undinet.c

Go to the documentation of this file.

/*
 * Copyright (C) 2007 Michael Brown <mbrown@fensystems.co.uk>.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 of the
 * License, or any later version.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
 * 02110-1301, USA.
 */

FILE_LICENCE ( GPL2_OR_LATER );

#include <string.h>
#include <unistd.h>
#include <byteswap.h>
#include <pxe.h>
#include <realmode.h>
#include <pic8259.h>
#include <biosint.h>
#include <pnpbios.h>
#include <basemem_packet.h>
#include <ipxe/io.h>
#include <ipxe/iobuf.h>
#include <ipxe/netdevice.h>
#include <ipxe/if_ether.h>
#include <ipxe/ethernet.h>
#include <ipxe/pci.h>
#include <ipxe/profile.h>
#include <undi.h>
#include <undinet.h>

/** @file
 *
 * UNDI network device driver
 *
 */

/** An UNDI NIC */
struct undi_nic {
        /** Device supports IRQs */
        int irq_supported;
        /** Assigned IRQ number */
        unsigned int irq;
        /** Currently processing ISR */
        int isr_processing;
        /** Bug workarounds */
        int hacks;
};

/* Disambiguate the various error causes */
#define EINFO_EPXECALL                                                  \
        __einfo_uniqify ( EINFO_EPLATFORM, 0x01,                        \
                          "External PXE API error" )
#define EPXECALL( status ) EPLATFORM ( EINFO_EPXECALL, status )

/**
 * @defgroup undi_hacks UNDI workarounds
 * @{
 */

/** Work around Etherboot 5.4 bugs */
#define UNDI_HACK_EB54          0x0001

/** @} */

/** Maximum number of times to retry PXENV_UNDI_INITIALIZE */
#define UNDI_INITIALIZE_RETRY_MAX 10

/** Delay between retries of PXENV_UNDI_INITIALIZE */
#define UNDI_INITIALIZE_RETRY_DELAY_MS 200

/** Maximum number of received packets per poll */
#define UNDI_RX_QUOTA 4

/** Alignment of received frame payload */
#define UNDI_RX_ALIGN 16

static void undinet_close ( struct net_device *netdev );

/**
 * UNDI parameter block
 *
 * Used as the parameter block for all UNDI API calls.  Resides in
 * base memory.
 */
static union u_PXENV_ANY __bss16 ( undinet_params );
#define undinet_params __use_data16 ( undinet_params )

/**
 * UNDI entry point
 *
 * Used as the indirection vector for all UNDI API calls.  Resides in
 * base memory.
 */
SEGOFF16_t __bss16 ( undinet_entry_point );
#define undinet_entry_point __use_data16 ( undinet_entry_point )

/* Read TSC in real mode only when profiling */
#if PROFILING
#define RDTSC_IF_PROFILING "rdtsc\n\t"
#else
#define RDTSC_IF_PROFILING ""
#endif

/** IRQ profiler */
static struct profiler undinet_irq_profiler __profiler =
        { .name = "undinet.irq" };

/** Receive profiler */
static struct profiler undinet_rx_profiler __profiler =
        { .name = "undinet.rx" };

/** A PXE API call breakdown profiler */
struct undinet_profiler {
        /** Total time spent performing REAL_CALL() */
        struct profiler total;
        /** Time spent transitioning to real mode */
        struct profiler p2r;
        /** Time spent in external code */
        struct profiler ext;
        /** Time spent transitioning back to protected mode */
        struct profiler r2p;
};

/** PXENV_UNDI_TRANSMIT profiler */
static struct undinet_profiler undinet_tx_profiler __profiler = {
        { .name = "undinet.tx" },
        { .name = "undinet.tx_p2r" },
        { .name = "undinet.tx_ext" },
        { .name = "undinet.tx_r2p" },
};

/** PXENV_UNDI_ISR profiler
 *
 * Note that this profiler will not see calls to
 * PXENV_UNDI_ISR_IN_START, which are handled by the UNDI ISR and do
 * not go via undinet_call().
 */
static struct undinet_profiler undinet_isr_profiler __profiler = {
        { .name = "undinet.isr" },
        { .name = "undinet.isr_p2r" },
        { .name = "undinet.isr_ext" },
        { .name = "undinet.isr_r2p" },
};

/** PXE unknown API call profiler
 *
 * This profiler can be used to measure the overhead of a dummy PXE
 * API call.
 */
static struct undinet_profiler undinet_unknown_profiler __profiler = {
        { .name = "undinet.unknown" },
        { .name = "undinet.unknown_p2r" },
        { .name = "undinet.unknown_ext" },
        { .name = "undinet.unknown_r2p" },
};

/** Miscellaneous PXE API call profiler */
static struct undinet_profiler undinet_misc_profiler __profiler = {
        { .name = "undinet.misc" },
        { .name = "undinet.misc_p2r" },
        { .name = "undinet.misc_ext" },
        { .name = "undinet.misc_r2p" },
};

/*****************************************************************************
 *
 * UNDI API call
 *
 *****************************************************************************
 */

/**
 * Name PXE API call
 *
 * @v function          API call number
 * @ret name            API call name
 */
static inline __attribute__ (( always_inline )) const char *
undinet_function_name ( unsigned int function ) {
        switch ( function ) {
        case PXENV_START_UNDI:
                return "PXENV_START_UNDI";
        case PXENV_STOP_UNDI:
                return "PXENV_STOP_UNDI";
        case PXENV_UNDI_STARTUP:
                return "PXENV_UNDI_STARTUP";
        case PXENV_UNDI_CLEANUP:
                return "PXENV_UNDI_CLEANUP";
        case PXENV_UNDI_INITIALIZE:
                return "PXENV_UNDI_INITIALIZE";
        case PXENV_UNDI_RESET_ADAPTER:
                return "PXENV_UNDI_RESET_ADAPTER";
        case PXENV_UNDI_SHUTDOWN:
                return "PXENV_UNDI_SHUTDOWN";
        case PXENV_UNDI_OPEN:
                return "PXENV_UNDI_OPEN";
        case PXENV_UNDI_CLOSE:
                return "PXENV_UNDI_CLOSE";
        case PXENV_UNDI_TRANSMIT:
                return "PXENV_UNDI_TRANSMIT";
        case PXENV_UNDI_SET_MCAST_ADDRESS:
                return "PXENV_UNDI_SET_MCAST_ADDRESS";
        case PXENV_UNDI_SET_STATION_ADDRESS:
                return "PXENV_UNDI_SET_STATION_ADDRESS";
        case PXENV_UNDI_SET_PACKET_FILTER:
                return "PXENV_UNDI_SET_PACKET_FILTER";
        case PXENV_UNDI_GET_INFORMATION:
                return "PXENV_UNDI_GET_INFORMATION";
        case PXENV_UNDI_GET_STATISTICS:
                return "PXENV_UNDI_GET_STATISTICS";
        case PXENV_UNDI_CLEAR_STATISTICS:
                return "PXENV_UNDI_CLEAR_STATISTICS";
        case PXENV_UNDI_INITIATE_DIAGS:
                return "PXENV_UNDI_INITIATE_DIAGS";
        case PXENV_UNDI_FORCE_INTERRUPT:
                return "PXENV_UNDI_FORCE_INTERRUPT";
        case PXENV_UNDI_GET_MCAST_ADDRESS:
                return "PXENV_UNDI_GET_MCAST_ADDRESS";
        case PXENV_UNDI_GET_NIC_TYPE:
                return "PXENV_UNDI_GET_NIC_TYPE";
        case PXENV_UNDI_GET_IFACE_INFO:
                return "PXENV_UNDI_GET_IFACE_INFO";
        /*
         * Duplicate case value; this is a bug in the PXE specification.
         *
         *      case PXENV_UNDI_GET_STATE:
         *              return "PXENV_UNDI_GET_STATE";
         */
        case PXENV_UNDI_ISR:
                return "PXENV_UNDI_ISR";
        case PXENV_GET_CACHED_INFO:
                return "PXENV_GET_CACHED_INFO";
        default:
                return "UNKNOWN API CALL";
        }
}

/**
 * Determine applicable profiler pair (for debugging)
 *
 * @v function          API call number
 * @ret profiler        Profiler
 */
static struct undinet_profiler * undinet_profiler ( unsigned int function ) {

        /* Determine applicable profiler */
        switch ( function ) {
        case PXENV_UNDI_TRANSMIT:
                return &undinet_tx_profiler;
        case PXENV_UNDI_ISR:
                return &undinet_isr_profiler;
        case PXENV_UNKNOWN:
                return &undinet_unknown_profiler;
        default:
                return &undinet_misc_profiler;
        }
}

/**
 * Issue UNDI API call
 *
 * @v undinic           UNDI NIC
 * @v function          API call number
 * @v params            PXE parameter block
 * @v params_len        Length of PXE parameter block
 * @ret rc              Return status code
 */
static int undinet_call ( struct undi_nic *undinic, unsigned int function,
                          void *params, size_t params_len ) {
        struct undinet_profiler *profiler = undinet_profiler ( function );
        PXENV_EXIT_t exit;
        uint32_t before;
        uint32_t started;
        uint32_t stopped;
        uint32_t after;
        int discard_D;
        int rc;

        /* Copy parameter block and entry point */
        assert ( params_len <= sizeof ( undinet_params ) );
        memcpy ( &undinet_params, params, params_len );

        /* Call real-mode entry point.  This calling convention will
         * work with both the !PXE and the PXENV+ entry points.
         */
        profile_start ( &profiler->total );
        __asm__ __volatile__ ( REAL_CODE ( "pushl %%ebp\n\t" /* gcc bug */
                                           RDTSC_IF_PROFILING
                                           "pushl %%eax\n\t"
                                           "pushw %%es\n\t"
                                           "pushw %%di\n\t"
                                           "pushw %%bx\n\t"
                                           "lcall *undinet_entry_point\n\t"
                                           "movw %%ax, %%bx\n\t"
                                           RDTSC_IF_PROFILING
                                           "addw $6, %%sp\n\t"
                                           "popl %%edx\n\t"
                                           "popl %%ebp\n\t" /* gcc bug */ )
                               : "=a" ( stopped ), "=d" ( started ),
                                 "=b" ( exit ), "=D" ( discard_D )
                               : "b" ( function ),
                                 "D" ( __from_data16 ( &undinet_params ) )
                               : "ecx", "esi" );
        profile_stop ( &profiler->total );
        before = profile_started ( &profiler->total );
        after = profile_stopped ( &profiler->total );
        profile_start_at ( &profiler->p2r, before );
        profile_stop_at ( &profiler->p2r, started );
        profile_start_at ( &profiler->ext, started );
        profile_stop_at ( &profiler->ext, stopped );
        profile_start_at ( &profiler->r2p, stopped );
        profile_stop_at ( &profiler->r2p, after );

        /* Determine return status code based on PXENV_EXIT and
         * PXENV_STATUS
         */
        rc = ( ( exit == PXENV_EXIT_SUCCESS ) ?
               0 : -EPXECALL ( undinet_params.Status ) );

        /* If anything goes wrong, print as much debug information as
         * it's possible to give.
         */
        if ( rc != 0 ) {
                SEGOFF16_t rm_params = {
                        .segment = rm_ds,
                        .offset = __from_data16 ( &undinet_params ),
                };

                DBGC ( undinic, "UNDINIC %p %s failed: %s\n", undinic,
                       undinet_function_name ( function ), strerror ( rc ) );
                DBGC ( undinic, "UNDINIC %p parameters at %04x:%04x length "
                       "%#02zx, entry point at %04x:%04x\n", undinic,
                       rm_params.segment, rm_params.offset, params_len,
                       undinet_entry_point.segment,
                       undinet_entry_point.offset );
                DBGC ( undinic, "UNDINIC %p parameters provided:\n", undinic );
                DBGC_HDA ( undinic, rm_params, params, params_len );
                DBGC ( undinic, "UNDINIC %p parameters returned:\n", undinic );
                DBGC_HDA ( undinic, rm_params, &undinet_params, params_len );
        }

        /* Copy parameter block back */
        memcpy ( params, &undinet_params, params_len );

        return rc;
}

/*****************************************************************************
 *
 * UNDI interrupt service routine
 *
 *****************************************************************************
 */

/**
 * UNDI interrupt service routine
 *
 * The UNDI ISR increments a counter (@c trigger_count) and exits.
 */
extern void undiisr ( void );

/** IRQ number */
uint8_t __data16 ( undiisr_irq );
#define undiisr_irq __use_data16 ( undiisr_irq )

/** IRQ mask register */
uint16_t __data16 ( undiisr_imr );
#define undiisr_imr __use_data16 ( undiisr_imr )

/** IRQ mask bit */
uint8_t __data16 ( undiisr_bit );
#define undiisr_bit __use_data16 ( undiisr_bit )

/** IRQ rearm flag */
uint8_t __data16 ( undiisr_rearm );
#define undiisr_rearm __use_data16 ( undiisr_rearm )

/** IRQ chain vector */
struct segoff __data16 ( undiisr_next_handler );
#define undiisr_next_handler __use_data16 ( undiisr_next_handler )

/** IRQ trigger count */
volatile uint8_t __data16 ( undiisr_trigger_count ) = 0;
#define undiisr_trigger_count __use_data16 ( undiisr_trigger_count )

/** Last observed trigger count */
static unsigned int last_trigger_count = 0;

/**
 * Hook UNDI interrupt service routine
 *
 * @v irq               IRQ number
 */
static void undinet_hook_isr ( unsigned int irq ) {

        assert ( irq <= IRQ_MAX );
        assert ( undiisr_irq == 0 );

        undiisr_irq = irq;
        undiisr_imr = IMR_REG ( irq );
        undiisr_bit = IMR_BIT ( irq );
        undiisr_rearm = 0;
        hook_bios_interrupt ( IRQ_INT ( irq ), ( ( intptr_t ) undiisr ),
                              &undiisr_next_handler );
}

/**
 * Unhook UNDI interrupt service routine
 *
 * @v irq               IRQ number
 */
static void undinet_unhook_isr ( unsigned int irq ) {

        assert ( irq <= IRQ_MAX );

        unhook_bios_interrupt ( IRQ_INT ( irq ), ( ( intptr_t ) undiisr ),
                                &undiisr_next_handler );
        undiisr_irq = 0;
}

/**
 * Test to see if UNDI ISR has been triggered
 *
 * @ret triggered       ISR has been triggered since last check
 */
static int undinet_isr_triggered ( void ) {
        unsigned int this_trigger_count;

        /* Read trigger_count.  Do this only once; it is volatile */
        this_trigger_count = undiisr_trigger_count;

        if ( this_trigger_count == last_trigger_count ) {
                /* Not triggered */
                return 0;
        } else {
                /* Triggered */
                last_trigger_count = this_trigger_count;
                return 1;
        }
}

/*****************************************************************************
 *
 * UNDI network device interface
 *
 *****************************************************************************
 */

/** UNDI transmit buffer descriptor */
static struct s_PXENV_UNDI_TBD __data16 ( undinet_tbd );
#define undinet_tbd __use_data16 ( undinet_tbd )

/** UNDI transmit destination address */
static uint8_t __data16_array ( undinet_destaddr, [ETH_ALEN] );
#define undinet_destaddr __use_data16 ( undinet_destaddr )

/**
 * Transmit packet
 *
 * @v netdev            Network device
 * @v iobuf             I/O buffer
 * @ret rc              Return status code
 */
static int undinet_transmit ( struct net_device *netdev,
                              struct io_buffer *iobuf ) {
        struct undi_nic *undinic = netdev->priv;
        struct s_PXENV_UNDI_TRANSMIT undi_transmit;
        const void *ll_dest;
        const void *ll_source;
        uint16_t net_proto;
        unsigned int flags;
        uint8_t protocol;
        size_t len;
        int rc;

        /* Technically, we ought to make sure that the previous
         * transmission has completed before we re-use the buffer.
         * However, many PXE stacks (including at least some Intel PXE
         * stacks and Etherboot 5.4) fail to generate TX completions.
         * In practice this won't be a problem, since our TX datapath
         * has a very low packet volume and we can get away with
         * assuming that a TX will be complete by the time we want to
         * transmit the next packet.
         */

        /* Some PXE stacks are unable to cope with P_UNKNOWN, and will
         * always try to prepend a link-layer header.  Work around
         * these stacks by stripping the existing link-layer header
         * and allowing the PXE stack to (re)construct the link-layer
         * header itself.
         */
        if ( ( rc = eth_pull ( netdev, iobuf, &ll_dest, &ll_source,
                               &net_proto, &flags ) ) != 0 ) {
                DBGC ( undinic, "UNDINIC %p could not strip Ethernet header: "
                       "%s\n", undinic, strerror ( rc ) );
                return rc;
        }
        memcpy ( undinet_destaddr, ll_dest, sizeof ( undinet_destaddr ) );
        switch ( net_proto ) {
        case htons ( ETH_P_IP ) :
                protocol = P_IP;
                break;
        case htons ( ETH_P_ARP ) :
                protocol = P_ARP;
                break;
        case htons ( ETH_P_RARP ) :
                protocol = P_RARP;
                break;
        default:
                /* Unknown protocol; restore the original link-layer header */
                iob_push ( iobuf, sizeof ( struct ethhdr ) );
                protocol = P_UNKNOWN;
                break;
        }

        /* Copy packet to UNDI I/O buffer */
        len = iob_len ( iobuf );
        if ( len > sizeof ( basemem_packet ) )
                len = sizeof ( basemem_packet );
        memcpy ( &basemem_packet, iobuf->data, len );

        /* Create PXENV_UNDI_TRANSMIT data structure */
        memset ( &undi_transmit, 0, sizeof ( undi_transmit ) );
        undi_transmit.Protocol = protocol;
        undi_transmit.XmitFlag = ( ( flags & LL_BROADCAST ) ?
                                   XMT_BROADCAST : XMT_DESTADDR );
        undi_transmit.DestAddr.segment = rm_ds;
        undi_transmit.DestAddr.offset = __from_data16 ( &undinet_destaddr );
        undi_transmit.TBD.segment = rm_ds;
        undi_transmit.TBD.offset = __from_data16 ( &undinet_tbd );

        /* Create PXENV_UNDI_TBD data structure */
        undinet_tbd.ImmedLength = len;
        undinet_tbd.Xmit.segment = rm_ds;
        undinet_tbd.Xmit.offset = __from_data16 ( basemem_packet );

        /* Issue PXE API call */
        if ( ( rc = undinet_call ( undinic, PXENV_UNDI_TRANSMIT, &undi_transmit,
                                   sizeof ( undi_transmit ) ) ) != 0 )
                goto done;

        /* Free I/O buffer */
        netdev_tx_complete ( netdev, iobuf );
 done:
        return rc;
}

/** 
 * Poll for received packets
 *
 * @v netdev            Network device
 *
 * Fun, fun, fun.  UNDI drivers don't use polling; they use
 * interrupts.  We therefore cheat and pretend that an interrupt has
 * occurred every time undinet_poll() is called.  This isn't too much
 * of a hack; PCI devices share IRQs and so the first thing that a
 * proper ISR should do is call PXENV_UNDI_ISR to determine whether or
 * not the UNDI NIC generated the interrupt; there is no harm done by
 * spurious calls to PXENV_UNDI_ISR.  Similarly, we wouldn't be
 * handling them any more rapidly than the usual rate of
 * undinet_poll() being called even if we did implement a full ISR.
 * So it should work.  Ha!
 *
 * Addendum (21/10/03).  Some cards don't play nicely with this trick,
 * so instead of doing it the easy way we have to go to all the hassle
 * of installing a genuine interrupt service routine and dealing with
 * the wonderful 8259 Programmable Interrupt Controller.  Joy.
 *
 * Addendum (10/07/07).  When doing things such as iSCSI boot, in
 * which we have to co-operate with a running OS, we can't get away
 * with the "ISR-just-increments-a-counter-and-returns" trick at all,
 * because it involves tying up the PIC for far too long, and other
 * interrupt-dependent components (e.g. local disks) start breaking.
 * We therefore implement a "proper" ISR which calls PXENV_UNDI_ISR
 * from within interrupt context in order to deassert the device
 * interrupt, and sends EOI if applicable.
 */
static void undinet_poll ( struct net_device *netdev ) {
        struct undi_nic *undinic = netdev->priv;
        struct s_PXENV_UNDI_ISR undi_isr;
        struct io_buffer *iobuf = NULL;
        unsigned int quota = UNDI_RX_QUOTA;
        size_t len;
        size_t reserve_len;
        size_t frag_len;
        size_t max_frag_len;
        int rc;

        if ( ! undinic->isr_processing ) {
                /* Allow interrupt to occur.  Do this even if
                 * interrupts are not known to be supported, since
                 * some cards erroneously report that they do not
                 * support interrupts.
                 */
                if ( ! undinet_isr_triggered() ) {

                        /* Rearm interrupt if needed */
                        if ( undiisr_rearm ) {
                                undiisr_rearm = 0;
                                assert ( undinic->irq != 0 );
                                enable_irq ( undinic->irq );
                        }

                        /* Allow interrupt to occur */
                        profile_start ( &undinet_irq_profiler );
                        __asm__ __volatile__ ( "sti\n\t"
                                               "nop\n\t"
                                               "nop\n\t"
                                               "cli\n\t" );
                        profile_stop ( &undinet_irq_profiler );

                        /* If interrupts are known to be supported,
                         * then do nothing on this poll; wait for the
                         * interrupt to be triggered.
                         */
                        if ( undinic->irq_supported )
                                return;
                }

                /* Start ISR processing */
                undinic->isr_processing = 1;
                undi_isr.FuncFlag = PXENV_UNDI_ISR_IN_PROCESS;
        } else {
                /* Continue ISR processing */
                undi_isr.FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
        }

        /* Run through the ISR loop */
        while ( quota ) {
                if ( ( rc = undinet_call ( undinic, PXENV_UNDI_ISR, &undi_isr,
                                           sizeof ( undi_isr ) ) ) != 0 ) {
                        netdev_rx_err ( netdev, NULL, rc );
                        break;
                }
                switch ( undi_isr.FuncFlag ) {
                case PXENV_UNDI_ISR_OUT_TRANSMIT:
                        /* We don't care about transmit completions */
                        break;
                case PXENV_UNDI_ISR_OUT_RECEIVE:
                        /* Packet fragment received */
                        profile_start ( &undinet_rx_profiler );
                        len = undi_isr.FrameLength;
                        frag_len = undi_isr.BufferLength;
                        reserve_len = ( -undi_isr.FrameHeaderLength &
                                        ( UNDI_RX_ALIGN - 1 ) );
                        if ( ( len == 0 ) || ( len < frag_len ) ) {
                                /* Don't laugh.  VMWare does it. */
                                DBGC ( undinic, "UNDINIC %p reported insane "
                                       "fragment (%zd of %zd bytes)\n",
                                       undinic, frag_len, len );
                                netdev_rx_err ( netdev, NULL, -EINVAL );
                                break;
                        }
                        if ( ! iobuf ) {
                                iobuf = alloc_iob ( reserve_len + len );
                                if ( ! iobuf ) {
                                        DBGC ( undinic, "UNDINIC %p could not "
                                               "allocate %zd bytes for RX "
                                               "buffer\n", undinic, len );
                                        /* Fragment will be dropped */
                                        netdev_rx_err ( netdev, NULL, -ENOMEM );
                                        goto done;
                                }
                                iob_reserve ( iobuf, reserve_len );
                        }
                        max_frag_len = iob_tailroom ( iobuf );
                        if ( frag_len > max_frag_len ) {
                                DBGC ( undinic, "UNDINIC %p fragment too big "
                                       "(%zd+%zd does not fit into %zd)\n",
                                       undinic, iob_len ( iobuf ), frag_len,
                                       ( iob_len ( iobuf ) + max_frag_len ) );
                                frag_len = max_frag_len;
                        }
                        copy_from_real ( iob_put ( iobuf, frag_len ),
                                         undi_isr.Frame.segment,
                                         undi_isr.Frame.offset, frag_len );
                        if ( iob_len ( iobuf ) == len ) {
                                /* Whole packet received; deliver it */
                                netdev_rx ( netdev, iob_disown ( iobuf ) );
                                quota--;
                                /* Etherboot 5.4 fails to return all packets
                                 * under mild load; pretend it retriggered.
                                 */
                                if ( undinic->hacks & UNDI_HACK_EB54 )
                                        --last_trigger_count;
                        }
                        profile_stop ( &undinet_rx_profiler );
                        break;
                case PXENV_UNDI_ISR_OUT_DONE:
                        /* Processing complete */
                        undinic->isr_processing = 0;
                        goto done;
                default:
                        /* Should never happen.  VMWare does it routinely. */
                        DBGC ( undinic, "UNDINIC %p ISR returned invalid "
                               "FuncFlag %04x\n", undinic, undi_isr.FuncFlag );
                        undinic->isr_processing = 0;
                        goto done;
                }
                undi_isr.FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
        }

 done:
        if ( iobuf ) {
                DBGC ( undinic, "UNDINIC %p returned incomplete packet "
                       "(%zd of %zd)\n", undinic, iob_len ( iobuf ),
                       ( iob_len ( iobuf ) + iob_tailroom ( iobuf ) ) );
                netdev_rx_err ( netdev, iobuf, -EINVAL );
        }
}

/**
 * Open NIC
 *
 * @v netdev            Net device
 * @ret rc              Return status code
 */
static int undinet_open ( struct net_device *netdev ) {
        struct undi_nic *undinic = netdev->priv;
        struct s_PXENV_UNDI_SET_STATION_ADDRESS undi_set_address;
        struct s_PXENV_UNDI_OPEN undi_open;
        int rc;

        /* Hook interrupt service routine and enable interrupt if applicable */
        if ( undinic->irq ) {
                undinet_hook_isr ( undinic->irq );
                enable_irq ( undinic->irq );
                send_eoi ( undinic->irq );
        }

        /* Set station address.  Required for some PXE stacks; will
         * spuriously fail on others.  Ignore failures.  We only ever
         * use it to set the MAC address to the card's permanent value
         * anyway.
         */
        memcpy ( undi_set_address.StationAddress, netdev->ll_addr,
                 sizeof ( undi_set_address.StationAddress ) );
        undinet_call ( undinic, PXENV_UNDI_SET_STATION_ADDRESS,
                       &undi_set_address, sizeof ( undi_set_address ) );

        /* Open NIC.  We ask for promiscuous operation, since it's the
         * only way to ask for all multicast addresses.  On any
         * switched network, it shouldn't really make a difference to
         * performance.
         */
        memset ( &undi_open, 0, sizeof ( undi_open ) );
        undi_open.PktFilter = ( FLTR_DIRECTED | FLTR_BRDCST | FLTR_PRMSCS );
        if ( ( rc = undinet_call ( undinic, PXENV_UNDI_OPEN, &undi_open,
                                   sizeof ( undi_open ) ) ) != 0 )
                goto err;

        DBGC ( undinic, "UNDINIC %p opened\n", undinic );
        return 0;

 err:
        undinet_close ( netdev );
        return rc;
}

/**
 * Close NIC
 *
 * @v netdev            Net device
 */
static void undinet_close ( struct net_device *netdev ) {
        struct undi_nic *undinic = netdev->priv;
        struct s_PXENV_UNDI_ISR undi_isr;
        struct s_PXENV_UNDI_CLOSE undi_close;
        int rc;

        /* Ensure ISR has exited cleanly */
        while ( undinic->isr_processing ) {
                undi_isr.FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
                if ( ( rc = undinet_call ( undinic, PXENV_UNDI_ISR, &undi_isr,
                                           sizeof ( undi_isr ) ) ) != 0 )
                        break;
                switch ( undi_isr.FuncFlag ) {
                case PXENV_UNDI_ISR_OUT_TRANSMIT:
                case PXENV_UNDI_ISR_OUT_RECEIVE:
                        /* Continue draining */
                        break;
                default:
                        /* Stop processing */
                        undinic->isr_processing = 0;
                        break;
                }
        }

        /* Close NIC */
        undinet_call ( undinic, PXENV_UNDI_CLOSE, &undi_close,
                       sizeof ( undi_close ) );

        /* Disable interrupt and unhook ISR if applicable */
        if ( undinic->irq ) {
                disable_irq ( undinic->irq );
                undinet_unhook_isr ( undinic->irq );
        }

        DBGC ( undinic, "UNDINIC %p closed\n", undinic );
}

/**
 * Enable/disable interrupts
 *
 * @v netdev            Net device
 * @v enable            Interrupts should be enabled
 */
static void undinet_irq ( struct net_device *netdev, int enable ) {
        struct undi_nic *undinic = netdev->priv;

        /* Cannot support interrupts yet */
        DBGC ( undinic, "UNDINIC %p cannot %s interrupts\n",
               undinic, ( enable ? "enable" : "disable" ) );
}

/** UNDI network device operations */
static struct net_device_operations undinet_operations = {
        .open           = undinet_open,
        .close          = undinet_close,
        .transmit       = undinet_transmit,
        .poll           = undinet_poll,
        .irq            = undinet_irq,
};

/** A device with broken support for generating interrupts */
struct undinet_irq_broken {
        /** PCI vendor ID */
        uint16_t pci_vendor;
        /** PCI device ID */
        uint16_t pci_device;
        /** PCI subsystem vendor ID */
        uint16_t pci_subsys_vendor;
        /** PCI subsystem ID */
        uint16_t pci_subsys;
};

/**
 * List of devices with broken support for generating interrupts
 *
 * Some PXE stacks are known to claim that IRQs are supported, but
 * then never generate interrupts.  No satisfactory solution has been
 * found to this problem; the workaround is to add the PCI vendor and
 * device IDs to this list.  This is something of a hack, since it
 * will generate false positives for identical devices with a working
 * PXE stack (e.g. those that have been reflashed with iPXE), but it's
 * an improvement on the current situation.
 */
static const struct undinet_irq_broken undinet_irq_broken_list[] = {
        /* HP XX70x laptops */
        { 0x8086, 0x1502, PCI_ANY_ID, PCI_ANY_ID },
        { 0x8086, 0x1503, PCI_ANY_ID, PCI_ANY_ID },
        /* HP 745 G3 laptop */
        { 0x14e4, 0x1687, PCI_ANY_ID, PCI_ANY_ID },
        /* ASUSTeK KNPA-U16 server */
        { 0x8086, 0x1521, 0x1043, PCI_ANY_ID },
};

/**
 * Check for devices with broken support for generating interrupts
 *
 * @v netdev            Net device
 * @ret irq_is_broken   Interrupt support is broken; no interrupts are generated
 */
static int undinet_irq_is_broken ( struct net_device *netdev ) {
        struct undi_nic *undinic = netdev->priv;
        struct device_description *desc = &netdev->dev->desc;
        const struct undinet_irq_broken *broken;
        struct pci_device pci;
        uint16_t subsys_vendor;
        uint16_t subsys;
        unsigned int i;

        /* Ignore non-PCI devices */
        if ( desc->bus_type != BUS_TYPE_PCI )
                return 0;

        /* Read subsystem IDs */
        pci_init ( &pci, desc->location );
        pci_read_config_word ( &pci, PCI_SUBSYSTEM_VENDOR_ID, &subsys_vendor );
        pci_read_config_word ( &pci, PCI_SUBSYSTEM_ID, &subsys );

        /* Check for a match against the broken device list */
        for ( i = 0 ; i < ( sizeof ( undinet_irq_broken_list ) /
                            sizeof ( undinet_irq_broken_list[0] ) ) ; i++ ) {
                broken = &undinet_irq_broken_list[i];
                if ( ( broken->pci_vendor == desc->vendor ) &&
                     ( broken->pci_device == desc->device ) &&
                     ( ( broken->pci_subsys_vendor == subsys_vendor ) ||
                       ( broken->pci_subsys_vendor == PCI_ANY_ID ) ) &&
                     ( ( broken->pci_subsys == subsys ) ||
                       ( broken->pci_subsys == PCI_ANY_ID ) ) ) {
                        DBGC ( undinic, "UNDINIC %p %04x:%04x subsys "
                               "%04x:%04x has broken interrupts\n",
                               undinic, desc->vendor, desc->device,
                               subsys_vendor, subsys );
                        return 1;
                }
        }

        /* Check for a PCI Express capability.  Given the number of
         * issues found with legacy INTx emulation on PCIe systems, we
         * assume that there is a high chance of interrupts not
         * working on any PCIe device.
         */
        if ( pci_find_capability ( &pci, PCI_CAP_ID_EXP ) ) {
                DBGC ( undinic, "UNDINIC %p is PCI Express: assuming "
                       "interrupts are unreliable\n", undinic );
                return 1;
        }

        return 0;
}

/**
 * Probe UNDI device
 *
 * @v undi              UNDI device
 * @v dev               Underlying generic device
 * @ret rc              Return status code
 */
int undinet_probe ( struct undi_device *undi, struct device *dev ) {
        struct net_device *netdev;
        struct undi_nic *undinic;
        struct s_PXENV_START_UNDI start_undi;
        struct s_PXENV_UNDI_STARTUP undi_startup;
        struct s_PXENV_UNDI_INITIALIZE undi_init;
        struct s_PXENV_UNDI_GET_INFORMATION undi_info;
        struct s_PXENV_UNDI_GET_IFACE_INFO undi_iface;
        struct s_PXENV_UNDI_SHUTDOWN undi_shutdown;
        struct s_PXENV_UNDI_CLEANUP undi_cleanup;
        struct s_PXENV_STOP_UNDI stop_undi;
        unsigned int retry;
        int rc;

        /* Allocate net device */
        netdev = alloc_etherdev ( sizeof ( *undinic ) );
        if ( ! netdev )
                return -ENOMEM;
        netdev_init ( netdev, &undinet_operations );
        undinic = netdev->priv;
        undi_set_drvdata ( undi, netdev );
        netdev->dev = dev;
        memset ( undinic, 0, sizeof ( *undinic ) );
        undinet_entry_point = undi->entry;
        DBGC ( undinic, "UNDINIC %p using UNDI %p\n", undinic, undi );

        /* Hook in UNDI stack */
        if ( ! ( undi->flags & UNDI_FL_STARTED ) ) {
                memset ( &start_undi, 0, sizeof ( start_undi ) );
                start_undi.AX = undi->pci_busdevfn;
                start_undi.BX = undi->isapnp_csn;
                start_undi.DX = undi->isapnp_read_port;
                start_undi.ES = BIOS_SEG;
                start_undi.DI = find_pnp_bios();
                if ( ( rc = undinet_call ( undinic, PXENV_START_UNDI,
                                           &start_undi,
                                           sizeof ( start_undi ) ) ) != 0 )
                        goto err_start_undi;
        }
        undi->flags |= UNDI_FL_STARTED;

        /* Bring up UNDI stack */
        if ( ! ( undi->flags & UNDI_FL_INITIALIZED ) ) {
                memset ( &undi_startup, 0, sizeof ( undi_startup ) );
                if ( ( rc = undinet_call ( undinic, PXENV_UNDI_STARTUP,
                                           &undi_startup,
                                           sizeof ( undi_startup ) ) ) != 0 )
                        goto err_undi_startup;
                /* On some PXE stacks, PXENV_UNDI_INITIALIZE may fail
                 * due to a transient condition (e.g. media test
                 * failing because the link has only just come out of
                 * reset).  We may therefore need to retry this call
                 * several times.
                 */
                for ( retry = 0 ; ; ) {
                        memset ( &undi_init, 0, sizeof ( undi_init ) );
                        if ( ( rc = undinet_call ( undinic,
                                                   PXENV_UNDI_INITIALIZE,
                                                   &undi_init,
                                                   sizeof ( undi_init ) ) ) ==0)
                                break;
                        if ( ++retry > UNDI_INITIALIZE_RETRY_MAX )
                                goto err_undi_initialize;
                        DBGC ( undinic, "UNDINIC %p retrying "
                               "PXENV_UNDI_INITIALIZE (retry %d)\n",
                               undinic, retry );
                        /* Delay to allow link to settle if necessary */
                        mdelay ( UNDI_INITIALIZE_RETRY_DELAY_MS );
                }
        }
        undi->flags |= UNDI_FL_INITIALIZED;

        /* Get device information */
        memset ( &undi_info, 0, sizeof ( undi_info ) );
        if ( ( rc = undinet_call ( undinic, PXENV_UNDI_GET_INFORMATION,
                                   &undi_info, sizeof ( undi_info ) ) ) != 0 )
                goto err_undi_get_information;
        memcpy ( netdev->hw_addr, undi_info.PermNodeAddress, ETH_ALEN );
        memcpy ( netdev->ll_addr, undi_info.CurrentNodeAddress, ETH_ALEN );
        undinic->irq = undi_info.IntNumber;
        if ( undinic->irq > IRQ_MAX ) {
                DBGC ( undinic, "UNDINIC %p ignoring invalid IRQ %d\n",
                       undinic, undinic->irq );
                undinic->irq = 0;
        }
        DBGC ( undinic, "UNDINIC %p has MAC address %s and IRQ %d\n",
               undinic, eth_ntoa ( netdev->hw_addr ), undinic->irq );
        if ( undinic->irq ) {
                /* Sanity check - prefix should have disabled the IRQ */
                assert ( ! irq_enabled ( undinic->irq ) );
        }

        /* Get interface information */
        memset ( &undi_iface, 0, sizeof ( undi_iface ) );
        if ( ( rc = undinet_call ( undinic, PXENV_UNDI_GET_IFACE_INFO,
                                   &undi_iface, sizeof ( undi_iface ) ) ) != 0 )
                goto err_undi_get_iface_info;
        DBGC ( undinic, "UNDINIC %p has type %s, speed %d, flags %08x\n",
               undinic, undi_iface.IfaceType, undi_iface.LinkSpeed,
               undi_iface.ServiceFlags );
        if ( ( undi_iface.ServiceFlags & SUPPORTED_IRQ ) &&
             ( undinic->irq != 0 ) ) {
                undinic->irq_supported = 1;
        }
        DBGC ( undinic, "UNDINIC %p using %s mode\n", undinic,
               ( undinic->irq_supported ? "interrupt" : "polling" ) );
        if ( strncmp ( ( ( char * ) undi_iface.IfaceType ), "Etherboot",
                       sizeof ( undi_iface.IfaceType ) ) == 0 ) {
                DBGC ( undinic, "UNDINIC %p Etherboot 5.4 workaround enabled\n",
                       undinic );
                undinic->hacks |= UNDI_HACK_EB54;
        }
        if ( undinet_irq_is_broken ( netdev ) ) {
                DBGC ( undinic, "UNDINIC %p forcing polling mode due to "
                       "broken interrupts\n", undinic );
                undinic->irq_supported = 0;
        }

        /* Register network device */
        if ( ( rc = register_netdev ( netdev ) ) != 0 )
                goto err_register;

        /* Mark as link up; we don't handle link state */
        netdev_link_up ( netdev );

        DBGC ( undinic, "UNDINIC %p added\n", undinic );
        return 0;

 err_register:
 err_undi_get_iface_info:
 err_undi_get_information:
 err_undi_initialize:
        /* Shut down UNDI stack */
        memset ( &undi_shutdown, 0, sizeof ( undi_shutdown ) );
        undinet_call ( undinic, PXENV_UNDI_SHUTDOWN, &undi_shutdown,
                       sizeof ( undi_shutdown ) );
        memset ( &undi_cleanup, 0, sizeof ( undi_cleanup ) );
        undinet_call ( undinic, PXENV_UNDI_CLEANUP, &undi_cleanup,
                       sizeof ( undi_cleanup ) );
        undi->flags &= ~UNDI_FL_INITIALIZED;
 err_undi_startup:
        /* Unhook UNDI stack */
        memset ( &stop_undi, 0, sizeof ( stop_undi ) );
        undinet_call ( undinic, PXENV_STOP_UNDI, &stop_undi,
                       sizeof ( stop_undi ) );
        undi->flags &= ~UNDI_FL_STARTED;
 err_start_undi:
        netdev_nullify ( netdev );
        netdev_put ( netdev );
        undi_set_drvdata ( undi, NULL );
        return rc;
}

/**
 * Remove UNDI device
 *
 * @v undi              UNDI device
 */
void undinet_remove ( struct undi_device *undi ) {
        struct net_device *netdev = undi_get_drvdata ( undi );
        struct undi_nic *undinic = netdev->priv;
        struct s_PXENV_UNDI_SHUTDOWN undi_shutdown;
        struct s_PXENV_UNDI_CLEANUP undi_cleanup;
        struct s_PXENV_STOP_UNDI stop_undi;

        /* Unregister net device */
        unregister_netdev ( netdev );

        /* If we are preparing for an OS boot, or if we cannot exit
         * via the PXE stack, then shut down the PXE stack.
         */
        if ( ! ( undi->flags & UNDI_FL_KEEP_ALL ) ) {

                /* Shut down UNDI stack */
                memset ( &undi_shutdown, 0, sizeof ( undi_shutdown ) );
                undinet_call ( undinic, PXENV_UNDI_SHUTDOWN,
                               &undi_shutdown, sizeof ( undi_shutdown ) );
                memset ( &undi_cleanup, 0, sizeof ( undi_cleanup ) );
                undinet_call ( undinic, PXENV_UNDI_CLEANUP,
                               &undi_cleanup, sizeof ( undi_cleanup ) );
                undi->flags &= ~UNDI_FL_INITIALIZED;

                /* Unhook UNDI stack */
                memset ( &stop_undi, 0, sizeof ( stop_undi ) );
                undinet_call ( undinic, PXENV_STOP_UNDI, &stop_undi,
                               sizeof ( stop_undi ) );
                undi->flags &= ~UNDI_FL_STARTED;
        }

        /* Clear entry point */
        memset ( &undinet_entry_point, 0, sizeof ( undinet_entry_point ) );

        /* Free network device */
        netdev_nullify ( netdev );
        netdev_put ( netdev );

        DBGC ( undinic, "UNDINIC %p removed\n", undinic );
}