linux/drivers/net/tokenring/3c359.c
Adrian Bunk 9b5587cdd9 tokenring/3c359.c:xl_init() must be __devinit
This patch fixes the following section mismatch:

<--  snip  -->

...
  MODPOST vmlinux
WARNING: drivers/built-in.o(.text+0x361ef1): Section mismatch: reference to .init.text:xl_init (between 'xl_probe' and 'xl_hw_reset')
...

<--  snip  -->

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
2007-07-10 12:24:42 -04:00

1830 lines
59 KiB
C

/*
* 3c359.c (c) 2000 Mike Phillips (mikep@linuxtr.net) All Rights Reserved
*
* Linux driver for 3Com 3c359 Tokenlink Velocity XL PCI NIC
*
* Base Driver Olympic:
* Written 1999 Peter De Schrijver & Mike Phillips
*
* This software may be used and distributed according to the terms
* of the GNU General Public License, incorporated herein by reference.
*
* 7/17/00 - Clean up, version number 0.9.0. Ready to release to the world.
*
* 2/16/01 - Port up to kernel 2.4.2 ready for submission into the kernel.
* 3/05/01 - Last clean up stuff before submission.
* 2/15/01 - Finally, update to new pci api.
*
* To Do:
*/
/*
* Technical Card Details
*
* All access to data is done with 16/8 bit transfers. The transfer
* method really sucks. You can only read or write one location at a time.
*
* Also, the microcode for the card must be uploaded if the card does not have
* the flashrom on board. This is a 28K bloat in the driver when compiled
* as a module.
*
* Rx is very simple, status into a ring of descriptors, dma data transfer,
* interrupts to tell us when a packet is received.
*
* Tx is a little more interesting. Similar scenario, descriptor and dma data
* transfers, but we don't have to interrupt the card to tell it another packet
* is ready for transmission, we are just doing simple memory writes, not io or mmio
* writes. The card can be set up to simply poll on the next
* descriptor pointer and when this value is non-zero will automatically download
* the next packet. The card then interrupts us when the packet is done.
*
*/
#define XL_DEBUG 0
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/timer.h>
#include <linux/in.h>
#include <linux/ioport.h>
#include <linux/string.h>
#include <linux/proc_fs.h>
#include <linux/ptrace.h>
#include <linux/skbuff.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
#include <linux/netdevice.h>
#include <linux/trdevice.h>
#include <linux/stddef.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/spinlock.h>
#include <linux/bitops.h>
#include <net/checksum.h>
#include <asm/io.h>
#include <asm/system.h>
#include "3c359.h"
static char version[] __devinitdata =
"3c359.c v1.2.0 2/17/01 - Mike Phillips (mikep@linuxtr.net)" ;
MODULE_AUTHOR("Mike Phillips <mikep@linuxtr.net>") ;
MODULE_DESCRIPTION("3Com 3C359 Velocity XL Token Ring Adapter Driver \n") ;
/* Module paramters */
/* Ring Speed 0,4,16
* 0 = Autosense
* 4,16 = Selected speed only, no autosense
* This allows the card to be the first on the ring
* and become the active monitor.
*
* WARNING: Some hubs will allow you to insert
* at the wrong speed.
*
* The adapter will _not_ fail to open if there are no
* active monitors on the ring, it will simply open up in
* its last known ringspeed if no ringspeed is specified.
*/
static int ringspeed[XL_MAX_ADAPTERS] = {0,} ;
module_param_array(ringspeed, int, NULL, 0);
MODULE_PARM_DESC(ringspeed,"3c359: Ringspeed selection - 4,16 or 0") ;
/* Packet buffer size */
static int pkt_buf_sz[XL_MAX_ADAPTERS] = {0,} ;
module_param_array(pkt_buf_sz, int, NULL, 0) ;
MODULE_PARM_DESC(pkt_buf_sz,"3c359: Initial buffer size") ;
/* Message Level */
static int message_level[XL_MAX_ADAPTERS] = {0,} ;
module_param_array(message_level, int, NULL, 0) ;
MODULE_PARM_DESC(message_level, "3c359: Level of reported messages \n") ;
/*
* This is a real nasty way of doing this, but otherwise you
* will be stuck with 1555 lines of hex #'s in the code.
*/
#include "3c359_microcode.h"
static struct pci_device_id xl_pci_tbl[] =
{
{PCI_VENDOR_ID_3COM,PCI_DEVICE_ID_3COM_3C359, PCI_ANY_ID, PCI_ANY_ID, },
{ } /* terminate list */
};
MODULE_DEVICE_TABLE(pci,xl_pci_tbl) ;
static int xl_init(struct net_device *dev);
static int xl_open(struct net_device *dev);
static int xl_open_hw(struct net_device *dev) ;
static int xl_hw_reset(struct net_device *dev);
static int xl_xmit(struct sk_buff *skb, struct net_device *dev);
static void xl_dn_comp(struct net_device *dev);
static int xl_close(struct net_device *dev);
static void xl_set_rx_mode(struct net_device *dev);
static irqreturn_t xl_interrupt(int irq, void *dev_id);
static struct net_device_stats * xl_get_stats(struct net_device *dev);
static int xl_set_mac_address(struct net_device *dev, void *addr) ;
static void xl_arb_cmd(struct net_device *dev);
static void xl_asb_cmd(struct net_device *dev) ;
static void xl_srb_cmd(struct net_device *dev, int srb_cmd) ;
static void xl_wait_misr_flags(struct net_device *dev) ;
static int xl_change_mtu(struct net_device *dev, int mtu);
static void xl_srb_bh(struct net_device *dev) ;
static void xl_asb_bh(struct net_device *dev) ;
static void xl_reset(struct net_device *dev) ;
static void xl_freemem(struct net_device *dev) ;
/* EEProm Access Functions */
static u16 xl_ee_read(struct net_device *dev, int ee_addr) ;
static void xl_ee_write(struct net_device *dev, int ee_addr, u16 ee_value) ;
/* Debugging functions */
#if XL_DEBUG
static void print_tx_state(struct net_device *dev) ;
static void print_rx_state(struct net_device *dev) ;
static void print_tx_state(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
struct xl_tx_desc *txd ;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
int i ;
printk("tx_ring_head: %d, tx_ring_tail: %d, free_ent: %d \n",xl_priv->tx_ring_head,
xl_priv->tx_ring_tail, xl_priv->free_ring_entries) ;
printk("Ring , Address , FSH , DnNextPtr, Buffer, Buffer_Len \n");
for (i = 0; i < 16; i++) {
txd = &(xl_priv->xl_tx_ring[i]) ;
printk("%d, %08lx, %08x, %08x, %08x, %08x \n", i, virt_to_bus(txd),
txd->framestartheader, txd->dnnextptr, txd->buffer, txd->buffer_length ) ;
}
printk("DNLISTPTR = %04x \n", readl(xl_mmio + MMIO_DNLISTPTR) );
printk("DmaCtl = %04x \n", readl(xl_mmio + MMIO_DMA_CTRL) );
printk("Queue status = %0x \n",netif_running(dev) ) ;
}
static void print_rx_state(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
struct xl_rx_desc *rxd ;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
int i ;
printk("rx_ring_tail: %d \n", xl_priv->rx_ring_tail) ;
printk("Ring , Address , FrameState , UPNextPtr, FragAddr, Frag_Len \n");
for (i = 0; i < 16; i++) {
/* rxd = (struct xl_rx_desc *)xl_priv->rx_ring_dma_addr + (i * sizeof(struct xl_rx_desc)) ; */
rxd = &(xl_priv->xl_rx_ring[i]) ;
printk("%d, %08lx, %08x, %08x, %08x, %08x \n", i, virt_to_bus(rxd),
rxd->framestatus, rxd->upnextptr, rxd->upfragaddr, rxd->upfraglen ) ;
}
printk("UPLISTPTR = %04x \n", readl(xl_mmio + MMIO_UPLISTPTR) );
printk("DmaCtl = %04x \n", readl(xl_mmio + MMIO_DMA_CTRL) );
printk("Queue status = %0x \n",netif_running(dev) ) ;
}
#endif
/*
* Read values from the on-board EEProm. This looks very strange
* but you have to wait for the EEProm to get/set the value before
* passing/getting the next value from the nic. As with all requests
* on this nic it has to be done in two stages, a) tell the nic which
* memory address you want to access and b) pass/get the value from the nic.
* With the EEProm, you have to wait before and inbetween access a) and b).
* As this is only read at initialization time and the wait period is very
* small we shouldn't have to worry about scheduling issues.
*/
static u16 xl_ee_read(struct net_device *dev, int ee_addr)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
/* Wait for EEProm to not be busy */
writel(IO_WORD_READ | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while ( readw(xl_mmio + MMIO_MACDATA) & EEBUSY ) ;
/* Tell EEProm what we want to do and where */
writel(IO_WORD_WRITE | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(EEREAD + ee_addr, xl_mmio + MMIO_MACDATA) ;
/* Wait for EEProm to not be busy */
writel(IO_WORD_READ | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while ( readw(xl_mmio + MMIO_MACDATA) & EEBUSY ) ;
/* Tell EEProm what we want to do and where */
writel(IO_WORD_WRITE | EECONTROL , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(EEREAD + ee_addr, xl_mmio + MMIO_MACDATA) ;
/* Finally read the value from the EEProm */
writel(IO_WORD_READ | EEDATA , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
return readw(xl_mmio + MMIO_MACDATA) ;
}
/*
* Write values to the onboard eeprom. As with eeprom read you need to
* set which location to write, wait, value to write, wait, with the
* added twist of having to enable eeprom writes as well.
*/
static void xl_ee_write(struct net_device *dev, int ee_addr, u16 ee_value)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
/* Wait for EEProm to not be busy */
writel(IO_WORD_READ | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while ( readw(xl_mmio + MMIO_MACDATA) & EEBUSY ) ;
/* Enable write/erase */
writel(IO_WORD_WRITE | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(EE_ENABLE_WRITE, xl_mmio + MMIO_MACDATA) ;
/* Wait for EEProm to not be busy */
writel(IO_WORD_READ | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while ( readw(xl_mmio + MMIO_MACDATA) & EEBUSY ) ;
/* Put the value we want to write into EEDATA */
writel(IO_WORD_WRITE | EEDATA, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(ee_value, xl_mmio + MMIO_MACDATA) ;
/* Tell EEProm to write eevalue into ee_addr */
writel(IO_WORD_WRITE | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(EEWRITE + ee_addr, xl_mmio + MMIO_MACDATA) ;
/* Wait for EEProm to not be busy, to ensure write gets done */
writel(IO_WORD_READ | EECONTROL, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while ( readw(xl_mmio + MMIO_MACDATA) & EEBUSY ) ;
return ;
}
static int __devinit xl_probe(struct pci_dev *pdev,
const struct pci_device_id *ent)
{
struct net_device *dev ;
struct xl_private *xl_priv ;
static int card_no = -1 ;
int i ;
card_no++ ;
if (pci_enable_device(pdev)) {
return -ENODEV ;
}
pci_set_master(pdev);
if ((i = pci_request_regions(pdev,"3c359"))) {
return i ;
} ;
/*
* Allowing init_trdev to allocate the dev->priv structure will align xl_private
* on a 32 bytes boundary which we need for the rx/tx descriptors
*/
dev = alloc_trdev(sizeof(struct xl_private)) ;
if (!dev) {
pci_release_regions(pdev) ;
return -ENOMEM ;
}
xl_priv = dev->priv ;
#if XL_DEBUG
printk("pci_device: %p, dev:%p, dev->priv: %p, ba[0]: %10x, ba[1]:%10x\n",
pdev, dev, dev->priv, (unsigned int)pdev->resource[0].start, (unsigned int)pdev->resource[1].start) ;
#endif
dev->irq=pdev->irq;
dev->base_addr=pci_resource_start(pdev,0) ;
xl_priv->xl_card_name = pci_name(pdev);
xl_priv->xl_mmio=ioremap(pci_resource_start(pdev,1), XL_IO_SPACE);
xl_priv->pdev = pdev ;
if ((pkt_buf_sz[card_no] < 100) || (pkt_buf_sz[card_no] > 18000) )
xl_priv->pkt_buf_sz = PKT_BUF_SZ ;
else
xl_priv->pkt_buf_sz = pkt_buf_sz[card_no] ;
dev->mtu = xl_priv->pkt_buf_sz - TR_HLEN ;
xl_priv->xl_ring_speed = ringspeed[card_no] ;
xl_priv->xl_message_level = message_level[card_no] ;
xl_priv->xl_functional_addr[0] = xl_priv->xl_functional_addr[1] = xl_priv->xl_functional_addr[2] = xl_priv->xl_functional_addr[3] = 0 ;
xl_priv->xl_copy_all_options = 0 ;
if((i = xl_init(dev))) {
iounmap(xl_priv->xl_mmio) ;
free_netdev(dev) ;
pci_release_regions(pdev) ;
return i ;
}
dev->open=&xl_open;
dev->hard_start_xmit=&xl_xmit;
dev->change_mtu=&xl_change_mtu;
dev->stop=&xl_close;
dev->do_ioctl=NULL;
dev->set_multicast_list=&xl_set_rx_mode;
dev->get_stats=&xl_get_stats ;
dev->set_mac_address=&xl_set_mac_address ;
SET_MODULE_OWNER(dev);
SET_NETDEV_DEV(dev, &pdev->dev);
pci_set_drvdata(pdev,dev) ;
if ((i = register_netdev(dev))) {
printk(KERN_ERR "3C359, register netdev failed\n") ;
pci_set_drvdata(pdev,NULL) ;
iounmap(xl_priv->xl_mmio) ;
free_netdev(dev) ;
pci_release_regions(pdev) ;
return i ;
}
printk(KERN_INFO "3C359: %s registered as: %s\n",xl_priv->xl_card_name,dev->name) ;
return 0;
}
static int __devinit xl_init(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
printk(KERN_INFO "%s \n", version);
printk(KERN_INFO "%s: I/O at %hx, MMIO at %p, using irq %d\n",
xl_priv->xl_card_name, (unsigned int)dev->base_addr ,xl_priv->xl_mmio, dev->irq);
spin_lock_init(&xl_priv->xl_lock) ;
return xl_hw_reset(dev) ;
}
/*
* Hardware reset. This needs to be a separate entity as we need to reset the card
* when we change the EEProm settings.
*/
static int xl_hw_reset(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
unsigned long t ;
u16 i ;
u16 result_16 ;
u8 result_8 ;
u16 start ;
int j ;
/*
* Reset the card. If the card has got the microcode on board, we have
* missed the initialization interrupt, so we must always do this.
*/
writew( GLOBAL_RESET, xl_mmio + MMIO_COMMAND ) ;
/*
* Must wait for cmdInProgress bit (12) to clear before continuing with
* card configuration.
*/
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 40*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL card not responding to global reset.\n", dev->name);
return -ENODEV;
}
}
/*
* Enable pmbar by setting bit in CPAttention
*/
writel( (IO_BYTE_READ | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
result_8 = readb(xl_mmio + MMIO_MACDATA) ;
result_8 = result_8 | CPA_PMBARVIS ;
writel( (IO_BYTE_WRITE | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(result_8, xl_mmio + MMIO_MACDATA) ;
/*
* Read cpHold bit in pmbar, if cleared we have got Flashrom on board.
* If not, we need to upload the microcode to the card
*/
writel( (IO_WORD_READ | PMBAR),xl_mmio + MMIO_MAC_ACCESS_CMD);
#if XL_DEBUG
printk(KERN_INFO "Read from PMBAR = %04x \n", readw(xl_mmio + MMIO_MACDATA)) ;
#endif
if ( readw( (xl_mmio + MMIO_MACDATA)) & PMB_CPHOLD ) {
/* Set PmBar, privateMemoryBase bits (8:2) to 0 */
writel( (IO_WORD_READ | PMBAR),xl_mmio + MMIO_MAC_ACCESS_CMD);
result_16 = readw(xl_mmio + MMIO_MACDATA) ;
result_16 = result_16 & ~((0x7F) << 2) ;
writel( (IO_WORD_WRITE | PMBAR), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(result_16,xl_mmio + MMIO_MACDATA) ;
/* Set CPAttention, memWrEn bit */
writel( (IO_BYTE_READ | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
result_8 = readb(xl_mmio + MMIO_MACDATA) ;
result_8 = result_8 | CPA_MEMWREN ;
writel( (IO_BYTE_WRITE | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(result_8, xl_mmio + MMIO_MACDATA) ;
/*
* Now to write the microcode into the shared ram
* The microcode must finish at position 0xFFFF, so we must subtract
* to get the start position for the code
*/
start = (0xFFFF - (mc_size) + 1 ) ; /* Looks strange but ensures compiler only uses 16 bit unsigned int for this */
printk(KERN_INFO "3C359: Uploading Microcode: ");
for (i = start, j = 0; j < mc_size; i++, j++) {
writel(MEM_BYTE_WRITE | 0XD0000 | i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(microcode[j],xl_mmio + MMIO_MACDATA) ;
if (j % 1024 == 0)
printk(".");
}
printk("\n") ;
for (i=0;i < 16; i++) {
writel( (MEM_BYTE_WRITE | 0xDFFF0) + i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(microcode[mc_size - 16 + i], xl_mmio + MMIO_MACDATA) ;
}
/*
* Have to write the start address of the upload to FFF4, but
* the address must be >> 4. You do not want to know how long
* it took me to discover this.
*/
writel(MEM_WORD_WRITE | 0xDFFF4, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(start >> 4, xl_mmio + MMIO_MACDATA);
/* Clear the CPAttention, memWrEn Bit */
writel( (IO_BYTE_READ | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
result_8 = readb(xl_mmio + MMIO_MACDATA) ;
result_8 = result_8 & ~CPA_MEMWREN ;
writel( (IO_BYTE_WRITE | CPATTENTION), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(result_8, xl_mmio + MMIO_MACDATA) ;
/* Clear the cpHold bit in pmbar */
writel( (IO_WORD_READ | PMBAR),xl_mmio + MMIO_MAC_ACCESS_CMD);
result_16 = readw(xl_mmio + MMIO_MACDATA) ;
result_16 = result_16 & ~PMB_CPHOLD ;
writel( (IO_WORD_WRITE | PMBAR), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(result_16,xl_mmio + MMIO_MACDATA) ;
} /* If microcode upload required */
/*
* The card should now go though a self test procedure and get itself ready
* to be opened, we must wait for an srb response with the initialization
* information.
*/
#if XL_DEBUG
printk(KERN_INFO "%s: Microcode uploaded, must wait for the self test to complete\n", dev->name);
#endif
writew(SETINDENABLE | 0xFFF, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while ( !(readw(xl_mmio + MMIO_INTSTATUS_AUTO) & INTSTAT_SRB) ) {
schedule();
if(jiffies-t > 15*HZ) {
printk(KERN_ERR "3COM 3C359 Velocity XL card not responding.\n");
return -ENODEV;
}
}
/*
* Write the RxBufArea with D000, RxEarlyThresh, TxStartThresh,
* DnPriReqThresh, read the tech docs if you want to know what
* values they need to be.
*/
writel(MMIO_WORD_WRITE | RXBUFAREA, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(0xD000, xl_mmio + MMIO_MACDATA) ;
writel(MMIO_WORD_WRITE | RXEARLYTHRESH, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(0X0020, xl_mmio + MMIO_MACDATA) ;
writew( SETTXSTARTTHRESH | 0x40 , xl_mmio + MMIO_COMMAND) ;
writeb(0x04, xl_mmio + MMIO_DNBURSTTHRESH) ;
writeb(0x04, xl_mmio + DNPRIREQTHRESH) ;
/*
* Read WRBR to provide the location of the srb block, have to use byte reads not word reads.
* Tech docs have this wrong !!!!
*/
writel(MMIO_BYTE_READ | WRBR, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
xl_priv->srb = readb(xl_mmio + MMIO_MACDATA) << 8 ;
writel( (MMIO_BYTE_READ | WRBR) + 1, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
xl_priv->srb = xl_priv->srb | readb(xl_mmio + MMIO_MACDATA) ;
#if XL_DEBUG
writel(IO_WORD_READ | SWITCHSETTINGS, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if ( readw(xl_mmio + MMIO_MACDATA) & 2) {
printk(KERN_INFO "Default ring speed 4 mbps \n") ;
} else {
printk(KERN_INFO "Default ring speed 16 mbps \n") ;
}
printk(KERN_INFO "%s: xl_priv->srb = %04x\n",xl_priv->xl_card_name, xl_priv->srb);
#endif
return 0;
}
static int xl_open(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
u8 i ;
u16 hwaddr[3] ; /* Should be u8[6] but we get word return values */
int open_err ;
u16 switchsettings, switchsettings_eeprom ;
if(request_irq(dev->irq, &xl_interrupt, IRQF_SHARED , "3c359", dev)) {
return -EAGAIN;
}
/*
* Read the information from the EEPROM that we need. I know we
* should use ntohs, but the word gets stored reversed in the 16
* bit field anyway and it all works its self out when we memcpy
* it into dev->dev_addr.
*/
hwaddr[0] = xl_ee_read(dev,0x10) ;
hwaddr[1] = xl_ee_read(dev,0x11) ;
hwaddr[2] = xl_ee_read(dev,0x12) ;
/* Ring speed */
switchsettings_eeprom = xl_ee_read(dev,0x08) ;
switchsettings = switchsettings_eeprom ;
if (xl_priv->xl_ring_speed != 0) {
if (xl_priv->xl_ring_speed == 4)
switchsettings = switchsettings | 0x02 ;
else
switchsettings = switchsettings & ~0x02 ;
}
/* Only write EEProm if there has been a change */
if (switchsettings != switchsettings_eeprom) {
xl_ee_write(dev,0x08,switchsettings) ;
/* Hardware reset after changing EEProm */
xl_hw_reset(dev) ;
}
memcpy(dev->dev_addr,hwaddr,dev->addr_len) ;
open_err = xl_open_hw(dev) ;
/*
* This really needs to be cleaned up with better error reporting.
*/
if (open_err != 0) { /* Something went wrong with the open command */
if (open_err & 0x07) { /* Wrong speed, retry at different speed */
printk(KERN_WARNING "%s: Open Error, retrying at different ringspeed \n", dev->name) ;
switchsettings = switchsettings ^ 2 ;
xl_ee_write(dev,0x08,switchsettings) ;
xl_hw_reset(dev) ;
open_err = xl_open_hw(dev) ;
if (open_err != 0) {
printk(KERN_WARNING "%s: Open error returned a second time, we're bombing out now\n", dev->name);
free_irq(dev->irq,dev) ;
return -ENODEV ;
}
} else {
printk(KERN_WARNING "%s: Open Error = %04x\n", dev->name, open_err) ;
free_irq(dev->irq,dev) ;
return -ENODEV ;
}
}
/*
* Now to set up the Rx and Tx buffer structures
*/
/* These MUST be on 8 byte boundaries */
xl_priv->xl_tx_ring = kmalloc((sizeof(struct xl_tx_desc) * XL_TX_RING_SIZE) + 7, GFP_DMA | GFP_KERNEL) ;
if (xl_priv->xl_tx_ring == NULL) {
printk(KERN_WARNING "%s: Not enough memory to allocate rx buffers.\n",
dev->name);
free_irq(dev->irq,dev);
return -ENOMEM;
}
xl_priv->xl_rx_ring = kmalloc((sizeof(struct xl_rx_desc) * XL_RX_RING_SIZE) +7, GFP_DMA | GFP_KERNEL) ;
if (xl_priv->xl_tx_ring == NULL) {
printk(KERN_WARNING "%s: Not enough memory to allocate rx buffers.\n",
dev->name);
free_irq(dev->irq,dev);
kfree(xl_priv->xl_tx_ring);
return -ENOMEM;
}
memset(xl_priv->xl_tx_ring,0,sizeof(struct xl_tx_desc) * XL_TX_RING_SIZE) ;
memset(xl_priv->xl_rx_ring,0,sizeof(struct xl_rx_desc) * XL_RX_RING_SIZE) ;
/* Setup Rx Ring */
for (i=0 ; i < XL_RX_RING_SIZE ; i++) {
struct sk_buff *skb ;
skb = dev_alloc_skb(xl_priv->pkt_buf_sz) ;
if (skb==NULL)
break ;
skb->dev = dev ;
xl_priv->xl_rx_ring[i].upfragaddr = pci_map_single(xl_priv->pdev, skb->data,xl_priv->pkt_buf_sz, PCI_DMA_FROMDEVICE) ;
xl_priv->xl_rx_ring[i].upfraglen = xl_priv->pkt_buf_sz | RXUPLASTFRAG;
xl_priv->rx_ring_skb[i] = skb ;
}
if (i==0) {
printk(KERN_WARNING "%s: Not enough memory to allocate rx buffers. Adapter disabled \n",dev->name) ;
free_irq(dev->irq,dev) ;
return -EIO ;
}
xl_priv->rx_ring_no = i ;
xl_priv->rx_ring_tail = 0 ;
xl_priv->rx_ring_dma_addr = pci_map_single(xl_priv->pdev,xl_priv->xl_rx_ring, sizeof(struct xl_rx_desc) * XL_RX_RING_SIZE, PCI_DMA_TODEVICE) ;
for (i=0;i<(xl_priv->rx_ring_no-1);i++) {
xl_priv->xl_rx_ring[i].upnextptr = xl_priv->rx_ring_dma_addr + (sizeof (struct xl_rx_desc) * (i+1)) ;
}
xl_priv->xl_rx_ring[i].upnextptr = 0 ;
writel(xl_priv->rx_ring_dma_addr, xl_mmio + MMIO_UPLISTPTR) ;
/* Setup Tx Ring */
xl_priv->tx_ring_dma_addr = pci_map_single(xl_priv->pdev,xl_priv->xl_tx_ring, sizeof(struct xl_tx_desc) * XL_TX_RING_SIZE,PCI_DMA_TODEVICE) ;
xl_priv->tx_ring_head = 1 ;
xl_priv->tx_ring_tail = 255 ; /* Special marker for first packet */
xl_priv->free_ring_entries = XL_TX_RING_SIZE ;
/*
* Setup the first dummy DPD entry for polling to start working.
*/
xl_priv->xl_tx_ring[0].framestartheader = TXDPDEMPTY ;
xl_priv->xl_tx_ring[0].buffer = 0 ;
xl_priv->xl_tx_ring[0].buffer_length = 0 ;
xl_priv->xl_tx_ring[0].dnnextptr = 0 ;
writel(xl_priv->tx_ring_dma_addr, xl_mmio + MMIO_DNLISTPTR) ;
writel(DNUNSTALL, xl_mmio + MMIO_COMMAND) ;
writel(UPUNSTALL, xl_mmio + MMIO_COMMAND) ;
writel(DNENABLE, xl_mmio + MMIO_COMMAND) ;
writeb(0x40, xl_mmio + MMIO_DNPOLL) ;
/*
* Enable interrupts on the card
*/
writel(SETINTENABLE | INT_MASK, xl_mmio + MMIO_COMMAND) ;
writel(SETINDENABLE | INT_MASK, xl_mmio + MMIO_COMMAND) ;
netif_start_queue(dev) ;
return 0;
}
static int xl_open_hw(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
u8 __iomem *xl_mmio = xl_priv->xl_mmio ;
u16 vsoff ;
char ver_str[33];
int open_err ;
int i ;
unsigned long t ;
/*
* Okay, let's build up the Open.NIC srb command
*
*/
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(OPEN_NIC, xl_mmio + MMIO_MACDATA) ;
/*
* Use this as a test byte, if it comes back with the same value, the command didn't work
*/
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb)+ 2, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0xff,xl_mmio + MMIO_MACDATA) ;
/* Open options */
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb) + 8, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0x00, xl_mmio + MMIO_MACDATA) ;
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb) + 9, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0x00, xl_mmio + MMIO_MACDATA) ;
/*
* Node address, be careful here, the docs say you can just put zeros here and it will use
* the hardware address, it doesn't, you must include the node address in the open command.
*/
if (xl_priv->xl_laa[0]) { /* If using a LAA address */
for (i=10;i<16;i++) {
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb) + i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(xl_priv->xl_laa[i],xl_mmio + MMIO_MACDATA) ;
}
memcpy(dev->dev_addr,xl_priv->xl_laa,dev->addr_len) ;
} else { /* Regular hardware address */
for (i=10;i<16;i++) {
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb) + i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(dev->dev_addr[i-10], xl_mmio + MMIO_MACDATA) ;
}
}
/* Default everything else to 0 */
for (i = 16; i < 34; i++) {
writel( (MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb) + i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0x00,xl_mmio + MMIO_MACDATA) ;
}
/*
* Set the csrb bit in the MISR register
*/
xl_wait_misr_flags(dev) ;
writel(MEM_BYTE_WRITE | MF_CSRB, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0xFF, xl_mmio + MMIO_MACDATA) ;
writel(MMIO_BYTE_WRITE | MISR_SET, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(MISR_CSRB , xl_mmio + MMIO_MACDATA) ;
/*
* Now wait for the command to run
*/
t=jiffies;
while (! (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_SRB)) {
schedule();
if(jiffies-t > 40*HZ) {
printk(KERN_ERR "3COM 3C359 Velocity XL card not responding.\n");
break ;
}
}
/*
* Let's interpret the open response
*/
writel( (MEM_BYTE_READ | 0xD0000 | xl_priv->srb)+2, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if (readb(xl_mmio + MMIO_MACDATA)!=0) {
open_err = readb(xl_mmio + MMIO_MACDATA) << 8 ;
writel( (MEM_BYTE_READ | 0xD0000 | xl_priv->srb) + 7, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
open_err |= readb(xl_mmio + MMIO_MACDATA) ;
return open_err ;
} else {
writel( (MEM_WORD_READ | 0xD0000 | xl_priv->srb) + 8, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
xl_priv->asb = ntohs(readw(xl_mmio + MMIO_MACDATA)) ;
printk(KERN_INFO "%s: Adapter Opened Details: ",dev->name) ;
printk("ASB: %04x",xl_priv->asb ) ;
writel( (MEM_WORD_READ | 0xD0000 | xl_priv->srb) + 10, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
printk(", SRB: %04x",ntohs(readw(xl_mmio + MMIO_MACDATA)) ) ;
writel( (MEM_WORD_READ | 0xD0000 | xl_priv->srb) + 12, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
xl_priv->arb = ntohs(readw(xl_mmio + MMIO_MACDATA)) ;
printk(", ARB: %04x \n",xl_priv->arb ) ;
writel( (MEM_WORD_READ | 0xD0000 | xl_priv->srb) + 14, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
vsoff = ntohs(readw(xl_mmio + MMIO_MACDATA)) ;
/*
* Interesting, sending the individual characters directly to printk was causing klogd to use
* use 100% of processor time, so we build up the string and print that instead.
*/
for (i=0;i<0x20;i++) {
writel( (MEM_BYTE_READ | 0xD0000 | vsoff) + i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
ver_str[i] = readb(xl_mmio + MMIO_MACDATA) ;
}
ver_str[i] = '\0' ;
printk(KERN_INFO "%s: Microcode version String: %s \n",dev->name,ver_str);
}
/*
* Issue the AckInterrupt
*/
writew(ACK_INTERRUPT | SRBRACK | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
return 0 ;
}
/*
* There are two ways of implementing rx on the 359 NIC, either
* interrupt driven or polling. We are going to uses interrupts,
* it is the easier way of doing things.
*
* The Rx works with a ring of Rx descriptors. At initialise time the ring
* entries point to the next entry except for the last entry in the ring
* which points to 0. The card is programmed with the location of the first
* available descriptor and keeps reading the next_ptr until next_ptr is set
* to 0. Hopefully with a ring size of 16 the card will never get to read a next_ptr
* of 0. As the Rx interrupt is received we copy the frame up to the protocol layers
* and then point the end of the ring to our current position and point our current
* position to 0, therefore making the current position the last position on the ring.
* The last position on the ring therefore loops continually loops around the rx ring.
*
* rx_ring_tail is the position on the ring to process next. (Think of a snake, the head
* expands as the card adds new packets and we go around eating the tail processing the
* packets.)
*
* Undoubtably it could be streamlined and improved upon, but at the moment it works
* and the fast path through the routine is fine.
*
* adv_rx_ring could be inlined to increase performance, but its called a *lot* of times
* in xl_rx so would increase the size of the function significantly.
*/
static void adv_rx_ring(struct net_device *dev) /* Advance rx_ring, cut down on bloat in xl_rx */
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
int prev_ring_loc ;
prev_ring_loc = (xl_priv->rx_ring_tail + XL_RX_RING_SIZE - 1) & (XL_RX_RING_SIZE - 1);
xl_priv->xl_rx_ring[prev_ring_loc].upnextptr = xl_priv->rx_ring_dma_addr + (sizeof (struct xl_rx_desc) * xl_priv->rx_ring_tail) ;
xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].framestatus = 0 ;
xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upnextptr = 0 ;
xl_priv->rx_ring_tail++ ;
xl_priv->rx_ring_tail &= (XL_RX_RING_SIZE-1) ;
return ;
}
static void xl_rx(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
struct sk_buff *skb, *skb2 ;
int frame_length = 0, copy_len = 0 ;
int temp_ring_loc ;
/*
* Receive the next frame, loop around the ring until all frames
* have been received.
*/
while (xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].framestatus & (RXUPDCOMPLETE | RXUPDFULL) ) { /* Descriptor to process */
if (xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].framestatus & RXUPDFULL ) { /* UpdFull, Multiple Descriptors used for the frame */
/*
* This is a pain, you need to go through all the descriptors until the last one
* for this frame to find the framelength
*/
temp_ring_loc = xl_priv->rx_ring_tail ;
while (xl_priv->xl_rx_ring[temp_ring_loc].framestatus & RXUPDFULL ) {
temp_ring_loc++ ;
temp_ring_loc &= (XL_RX_RING_SIZE-1) ;
}
frame_length = xl_priv->xl_rx_ring[temp_ring_loc].framestatus & 0x7FFF ;
skb = dev_alloc_skb(frame_length) ;
if (skb==NULL) { /* No memory for frame, still need to roll forward the rx ring */
printk(KERN_WARNING "%s: dev_alloc_skb failed - multi buffer !\n", dev->name) ;
while (xl_priv->rx_ring_tail != temp_ring_loc)
adv_rx_ring(dev) ;
adv_rx_ring(dev) ; /* One more time just for luck :) */
xl_priv->xl_stats.rx_dropped++ ;
writel(ACK_INTERRUPT | UPCOMPACK | LATCH_ACK , xl_mmio + MMIO_COMMAND) ;
return ;
}
while (xl_priv->rx_ring_tail != temp_ring_loc) {
copy_len = xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfraglen & 0x7FFF ;
frame_length -= copy_len ;
pci_dma_sync_single_for_cpu(xl_priv->pdev,xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr,xl_priv->pkt_buf_sz,PCI_DMA_FROMDEVICE) ;
skb_copy_from_linear_data(xl_priv->rx_ring_skb[xl_priv->rx_ring_tail],
skb_put(skb, copy_len),
copy_len);
pci_dma_sync_single_for_device(xl_priv->pdev,xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr,xl_priv->pkt_buf_sz,PCI_DMA_FROMDEVICE) ;
adv_rx_ring(dev) ;
}
/* Now we have found the last fragment */
pci_dma_sync_single_for_cpu(xl_priv->pdev,xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr,xl_priv->pkt_buf_sz,PCI_DMA_FROMDEVICE) ;
skb_copy_from_linear_data(xl_priv->rx_ring_skb[xl_priv->rx_ring_tail],
skb_put(skb,copy_len), frame_length);
/* memcpy(skb_put(skb,frame_length), bus_to_virt(xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr), frame_length) ; */
pci_dma_sync_single_for_device(xl_priv->pdev,xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr,xl_priv->pkt_buf_sz,PCI_DMA_FROMDEVICE) ;
adv_rx_ring(dev) ;
skb->protocol = tr_type_trans(skb,dev) ;
netif_rx(skb) ;
} else { /* Single Descriptor Used, simply swap buffers over, fast path */
frame_length = xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].framestatus & 0x7FFF ;
skb = dev_alloc_skb(xl_priv->pkt_buf_sz) ;
if (skb==NULL) { /* Still need to fix the rx ring */
printk(KERN_WARNING "%s: dev_alloc_skb failed in rx, single buffer \n",dev->name) ;
adv_rx_ring(dev) ;
xl_priv->xl_stats.rx_dropped++ ;
writel(ACK_INTERRUPT | UPCOMPACK | LATCH_ACK , xl_mmio + MMIO_COMMAND) ;
return ;
}
skb2 = xl_priv->rx_ring_skb[xl_priv->rx_ring_tail] ;
pci_unmap_single(xl_priv->pdev, xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr, xl_priv->pkt_buf_sz,PCI_DMA_FROMDEVICE) ;
skb_put(skb2, frame_length) ;
skb2->protocol = tr_type_trans(skb2,dev) ;
xl_priv->rx_ring_skb[xl_priv->rx_ring_tail] = skb ;
xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr = pci_map_single(xl_priv->pdev,skb->data,xl_priv->pkt_buf_sz, PCI_DMA_FROMDEVICE) ;
xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfraglen = xl_priv->pkt_buf_sz | RXUPLASTFRAG ;
adv_rx_ring(dev) ;
xl_priv->xl_stats.rx_packets++ ;
xl_priv->xl_stats.rx_bytes += frame_length ;
netif_rx(skb2) ;
} /* if multiple buffers */
dev->last_rx = jiffies ;
} /* while packet to do */
/* Clear the updComplete interrupt */
writel(ACK_INTERRUPT | UPCOMPACK | LATCH_ACK , xl_mmio + MMIO_COMMAND) ;
return ;
}
/*
* This is ruthless, it doesn't care what state the card is in it will
* completely reset the adapter.
*/
static void xl_reset(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
unsigned long t;
writew( GLOBAL_RESET, xl_mmio + MMIO_COMMAND ) ;
/*
* Must wait for cmdInProgress bit (12) to clear before continuing with
* card configuration.
*/
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
if(jiffies-t > 40*HZ) {
printk(KERN_ERR "3COM 3C359 Velocity XL card not responding.\n");
break ;
}
}
}
static void xl_freemem(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv ;
int i ;
for (i=0;i<XL_RX_RING_SIZE;i++) {
dev_kfree_skb_irq(xl_priv->rx_ring_skb[xl_priv->rx_ring_tail]) ;
pci_unmap_single(xl_priv->pdev,xl_priv->xl_rx_ring[xl_priv->rx_ring_tail].upfragaddr,xl_priv->pkt_buf_sz, PCI_DMA_FROMDEVICE) ;
xl_priv->rx_ring_tail++ ;
xl_priv->rx_ring_tail &= XL_RX_RING_SIZE-1;
}
/* unmap ring */
pci_unmap_single(xl_priv->pdev,xl_priv->rx_ring_dma_addr, sizeof(struct xl_rx_desc) * XL_RX_RING_SIZE, PCI_DMA_FROMDEVICE) ;
pci_unmap_single(xl_priv->pdev,xl_priv->tx_ring_dma_addr, sizeof(struct xl_tx_desc) * XL_TX_RING_SIZE, PCI_DMA_TODEVICE) ;
kfree(xl_priv->xl_rx_ring) ;
kfree(xl_priv->xl_tx_ring) ;
return ;
}
static irqreturn_t xl_interrupt(int irq, void *dev_id)
{
struct net_device *dev = (struct net_device *)dev_id;
struct xl_private *xl_priv =(struct xl_private *)dev->priv;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
u16 intstatus, macstatus ;
if (!dev) {
printk(KERN_WARNING "Device structure dead, aaahhhh !\n") ;
return IRQ_NONE;
}
intstatus = readw(xl_mmio + MMIO_INTSTATUS) ;
if (!(intstatus & 1)) /* We didn't generate the interrupt */
return IRQ_NONE;
spin_lock(&xl_priv->xl_lock) ;
/*
* Process the interrupt
*/
/*
* Something fishy going on here, we shouldn't get 0001 ints, not fatal though.
*/
if (intstatus == 0x0001) {
writel(ACK_INTERRUPT | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
printk(KERN_INFO "%s: 00001 int received \n",dev->name) ;
} else {
if (intstatus & (HOSTERRINT | SRBRINT | ARBCINT | UPCOMPINT | DNCOMPINT | HARDERRINT | (1<<8) | TXUNDERRUN | ASBFINT)) {
/*
* Host Error.
* It may be possible to recover from this, but usually it means something
* is seriously fubar, so we just close the adapter.
*/
if (intstatus & HOSTERRINT) {
printk(KERN_WARNING "%s: Host Error, performing global reset, intstatus = %04x \n",dev->name,intstatus) ;
writew( GLOBAL_RESET, xl_mmio + MMIO_COMMAND ) ;
printk(KERN_WARNING "%s: Resetting hardware: \n", dev->name);
netif_stop_queue(dev) ;
xl_freemem(dev) ;
free_irq(dev->irq,dev);
xl_reset(dev) ;
writel(ACK_INTERRUPT | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
spin_unlock(&xl_priv->xl_lock) ;
return IRQ_HANDLED;
} /* Host Error */
if (intstatus & SRBRINT ) { /* Srbc interrupt */
writel(ACK_INTERRUPT | SRBRACK | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
if (xl_priv->srb_queued)
xl_srb_bh(dev) ;
} /* SRBR Interrupt */
if (intstatus & TXUNDERRUN) { /* Issue DnReset command */
writel(DNRESET, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) { /* Wait for command to run */
/* !!! FIX-ME !!!!
Must put a timeout check here ! */
/* Empty Loop */
}
printk(KERN_WARNING "%s: TX Underrun received \n",dev->name) ;
writel(ACK_INTERRUPT | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
} /* TxUnderRun */
if (intstatus & ARBCINT ) { /* Arbc interrupt */
xl_arb_cmd(dev) ;
} /* Arbc */
if (intstatus & ASBFINT) {
if (xl_priv->asb_queued == 1) {
xl_asb_cmd(dev) ;
} else if (xl_priv->asb_queued == 2) {
xl_asb_bh(dev) ;
} else {
writel(ACK_INTERRUPT | LATCH_ACK | ASBFACK, xl_mmio + MMIO_COMMAND) ;
}
} /* Asbf */
if (intstatus & UPCOMPINT ) /* UpComplete */
xl_rx(dev) ;
if (intstatus & DNCOMPINT ) /* DnComplete */
xl_dn_comp(dev) ;
if (intstatus & HARDERRINT ) { /* Hardware error */
writel(MMIO_WORD_READ | MACSTATUS, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
macstatus = readw(xl_mmio + MMIO_MACDATA) ;
printk(KERN_WARNING "%s: MacStatusError, details: ", dev->name);
if (macstatus & (1<<14))
printk(KERN_WARNING "tchk error: Unrecoverable error \n") ;
if (macstatus & (1<<3))
printk(KERN_WARNING "eint error: Internal watchdog timer expired \n") ;
if (macstatus & (1<<2))
printk(KERN_WARNING "aint error: Host tried to perform invalid operation \n") ;
printk(KERN_WARNING "Instatus = %02x, macstatus = %02x\n",intstatus,macstatus) ;
printk(KERN_WARNING "%s: Resetting hardware: \n", dev->name);
netif_stop_queue(dev) ;
xl_freemem(dev) ;
free_irq(dev->irq,dev);
unregister_netdev(dev) ;
free_netdev(dev) ;
xl_reset(dev) ;
writel(ACK_INTERRUPT | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
spin_unlock(&xl_priv->xl_lock) ;
return IRQ_HANDLED;
}
} else {
printk(KERN_WARNING "%s: Received Unknown interrupt : %04x \n", dev->name, intstatus) ;
writel(ACK_INTERRUPT | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
}
}
/* Turn interrupts back on */
writel( SETINDENABLE | INT_MASK, xl_mmio + MMIO_COMMAND) ;
writel( SETINTENABLE | INT_MASK, xl_mmio + MMIO_COMMAND) ;
spin_unlock(&xl_priv->xl_lock) ;
return IRQ_HANDLED;
}
/*
* Tx - Polling configuration
*/
static int xl_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
struct xl_tx_desc *txd ;
int tx_head, tx_tail, tx_prev ;
unsigned long flags ;
spin_lock_irqsave(&xl_priv->xl_lock,flags) ;
netif_stop_queue(dev) ;
if (xl_priv->free_ring_entries > 1 ) {
/*
* Set up the descriptor for the packet
*/
tx_head = xl_priv->tx_ring_head ;
tx_tail = xl_priv->tx_ring_tail ;
txd = &(xl_priv->xl_tx_ring[tx_head]) ;
txd->dnnextptr = 0 ;
txd->framestartheader = skb->len | TXDNINDICATE ;
txd->buffer = pci_map_single(xl_priv->pdev, skb->data, skb->len, PCI_DMA_TODEVICE) ;
txd->buffer_length = skb->len | TXDNFRAGLAST ;
xl_priv->tx_ring_skb[tx_head] = skb ;
xl_priv->xl_stats.tx_packets++ ;
xl_priv->xl_stats.tx_bytes += skb->len ;
/*
* Set the nextptr of the previous descriptor equal to this descriptor, add XL_TX_RING_SIZE -1
* to ensure no negative numbers in unsigned locations.
*/
tx_prev = (xl_priv->tx_ring_head + XL_TX_RING_SIZE - 1) & (XL_TX_RING_SIZE - 1) ;
xl_priv->tx_ring_head++ ;
xl_priv->tx_ring_head &= (XL_TX_RING_SIZE - 1) ;
xl_priv->free_ring_entries-- ;
xl_priv->xl_tx_ring[tx_prev].dnnextptr = xl_priv->tx_ring_dma_addr + (sizeof (struct xl_tx_desc) * tx_head) ;
/* Sneaky, by doing a read on DnListPtr we can force the card to poll on the DnNextPtr */
/* readl(xl_mmio + MMIO_DNLISTPTR) ; */
netif_wake_queue(dev) ;
spin_unlock_irqrestore(&xl_priv->xl_lock,flags) ;
return 0;
} else {
spin_unlock_irqrestore(&xl_priv->xl_lock,flags) ;
return 1;
}
}
/*
* The NIC has told us that a packet has been downloaded onto the card, we must
* find out which packet it has done, clear the skb and information for the packet
* then advance around the ring for all tranmitted packets
*/
static void xl_dn_comp(struct net_device *dev)
{
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
struct xl_tx_desc *txd ;
if (xl_priv->tx_ring_tail == 255) {/* First time */
xl_priv->xl_tx_ring[0].framestartheader = 0 ;
xl_priv->xl_tx_ring[0].dnnextptr = 0 ;
xl_priv->tx_ring_tail = 1 ;
}
while (xl_priv->xl_tx_ring[xl_priv->tx_ring_tail].framestartheader & TXDNCOMPLETE ) {
txd = &(xl_priv->xl_tx_ring[xl_priv->tx_ring_tail]) ;
pci_unmap_single(xl_priv->pdev,txd->buffer, xl_priv->tx_ring_skb[xl_priv->tx_ring_tail]->len, PCI_DMA_TODEVICE) ;
txd->framestartheader = 0 ;
txd->buffer = 0xdeadbeef ;
txd->buffer_length = 0 ;
dev_kfree_skb_irq(xl_priv->tx_ring_skb[xl_priv->tx_ring_tail]) ;
xl_priv->tx_ring_tail++ ;
xl_priv->tx_ring_tail &= (XL_TX_RING_SIZE - 1) ;
xl_priv->free_ring_entries++ ;
}
netif_wake_queue(dev) ;
writel(ACK_INTERRUPT | DNCOMPACK | LATCH_ACK , xl_mmio + MMIO_COMMAND) ;
}
/*
* Close the adapter properly.
* This srb reply cannot be handled from interrupt context as we have
* to free the interrupt from the driver.
*/
static int xl_close(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
unsigned long t ;
netif_stop_queue(dev) ;
/*
* Close the adapter, need to stall the rx and tx queues.
*/
writew(DNSTALL, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-DNSTALL not responding.\n", dev->name);
break ;
}
}
writew(DNDISABLE, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-DNDISABLE not responding.\n", dev->name);
break ;
}
}
writew(UPSTALL, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-UPSTALL not responding.\n", dev->name);
break ;
}
}
/* Turn off interrupts, we will still get the indication though
* so we can trap it
*/
writel(SETINTENABLE, xl_mmio + MMIO_COMMAND) ;
xl_srb_cmd(dev,CLOSE_NIC) ;
t=jiffies;
while (!(readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_SRB)) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-CLOSENIC not responding.\n", dev->name);
break ;
}
}
/* Read the srb response from the adapter */
writel(MEM_BYTE_READ | 0xd0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD);
if (readb(xl_mmio + MMIO_MACDATA) != CLOSE_NIC) {
printk(KERN_INFO "%s: CLOSE_NIC did not get a CLOSE_NIC response \n",dev->name) ;
} else {
writel((MEM_BYTE_READ | 0xd0000 | xl_priv->srb) +2, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if (readb(xl_mmio + MMIO_MACDATA)==0) {
printk(KERN_INFO "%s: Adapter has been closed \n",dev->name) ;
writew(ACK_INTERRUPT | SRBRACK | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
xl_freemem(dev) ;
free_irq(dev->irq,dev) ;
} else {
printk(KERN_INFO "%s: Close nic command returned error code %02x\n",dev->name, readb(xl_mmio + MMIO_MACDATA)) ;
}
}
/* Reset the upload and download logic */
writew(UPRESET, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-UPRESET not responding.\n", dev->name);
break ;
}
}
writew(DNRESET, xl_mmio + MMIO_COMMAND) ;
t=jiffies;
while (readw(xl_mmio + MMIO_INTSTATUS) & INTSTAT_CMD_IN_PROGRESS) {
schedule();
if(jiffies-t > 10*HZ) {
printk(KERN_ERR "%s: 3COM 3C359 Velocity XL-DNRESET not responding.\n", dev->name);
break ;
}
}
xl_hw_reset(dev) ;
return 0 ;
}
static void xl_set_rx_mode(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
struct dev_mc_list *dmi ;
unsigned char dev_mc_address[4] ;
u16 options ;
int i ;
if (dev->flags & IFF_PROMISC)
options = 0x0004 ;
else
options = 0x0000 ;
if (options ^ xl_priv->xl_copy_all_options) { /* Changed, must send command */
xl_priv->xl_copy_all_options = options ;
xl_srb_cmd(dev, SET_RECEIVE_MODE) ;
return ;
}
dev_mc_address[0] = dev_mc_address[1] = dev_mc_address[2] = dev_mc_address[3] = 0 ;
for (i=0,dmi=dev->mc_list;i < dev->mc_count; i++,dmi = dmi->next) {
dev_mc_address[0] |= dmi->dmi_addr[2] ;
dev_mc_address[1] |= dmi->dmi_addr[3] ;
dev_mc_address[2] |= dmi->dmi_addr[4] ;
dev_mc_address[3] |= dmi->dmi_addr[5] ;
}
if (memcmp(xl_priv->xl_functional_addr,dev_mc_address,4) != 0) { /* Options have changed, run the command */
memcpy(xl_priv->xl_functional_addr, dev_mc_address,4) ;
xl_srb_cmd(dev, SET_FUNC_ADDRESS) ;
}
return ;
}
/*
* We issued an srb command and now we must read
* the response from the completed command.
*/
static void xl_srb_bh(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
u8 srb_cmd, ret_code ;
int i ;
writel(MEM_BYTE_READ | 0xd0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
srb_cmd = readb(xl_mmio + MMIO_MACDATA) ;
writel((MEM_BYTE_READ | 0xd0000 | xl_priv->srb) +2, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
ret_code = readb(xl_mmio + MMIO_MACDATA) ;
/* Ret_code is standard across all commands */
switch (ret_code) {
case 1:
printk(KERN_INFO "%s: Command: %d - Invalid Command code\n",dev->name,srb_cmd) ;
break ;
case 4:
printk(KERN_INFO "%s: Command: %d - Adapter is closed, must be open for this command \n",dev->name,srb_cmd) ;
break ;
case 6:
printk(KERN_INFO "%s: Command: %d - Options Invalid for command \n",dev->name,srb_cmd) ;
break ;
case 0: /* Successful command execution */
switch (srb_cmd) {
case READ_LOG: /* Returns 14 bytes of data from the NIC */
if(xl_priv->xl_message_level)
printk(KERN_INFO "%s: READ.LOG 14 bytes of data ",dev->name) ;
/*
* We still have to read the log even if message_level = 0 and we don't want
* to see it
*/
for (i=0;i<14;i++) {
writel(MEM_BYTE_READ | 0xd0000 | xl_priv->srb | i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if(xl_priv->xl_message_level)
printk("%02x:",readb(xl_mmio + MMIO_MACDATA)) ;
}
printk("\n") ;
break ;
case SET_FUNC_ADDRESS:
if(xl_priv->xl_message_level)
printk(KERN_INFO "%s: Functional Address Set \n",dev->name) ;
break ;
case CLOSE_NIC:
if(xl_priv->xl_message_level)
printk(KERN_INFO "%s: Received CLOSE_NIC interrupt in interrupt handler \n",dev->name) ;
break ;
case SET_MULTICAST_MODE:
if(xl_priv->xl_message_level)
printk(KERN_INFO "%s: Multicast options successfully changed\n",dev->name) ;
break ;
case SET_RECEIVE_MODE:
if(xl_priv->xl_message_level) {
if (xl_priv->xl_copy_all_options == 0x0004)
printk(KERN_INFO "%s: Entering promiscuous mode \n", dev->name) ;
else
printk(KERN_INFO "%s: Entering normal receive mode \n",dev->name) ;
}
break ;
} /* switch */
break ;
} /* switch */
return ;
}
static struct net_device_stats * xl_get_stats(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
return (struct net_device_stats *) &xl_priv->xl_stats;
}
static int xl_set_mac_address (struct net_device *dev, void *addr)
{
struct sockaddr *saddr = addr ;
struct xl_private *xl_priv = (struct xl_private *)dev->priv ;
if (netif_running(dev)) {
printk(KERN_WARNING "%s: Cannot set mac/laa address while card is open\n", dev->name) ;
return -EIO ;
}
memcpy(xl_priv->xl_laa, saddr->sa_data,dev->addr_len) ;
if (xl_priv->xl_message_level) {
printk(KERN_INFO "%s: MAC/LAA Set to = %x.%x.%x.%x.%x.%x\n",dev->name, xl_priv->xl_laa[0],
xl_priv->xl_laa[1], xl_priv->xl_laa[2],
xl_priv->xl_laa[3], xl_priv->xl_laa[4],
xl_priv->xl_laa[5]);
}
return 0 ;
}
static void xl_arb_cmd(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
u8 arb_cmd ;
u16 lan_status, lan_status_diff ;
writel( ( MEM_BYTE_READ | 0xD0000 | xl_priv->arb), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
arb_cmd = readb(xl_mmio + MMIO_MACDATA) ;
if (arb_cmd == RING_STATUS_CHANGE) { /* Ring.Status.Change */
writel( ( (MEM_WORD_READ | 0xD0000 | xl_priv->arb) + 6), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
printk(KERN_INFO "%s: Ring Status Change: New Status = %04x\n", dev->name, ntohs(readw(xl_mmio + MMIO_MACDATA) )) ;
lan_status = ntohs(readw(xl_mmio + MMIO_MACDATA));
/* Acknowledge interrupt, this tells nic we are done with the arb */
writel(ACK_INTERRUPT | ARBCACK | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
lan_status_diff = xl_priv->xl_lan_status ^ lan_status ;
if (lan_status_diff & (LSC_LWF | LSC_ARW | LSC_FPE | LSC_RR) ) {
if (lan_status_diff & LSC_LWF)
printk(KERN_WARNING "%s: Short circuit detected on the lobe\n",dev->name);
if (lan_status_diff & LSC_ARW)
printk(KERN_WARNING "%s: Auto removal error\n",dev->name);
if (lan_status_diff & LSC_FPE)
printk(KERN_WARNING "%s: FDX Protocol Error\n",dev->name);
if (lan_status_diff & LSC_RR)
printk(KERN_WARNING "%s: Force remove MAC frame received\n",dev->name);
/* Adapter has been closed by the hardware */
netif_stop_queue(dev);
xl_freemem(dev) ;
free_irq(dev->irq,dev);
printk(KERN_WARNING "%s: Adapter has been closed \n", dev->name) ;
} /* If serious error */
if (xl_priv->xl_message_level) {
if (lan_status_diff & LSC_SIG_LOSS)
printk(KERN_WARNING "%s: No receive signal detected \n", dev->name) ;
if (lan_status_diff & LSC_HARD_ERR)
printk(KERN_INFO "%s: Beaconing \n",dev->name);
if (lan_status_diff & LSC_SOFT_ERR)
printk(KERN_WARNING "%s: Adapter transmitted Soft Error Report Mac Frame \n",dev->name);
if (lan_status_diff & LSC_TRAN_BCN)
printk(KERN_INFO "%s: We are tranmitting the beacon, aaah\n",dev->name);
if (lan_status_diff & LSC_SS)
printk(KERN_INFO "%s: Single Station on the ring \n", dev->name);
if (lan_status_diff & LSC_RING_REC)
printk(KERN_INFO "%s: Ring recovery ongoing\n",dev->name);
if (lan_status_diff & LSC_FDX_MODE)
printk(KERN_INFO "%s: Operating in FDX mode\n",dev->name);
}
if (lan_status_diff & LSC_CO) {
if (xl_priv->xl_message_level)
printk(KERN_INFO "%s: Counter Overflow \n", dev->name);
/* Issue READ.LOG command */
xl_srb_cmd(dev, READ_LOG) ;
}
/* There is no command in the tech docs to issue the read_sr_counters */
if (lan_status_diff & LSC_SR_CO) {
if (xl_priv->xl_message_level)
printk(KERN_INFO "%s: Source routing counters overflow\n", dev->name);
}
xl_priv->xl_lan_status = lan_status ;
} /* Lan.change.status */
else if ( arb_cmd == RECEIVE_DATA) { /* Received.Data */
#if XL_DEBUG
printk(KERN_INFO "Received.Data \n") ;
#endif
writel( ((MEM_WORD_READ | 0xD0000 | xl_priv->arb) + 6), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
xl_priv->mac_buffer = ntohs(readw(xl_mmio + MMIO_MACDATA)) ;
/* Now we are going to be really basic here and not do anything
* with the data at all. The tech docs do not give me enough
* information to calculate the buffers properly so we're
* just going to tell the nic that we've dealt with the frame
* anyway.
*/
dev->last_rx = jiffies ;
/* Acknowledge interrupt, this tells nic we are done with the arb */
writel(ACK_INTERRUPT | ARBCACK | LATCH_ACK, xl_mmio + MMIO_COMMAND) ;
/* Is the ASB free ? */
xl_priv->asb_queued = 0 ;
writel( ((MEM_BYTE_READ | 0xD0000 | xl_priv->asb) + 2), xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if (readb(xl_mmio + MMIO_MACDATA) != 0xff) {
xl_priv->asb_queued = 1 ;
xl_wait_misr_flags(dev) ;
writel(MEM_BYTE_WRITE | MF_ASBFR, xl_mmio + MMIO_MAC_ACCESS_CMD);
writeb(0xff, xl_mmio + MMIO_MACDATA) ;
writel(MMIO_BYTE_WRITE | MISR_SET, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(MISR_ASBFR, xl_mmio + MMIO_MACDATA) ;
return ;
/* Drop out and wait for the bottom half to be run */
}
xl_asb_cmd(dev) ;
} else {
printk(KERN_WARNING "%s: Received unknown arb (xl_priv) command: %02x \n",dev->name,arb_cmd) ;
}
/* Acknowledge the arb interrupt */
writel(ACK_INTERRUPT | ARBCACK | LATCH_ACK , xl_mmio + MMIO_COMMAND) ;
return ;
}
/*
* There is only one asb command, but we can get called from different
* places.
*/
static void xl_asb_cmd(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
if (xl_priv->asb_queued == 1)
writel(ACK_INTERRUPT | LATCH_ACK | ASBFACK, xl_mmio + MMIO_COMMAND) ;
writel(MEM_BYTE_WRITE | 0xd0000 | xl_priv->asb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0x81, xl_mmio + MMIO_MACDATA) ;
writel(MEM_WORD_WRITE | 0xd0000 | xl_priv->asb | 6, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(ntohs(xl_priv->mac_buffer), xl_mmio + MMIO_MACDATA) ;
xl_wait_misr_flags(dev) ;
writel(MEM_BYTE_WRITE | MF_RASB, xl_mmio + MMIO_MAC_ACCESS_CMD);
writeb(0xff, xl_mmio + MMIO_MACDATA) ;
writel(MMIO_BYTE_WRITE | MISR_SET, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(MISR_RASB, xl_mmio + MMIO_MACDATA) ;
xl_priv->asb_queued = 2 ;
return ;
}
/*
* This will only get called if there was an error
* from the asb cmd.
*/
static void xl_asb_bh(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
u8 ret_code ;
writel(MMIO_BYTE_READ | 0xd0000 | xl_priv->asb | 2, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
ret_code = readb(xl_mmio + MMIO_MACDATA) ;
switch (ret_code) {
case 0x01:
printk(KERN_INFO "%s: ASB Command, unrecognized command code \n",dev->name) ;
break ;
case 0x26:
printk(KERN_INFO "%s: ASB Command, unexpected receive buffer \n", dev->name) ;
break ;
case 0x40:
printk(KERN_INFO "%s: ASB Command, Invalid Station ID \n", dev->name) ;
break ;
}
xl_priv->asb_queued = 0 ;
writel(ACK_INTERRUPT | LATCH_ACK | ASBFACK, xl_mmio + MMIO_COMMAND) ;
return ;
}
/*
* Issue srb commands to the nic
*/
static void xl_srb_cmd(struct net_device *dev, int srb_cmd)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
switch (srb_cmd) {
case READ_LOG:
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(READ_LOG, xl_mmio + MMIO_MACDATA) ;
break;
case CLOSE_NIC:
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(CLOSE_NIC, xl_mmio + MMIO_MACDATA) ;
break ;
case SET_RECEIVE_MODE:
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(SET_RECEIVE_MODE, xl_mmio + MMIO_MACDATA) ;
writel(MEM_WORD_WRITE | 0xD0000 | xl_priv->srb | 4, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writew(xl_priv->xl_copy_all_options, xl_mmio + MMIO_MACDATA) ;
break ;
case SET_FUNC_ADDRESS:
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(SET_FUNC_ADDRESS, xl_mmio + MMIO_MACDATA) ;
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb | 6 , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(xl_priv->xl_functional_addr[0], xl_mmio + MMIO_MACDATA) ;
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb | 7 , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(xl_priv->xl_functional_addr[1], xl_mmio + MMIO_MACDATA) ;
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb | 8 , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(xl_priv->xl_functional_addr[2], xl_mmio + MMIO_MACDATA) ;
writel(MEM_BYTE_WRITE | 0xD0000 | xl_priv->srb | 9 , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(xl_priv->xl_functional_addr[3], xl_mmio + MMIO_MACDATA) ;
break ;
} /* switch */
xl_wait_misr_flags(dev) ;
/* Write 0xff to the CSRB flag */
writel(MEM_BYTE_WRITE | MF_CSRB , xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0xFF, xl_mmio + MMIO_MACDATA) ;
/* Set csrb bit in MISR register to process command */
writel(MMIO_BYTE_WRITE | MISR_SET, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(MISR_CSRB, xl_mmio + MMIO_MACDATA) ;
xl_priv->srb_queued = 1 ;
return ;
}
/*
* This is nasty, to use the MISR command you have to wait for 6 memory locations
* to be zero. This is the way the driver does on other OS'es so we should be ok with
* the empty loop.
*/
static void xl_wait_misr_flags(struct net_device *dev)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv ;
u8 __iomem * xl_mmio = xl_priv->xl_mmio ;
int i ;
writel(MMIO_BYTE_READ | MISR_RW, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
if (readb(xl_mmio + MMIO_MACDATA) != 0) { /* Misr not clear */
for (i=0; i<6; i++) {
writel(MEM_BYTE_READ | 0xDFFE0 | i, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
while (readb(xl_mmio + MMIO_MACDATA) != 0 ) {} ; /* Empty Loop */
}
}
writel(MMIO_BYTE_WRITE | MISR_AND, xl_mmio + MMIO_MAC_ACCESS_CMD) ;
writeb(0x80, xl_mmio + MMIO_MACDATA) ;
return ;
}
/*
* Change mtu size, this should work the same as olympic
*/
static int xl_change_mtu(struct net_device *dev, int mtu)
{
struct xl_private *xl_priv = (struct xl_private *) dev->priv;
u16 max_mtu ;
if (xl_priv->xl_ring_speed == 4)
max_mtu = 4500 ;
else
max_mtu = 18000 ;
if (mtu > max_mtu)
return -EINVAL ;
if (mtu < 100)
return -EINVAL ;
dev->mtu = mtu ;
xl_priv->pkt_buf_sz = mtu + TR_HLEN ;
return 0 ;
}
static void __devexit xl_remove_one (struct pci_dev *pdev)
{
struct net_device *dev = pci_get_drvdata(pdev);
struct xl_private *xl_priv=(struct xl_private *)dev->priv;
unregister_netdev(dev);
iounmap(xl_priv->xl_mmio) ;
pci_release_regions(pdev) ;
pci_set_drvdata(pdev,NULL) ;
free_netdev(dev);
return ;
}
static struct pci_driver xl_3c359_driver = {
.name = "3c359",
.id_table = xl_pci_tbl,
.probe = xl_probe,
.remove = __devexit_p(xl_remove_one),
};
static int __init xl_pci_init (void)
{
return pci_register_driver(&xl_3c359_driver);
}
static void __exit xl_pci_cleanup (void)
{
pci_unregister_driver (&xl_3c359_driver);
}
module_init(xl_pci_init);
module_exit(xl_pci_cleanup);
MODULE_LICENSE("GPL") ;