ipath_verbs.c

来自「LINUX 2.6.17.4的源码」· C语言 代码 · 共 1,205 行 · 第 1/3 页

	if (qp_num == IPS_MULTICAST_QPN) {
		struct ipath_mcast *mcast;
		struct ipath_mcast_qp *p;

		mcast = ipath_mcast_find(&hdr->u.l.grh.dgid);
		if (mcast == NULL) {
			dev->n_pkt_drops++;
			goto bail;
		}
		dev->n_multicast_rcv++;
		list_for_each_entry_rcu(p, &mcast->qp_list, list)
			ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
				     tlen, p->qp);
		/*
		 * Notify ipath_multicast_detach() if it is waiting for us
		 * to finish.
		 */
		if (atomic_dec_return(&mcast->refcount) <= 1)
			wake_up(&mcast->wait);
	} else {
		qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
		if (qp) {
			dev->n_unicast_rcv++;
			ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
				     tlen, qp);
			/*
			 * Notify ipath_destroy_qp() if it is waiting
			 * for us to finish.
			 */
			if (atomic_dec_and_test(&qp->refcount))
				wake_up(&qp->wait);
		} else
			dev->n_pkt_drops++;
	}

bail:;
}

/**
 * ipath_ib_timer - verbs timer
 * @arg: the device pointer
 *
 * This is called from ipath_do_rcv_timer() at interrupt level to check for
 * QPs which need retransmits and to collect performance numbers.
 */
static void ipath_ib_timer(void *arg)
{
	struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
	struct ipath_qp *resend = NULL;
	struct list_head *last;
	struct ipath_qp *qp;
	unsigned long flags;

	if (dev == NULL)
		return;

	spin_lock_irqsave(&dev->pending_lock, flags);
	/* Start filling the next pending queue. */
	if (++dev->pending_index >= ARRAY_SIZE(dev->pending))
		dev->pending_index = 0;
	/* Save any requests still in the new queue, they have timed out. */
	last = &dev->pending[dev->pending_index];
	while (!list_empty(last)) {
		qp = list_entry(last->next, struct ipath_qp, timerwait);
		list_del_init(&qp->timerwait);
		qp->timer_next = resend;
		resend = qp;
		atomic_inc(&qp->refcount);
	}
	last = &dev->rnrwait;
	if (!list_empty(last)) {
		qp = list_entry(last->next, struct ipath_qp, timerwait);
		if (--qp->s_rnr_timeout == 0) {
			do {
				list_del_init(&qp->timerwait);
				tasklet_hi_schedule(&qp->s_task);
				if (list_empty(last))
					break;
				qp = list_entry(last->next, struct ipath_qp,
						timerwait);
			} while (qp->s_rnr_timeout == 0);
		}
	}
	/*
	 * We should only be in the started state if pma_sample_start != 0
	 */
	if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED &&
	    --dev->pma_sample_start == 0) {
		dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
		ipath_layer_snapshot_counters(dev->dd, &dev->ipath_sword,
					      &dev->ipath_rword,
					      &dev->ipath_spkts,
					      &dev->ipath_rpkts,
					      &dev->ipath_xmit_wait);
	}
	if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) {
		if (dev->pma_sample_interval == 0) {
			u64 ta, tb, tc, td, te;

			dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
			ipath_layer_snapshot_counters(dev->dd, &ta, &tb,
						      &tc, &td, &te);

			dev->ipath_sword = ta - dev->ipath_sword;
			dev->ipath_rword = tb - dev->ipath_rword;
			dev->ipath_spkts = tc - dev->ipath_spkts;
			dev->ipath_rpkts = td - dev->ipath_rpkts;
			dev->ipath_xmit_wait = te - dev->ipath_xmit_wait;
		}
		else
			dev->pma_sample_interval--;
	}
	spin_unlock_irqrestore(&dev->pending_lock, flags);

	/* XXX What if timer fires again while this is running? */
	for (qp = resend; qp != NULL; qp = qp->timer_next) {
		struct ib_wc wc;

		spin_lock_irqsave(&qp->s_lock, flags);
		if (qp->s_last != qp->s_tail && qp->state == IB_QPS_RTS) {
			dev->n_timeouts++;
			ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
		}
		spin_unlock_irqrestore(&qp->s_lock, flags);

		/* Notify ipath_destroy_qp() if it is waiting. */
		if (atomic_dec_and_test(&qp->refcount))
			wake_up(&qp->wait);
	}
}

/**
 * ipath_ib_piobufavail - callback when a PIO buffer is available
 * @arg: the device pointer
 *
 * This is called from ipath_intr() at interrupt level when a PIO buffer is
 * available after ipath_verbs_send() returned an error that no buffers were
 * available.  Return 1 if we consumed all the PIO buffers and we still have
 * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and
 * return zero).
 */
static int ipath_ib_piobufavail(void *arg)
{
	struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
	struct ipath_qp *qp;
	unsigned long flags;

	if (dev == NULL)
		goto bail;

	spin_lock_irqsave(&dev->pending_lock, flags);
	while (!list_empty(&dev->piowait)) {
		qp = list_entry(dev->piowait.next, struct ipath_qp,
				piowait);
		list_del_init(&qp->piowait);
		tasklet_hi_schedule(&qp->s_task);
	}
	spin_unlock_irqrestore(&dev->pending_lock, flags);

bail:
	return 0;
}

static int ipath_query_device(struct ib_device *ibdev,
			      struct ib_device_attr *props)
{
	struct ipath_ibdev *dev = to_idev(ibdev);
	u32 vendor, boardrev, majrev, minrev;

	memset(props, 0, sizeof(*props));

	props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
		IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
		IB_DEVICE_SYS_IMAGE_GUID;
	ipath_layer_query_device(dev->dd, &vendor, &boardrev,
				 &majrev, &minrev);
	props->vendor_id = vendor;
	props->vendor_part_id = boardrev;
	props->hw_ver = boardrev << 16 | majrev << 8 | minrev;

	props->sys_image_guid = dev->sys_image_guid;

	props->max_mr_size = ~0ull;
	props->max_qp = 0xffff;
	props->max_qp_wr = 0xffff;
	props->max_sge = 255;
	props->max_cq = 0xffff;
	props->max_cqe = 0xffff;
	props->max_mr = 0xffff;
	props->max_pd = 0xffff;
	props->max_qp_rd_atom = 1;
	props->max_qp_init_rd_atom = 1;
	/* props->max_res_rd_atom */
	props->max_srq = 0xffff;
	props->max_srq_wr = 0xffff;
	props->max_srq_sge = 255;
	/* props->local_ca_ack_delay */
	props->atomic_cap = IB_ATOMIC_HCA;
	props->max_pkeys = ipath_layer_get_npkeys(dev->dd);
	props->max_mcast_grp = 0xffff;
	props->max_mcast_qp_attach = 0xffff;
	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
		props->max_mcast_grp;

	return 0;
}

const u8 ipath_cvt_physportstate[16] = {
	[INFINIPATH_IBCS_LT_STATE_DISABLED] = 3,
	[INFINIPATH_IBCS_LT_STATE_LINKUP] = 5,
	[INFINIPATH_IBCS_LT_STATE_POLLACTIVE] = 2,
	[INFINIPATH_IBCS_LT_STATE_POLLQUIET] = 2,
	[INFINIPATH_IBCS_LT_STATE_SLEEPDELAY] = 1,
	[INFINIPATH_IBCS_LT_STATE_SLEEPQUIET] = 1,
	[INFINIPATH_IBCS_LT_STATE_CFGDEBOUNCE] = 4,
	[INFINIPATH_IBCS_LT_STATE_CFGRCVFCFG] = 4,
	[INFINIPATH_IBCS_LT_STATE_CFGWAITRMT] = 4,
	[INFINIPATH_IBCS_LT_STATE_CFGIDLE] = 4,
	[INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN] = 6,
	[INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT] = 6,
	[INFINIPATH_IBCS_LT_STATE_RECOVERIDLE] = 6,
};

static int ipath_query_port(struct ib_device *ibdev,
			    u8 port, struct ib_port_attr *props)
{
	struct ipath_ibdev *dev = to_idev(ibdev);
	enum ib_mtu mtu;
	u16 lid = ipath_layer_get_lid(dev->dd);
	u64 ibcstat;

	memset(props, 0, sizeof(*props));
	props->lid = lid ? lid : __constant_be16_to_cpu(IB_LID_PERMISSIVE);
	props->lmc = dev->mkeyprot_resv_lmc & 7;
	props->sm_lid = dev->sm_lid;
	props->sm_sl = dev->sm_sl;
	ibcstat = ipath_layer_get_lastibcstat(dev->dd);
	props->state = ((ibcstat >> 4) & 0x3) + 1;
	/* See phys_state_show() */
	props->phys_state = ipath_cvt_physportstate[
		ipath_layer_get_lastibcstat(dev->dd) & 0xf];
	props->port_cap_flags = dev->port_cap_flags;
	props->gid_tbl_len = 1;
	props->max_msg_sz = 4096;
	props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
	props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
		dev->n_pkey_violations;
	props->qkey_viol_cntr = dev->qkey_violations;
	props->active_width = IB_WIDTH_4X;
	/* See rate_show() */
	props->active_speed = 1;	/* Regular 10Mbs speed. */
	props->max_vl_num = 1;		/* VLCap = VL0 */
	props->init_type_reply = 0;

	props->max_mtu = IB_MTU_4096;
	switch (ipath_layer_get_ibmtu(dev->dd)) {
	case 4096:
		mtu = IB_MTU_4096;
		break;
	case 2048:
		mtu = IB_MTU_2048;
		break;
	case 1024:
		mtu = IB_MTU_1024;
		break;
	case 512:
		mtu = IB_MTU_512;
		break;
	case 256:
		mtu = IB_MTU_256;
		break;
	default:
		mtu = IB_MTU_2048;
	}
	props->active_mtu = mtu;
	props->subnet_timeout = dev->subnet_timeout;

	return 0;
}

static int ipath_modify_device(struct ib_device *device,
			       int device_modify_mask,
			       struct ib_device_modify *device_modify)
{
	int ret;

	if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
				   IB_DEVICE_MODIFY_NODE_DESC)) {
		ret = -EOPNOTSUPP;
		goto bail;
	}

	if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC)
		memcpy(device->node_desc, device_modify->node_desc, 64);

	if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
		to_idev(device)->sys_image_guid =
			cpu_to_be64(device_modify->sys_image_guid);

	ret = 0;

bail:
	return ret;
}

static int ipath_modify_port(struct ib_device *ibdev,
			     u8 port, int port_modify_mask,
			     struct ib_port_modify *props)
{
	struct ipath_ibdev *dev = to_idev(ibdev);

	dev->port_cap_flags |= props->set_port_cap_mask;
	dev->port_cap_flags &= ~props->clr_port_cap_mask;
	if (port_modify_mask & IB_PORT_SHUTDOWN)
		ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKDOWN);
	if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
		dev->qkey_violations = 0;
	return 0;
}

static int ipath_query_gid(struct ib_device *ibdev, u8 port,
			   int index, union ib_gid *gid)
{
	struct ipath_ibdev *dev = to_idev(ibdev);
	int ret;

	if (index >= 1) {
		ret = -EINVAL;
		goto bail;
	}
	gid->global.subnet_prefix = dev->gid_prefix;
	gid->global.interface_id = ipath_layer_get_guid(dev->dd);

	ret = 0;

bail:
	return ret;
}

static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev,
				    struct ib_ucontext *context,
				    struct ib_udata *udata)
{
	struct ipath_pd *pd;
	struct ib_pd *ret;

	pd = kmalloc(sizeof *pd, GFP_KERNEL);
	if (!pd) {
		ret = ERR_PTR(-ENOMEM);
		goto bail;
	}

	/* ib_alloc_pd() will initialize pd->ibpd. */
	pd->user = udata != NULL;

	ret = &pd->ibpd;

bail:
	return ret;
}

static int ipath_dealloc_pd(struct ib_pd *ibpd)
{
	struct ipath_pd *pd = to_ipd(ibpd);

	kfree(pd);

	return 0;
}

/**
 * ipath_create_ah - create an address handle
 * @pd: the protection domain
 * @ah_attr: the attributes of the AH
 *
 * This may be called from interrupt context.
 */
static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
				     struct ib_ah_attr *ah_attr)
{
	struct ipath_ah *ah;
	struct ib_ah *ret;

	/* A multicast address requires a GRH (see ch. 8.4.1). */
	if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
	    ah_attr->dlid != IPS_PERMISSIVE_LID &&
	    !(ah_attr->ah_flags & IB_AH_GRH)) {
		ret = ERR_PTR(-EINVAL);
		goto bail;
	}

	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
	if (!ah) {
		ret = ERR_PTR(-ENOMEM);
		goto bail;
	}

	/* ib_create_ah() will initialize ah->ibah. */
	ah->attr = *ah_attr;

	ret = &ah->ibah;