clients/eoib/eib_mac.c

	eib_mac.c revision b494511a9cf72b1fc4eb13a0e593f55c624ab829
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
 */

#include <sys/types.h>
#include <sys/conf.h>
#include <sys/devops.h>
#include <sys/kmem.h>
#include <sys/ksynch.h>
#include <sys/modctl.h>
#include <sys/stat.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/mac_provider.h>
#include <sys/mac_ether.h>

#include <sys/ib/clients/eoib/eib_impl.h>

/*
 * Declarations private to this file
 */
static void eib_rb_mac_start(eib_t *, eib_vnic_t *);

/*
 * This set of routines are used to set/clear the condition that the
 * caller is about to do something that affects the state of the nic.
 * If there's already someone doing either a start or a stop (possibly
 * due to the async handler, a plumb or a dlpi_open happening, or an
 * unplumb or dlpi_close coming in), we wait until that's done.
 */
void
eib_mac_set_nic_state(eib_t *ss, uint_t flags)
{
    eib_node_state_t *ns = ss->ei_node_state;

    mutex_enter(&ns->ns_lock);

    while ((ns->ns_nic_state & EIB_NIC_STARTING) ||
        (ns->ns_nic_state & EIB_NIC_STOPPING)) {
        cv_wait(&ns->ns_cv, &ns->ns_lock);
    }
    ns->ns_nic_state |= flags;

    mutex_exit(&ns->ns_lock);
}

void
eib_mac_clr_nic_state(eib_t *ss, uint_t flags)
{
    eib_node_state_t *ns = ss->ei_node_state;

    mutex_enter(&ns->ns_lock);

    ns->ns_nic_state &= (~flags);

    cv_broadcast(&ns->ns_cv);
    mutex_exit(&ns->ns_lock);
}

void
eib_mac_upd_nic_state(eib_t *ss, uint_t clr_flags, uint_t set_flags)
{
    eib_node_state_t *ns = ss->ei_node_state;

    mutex_enter(&ns->ns_lock);

    ns->ns_nic_state &= (~clr_flags);
    ns->ns_nic_state |= set_flags;

    cv_broadcast(&ns->ns_cv);
    mutex_exit(&ns->ns_lock);
}

uint_t
eib_mac_get_nic_state(eib_t *ss)
{
    eib_node_state_t *ns = ss->ei_node_state;
    uint_t nic_state;

    mutex_enter(&ns->ns_lock);
    nic_state = ns->ns_nic_state;
    mutex_exit(&ns->ns_lock);

    return (nic_state);
}

void
eib_mac_link_state(eib_t *ss, link_state_t new_link_state,
    boolean_t force)
{
    eib_node_state_t *ns = ss->ei_node_state;
    boolean_t state_changed = B_FALSE;

    mutex_enter(&ns->ns_lock);

    /*
     * We track the link state only if the current link state is
     * not unknown.  Obviously therefore, the first calls to set
     * the link state from eib_mac_start() have to pass an explicit
     * 'force' flag to force the state change tracking.
     */
    if (ns->ns_link_state != LINK_STATE_UNKNOWN)
        force = B_TRUE;

    if ((force) && (new_link_state != ns->ns_link_state)) {
        ns->ns_link_state = new_link_state;
        state_changed = B_TRUE;
    }
    mutex_exit(&ns->ns_lock);

    if (state_changed) {
        EIB_DPRINTF_DEBUG(ss->ei_instance,
            "eib_mac_link_state: changing link state to %d",
            new_link_state);

        mac_link_update(ss->ei_mac_hdl, new_link_state);
    } else  {
        EIB_DPRINTF_DEBUG(ss->ei_instance,
            "eib_mac_link_state: link state already %d",
            new_link_state);
    }
}

void
eib_mac_link_up(eib_t *ss, boolean_t force)
{
    eib_mac_link_state(ss, LINK_STATE_UP, force);
}

void
eib_mac_link_down(eib_t *ss, boolean_t force)
{
    eib_mac_link_state(ss, LINK_STATE_DOWN, force);
}

int
eib_mac_start(eib_t *ss)
{
    eib_vnic_t *vnic0 = NULL;
    eib_login_data_t *ld;
    int err;

    /*
     * Perform HCA related initializations
     */
    if (eib_ibt_hca_init(ss) != EIB_E_SUCCESS)
        goto start_fail;

    /*
     * Make sure port is up. Also record the port base lid if it's up.
     */
    if (eib_mac_hca_portstate(ss, &ss->ei_props->ep_blid,
        &err) != EIB_E_SUCCESS) {
        goto start_fail;
    }

    /*
     * Set up tx and rx buffer pools
     */
    if (eib_rsrc_setup_bufs(ss, &err) != EIB_E_SUCCESS)
        goto start_fail;

    /*
     * Set up admin qp for logins and logouts
     */
    if (eib_adm_setup_qp(ss, &err) != EIB_E_SUCCESS)
        goto start_fail;

    /*
     * Create the vnic for physlink (instance 0)
     */
    if (eib_vnic_create(ss, 0, 0, &vnic0, &err) != EIB_E_SUCCESS)
        goto start_fail;

    /*
     * Update the mac layer about the correct values for MTU and
     * unicast MAC address.  Note that we've already verified that the
     * vhub mtu (plus the eoib encapsulation header) is not greater
     * than our port mtu, so we can go ahead and report the vhub mtu
     * (of vnic0) directly.
     */
    ld = &(vnic0->vn_login_data);
    (void) mac_maxsdu_update(ss->ei_mac_hdl, ld->ld_vhub_mtu);
    mac_unicst_update(ss->ei_mac_hdl, ld->ld_assigned_mac);

    /*
     * Report that the link is up and ready
     */
    eib_mac_link_up(ss, B_TRUE);
    return (0);

start_fail:
    eib_rb_mac_start(ss, vnic0);
    eib_mac_link_down(ss, B_TRUE);
    return (err);
}

void
eib_mac_stop(eib_t *ss)
{
    eib_vnic_t *vnic;
    link_state_t cur_link_state = ss->ei_node_state->ns_link_state;
    int ndx;

    /*
     * Stopping an EoIB device instance is somewhat different from starting
     * it. Between the time the device instance was started and the call to
     * eib_m_stop() now, a number of vnics could've been created. All of
     * these will need to be destroyed before we can stop the device.
     */
    for (ndx = EIB_MAX_VNICS - 1; ndx >= 0; ndx--) {
        if ((vnic = ss->ei_vnic[ndx]) != NULL)
            eib_vnic_delete(ss, vnic);
    }

    /*
     * And now, to undo the things we did in start (other than creation
     * of vnics itself)
     */
    eib_rb_mac_start(ss, NULL);

    /*
     * Now that we're completed stopped, there's no mac address assigned
     * to us.  Update the mac layer with this information. Note that we
     * can let the old max mtu information remain as-is, since we're likely
     * to get that same mtu on a later plumb.
     */
    mac_unicst_update(ss->ei_mac_hdl, eib_zero_mac);

    /*
     * If our link state was up when the eib_m_stop() callback was called,
     * we'll mark the link state as unknown now.  Otherwise, we'll leave
     * the link state as-is (down).
     */
    if (cur_link_state == LINK_STATE_UP)
        eib_mac_link_state(ss, LINK_STATE_UNKNOWN, B_TRUE);
}

int
eib_mac_multicast(eib_t *ss, boolean_t add, uint8_t *mcast_mac)
{
    int ret = EIB_E_SUCCESS;
    int err = 0;

    /*
     * If it's a broadcast group join, each vnic needs to and is always
     * joined to the broadcast address, so we return success immediately.
     * If it's a broadcast group leave, we fail immediately for the same
     * reason as above.
     */
    if (bcmp(mcast_mac, eib_broadcast_mac, ETHERADDRL) == 0) {
        if (add)
            return (0);
        else
            return (EINVAL);
    }

    if (ss->ei_vnic[0]) {
        if (add) {
            ret = eib_vnic_join_data_mcg(ss, ss->ei_vnic[0],
                mcast_mac, B_FALSE, &err);
        } else {
            eib_vnic_leave_data_mcg(ss, ss->ei_vnic[0], mcast_mac);
            ret = EIB_E_SUCCESS;
        }
    }

    if (ret == EIB_E_SUCCESS)
        return (0);
    else
        return (err);
}

int
eib_mac_promisc(eib_t *ss, boolean_t set)
{
    int ret = EIB_E_SUCCESS;
    int err = 0;

    if (ss->ei_vnic[0]) {
        if (set) {
            ret = eib_vnic_join_data_mcg(ss, ss->ei_vnic[0],
                eib_zero_mac, B_FALSE, &err);
        } else {
            eib_vnic_leave_data_mcg(ss, ss->ei_vnic[0],
                eib_zero_mac);
            ret = EIB_E_SUCCESS;
        }
    }

    if (ret == EIB_E_SUCCESS)
        return (0);
    else
        return (err);
}

int
eib_mac_tx(eib_t *ss, mblk_t *mp)
{
    eib_ether_hdr_t evh;
    eib_vnic_t *vnic = NULL;
    eib_wqe_t *swqe = NULL;
    boolean_t failed_vnic;
    int found;
    int ret;

    /*
     * Grab a send wqe.  If we cannot get one, wake up a service
     * thread to monitor the swqe status and let the mac layer know
     * as soon as we have enough tx wqes to start the traffic again.
     */
    if ((swqe = eib_rsrc_grab_swqe(ss, EIB_WPRI_LO)) == NULL) {
        EIB_DPRINTF_WARN(ss->ei_instance, "eib_mac_tx: "
            "no swqe available, holding tx until resource "
            "becomes available");
        eib_rsrc_txwqes_needed(ss);
        return (EIB_E_FAILURE);
    }

    /*
     * Determine dmac, smac and vlan information
     */
    eib_data_parse_ether_hdr(mp, &evh);

    /*
     * Lookup the {smac, vlan} tuple in our vnic list. If it isn't
     * there, this is obviously a new packet on a vnic/vlan that
     * we haven't been informed about. So go ahead and file a request
     * to create a new vnic. This is obviously not a clean thing to
     * do - we should be informed when a vnic/vlan is being created
     * and should be given a proper opportunity to login to the gateway
     * and do the creation.  But we don't have that luxury now, and
     * this is the next best thing to do.  Note that we return failure
     * from here, so tx flow control should prevent further packets
     * from coming in until the vnic creation has completed.
     */
    found = eib_data_lookup_vnic(ss, evh.eh_smac, evh.eh_vlan, &vnic,
        &failed_vnic);
    if (found != EIB_E_SUCCESS) {
        uint8_t *m = evh.eh_smac;

        /*
         * Return the swqe back to the pool
         */
        eib_rsrc_return_swqe(ss, swqe, NULL);

        /*
         * If we had previously tried creating this vnic and had
         * failed, we'll simply drop the packets on this vnic.
         * Otherwise, we'll queue up a request to create this vnic.
         */
        if (failed_vnic) {
            EIB_DPRINTF_VERBOSE(ss->ei_instance, "eib_mac_tx: "
                "vnic creation for mac=%x:%x:%x:%x:%x:%x "
                "vlan=0x%x failed previously, dropping pkt",
                m[0], m[1], m[2], m[3], m[4], m[5], evh.eh_vlan);
            return (EIB_E_SUCCESS);
        } else {
            eib_vnic_need_new(ss, evh.eh_smac, evh.eh_vlan);
            return (EIB_E_FAILURE);
        }
    }

    /*
     * We'll try to setup the destination in the swqe for this dmac
     * and vlan.  If we don't succeed, there's no need to undo any
     * vnic-creation we might've made above (if we didn't find the
     * vnic corresponding to the {smac, vlan} originally). Note that
     * this is not a resource issue, so we'll issue a warning and
     * drop the packet, but won't return failure from here.
     */
    ret = eib_vnic_setup_dest(vnic, swqe, evh.eh_dmac, evh.eh_vlan);
    if (ret != EIB_E_SUCCESS) {
        uint8_t *dmac;

        dmac = evh.eh_dmac;
        EIB_DPRINTF_WARN(ss->ei_instance, "eib_mac_tx: "
            "eib_vnic_setup_dest() failed for mac=%x:%x:%x:%x:%x:%x, "
            "vlan=0x%x, dropping pkt", dmac[0], dmac[1], dmac[2],
            dmac[3], dmac[4], dmac[5]);

        eib_rsrc_return_swqe(ss, swqe, NULL);
        return (EIB_E_SUCCESS);
    }

    /*
     * The only reason why this would fail is if we needed LSO buffer(s)
     * to prepare this frame and couldn't find enough of those.
     */
    ret = eib_data_prepare_frame(vnic, swqe, mp, &evh);
    if (ret != EIB_E_SUCCESS) {
        EIB_DPRINTF_WARN(ss->ei_instance, "eib_mac_tx: "
            "eib_data_prepare_frame() failed (no LSO bufs?), "
            "holding tx until resource becomes available");

        eib_rsrc_return_swqe(ss, swqe, NULL);
        eib_rsrc_lsobufs_needed(ss);
        return (EIB_E_FAILURE);
    }

    eib_data_post_tx(vnic, swqe);

    return (EIB_E_SUCCESS);
}

int
eib_mac_hca_portstate(eib_t *ss, ib_lid_t *blid, int *err)
{
    ibt_hca_portinfo_t *pi;
    ibt_status_t ret;
    uint_t num_pi;
    uint_t sz_pi;

    ret = ibt_query_hca_ports(ss->ei_hca_hdl, ss->ei_props->ep_port_num,
        &pi, &num_pi, &sz_pi);
    if (ret != IBT_SUCCESS) {
        EIB_DPRINTF_ERR(ss->ei_instance,
            "ibt_query_hca_ports(hca_hdl=0x%llx, "
            "port=0x%x) failed, ret=%d", ss->ei_hca_hdl,
            ss->ei_props->ep_port_num, ret);
        goto mac_hca_portstate_fail;
    }
    if (num_pi != 1) {
        EIB_DPRINTF_ERR(ss->ei_instance,
            "ibt_query_hca_ports(hca_hdl=0x%llx, "
            "port=0x%x) returned num_pi=%d", ss->ei_hca_hdl,
            ss->ei_props->ep_port_num, num_pi);
        goto mac_hca_portstate_fail;
    }

    if (pi->p_linkstate != IBT_PORT_ACTIVE)
        goto mac_hca_portstate_fail;

    /*
     * Return the port's base lid if asked
     */
    if (blid) {
        *blid = pi->p_base_lid;
    }

    ibt_free_portinfo(pi, sz_pi);
    return (EIB_E_SUCCESS);

mac_hca_portstate_fail:
    if (pi) {
        ibt_free_portinfo(pi, sz_pi);
    }
    if (err) {
        *err = ENETDOWN;
    }
    return (EIB_E_FAILURE);
}

static void
eib_rb_mac_start(eib_t *ss, eib_vnic_t *vnic0)
{
    int ntries;

    /*
     * If vnic0 is non-null, delete it
     */
    if (vnic0) {
        eib_rb_vnic_create(ss, vnic0, ~0);
    }

    /*
     * At this point, we're pretty much done with all communication that
     * we need to do for vnic-logout, etc. so we can get rid of any address
     * vectors we might've allocated to send control/data packets.
     */
    eib_ibt_free_avects(ss);

    /*
     * Tear down the rest of it
     */
    if (ss->ei_admin_chan) {
        eib_rb_adm_setup_qp(ss);
    }

    /*
     * If (say) the network layer has been holding onto our rx buffers, we
     * wait a reasonable time for it to hand them back to us.  If we don't
     * get it still, we have nothing to do but avoid rolling back hca init
     * since we cannot unregister the memory, release the pd or close the
     * hca.  We'll try to reuse it if there's a plumb again.
     */
    for (ntries = 0; ntries < EIB_MAX_ATTEMPTS; ntries++) {
        eib_rb_rsrc_setup_bufs(ss, B_FALSE);
        if ((ss->ei_tx == NULL) && (ss->ei_rx == NULL) &&
            (ss->ei_lso == NULL)) {
            break;
        }

        delay(drv_usectohz(EIB_DELAY_HALF_SECOND));
    }

    if (ntries == EIB_MAX_ATTEMPTS) {
        EIB_DPRINTF_WARN(ss->ei_instance, "eib_rb_mac_start: "
            "bufs outstanding, tx=0x%llx, rx=0x%llx, lso=0x%llx",
            ss->ei_tx, ss->ei_rx, ss->ei_lso);
    } else if (ss->ei_hca_hdl) {
        eib_rb_ibt_hca_init(ss, ~0);
    }
    ss->ei_props->ep_blid = 0;

    /*
     * Pending vnic creation requests (and failed-vnic records) will have
     * to be cleaned up in any case
     */
    eib_flush_vnic_reqs(ss);
}