kernel_optimize_test/drivers/net/ni65.c

1274 lines
30 KiB
C
Raw Normal View History

/*
* ni6510 (am7990 'lance' chip) driver for Linux-net-3
* BETAcode v0.71 (96/09/29) for 2.0.0 (or later)
* copyrights (c) 1994,1995,1996 by M.Hipp
*
* This driver can handle the old ni6510 board and the newer ni6510
* EtherBlaster. (probably it also works with every full NE2100
* compatible card)
*
* To compile as module, type:
* gcc -O2 -fomit-frame-pointer -m486 -D__KERNEL__ -DMODULE -c ni65.c
* driver probes: io: 0x360,0x300,0x320,0x340 / dma: 3,5,6,7
*
* This is an extension to the Linux operating system, and is covered by the
* same GNU General Public License that covers the Linux-kernel.
*
* comments/bugs/suggestions can be sent to:
* Michael Hipp
* email: hippm@informatik.uni-tuebingen.de
*
* sources:
* some things are from the 'ni6510-packet-driver for dos by Russ Nelson'
* and from the original drivers by D.Becker
*
* known problems:
* - on some PCI boards (including my own) the card/board/ISA-bridge has
* problems with bus master DMA. This results in lotsa overruns.
* It may help to '#define RCV_PARANOIA_CHECK' or try to #undef
* the XMT and RCV_VIA_SKB option .. this reduces driver performance.
* Or just play with your BIOS options to optimize ISA-DMA access.
* Maybe you also wanna play with the LOW_PERFORAMCE and MID_PERFORMANCE
* defines -> please report me your experience then
* - Harald reported for ASUS SP3G mainboards, that you should use
* the 'optimal settings' from the user's manual on page 3-12!
*
* credits:
* thanx to Jason Sullivan for sending me a ni6510 card!
* lot of debug runs with ASUS SP3G Boards (Intel Saturn) by Harald Koenig
*
* simple performance test: (486DX-33/Ni6510-EB receives from 486DX4-100/Ni6510-EB)
* average: FTP -> 8384421 bytes received in 8.5 seconds
* (no RCV_VIA_SKB,no XMT_VIA_SKB,PARANOIA_CHECK,4 XMIT BUFS, 8 RCV_BUFFS)
* peak: FTP -> 8384421 bytes received in 7.5 seconds
* (RCV_VIA_SKB,XMT_VIA_SKB,no PARANOIA_CHECK,1(!) XMIT BUF, 16 RCV BUFFS)
*/
/*
* 99.Jun.8: added support for /proc/net/dev byte count for xosview (HK)
* 96.Sept.29: virt_to_bus stuff added for new memory modell
* 96.April.29: Added Harald Koenig's Patches (MH)
* 96.April.13: enhanced error handling .. more tests (MH)
* 96.April.5/6: a lot of performance tests. Got it stable now (hopefully) (MH)
* 96.April.1: (no joke ;) .. added EtherBlaster and Module support (MH)
* 96.Feb.19: fixed a few bugs .. cleanups .. tested for 1.3.66 (MH)
* hopefully no more 16MB limit
*
* 95.Nov.18: multicast tweaked (AC).
*
* 94.Aug.22: changes in xmit_intr (ack more than one xmitted-packet), ni65_send_packet (p->lock) (MH)
*
* 94.July.16: fixed bugs in recv_skb and skb-alloc stuff (MH)
*/
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/ioport.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/module.h>
#include <linux/bitops.h>
#include <asm/io.h>
#include <asm/dma.h>
#include "ni65.h"
/*
* the current setting allows an acceptable performance
* for 'RCV_PARANOIA_CHECK' read the 'known problems' part in
* the header of this file
* 'invert' the defines for max. performance. This may cause DMA problems
* on some boards (e.g on my ASUS SP3G)
*/
#undef XMT_VIA_SKB
#undef RCV_VIA_SKB
#define RCV_PARANOIA_CHECK
#define MID_PERFORMANCE
#if defined( LOW_PERFORMANCE )
static int isa0=7,isa1=7,csr80=0x0c10;
#elif defined( MID_PERFORMANCE )
static int isa0=5,isa1=5,csr80=0x2810;
#else /* high performance */
static int isa0=4,isa1=4,csr80=0x0017;
#endif
/*
* a few card/vendor specific defines
*/
#define NI65_ID0 0x00
#define NI65_ID1 0x55
#define NI65_EB_ID0 0x52
#define NI65_EB_ID1 0x44
#define NE2100_ID0 0x57
#define NE2100_ID1 0x57
#define PORT p->cmdr_addr
/*
* buffer configuration
*/
#if 1
#define RMDNUM 16
#define RMDNUMMASK 0x80000000
#else
#define RMDNUM 8
#define RMDNUMMASK 0x60000000 /* log2(RMDNUM)<<29 */
#endif
#if 0
#define TMDNUM 1
#define TMDNUMMASK 0x00000000
#else
#define TMDNUM 4
#define TMDNUMMASK 0x40000000 /* log2(TMDNUM)<<29 */
#endif
/* slightly oversized */
#define R_BUF_SIZE 1544
#define T_BUF_SIZE 1544
/*
* lance register defines
*/
#define L_DATAREG 0x00
#define L_ADDRREG 0x02
#define L_RESET 0x04
#define L_CONFIG 0x05
#define L_BUSIF 0x06
/*
* to access the lance/am7990-regs, you have to write
* reg-number into L_ADDRREG, then you can access it using L_DATAREG
*/
#define CSR0 0x00
#define CSR1 0x01
#define CSR2 0x02
#define CSR3 0x03
#define INIT_RING_BEFORE_START 0x1
#define FULL_RESET_ON_ERROR 0x2
#if 0
#define writereg(val,reg) {outw(reg,PORT+L_ADDRREG);inw(PORT+L_ADDRREG); \
outw(val,PORT+L_DATAREG);inw(PORT+L_DATAREG);}
#define readreg(reg) (outw(reg,PORT+L_ADDRREG),inw(PORT+L_ADDRREG),\
inw(PORT+L_DATAREG))
#if 0
#define writedatareg(val) {outw(val,PORT+L_DATAREG);inw(PORT+L_DATAREG);}
#else
#define writedatareg(val) { writereg(val,CSR0); }
#endif
#else
#define writereg(val,reg) {outw(reg,PORT+L_ADDRREG);outw(val,PORT+L_DATAREG);}
#define readreg(reg) (outw(reg,PORT+L_ADDRREG),inw(PORT+L_DATAREG))
#define writedatareg(val) { writereg(val,CSR0); }
#endif
static unsigned char ni_vendor[] = { 0x02,0x07,0x01 };
static struct card {
unsigned char id0,id1;
short id_offset;
short total_size;
short cmd_offset;
short addr_offset;
unsigned char *vendor_id;
char *cardname;
long config;
} cards[] = {
{
.id0 = NI65_ID0,
.id1 = NI65_ID1,
.id_offset = 0x0e,
.total_size = 0x10,
.cmd_offset = 0x0,
.addr_offset = 0x8,
.vendor_id = ni_vendor,
.cardname = "ni6510",
.config = 0x1,
},
{
.id0 = NI65_EB_ID0,
.id1 = NI65_EB_ID1,
.id_offset = 0x0e,
.total_size = 0x18,
.cmd_offset = 0x10,
.addr_offset = 0x0,
.vendor_id = ni_vendor,
.cardname = "ni6510 EtherBlaster",
.config = 0x2,
},
{
.id0 = NE2100_ID0,
.id1 = NE2100_ID1,
.id_offset = 0x0e,
.total_size = 0x18,
.cmd_offset = 0x10,
.addr_offset = 0x0,
.vendor_id = NULL,
.cardname = "generic NE2100",
.config = 0x0,
},
};
#define NUM_CARDS 3
struct priv
{
struct rmd rmdhead[RMDNUM];
struct tmd tmdhead[TMDNUM];
struct init_block ib;
int rmdnum;
int tmdnum,tmdlast;
#ifdef RCV_VIA_SKB
struct sk_buff *recv_skb[RMDNUM];
#else
void *recvbounce[RMDNUM];
#endif
#ifdef XMT_VIA_SKB
struct sk_buff *tmd_skb[TMDNUM];
#endif
void *tmdbounce[TMDNUM];
int tmdbouncenum;
int lock,xmit_queued;
struct net_device_stats stats;
void *self;
int cmdr_addr;
int cardno;
int features;
spinlock_t ring_lock;
};
static int ni65_probe1(struct net_device *dev,int);
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 21:55:46 +08:00
static irqreturn_t ni65_interrupt(int irq, void * dev_id);
static void ni65_recv_intr(struct net_device *dev,int);
static void ni65_xmit_intr(struct net_device *dev,int);
static int ni65_open(struct net_device *dev);
static int ni65_lance_reinit(struct net_device *dev);
static void ni65_init_lance(struct priv *p,unsigned char*,int,int);
static int ni65_send_packet(struct sk_buff *skb, struct net_device *dev);
static void ni65_timeout(struct net_device *dev);
static int ni65_close(struct net_device *dev);
static int ni65_alloc_buffer(struct net_device *dev);
static void ni65_free_buffer(struct priv *p);
static struct net_device_stats *ni65_get_stats(struct net_device *);
static void set_multicast_list(struct net_device *dev);
static int irqtab[] __initdata = { 9,12,15,5 }; /* irq config-translate */
static int dmatab[] __initdata = { 0,3,5,6,7 }; /* dma config-translate and autodetect */
static int debuglevel = 1;
/*
* set 'performance' registers .. we must STOP lance for that
*/
static void ni65_set_performance(struct priv *p)
{
writereg(CSR0_STOP | CSR0_CLRALL,CSR0); /* STOP */
if( !(cards[p->cardno].config & 0x02) )
return;
outw(80,PORT+L_ADDRREG);
if(inw(PORT+L_ADDRREG) != 80)
return;
writereg( (csr80 & 0x3fff) ,80); /* FIFO watermarks */
outw(0,PORT+L_ADDRREG);
outw((short)isa0,PORT+L_BUSIF); /* write ISA 0: DMA_R : isa0 * 50ns */
outw(1,PORT+L_ADDRREG);
outw((short)isa1,PORT+L_BUSIF); /* write ISA 1: DMA_W : isa1 * 50ns */
outw(CSR0,PORT+L_ADDRREG); /* switch back to CSR0 */
}
/*
* open interface (up)
*/
static int ni65_open(struct net_device *dev)
{
struct priv *p = (struct priv *) dev->priv;
int irqval = request_irq(dev->irq, &ni65_interrupt,0,
cards[p->cardno].cardname,dev);
if (irqval) {
printk(KERN_ERR "%s: unable to get IRQ %d (irqval=%d).\n",
dev->name,dev->irq, irqval);
return -EAGAIN;
}
if(ni65_lance_reinit(dev))
{
netif_start_queue(dev);
return 0;
}
else
{
free_irq(dev->irq,dev);
return -EAGAIN;
}
}
/*
* close interface (down)
*/
static int ni65_close(struct net_device *dev)
{
struct priv *p = (struct priv *) dev->priv;
netif_stop_queue(dev);
outw(inw(PORT+L_RESET),PORT+L_RESET); /* that's the hard way */
#ifdef XMT_VIA_SKB
{
int i;
for(i=0;i<TMDNUM;i++)
{
if(p->tmd_skb[i]) {
dev_kfree_skb(p->tmd_skb[i]);
p->tmd_skb[i] = NULL;
}
}
}
#endif
free_irq(dev->irq,dev);
return 0;
}
static void cleanup_card(struct net_device *dev)
{
struct priv *p = (struct priv *) dev->priv;
disable_dma(dev->dma);
free_dma(dev->dma);
release_region(dev->base_addr, cards[p->cardno].total_size);
ni65_free_buffer(p);
}
/* set: io,irq,dma or set it when calling insmod */
static int irq;
static int io;
static int dma;
/*
* Probe The Card (not the lance-chip)
*/
struct net_device * __init ni65_probe(int unit)
{
struct net_device *dev = alloc_etherdev(0);
static int ports[] = {0x360,0x300,0x320,0x340, 0};
int *port;
int err = 0;
if (!dev)
return ERR_PTR(-ENOMEM);
if (unit >= 0) {
sprintf(dev->name, "eth%d", unit);
netdev_boot_setup_check(dev);
irq = dev->irq;
dma = dev->dma;
} else {
dev->base_addr = io;
}
if (dev->base_addr > 0x1ff) { /* Check a single specified location. */
err = ni65_probe1(dev, dev->base_addr);
} else if (dev->base_addr > 0) { /* Don't probe at all. */
err = -ENXIO;
} else {
for (port = ports; *port && ni65_probe1(dev, *port); port++)
;
if (!*port)
err = -ENODEV;
}
if (err)
goto out;
err = register_netdev(dev);
if (err)
goto out1;
return dev;
out1:
cleanup_card(dev);
out:
free_netdev(dev);
return ERR_PTR(err);
}
/*
* this is the real card probe ..
*/
static int __init ni65_probe1(struct net_device *dev,int ioaddr)
{
int i,j;
struct priv *p;
unsigned long flags;
dev->irq = irq;
dev->dma = dma;
for(i=0;i<NUM_CARDS;i++) {
if(!request_region(ioaddr, cards[i].total_size, cards[i].cardname))
continue;
if(cards[i].id_offset >= 0) {
if(inb(ioaddr+cards[i].id_offset+0) != cards[i].id0 ||
inb(ioaddr+cards[i].id_offset+1) != cards[i].id1) {
release_region(ioaddr, cards[i].total_size);
continue;
}
}
if(cards[i].vendor_id) {
for(j=0;j<3;j++)
if(inb(ioaddr+cards[i].addr_offset+j) != cards[i].vendor_id[j]) {
release_region(ioaddr, cards[i].total_size);
continue;
}
}
break;
}
if(i == NUM_CARDS)
return -ENODEV;
for(j=0;j<6;j++)
dev->dev_addr[j] = inb(ioaddr+cards[i].addr_offset+j);
if( (j=ni65_alloc_buffer(dev)) < 0) {
release_region(ioaddr, cards[i].total_size);
return j;
}
p = (struct priv *) dev->priv;
p->cmdr_addr = ioaddr + cards[i].cmd_offset;
p->cardno = i;
spin_lock_init(&p->ring_lock);
printk(KERN_INFO "%s: %s found at %#3x, ", dev->name, cards[p->cardno].cardname , ioaddr);
outw(inw(PORT+L_RESET),PORT+L_RESET); /* first: reset the card */
if( (j=readreg(CSR0)) != 0x4) {
printk("failed.\n");
printk(KERN_ERR "%s: Can't RESET card: %04x\n", dev->name, j);
ni65_free_buffer(p);
release_region(ioaddr, cards[p->cardno].total_size);
return -EAGAIN;
}
outw(88,PORT+L_ADDRREG);
if(inw(PORT+L_ADDRREG) == 88) {
unsigned long v;
v = inw(PORT+L_DATAREG);
v <<= 16;
outw(89,PORT+L_ADDRREG);
v |= inw(PORT+L_DATAREG);
printk("Version %#08lx, ",v);
p->features = INIT_RING_BEFORE_START;
}
else {
printk("ancient LANCE, ");
p->features = 0x0;
}
if(test_bit(0,&cards[i].config)) {
dev->irq = irqtab[(inw(ioaddr+L_CONFIG)>>2)&3];
dev->dma = dmatab[inw(ioaddr+L_CONFIG)&3];
printk("IRQ %d (from card), DMA %d (from card).\n",dev->irq,dev->dma);
}
else {
if(dev->dma == 0) {
/* 'stuck test' from lance.c */
long dma_channels = ((inb(DMA1_STAT_REG) >> 4) & 0x0f) |
(inb(DMA2_STAT_REG) & 0xf0);
for(i=1;i<5;i++) {
int dma = dmatab[i];
if(test_bit(dma,&dma_channels) || request_dma(dma,"ni6510"))
continue;
flags=claim_dma_lock();
disable_dma(dma);
set_dma_mode(dma,DMA_MODE_CASCADE);
enable_dma(dma);
release_dma_lock(flags);
ni65_init_lance(p,dev->dev_addr,0,0); /* trigger memory access */
flags=claim_dma_lock();
disable_dma(dma);
free_dma(dma);
release_dma_lock(flags);
if(readreg(CSR0) & CSR0_IDON)
break;
}
if(i == 5) {
printk("failed.\n");
printk(KERN_ERR "%s: Can't detect DMA channel!\n", dev->name);
ni65_free_buffer(p);
release_region(ioaddr, cards[p->cardno].total_size);
return -EAGAIN;
}
dev->dma = dmatab[i];
printk("DMA %d (autodetected), ",dev->dma);
}
else
printk("DMA %d (assigned), ",dev->dma);
if(dev->irq < 2)
{
unsigned long irq_mask;
ni65_init_lance(p,dev->dev_addr,0,0);
irq_mask = probe_irq_on();
writereg(CSR0_INIT|CSR0_INEA,CSR0); /* trigger interrupt */
msleep(20);
dev->irq = probe_irq_off(irq_mask);
if(!dev->irq)
{
printk("Failed to detect IRQ line!\n");
ni65_free_buffer(p);
release_region(ioaddr, cards[p->cardno].total_size);
return -EAGAIN;
}
printk("IRQ %d (autodetected).\n",dev->irq);
}
else
printk("IRQ %d (assigned).\n",dev->irq);
}
if(request_dma(dev->dma, cards[p->cardno].cardname ) != 0)
{
printk(KERN_ERR "%s: Can't request dma-channel %d\n",dev->name,(int) dev->dma);
ni65_free_buffer(p);
release_region(ioaddr, cards[p->cardno].total_size);
return -EAGAIN;
}
dev->base_addr = ioaddr;
SET_MODULE_OWNER(dev);
dev->open = ni65_open;
dev->stop = ni65_close;
dev->hard_start_xmit = ni65_send_packet;
dev->tx_timeout = ni65_timeout;
dev->watchdog_timeo = HZ/2;
dev->get_stats = ni65_get_stats;
dev->set_multicast_list = set_multicast_list;
return 0; /* everything is OK */
}
/*
* set lance register and trigger init
*/
static void ni65_init_lance(struct priv *p,unsigned char *daddr,int filter,int mode)
{
int i;
u32 pib;
writereg(CSR0_CLRALL|CSR0_STOP,CSR0);
for(i=0;i<6;i++)
p->ib.eaddr[i] = daddr[i];
for(i=0;i<8;i++)
p->ib.filter[i] = filter;
p->ib.mode = mode;
p->ib.trp = (u32) isa_virt_to_bus(p->tmdhead) | TMDNUMMASK;
p->ib.rrp = (u32) isa_virt_to_bus(p->rmdhead) | RMDNUMMASK;
writereg(0,CSR3); /* busmaster/no word-swap */
pib = (u32) isa_virt_to_bus(&p->ib);
writereg(pib & 0xffff,CSR1);
writereg(pib >> 16,CSR2);
writereg(CSR0_INIT,CSR0); /* this changes L_ADDRREG to CSR0 */
for(i=0;i<32;i++)
{
mdelay(4);
if(inw(PORT+L_DATAREG) & (CSR0_IDON | CSR0_MERR) )
break; /* init ok ? */
}
}
/*
* allocate memory area and check the 16MB border
*/
static void *ni65_alloc_mem(struct net_device *dev,char *what,int size,int type)
{
struct sk_buff *skb=NULL;
unsigned char *ptr;
void *ret;
if(type) {
ret = skb = alloc_skb(2+16+size,GFP_KERNEL|GFP_DMA);
if(!skb) {
printk(KERN_WARNING "%s: unable to allocate %s memory.\n",dev->name,what);
return NULL;
}
skb_reserve(skb,2+16);
skb_put(skb,R_BUF_SIZE); /* grab the whole space .. (not necessary) */
ptr = skb->data;
}
else {
ret = ptr = kmalloc(T_BUF_SIZE,GFP_KERNEL | GFP_DMA);
if(!ret) {
printk(KERN_WARNING "%s: unable to allocate %s memory.\n",dev->name,what);
return NULL;
}
}
if( (u32) virt_to_phys(ptr+size) > 0x1000000) {
printk(KERN_WARNING "%s: unable to allocate %s memory in lower 16MB!\n",dev->name,what);
if(type)
kfree_skb(skb);
else
kfree(ptr);
return NULL;
}
return ret;
}
/*
* allocate all memory structures .. send/recv buffers etc ...
*/
static int ni65_alloc_buffer(struct net_device *dev)
{
unsigned char *ptr;
struct priv *p;
int i;
/*
* we need 8-aligned memory ..
*/
ptr = ni65_alloc_mem(dev,"BUFFER",sizeof(struct priv)+8,0);
if(!ptr)
return -ENOMEM;
p = dev->priv = (struct priv *) (((unsigned long) ptr + 7) & ~0x7);
memset((char *) dev->priv,0,sizeof(struct priv));
p->self = ptr;
for(i=0;i<TMDNUM;i++)
{
#ifdef XMT_VIA_SKB
p->tmd_skb[i] = NULL;
#endif
p->tmdbounce[i] = ni65_alloc_mem(dev,"XMIT",T_BUF_SIZE,0);
if(!p->tmdbounce[i]) {
ni65_free_buffer(p);
return -ENOMEM;
}
}
for(i=0;i<RMDNUM;i++)
{
#ifdef RCV_VIA_SKB
p->recv_skb[i] = ni65_alloc_mem(dev,"RECV",R_BUF_SIZE,1);
if(!p->recv_skb[i]) {
ni65_free_buffer(p);
return -ENOMEM;
}
#else
p->recvbounce[i] = ni65_alloc_mem(dev,"RECV",R_BUF_SIZE,0);
if(!p->recvbounce[i]) {
ni65_free_buffer(p);
return -ENOMEM;
}
#endif
}
return 0; /* everything is OK */
}
/*
* free buffers and private struct
*/
static void ni65_free_buffer(struct priv *p)
{
int i;
if(!p)
return;
for(i=0;i<TMDNUM;i++) {
kfree(p->tmdbounce[i]);
#ifdef XMT_VIA_SKB
if(p->tmd_skb[i])
dev_kfree_skb(p->tmd_skb[i]);
#endif
}
for(i=0;i<RMDNUM;i++)
{
#ifdef RCV_VIA_SKB
if(p->recv_skb[i])
dev_kfree_skb(p->recv_skb[i]);
#else
kfree(p->recvbounce[i]);
#endif
}
kfree(p->self);
}
/*
* stop and (re)start lance .. e.g after an error
*/
static void ni65_stop_start(struct net_device *dev,struct priv *p)
{
int csr0 = CSR0_INEA;
writedatareg(CSR0_STOP);
if(debuglevel > 1)
printk(KERN_DEBUG "ni65_stop_start\n");
if(p->features & INIT_RING_BEFORE_START) {
int i;
#ifdef XMT_VIA_SKB
struct sk_buff *skb_save[TMDNUM];
#endif
unsigned long buffer[TMDNUM];
short blen[TMDNUM];
if(p->xmit_queued) {
while(1) {
if((p->tmdhead[p->tmdlast].u.s.status & XMIT_OWN))
break;
p->tmdlast = (p->tmdlast + 1) & (TMDNUM-1);
if(p->tmdlast == p->tmdnum)
break;
}
}
for(i=0;i<TMDNUM;i++) {
struct tmd *tmdp = p->tmdhead + i;
#ifdef XMT_VIA_SKB
skb_save[i] = p->tmd_skb[i];
#endif
buffer[i] = (u32) isa_bus_to_virt(tmdp->u.buffer);
blen[i] = tmdp->blen;
tmdp->u.s.status = 0x0;
}
for(i=0;i<RMDNUM;i++) {
struct rmd *rmdp = p->rmdhead + i;
rmdp->u.s.status = RCV_OWN;
}
p->tmdnum = p->xmit_queued = 0;
writedatareg(CSR0_STRT | csr0);
for(i=0;i<TMDNUM;i++) {
int num = (i + p->tmdlast) & (TMDNUM-1);
p->tmdhead[i].u.buffer = (u32) isa_virt_to_bus((char *)buffer[num]); /* status is part of buffer field */
p->tmdhead[i].blen = blen[num];
if(p->tmdhead[i].u.s.status & XMIT_OWN) {
p->tmdnum = (p->tmdnum + 1) & (TMDNUM-1);
p->xmit_queued = 1;
writedatareg(CSR0_TDMD | CSR0_INEA | csr0);
}
#ifdef XMT_VIA_SKB
p->tmd_skb[i] = skb_save[num];
#endif
}
p->rmdnum = p->tmdlast = 0;
if(!p->lock)
if (p->tmdnum || !p->xmit_queued)
netif_wake_queue(dev);
dev->trans_start = jiffies;
}
else
writedatareg(CSR0_STRT | csr0);
}
/*
* init lance (write init-values .. init-buffers) (open-helper)
*/
static int ni65_lance_reinit(struct net_device *dev)
{
int i;
struct priv *p = (struct priv *) dev->priv;
unsigned long flags;
p->lock = 0;
p->xmit_queued = 0;
flags=claim_dma_lock();
disable_dma(dev->dma); /* I've never worked with dma, but we do it like the packetdriver */
set_dma_mode(dev->dma,DMA_MODE_CASCADE);
enable_dma(dev->dma);
release_dma_lock(flags);
outw(inw(PORT+L_RESET),PORT+L_RESET); /* first: reset the card */
if( (i=readreg(CSR0) ) != 0x4)
{
printk(KERN_ERR "%s: can't RESET %s card: %04x\n",dev->name,
cards[p->cardno].cardname,(int) i);
flags=claim_dma_lock();
disable_dma(dev->dma);
release_dma_lock(flags);
return 0;
}
p->rmdnum = p->tmdnum = p->tmdlast = p->tmdbouncenum = 0;
for(i=0;i<TMDNUM;i++)
{
struct tmd *tmdp = p->tmdhead + i;
#ifdef XMT_VIA_SKB
if(p->tmd_skb[i]) {
dev_kfree_skb(p->tmd_skb[i]);
p->tmd_skb[i] = NULL;
}
#endif
tmdp->u.buffer = 0x0;
tmdp->u.s.status = XMIT_START | XMIT_END;
tmdp->blen = tmdp->status2 = 0;
}
for(i=0;i<RMDNUM;i++)
{
struct rmd *rmdp = p->rmdhead + i;
#ifdef RCV_VIA_SKB
rmdp->u.buffer = (u32) isa_virt_to_bus(p->recv_skb[i]->data);
#else
rmdp->u.buffer = (u32) isa_virt_to_bus(p->recvbounce[i]);
#endif
rmdp->blen = -(R_BUF_SIZE-8);
rmdp->mlen = 0;
rmdp->u.s.status = RCV_OWN;
}
if(dev->flags & IFF_PROMISC)
ni65_init_lance(p,dev->dev_addr,0x00,M_PROM);
else if(dev->mc_count || dev->flags & IFF_ALLMULTI)
ni65_init_lance(p,dev->dev_addr,0xff,0x0);
else
ni65_init_lance(p,dev->dev_addr,0x00,0x00);
/*
* ni65_set_lance_mem() sets L_ADDRREG to CSR0
* NOW, WE WILL NEVER CHANGE THE L_ADDRREG, CSR0 IS ALWAYS SELECTED
*/
if(inw(PORT+L_DATAREG) & CSR0_IDON) {
ni65_set_performance(p);
/* init OK: start lance , enable interrupts */
writedatareg(CSR0_CLRALL | CSR0_INEA | CSR0_STRT);
return 1; /* ->OK */
}
printk(KERN_ERR "%s: can't init lance, status: %04x\n",dev->name,(int) inw(PORT+L_DATAREG));
flags=claim_dma_lock();
disable_dma(dev->dma);
release_dma_lock(flags);
return 0; /* ->Error */
}
/*
* interrupt handler
*/
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 21:55:46 +08:00
static irqreturn_t ni65_interrupt(int irq, void * dev_id)
{
int csr0 = 0;
struct net_device *dev = dev_id;
struct priv *p;
int bcnt = 32;
p = (struct priv *) dev->priv;
spin_lock(&p->ring_lock);
while(--bcnt) {
csr0 = inw(PORT+L_DATAREG);
#if 0
writedatareg( (csr0 & CSR0_CLRALL) ); /* ack interrupts, disable int. */
#else
writedatareg( (csr0 & CSR0_CLRALL) | CSR0_INEA ); /* ack interrupts, interrupts enabled */
#endif
if(!(csr0 & (CSR0_ERR | CSR0_RINT | CSR0_TINT)))
break;
if(csr0 & CSR0_RINT) /* RECV-int? */
ni65_recv_intr(dev,csr0);
if(csr0 & CSR0_TINT) /* XMIT-int? */
ni65_xmit_intr(dev,csr0);
if(csr0 & CSR0_ERR)
{
struct priv *p = (struct priv *) dev->priv;
if(debuglevel > 1)
printk(KERN_ERR "%s: general error: %04x.\n",dev->name,csr0);
if(csr0 & CSR0_BABL)
p->stats.tx_errors++;
if(csr0 & CSR0_MISS) {
int i;
for(i=0;i<RMDNUM;i++)
printk("%02x ",p->rmdhead[i].u.s.status);
printk("\n");
p->stats.rx_errors++;
}
if(csr0 & CSR0_MERR) {
if(debuglevel > 1)
printk(KERN_ERR "%s: Ooops .. memory error: %04x.\n",dev->name,csr0);
ni65_stop_start(dev,p);
}
}
}
#ifdef RCV_PARANOIA_CHECK
{
int j;
for(j=0;j<RMDNUM;j++)
{
struct priv *p = (struct priv *) dev->priv;
int i,k,num1,num2;
for(i=RMDNUM-1;i>0;i--) {
num2 = (p->rmdnum + i) & (RMDNUM-1);
if(!(p->rmdhead[num2].u.s.status & RCV_OWN))
break;
}
if(i) {
for(k=0;k<RMDNUM;k++) {
num1 = (p->rmdnum + k) & (RMDNUM-1);
if(!(p->rmdhead[num1].u.s.status & RCV_OWN))
break;
}
if(!k)
break;
if(debuglevel > 0)
{
char buf[256],*buf1;
int k;
buf1 = buf;
for(k=0;k<RMDNUM;k++) {
sprintf(buf1,"%02x ",(p->rmdhead[k].u.s.status)); /* & RCV_OWN) ); */
buf1 += 3;
}
*buf1 = 0;
printk(KERN_ERR "%s: Ooops, receive ring corrupted %2d %2d | %s\n",dev->name,p->rmdnum,i,buf);
}
p->rmdnum = num1;
ni65_recv_intr(dev,csr0);
if((p->rmdhead[num2].u.s.status & RCV_OWN))
break; /* ok, we are 'in sync' again */
}
else
break;
}
}
#endif
if( (csr0 & (CSR0_RXON | CSR0_TXON)) != (CSR0_RXON | CSR0_TXON) ) {
printk(KERN_DEBUG "%s: RX or TX was offline -> restart\n",dev->name);
ni65_stop_start(dev,p);
}
else
writedatareg(CSR0_INEA);
spin_unlock(&p->ring_lock);
return IRQ_HANDLED;
}
/*
* We have received an Xmit-Interrupt ..
* send a new packet if necessary
*/
static void ni65_xmit_intr(struct net_device *dev,int csr0)
{
struct priv *p = (struct priv *) dev->priv;
while(p->xmit_queued)
{
struct tmd *tmdp = p->tmdhead + p->tmdlast;
int tmdstat = tmdp->u.s.status;
if(tmdstat & XMIT_OWN)
break;
if(tmdstat & XMIT_ERR)
{
#if 0
if(tmdp->status2 & XMIT_TDRMASK && debuglevel > 3)
printk(KERN_ERR "%s: tdr-problems (e.g. no resistor)\n",dev->name);
#endif
/* checking some errors */
if(tmdp->status2 & XMIT_RTRY)
p->stats.tx_aborted_errors++;
if(tmdp->status2 & XMIT_LCAR)
p->stats.tx_carrier_errors++;
if(tmdp->status2 & (XMIT_BUFF | XMIT_UFLO )) {
/* this stops the xmitter */
p->stats.tx_fifo_errors++;
if(debuglevel > 0)
printk(KERN_ERR "%s: Xmit FIFO/BUFF error\n",dev->name);
if(p->features & INIT_RING_BEFORE_START) {
tmdp->u.s.status = XMIT_OWN | XMIT_START | XMIT_END; /* test: resend this frame */
ni65_stop_start(dev,p);
break; /* no more Xmit processing .. */
}
else
ni65_stop_start(dev,p);
}
if(debuglevel > 2)
printk(KERN_ERR "%s: xmit-error: %04x %02x-%04x\n",dev->name,csr0,(int) tmdstat,(int) tmdp->status2);
if(!(csr0 & CSR0_BABL)) /* don't count errors twice */
p->stats.tx_errors++;
tmdp->status2 = 0;
}
else {
p->stats.tx_bytes -= (short)(tmdp->blen);
p->stats.tx_packets++;
}
#ifdef XMT_VIA_SKB
if(p->tmd_skb[p->tmdlast]) {
dev_kfree_skb_irq(p->tmd_skb[p->tmdlast]);
p->tmd_skb[p->tmdlast] = NULL;
}
#endif
p->tmdlast = (p->tmdlast + 1) & (TMDNUM-1);
if(p->tmdlast == p->tmdnum)
p->xmit_queued = 0;
}
netif_wake_queue(dev);
}
/*
* We have received a packet
*/
static void ni65_recv_intr(struct net_device *dev,int csr0)
{
struct rmd *rmdp;
int rmdstat,len;
int cnt=0;
struct priv *p = (struct priv *) dev->priv;
rmdp = p->rmdhead + p->rmdnum;
while(!( (rmdstat = rmdp->u.s.status) & RCV_OWN))
{
cnt++;
if( (rmdstat & (RCV_START | RCV_END | RCV_ERR)) != (RCV_START | RCV_END) ) /* error or oversized? */
{
if(!(rmdstat & RCV_ERR)) {
if(rmdstat & RCV_START)
{
p->stats.rx_length_errors++;
printk(KERN_ERR "%s: recv, packet too long: %d\n",dev->name,rmdp->mlen & 0x0fff);
}
}
else {
if(debuglevel > 2)
printk(KERN_ERR "%s: receive-error: %04x, lance-status: %04x/%04x\n",
dev->name,(int) rmdstat,csr0,(int) inw(PORT+L_DATAREG) );
if(rmdstat & RCV_FRAM)
p->stats.rx_frame_errors++;
if(rmdstat & RCV_OFLO)
p->stats.rx_over_errors++;
if(rmdstat & RCV_CRC)
p->stats.rx_crc_errors++;
if(rmdstat & RCV_BUF_ERR)
p->stats.rx_fifo_errors++;
}
if(!(csr0 & CSR0_MISS)) /* don't count errors twice */
p->stats.rx_errors++;
}
else if( (len = (rmdp->mlen & 0x0fff) - 4) >= 60)
{
#ifdef RCV_VIA_SKB
struct sk_buff *skb = alloc_skb(R_BUF_SIZE+2+16,GFP_ATOMIC);
if (skb)
skb_reserve(skb,16);
#else
struct sk_buff *skb = dev_alloc_skb(len+2);
#endif
if(skb)
{
skb_reserve(skb,2);
#ifdef RCV_VIA_SKB
if( (unsigned long) (skb->data + R_BUF_SIZE) > 0x1000000) {
skb_put(skb,len);
skb_copy_to_linear_data(skb, (unsigned char *)(p->recv_skb[p->rmdnum]->data),len);
}
else {
struct sk_buff *skb1 = p->recv_skb[p->rmdnum];
skb_put(skb,R_BUF_SIZE);
p->recv_skb[p->rmdnum] = skb;
rmdp->u.buffer = (u32) isa_virt_to_bus(skb->data);
skb = skb1;
skb_trim(skb,len);
}
#else
skb_put(skb,len);
skb_copy_to_linear_data(skb, (unsigned char *) p->recvbounce[p->rmdnum],len);
#endif
p->stats.rx_packets++;
p->stats.rx_bytes += len;
skb->protocol=eth_type_trans(skb,dev);
netif_rx(skb);
dev->last_rx = jiffies;
}
else
{
printk(KERN_ERR "%s: can't alloc new sk_buff\n",dev->name);
p->stats.rx_dropped++;
}
}
else {
printk(KERN_INFO "%s: received runt packet\n",dev->name);
p->stats.rx_errors++;
}
rmdp->blen = -(R_BUF_SIZE-8);
rmdp->mlen = 0;
rmdp->u.s.status = RCV_OWN; /* change owner */
p->rmdnum = (p->rmdnum + 1) & (RMDNUM-1);
rmdp = p->rmdhead + p->rmdnum;
}
}
/*
* kick xmitter ..
*/
static void ni65_timeout(struct net_device *dev)
{
int i;
struct priv *p = (struct priv *) dev->priv;
printk(KERN_ERR "%s: xmitter timed out, try to restart!\n",dev->name);
for(i=0;i<TMDNUM;i++)
printk("%02x ",p->tmdhead[i].u.s.status);
printk("\n");
ni65_lance_reinit(dev);
dev->trans_start = jiffies;
netif_wake_queue(dev);
}
/*
* Send a packet
*/
static int ni65_send_packet(struct sk_buff *skb, struct net_device *dev)
{
struct priv *p = (struct priv *) dev->priv;
netif_stop_queue(dev);
if (test_and_set_bit(0, (void*)&p->lock)) {
printk(KERN_ERR "%s: Queue was locked.\n", dev->name);
return 1;
}
{
short len = ETH_ZLEN < skb->len ? skb->len : ETH_ZLEN;
struct tmd *tmdp;
unsigned long flags;
#ifdef XMT_VIA_SKB
if( (unsigned long) (skb->data + skb->len) > 0x1000000) {
#endif
skb_copy_from_linear_data(skb, p->tmdbounce[p->tmdbouncenum],
skb->len > T_BUF_SIZE ? T_BUF_SIZE :
skb->len);
if (len > skb->len)
memset((char *)p->tmdbounce[p->tmdbouncenum]+skb->len, 0, len-skb->len);
dev_kfree_skb (skb);
spin_lock_irqsave(&p->ring_lock, flags);
tmdp = p->tmdhead + p->tmdnum;
tmdp->u.buffer = (u32) isa_virt_to_bus(p->tmdbounce[p->tmdbouncenum]);
p->tmdbouncenum = (p->tmdbouncenum + 1) & (TMDNUM - 1);
#ifdef XMT_VIA_SKB
}
else {
spin_lock_irqsave(&p->ring_lock, flags);
tmdp = p->tmdhead + p->tmdnum;
tmdp->u.buffer = (u32) isa_virt_to_bus(skb->data);
p->tmd_skb[p->tmdnum] = skb;
}
#endif
tmdp->blen = -len;
tmdp->u.s.status = XMIT_OWN | XMIT_START | XMIT_END;
writedatareg(CSR0_TDMD | CSR0_INEA); /* enable xmit & interrupt */
p->xmit_queued = 1;
p->tmdnum = (p->tmdnum + 1) & (TMDNUM-1);
if(p->tmdnum != p->tmdlast)
netif_wake_queue(dev);
p->lock = 0;
dev->trans_start = jiffies;
spin_unlock_irqrestore(&p->ring_lock, flags);
}
return 0;
}
static struct net_device_stats *ni65_get_stats(struct net_device *dev)
{
#if 0
int i;
struct priv *p = (struct priv *) dev->priv;
for(i=0;i<RMDNUM;i++)
{
struct rmd *rmdp = p->rmdhead + ((p->rmdnum + i) & (RMDNUM-1));
printk("%02x ",rmdp->u.s.status);
}
printk("\n");
#endif
return &((struct priv *) dev->priv)->stats;
}
static void set_multicast_list(struct net_device *dev)
{
if(!ni65_lance_reinit(dev))
printk(KERN_ERR "%s: Can't switch card into MC mode!\n",dev->name);
netif_wake_queue(dev);
}
#ifdef MODULE
static struct net_device *dev_ni65;
module_param(irq, int, 0);
module_param(io, int, 0);
module_param(dma, int, 0);
MODULE_PARM_DESC(irq, "ni6510 IRQ number (ignored for some cards)");
MODULE_PARM_DESC(io, "ni6510 I/O base address");
MODULE_PARM_DESC(dma, "ni6510 ISA DMA channel (ignored for some cards)");
int __init init_module(void)
{
dev_ni65 = ni65_probe(-1);
return IS_ERR(dev_ni65) ? PTR_ERR(dev_ni65) : 0;
}
void __exit cleanup_module(void)
{
unregister_netdev(dev_ni65);
cleanup_card(dev_ni65);
free_netdev(dev_ni65);
}
#endif /* MODULE */
MODULE_LICENSE("GPL");
/*
* END of ni65.c
*/