kernel - kqueue - major refactoring

[dragonfly.git] / sys / kern / kern_poll.c

/*-
 * Copyright (c) 2001-2002 Luigi Rizzo
 *
 * Supported by: the Xorp Project (www.xorp.org)
 *
 * 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 AUTHORS 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 AUTHORS 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.
 *
 * $FreeBSD: src/sys/kern/kern_poll.c,v 1.2.2.4 2002/06/27 23:26:33 luigi Exp $
 * $DragonFly: src/sys/kern/kern_poll.c,v 1.48 2008/09/24 12:07:19 sephe Exp $
 */

#include "opt_polling.h"

#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/socket.h>                 /* needed by net/if.h           */
#include <sys/sysctl.h>

#include <sys/thread2.h>
#include <sys/msgport2.h>

#include <net/if.h>                     /* for IFF_* flags              */
#include <net/netmsg2.h>

/*
 * Polling support for [network] device drivers.
 *
 * Drivers which support this feature try to register with the
 * polling code.
 *
 * If registration is successful, the driver must disable interrupts,
 * and further I/O is performed through the handler, which is invoked
 * (at least once per clock tick) with 3 arguments: the "arg" passed at
 * register time (a struct ifnet pointer), a command, and a "count" limit.
 *
 * The command can be one of the following:
 *  POLL_ONLY: quick move of "count" packets from input/output queues.
 *  POLL_AND_CHECK_STATUS: as above, plus check status registers or do
 *      other more expensive operations. This command is issued periodically
 *      but less frequently than POLL_ONLY.
 *  POLL_DEREGISTER: deregister and return to interrupt mode.
 *  POLL_REGISTER: register and disable interrupts
 *
 * The first two commands are only issued if the interface is marked as
 * 'IFF_UP, IFF_RUNNING and IFF_POLLING', the last two only if IFF_RUNNING
 * is set.
 *
 * The count limit specifies how much work the handler can do during the
 * call -- typically this is the number of packets to be received, or
 * transmitted, etc. (drivers are free to interpret this number, as long
 * as the max time spent in the function grows roughly linearly with the
 * count).
 *
 * Deregistration can be requested by the driver itself (typically in the
 * *_stop() routine), or by the polling code, by invoking the handler.
 *
 * Polling can be enabled or disabled on particular CPU_X with the sysctl
 * variable kern.polling.X.enable (default is 1, enabled)
 *
 * A second variable controls the sharing of CPU between polling/kernel
 * network processing, and other activities (typically userlevel tasks):
 * kern.polling.X.user_frac (between 0 and 100, default 50) sets the share
 * of CPU allocated to user tasks. CPU is allocated proportionally to the
 * shares, by dynamically adjusting the "count" (poll_burst).
 *
 * Other parameters can should be left to their default values.
 * The following constraints hold
 *
 *      1 <= poll_burst <= poll_burst_max
 *      1 <= poll_each_burst <= poll_burst_max
 *      MIN_POLL_BURST_MAX <= poll_burst_max <= MAX_POLL_BURST_MAX
 */

#define MIN_POLL_BURST_MAX      10
#define MAX_POLL_BURST_MAX      1000
#define POLL_BURST_MAX          150     /* good for 100Mbit net and HZ=1000 */
#define POLL_EACH_BURST         5

#ifndef DEVICE_POLLING_FREQ_MAX
#define DEVICE_POLLING_FREQ_MAX         30000
#endif
#define DEVICE_POLLING_FREQ_DEFAULT     2000

#define POLL_LIST_LEN  128
struct pollrec {
        struct ifnet    *ifp;
};

#define POLLCTX_MAX     32

struct pollctx {
        struct sysctl_ctx_list  poll_sysctl_ctx;
        struct sysctl_oid       *poll_sysctl_tree;

        uint32_t                poll_burst;             /* state */
        uint32_t                poll_each_burst;        /* tunable */
        uint32_t                poll_burst_max;         /* tunable */
        uint32_t                user_frac;              /* tunable */
        int                     reg_frac_count;         /* state */
        uint32_t                reg_frac;               /* tunable */
        uint32_t                short_ticks;            /* statistics */
        uint32_t                lost_polls;             /* statistics */
        uint32_t                pending_polls;          /* state */
        int                     residual_burst;         /* state */
        uint32_t                phase;                  /* state */
        uint32_t                suspect;                /* statistics */
        uint32_t                stalled;                /* statistics */
        struct timeval          poll_start_t;           /* state */
        struct timeval          prev_t;                 /* state */

        uint32_t                poll_handlers; /* next free entry in pr[]. */
        struct pollrec          pr[POLL_LIST_LEN];

        int                     poll_cpuid;
        struct systimer         pollclock;
        int                     polling_enabled;        /* tunable */
        int                     pollhz;                 /* tunable */

        struct netmsg           poll_netmsg;
        struct netmsg           poll_more_netmsg;
};

static struct pollctx   *poll_context[POLLCTX_MAX];

SYSCTL_NODE(_kern, OID_AUTO, polling, CTLFLAG_RW, 0,
        "Device polling parameters");

static int      poll_defcpu = -1;
SYSCTL_INT(_kern_polling, OID_AUTO, defcpu, CTLFLAG_RD,
        &poll_defcpu, 0, "default CPU to run device polling");

static uint32_t poll_cpumask0 = 0xffffffff;
TUNABLE_INT("kern.polling.cpumask", (int *)&poll_cpumask0);

static uint32_t poll_cpumask;
SYSCTL_INT(_kern_polling, OID_AUTO, cpumask, CTLFLAG_RD,
        &poll_cpumask, 0, "CPUs that can run device polling");

static int      polling_enabled = 1;    /* global polling enable */
TUNABLE_INT("kern.polling.enable", &polling_enabled);

static int      pollhz = DEVICE_POLLING_FREQ_DEFAULT;
TUNABLE_INT("kern.polling.pollhz", &pollhz);

static int      poll_burst_max = POLL_BURST_MAX;
TUNABLE_INT("kern.polling.burst_max", &poll_burst_max);

static int      poll_each_burst = POLL_EACH_BURST;
TUNABLE_INT("kern.polling.each_burst", &poll_each_burst);

/* Netisr handlers */
static void     netisr_poll(struct netmsg *);
static void     netisr_pollmore(struct netmsg *);
static void     poll_register(struct netmsg *);
static void     poll_deregister(struct netmsg *);
static void     poll_sysctl_pollhz(struct netmsg *);
static void     poll_sysctl_polling(struct netmsg *);
static void     poll_sysctl_regfrac(struct netmsg *);
static void     poll_sysctl_burstmax(struct netmsg *);
static void     poll_sysctl_eachburst(struct netmsg *);

/* Systimer handler */
static void     pollclock(systimer_t, struct intrframe *);

/* Sysctl handlers */
static int      sysctl_pollhz(SYSCTL_HANDLER_ARGS);
static int      sysctl_polling(SYSCTL_HANDLER_ARGS);
static int      sysctl_regfrac(SYSCTL_HANDLER_ARGS);
static int      sysctl_burstmax(SYSCTL_HANDLER_ARGS);
static int      sysctl_eachburst(SYSCTL_HANDLER_ARGS);
static void     poll_add_sysctl(struct sysctl_ctx_list *,
                                struct sysctl_oid_list *, struct pollctx *);

static void     schedpoll_oncpu(struct netmsg *);

void            init_device_poll_pcpu(int);     /* per-cpu init routine */

#define POLL_KTR_STRING         "ifp=%p"
#define POLL_KTR_ARG_SIZE       (sizeof(void *))

#ifndef KTR_POLLING
#define KTR_POLLING     KTR_ALL
#endif
KTR_INFO_MASTER(poll);
KTR_INFO(KTR_POLLING, poll, beg, 0, POLL_KTR_STRING, POLL_KTR_ARG_SIZE);
KTR_INFO(KTR_POLLING, poll, end, 1, POLL_KTR_STRING, POLL_KTR_ARG_SIZE);

#define logpoll(name, arg)      KTR_LOG(poll_ ## name, arg)

static __inline void
poll_reset_state(struct pollctx *pctx)
{
        crit_enter();
        pctx->poll_burst = 5;
        pctx->reg_frac_count = 0;
        pctx->pending_polls = 0;
        pctx->residual_burst = 0;
        pctx->phase = 0;
        bzero(&pctx->poll_start_t, sizeof(pctx->poll_start_t));
        bzero(&pctx->prev_t, sizeof(pctx->prev_t));
        crit_exit();
}

/*
 * Initialize per-cpu polling(4) context.  Called from kern_clock.c:
 */
void
init_device_poll_pcpu(int cpuid)
{
        struct pollctx *pctx;
        char cpuid_str[3];

        if (cpuid >= POLLCTX_MAX)
                return;

        if (((1 << cpuid) & poll_cpumask0) == 0)
                return;

        if (poll_burst_max < MIN_POLL_BURST_MAX)
                poll_burst_max = MIN_POLL_BURST_MAX;
        else if (poll_burst_max > MAX_POLL_BURST_MAX)
                poll_burst_max = MAX_POLL_BURST_MAX;

        if (poll_each_burst > poll_burst_max)
                poll_each_burst = poll_burst_max;

        poll_cpumask |= (1 << cpuid);

        pctx = kmalloc(sizeof(*pctx), M_DEVBUF, M_WAITOK | M_ZERO);

        pctx->poll_each_burst = poll_each_burst;
        pctx->poll_burst_max = poll_burst_max;
        pctx->user_frac = 50;
        pctx->reg_frac = 20;
        pctx->polling_enabled = polling_enabled;
        pctx->pollhz = pollhz;
        pctx->poll_cpuid = cpuid;
        poll_reset_state(pctx);

        netmsg_init(&pctx->poll_netmsg, NULL, &netisr_adone_rport,
                    MSGF_MPSAFE, netisr_poll);
#ifdef INVARIANTS
        pctx->poll_netmsg.nm_lmsg.u.ms_resultp = pctx;
#endif

        netmsg_init(&pctx->poll_more_netmsg, NULL, &netisr_adone_rport,
                    MSGF_MPSAFE, netisr_pollmore);
#ifdef INVARIANTS
        pctx->poll_more_netmsg.nm_lmsg.u.ms_resultp = pctx;
#endif

        KASSERT(cpuid < POLLCTX_MAX, ("cpu id must < %d", cpuid));
        poll_context[cpuid] = pctx;

        if (poll_defcpu < 0) {
                poll_defcpu = cpuid;

                /*
                 * Initialize global sysctl nodes, for compat
                 */
                poll_add_sysctl(NULL, SYSCTL_STATIC_CHILDREN(_kern_polling),
                                pctx);
        }

        /*
         * Initialize per-cpu sysctl nodes
         */
        ksnprintf(cpuid_str, sizeof(cpuid_str), "%d", pctx->poll_cpuid);

        sysctl_ctx_init(&pctx->poll_sysctl_ctx);
        pctx->poll_sysctl_tree = SYSCTL_ADD_NODE(&pctx->poll_sysctl_ctx,
                                 SYSCTL_STATIC_CHILDREN(_kern_polling),
                                 OID_AUTO, cpuid_str, CTLFLAG_RD, 0, "");
        poll_add_sysctl(&pctx->poll_sysctl_ctx,
                        SYSCTL_CHILDREN(pctx->poll_sysctl_tree), pctx);

        /*
         * Initialize systimer
         */
        systimer_init_periodic_nq(&pctx->pollclock, pollclock, pctx, 1);
}

static void
schedpoll_oncpu(struct netmsg *msg)
{
        if (msg->nm_lmsg.ms_flags & MSGF_DONE)
                lwkt_sendmsg(cpu_portfn(mycpuid), &msg->nm_lmsg);
}

static __inline void
schedpoll(struct pollctx *pctx)
{
        crit_enter();
        schedpoll_oncpu(&pctx->poll_netmsg);
        crit_exit();
}

static __inline void
schedpollmore(struct pollctx *pctx)
{
        schedpoll_oncpu(&pctx->poll_more_netmsg);
}

/*
 * Set the polling frequency
 */
static int
sysctl_pollhz(SYSCTL_HANDLER_ARGS)
{
        struct pollctx *pctx = arg1;
        struct netmsg msg;
        lwkt_port_t port;
        int error, phz;

        phz = pctx->pollhz;
        error = sysctl_handle_int(oidp, &phz, 0, req);
        if (error || req->newptr == NULL)
                return error;
        if (phz <= 0)
                return EINVAL;
        else if (phz > DEVICE_POLLING_FREQ_MAX)
                phz = DEVICE_POLLING_FREQ_MAX;

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_sysctl_pollhz);
        msg.nm_lmsg.u.ms_result = phz;

        port = cpu_portfn(pctx->poll_cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);
        return 0;
}

/*
 * Master enable.
 */
static int
sysctl_polling(SYSCTL_HANDLER_ARGS)
{
        struct pollctx *pctx = arg1;
        struct netmsg msg;
        lwkt_port_t port;
        int error, enabled;

        enabled = pctx->polling_enabled;
        error = sysctl_handle_int(oidp, &enabled, 0, req);
        if (error || req->newptr == NULL)
                return error;

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_sysctl_polling);
        msg.nm_lmsg.u.ms_result = enabled;

        port = cpu_portfn(pctx->poll_cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);
        return 0;
}

static int
sysctl_regfrac(SYSCTL_HANDLER_ARGS)
{
        struct pollctx *pctx = arg1;
        struct netmsg msg;
        lwkt_port_t port;
        uint32_t reg_frac;
        int error;

        reg_frac = pctx->reg_frac;
        error = sysctl_handle_int(oidp, &reg_frac, 0, req);
        if (error || req->newptr == NULL)
                return error;

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_sysctl_regfrac);
        msg.nm_lmsg.u.ms_result = reg_frac;

        port = cpu_portfn(pctx->poll_cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);
        return 0;
}

static int
sysctl_burstmax(SYSCTL_HANDLER_ARGS)
{
        struct pollctx *pctx = arg1;
        struct netmsg msg;
        lwkt_port_t port;
        uint32_t burst_max;
        int error;

        burst_max = pctx->poll_burst_max;
        error = sysctl_handle_int(oidp, &burst_max, 0, req);
        if (error || req->newptr == NULL)
                return error;
        if (burst_max < MIN_POLL_BURST_MAX)
                burst_max = MIN_POLL_BURST_MAX;
        else if (burst_max > MAX_POLL_BURST_MAX)
                burst_max = MAX_POLL_BURST_MAX;

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_sysctl_burstmax);
        msg.nm_lmsg.u.ms_result = burst_max;

        port = cpu_portfn(pctx->poll_cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);
        return 0;
}

static int
sysctl_eachburst(SYSCTL_HANDLER_ARGS)
{
        struct pollctx *pctx = arg1;
        struct netmsg msg;
        lwkt_port_t port;
        uint32_t each_burst;
        int error;

        each_burst = pctx->poll_each_burst;
        error = sysctl_handle_int(oidp, &each_burst, 0, req);
        if (error || req->newptr == NULL)
                return error;

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_sysctl_eachburst);
        msg.nm_lmsg.u.ms_result = each_burst;

        port = cpu_portfn(pctx->poll_cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);
        return 0;
}

/*
 * Hook from polling systimer. Tries to schedule a netisr, but keeps
 * track of lost ticks due to the previous handler taking too long.
 * Normally, this should not happen, because polling handler should
 * run for a short time. However, in some cases (e.g. when there are
 * changes in link status etc.) the drivers take a very long time
 * (even in the order of milliseconds) to reset and reconfigure the
 * device, causing apparent lost polls.
 *
 * The first part of the code is just for debugging purposes, and tries
 * to count how often hardclock ticks are shorter than they should,
 * meaning either stray interrupts or delayed events.
 *
 * WARNING! called from fastint or IPI, the MP lock might not be held.
 */
static void
pollclock(systimer_t info, struct intrframe *frame __unused)
{
        struct pollctx *pctx = info->data;
        struct timeval t;
        int delta;

        if (pctx->poll_handlers == 0)
                return;

        microuptime(&t);
        delta = (t.tv_usec - pctx->prev_t.tv_usec) +
                (t.tv_sec - pctx->prev_t.tv_sec)*1000000;
        if (delta * pctx->pollhz < 500000)
                pctx->short_ticks++;
        else
                pctx->prev_t = t;

        if (pctx->pending_polls > 100) {
                /*
                 * Too much, assume it has stalled (not always true
                 * see comment above).
                 */
                pctx->stalled++;
                pctx->pending_polls = 0;
                pctx->phase = 0;
        }

        if (pctx->phase <= 2) {
                if (pctx->phase != 0)
                        pctx->suspect++;
                pctx->phase = 1;
                schedpoll(pctx);
                pctx->phase = 2;
        }
        if (pctx->pending_polls++ > 0)
                pctx->lost_polls++;
}

/*
 * netisr_pollmore is called after other netisr's, possibly scheduling
 * another NETISR_POLL call, or adapting the burst size for the next cycle.
 *
 * It is very bad to fetch large bursts of packets from a single card at once,
 * because the burst could take a long time to be completely processed leading
 * to unfairness. To reduce the problem, and also to account better for time
 * spent in network-related processing, we split the burst in smaller chunks
 * of fixed size, giving control to the other netisr's between chunks.  This
 * helps in improving the fairness, reducing livelock (because we emulate more
 * closely the "process to completion" that we have with fastforwarding) and
 * accounting for the work performed in low level handling and forwarding.
 */

/* ARGSUSED */
static void
netisr_pollmore(struct netmsg *msg)
{
        struct pollctx *pctx;
        struct timeval t;
        int kern_load, cpuid;
        uint32_t pending_polls;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);
        KKASSERT(pctx == msg->nm_lmsg.u.ms_resultp);

        lwkt_replymsg(&msg->nm_lmsg, 0);

        if (pctx->poll_handlers == 0)
                return;

        KASSERT(pctx->polling_enabled,
                ("# of registered poll handlers are not zero, "
                 "but polling is not enabled\n"));

        pctx->phase = 5;
        if (pctx->residual_burst > 0) {
                schedpoll(pctx);
                /* will run immediately on return, followed by netisrs */
                return;
        }
        /* here we can account time spent in netisr's in this tick */
        microuptime(&t);
        kern_load = (t.tv_usec - pctx->poll_start_t.tv_usec) +
                (t.tv_sec - pctx->poll_start_t.tv_sec)*1000000; /* us */
        kern_load = (kern_load * pctx->pollhz) / 10000;         /* 0..100 */
        if (kern_load > (100 - pctx->user_frac)) { /* try decrease ticks */
                if (pctx->poll_burst > 1)
                        pctx->poll_burst--;
        } else {
                if (pctx->poll_burst < pctx->poll_burst_max)
                        pctx->poll_burst++;
        }

        crit_enter();
        pctx->pending_polls--;
        pending_polls = pctx->pending_polls;
        crit_exit();

        if (pending_polls == 0) {       /* we are done */
                pctx->phase = 0;
        } else {
                /*
                 * Last cycle was long and caused us to miss one or more
                 * hardclock ticks. Restart processing again, but slightly
                 * reduce the burst size to prevent that this happens again.
                 */
                pctx->poll_burst -= (pctx->poll_burst / 8);
                if (pctx->poll_burst < 1)
                        pctx->poll_burst = 1;
                schedpoll(pctx);
                pctx->phase = 6;
        }
}

/*
 * netisr_poll is scheduled by schedpoll when appropriate, typically once
 * per polling systimer tick.
 *
 * Note that the message is replied immediately in order to allow a new
 * ISR to be scheduled in the handler.
 *
 * XXX each registration should indicate whether it needs a critical
 * section to operate.
 */
/* ARGSUSED */
static void
netisr_poll(struct netmsg *msg)
{
        struct pollctx *pctx;
        int i, cycles, cpuid;
        enum poll_cmd arg = POLL_ONLY;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);
        KKASSERT(pctx == msg->nm_lmsg.u.ms_resultp);

        crit_enter();
        lwkt_replymsg(&msg->nm_lmsg, 0);
        crit_exit();

        if (pctx->poll_handlers == 0)
                return;

        KASSERT(pctx->polling_enabled,
                ("# of registered poll handlers are not zero, "
                 "but polling is not enabled\n"));

        pctx->phase = 3;
        if (pctx->residual_burst == 0) { /* first call in this tick */
                microuptime(&pctx->poll_start_t);

                if (pctx->reg_frac_count-- == 0) {
                        arg = POLL_AND_CHECK_STATUS;
                        pctx->reg_frac_count = pctx->reg_frac - 1;
                }

                pctx->residual_burst = pctx->poll_burst;
        }
        cycles = (pctx->residual_burst < pctx->poll_each_burst) ?
                pctx->residual_burst : pctx->poll_each_burst;
        pctx->residual_burst -= cycles;

        for (i = 0 ; i < pctx->poll_handlers ; i++) {
                struct ifnet *ifp = pctx->pr[i].ifp;

                if (!ifnet_tryserialize_main(ifp))
                        continue;

                if ((ifp->if_flags & (IFF_UP|IFF_RUNNING|IFF_POLLING))
                    == (IFF_UP|IFF_RUNNING|IFF_POLLING)) {
                        logpoll(beg, ifp);
                        crit_enter();
                        ifp->if_poll(ifp, arg, cycles);
                        crit_exit();
                        logpoll(end, ifp);
                }

                ifnet_deserialize_main(ifp);
        }

        schedpollmore(pctx);
        pctx->phase = 4;
}

static void
poll_register(struct netmsg *msg)
{
        struct ifnet *ifp = msg->nm_lmsg.u.ms_resultp;
        struct pollctx *pctx;
        int rc, cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        if (pctx->polling_enabled == 0) {
                /* Polling disabled, cannot register */
                rc = EOPNOTSUPP;
                goto back;
        }

        /*
         * Check if there is room.
         */
        if (pctx->poll_handlers >= POLL_LIST_LEN) {
                /*
                 * List full, cannot register more entries.
                 * This should never happen; if it does, it is probably a
                 * broken driver trying to register multiple times. Checking
                 * this at runtime is expensive, and won't solve the problem
                 * anyways, so just report a few times and then give up.
                 */
                static int verbose = 10;        /* XXX */
                if (verbose >0) {
                        kprintf("poll handlers list full, "
                                "maybe a broken driver ?\n");
                        verbose--;
                }
                rc = ENOMEM;
        } else {
                pctx->pr[pctx->poll_handlers].ifp = ifp;
                pctx->poll_handlers++;
                rc = 0;

                if (pctx->poll_handlers == 1) {
                        KKASSERT(pctx->polling_enabled);
                        systimer_adjust_periodic(&pctx->pollclock,
                                                 pctx->pollhz);
                }
        }
back:
        lwkt_replymsg(&msg->nm_lmsg, rc);
}

/*
 * Try to register routine for polling. Returns 1 if successful
 * (and polling should be enabled), 0 otherwise.
 *
 * Called from mainline code only, not called from an interrupt.
 */
int
ether_poll_register(struct ifnet *ifp)
{
        if (poll_defcpu < 0)
                return 0;
        KKASSERT(poll_defcpu < POLLCTX_MAX);

        return ether_pollcpu_register(ifp, poll_defcpu);
}

int
ether_pollcpu_register(struct ifnet *ifp, int cpuid)
{
        struct netmsg msg;
        lwkt_port_t port;
        int rc;

        if (ifp->if_poll == NULL) {
                /* Device does not support polling */
                return 0;
        }

        if (cpuid < 0 || cpuid >= POLLCTX_MAX)
                return 0;

        if (((1 << cpuid) & poll_cpumask) == 0) {
                /* Polling is not supported on 'cpuid' */
                return 0;
        }
        KKASSERT(poll_context[cpuid] != NULL);

        /*
         * Attempt to register.  Interlock with IFF_POLLING.
         */
        crit_enter();   /* XXX MP - not mp safe */

        ifnet_serialize_all(ifp);
        if (ifp->if_flags & IFF_POLLING) {
                /* Already polling */
                KKASSERT(ifp->if_poll_cpuid >= 0);
                ifnet_deserialize_all(ifp);
                crit_exit();
                return 0;
        }
        KKASSERT(ifp->if_poll_cpuid < 0);
        ifp->if_flags |= IFF_POLLING;
        ifp->if_poll_cpuid = cpuid;
        if (ifp->if_flags & IFF_RUNNING)
                ifp->if_poll(ifp, POLL_REGISTER, 0);
        ifnet_deserialize_all(ifp);

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_register);
        msg.nm_lmsg.u.ms_resultp = ifp;

        port = cpu_portfn(cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);

        if (msg.nm_lmsg.ms_error) {
                ifnet_serialize_all(ifp);
                ifp->if_flags &= ~IFF_POLLING;
                ifp->if_poll_cpuid = -1;
                if (ifp->if_flags & IFF_RUNNING)
                        ifp->if_poll(ifp, POLL_DEREGISTER, 0);
                ifnet_deserialize_all(ifp);
                rc = 0;
        } else {
                rc = 1;
        }

        crit_exit();
        return rc;
}

static void
poll_deregister(struct netmsg *msg)
{
        struct ifnet *ifp = msg->nm_lmsg.u.ms_resultp;
        struct pollctx *pctx;
        int rc, i, cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        for (i = 0 ; i < pctx->poll_handlers ; i++) {
                if (pctx->pr[i].ifp == ifp) /* Found it */
                        break;
        }
        if (i == pctx->poll_handlers) {
                kprintf("ether_poll_deregister: ifp not found!!!\n");
                rc = ENOENT;
        } else {
                pctx->poll_handlers--;
                if (i < pctx->poll_handlers) {
                        /* Last entry replaces this one. */
                        pctx->pr[i].ifp = pctx->pr[pctx->poll_handlers].ifp;
                }

                if (pctx->poll_handlers == 0) {
                        systimer_adjust_periodic(&pctx->pollclock, 1);
                        poll_reset_state(pctx);
                }
                rc = 0;
        }
        lwkt_replymsg(&msg->nm_lmsg, rc);
}

/*
 * Remove interface from the polling list.  Occurs when polling is turned
 * off.  Called from mainline code only, not called from an interrupt.
 */
int
ether_poll_deregister(struct ifnet *ifp)
{
        struct netmsg msg;
        lwkt_port_t port;
        int rc, cpuid;

        KKASSERT(ifp != NULL);

        if (ifp->if_poll == NULL)
                return 0;

        crit_enter();

        ifnet_serialize_all(ifp);
        if ((ifp->if_flags & IFF_POLLING) == 0) {
                KKASSERT(ifp->if_poll_cpuid < 0);
                ifnet_deserialize_all(ifp);
                crit_exit();
                return 0;
        }

        cpuid = ifp->if_poll_cpuid;
        KKASSERT(cpuid >= 0);
        KKASSERT(poll_context[cpuid] != NULL);

        ifp->if_flags &= ~IFF_POLLING;
        ifp->if_poll_cpuid = -1;
        ifnet_deserialize_all(ifp);

        netmsg_init(&msg, NULL, &curthread->td_msgport,
                    MSGF_MPSAFE, poll_deregister);
        msg.nm_lmsg.u.ms_resultp = ifp;

        port = cpu_portfn(cpuid);
        lwkt_domsg(port, &msg.nm_lmsg, 0);

        if (!msg.nm_lmsg.ms_error) {
                ifnet_serialize_all(ifp);
                if (ifp->if_flags & IFF_RUNNING)
                        ifp->if_poll(ifp, POLL_DEREGISTER, 1);
                ifnet_deserialize_all(ifp);
                rc = 1;
        } else {
                rc = 0;
        }

        crit_exit();
        return rc;
}

static void
poll_add_sysctl(struct sysctl_ctx_list *ctx, struct sysctl_oid_list *parent,
                struct pollctx *pctx)
{
        SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, "enable",
                        CTLTYPE_INT | CTLFLAG_RW, pctx, 0, sysctl_polling,
                        "I", "Polling enabled");

        SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, "pollhz",
                        CTLTYPE_INT | CTLFLAG_RW, pctx, 0, sysctl_pollhz,
                        "I", "Device polling frequency");

        SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, "reg_frac",
                        CTLTYPE_UINT | CTLFLAG_RW, pctx, 0, sysctl_regfrac,
                        "IU", "Every this many cycles poll register");

        SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, "burst_max",
                        CTLTYPE_UINT | CTLFLAG_RW, pctx, 0, sysctl_burstmax,
                        "IU", "Max Polling burst size");

        SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, "each_burst",
                        CTLTYPE_UINT | CTLFLAG_RW, pctx, 0, sysctl_eachburst,
                        "IU", "Max size of each burst");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "phase", CTLFLAG_RD,
                        &pctx->phase, 0, "Polling phase");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "suspect", CTLFLAG_RW,
                        &pctx->suspect, 0, "suspect event");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "stalled", CTLFLAG_RW,
                        &pctx->stalled, 0, "potential stalls");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "burst", CTLFLAG_RD,
                        &pctx->poll_burst, 0, "Current polling burst size");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "user_frac", CTLFLAG_RW,
                        &pctx->user_frac, 0,
                        "Desired user fraction of cpu time");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "short_ticks", CTLFLAG_RW,
                        &pctx->short_ticks, 0,
                        "Hardclock ticks shorter than they should be");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "lost_polls", CTLFLAG_RW,
                        &pctx->lost_polls, 0,
                        "How many times we would have lost a poll tick");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "pending_polls", CTLFLAG_RD,
                        &pctx->pending_polls, 0, "Do we need to poll again");

        SYSCTL_ADD_INT(ctx, parent, OID_AUTO, "residual_burst", CTLFLAG_RD,
                       &pctx->residual_burst, 0,
                       "# of residual cycles in burst");

        SYSCTL_ADD_UINT(ctx, parent, OID_AUTO, "handlers", CTLFLAG_RD,
                        &pctx->poll_handlers, 0,
                        "Number of registered poll handlers");
}

static void
poll_sysctl_pollhz(struct netmsg *msg)
{
        struct pollctx *pctx;
        int cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        /*
         * If polling is disabled or there is no device registered,
         * don't adjust polling systimer frequency.
         * Polling systimer frequency will be adjusted once polling
         * is enabled and there are registered devices.
         */
        pctx->pollhz = msg->nm_lmsg.u.ms_result;
        if (pctx->polling_enabled && pctx->poll_handlers)
                systimer_adjust_periodic(&pctx->pollclock, pctx->pollhz);

        /*
         * Make sure that reg_frac and reg_frac_count are within valid range.
         */
        if (pctx->reg_frac > pctx->pollhz) {
                pctx->reg_frac = pctx->pollhz;
                if (pctx->reg_frac_count > pctx->reg_frac)
                        pctx->reg_frac_count = pctx->reg_frac - 1;
        }

        lwkt_replymsg(&msg->nm_lmsg, 0);
}

static void
poll_sysctl_polling(struct netmsg *msg)
{
        struct pollctx *pctx;
        int cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        /*
         * If polling is disabled or there is no device registered,
         * cut the polling systimer frequency to 1hz.
         */
        pctx->polling_enabled = msg->nm_lmsg.u.ms_result;
        if (pctx->polling_enabled && pctx->poll_handlers) {
                systimer_adjust_periodic(&pctx->pollclock, pctx->pollhz);
        } else {
                systimer_adjust_periodic(&pctx->pollclock, 1);
                poll_reset_state(pctx);
        }

        if (!pctx->polling_enabled && pctx->poll_handlers != 0) {
                int i;

                for (i = 0 ; i < pctx->poll_handlers ; i++) {
                        struct ifnet *ifp = pctx->pr[i].ifp;

                        ifnet_serialize_all(ifp);

                        if ((ifp->if_flags & IFF_POLLING) == 0) {
                                KKASSERT(ifp->if_poll_cpuid < 0);
                                ifnet_deserialize_all(ifp);
                                continue;
                        }
                        ifp->if_flags &= ~IFF_POLLING;
                        ifp->if_poll_cpuid = -1;

                        /*
                         * Only call the interface deregistration
                         * function if the interface is still 
                         * running.
                         */
                        if (ifp->if_flags & IFF_RUNNING)
                                ifp->if_poll(ifp, POLL_DEREGISTER, 1);

                        ifnet_deserialize_all(ifp);
                }
                pctx->poll_handlers = 0;
        }

        lwkt_replymsg(&msg->nm_lmsg, 0);
}

static void
poll_sysctl_regfrac(struct netmsg *msg)
{
        struct pollctx *pctx;
        uint32_t reg_frac;
        int cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        reg_frac = msg->nm_lmsg.u.ms_result;
        if (reg_frac > pctx->pollhz)
                reg_frac = pctx->pollhz;
        else if (reg_frac < 1)
                reg_frac = 1;

        pctx->reg_frac = reg_frac;
        if (pctx->reg_frac_count > pctx->reg_frac)
                pctx->reg_frac_count = pctx->reg_frac - 1;

        lwkt_replymsg(&msg->nm_lmsg, 0);
}

static void
poll_sysctl_burstmax(struct netmsg *msg)
{
        struct pollctx *pctx;
        int cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        pctx->poll_burst_max = msg->nm_lmsg.u.ms_result;
        if (pctx->poll_each_burst > pctx->poll_burst_max)
                pctx->poll_each_burst = pctx->poll_burst_max;
        if (pctx->poll_burst > pctx->poll_burst_max)
                pctx->poll_burst = pctx->poll_burst_max;
        if (pctx->residual_burst > pctx->poll_burst_max)
                pctx->residual_burst = pctx->poll_burst_max;

        lwkt_replymsg(&msg->nm_lmsg, 0);
}

static void
poll_sysctl_eachburst(struct netmsg *msg)
{
        struct pollctx *pctx;
        uint32_t each_burst;
        int cpuid;

        cpuid = mycpu->gd_cpuid;
        KKASSERT(cpuid < POLLCTX_MAX);

        pctx = poll_context[cpuid];
        KKASSERT(pctx != NULL);
        KKASSERT(pctx->poll_cpuid == cpuid);

        each_burst = msg->nm_lmsg.u.ms_result;
        if (each_burst > pctx->poll_burst_max)
                each_burst = pctx->poll_burst_max;
        else if (each_burst < 1)
                each_burst = 1;
        pctx->poll_each_burst = each_burst;

        lwkt_replymsg(&msg->nm_lmsg, 0);
}