From mboxrd@z Thu Jan  1 00:00:00 1970
From: quanstro@quanstro.net (erik quanstrom)
Date: Fri, 25 Nov 2011 17:42:22 -0500
Subject: [9fans] Forks of Plan 9 (Was: 9vx instability)
In-Reply-To: <CAHfqX-ot8PRUNW5aH+5zKFQXvd_O+otfE4pBS0HfjfD6-6HQSA@mail.gmail.c>
References: <CA+N-5bYRKyGoL_F1_eByiKTcJ3FVro1qoE12kYYLyYmy+iecsA@mail.gmail.com>
	<fdcb0da874d9ce0f7ad970d34e5c8587@proxima.alt.za>
	<CAHfqX-pxHzzHQdf2Rg=Fo+yKOSxYVX7UdU7j4hQJKYS2c7e4tQ@mail.gmail.c>
	<c2f05f243284ea6334d85f4eb34d40a9@chula.quanstro.net>
	<CAHfqX-ot8PRUNW5aH+5zKFQXvd_O+otfE4pBS0HfjfD6-6HQSA@mail.gmail.c>
Message-ID: <22cc46aeabe9242035819a85e84179a6@chula.quanstro.net>
Topicbox-Message-UUID: 47863cc0-ead7-11e9-9d60-3106f5b1d025

still not working; doesn't even reset.

i've taken a pass at cleaning up the code.  the biggest change
is using style(6) formatting.  also, coherence() is missing when
entering ring entries, since the posted write of the ring ptr can
pass the memory writes of the ring entry, from the perspective
of the pci root complex.  and transmit wasn't careful enough
not to run into its head.  (checking the chip isn't good enough.
you have to not run over your own unfreed bps.)

- erik
-------------- next part --------------
/*
 * Broadcom BCM57xx
 * Not implemented:
 *  proper fatal error handling
 *  multiple rings
 *  QoS
 *  checksum offloading
 */

/* this driver needs some work, and probablly doesn't conform to style(6) */

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

#include "etherif.h"

#define dprint(...)	do{ if(debug)print(__VA_ARGS__); }while(0)
#define Rbsz		ROUNDUP(sizeof(Etherpkt)+4, 4)
#define	Pciwaddrl(x)	PCIWADDR(x)
#define Pciwaddrh(x)	(sizeof(uintptr)>4? (uvlong)PCIWADDR(x)>>32: 0)

typedef struct Ctlr Ctlr;
struct Ctlr {
	Lock txlock, imlock;
	Ctlr *next;
	Pcidev *pdev;
	ulong *nic, *status;

	ulong *recvret, *recvprod, *sendr;
	ulong port;
	ulong recvreti, recvprodi, sendri, sendcleani;
	Block **sends;
	Block **rxs;
	int active, duplex;

	uint	nobuf;
	uint	partial;
	uint	rxerr;
	uint	qfull;
	uint	dmaerr;
};

enum {
	RecvRetRingLen = 0x200,
	RecvProdRingLen = 0x200,
	SendRingLen = 0x200,

	Reset = 1<<0,
	Enable = 1<<1,
	Attn = 1<<2,
	
	PowerControlStatus = 0x4C,

	MiscHostCtl = 0x68,
	ClearIntA = 1<<0,
	MaskPCIInt = 1<<1,
	IndirectAccessEnable = 1<<7,
	EnablePCIStateRegister = 1<<4,
	EnableClockControlRegister = 1<<5,
	TaggedStatus = 1<<9,
	
	Fwmbox		= 0x0b50,	/* magic value exchange */
	Fwmagic		= 0x4b657654,

	DMARWControl = 0x6C,
	DMAWatermarkMask = ~(7<<19),
	DMAWatermarkValue = 3<<19,

	MemoryWindow = 0x7C,
	MemoryWindowData = 0x84,
	
	SendRCB = 0x100,
	RecvRetRCB = 0x200,
	
	InterruptMailbox = 0x204,
	
	RecvProdBDRingIndex = 0x26c,
	RecvBDRetRingIndex = 0x284,
	SendBDRingHostIndex = 0x304,
	
	MACMode = 0x400,
	MACPortMask = ~((1<<3)|(1<<2)),
	MACPortGMII = 1<<3,
	MACPortMII = 1<<2,
	MACEnable = (1<<23) | (1<<22) | (1<<21) | (1 << 15) | (1 << 14) | (1<<12) | (1<<11),
	MACHalfDuplex = 1<<1,
	
	MACEventStatus = 0x404,
	MACEventEnable = 0x408,
	MACAddress = 0x410,
	EthernetRandomBackoff = 0x438,
	ReceiveMTU = 0x43C,
	MIComm = 0x44C,
	MIStatus = 0x450,
	MIMode = 0x454,
	ReceiveMACMode = 0x468,
	TransmitMACMode = 0x45C,
	TransmitMACLengths = 0x464,
	MACHash = 0x470,
	ReceiveRules = 0x480,
	
	ReceiveRulesConfiguration = 0x500,
	LowWatermarkMaximum = 0x504,
	LowWatermarkMaxMask = ~0xFFFF,
	LowWatermarkMaxValue = 2,

	SendDataInitiatorMode = 0xC00,
	SendInitiatorConfiguration = 0x0C08,
	SendStats = 1<<0,
	SendInitiatorMask = 0x0C0C,
	
	SendDataCompletionMode = 0x1000,
	SendBDSelectorMode = 0x1400,
	SendBDInitiatorMode = 0x1800,
	SendBDCompletionMode = 0x1C00,
	
	ReceiveListPlacementMode = 0x2000,
	ReceiveListPlacement = 0x2010,
	ReceiveListPlacementConfiguration = 0x2014,
	ReceiveStats = 1<<0,
	ReceiveListPlacementMask = 0x2018,
	
	ReceiveDataBDInitiatorMode = 0x2400,
	ReceiveBDHostAddr = 0x2450,
	ReceiveBDFlags = 0x2458,
	ReceiveBDNIC = 0x245C,
	ReceiveDataCompletionMode = 0x2800,
	ReceiveBDInitiatorMode = 0x2C00,
	ReceiveBDRepl = 0x2C18,
	
	ReceiveBDCompletionMode = 0x3000,
	HostCoalescingMode = 0x3C00,
	HostCoalescingRecvTicks = 0x3C08,
	HostCoalescingSendTicks = 0x3C0C,
	RecvMaxCoalescedFrames = 0x3C10,
	SendMaxCoalescedFrames = 0x3C14,
	RecvMaxCoalescedFramesInt = 0x3C20,
	SendMaxCoalescedFramesInt = 0x3C24,
	StatusBlockHostAddr = 0x3C38,
	FlowAttention = 0x3C48,

	MemArbiterMode = 0x4000,
	
	BufferManMode = 0x4400,
	
	MBUFLowWatermark = 0x4414,
	MBUFHighWatermark = 0x4418,
	
	ReadDMAMode = 0x4800,
	ReadDMAStatus = 0x4804,
	WriteDMAMode = 0x4C00,
	WriteDMAStatus = 0x4C04,
	
	RISCState = 0x5004,
	FTQReset = 0x5C00,
	MSIMode = 0x6000,
	
	ModeControl = 0x6800,
	ByteWordSwap = (1<<4)|(1<<5)|(1<<2),//|(1<<1),
	HostStackUp = 1<<16,
	HostSendBDs = 1<<17,
	InterruptOnMAC = 1<<26,
	
	MiscConfiguration = 0x6804,
	CoreClockBlocksReset = 1<<0,
	GPHYPowerDownOverride = 1<<26,
	DisableGRCResetOnPCIE = 1<<29,
	TimerMask = ~0xFF,
	TimerValue = 65<<1,
	MiscLocalControl = 0x6808,
	InterruptOnAttn = 1<<3,
	AutoSEEPROM = 1<<24,
	
	SwArbitration = 0x7020,
	SwArbitSet1 = 1<<1,
	SwArbitWon1 = 1<<9,
	TLPControl = 0x7C00,
	
	PhyControl = 0x00,
	PhyStatus = 0x01,
	PhyLinkStatus = 1<<2,
	PhyAutoNegComplete = 1<<5,
	PhyPartnerStatus = 0x05,
	Phy100FD = 1<<8,
	Phy100HD = 1<<7,
	Phy10FD = 1<<6,
	Phy10HD = 1<<5,
	PhyGbitStatus = 0x0A,
	Phy1000FD = 1<<12,
	Phy1000HD = 1<<11,
	PhyAuxControl = 0x18,
	PhyIntStatus = 0x1A,
	PhyIntMask = 0x1B,
	
	Updated = 1<<0,
	LinkStateChange = 1<<1,
	Error = 1<<2,
	
	PacketEnd = 1<<2,
	FrameError = 1<<10,
};

#define csr32(c, r)	((c)->nic[(r)/4])
#define mem32(c, r)	csr32(c, (r)+0x8000)

static Ctlr *bcmhead;
static int debug=1;

static long
bcmifstat(Ether *edev, void *a, long n, ulong offset)
{
	char *s, *p, *e;
	Ctlr *c;

	c = edev->ctlr;
	p = s = malloc(READSTR);
	e = p + READSTR;

	p = seprint(p, e, "nobuf	%ud\n", c->nobuf);
	p = seprint(p, e, "partial	%ud\n", c->partial);
	p = seprint(p, e, "rxerr	%ud\n", c->rxerr);
	p = seprint(p, e, "qfull	%ud\n", c->qfull);
	p = seprint(p, e, "dmaerr	%ud\n", c->dmaerr);

	USED(p);
	n = readstr(offset, a, n, s);
	free(s);

	return n;
}

static int
miir(Ctlr *ctlr, int ra)
{
	while(csr32(ctlr, MIComm) & (1<<29))
		;
	csr32(ctlr, MIComm) = (ra << 16) | (1 << 21) | (1 << 27) | (1 << 29);
	while(csr32(ctlr, MIComm) & (1<<29))
		;
	if(csr32(ctlr, MIComm) & (1<<28))
		return -1;
	return csr32(ctlr, MIComm) & 0xFFFF;
}

static int
miiw(Ctlr *ctlr, int ra, int value)
{
	while(csr32(ctlr, MIComm) & (1<<29))
		;
	csr32(ctlr, MIComm) = (value & 0xFFFF) | (ra << 16) | (1 << 21) | (1 << 27) | (1 << 29);
	while(csr32(ctlr, MIComm) & (1<<29))
		;
	return 0;
}

static void
checklink(Ether *edev)
{
	ulong i;
	Ctlr *ctlr;

	ctlr = edev->ctlr;
	miir(ctlr, PhyStatus); /* dummy read necessary */
	if(!(miir(ctlr, PhyStatus) & PhyLinkStatus)) {
		edev->link = 0;
		edev->mbps = 1000;
		ctlr->duplex = 1;
		dprint("bcm: no link\n");
		goto out;
	}
	edev->link = 1;
	while((miir(ctlr, PhyStatus) & PhyAutoNegComplete) == 0)
		;
	i = miir(ctlr, PhyGbitStatus);
	if(i & (Phy1000FD | Phy1000HD)) {
		edev->mbps = 1000;
		ctlr->duplex = (i & Phy1000FD) != 0;
	} else if(i = miir(ctlr, PhyPartnerStatus), i & (Phy100FD | Phy100HD)) {
		edev->mbps = 100;
		ctlr->duplex = (i & Phy100FD) != 0;
	} else if(i & (Phy10FD | Phy10HD)) {
		edev->mbps = 10;
		ctlr->duplex = (i & Phy10FD) != 0;
	} else {
		edev->link = 0;
		edev->mbps = 1000;
		ctlr->duplex = 1;
		dprint("bcm: link partner supports neither 10/100/1000 Mbps\n"); 
		goto out;
	}
	dprint("bcm: %d Mbps link, %s duplex\n", edev->mbps, ctlr->duplex ? "full" : "half");
out:
	if(ctlr->duplex)
		csr32(ctlr, MACMode) &= ~MACHalfDuplex;
	else
		csr32(ctlr, MACMode) |= MACHalfDuplex;
	if(edev->mbps >= 1000)
		csr32(ctlr, MACMode) = (csr32(ctlr, MACMode) & MACPortMask) | MACPortGMII;
	else
		csr32(ctlr, MACMode) = (csr32(ctlr, MACMode) & MACPortMask) | MACPortMII;
	csr32(ctlr, MACEventStatus) |= (1<<4) | (1<<3); /* undocumented bits (sync and config changed) */
}

static ulong*
currentrecvret(Ctlr *ctlr)
{
	if(ctlr->recvreti == (ctlr->status[4] & 0xFFFF))
		return 0;
	return ctlr->recvret + ctlr->recvreti * 8;
}

static void
consumerecvret(Ctlr *ctlr)
{
	ctlr->recvreti = ctlr->recvreti+1 & RecvRetRingLen-1;
	csr32(ctlr, RecvBDRetRingIndex) = ctlr->recvreti;
}

static int
replenish(Ctlr *ctlr)
{
	ulong *next, incr;
	Block *bp;
	
	incr = (ctlr->recvprodi + 1) & (RecvProdRingLen - 1);
	if(incr == (ctlr->status[2] >> 16))
		return -1;
	bp = iallocb(Rbsz);
	if(bp == nil) {
		/* iallocb never fails.  this code is unnecessary */
		dprint("bcm: out of memory for receive buffers\n");
		ctlr->nobuf++;
		return -1;
	}
	next = ctlr->recvprod + ctlr->recvprodi * 8;
	memset(next, 0, 32);
	next[0] = Pciwaddrh(bp->rp);
	next[1] = Pciwaddrl(bp->rp);
	next[2] = Rbsz;
	next[7] = ctlr->recvprodi;
	ctlr->rxs[ctlr->recvprodi] = bp;
	coherence();
	csr32(ctlr, RecvProdBDRingIndex) = ctlr->recvprodi = incr;
	return 0;
}

static void
bcmreceive(Ether *edev)
{
	ulong *pkt, len;
	Ctlr *ctlr;
	Block *bp;
	
	ctlr = edev->ctlr;
	for(; pkt = currentrecvret(ctlr); replenish(ctlr), consumerecvret(ctlr)) {
		bp = ctlr->rxs[pkt[7]];
		len = pkt[2] & 0xFFFF;
		bp->wp = bp->rp + len;
		if((pkt[3] & PacketEnd) == 0){
			dprint("bcm: partial frame received -- shouldn't happen\n");
			ctlr->partial++;
			freeb(bp);
			continue;
		}
		if(pkt[3] & FrameError){
			ctlr->rxerr++;
			freeb(bp);
			continue;
		}
		etheriq(edev, bp, 1);
	}
}

static void
bcmtransclean(Ether *edev)
{
	Ctlr *ctlr;
	
	ctlr = edev->ctlr;
	ilock(&ctlr->txlock);
	while(ctlr->sendcleani != (ctlr->status[4] >> 16)) {
		freeb(ctlr->sends[ctlr->sendcleani]);
		ctlr->sends[ctlr->sendcleani] = nil;
		ctlr->sendcleani = (ctlr->sendcleani + 1) & (SendRingLen - 1);
	}
	iunlock(&ctlr->txlock);
}

static void
bcmtransmit(Ether *edev)
{
	ulong *next, incr;
	Ctlr *ctlr;
	Block *bp;
	
	ctlr = edev->ctlr;
	ilock(&ctlr->txlock);
	for(;;){
		incr = (ctlr->sendri + 1) & (SendRingLen - 1);
		if(incr == ctlr->sendcleani) {
			dprint("bcm: send queue full\n");
			ctlr->qfull++;
			break;
		}
		bp = qget(edev->oq);
		if(bp == nil)
			break;
		next = ctlr->sendr + ctlr->sendri * 4;
		next[0] = Pciwaddrh(bp->rp);
		next[1] = Pciwaddrl(bp->rp);
		next[2] = (BLEN(bp) << 16) | PacketEnd;
		next[3] = 0;
		ctlr->sends[ctlr->sendri] = bp;
		coherence();
		csr32(ctlr, SendBDRingHostIndex) = ctlr->sendri = incr;
	}
	iunlock(&ctlr->txlock);
}

static void
bcmerror(Ether *edev)
{
	Ctlr *ctlr;
	
	ctlr = edev->ctlr;
	if(csr32(ctlr, FlowAttention)) {
		if(csr32(ctlr, FlowAttention) & 0xf8ff8080)
			print("bcm: fatal error %#.8lux", csr32(ctlr, FlowAttention));
		csr32(ctlr, FlowAttention) = 0;
	}
	csr32(ctlr, MACEventStatus) = 0; /* worth ignoring */
	if(csr32(ctlr, ReadDMAStatus) || csr32(ctlr, WriteDMAStatus)) {
		dprint("bcm: DMA error\n");
		ctlr->dmaerr++;
		csr32(ctlr, ReadDMAStatus) = 0;
		csr32(ctlr, WriteDMAStatus) = 0;
	}
	if(csr32(ctlr, RISCState)) {
		if(csr32(ctlr, RISCState) & 0x78000403)
			print("bcm: RISC halted %#.8lux", csr32(ctlr, RISCState));
		csr32(ctlr, RISCState) = 0;
	}
}

static void
bcminterrupt(Ureg*, void *arg)
{
	ulong status, tag, dummy;
	Ether *edev;
	Ctlr *ctlr;
	
	edev = arg;
	ctlr = edev->ctlr;
	ilock(&ctlr->imlock);
	dummy = csr32(ctlr, InterruptMailbox);
	USED(dummy);
	csr32(ctlr, InterruptMailbox) = 1;
	status = ctlr->status[0];
	tag = ctlr->status[1];
	ctlr->status[0] = 0;
	if(status & Error)
		bcmerror(edev);
	if(status & LinkStateChange)
		checklink(edev);
	iprint("bcm: interrupt %8lux %8lux\n", ctlr->status[2], ctlr->status[4]);
	bcmreceive(edev);
	bcmtransclean(edev);
	bcmtransmit(edev);
	csr32(ctlr, InterruptMailbox) = tag << 24;
	iunlock(&ctlr->imlock);
}

static int
bcminit(Ether *edev)
{
	ulong i, j;
	Ctlr *ctlr;
	
	ctlr = edev->ctlr;
	dprint("bcm: reset\n");
	/* initialization procedure according to the datasheet */
	csr32(ctlr, MiscHostCtl) |= MaskPCIInt | ClearIntA;
	csr32(ctlr, SwArbitration) |= SwArbitSet1;
	for(i = 0;; i += 100){
		if((csr32(ctlr, SwArbitration) & SwArbitWon1))
			break;
		if(i == 2000 /* ?s */){
			print("bcm: arbiter failed to respond\n");
			return -1;
		}
		microdelay(100);
	}
	csr32(ctlr, MemArbiterMode) |= Enable;
	csr32(ctlr, MiscHostCtl) |= IndirectAccessEnable | EnablePCIStateRegister | EnableClockControlRegister;
	csr32(ctlr, MemoryWindow) = 0;
	mem32(ctlr, Fwmbox) = Fwmagic;
	csr32(ctlr, MiscConfiguration) |= GPHYPowerDownOverride | DisableGRCResetOnPCIE;
	csr32(ctlr, MiscConfiguration) |= CoreClockBlocksReset;
	delay(100);
	ctlr->pdev->pcr |= 1<<1;
	pcisetbme(ctlr->pdev);
	csr32(ctlr, MiscHostCtl) |= MaskPCIInt;
	csr32(ctlr, MemArbiterMode) |= Enable;
	csr32(ctlr, MiscHostCtl) |= IndirectAccessEnable | EnablePCIStateRegister | EnableClockControlRegister | TaggedStatus;
	csr32(ctlr, ModeControl) |= ByteWordSwap;
	csr32(ctlr, MACMode) = (csr32(ctlr, MACMode) & MACPortMask) | MACPortGMII;
	delay(40);
	for(i = 0;; i += 100){
		if(mem32(ctlr, Fwmbox) == ~Fwmagic)
			break;
		if(i == 20*10000 /* ?s */){
			print("bcm: fw failed to respond %#.8lux\n", mem32(ctlr, Fwmbox));
			break; //return -1;
		}
		microdelay(100);
	}
	csr32(ctlr, TLPControl) |= (1<<25) | (1<<29);
	memset(ctlr->status, 0, 20);
	csr32(ctlr, DMARWControl) = (csr32(ctlr, DMARWControl) & DMAWatermarkMask) | DMAWatermarkValue;
	csr32(ctlr, ModeControl) |= HostSendBDs | HostStackUp | InterruptOnMAC;
	csr32(ctlr, MiscConfiguration) = (csr32(ctlr, MiscConfiguration) & TimerMask) | TimerValue;
	csr32(ctlr, MBUFLowWatermark) = 0x20;
	csr32(ctlr, MBUFHighWatermark) = 0x60;
	csr32(ctlr, LowWatermarkMaximum) = (csr32(ctlr, LowWatermarkMaximum) & LowWatermarkMaxMask) | LowWatermarkMaxValue;
	csr32(ctlr, BufferManMode) |= Enable | Attn;
	while((csr32(ctlr, BufferManMode) & Enable) == 0)
		;
	csr32(ctlr, FTQReset) = -1;
	csr32(ctlr, FTQReset) = 0;
	while(csr32(ctlr, FTQReset))
		;
	csr32(ctlr, ReceiveBDHostAddr) = Pciwaddrh(ctlr->recvprod);
	csr32(ctlr, ReceiveBDHostAddr + 4) = Pciwaddrl(ctlr->recvprod);
	csr32(ctlr, ReceiveBDFlags) = RecvProdRingLen << 16;
	csr32(ctlr, ReceiveBDNIC) = 0x6000;
	csr32(ctlr, ReceiveBDRepl) = 25;
	csr32(ctlr, SendBDRingHostIndex) = 0;
	csr32(ctlr, SendBDRingHostIndex+4) = 0;
	mem32(ctlr, SendRCB) = Pciwaddrh(ctlr->sendr);
	mem32(ctlr, SendRCB + 4) = Pciwaddrl(ctlr->sendr);
	mem32(ctlr, SendRCB + 8) = SendRingLen << 16;
	mem32(ctlr, SendRCB + 12) = 0x4000;
	for(i=1;i<4;i++)
		mem32(ctlr, RecvRetRCB + i * 0x10 + 8) = 2;
	mem32(ctlr, RecvRetRCB) = Pciwaddrh(ctlr->recvret);
	mem32(ctlr, RecvRetRCB + 4) = Pciwaddrl(ctlr->recvret);
	mem32(ctlr, RecvRetRCB + 8) = RecvRetRingLen << 16;
	csr32(ctlr, RecvProdBDRingIndex) = 0;
	csr32(ctlr, RecvProdBDRingIndex+4) = 0;
	/* this delay is not in the datasheet, but necessary */
	delay(1); 
	i = csr32(ctlr, MACAddress);
	j = edev->ea[0] = i >> 8;
	j += edev->ea[1] = i;
	i = csr32(ctlr, MACAddress + 4);
	j += edev->ea[2] = i >> 24;
	j += edev->ea[3] = i >> 16;
	j += edev->ea[4] = i >> 8;
	j += edev->ea[5] = i;
	csr32(ctlr, EthernetRandomBackoff) = j & 0x3FF;
	csr32(ctlr, ReceiveMTU) = Rbsz;
	csr32(ctlr, TransmitMACLengths) = 0x2620;
	csr32(ctlr, ReceiveListPlacement) = 1<<3; /* one list */
	csr32(ctlr, ReceiveListPlacementMask) = 0xFFFFFF;
	csr32(ctlr, ReceiveListPlacementConfiguration) |= ReceiveStats;
	csr32(ctlr, SendInitiatorMask) = 0xFFFFFF;
	csr32(ctlr, SendInitiatorConfiguration) |= SendStats;
	csr32(ctlr, HostCoalescingMode) = 0;
	while(csr32(ctlr, HostCoalescingMode) != 0)
		;
	csr32(ctlr, HostCoalescingRecvTicks) = 150;
	csr32(ctlr, HostCoalescingSendTicks) = 150;
	csr32(ctlr, RecvMaxCoalescedFrames) = 10;
	csr32(ctlr, SendMaxCoalescedFrames) = 10;
	csr32(ctlr, RecvMaxCoalescedFramesInt) = 0;
	csr32(ctlr, SendMaxCoalescedFramesInt) = 0;
	csr32(ctlr, StatusBlockHostAddr) = Pciwaddrh(ctlr->status);
	csr32(ctlr, StatusBlockHostAddr + 4) = Pciwaddrl(ctlr->status);
	csr32(ctlr, HostCoalescingMode) |= Enable;
	csr32(ctlr, ReceiveBDCompletionMode) |= Enable | Attn;
	csr32(ctlr, ReceiveListPlacementMode) |= Enable;
	csr32(ctlr, MACMode) |= MACEnable;
	csr32(ctlr, MiscLocalControl) |= InterruptOnAttn | AutoSEEPROM;
	csr32(ctlr, InterruptMailbox) = 0;
	csr32(ctlr, WriteDMAMode) |= 0x200003fe; /* pulled out of my nose */
	csr32(ctlr, ReadDMAMode) |= 0x3fe;
	csr32(ctlr, ReceiveDataCompletionMode) |= Enable | Attn;
	csr32(ctlr, SendDataCompletionMode) |= Enable;
	csr32(ctlr, SendBDCompletionMode) |= Enable | Attn;
	csr32(ctlr, ReceiveBDInitiatorMode) |= Enable | Attn;
	csr32(ctlr, ReceiveDataBDInitiatorMode) |= Enable | (1<<4);
	csr32(ctlr, SendDataInitiatorMode) |= Enable;
	csr32(ctlr, SendBDInitiatorMode) |= Enable | Attn;
	csr32(ctlr, SendBDSelectorMode) |= Enable | Attn;
	ctlr->recvprodi = 0;
	while(replenish(ctlr) >= 0)
		;
	csr32(ctlr, TransmitMACMode) |= Enable;
	csr32(ctlr, ReceiveMACMode) |= Enable;
	csr32(ctlr, PowerControlStatus) &= ~3;
	csr32(ctlr, MIStatus) |= 1<<0;
	csr32(ctlr, MACEventEnable) = 0;
	csr32(ctlr, MACEventStatus) |= (1<<12);
	csr32(ctlr, MIMode) = 0xC0000;
	microdelay(40);
	miiw(ctlr, PhyControl, 1<<15);
	while(miir(ctlr, PhyControl) & (1<<15))
		;
	for(i = 0;; i += 100){
		if((miir(ctlr, PhyControl) & (1<<15)) == 0)
			break;
		if(i == 10000 /* ?s */){
			print("bcm: phy reset failure\n");
			return -1;
		}
		microdelay(100);
	}
	miiw(ctlr, PhyAuxControl, 2);
	miir(ctlr, PhyIntStatus);
	miir(ctlr, PhyIntStatus);
	miiw(ctlr, PhyIntMask, ~(1<<1));
	checklink(edev);
	csr32(ctlr, MACEventEnable) |= 1<<12;
	csr32(ctlr, MACHash) = -1;
	csr32(ctlr, MACHash+4) = -1;
	csr32(ctlr, MACHash+8) = -1;
	csr32(ctlr, MACHash+12) = -1;
	for(i = 0; i < 8; i++)
		csr32(ctlr, ReceiveRules + 8 * i) = 0;
	csr32(ctlr, ReceiveRulesConfiguration) = 1 << 3;
	csr32(ctlr, MSIMode) |= Enable;
	csr32(ctlr, MiscHostCtl) &= ~(MaskPCIInt | ClearIntA);
	return 0;
}

static void
bcmpci(void)
{
	void *mem;
	Ctlr *ctlr, **xx;
	Pcidev *p;

	xx = &bcmhead;
	for(p = nil; p = pcimatch(p, 0, 0); ) {
		if(p->ccrb != 2 || p->ccru != 0)
			continue;
		
		switch(p->vid<<16 | p->did){
		default:
			continue;
		case 0x14e4165a:
		case 0x14e4167d:
		case 0x14e41670:
		case 0x14e41672:
		case 0x14e41673:
		case 0x14e41674:
		case 0x14e41677:
		case 0x14e4167a:
		case 0x14e4167b:
		case 0x14e41693:
		case 0x14e4169b:
		case 0x14e41712:
		case 0x14e41713:
			break;
		}
		pcisetbme(p);
		pcisetpms(p, 0);
		ctlr = malloc(sizeof(Ctlr));
		if(ctlr == nil)
			continue;
		ctlr->port = p->mem[0].bar & ~0x0F;
		mem = vmap(ctlr->port, p->mem[0].size);
		if(mem == nil) {
			print("bcm: can't map %#p\n", ctlr->port);
			free(ctlr);
			continue;
		}
		ctlr->pdev = p;
		ctlr->nic = mem;
		ctlr->status = xspanalloc(20, 16, 0);
		ctlr->recvprod = xspanalloc(32 * RecvProdRingLen, 16, 0);
		ctlr->recvret = xspanalloc(32 * RecvRetRingLen, 16, 0);
		ctlr->sendr = xspanalloc(16 * SendRingLen, 16, 0);
		ctlr->sends = malloc(sizeof *ctlr->sends * SendRingLen);
		ctlr->rxs = malloc(sizeof *ctlr->sends * SendRingLen);
		*xx = ctlr;
		xx = &ctlr->next;
	}
}

static void
bcmpromiscuous(void* arg, int on)
{
	Ctlr *ctlr;
	
	ctlr = ((Ether*)arg)->ctlr;
	if(on)
		csr32(ctlr, ReceiveMACMode) |= 1<<8;
	else
		csr32(ctlr, ReceiveMACMode) &= ~(1<<8);
}

static void
bcmmulticast(void*, uchar*, int)
{
}

static int
bcmpnp(Ether* edev)
{
	Ctlr *ctlr;
	static int done;

	if(done == 0){
		bcmpci();
		done = 1;
	}
	
redux:
	for(ctlr = bcmhead; ; ctlr = ctlr->next) {
		if(ctlr == nil)
			return -1;
		if(ctlr->active)
			continue;
		if(edev->port == 0 || edev->port == ctlr->port) {
			ctlr->active = 1;
			break;
		}
	}

	edev->ctlr = ctlr;
	edev->port = ctlr->port;
	edev->irq = ctlr->pdev->intl;
	edev->tbdf = ctlr->pdev->tbdf;
	edev->interrupt = bcminterrupt;
	edev->ifstat = bcmifstat;
	edev->transmit = bcmtransmit;
	edev->multicast = bcmmulticast;
	edev->promiscuous = bcmpromiscuous;
	edev->arg = edev;
	edev->mbps = 1000;
	
	if(bcminit(edev) == -1)
		goto redux;
	return 0;
}

void
etherbcmlink(void)
{
	addethercard("bcm57xx", bcmpnp);
}