ethervirtio.c : virtio-net for 9front, able to send ping, receive dhcp but dead when received more than 255 frames

gistfile1.c

#include "u.h"
#include "../port/lib.h"
#include "mem.h"
#include "dat.h"
#include "fns.h"
#include "io.h"
#include "ureg.h"
#include "../port/error.h"

#include "../port/netif.h"
#include "etherif.h"
#include "ethermii.h"
//
//
// Virtio
//
//
typedef struct Vioreqhdr Vioreqhdr;
typedef struct Vringhdr Vringhdr;
typedef struct Vdesc Vdesc;
typedef struct Vused Vused;
typedef struct Vqueue Vqueue;
typedef struct Vdev Vdev;

enum {
	Acknowledge	= 1,
	Driver		= 2,
	DriverOk	= 4,
	Failed		= 128,
};

enum {
	Devfeat = 0,
	Drvfeat = 4,
	Qaddr = 8,
	Qsize = 12,
	Qselect = 14,
	Qnotify = 16,
	Status = 18,
	Isr = 19,

	Devspec = 20,
};

enum {
	Next = 1,
	Write = 2,
	Indirect = 4,
};	

enum {
	VIRTIO_F_NOTIFY_ON_EMPTY	= (1 << 24),
	VIRTIO_F_RING_INDIRECT_DESC	= (1 << 28),
	VIRTIO_F_RING_EVENT_IDX		= (1 << 29),
};

enum {
	VRING_AVAIL_F_NO_INTERRUPT	= (1 << 0),
};

struct Vioreqhdr
{
	u32int	typ;
	u32int	prio;
	u64int	lba;
};

struct Vringhdr
{
	u16int	flags;
	u16int	idx;
};

struct Vdesc
{
	u64int	addr;
	u32int	len;
	u16int	flags;
	u16int	next;
};

struct Vused
{
	u32int	id;
	u32int	len;
};

struct Key
{
	void*		kaddr;
	ulong		paddr;
};
typedef struct Key	Key;

struct Vqueue
{
	int		size;

	int		free;
	int		nfree;

	Vdesc		*desc;

	Vringhdr	*avail;
	u16int		*availent;
	u16int		*availevent;

	Vringhdr	*used;
	Vused		*usedent;
	u16int		*usedevent;

	u16int		lastused;

	Key		*key;

	Rendez;
	QLock;
	Lock;
};

struct Vdev
{
	int	typ;

	Pcidev	*pci;

	ulong	port;
	ulong	features;

	int	nqueue;
	Vqueue	*queue[16];
	
	int 	active;

	Vdev	*next;
};

static Vqueue*
mkvqueue(int size)
{
	Vqueue *q;
	uchar *p;
	int i;

	q = malloc(sizeof(*q));
	p = mallocalign(
		PGROUND(sizeof(Vdesc)*size + 
			sizeof(Vringhdr) + 
			sizeof(u16int)*size + 
			sizeof(u16int)) +
		PGROUND(sizeof(Vringhdr) + 
			sizeof(Vused)*size + 
			sizeof(u16int)), 
		BY2PG, 0, 0);
	if(p == nil || q == nil){
		print("mkvqueue: no memory for Vqueue\n");
		free(p);
		free(q);
		return nil;
	}

	q->desc = (void*)p;
	p += sizeof(Vdesc)*size;
	q->avail = (void*)p;
	p += sizeof(Vringhdr);
	q->availent = (void*)p;
	p += sizeof(u16int)*size;
	q->availevent = (void*)p;
	p += sizeof(u16int);

	p = (uchar*)PGROUND((ulong)p);
	q->used = (void*)p;
	p += sizeof(Vringhdr);
	q->usedent = (void*)p;
	p += sizeof(Vused)*size;
	q->usedevent = (void*)p;

	q->key = malloc(sizeof(Key) * size);

	q->free = -1;
	q->nfree = q->size = size;
	for(i=0; i<size; i++){
		q->desc[i].next = q->free;
		q->free = i;
	}

	return q;
}

//
//
// Virtio-Net specific
//
//
typedef struct Vnetcfg Vnetcfg;
typedef struct Vnetctrl Vnetctrl;

/* Virtio-net queue number */
enum {
	VIRTIO_NET_Q_RX			= 0,
	VIRTIO_NET_Q_TX			= 1,
	VIRTIO_NET_Q_CTLR		= 2,
};

/* Virtio-net specific Device features */
enum {
	VIRTIO_NET_F_CSUM		= (1 << 0),
	VIRTIO_NET_F_GUEST_CSUM		= (1 << 1),
	VIRTIO_NET_F_MAC		= (1 << 5),
	VIRTIO_NET_F_GSO		= (1 << 6),
	VIRTIO_NET_F_GUEST_TSO4		= (1 << 7),
	VIRTIO_NET_F_GUEST_TSO6		= (1 << 8),
	VIRTIO_NET_F_GUEST_ECN		= (1 << 9),
	VIRTIO_NET_F_GUEST_UFO		= (1 << 10),
	VIRTIO_NET_F_HOST_TSO4		= (1 << 11),
	VIRTIO_NET_F_HOST_TSO6		= (1 << 12),
	VIRTIO_NET_F_HOST_ECN		= (1 << 13),
	VIRTIO_NET_F_HOST_UFO		= (1 << 14),
	VIRTIO_NET_F_MRG_RXBUF		= (1 << 15),
	VIRTIO_NET_F_STATUS		= (1 << 16),
	VIRTIO_NET_F_CTRL_VQ		= (1 << 17),
	VIRTIO_NET_F_CTRL_RX		= (1 << 18),
	VIRTIO_NET_F_CTRL_VLAN		= (1 << 19),
	VIRTIO_NET_F_GUEST_ANNOUNCE	= (1 << 21),
};

/* Virtio-net Device Status */
enum {
	VIRTIO_NET_S_LINK_UP		= 1,
	VIRTIO_NET_S_ANNOUNCE		= 2,
};

/* Not used : Virtio-net Device Specific Region */
struct Vnetcfg {
	u8int	mac[6];
	u16int	status;
};

/* Vnetreqhdr, flags */
enum {
	VIRTIO_NET_HDR_F_NEEDS_CSUM	=~1,
};

/* Vnetreqhdr, gso_type */
enum {
	VIRTIO_NET_HDR_GSO_NONE		= 0,
	VIRTIO_NET_HDR_GSO_TCPV4	= 1,
	VIRTIO_NET_HDR_GSO_UDP		= 3,
	VIRTIO_NET_HDR_GSO_TCPV6	= 4,
	VIRTIO_NET_HDR_GSO_ECN		= 0x80,
};

/* Virtio-net request header */
#pragma pack on 
typedef struct Vnetreqhdr Vnetreqhdr;
struct Vnetreqhdr {
	u8int	flags;
	u8int	gso_type;
	u16int	hdr_len;
	u16int	gso_size;
	u16int	csum_start;
	u16int	csum_offset;
};
#pragma pack off
int Vnetreqhdrsize = 10;

/* Virtio-Net Control Requext Class */
enum {
	VIRTIO_NET_CTRL_RX_CLASS	= 0,
};

/* Virtio-net Control Request Command */
enum {
	VIRTIO_NET_CTRL_RX_CMD_PROMISC	= 0,
	VIRTIO_NET_CTRL_RX_CMD_ALLMULTI	= 1,
};

/* on off */
enum {
	VIRTIO_NET_CTRL_RX_OFF		= 0,
	VIRTIO_NET_CTRL_RX_ON		= 1,
};

/* ack value */
enum {
	VIRTIO_NET_CTRL_ACK_OK		= 0,
	VIRTIO_NET_CTRL_ACK_ERR		= 1,
};

/* Virtio-net control request */
struct Vnetctrl {
	u8int	class;
	u8int	command;
};
/* followed by u8int onoff, u8int ack */

Vdev *vionethead = nil;
Vdev *vionettail = nil;

/////////////////////////////////////////////////////////////////////
// Virtio-net helper funcs
/////////////////////////////////////////////////////////////////////
/*********** Interrupt ***********/
/*
 * Free descriptors used for this transmit
 */
static void
vionettxfree(Vqueue* q, int head)
{
	Vdesc *d;
	int bufid;
	Vdesc *bufd;
	//ulong paddr;
	//void* kaddr;
	Block *bp;
	Vnetreqhdr *vnetreqhdr;
	print("txfree q %p head %d\n", q, head);

	/* get Vnetreqhdr */
	d = &q->desc[head];

	/* Vnetreqhdr is followed by real buffer descriptor */
	bufid = d->next;
	bufd = &q->desc[bufid];

	/* free header */
	vnetreqhdr = q->key[head].kaddr;
	free(vnetreqhdr);
	d->len = d->flags = 0;
	q->key[head].kaddr = nil;

	/* calculate original Block address and free */
	bp = q->key[bufid].kaddr;
	freeb(bp);
	q->key[bufid].kaddr = nil;

	/* put vdesc for header back */
	d->next = q->free;
	q->free = head;
	q->nfree++;

	/* pub vdesc for buffer back */
	bufd->next = q->free;
	q->free = bufid;
	q->nfree++;
}	

/*
 * Interrupt on Tx vq.
 * responsible for checking vq change and cleanup
 */
static void
vionettxinterrupt(Vdev* vd)
{
	Vqueue* q;
	int id, m;

	print("txinterrupt : in\n");
	q = vd->queue[VIRTIO_NET_Q_TX];
	print("txinterrupt : vd = %p\n", vd);
	print("txinterrupt : q= %p\n", q);
	m = (q->size-1);

	//lock(q);
	print("txinterrupt: %d ?? %d\n",
		(q->lastused % m), (q->used->idx &m));
	while((q->lastused % m) != (q->used->idx % m)) {
		id = q->usedent[q->lastused++ % (q->size-1)].id;
		vionettxfree(q, id);
	}
	//unlock(q);
	print("txinterrupt : out\n");
}

static void
vionetrxproc(Ether* edev, Vqueue* q, int id)
{
	Vdesc* d;
	Vdesc* bufd;
	void* buf;
	Block *bp;
	Vnetreqhdr* vnetreqhdr;

	d = &q->desc[id];
	bufd = &q->desc[d->next];

	vnetreqhdr = (Vnetreqhdr*)q->key[id].kaddr;

	print("rxproc : d %p len %d\n", d, d->len);
	print("rxporc : vnetreqhdr %p flags %d len %d\n",
		vnetreqhdr, vnetreqhdr->flags, vnetreqhdr->hdr_len);
	print("rxproc : bufd %p size %d\n", bufd, bufd->len);

	buf = (void*)q->key[d->next].kaddr;

	bp = iallocb(ETHERMAXTU);  // XXX
	memmove(bp->rp, buf, ETHERMAXTU); // XXX

	bp->wp += ETHERMAXTU;
	bp->lim = bp->wp; /* lie like a dog */

	/* go up */
	etheriq(edev, bp, 1);

	/* clean current buffer : XXX */
	d->len = sizeof(Vnetreqhdr);
	d->flags = Next|Write;
	vnetreqhdr->hdr_len = ETHERMAXTU;
	bufd->len = ETHERMAXTU;
	bufd->flags = Write;
}

static void
vionetrxinterrupt(Ether* edev)
{
	Vqueue* q;
	int id, m;
	Vdev* vd;
	int i, head;

	print("vionetrxinterrupt: in, edev %p\n", edev);

	vd = edev->ctlr;
	print("vionetrxinterrupt: in, vd %p\n", vd);
	q = vd->queue[VIRTIO_NET_Q_RX];
	print("vionetrxinterrupt: in, q %p\n", q);
	m = (q->size-1);

	//lock(q);
	print("vionetrxinterrupt: forward and process lastused %d to used->idx %d\n", 
	q->lastused, q->used->idx);
	i = 0;
	while((q->lastused % m) != (q->used->idx % m)) {
		id = q->usedent[q->lastused++ % m].id;
		vionetrxproc(edev, q, id);
		/* 
		 * here id is free. we do push back to free list?
		 * NO. push back to avail ring as is.
		 */
		coherence();
		i++;
		q->availent[(q->avail->idx++) % m] = id;
	}
	coherence();
	if (i) {
		coherence();
		outs(vd->port+Qnotify, VIRTIO_NET_Q_RX);
	}
	print("vionetrxinterrupt: fixedup receive buffer\n");

	print("vionetrxinterrupt: out\n");
}

/************ Initialize ****************/
static void
vionettxinit(Vdev* vd)
{
	print("vionettxinit: in\n");
	if (vd->nqueue < 2) {
		return; 
	}

	/* no interrupt */
	print("vionettxinit: in\n");
//	vd->queue[VIRTIO_NET_Q_RX]->avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
	print("vionettxinit: in\n");

	vd->queue[VIRTIO_NET_Q_RX]->lastused = 0;
	print("vionettxinit: out\n");
}

static int
vionetrxinithead(Vqueue* q)
{
	Vdesc* d;
	int head, free;
	Vnetreqhdr* vnetreqhdr;
	void* buf;

	if (q->nfree < 2)
		return -1;

	head = free = q->free;
	d = &q->desc[free];
	/* setup Vnethdr */
	vnetreqhdr = malloc(sizeof(Vnetreqhdr));
	vnetreqhdr->hdr_len = ETHERMAXTU;//sizeof(Vnetreqhdr);
	vnetreqhdr->flags = 0; // XXX
//	vnetreqhdr->num_buffers = 1;
	/* setup vdesc for Vnetreqhdr */
	d->addr = PADDR(vnetreqhdr);
	d->len = Vnetreqhdrsize;//sizeof(Vnetreqhdr);
	d->flags = Next|Write ;
	/* save key */
	q->key[head].paddr = d->addr;
	q->key[head].kaddr = vnetreqhdr;
	free = d->next;

	/* get vdesc for data buffer : a real frame */
	d = &q->desc[free];
	/* setup vdesc */
	buf = malloc(ETHERMAXTU);
	d->addr = PADDR(buf);
	d->len = ETHERMAXTU;
	d->flags = Write;
	/* save key */
	q->key[free].paddr = d->addr;
	q->key[free].kaddr = buf;
	free = d->next;

	q->nfree -= 2;
	q->free = free;

	return head;
}

static void
vionetrxinit(Vdev* vd)
{
	Vqueue *q;
	Vdesc *d;
	int nadded;
	int head, free;
	Vnetreqhdr* vnetreqhdr;
	void* buf;
	

	print("vionetrxinit: in\n");
	if (vd->nqueue < 1) {
		return;
	}

	print("vionetrxinit: fetch queuen\n");
	q = vd->queue[VIRTIO_NET_Q_RX];

	print("vionetrxinit: each vdesc has %d byte buffer\n", ETHERMAXTU);
/*
	q->avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
*/

	print("vionetrxinit: allocating buffer\n");
	/* allocate receive buffer */
	nadded = 0;
	while(1) {
		head = vionetrxinithead(q);
		if (head < 0)
			break;

		coherence();
		/* put head's number into avail ring */
		nadded++;
		q->availent[(q->avail->idx++) % (q->size-1)] = head;
		coherence();
	}
	if (nadded > 0) {
		coherence();
		/* update avail idx */
		coherence();
		outs(vd->port+Qnotify, VIRTIO_NET_Q_RX);
	}
	print("vionetrxinit: allocated %d buffers\n", nadded);

	q->lastused = 0;

	print("vionettxinit: out\n");
}

static void
vionetvqinit(Vdev* vd)
{
	int i;
	int size;
	u32int a;


	for (i = 0; i < nelem(vd->queue) ; i++) {
		/* select operating queue */
		outs(vd->port + Qselect, i);
		/* get operating queue size (0 : disabled) */
		if ((size = ins(vd->port + Qsize)) == 0)
			break;
		if ((vd->queue[i] = mkvqueue(size)) == nil)
			break;
		coherence();
		a = PADDR(vd->queue[i]->desc)/BY2PG;
		/* write addr of operating queue at Qaddr */
		outl(vd->port + Qaddr, a);
	}
	vd->nqueue = i;

	/* setup rx */
	vionettxinit(vd);

	/* setup tx */
	vionetrxinit(vd);
}

/*
 * Negotiates Device features and given Driver features
 */
static ulong
vio_negotiate_feature(Vdev *vd, ulong drvfeat)
{
	ulong features;

	print("vio neto fet: in %ld + %d\n", vd->port, Devfeat);
	/* read device features */
	features = inl(vd->port + Devfeat);

	print("vio neto fet: in sec\n");
	features &= drvfeat;

	print("vio neto fet: out %ld %d\n", vd->port, Drvfeat);
	/* write back features */
	outl(vd->port + Drvfeat, features);

	print("vio neto fet: out\n");
	return features;
}

static void
vionetinit1(Ether* edev, Vdev* vd)
{
	int i;

	print("vionetinit1: in\n");
	/* check feature bit : F_MAC */
	if (vd->features & VIRTIO_NET_F_MAC) {
		for (i = 0; i < 6; i++) {
			edev->ea[i] = inb(vd->port + Devspec + i);
		}
	} else {
		for (i = 0; i < 6; i++) {
			edev->ea[i] = 0xff;
			outb(vd->port + Devspec + i, edev->ea[i]);
		}
	}


	/* check feature bit : F_STATUS */
	if (vd->features & VIRTIO_NET_F_STATUS) {
		/* check and up the link */
		edev->link = (0x01 & ins(vd->port + Devspec + Eaddrlen));
		print("vionetinit1: link is %d\n", edev->link);
		edev->link &= 0x01;
		print("vionetinit1: link is %d\n", edev->link);
		outs(vd->port + Devspec + Eaddrlen, edev->link);
	} else {
		/* spec says assume active */
		edev->link = 0x01;
	}
	print("vionetinit1: out\n");
}

/*
 * Initialize device&driver
 */
static void
vionetinit(Vdev* vd)
{
	int is_ctrl_vq = 0;

	print("vionetinit: in\n");
	/* negotiate features */
	vd->features = vio_negotiate_feature(vd,
		VIRTIO_F_NOTIFY_ON_EMPTY |
		VIRTIO_NET_F_MAC |
		VIRTIO_NET_F_STATUS |
//		VIRTIO_NET_F_CTRL_VQ |
		VIRTIO_NET_F_CTRL_RX |
		0
		);

	print("vionetinit: negotiated\n");
	/* negotiate features */
	/* check feature bit : F_CTRL_VQ */
	if (vd->features & VIRTIO_NET_F_CTRL_VQ) {
		is_ctrl_vq = 1;
	}

	/* setup virtqueue */
	print("vionetinit: call vionetvqinit\n");
	vionetvqinit(vd);
	print("vionetinit: called vionetvqinit\n");

	/* check if ctrlvq is valid */
	if (is_ctrl_vq && vd->nqueue != 3) {
		print("vionetinit(): ctrl queue is invalid\n");
	}

	print("vionetinit: out\n");
	return;
}

/////////////////////////////////////////////////////////////////////
// Generic Network driver funcs
/////////////////////////////////////////////////////////////////////
static void
vionetattach(Ether*)
{
	/* nothing to do */
}

 void
vionettransmit(Ether* edev)
{
	Block* bp; /* bp->rp has address */
	Vdev* vd;
	int head, free;
	Vdesc *d;
	Vqueue *q;
	int nadded = 0;
	Vnetreqhdr* vnetreqhdr;

	vd = edev->ctlr;
	q = vd->queue[VIRTIO_NET_Q_TX];

	print("vionettransmit: in\n");
	//lock(q);

	for (;;) {
		/* needs 2 entries for transmit */
		if (q->nfree < 2) {
			//unlock(q);
			//error("out of virtio descriptors");
			print("vionettransmit: run out of queue : break;\n");
			break;
		}

		/* get next frame to send */
		if ((bp = qget(edev->oq)) == nil) {
			print("vionettransmit: run out of oq : break;\n");
			break;
		}

		head = free = q->free;

		print("vionettransmit: head is %d\n", head);
		/* get vdesc for Vnethdr */
		d = &q->desc[free]; free = d->next;

		/* setup Vnethdr */
		vnetreqhdr = malloc(Vnetreqhdrsize/*sizeof(Vnetreqhdr)*/);
//		vnetreqhdr->hdr_len = Vnetreqhdrsize;//sizeof(Vnetreqhdr);
		vnetreqhdr->flags = 0; // XXX
//		vnetreqhdr->num_buffers = 1;

		/* setup vdesc for Vnethdr */
		d->addr = PADDR(vnetreqhdr);
		d->len = Vnetreqhdrsize;//sizeof(Vnetreqhdr);
		d->flags = Next;
		
		/* save paddr and kaddr */
		q->key[head].paddr = d->addr;
		q->key[head].kaddr = vnetreqhdr;

		/* get vdesc for data buffer : a real frame */
		print("vionettransmit: buffer is %d\n", free);
		d = &q->desc[free]; free = d->next;

		/* setup vdesc for data buffer */
		d->addr = PADDR(bp->rp);
		d->len = BLEN(bp);
		d->flags = 0;

		/* save paddr and kaddr , paddr */
		q->key[free].paddr = d->addr;
		q->key[free].kaddr = bp;

		print("vionettransmit: q->free %d, q->nfree %d\n", q->free, q->nfree);
		q->nfree -= 2;
		q->free = free;

		coherence();
		/* put head's number into avail ring */
		print("vionettransmit: set head %d at idx %d + nadded %d q->size %d\n", 
			head, q->avail->idx, nadded, q->size);
		nadded++;
		/* update avail idx */
		q->availent[(q->avail->idx++) & (q->size-1)] = head;
		coherence();
		print("vionettransmit: nadded is %d q->free %d, q->nfree %d\n", 
			nadded, q->free, q->nfree);
	}

	coherence();
	//unlock(q);
	if (nadded > 0)  {
		print("vionettransmit: updated idx to %d\n", q->avail->idx);
		/* notify */
		outs(vd->port+Qnotify, VIRTIO_NET_Q_TX);
	}

	print("vionettransmit: done\n");
	print("vionettransmit: out\n");
}

static void
vionetinterrupt(Ureg*, void* arg)
{
	Vdev *vd;
	Ether *edev;

	edev = arg;
	vd = edev->ctlr;

	if (edev == nil || vd == nil)
		return;
	
	print("vionetinterrupt: in edev %p vd %p\n", edev, vd);
	if (inb(vd->port + Isr) & 1) {
		print("vionetinterrupt: in Isr\n");
		/* wakeup ctrl op */
//		if (vd->nqueue > VIRTIO_NET_Q_CTLR) {
//			wakeup(vd->queue[VIRTIO_NET_Q_CTLR]);
//		}

		/* control vq */
//		vionetctrlinterrupt(vd);

		/* transmit vq */
		print("vionetinterrupt: in tx\n");
		vionettxinterrupt(vd);

		/* receive vq */
		print("vionetinterrupt: in rx\n");
		vionetrxinterrupt(edev);
		print("vionetinterrupt: in done\n");
	}

	if (inb(vd->port + Isr) & 2) {
		/* config space has changed */
	print("vionetinterrupt: config changed\n");
	}
	print("vionetinterrupt: out\n");
}

static long
vionetifstat(Ether*, void*, long, ulong)
{
	return 0;
}

static void
vionetshutdown()
{
	/* do cleanup */
	return;
}

static long
vionetctl(Ether*, void*, long)
{
	/* do nothing */
	return 0;
}

/*
static void
vionetdone(void *arg)
{
	struct Rock *r;
	Vqueue *q;
	u16int i;

	r = arg;
	q = r->q;

	for (i = q->lastused; i != q->used->idx; i++) {
		if (q->usedent[i % q->size].id == r->id) {
			if (i == q->lastused)
				q->lastused++;
			r->done =1;
			break;
		}
	}

	return r->done;
}
*/


/*
static void
vionetwait(Vqueue *q, int id)
{
	struct Rock r;

	r.q = q;
	r.id = id;
	r.done = 0;
	do {
		//qlock(q);
		while(waserror());

		sleep(q, vionetdone, &r);
		poperror();
		//qunlock(q);
	} while(!r.done);
}
*/

/*
 * Send Control request to device.
 */
static void
vionetctrl(Ether* edev, u8int cmd, u8int on)
{
	Vnetctrl vnetctrl;
	Vdev* vd;
	Vqueue* q;
	int free, head;
	Vdesc *d;
	u8int ack;
	u8int onoff;

	vd = edev->ctlr;
	onoff = on ? VIRTIO_NET_CTRL_RX_ON : VIRTIO_NET_CTRL_RX_OFF;
	ack = VIRTIO_NET_CTRL_ACK_ERR;

	q = vd->queue[VIRTIO_NET_Q_CTLR];
	//lock(q);

	/* get idx of  head of free entry */
	head = free = q->free;

	/* allocate Vnetctrl entry */
	
	vnetctrl.class = VIRTIO_NET_CTRL_RX_CLASS;
	vnetctrl.command = cmd;

	/* push vnetctrl */
	d = &q->desc[free]; free = d->next;
	d->addr = PADDR(&vnetctrl);
	d->len = sizeof(vnetctrl);
	d->flags = Next;

	/* push onoff */
	d = &q->desc[free]; free = d->next;
	d->addr = PADDR(&onoff);
	d->len = sizeof(u8int);
	d->flags = Next;

	/* push ack */
	d = &q->desc[free]; free = d->next;
	d->addr = PADDR(&ack);
	d->len = sizeof(u8int);
	d->flags = Write;
	d->next = -1;

	q->free = free;
	q->nfree -= 3; /* pushed 3 */

	coherence(); /* sync */
	
	q->availent[q->avail->idx++ & q->size] = head;

	//unlock(q);

	coherence();

	/* commit notify */
	outs(vd->port + Qnotify, VIRTIO_NET_Q_CTLR);

	/* blocking */
	//vionetwait(q, head);

	if (ack != VIRTIO_NET_CTRL_RX_ON) {
		print("vionetctrl: cmd = %d ack is ERR\n", cmd);
		error(Eio);
	}

	return;
}

static void
vionetpromiscuous(void* arg, int on)
{
	Ether* edev;

	edev = arg;
	vionetctrl(edev, VIRTIO_NET_CTRL_RX_CMD_PROMISC, on);

	return;
}

static void
vionetmulticast(void* arg, uchar*, int on)
{
	Ether* edev;

	edev = arg;
	vionetctrl(edev, VIRTIO_NET_CTRL_RX_CMD_ALLMULTI, on);

	return;
}

/*
 * Finds all virtio-net devices and list up in vionethead to vionettail.
 * Do virtio-dev init and mkvqueue
 */
static void
vionetpci(void)
{
	Pcidev* p;
	Vdev *vd;

	print("vionetpci: in %d %d \n", 
		sizeof(Vnetreqhdr), sizeof(Vnetctrl));
	for (p = nil; p = pcimatch(p, 0, 0);) {
		if (p->vid != 0x1af4)
			continue;
		if ((p->did < 0x1000) || (p->did >= 0x1040))
			continue;
		if (p->rid != 0)
			continue;
		if (pcicfgr16(p, 0x2e) != 1 /* virtio-net subid */)
			continue;
		print("vionetpci: found device\n");
		if ((vd = malloc(sizeof(*vd))) == nil) {
			print("vionetpci(): cannnot allocate Vdev\n");
			break;
		}
		print("vionetpci: vd alloced\n");

		vd->port = p->mem[0].bar & ~0x1;
		if (ioalloc(vd->port, p->mem[0].size, 0, "virtio") < 0) {
			print("virtio: port %lux in use\n", vd->port);
			free(vd);
			continue;
		}
			
		vd->typ = 1 /* nic */;
		vd->pci = p;

		/* reset */
		outb(vd->port + Status, 0);

		/* set Acknowkedge and Driver */
		print("vionetpci: Acknowledge and Driver to device\n");
		outl(vd->port+Status, inb(vd->port+Status)|Acknowledge|Driver);
		print("vionetpci: sent Acknowledge and Driver to device\n");

		/* prepare vionet device */
		print("vionetpci: call vionetinit()\n");
		vionetinit(vd);
		print("vionetpci: called vionetinit()\n");

		print("vionetpnp: enable interrupt¥n");
		//intrenable(vd->pci->intl, vionetinterrupt, vd, vd->pci->tbdf, "ethervirtio");
		print("vionetpnp: enabled interrupt¥n");
	
		/* done initialization */
		print("vionetpnp: set status DriverOk\n");
		outb(vd->port + Status, inb(vd->port + Status) | DriverOk);
		print("vionetpnp: seted status DriverOk\n");

		vd->next = nil;
		if (vionethead == nil)
			vionethead = vd;
		else
			vionettail->next = vd;
		vionettail = vd;
		print("inited : vd %p\n", vd);
		print("inited : vd q0 %p\n", vd->queue[0]);
		print("inited : vd q1 %p\n", vd->queue[1]);
	}
	print("vionetpci: out\n");
}

static int
vionetpnp(Ether* edev)
{
	Vdev* vd;
	static int done;

	if (!done) {
		vionetpci(); /* do initialize each device */
		done = 1;
	}

	for (vd = vionethead; ; vd = vd->next) {
		if (vd == nil)
			return -1;
		if (vd->active)
			continue;
		if (edev->port == 0 || edev->port == vd->port) {
			vd->active = 1;
			break;
		}
	}

	if (vd == nil)
		return -1;

	edev->ctlr = vd;
	edev->port = vd->port;
	edev->irq = vd->pci->intl;
	edev->tbdf = vd->pci->tbdf;
	edev->mbps = 1000;

	print("vionetpnp: edev %p\n", edev);
	print("vionetpnp: vd %p\n", vd);
	print("vionetpnp: call vionetinit1\n");
	vionetinit1(edev, vd);
	print("vionetpnp: called vionetinit1\n");
	/*
	 * set handler functions
	 */
	edev->attach = vionetattach;
	edev->transmit = vionettransmit;
	edev->interrupt = vionetinterrupt;
	edev->ifstat = vionetifstat;
	edev->shutdown = vionetshutdown;
	edev->ctl = vionetctl;
	edev->arg = edev;
/*
	edev->promiscuous = vionetpromiscuous;
	edev->multicast = vionetmulticast;
*/

	print("vionetpnp: out\n");
	return 0;
}

void
ethervirtiolink(void)
{
	addethercard("virtio", vionetpnp);
}