_id && !is_vlan_dev(attr->ndev)) ||
		     (vlan_id && is_vlan_dev(attr->ndev) &&
		      vlan_dev_vlan_id(attr->ndev) == vlan_id)) &&
		    attr->gid_type == IB_GID_TYPE_ROCE) {
			rcu_read_unlock();
			if (gid)
				memcpy(gid, &attr->gid, SMC_GID_SIZE);
			if (sgid_index)
				*sgid_index = attr->index;
			rdma_put_gid_attr(attr);
			return 0;
		}
		rcu_read_unlock();
		rdma_put_gid_attr(attr);
	}
	return -ENODEV;
}

static int smc_ib_remember_port_attr(struct smc_ib_device *smcibdev, u8 ibport)
{
	int rc;

	memset(&smcibdev->pattr[ibport - 1], 0,
	       sizeof(smcibdev->pattr[ibport - 1]));
	rc = ib_query_port(smcibdev->ibdev, ibport,
			   &smcibdev->pattr[ibport - 1]);
	if (rc)
		goto out;
	/* the SMC protocol requires specification of the RoCE MAC address */
	rc = smc_ib_fill_mac(smcibdev, ibport);
	if (rc)
		goto out;
	if (!smc_ib_is_valid_local_systemid() &&
	    smc_ib_port_active(smcibdev, ibport))
		/* create unique system identifier */
		smc_ib_define_local_systemid(smcibdev, ibport);
out:
	return rc;
}

/* process context wrapper for might_sleep smc_ib_remember_port_attr */
static void smc_ib_port_event_work(struct work_struct *work)
{
	struct smc_ib_device *smcibdev = container_of(
		work, struct smc_ib_device, port_event_work);
	u8 port_idx;

	for_each_set_bit(port_idx, &smcibdev->port_event_mask, SMC_MAX_PORTS) {
		smc_ib_remember_port_attr(smcibdev, port_idx + 1);
		clear_bit(port_idx, &smcibdev->port_event_mask);
		if (!smc_ib_port_active(smcibdev, port_idx + 1)) {
			set_bit(port_idx, smcibdev->ports_going_away);
			smc_port_terminate(smcibdev, port_idx + 1);
		} else {
			clear_bit(port_idx, smcibdev->ports_going_away);
		}
	}
}

/* can be called in IRQ context */
static void smc_ib_global_event_handler(struct ib_event_handler *handler,
					struct ib_event *ibevent)
{
	struct smc_ib_device *smcibdev;
	bool schedule = false;
	u8 port_idx;

	smcibdev = container_of(handler, struct smc_ib_device, event_handler);

	switch (ibevent->event) {
	case IB_EVENT_DEVICE_FATAL:
		/* terminate all ports on device */
		for (port_idx = 0; port_idx < SMC_MAX_PORTS; port_idx++) {
			set_bit(port_idx, &smcibdev->port_event_mask);
			if (!test_and_set_bit(port_idx,
					      smcibdev->ports_going_away))
				schedule = true;
		}
		if (schedule)
			schedule_work(&smcibdev->port_event_work);
		break;
	case IB_EVENT_PORT_ACTIVE:
		port_idx = ibevent->element.port_num - 1;
		if (port_idx >= SMC_MAX_PORTS)
			break;
		set_bit(port_idx, &smcibdev->port_event_mask);
		if (test_and_clear_bit(port_idx, smcibdev->ports_going_away))
			schedule_work(&smcibdev->port_event_work);
		break;
	case IB_EVENT_PORT_ERR:
		port_idx = ibevent->element.port_num - 1;
		if (port_idx >= SMC_MAX_PORTS)
			break;
		set_bit(port_idx, &smcibdev->port_event_mask);
		if (!test_and_set_bit(port_idx, smcibdev->ports_going_away))
			schedule_work(&smcibdev->port_event_work);
		break;
	case IB_EVENT_GID_CHANGE:
		port_idx = ibevent->element.port_num - 1;
		if (port_idx >= SMC_MAX_PORTS)
			break;
		set_bit(port_idx, &smcibdev->port_event_mask);
		schedule_work(&smcibdev->port_event_work);
		break;
	default:
		break;
	}
}

void smc_ib_dealloc_protection_domain(struct smc_link *lnk)
{
	if (lnk->roce_pd)
		ib_dealloc_pd(lnk->roce_pd);
	lnk->roce_pd = NULL;
}

int smc_ib_create_protection_domain(struct smc_link *lnk)
{
	int rc;

	lnk->roce_pd = ib_alloc_pd(lnk->smcibdev->ibdev, 0);
	rc = PTR_ERR_OR_ZERO(lnk->roce_pd);
	if (IS_ERR(lnk->roce_pd))
		lnk->roce_pd = NULL;
	return rc;
}

static void smc_ib_qp_event_handler(struct ib_event *ibevent, void *priv)
{
	struct smc_link *lnk = (struct smc_link *)priv;
	struct smc_ib_device *smcibdev = lnk->smcibdev;
	u8 port_idx;

	switch (ibevent->event) {
	case IB_EVENT_QP_FATAL:
	case IB_EVENT_QP_ACCESS_ERR:
		port_idx = ibevent->element.qp->port - 1;
		if (port_idx >= SMC_MAX_PORTS)
			break;
		set_bit(port_idx, &smcibdev->port_event_mask);
		if (!test_and_set_bit(port_idx, smcibdev->ports_going_away))
			schedule_work(&smcibdev->port_event_work);
		break;
	default:
		break;
	}
}

void smc_ib_destroy_queue_pair(struct smc_link *lnk)
{
	if (lnk->roce_qp)
		ib_destroy_qp(lnk->roce_qp);
	lnk->roce_qp = NULL;
}

/* create a queue pair within the protection domain for a link */
int smc_ib_create_queue_pair(struct smc_link *lnk)
{
	struct ib_qp_init_attr qp_attr = {
		.event_handler = smc_ib_qp_event_handler,
		.qp_context = lnk,
		.send_cq = lnk->smcibdev->roce_cq_send,
		.recv_cq = lnk->smcibdev->roce_cq_recv,
		.srq = NULL,
		.cap = {
				/* include unsolicited rdma_writes as well,
				 * there are max. 2 RDMA_WRITE per 1 WR_SEND
				 */
			.max_send_wr = SMC_WR_BUF_CNT * 3,
			.max_recv_wr = SMC_WR_BUF_CNT * 3,
			.max_send_sge = SMC_IB_MAX_SEND_SGE,
			.max_recv_sge = 1,
		},
		.sq_sig_type = IB_SIGNAL_REQ_WR,
		.qp_type = IB_QPT_RC,
	};
	int rc;

	lnk->roce_qp = ib_create_qp(lnk->roce_pd, &qp_attr);
	rc = PTR_ERR_OR_ZERO(lnk->roce_qp);
	if (IS_ERR(lnk->roce_qp))
		lnk->roce_qp = NULL;
	else
		smc_wr_remember_qp_attr(lnk);
	return rc;
}

void smc_ib_put_memory_region(struct ib_mr *mr)
{
	ib_dereg_mr(mr);
}

static int smc_ib_map_mr_sg(struct smc_buf_desc *buf_slot)
{
	unsigned int offset = 0;
	int sg_num;

	/* map the largest prefix of a dma mapped SG list */
	sg_num = ib_map_mr_sg(buf_slot->mr_rx[SMC_SINGLE_LINK],
			      buf_slot->sgt[SMC_SINGLE_LINK].sgl,
			      buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
			      &offset, PAGE_SIZE);

	return sg_num;
}

/* Allocate a memory region and map the dma mapped SG list of buf_slot */
int smc_ib_get_memory_region(struct ib_pd *pd, int access_flags,
			     struct smc_buf_desc *buf_slot)
{
	if (buf_slot->mr_rx[SMC_SINGLE_LINK])
		return 0; /* already done */

	buf_slot->mr_rx[SMC_SINGLE_LINK] =
		ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, 1 << buf_slot->order);
	if (IS_ERR(buf_slot->mr_rx[SMC_SINGLE_LINK])) {
		int rc;

		rc = PTR_ERR(buf_slot->mr_rx[SMC_SINGLE_LINK]);
		buf_slot->mr_rx[SMC_SINGLE_LINK] = NULL;
		return rc;
	}

	if (smc_ib_map_mr_sg(buf_slot) != 1)
		return -EINVAL;

	return 0;
}

/* synchronize buffer usage for cpu access */
void smc_ib_sync_sg_for_cpu(struct smc_ib_device *smcibdev,
			    struct smc_buf_desc *buf_slot,
			    enum dma_data_direction data_direction)
{
	struct scatterlist *sg;
	unsigned int i;

	/* for now there is just one DMA address */
	for_each_sg(buf_slot->sgt[SMC_SINGLE_LINK].sgl, sg,
		    buf_slot->sgt[SMC_SINGLE_LINK].nents, i) {
		if (!sg_dma_len(sg))
			break;
		ib_dma_sync_single_for_cpu(smcibdev->ibdev,
					   sg_dma_address(sg),
					   sg_dma_len(sg),
					   data_direction);
	}
}

/* synchronize buffer usage for device access */
void smc_ib_sync_sg_for_device(struct smc_ib_device *smcibdev,
			       struct smc_buf_desc *buf_slot,
			       enum dma_data_direction data_direction)
{
	struct scatterlist *sg;
	unsigned int i;

	/* for now there is just one DMA address */
	for_each_sg(buf_slot->sgt[SMC_SINGLE_LINK].sgl, sg,
		    buf_slot->sgt[SMC_SINGLE_LINK].nents, i) {
		if (!sg_dma_len(sg))
			break;
		ib_dma_sync_single_for_device(smcibdev->ibdev,
					      sg_dma_address(sg),
					      sg_dma_len(sg),
					      data_direction);
	}
}

/* Map a new TX or RX buffer SG-table to DMA */
int smc_ib_buf_map_sg(struct smc_ib_device *smcibdev,
		      struct smc_buf_desc *buf_slot,
		      enum dma_data_direction data_direction)
{
	int mapped_nents;

	mapped_nents = ib_dma_map_sg(smcibdev->ibdev,
				     buf_slot->sgt[SMC_SINGLE_LINK].sgl,
				     buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
				     data_direction);
	if (!mapped_nents)
		return -ENOMEM;

	return mapped_nents;
}

void smc_ib_buf_unmap_sg(struct smc_ib_device *smcibdev,
			 struct smc_buf_desc *buf_slot,
			 enum dma_data_direction data_direction)
{
	if (!buf_slot->sgt[SMC_SINGLE_LINK].sgl->dma_address)
		return; /* already unmapped */

	ib_dma_unmap_sg(smcibdev->ibdev,
			buf_slot->sgt[SMC_SINGLE_LINK].sgl,
			buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
			data_direction);
	buf_slot->sgt[SMC_SINGLE_LINK].sgl->dma_address = 0;
}

long smc_ib_setup_per_ibdev(struct smc_ib_device *smcibdev)
{
	struct ib_cq_init_attr cqattr =	{
		.cqe = SMC_MAX_CQE, .comp_vector = 0 };
	int cqe_size_order, smc_order;
	long rc;

	/* the calculated number of cq entries fits to mlx5 cq allocation */
	cqe_size_order = cache_line_size() == 128 ? 7 : 6;
	smc_order = MAX_ORDER - cqe_size_order - 1;
	if (SMC_MAX_CQE + 2 > (0x00000001 << smc_order) * PAGE_SIZE)
		cqattr.cqe = (0x00000001 << smc_order) * PAGE_SIZE - 2;
	smcibdev->roce_cq_send = ib_create_cq(smcibdev->ibdev,
					      smc_wr_tx_cq_handler, NULL,
					      smcibdev, &cqattr);
	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_send);
	if (IS_ERR(smcibdev->roce_cq_send)) {
		smcibdev->roce_cq_send = NULL;
		return rc;
	}
	smcibdev->roce_cq_recv = ib_create_cq(smcibdev->ibdev,
					      smc_wr_rx_cq_handler, NULL,
					      smcibdev, &cqattr);
	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_recv);
	if (IS_ERR(smcibdev->roce_cq_recv)) {
		smcibdev->roce_cq_recv = NULL;
		goto err;
	}
	smc_wr_add_dev(smcibdev);
	smcibdev->initialized = 1;
	return rc;

err:
	ib_destroy_cq(smcibdev->roce_cq_send);
	return rc;
}

static void smc_ib_cleanup_per_ibdev(struct smc_ib_device *smcibdev)
{
	if (!smcibdev->initialized)
		return;
	smcibdev->initialized = 0;
	ib_destroy_cq(smcibdev->roce_cq_recv);
	ib_destroy_cq(smcibdev->roce_cq_send);
	smc_wr_remove_dev(smcibdev);
}

static struct ib_client smc_ib_client;

/* callback function for ib_register_client() */
static void smc_ib_add_dev(struct ib_device *ibdev)
{
	struct smc_ib_device *smcibdev;
	u8 port_cnt;
	int i;

	if (ibdev->node_type != RDMA_NODE_IB_CA)
		return;

	smcibdev = kzalloc(sizeof(*smcibdev), GFP_KERNEL);
	if (!smcibdev)
		return;

	smcibdev->ibdev = ibdev;
	INIT_WORK(&smcibdev->port_event_work, smc_ib_port_event_work);
	atomic_set(&smcibdev->lnk_cnt, 0);
	init_waitqueue_head(&smcibdev->lnks_deleted);
	spin_lock(&smc_ib_devices.lock);
	list_add_tail(&smcibdev->list, &smc_ib_devices.list);
	spin_unlock(&smc_ib_devices.lock);
	ib_set_client_data(ibdev, &smc_ib_client, smcibdev);
	INIT_IB_EVENT_HANDLER(&smcibdev->event_handler, smcibdev->ibdev,
			      smc_ib_global_event_handler);
	ib_register_event_handler(&smcibdev->event_handler);

	/* trigger reading of the port attributes */
	port_cnt = smcibdev->ibdev->phys_port_cnt;
	for (i = 0;
	     i < min_t(size_t, port_cnt, SMC_MAX_PORTS);
	     i++) {
		set_bit(i, &smcibdev->port_event_mask);
		/* determine pnetids of the port */
		smc_pnetid_by_dev_port(ibdev->dev.parent, i,
				       smcibdev->pnetid[i]);
	}
	schedule_work(&smcibdev->port_event_work);
}

/* callback function for ib_unregister_client() */
static void smc_ib_remove_dev(struct ib_device *ibdev, void *client_data)
{
	struct smc_ib_device *smcibdev;

	smcibdev = ib_get_client_data(ibdev, &smc_ib_client);
	if (!smcibdev || smcibdev->ibdev != ibdev)
		return;
	ib_set_client_data(ibdev, &smc_ib_client, NULL);
	spin_lock(&smc_ib_devices.lock);
	list_del_init(&smcibdev->list); /* remove from smc_ib_devices */
	spin_unlock(&smc_ib_devices.lock);
	smc_smcr_terminate_all(smcibdev);
	smc_ib_cleanup_per_ibdev(smcibdev);
	ib_unregister_event_handler(&smcibdev->event_handler);
	cancel_work_sync(&smcibdev->port_event_work);
	kfree(smcibdev);
}

static struct ib_client smc_ib_client = {
	.name	= "smc_ib",
	.add	= smc_ib_add_dev,
	.remove = smc_ib_remove_dev,
};

int __init smc_ib_register_client(void)
{
	smc_ib_init_local_systemid();
	return ib_register_client(&smc_ib_client);
}

void smc_ib_unregister_client(void)
{
	ib_unregister_client(&smc_ib_client);
}
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
a4cf0443 UB	2	/*
	3	* Shared Memory Communications over RDMA (SMC-R) and RoCE
	4	*
	5	* IB infrastructure:
	6	* Establish SMC-R as an Infiniband Client to be notified about added and
	7	* removed IB devices of type RDMA.
	8	* Determine device and port characteristics for these IB devices.
	9	*
	10	* Copyright IBM Corp. 2016
	11	*
	12	* Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
	13	*/
	14
	15	#include <linux/random.h>
bd4ad577	16	#include <linux/workqueue.h>
10428dd8	17	#include <linux/scatterlist.h>
6dabd405	18	#include <linux/wait.h>
a4cf0443	19	#include <rdma/ib_verbs.h>
ddb457c6	20	#include <rdma/ib_cache.h>
a4cf0443	21
6812baab	22	#include "smc_pnet.h"
a4cf0443	23	#include "smc_ib.h"
cd6851f3	24	#include "smc_core.h"
f38ba179	25	#include "smc_wr.h"
a4cf0443 UB	26	#include "smc.h"
a4cf0443 UB	27
c9f4c6cf UB	28	#define SMC_MAX_CQE 32766 /* max. # of completion queue elements */
c9f4c6cf UB	29
bd4ad577 UB	30	#define SMC_QP_MIN_RNR_TIMER 5
	31	#define SMC_QP_TIMEOUT 15 /* 4096 * 2 ** timeout usec */
	32	#define SMC_QP_RETRY_CNT 7 /* 7: infinite */
	33	#define SMC_QP_RNR_RETRY 7 /* 7: infinite */
	34
a4cf0443 UB	35	struct smc_ib_devices smc_ib_devices = { /* smc-registered ib devices */
	36	.lock = __SPIN_LOCK_UNLOCKED(smc_ib_devices.lock),
	37	.list = LIST_HEAD_INIT(smc_ib_devices.list),
	38	};
	39
366bb249	40	u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
a4cf0443	41
bd4ad577 UB	42	static int smc_ib_modify_qp_init(struct smc_link *lnk)
	43	{
	44	struct ib_qp_attr qp_attr;
	45
	46	memset(&qp_attr, 0, sizeof(qp_attr));
	47	qp_attr.qp_state = IB_QPS_INIT;
	48	qp_attr.pkey_index = 0;
	49	qp_attr.port_num = lnk->ibport;
	50	qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE
	51	\| IB_ACCESS_REMOTE_WRITE;
	52	return ib_modify_qp(lnk->roce_qp, &qp_attr,
	53	IB_QP_STATE \| IB_QP_PKEY_INDEX \|
	54	IB_QP_ACCESS_FLAGS \| IB_QP_PORT);
	55	}
	56
	57	static int smc_ib_modify_qp_rtr(struct smc_link *lnk)
	58	{
	59	enum ib_qp_attr_mask qp_attr_mask =
	60	IB_QP_STATE \| IB_QP_AV \| IB_QP_PATH_MTU \| IB_QP_DEST_QPN \|
	61	IB_QP_RQ_PSN \| IB_QP_MAX_DEST_RD_ATOMIC \| IB_QP_MIN_RNR_TIMER;
	62	struct ib_qp_attr qp_attr;
	63
	64	memset(&qp_attr, 0, sizeof(qp_attr));
	65	qp_attr.qp_state = IB_QPS_RTR;
	66	qp_attr.path_mtu = min(lnk->path_mtu, lnk->peer_mtu);
44c58487	67	qp_attr.ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
d8966fcd	68	rdma_ah_set_port_num(&qp_attr.ah_attr, lnk->ibport);
7005ada6	69	rdma_ah_set_grh(&qp_attr.ah_attr, NULL, 0, lnk->sgid_index, 1, 0);
d8966fcd	70	rdma_ah_set_dgid_raw(&qp_attr.ah_attr, lnk->peer_gid);
44c58487	71	memcpy(&qp_attr.ah_attr.roce.dmac, lnk->peer_mac,
bd4ad577 UB	72	sizeof(lnk->peer_mac));
	73	qp_attr.dest_qp_num = lnk->peer_qpn;
	74	qp_attr.rq_psn = lnk->peer_psn; /* starting receive packet seq # */
	75	qp_attr.max_dest_rd_atomic = 1; /* max # of resources for incoming
	76	* requests
	77	*/
	78	qp_attr.min_rnr_timer = SMC_QP_MIN_RNR_TIMER;
	79
	80	return ib_modify_qp(lnk->roce_qp, &qp_attr, qp_attr_mask);
	81	}
	82
	83	int smc_ib_modify_qp_rts(struct smc_link *lnk)
	84	{
	85	struct ib_qp_attr qp_attr;
	86
	87	memset(&qp_attr, 0, sizeof(qp_attr));
	88	qp_attr.qp_state = IB_QPS_RTS;
	89	qp_attr.timeout = SMC_QP_TIMEOUT; /* local ack timeout */
	90	qp_attr.retry_cnt = SMC_QP_RETRY_CNT; /* retry count */
	91	qp_attr.rnr_retry = SMC_QP_RNR_RETRY; /* RNR retries, 7=infinite */
	92	qp_attr.sq_psn = lnk->psn_initial; /* starting send packet seq # */
	93	qp_attr.max_rd_atomic = 1; /* # of outstanding RDMA reads and
	94	* atomic ops allowed
	95	*/
	96	return ib_modify_qp(lnk->roce_qp, &qp_attr,
	97	IB_QP_STATE \| IB_QP_TIMEOUT \| IB_QP_RETRY_CNT \|
	98	IB_QP_SQ_PSN \| IB_QP_RNR_RETRY \|
	99	IB_QP_MAX_QP_RD_ATOMIC);
	100	}
	101
	102	int smc_ib_modify_qp_reset(struct smc_link *lnk)
	103	{
	104	struct ib_qp_attr qp_attr;
	105
	106	memset(&qp_attr, 0, sizeof(qp_attr));
	107	qp_attr.qp_state = IB_QPS_RESET;
	108	return ib_modify_qp(lnk->roce_qp, &qp_attr, IB_QP_STATE);
	109	}
	110
	111	int smc_ib_ready_link(struct smc_link *lnk)
	112	{
00e5fb26	113	struct smc_link_group *lgr = smc_get_lgr(lnk);
bd4ad577 UB	114	int rc = 0;
	115
	116	rc = smc_ib_modify_qp_init(lnk);
	117	if (rc)
	118	goto out;
	119
	120	rc = smc_ib_modify_qp_rtr(lnk);
	121	if (rc)
	122	goto out;
	123	smc_wr_remember_qp_attr(lnk);
	124	rc = ib_req_notify_cq(lnk->smcibdev->roce_cq_recv,
	125	IB_CQ_SOLICITED_MASK);
	126	if (rc)
	127	goto out;
	128	rc = smc_wr_rx_post_init(lnk);
	129	if (rc)
	130	goto out;
	131	smc_wr_remember_qp_attr(lnk);
	132
	133	if (lgr->role == SMC_SERV) {
	134	rc = smc_ib_modify_qp_rts(lnk);
	135	if (rc)
	136	goto out;
	137	smc_wr_remember_qp_attr(lnk);
	138	}
	139	out:
	140	return rc;
	141	}
	142
7005ada6	143	static int smc_ib_fill_mac(struct smc_ib_device *smcibdev, u8 ibport)
be6a3f38	144	{
b4c296f9	145	const struct ib_gid_attr *attr;
5102eca9	146	int rc;
be6a3f38	147
b4c296f9 JG	148	attr = rdma_get_gid_attr(smcibdev->ibdev, ibport, 0);
b4c296f9 JG	149	if (IS_ERR(attr))
be6a3f38 UB	150	return -ENODEV;
be6a3f38 UB	151
5102eca9	152	rc = rdma_read_gid_l2_fields(attr, NULL, smcibdev->mac[ibport - 1]);
b4c296f9 JG	153	rdma_put_gid_attr(attr);
b4c296f9 JG	154	return rc;
be6a3f38 UB	155	}
	156
	157	/* Create an identifier unique for this instance of SMC-R.
	158	* The MAC-address of the first active registered IB device
	159	* plus a random 2-byte number is used to create this identifier.
	160	* This name is delivered to the peer during connection initialization.
	161	*/
	162	static inline void smc_ib_define_local_systemid(struct smc_ib_device *smcibdev,
	163	u8 ibport)
	164	{
	165	memcpy(&local_systemid[2], &smcibdev->mac[ibport - 1],
	166	sizeof(smcibdev->mac[ibport - 1]));
366bb249 HW	167	}
366bb249 HW	168
a082ec89	169	bool smc_ib_is_valid_local_systemid(void)
366bb249 HW	170	{
	171	return !is_zero_ether_addr(&local_systemid[2]);
	172	}
	173
	174	static void smc_ib_init_local_systemid(void)
	175	{
be6a3f38 UB	176	get_random_bytes(&local_systemid[0], 2);
	177	}
	178
	179	bool smc_ib_port_active(struct smc_ib_device *smcibdev, u8 ibport)
	180	{
	181	return smcibdev->pattr[ibport - 1].state == IB_PORT_ACTIVE;
	182	}
	183
7005ada6 UB	184	/* determine the gid for an ib-device port and vlan id */
	185	int smc_ib_determine_gid(struct smc_ib_device *smcibdev, u8 ibport,
	186	unsigned short vlan_id, u8 gid[], u8 *sgid_index)
	187	{
b4c296f9	188	const struct ib_gid_attr *attr;
5102eca9	189	const struct net_device *ndev;
7005ada6 UB	190	int i;
	191
	192	for (i = 0; i < smcibdev->pattr[ibport - 1].gid_tbl_len; i++) {
b4c296f9 JG	193	attr = rdma_get_gid_attr(smcibdev->ibdev, ibport, i);
b4c296f9 JG	194	if (IS_ERR(attr))
7005ada6	195	continue;
b4c296f9	196
5102eca9 PP	197	rcu_read_lock();
	198	ndev = rdma_read_gid_attr_ndev_rcu(attr);
	199	if (!IS_ERR(ndev) &&
b4c296f9 JG	200	((!vlan_id && !is_vlan_dev(attr->ndev)) \|\|
	201	(vlan_id && is_vlan_dev(attr->ndev) &&
	202	vlan_dev_vlan_id(attr->ndev) == vlan_id)) &&
	203	attr->gid_type == IB_GID_TYPE_ROCE) {
5102eca9	204	rcu_read_unlock();
7005ada6	205	if (gid)
b4c296f9	206	memcpy(gid, &attr->gid, SMC_GID_SIZE);
7005ada6	207	if (sgid_index)
b4c296f9 JG	208	*sgid_index = attr->index;
b4c296f9 JG	209	rdma_put_gid_attr(attr);
7005ada6 UB	210	return 0;
7005ada6 UB	211	}
5102eca9	212	rcu_read_unlock();
b4c296f9	213	rdma_put_gid_attr(attr);
7005ada6 UB	214	}
	215	return -ENODEV;
	216	}
	217
be6a3f38 UB	218	static int smc_ib_remember_port_attr(struct smc_ib_device *smcibdev, u8 ibport)
	219	{
	220	int rc;
	221
	222	memset(&smcibdev->pattr[ibport - 1], 0,
	223	sizeof(smcibdev->pattr[ibport - 1]));
	224	rc = ib_query_port(smcibdev->ibdev, ibport,
	225	&smcibdev->pattr[ibport - 1]);
	226	if (rc)
	227	goto out;
	228	/* the SMC protocol requires specification of the RoCE MAC address */
7005ada6	229	rc = smc_ib_fill_mac(smcibdev, ibport);
be6a3f38 UB	230	if (rc)
be6a3f38 UB	231	goto out;
366bb249	232	if (!smc_ib_is_valid_local_systemid() &&
be6a3f38 UB	233	smc_ib_port_active(smcibdev, ibport))
	234	/* create unique system identifier */
	235	smc_ib_define_local_systemid(smcibdev, ibport);
	236	out:
	237	return rc;
	238	}
	239
bd4ad577 UB	240	/* process context wrapper for might_sleep smc_ib_remember_port_attr */
	241	static void smc_ib_port_event_work(struct work_struct *work)
	242	{
	243	struct smc_ib_device *smcibdev = container_of(
	244	work, struct smc_ib_device, port_event_work);
	245	u8 port_idx;
	246
	247	for_each_set_bit(port_idx, &smcibdev->port_event_mask, SMC_MAX_PORTS) {
	248	smc_ib_remember_port_attr(smcibdev, port_idx + 1);
	249	clear_bit(port_idx, &smcibdev->port_event_mask);
c3d9494e UB	250	if (!smc_ib_port_active(smcibdev, port_idx + 1)) {
c3d9494e UB	251	set_bit(port_idx, smcibdev->ports_going_away);
9fda3510	252	smc_port_terminate(smcibdev, port_idx + 1);
c3d9494e UB	253	} else {
	254	clear_bit(port_idx, smcibdev->ports_going_away);
	255	}
bd4ad577 UB	256	}
	257	}
	258
	259	/* can be called in IRQ context */
	260	static void smc_ib_global_event_handler(struct ib_event_handler *handler,
	261	struct ib_event *ibevent)
	262	{
	263	struct smc_ib_device *smcibdev;
5613f20c	264	bool schedule = false;
bd4ad577 UB	265	u8 port_idx;
	266
	267	smcibdev = container_of(handler, struct smc_ib_device, event_handler);
bd4ad577 UB	268
bd4ad577 UB	269	switch (ibevent->event) {
bd4ad577	270	case IB_EVENT_DEVICE_FATAL:
81cf6430	271	/* terminate all ports on device */
c3d9494e	272	for (port_idx = 0; port_idx < SMC_MAX_PORTS; port_idx++) {
81cf6430	273	set_bit(port_idx, &smcibdev->port_event_mask);
5613f20c UB	274	if (!test_and_set_bit(port_idx,
	275	smcibdev->ports_going_away))
	276	schedule = true;
c3d9494e	277	}
5613f20c UB	278	if (schedule)
5613f20c UB	279	schedule_work(&smcibdev->port_event_work);
81cf6430	280	break;
bd4ad577 UB	281	case IB_EVENT_PORT_ACTIVE:
bd4ad577 UB	282	port_idx = ibevent->element.port_num - 1;
5613f20c UB	283	if (port_idx >= SMC_MAX_PORTS)
	284	break;
	285	set_bit(port_idx, &smcibdev->port_event_mask);
	286	if (test_and_clear_bit(port_idx, smcibdev->ports_going_away))
	287	schedule_work(&smcibdev->port_event_work);
	288	break;
	289	case IB_EVENT_PORT_ERR:
	290	port_idx = ibevent->element.port_num - 1;
	291	if (port_idx >= SMC_MAX_PORTS)
	292	break;
	293	set_bit(port_idx, &smcibdev->port_event_mask);
	294	if (!test_and_set_bit(port_idx, smcibdev->ports_going_away))
81cf6430	295	schedule_work(&smcibdev->port_event_work);
5613f20c UB	296	break;
	297	case IB_EVENT_GID_CHANGE:
	298	port_idx = ibevent->element.port_num - 1;
	299	if (port_idx >= SMC_MAX_PORTS)
	300	break;
	301	set_bit(port_idx, &smcibdev->port_event_mask);
	302	schedule_work(&smcibdev->port_event_work);
bd4ad577 UB	303	break;
	304	default:
	305	break;
	306	}
	307	}
	308
f38ba179 UB	309	void smc_ib_dealloc_protection_domain(struct smc_link *lnk)
f38ba179 UB	310	{
da05bf29 UB	311	if (lnk->roce_pd)
da05bf29 UB	312	ib_dealloc_pd(lnk->roce_pd);
f38ba179 UB	313	lnk->roce_pd = NULL;
	314	}
	315
	316	int smc_ib_create_protection_domain(struct smc_link *lnk)
	317	{
	318	int rc;
	319
897e1c24	320	lnk->roce_pd = ib_alloc_pd(lnk->smcibdev->ibdev, 0);
f38ba179 UB	321	rc = PTR_ERR_OR_ZERO(lnk->roce_pd);
	322	if (IS_ERR(lnk->roce_pd))
	323	lnk->roce_pd = NULL;
	324	return rc;
	325	}
	326
	327	static void smc_ib_qp_event_handler(struct ib_event ibevent, void priv)
	328	{
e5f3aa04 KG	329	struct smc_link lnk = (struct smc_link )priv;
e5f3aa04 KG	330	struct smc_ib_device *smcibdev = lnk->smcibdev;
da05bf29 UB	331	u8 port_idx;
da05bf29 UB	332
f38ba179	333	switch (ibevent->event) {
81cf6430	334	case IB_EVENT_QP_FATAL:
f38ba179	335	case IB_EVENT_QP_ACCESS_ERR:
e5f3aa04	336	port_idx = ibevent->element.qp->port - 1;
5613f20c UB	337	if (port_idx >= SMC_MAX_PORTS)
	338	break;
	339	set_bit(port_idx, &smcibdev->port_event_mask);
	340	if (!test_and_set_bit(port_idx, smcibdev->ports_going_away))
81cf6430	341	schedule_work(&smcibdev->port_event_work);
f38ba179 UB	342	break;
	343	default:
	344	break;
	345	}
	346	}
	347
	348	void smc_ib_destroy_queue_pair(struct smc_link *lnk)
	349	{
da05bf29 UB	350	if (lnk->roce_qp)
da05bf29 UB	351	ib_destroy_qp(lnk->roce_qp);
f38ba179 UB	352	lnk->roce_qp = NULL;
	353	}
	354
	355	/* create a queue pair within the protection domain for a link */
	356	int smc_ib_create_queue_pair(struct smc_link *lnk)
	357	{
	358	struct ib_qp_init_attr qp_attr = {
	359	.event_handler = smc_ib_qp_event_handler,
	360	.qp_context = lnk,
	361	.send_cq = lnk->smcibdev->roce_cq_send,
	362	.recv_cq = lnk->smcibdev->roce_cq_recv,
	363	.srq = NULL,
	364	.cap = {
f38ba179 UB	365	/* include unsolicited rdma_writes as well,
	366	* there are max. 2 RDMA_WRITE per 1 WR_SEND
	367	*/
652a1e41	368	.max_send_wr = SMC_WR_BUF_CNT * 3,
f38ba179 UB	369	.max_recv_wr = SMC_WR_BUF_CNT * 3,
	370	.max_send_sge = SMC_IB_MAX_SEND_SGE,
	371	.max_recv_sge = 1,
f38ba179 UB	372	},
	373	.sq_sig_type = IB_SIGNAL_REQ_WR,
	374	.qp_type = IB_QPT_RC,
	375	};
	376	int rc;
	377
	378	lnk->roce_qp = ib_create_qp(lnk->roce_pd, &qp_attr);
	379	rc = PTR_ERR_OR_ZERO(lnk->roce_qp);
	380	if (IS_ERR(lnk->roce_qp))
	381	lnk->roce_qp = NULL;
	382	else
	383	smc_wr_remember_qp_attr(lnk);
	384	return rc;
	385	}
	386
897e1c24 UB	387	void smc_ib_put_memory_region(struct ib_mr *mr)
	388	{
	389	ib_dereg_mr(mr);
	390	}
	391
	392	static int smc_ib_map_mr_sg(struct smc_buf_desc *buf_slot)
	393	{
	394	unsigned int offset = 0;
	395	int sg_num;
	396
	397	/* map the largest prefix of a dma mapped SG list */
	398	sg_num = ib_map_mr_sg(buf_slot->mr_rx[SMC_SINGLE_LINK],
	399	buf_slot->sgt[SMC_SINGLE_LINK].sgl,
	400	buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
	401	&offset, PAGE_SIZE);
	402
	403	return sg_num;
	404	}
	405
	406	/* Allocate a memory region and map the dma mapped SG list of buf_slot */
	407	int smc_ib_get_memory_region(struct ib_pd *pd, int access_flags,
	408	struct smc_buf_desc *buf_slot)
	409	{
	410	if (buf_slot->mr_rx[SMC_SINGLE_LINK])
	411	return 0; /* already done */
	412
	413	buf_slot->mr_rx[SMC_SINGLE_LINK] =
	414	ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, 1 << buf_slot->order);
	415	if (IS_ERR(buf_slot->mr_rx[SMC_SINGLE_LINK])) {
	416	int rc;
	417
	418	rc = PTR_ERR(buf_slot->mr_rx[SMC_SINGLE_LINK]);
	419	buf_slot->mr_rx[SMC_SINGLE_LINK] = NULL;
	420	return rc;
	421	}
	422
	423	if (smc_ib_map_mr_sg(buf_slot) != 1)
	424	return -EINVAL;
	425
	426	return 0;
	427	}
	428
10428dd8 UB	429	/* synchronize buffer usage for cpu access */
	430	void smc_ib_sync_sg_for_cpu(struct smc_ib_device *smcibdev,
	431	struct smc_buf_desc *buf_slot,
	432	enum dma_data_direction data_direction)
	433	{
	434	struct scatterlist *sg;
	435	unsigned int i;
	436
	437	/* for now there is just one DMA address */
	438	for_each_sg(buf_slot->sgt[SMC_SINGLE_LINK].sgl, sg,
	439	buf_slot->sgt[SMC_SINGLE_LINK].nents, i) {
	440	if (!sg_dma_len(sg))
	441	break;
	442	ib_dma_sync_single_for_cpu(smcibdev->ibdev,
	443	sg_dma_address(sg),
	444	sg_dma_len(sg),
	445	data_direction);
	446	}
	447	}
	448
	449	/* synchronize buffer usage for device access */
	450	void smc_ib_sync_sg_for_device(struct smc_ib_device *smcibdev,
	451	struct smc_buf_desc *buf_slot,
	452	enum dma_data_direction data_direction)
	453	{
	454	struct scatterlist *sg;
	455	unsigned int i;
	456
	457	/* for now there is just one DMA address */
	458	for_each_sg(buf_slot->sgt[SMC_SINGLE_LINK].sgl, sg,
	459	buf_slot->sgt[SMC_SINGLE_LINK].nents, i) {
	460	if (!sg_dma_len(sg))
	461	break;
	462	ib_dma_sync_single_for_device(smcibdev->ibdev,
	463	sg_dma_address(sg),
	464	sg_dma_len(sg),
	465	data_direction);
	466	}
	467	}
	468
a3fe3d01 UB	469	/* Map a new TX or RX buffer SG-table to DMA */
	470	int smc_ib_buf_map_sg(struct smc_ib_device *smcibdev,
	471	struct smc_buf_desc *buf_slot,
	472	enum dma_data_direction data_direction)
	473	{
	474	int mapped_nents;
	475
	476	mapped_nents = ib_dma_map_sg(smcibdev->ibdev,
	477	buf_slot->sgt[SMC_SINGLE_LINK].sgl,
	478	buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
	479	data_direction);
	480	if (!mapped_nents)
	481	return -ENOMEM;
	482
	483	return mapped_nents;
	484	}
	485
	486	void smc_ib_buf_unmap_sg(struct smc_ib_device *smcibdev,
	487	struct smc_buf_desc *buf_slot,
	488	enum dma_data_direction data_direction)
	489	{
	490	if (!buf_slot->sgt[SMC_SINGLE_LINK].sgl->dma_address)
	491	return; /* already unmapped */
	492
	493	ib_dma_unmap_sg(smcibdev->ibdev,
	494	buf_slot->sgt[SMC_SINGLE_LINK].sgl,
	495	buf_slot->sgt[SMC_SINGLE_LINK].orig_nents,
	496	data_direction);
	497	buf_slot->sgt[SMC_SINGLE_LINK].sgl->dma_address = 0;
	498	}
	499
bd4ad577 UB	500	long smc_ib_setup_per_ibdev(struct smc_ib_device *smcibdev)
	501	{
	502	struct ib_cq_init_attr cqattr = {
c9f4c6cf UB	503	.cqe = SMC_MAX_CQE, .comp_vector = 0 };
c9f4c6cf UB	504	int cqe_size_order, smc_order;
bd4ad577 UB	505	long rc;
bd4ad577 UB	506
c9f4c6cf UB	507	/* the calculated number of cq entries fits to mlx5 cq allocation */
	508	cqe_size_order = cache_line_size() == 128 ? 7 : 6;
	509	smc_order = MAX_ORDER - cqe_size_order - 1;
	510	if (SMC_MAX_CQE + 2 > (0x00000001 << smc_order) * PAGE_SIZE)
	511	cqattr.cqe = (0x00000001 << smc_order) * PAGE_SIZE - 2;
bd4ad577 UB	512	smcibdev->roce_cq_send = ib_create_cq(smcibdev->ibdev,
	513	smc_wr_tx_cq_handler, NULL,
	514	smcibdev, &cqattr);
	515	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_send);
	516	if (IS_ERR(smcibdev->roce_cq_send)) {
	517	smcibdev->roce_cq_send = NULL;
	518	return rc;
	519	}
	520	smcibdev->roce_cq_recv = ib_create_cq(smcibdev->ibdev,
	521	smc_wr_rx_cq_handler, NULL,
	522	smcibdev, &cqattr);
	523	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_recv);
	524	if (IS_ERR(smcibdev->roce_cq_recv)) {
	525	smcibdev->roce_cq_recv = NULL;
	526	goto err;
	527	}
bd4ad577 UB	528	smc_wr_add_dev(smcibdev);
	529	smcibdev->initialized = 1;
	530	return rc;
	531
	532	err:
	533	ib_destroy_cq(smcibdev->roce_cq_send);
	534	return rc;
	535	}
	536
	537	static void smc_ib_cleanup_per_ibdev(struct smc_ib_device *smcibdev)
	538	{
	539	if (!smcibdev->initialized)
	540	return;
da05bf29	541	smcibdev->initialized = 0;
bd4ad577 UB	542	ib_destroy_cq(smcibdev->roce_cq_recv);
bd4ad577 UB	543	ib_destroy_cq(smcibdev->roce_cq_send);
6a37ad3d	544	smc_wr_remove_dev(smcibdev);
bd4ad577 UB	545	}
bd4ad577 UB	546
a4cf0443 UB	547	static struct ib_client smc_ib_client;
	548
	549	/* callback function for ib_register_client() */
	550	static void smc_ib_add_dev(struct ib_device *ibdev)
	551	{
	552	struct smc_ib_device *smcibdev;
be6a3f38 UB	553	u8 port_cnt;
be6a3f38 UB	554	int i;
a4cf0443 UB	555
	556	if (ibdev->node_type != RDMA_NODE_IB_CA)
	557	return;
	558
	559	smcibdev = kzalloc(sizeof(*smcibdev), GFP_KERNEL);
	560	if (!smcibdev)
	561	return;
	562
	563	smcibdev->ibdev = ibdev;
bd4ad577	564	INIT_WORK(&smcibdev->port_event_work, smc_ib_port_event_work);
6dabd405 UB	565	atomic_set(&smcibdev->lnk_cnt, 0);
6dabd405 UB	566	init_waitqueue_head(&smcibdev->lnks_deleted);
a4cf0443 UB	567	spin_lock(&smc_ib_devices.lock);
	568	list_add_tail(&smcibdev->list, &smc_ib_devices.list);
	569	spin_unlock(&smc_ib_devices.lock);
	570	ib_set_client_data(ibdev, &smc_ib_client, smcibdev);
be6a3f38 UB	571	INIT_IB_EVENT_HANDLER(&smcibdev->event_handler, smcibdev->ibdev,
	572	smc_ib_global_event_handler);
	573	ib_register_event_handler(&smcibdev->event_handler);
	574
	575	/* trigger reading of the port attributes */
	576	port_cnt = smcibdev->ibdev->phys_port_cnt;
	577	for (i = 0;
	578	i < min_t(size_t, port_cnt, SMC_MAX_PORTS);
0afff91c	579	i++) {
be6a3f38	580	set_bit(i, &smcibdev->port_event_mask);
0afff91c UB	581	/* determine pnetids of the port */
	582	smc_pnetid_by_dev_port(ibdev->dev.parent, i,
	583	smcibdev->pnetid[i]);
	584	}
be6a3f38	585	schedule_work(&smcibdev->port_event_work);
a4cf0443 UB	586	}
a4cf0443 UB	587
0b29ec64	588	/* callback function for ib_unregister_client() */
a4cf0443 UB	589	static void smc_ib_remove_dev(struct ib_device ibdev, void client_data)
	590	{
	591	struct smc_ib_device *smcibdev;
	592
	593	smcibdev = ib_get_client_data(ibdev, &smc_ib_client);
a2f2ef4a KG	594	if (!smcibdev \|\| smcibdev->ibdev != ibdev)
a2f2ef4a KG	595	return;
a4cf0443 UB	596	ib_set_client_data(ibdev, &smc_ib_client, NULL);
	597	spin_lock(&smc_ib_devices.lock);
	598	list_del_init(&smcibdev->list); /* remove from smc_ib_devices */
	599	spin_unlock(&smc_ib_devices.lock);
0b29ec64	600	smc_smcr_terminate_all(smcibdev);
bd4ad577	601	smc_ib_cleanup_per_ibdev(smcibdev);
be6a3f38	602	ib_unregister_event_handler(&smcibdev->event_handler);
ece0d7bd	603	cancel_work_sync(&smcibdev->port_event_work);
a4cf0443 UB	604	kfree(smcibdev);
	605	}
	606
	607	static struct ib_client smc_ib_client = {
	608	.name = "smc_ib",
	609	.add = smc_ib_add_dev,
	610	.remove = smc_ib_remove_dev,
	611	};
	612
	613	int __init smc_ib_register_client(void)
	614	{
366bb249	615	smc_ib_init_local_systemid();
a4cf0443 UB	616	return ib_register_client(&smc_ib_client);
	617	}
	618
	619	void smc_ib_unregister_client(void)
	620	{
	621	ib_unregister_client(&smc_ib_client);
	622	}