linux/drivers/net/arm/ether1.c

1104 lines
27 KiB
C
Raw Normal View History

/*
* linux/drivers/acorn/net/ether1.c
*
* Copyright (C) 1996-2000 Russell King
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*
* Acorn ether1 driver (82586 chip) for Acorn machines
*
* We basically keep two queues in the cards memory - one for transmit
* and one for receive. Each has a head and a tail. The head is where
* we/the chip adds packets to be transmitted/received, and the tail
* is where the transmitter has got to/where the receiver will stop.
* Both of these queues are circular, and since the chip is running
* all the time, we have to be careful when we modify the pointers etc
* so that the buffer memory contents is valid all the time.
*
* Change log:
* 1.00 RMK Released
* 1.01 RMK 19/03/1996 Transfers the last odd byte onto/off of the card now.
* 1.02 RMK 25/05/1997 Added code to restart RU if it goes not ready
* 1.03 RMK 14/09/1997 Cleaned up the handling of a reset during the TX interrupt.
* Should prevent lockup.
* 1.04 RMK 17/09/1997 Added more info when initialsation of chip goes wrong.
* TDR now only reports failure when chip reports non-zero
* TDR time-distance.
* 1.05 RMK 31/12/1997 Removed calls to dev_tint for 2.1
* 1.06 RMK 10/02/2000 Updated for 2.3.43
* 1.07 RMK 13/05/2000 Updated for 2.3.99-pre8
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/fcntl.h>
#include <linux/interrupt.h>
#include <linux/ptrace.h>
#include <linux/ioport.h>
#include <linux/in.h>
#include <linux/slab.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/device.h>
#include <linux/init.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/bitops.h>
#include <asm/system.h>
#include <asm/io.h>
#include <asm/dma.h>
#include <asm/ecard.h>
#define __ETHER1_C
#include "ether1.h"
static unsigned int net_debug = NET_DEBUG;
#define BUFFER_SIZE 0x10000
#define TX_AREA_START 0x00100
#define TX_AREA_END 0x05000
#define RX_AREA_START 0x05000
#define RX_AREA_END 0x0fc00
static int ether1_open(struct net_device *dev);
static int ether1_sendpacket(struct sk_buff *skb, struct net_device *dev);
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 13:55:46 +00:00
static irqreturn_t ether1_interrupt(int irq, void *dev_id);
static int ether1_close(struct net_device *dev);
static struct net_device_stats *ether1_getstats(struct net_device *dev);
static void ether1_setmulticastlist(struct net_device *dev);
static void ether1_timeout(struct net_device *dev);
/* ------------------------------------------------------------------------- */
static char version[] __initdata = "ether1 ethernet driver (c) 2000 Russell King v1.07\n";
#define BUS_16 16
#define BUS_8 8
/* ------------------------------------------------------------------------- */
#define DISABLEIRQS 1
#define NORMALIRQS 0
#define ether1_readw(dev, addr, type, offset, svflgs) ether1_inw_p (dev, addr + (int)(&((type *)0)->offset), svflgs)
#define ether1_writew(dev, val, addr, type, offset, svflgs) ether1_outw_p (dev, val, addr + (int)(&((type *)0)->offset), svflgs)
static inline unsigned short
ether1_inw_p (struct net_device *dev, int addr, int svflgs)
{
unsigned long flags;
unsigned short ret;
if (svflgs)
local_irq_save (flags);
writeb(addr >> 12, REG_PAGE);
ret = readw(ETHER1_RAM + ((addr & 4095) << 1));
if (svflgs)
local_irq_restore (flags);
return ret;
}
static inline void
ether1_outw_p (struct net_device *dev, unsigned short val, int addr, int svflgs)
{
unsigned long flags;
if (svflgs)
local_irq_save (flags);
writeb(addr >> 12, REG_PAGE);
writew(val, ETHER1_RAM + ((addr & 4095) << 1));
if (svflgs)
local_irq_restore (flags);
}
/*
* Some inline assembler to allow fast transfers on to/off of the card.
* Since this driver depends on some features presented by the ARM
* specific architecture, and that you can't configure this driver
* without specifiing ARM mode, this is not a problem.
*
* This routine is essentially an optimised memcpy from the card's
* onboard RAM to kernel memory.
*/
static void
ether1_writebuffer (struct net_device *dev, void *data, unsigned int start, unsigned int length)
{
unsigned int page, thislen, offset;
void __iomem *addr;
offset = start & 4095;
page = start >> 12;
addr = ETHER1_RAM + (offset << 1);
if (offset + length > 4096)
thislen = 4096 - offset;
else
thislen = length;
do {
int used;
writeb(page, REG_PAGE);
length -= thislen;
__asm__ __volatile__(
"subs %3, %3, #2\n\
bmi 2f\n\
1: ldr %0, [%1], #2\n\
mov %0, %0, lsl #16\n\
orr %0, %0, %0, lsr #16\n\
str %0, [%2], #4\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%1], #2\n\
mov %0, %0, lsl #16\n\
orr %0, %0, %0, lsr #16\n\
str %0, [%2], #4\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%1], #2\n\
mov %0, %0, lsl #16\n\
orr %0, %0, %0, lsr #16\n\
str %0, [%2], #4\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%1], #2\n\
mov %0, %0, lsl #16\n\
orr %0, %0, %0, lsr #16\n\
str %0, [%2], #4\n\
subs %3, %3, #2\n\
bpl 1b\n\
2: adds %3, %3, #1\n\
ldreqb %0, [%1]\n\
streqb %0, [%2]"
: "=&r" (used), "=&r" (data)
: "r" (addr), "r" (thislen), "1" (data));
addr = ETHER1_RAM;
thislen = length;
if (thislen > 4096)
thislen = 4096;
page++;
} while (thislen);
}
static void
ether1_readbuffer (struct net_device *dev, void *data, unsigned int start, unsigned int length)
{
unsigned int page, thislen, offset;
void __iomem *addr;
offset = start & 4095;
page = start >> 12;
addr = ETHER1_RAM + (offset << 1);
if (offset + length > 4096)
thislen = 4096 - offset;
else
thislen = length;
do {
int used;
writeb(page, REG_PAGE);
length -= thislen;
__asm__ __volatile__(
"subs %3, %3, #2\n\
bmi 2f\n\
1: ldr %0, [%2], #4\n\
strb %0, [%1], #1\n\
mov %0, %0, lsr #8\n\
strb %0, [%1], #1\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%2], #4\n\
strb %0, [%1], #1\n\
mov %0, %0, lsr #8\n\
strb %0, [%1], #1\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%2], #4\n\
strb %0, [%1], #1\n\
mov %0, %0, lsr #8\n\
strb %0, [%1], #1\n\
subs %3, %3, #2\n\
bmi 2f\n\
ldr %0, [%2], #4\n\
strb %0, [%1], #1\n\
mov %0, %0, lsr #8\n\
strb %0, [%1], #1\n\
subs %3, %3, #2\n\
bpl 1b\n\
2: adds %3, %3, #1\n\
ldreqb %0, [%2]\n\
streqb %0, [%1]"
: "=&r" (used), "=&r" (data)
: "r" (addr), "r" (thislen), "1" (data));
addr = ETHER1_RAM;
thislen = length;
if (thislen > 4096)
thislen = 4096;
page++;
} while (thislen);
}
static int __devinit
ether1_ramtest(struct net_device *dev, unsigned char byte)
{
unsigned char *buffer = kmalloc (BUFFER_SIZE, GFP_KERNEL);
int i, ret = BUFFER_SIZE;
int max_errors = 15;
int bad = -1;
int bad_start = 0;
if (!buffer)
return 1;
memset (buffer, byte, BUFFER_SIZE);
ether1_writebuffer (dev, buffer, 0, BUFFER_SIZE);
memset (buffer, byte ^ 0xff, BUFFER_SIZE);
ether1_readbuffer (dev, buffer, 0, BUFFER_SIZE);
for (i = 0; i < BUFFER_SIZE; i++) {
if (buffer[i] != byte) {
if (max_errors >= 0 && bad != buffer[i]) {
if (bad != -1)
printk ("\n");
printk (KERN_CRIT "%s: RAM failed with (%02X instead of %02X) at 0x%04X",
dev->name, buffer[i], byte, i);
ret = -ENODEV;
max_errors --;
bad = buffer[i];
bad_start = i;
}
} else {
if (bad != -1) {
if (bad_start == i - 1)
printk ("\n");
else
printk (" - 0x%04X\n", i - 1);
bad = -1;
}
}
}
if (bad != -1)
printk (" - 0x%04X\n", BUFFER_SIZE);
kfree (buffer);
return ret;
}
static int
ether1_reset (struct net_device *dev)
{
writeb(CTRL_RST|CTRL_ACK, REG_CONTROL);
return BUS_16;
}
static int __devinit
ether1_init_2(struct net_device *dev)
{
int i;
dev->mem_start = 0;
i = ether1_ramtest (dev, 0x5a);
if (i > 0)
i = ether1_ramtest (dev, 0x1e);
if (i <= 0)
return -ENODEV;
dev->mem_end = i;
return 0;
}
/*
* These are the structures that are loaded into the ether RAM card to
* initialise the 82586
*/
/* at 0x0100 */
#define NOP_ADDR (TX_AREA_START)
#define NOP_SIZE (0x06)
static nop_t init_nop = {
0,
CMD_NOP,
NOP_ADDR
};
/* at 0x003a */
#define TDR_ADDR (0x003a)
#define TDR_SIZE (0x08)
static tdr_t init_tdr = {
0,
CMD_TDR | CMD_INTR,
NOP_ADDR,
0
};
/* at 0x002e */
#define MC_ADDR (0x002e)
#define MC_SIZE (0x0c)
static mc_t init_mc = {
0,
CMD_SETMULTICAST,
TDR_ADDR,
0,
{ { 0, } }
};
/* at 0x0022 */
#define SA_ADDR (0x0022)
#define SA_SIZE (0x0c)
static sa_t init_sa = {
0,
CMD_SETADDRESS,
MC_ADDR,
{ 0, }
};
/* at 0x0010 */
#define CFG_ADDR (0x0010)
#define CFG_SIZE (0x12)
static cfg_t init_cfg = {
0,
CMD_CONFIG,
SA_ADDR,
8,
8,
CFG8_SRDY,
CFG9_PREAMB8 | CFG9_ADDRLENBUF | CFG9_ADDRLEN(6),
0,
0x60,
0,
CFG13_RETRY(15) | CFG13_SLOTH(2),
0,
};
/* at 0x0000 */
#define SCB_ADDR (0x0000)
#define SCB_SIZE (0x10)
static scb_t init_scb = {
0,
SCB_CMDACKRNR | SCB_CMDACKCNA | SCB_CMDACKFR | SCB_CMDACKCX,
CFG_ADDR,
RX_AREA_START,
0,
0,
0,
0
};
/* at 0xffee */
#define ISCP_ADDR (0xffee)
#define ISCP_SIZE (0x08)
static iscp_t init_iscp = {
1,
SCB_ADDR,
0x0000,
0x0000
};
/* at 0xfff6 */
#define SCP_ADDR (0xfff6)
#define SCP_SIZE (0x0a)
static scp_t init_scp = {
SCP_SY_16BBUS,
{ 0, 0 },
ISCP_ADDR,
0
};
#define RFD_SIZE (0x16)
static rfd_t init_rfd = {
0,
0,
0,
0,
{ 0, },
{ 0, },
0
};
#define RBD_SIZE (0x0a)
static rbd_t init_rbd = {
0,
0,
0,
0,
ETH_FRAME_LEN + 8
};
#define TX_SIZE (0x08)
#define TBD_SIZE (0x08)
static int
ether1_init_for_open (struct net_device *dev)
{
int i, status, addr, next, next2;
int failures = 0;
unsigned long timeout;
writeb(CTRL_RST|CTRL_ACK, REG_CONTROL);
for (i = 0; i < 6; i++)
init_sa.sa_addr[i] = dev->dev_addr[i];
/* load data structures into ether1 RAM */
ether1_writebuffer (dev, &init_scp, SCP_ADDR, SCP_SIZE);
ether1_writebuffer (dev, &init_iscp, ISCP_ADDR, ISCP_SIZE);
ether1_writebuffer (dev, &init_scb, SCB_ADDR, SCB_SIZE);
ether1_writebuffer (dev, &init_cfg, CFG_ADDR, CFG_SIZE);
ether1_writebuffer (dev, &init_sa, SA_ADDR, SA_SIZE);
ether1_writebuffer (dev, &init_mc, MC_ADDR, MC_SIZE);
ether1_writebuffer (dev, &init_tdr, TDR_ADDR, TDR_SIZE);
ether1_writebuffer (dev, &init_nop, NOP_ADDR, NOP_SIZE);
if (ether1_readw(dev, CFG_ADDR, cfg_t, cfg_command, NORMALIRQS) != CMD_CONFIG) {
printk (KERN_ERR "%s: detected either RAM fault or compiler bug\n",
dev->name);
return 1;
}
/*
* setup circularly linked list of { rfd, rbd, buffer }, with
* all rfds circularly linked, rbds circularly linked.
* First rfd is linked to scp, first rbd is linked to first
* rfd. Last rbd has a suspend command.
*/
addr = RX_AREA_START;
do {
next = addr + RFD_SIZE + RBD_SIZE + ETH_FRAME_LEN + 10;
next2 = next + RFD_SIZE + RBD_SIZE + ETH_FRAME_LEN + 10;
if (next2 >= RX_AREA_END) {
next = RX_AREA_START;
init_rfd.rfd_command = RFD_CMDEL | RFD_CMDSUSPEND;
priv(dev)->rx_tail = addr;
} else
init_rfd.rfd_command = 0;
if (addr == RX_AREA_START)
init_rfd.rfd_rbdoffset = addr + RFD_SIZE;
else
init_rfd.rfd_rbdoffset = 0;
init_rfd.rfd_link = next;
init_rbd.rbd_link = next + RFD_SIZE;
init_rbd.rbd_bufl = addr + RFD_SIZE + RBD_SIZE;
ether1_writebuffer (dev, &init_rfd, addr, RFD_SIZE);
ether1_writebuffer (dev, &init_rbd, addr + RFD_SIZE, RBD_SIZE);
addr = next;
} while (next2 < RX_AREA_END);
priv(dev)->tx_link = NOP_ADDR;
priv(dev)->tx_head = NOP_ADDR + NOP_SIZE;
priv(dev)->tx_tail = TDR_ADDR;
priv(dev)->rx_head = RX_AREA_START;
/* release reset & give 586 a prod */
priv(dev)->resetting = 1;
priv(dev)->initialising = 1;
writeb(CTRL_RST, REG_CONTROL);
writeb(0, REG_CONTROL);
writeb(CTRL_CA, REG_CONTROL);
/* 586 should now unset iscp.busy */
timeout = jiffies + HZ/2;
while (ether1_readw(dev, ISCP_ADDR, iscp_t, iscp_busy, DISABLEIRQS) == 1) {
if (time_after(jiffies, timeout)) {
printk (KERN_WARNING "%s: can't initialise 82586: iscp is busy\n", dev->name);
return 1;
}
}
/* check status of commands that we issued */
timeout += HZ/10;
while (((status = ether1_readw(dev, CFG_ADDR, cfg_t, cfg_status, DISABLEIRQS))
& STAT_COMPLETE) == 0) {
if (time_after(jiffies, timeout))
break;
}
if ((status & (STAT_COMPLETE | STAT_OK)) != (STAT_COMPLETE | STAT_OK)) {
printk (KERN_WARNING "%s: can't initialise 82586: config status %04X\n", dev->name, status);
printk (KERN_DEBUG "%s: SCB=[STS=%04X CMD=%04X CBL=%04X RFA=%04X]\n", dev->name,
ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_command, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_rfa_offset, NORMALIRQS));
failures += 1;
}
timeout += HZ/10;
while (((status = ether1_readw(dev, SA_ADDR, sa_t, sa_status, DISABLEIRQS))
& STAT_COMPLETE) == 0) {
if (time_after(jiffies, timeout))
break;
}
if ((status & (STAT_COMPLETE | STAT_OK)) != (STAT_COMPLETE | STAT_OK)) {
printk (KERN_WARNING "%s: can't initialise 82586: set address status %04X\n", dev->name, status);
printk (KERN_DEBUG "%s: SCB=[STS=%04X CMD=%04X CBL=%04X RFA=%04X]\n", dev->name,
ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_command, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_rfa_offset, NORMALIRQS));
failures += 1;
}
timeout += HZ/10;
while (((status = ether1_readw(dev, MC_ADDR, mc_t, mc_status, DISABLEIRQS))
& STAT_COMPLETE) == 0) {
if (time_after(jiffies, timeout))
break;
}
if ((status & (STAT_COMPLETE | STAT_OK)) != (STAT_COMPLETE | STAT_OK)) {
printk (KERN_WARNING "%s: can't initialise 82586: set multicast status %04X\n", dev->name, status);
printk (KERN_DEBUG "%s: SCB=[STS=%04X CMD=%04X CBL=%04X RFA=%04X]\n", dev->name,
ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_command, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_rfa_offset, NORMALIRQS));
failures += 1;
}
timeout += HZ;
while (((status = ether1_readw(dev, TDR_ADDR, tdr_t, tdr_status, DISABLEIRQS))
& STAT_COMPLETE) == 0) {
if (time_after(jiffies, timeout))
break;
}
if ((status & (STAT_COMPLETE | STAT_OK)) != (STAT_COMPLETE | STAT_OK)) {
printk (KERN_WARNING "%s: can't tdr (ignored)\n", dev->name);
printk (KERN_DEBUG "%s: SCB=[STS=%04X CMD=%04X CBL=%04X RFA=%04X]\n", dev->name,
ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_command, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS),
ether1_readw(dev, SCB_ADDR, scb_t, scb_rfa_offset, NORMALIRQS));
} else {
status = ether1_readw(dev, TDR_ADDR, tdr_t, tdr_result, DISABLEIRQS);
if (status & TDR_XCVRPROB)
printk (KERN_WARNING "%s: i/f failed tdr: transceiver problem\n", dev->name);
else if ((status & (TDR_SHORT|TDR_OPEN)) && (status & TDR_TIME)) {
#ifdef FANCY
printk (KERN_WARNING "%s: i/f failed tdr: cable %s %d.%d us away\n", dev->name,
status & TDR_SHORT ? "short" : "open", (status & TDR_TIME) / 10,
(status & TDR_TIME) % 10);
#else
printk (KERN_WARNING "%s: i/f failed tdr: cable %s %d clks away\n", dev->name,
status & TDR_SHORT ? "short" : "open", (status & TDR_TIME));
#endif
}
}
if (failures)
ether1_reset (dev);
return failures ? 1 : 0;
}
/* ------------------------------------------------------------------------- */
static int
ether1_txalloc (struct net_device *dev, int size)
{
int start, tail;
size = (size + 1) & ~1;
tail = priv(dev)->tx_tail;
if (priv(dev)->tx_head + size > TX_AREA_END) {
if (tail > priv(dev)->tx_head)
return -1;
start = TX_AREA_START;
if (start + size > tail)
return -1;
priv(dev)->tx_head = start + size;
} else {
if (priv(dev)->tx_head < tail && (priv(dev)->tx_head + size) > tail)
return -1;
start = priv(dev)->tx_head;
priv(dev)->tx_head += size;
}
return start;
}
static int
ether1_open (struct net_device *dev)
{
if (!is_valid_ether_addr(dev->dev_addr)) {
printk(KERN_WARNING "%s: invalid ethernet MAC address\n",
dev->name);
return -EINVAL;
}
if (request_irq(dev->irq, ether1_interrupt, 0, "ether1", dev))
return -EAGAIN;
memset (&priv(dev)->stats, 0, sizeof (struct net_device_stats));
if (ether1_init_for_open (dev)) {
free_irq (dev->irq, dev);
return -EAGAIN;
}
netif_start_queue(dev);
return 0;
}
static void
ether1_timeout(struct net_device *dev)
{
printk(KERN_WARNING "%s: transmit timeout, network cable problem?\n",
dev->name);
printk(KERN_WARNING "%s: resetting device\n", dev->name);
ether1_reset (dev);
if (ether1_init_for_open (dev))
printk (KERN_ERR "%s: unable to restart interface\n", dev->name);
priv(dev)->stats.tx_errors++;
netif_wake_queue(dev);
}
static int
ether1_sendpacket (struct sk_buff *skb, struct net_device *dev)
{
int tmp, tst, nopaddr, txaddr, tbdaddr, dataddr;
unsigned long flags;
tx_t tx;
tbd_t tbd;
nop_t nop;
if (priv(dev)->restart) {
printk(KERN_WARNING "%s: resetting device\n", dev->name);
ether1_reset(dev);
if (ether1_init_for_open(dev))
printk(KERN_ERR "%s: unable to restart interface\n", dev->name);
else
priv(dev)->restart = 0;
}
if (skb->len < ETH_ZLEN) {
if (skb_padto(skb, ETH_ZLEN))
goto out;
}
/*
* insert packet followed by a nop
*/
txaddr = ether1_txalloc (dev, TX_SIZE);
tbdaddr = ether1_txalloc (dev, TBD_SIZE);
dataddr = ether1_txalloc (dev, skb->len);
nopaddr = ether1_txalloc (dev, NOP_SIZE);
tx.tx_status = 0;
tx.tx_command = CMD_TX | CMD_INTR;
tx.tx_link = nopaddr;
tx.tx_tbdoffset = tbdaddr;
tbd.tbd_opts = TBD_EOL | skb->len;
tbd.tbd_link = I82586_NULL;
tbd.tbd_bufl = dataddr;
tbd.tbd_bufh = 0;
nop.nop_status = 0;
nop.nop_command = CMD_NOP;
nop.nop_link = nopaddr;
local_irq_save(flags);
ether1_writebuffer (dev, &tx, txaddr, TX_SIZE);
ether1_writebuffer (dev, &tbd, tbdaddr, TBD_SIZE);
ether1_writebuffer (dev, skb->data, dataddr, skb->len);
ether1_writebuffer (dev, &nop, nopaddr, NOP_SIZE);
tmp = priv(dev)->tx_link;
priv(dev)->tx_link = nopaddr;
/* now reset the previous nop pointer */
ether1_writew(dev, txaddr, tmp, nop_t, nop_link, NORMALIRQS);
local_irq_restore(flags);
/* handle transmit */
dev->trans_start = jiffies;
/* check to see if we have room for a full sized ether frame */
tmp = priv(dev)->tx_head;
tst = ether1_txalloc (dev, TX_SIZE + TBD_SIZE + NOP_SIZE + ETH_FRAME_LEN);
priv(dev)->tx_head = tmp;
dev_kfree_skb (skb);
if (tst == -1)
netif_stop_queue(dev);
out:
return 0;
}
static void
ether1_xmit_done (struct net_device *dev)
{
nop_t nop;
int caddr, tst;
caddr = priv(dev)->tx_tail;
again:
ether1_readbuffer (dev, &nop, caddr, NOP_SIZE);
switch (nop.nop_command & CMD_MASK) {
case CMD_TDR:
/* special case */
if (ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS)
!= (unsigned short)I82586_NULL) {
ether1_writew(dev, SCB_CMDCUCSTART | SCB_CMDRXSTART, SCB_ADDR, scb_t,
scb_command, NORMALIRQS);
writeb(CTRL_CA, REG_CONTROL);
}
priv(dev)->tx_tail = NOP_ADDR;
return;
case CMD_NOP:
if (nop.nop_link == caddr) {
if (priv(dev)->initialising == 0)
printk (KERN_WARNING "%s: strange command complete with no tx command!\n", dev->name);
else
priv(dev)->initialising = 0;
return;
}
if (caddr == nop.nop_link)
return;
caddr = nop.nop_link;
goto again;
case CMD_TX:
if (nop.nop_status & STAT_COMPLETE)
break;
printk (KERN_ERR "%s: strange command complete without completed command\n", dev->name);
priv(dev)->restart = 1;
return;
default:
printk (KERN_WARNING "%s: strange command %d complete! (offset %04X)", dev->name,
nop.nop_command & CMD_MASK, caddr);
priv(dev)->restart = 1;
return;
}
while (nop.nop_status & STAT_COMPLETE) {
if (nop.nop_status & STAT_OK) {
priv(dev)->stats.tx_packets ++;
priv(dev)->stats.collisions += (nop.nop_status & STAT_COLLISIONS);
} else {
priv(dev)->stats.tx_errors ++;
if (nop.nop_status & STAT_COLLAFTERTX)
priv(dev)->stats.collisions ++;
if (nop.nop_status & STAT_NOCARRIER)
priv(dev)->stats.tx_carrier_errors ++;
if (nop.nop_status & STAT_TXLOSTCTS)
printk (KERN_WARNING "%s: cts lost\n", dev->name);
if (nop.nop_status & STAT_TXSLOWDMA)
priv(dev)->stats.tx_fifo_errors ++;
if (nop.nop_status & STAT_COLLEXCESSIVE)
priv(dev)->stats.collisions += 16;
}
if (nop.nop_link == caddr) {
printk (KERN_ERR "%s: tx buffer chaining error: tx command points to itself\n", dev->name);
break;
}
caddr = nop.nop_link;
ether1_readbuffer (dev, &nop, caddr, NOP_SIZE);
if ((nop.nop_command & CMD_MASK) != CMD_NOP) {
printk (KERN_ERR "%s: tx buffer chaining error: no nop after tx command\n", dev->name);
break;
}
if (caddr == nop.nop_link)
break;
caddr = nop.nop_link;
ether1_readbuffer (dev, &nop, caddr, NOP_SIZE);
if ((nop.nop_command & CMD_MASK) != CMD_TX) {
printk (KERN_ERR "%s: tx buffer chaining error: no tx command after nop\n", dev->name);
break;
}
}
priv(dev)->tx_tail = caddr;
caddr = priv(dev)->tx_head;
tst = ether1_txalloc (dev, TX_SIZE + TBD_SIZE + NOP_SIZE + ETH_FRAME_LEN);
priv(dev)->tx_head = caddr;
if (tst != -1)
netif_wake_queue(dev);
}
static void
ether1_recv_done (struct net_device *dev)
{
int status;
int nexttail, rbdaddr;
rbd_t rbd;
do {
status = ether1_readw(dev, priv(dev)->rx_head, rfd_t, rfd_status, NORMALIRQS);
if ((status & RFD_COMPLETE) == 0)
break;
rbdaddr = ether1_readw(dev, priv(dev)->rx_head, rfd_t, rfd_rbdoffset, NORMALIRQS);
ether1_readbuffer (dev, &rbd, rbdaddr, RBD_SIZE);
if ((rbd.rbd_status & (RBD_EOF | RBD_ACNTVALID)) == (RBD_EOF | RBD_ACNTVALID)) {
int length = rbd.rbd_status & RBD_ACNT;
struct sk_buff *skb;
length = (length + 1) & ~1;
skb = dev_alloc_skb (length + 2);
if (skb) {
skb_reserve (skb, 2);
ether1_readbuffer (dev, skb_put (skb, length), rbd.rbd_bufl, length);
skb->protocol = eth_type_trans (skb, dev);
netif_rx (skb);
priv(dev)->stats.rx_packets ++;
} else
priv(dev)->stats.rx_dropped ++;
} else {
printk(KERN_WARNING "%s: %s\n", dev->name,
(rbd.rbd_status & RBD_EOF) ? "oversized packet" : "acnt not valid");
priv(dev)->stats.rx_dropped ++;
}
nexttail = ether1_readw(dev, priv(dev)->rx_tail, rfd_t, rfd_link, NORMALIRQS);
/* nexttail should be rx_head */
if (nexttail != priv(dev)->rx_head)
printk(KERN_ERR "%s: receiver buffer chaining error (%04X != %04X)\n",
dev->name, nexttail, priv(dev)->rx_head);
ether1_writew(dev, RFD_CMDEL | RFD_CMDSUSPEND, nexttail, rfd_t, rfd_command, NORMALIRQS);
ether1_writew(dev, 0, priv(dev)->rx_tail, rfd_t, rfd_command, NORMALIRQS);
ether1_writew(dev, 0, priv(dev)->rx_tail, rfd_t, rfd_status, NORMALIRQS);
ether1_writew(dev, 0, priv(dev)->rx_tail, rfd_t, rfd_rbdoffset, NORMALIRQS);
priv(dev)->rx_tail = nexttail;
priv(dev)->rx_head = ether1_readw(dev, priv(dev)->rx_head, rfd_t, rfd_link, NORMALIRQS);
} while (1);
}
static irqreturn_t
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 13:55:46 +00:00
ether1_interrupt (int irq, void *dev_id)
{
struct net_device *dev = (struct net_device *)dev_id;
int status;
status = ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS);
if (status) {
ether1_writew(dev, status & (SCB_STRNR | SCB_STCNA | SCB_STFR | SCB_STCX),
SCB_ADDR, scb_t, scb_command, NORMALIRQS);
writeb(CTRL_CA | CTRL_ACK, REG_CONTROL);
if (status & SCB_STCX) {
ether1_xmit_done (dev);
}
if (status & SCB_STCNA) {
if (priv(dev)->resetting == 0)
printk (KERN_WARNING "%s: CU went not ready ???\n", dev->name);
else
priv(dev)->resetting += 1;
if (ether1_readw(dev, SCB_ADDR, scb_t, scb_cbl_offset, NORMALIRQS)
!= (unsigned short)I82586_NULL) {
ether1_writew(dev, SCB_CMDCUCSTART, SCB_ADDR, scb_t, scb_command, NORMALIRQS);
writeb(CTRL_CA, REG_CONTROL);
}
if (priv(dev)->resetting == 2)
priv(dev)->resetting = 0;
}
if (status & SCB_STFR) {
ether1_recv_done (dev);
}
if (status & SCB_STRNR) {
if (ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS) & SCB_STRXSUSP) {
printk (KERN_WARNING "%s: RU went not ready: RU suspended\n", dev->name);
ether1_writew(dev, SCB_CMDRXRESUME, SCB_ADDR, scb_t, scb_command, NORMALIRQS);
writeb(CTRL_CA, REG_CONTROL);
priv(dev)->stats.rx_dropped ++; /* we suspended due to lack of buffer space */
} else
printk(KERN_WARNING "%s: RU went not ready: %04X\n", dev->name,
ether1_readw(dev, SCB_ADDR, scb_t, scb_status, NORMALIRQS));
printk (KERN_WARNING "RU ptr = %04X\n", ether1_readw(dev, SCB_ADDR, scb_t, scb_rfa_offset,
NORMALIRQS));
}
} else
writeb(CTRL_ACK, REG_CONTROL);
return IRQ_HANDLED;
}
static int
ether1_close (struct net_device *dev)
{
ether1_reset (dev);
free_irq(dev->irq, dev);
return 0;
}
static struct net_device_stats *
ether1_getstats (struct net_device *dev)
{
return &priv(dev)->stats;
}
/*
* Set or clear the multicast filter for this adaptor.
* num_addrs == -1 Promiscuous mode, receive all packets.
* num_addrs == 0 Normal mode, clear multicast list.
* num_addrs > 0 Multicast mode, receive normal and MC packets, and do
* best-effort filtering.
*/
static void
ether1_setmulticastlist (struct net_device *dev)
{
}
/* ------------------------------------------------------------------------- */
static void __devinit ether1_banner(void)
{
static unsigned int version_printed = 0;
if (net_debug && version_printed++ == 0)
printk(KERN_INFO "%s", version);
}
static int __devinit
ether1_probe(struct expansion_card *ec, const struct ecard_id *id)
{
struct net_device *dev;
int i, ret = 0;
ether1_banner();
ret = ecard_request_resources(ec);
if (ret)
goto out;
dev = alloc_etherdev(sizeof(struct ether1_priv));
if (!dev) {
ret = -ENOMEM;
goto release;
}
SET_MODULE_OWNER(dev);
SET_NETDEV_DEV(dev, &ec->dev);
dev->irq = ec->irq;
priv(dev)->base = ecardm_iomap(ec, ECARD_RES_IOCFAST, 0, 0);
if (!priv(dev)->base) {
ret = -ENOMEM;
goto free;
}
if ((priv(dev)->bus_type = ether1_reset(dev)) == 0) {
ret = -ENODEV;
goto free;
}
for (i = 0; i < 6; i++)
dev->dev_addr[i] = readb(IDPROM_ADDRESS + (i << 2));
if (ether1_init_2(dev)) {
ret = -ENODEV;
goto free;
}
dev->open = ether1_open;
dev->stop = ether1_close;
dev->hard_start_xmit = ether1_sendpacket;
dev->get_stats = ether1_getstats;
dev->set_multicast_list = ether1_setmulticastlist;
dev->tx_timeout = ether1_timeout;
dev->watchdog_timeo = 5 * HZ / 100;
ret = register_netdev(dev);
if (ret)
goto free;
printk(KERN_INFO "%s: ether1 in slot %d, ",
dev->name, ec->slot_no);
for (i = 0; i < 6; i++)
printk ("%2.2x%c", dev->dev_addr[i], i == 5 ? '\n' : ':');
ecard_set_drvdata(ec, dev);
return 0;
free:
free_netdev(dev);
release:
ecard_release_resources(ec);
out:
return ret;
}
static void __devexit ether1_remove(struct expansion_card *ec)
{
struct net_device *dev = ecard_get_drvdata(ec);
ecard_set_drvdata(ec, NULL);
unregister_netdev(dev);
free_netdev(dev);
ecard_release_resources(ec);
}
static const struct ecard_id ether1_ids[] = {
{ MANU_ACORN, PROD_ACORN_ETHER1 },
{ 0xffff, 0xffff }
};
static struct ecard_driver ether1_driver = {
.probe = ether1_probe,
.remove = __devexit_p(ether1_remove),
.id_table = ether1_ids,
.drv = {
.name = "ether1",
},
};
static int __init ether1_init(void)
{
return ecard_register_driver(&ether1_driver);
}
static void __exit ether1_exit(void)
{
ecard_remove_driver(&ether1_driver);
}
module_init(ether1_init);
module_exit(ether1_exit);
MODULE_LICENSE("GPL");