1
linux/drivers/net/dgrs.c

1616 lines
38 KiB
C
Raw Normal View History

/*
* Digi RightSwitch SE-X loadable device driver for Linux
*
* The RightSwitch is a 4 (EISA) or 6 (PCI) port etherswitch and
* a NIC on an internal board.
*
* Author: Rick Richardson, rick@remotepoint.com
* Derived from the SVR4.2 (UnixWare) driver for the same card.
*
* Copyright 1995-1996 Digi International Inc.
*
* This software may be used and distributed according to the terms
* of the GNU General Public License, incorporated herein by reference.
*
* For information on purchasing a RightSwitch SE-4 or SE-6
* board, please contact Digi's sales department at 1-612-912-3444
* or 1-800-DIGIBRD. Outside the U.S., please check our Web page
* at http://www.dgii.com for sales offices worldwide.
*
* OPERATION:
* When compiled as a loadable module, this driver can operate
* the board as either a 4/6 port switch with a 5th or 7th port
* that is a conventional NIC interface as far as the host is
* concerned, OR as 4/6 independent NICs. To select multi-NIC
* mode, add "nicmode=1" on the insmod load line for the driver.
*
* This driver uses the "dev" common ethernet device structure
* and a private "priv" (dev->priv) structure that contains
* mostly DGRS-specific information and statistics. To keep
* the code for both the switch mode and the multi-NIC mode
* as similar as possible, I have introduced the concept of
* "dev0"/"priv0" and "devN"/"privN" pointer pairs in subroutines
* where needed. The first pair of pointers points to the
* "dev" and "priv" structures of the zeroth (0th) device
* interface associated with a board. The second pair of
* pointers points to the current (Nth) device interface
* for the board: the one for which we are processing data.
*
* In switch mode, the pairs of pointers are always the same,
* that is, dev0 == devN and priv0 == privN. This is just
* like previous releases of this driver which did not support
* NIC mode.
*
* In multi-NIC mode, the pairs of pointers may be different.
* We use the devN and privN pointers to reference just the
* name, port number, and statistics for the current interface.
* We use the dev0 and priv0 pointers to access the variables
* that control access to the board, such as board address
* and simulated 82596 variables. This is because there is
* only one "fake" 82596 that serves as the interface to
* the board. We do not want to try to keep the variables
* associated with this 82596 in sync across all devices.
*
* This scheme works well. As you will see, except for
* initialization, there is very little difference between
* the two modes as far as this driver is concerned. On the
* receive side in NIC mode, the interrupt *always* comes in on
* the 0th interface (dev0/priv0). We then figure out which
* real 82596 port it came in on from looking at the "chan"
* member that the board firmware adds at the end of each
* RBD (a.k.a. TBD). We get the channel number like this:
* int chan = ((I596_RBD *) S2H(cbp->xmit.tbdp))->chan;
*
* On the transmit side in multi-NIC mode, we specify the
* output 82596 port by setting the new "dstchan" structure
* member that is at the end of the RFD, like this:
* priv0->rfdp->dstchan = privN->chan;
*
* TODO:
* - Multi-NIC mode is not yet supported when the driver is linked
* into the kernel.
* - Better handling of multicast addresses.
*
* Fixes:
* Arnaldo Carvalho de Melo <acme@conectiva.com.br> - 11/01/2001
* - fix dgrs_found_device wrt checking kmalloc return and
* rollbacking the partial steps of the whole process when
* one of the devices can't be allocated. Fix SET_MODULE_OWNER
* on the loop to use devN instead of repeated calls to dev.
*
* davej <davej@suse.de> - 9/2/2001
* - Enable PCI device before reading ioaddr/irq
*
*/
#include <linux/module.h>
#include <linux/eisa.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/delay.h>
#include <linux/errno.h>
#include <linux/ioport.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/pci.h>
#include <linux/init.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/bitops.h>
#include <asm/io.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>
static char version[] __initdata =
"$Id: dgrs.c,v 1.13 2000/06/06 04:07:00 rick Exp $";
/*
* DGRS include files
*/
typedef unsigned char uchar;
#define vol volatile
#include "dgrs.h"
#include "dgrs_es4h.h"
#include "dgrs_plx9060.h"
#include "dgrs_i82596.h"
#include "dgrs_ether.h"
#include "dgrs_asstruct.h"
#include "dgrs_bcomm.h"
#ifdef CONFIG_PCI
static struct pci_device_id dgrs_pci_tbl[] = {
{ SE6_PCI_VENDOR_ID, SE6_PCI_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, },
{ } /* Terminating entry */
};
MODULE_DEVICE_TABLE(pci, dgrs_pci_tbl);
#endif
#ifdef CONFIG_EISA
static struct eisa_device_id dgrs_eisa_tbl[] = {
{ "DBI0A01" },
{ }
};
MODULE_DEVICE_TABLE(eisa, dgrs_eisa_tbl);
#endif
MODULE_LICENSE("GPL");
/*
* Firmware. Compiled separately for local compilation,
* but #included for Linux distribution.
*/
#ifndef NOFW
#include "dgrs_firmware.c"
#else
extern int dgrs_firmnum;
extern char dgrs_firmver[];
extern char dgrs_firmdate[];
extern uchar dgrs_code[];
extern int dgrs_ncode;
#endif
/*
* Linux out*() is backwards from all other operating systems
*/
#define OUTB(ADDR, VAL) outb(VAL, ADDR)
#define OUTW(ADDR, VAL) outw(VAL, ADDR)
#define OUTL(ADDR, VAL) outl(VAL, ADDR)
/*
* Macros to convert switch to host and host to switch addresses
* (assumes a local variable priv points to board dependent struct)
*/
#define S2H(A) ( ((unsigned long)(A)&0x00ffffff) + priv0->vmem )
#define S2HN(A) ( ((unsigned long)(A)&0x00ffffff) + privN->vmem )
#define H2S(A) ( ((char *) (A) - priv0->vmem) + 0xA3000000 )
/*
* Convert a switch address to a "safe" address for use with the
* PLX 9060 DMA registers and the associated HW kludge that allows
* for host access of the DMA registers.
*/
#define S2DMA(A) ( (unsigned long)(A) & 0x00ffffff)
/*
* "Space.c" variables, now settable from module interface
* Use the name below, minus the "dgrs_" prefix. See init_module().
*/
static int dgrs_debug = 1;
static int dgrs_dma = 1;
static int dgrs_spantree = -1;
static int dgrs_hashexpire = -1;
static uchar dgrs_ipaddr[4] = { 0xff, 0xff, 0xff, 0xff};
static uchar dgrs_iptrap[4] = { 0xff, 0xff, 0xff, 0xff};
static __u32 dgrs_ipxnet = -1;
static int dgrs_nicmode;
/*
* Private per-board data structure (dev->priv)
*/
typedef struct
{
/*
* Stuff for generic ethercard I/F
*/
struct net_device_stats stats;
/*
* DGRS specific data
*/
char *vmem;
struct bios_comm *bcomm; /* Firmware BIOS comm structure */
PORT *port; /* Ptr to PORT[0] struct in VM */
I596_SCB *scbp; /* Ptr to SCB struct in VM */
I596_RFD *rfdp; /* Current RFD list */
I596_RBD *rbdp; /* Current RBD list */
volatile int intrcnt; /* Count of interrupts */
/*
* SE-4 (EISA) board variables
*/
uchar is_reg; /* EISA: Value for ES4H_IS reg */
/*
* SE-6 (PCI) board variables
*
* The PLX "expansion rom" space is used for DMA register
* access from the host on the SE-6. These are the physical
* and virtual addresses of that space.
*/
ulong plxreg; /* Phys address of PLX chip */
char *vplxreg; /* Virtual address of PLX chip */
ulong plxdma; /* Phys addr of PLX "expansion rom" */
ulong volatile *vplxdma; /* Virtual addr of "expansion rom" */
int use_dma; /* Flag: use DMA */
DMACHAIN *dmadesc_s; /* area for DMA chains (SW addr.) */
DMACHAIN *dmadesc_h; /* area for DMA chains (Host Virtual) */
/*
* Multi-NIC mode variables
*
* All entries of the devtbl[] array are valid for the 0th
* device (i.e. eth0, but not eth1...eth5). devtbl[0] is
* valid for all devices (i.e. eth0, eth1, ..., eth5).
*/
int nports; /* Number of physical ports (4 or 6) */
int chan; /* Channel # (1-6) for this device */
struct net_device *devtbl[6]; /* Ptrs to N device structs */
} DGRS_PRIV;
/*
* reset or un-reset the IDT processor
*/
static void
proc_reset(struct net_device *dev0, int reset)
{
DGRS_PRIV *priv0 = (DGRS_PRIV *) dev0->priv;
if (priv0->plxreg)
{
ulong val;
val = inl(dev0->base_addr + PLX_MISC_CSR);
if (reset)
val |= SE6_RESET;
else
val &= ~SE6_RESET;
OUTL(dev0->base_addr + PLX_MISC_CSR, val);
}
else
{
OUTB(dev0->base_addr + ES4H_PC, reset ? ES4H_PC_RESET : 0);
}
}
/*
* See if the board supports bus master DMA
*/
static int
check_board_dma(struct net_device *dev0)
{
DGRS_PRIV *priv0 = (DGRS_PRIV *) dev0->priv;
ulong x;
/*
* If Space.c says not to use DMA, or if it's not a PLX based
* PCI board, or if the expansion ROM space is not PCI
* configured, then return false.
*/
if (!dgrs_dma || !priv0->plxreg || !priv0->plxdma)
return (0);
/*
* Set the local address remap register of the "expansion rom"
* area to 0x80000000 so that we can use it to access the DMA
* registers from the host side.
*/
OUTL(dev0->base_addr + PLX_ROM_BASE_ADDR, 0x80000000);
/*
* Set the PCI region descriptor to:
* Space 0:
* disable read-prefetch
* enable READY
* enable BURST
* 0 internal wait states
* Expansion ROM: (used for host DMA register access)
* disable read-prefetch
* enable READY
* disable BURST
* 0 internal wait states
*/
OUTL(dev0->base_addr + PLX_BUS_REGION, 0x49430343);
/*
* Now map the DMA registers into our virtual space
*/
priv0->vplxdma = (ulong *) ioremap (priv0->plxdma, 256);
if (!priv0->vplxdma)
{
printk("%s: can't *remap() the DMA regs\n", dev0->name);
return (0);
}
/*
* Now test to see if we can access the DMA registers
* If we write -1 and get back 1FFF, then we accessed the
* DMA register. Otherwise, we probably have an old board
* and wrote into regular RAM.
*/
priv0->vplxdma[PLX_DMA0_MODE/4] = 0xFFFFFFFF;
x = priv0->vplxdma[PLX_DMA0_MODE/4];
if (x != 0x00001FFF) {
iounmap((void *)priv0->vplxdma);
return (0);
}
return (1);
}
/*
* Initiate DMA using PLX part on PCI board. Spin the
* processor until completed. All addresses are physical!
*
* If pciaddr is NULL, then it's a chaining DMA, and lcladdr is
* the address of the first DMA descriptor in the chain.
*
* If pciaddr is not NULL, then it's a single DMA.
*
* In either case, "lcladdr" must have been fixed up to make
* sure the MSB isn't set using the S2DMA macro before passing
* the address to this routine.
*/
static int
do_plx_dma(
struct net_device *dev,
ulong pciaddr,
ulong lcladdr,
int len,
int to_host
)
{
int i;
ulong csr = 0;
DGRS_PRIV *priv = (DGRS_PRIV *) dev->priv;
if (pciaddr)
{
/*
* Do a single, non-chain DMA
*/
priv->vplxdma[PLX_DMA0_PCI_ADDR/4] = pciaddr;
priv->vplxdma[PLX_DMA0_LCL_ADDR/4] = lcladdr;
priv->vplxdma[PLX_DMA0_SIZE/4] = len;
priv->vplxdma[PLX_DMA0_DESCRIPTOR/4] = to_host
? PLX_DMA_DESC_TO_HOST
: PLX_DMA_DESC_TO_BOARD;
priv->vplxdma[PLX_DMA0_MODE/4] =
PLX_DMA_MODE_WIDTH32
| PLX_DMA_MODE_WAITSTATES(0)
| PLX_DMA_MODE_READY
| PLX_DMA_MODE_NOBTERM
| PLX_DMA_MODE_BURST
| PLX_DMA_MODE_NOCHAIN;
}
else
{
/*
* Do a chaining DMA
*/
priv->vplxdma[PLX_DMA0_MODE/4] =
PLX_DMA_MODE_WIDTH32
| PLX_DMA_MODE_WAITSTATES(0)
| PLX_DMA_MODE_READY
| PLX_DMA_MODE_NOBTERM
| PLX_DMA_MODE_BURST
| PLX_DMA_MODE_CHAIN;
priv->vplxdma[PLX_DMA0_DESCRIPTOR/4] = lcladdr;
}
priv->vplxdma[PLX_DMA_CSR/4] =
PLX_DMA_CSR_0_ENABLE | PLX_DMA_CSR_0_START;
/*
* Wait for DMA to complete
*/
for (i = 0; i < 1000000; ++i)
{
/*
* Spin the host CPU for 1 usec, so we don't thrash
* the PCI bus while the PLX 9060 is doing DMA.
*/
udelay(1);
csr = (volatile unsigned long) priv->vplxdma[PLX_DMA_CSR/4];
if (csr & PLX_DMA_CSR_0_DONE)
break;
}
if ( ! (csr & PLX_DMA_CSR_0_DONE) )
{
printk("%s: DMA done never occurred. DMA disabled.\n",
dev->name);
priv->use_dma = 0;
return 1;
}
return 0;
}
/*
* dgrs_rcv_frame()
*
* Process a received frame. This is called from the interrupt
* routine, and works for both switch mode and multi-NIC mode.
*
* Note that when in multi-NIC mode, we want to always access the
* hardware using the dev and priv structures of the first port,
* so that we are using only one set of variables to maintain
* the board interface status, but we want to use the Nth port
* dev and priv structures to maintain statistics and to pass
* the packet up.
*
* Only the first device structure is attached to the interrupt.
* We use the special "chan" variable at the end of the first RBD
* to select the Nth device in multi-NIC mode.
*
* We currently do chained DMA on a per-packet basis when the
* packet is "long", and we spin the CPU a short time polling
* for DMA completion. This avoids a second interrupt overhead,
* and gives the best performance for light traffic to the host.
*
* However, a better scheme that could be implemented would be
* to see how many packets are outstanding for the host, and if
* the number is "large", create a long chain to DMA several
* packets into the host in one go. In this case, we would set
* up some state variables to let the host CPU continue doing
* other things until a DMA completion interrupt comes along.
*/
static void
dgrs_rcv_frame(
struct net_device *dev0,
DGRS_PRIV *priv0,
I596_CB *cbp
)
{
int len;
I596_TBD *tbdp;
struct sk_buff *skb;
uchar *putp;
uchar *p;
struct net_device *devN;
DGRS_PRIV *privN;
/*
* Determine Nth priv and dev structure pointers
*/
if (dgrs_nicmode)
{ /* Multi-NIC mode */
int chan = ((I596_RBD *) S2H(cbp->xmit.tbdp))->chan;
devN = priv0->devtbl[chan-1];
/*
* If devN is null, we got an interrupt before the I/F
* has been initialized. Pitch the packet.
*/
if (devN == NULL)
goto out;
privN = (DGRS_PRIV *) devN->priv;
}
else
{ /* Switch mode */
devN = dev0;
privN = priv0;
}
if (0) printk("%s: rcv len=%ld\n", devN->name, cbp->xmit.count);
/*
* Allocate a message block big enough to hold the whole frame
*/
len = cbp->xmit.count;
if ((skb = dev_alloc_skb(len+5)) == NULL)
{
printk("%s: dev_alloc_skb failed for rcv buffer\n", devN->name);
++privN->stats.rx_dropped;
/* discarding the frame */
goto out;
}
skb_reserve(skb, 2); /* Align IP header */
again:
putp = p = skb_put(skb, len);
/*
* There are three modes here for doing the packet copy.
* If we have DMA, and the packet is "long", we use the
* chaining mode of DMA. If it's shorter, we use single
* DMA's. Otherwise, we use memcpy().
*/
if (priv0->use_dma && priv0->dmadesc_h && len > 64)
{
/*
* If we can use DMA and it's a long frame, copy it using
* DMA chaining.
*/
DMACHAIN *ddp_h; /* Host virtual DMA desc. pointer */
DMACHAIN *ddp_s; /* Switch physical DMA desc. pointer */
uchar *phys_p;
/*
* Get the physical address of the STREAMS buffer.
* NOTE: allocb() guarantees that the whole buffer
* is in a single page if the length < 4096.
*/
phys_p = (uchar *) virt_to_phys(putp);
ddp_h = priv0->dmadesc_h;
ddp_s = priv0->dmadesc_s;
tbdp = (I596_TBD *) S2H(cbp->xmit.tbdp);
for (;;)
{
int count;
int amt;
count = tbdp->count;
amt = count & 0x3fff;
if (amt == 0)
break; /* For safety */
if ( (p-putp) >= len)
{
printk("%s: cbp = %lx\n", devN->name, (long) H2S(cbp));
proc_reset(dev0, 1); /* Freeze IDT */
break; /* For Safety */
}
ddp_h->pciaddr = (ulong) phys_p;
ddp_h->lcladdr = S2DMA(tbdp->buf);
ddp_h->len = amt;
phys_p += amt;
p += amt;
if (count & I596_TBD_EOF)
{
ddp_h->next = PLX_DMA_DESC_TO_HOST
| PLX_DMA_DESC_EOC;
++ddp_h;
break;
}
else
{
++ddp_s;
ddp_h->next = PLX_DMA_DESC_TO_HOST
| (ulong) ddp_s;
tbdp = (I596_TBD *) S2H(tbdp->next);
++ddp_h;
}
}
if (ddp_h - priv0->dmadesc_h)
{
int rc;
rc = do_plx_dma(dev0,
0, (ulong) priv0->dmadesc_s, len, 0);
if (rc)
{
printk("%s: Chained DMA failure\n", devN->name);
goto again;
}
}
}
else if (priv0->use_dma)
{
/*
* If we can use DMA and it's a shorter frame, copy it
* using single DMA transfers.
*/
uchar *phys_p;
/*
* Get the physical address of the STREAMS buffer.
* NOTE: allocb() guarantees that the whole buffer
* is in a single page if the length < 4096.
*/
phys_p = (uchar *) virt_to_phys(putp);
tbdp = (I596_TBD *) S2H(cbp->xmit.tbdp);
for (;;)
{
int count;
int amt;
int rc;
count = tbdp->count;
amt = count & 0x3fff;
if (amt == 0)
break; /* For safety */
if ( (p-putp) >= len)
{
printk("%s: cbp = %lx\n", devN->name, (long) H2S(cbp));
proc_reset(dev0, 1); /* Freeze IDT */
break; /* For Safety */
}
rc = do_plx_dma(dev0, (ulong) phys_p,
S2DMA(tbdp->buf), amt, 1);
if (rc)
{
memcpy(p, S2H(tbdp->buf), amt);
printk("%s: Single DMA failed\n", devN->name);
}
phys_p += amt;
p += amt;
if (count & I596_TBD_EOF)
break;
tbdp = (I596_TBD *) S2H(tbdp->next);
}
}
else
{
/*
* Otherwise, copy it piece by piece using memcpy()
*/
tbdp = (I596_TBD *) S2H(cbp->xmit.tbdp);
for (;;)
{
int count;
int amt;
count = tbdp->count;
amt = count & 0x3fff;
if (amt == 0)
break; /* For safety */
if ( (p-putp) >= len)
{
printk("%s: cbp = %lx\n", devN->name, (long) H2S(cbp));
proc_reset(dev0, 1); /* Freeze IDT */
break; /* For Safety */
}
memcpy(p, S2H(tbdp->buf), amt);
p += amt;
if (count & I596_TBD_EOF)
break;
tbdp = (I596_TBD *) S2H(tbdp->next);
}
}
/*
* Pass the frame to upper half
*/
skb->protocol = eth_type_trans(skb, devN);
netif_rx(skb);
devN->last_rx = jiffies;
++privN->stats.rx_packets;
privN->stats.rx_bytes += len;
out:
cbp->xmit.status = I596_CB_STATUS_C | I596_CB_STATUS_OK;
}
/*
* Start transmission of a frame
*
* The interface to the board is simple: we pretend that we are
* a fifth 82596 ethernet controller 'receiving' data, and copy the
* data into the same structures that a real 82596 would. This way,
* the board firmware handles the host 'port' the same as any other.
*
* NOTE: we do not use Bus master DMA for this routine. Turns out
* that it is not needed. Slave writes over the PCI bus are about
* as fast as DMA, due to the fact that the PLX part can do burst
* writes. The same is not true for data being read from the board.
*
* For multi-NIC mode, we tell the firmware the desired 82596
* output port by setting the special "dstchan" member at the
* end of the traditional 82596 RFD structure.
*/
static int dgrs_start_xmit(struct sk_buff *skb, struct net_device *devN)
{
DGRS_PRIV *privN = (DGRS_PRIV *) devN->priv;
struct net_device *dev0;
DGRS_PRIV *priv0;
I596_RBD *rbdp;
int count;
int i, len, amt;
/*
* Determine 0th priv and dev structure pointers
*/
if (dgrs_nicmode)
{
dev0 = privN->devtbl[0];
priv0 = (DGRS_PRIV *) dev0->priv;
}
else
{
dev0 = devN;
priv0 = privN;
}
if (dgrs_debug > 1)
printk("%s: xmit len=%d\n", devN->name, (int) skb->len);
devN->trans_start = jiffies;
netif_start_queue(devN);
if (priv0->rfdp->cmd & I596_RFD_EL)
{ /* Out of RFD's */
if (0) printk("%s: NO RFD's\n", devN->name);
goto no_resources;
}
rbdp = priv0->rbdp;
count = 0;
priv0->rfdp->rbdp = (I596_RBD *) H2S(rbdp);
i = 0; len = skb->len;
for (;;)
{
if (rbdp->size & I596_RBD_EL)
{ /* Out of RBD's */
if (0) printk("%s: NO RBD's\n", devN->name);
goto no_resources;
}
amt = min_t(unsigned int, len, rbdp->size - count);
skb_copy_from_linear_data_offset(skb, i, S2H(rbdp->buf) + count, amt);
i += amt;
count += amt;
len -= amt;
if (len == 0)
{
if (skb->len < 60)
rbdp->count = 60 | I596_RBD_EOF;
else
rbdp->count = count | I596_RBD_EOF;
rbdp = (I596_RBD *) S2H(rbdp->next);
goto frame_done;
}
else if (count < 32)
{
/* More data to come, but we used less than 32
* bytes of this RBD. Keep filling this RBD.
*/
{} /* Yes, we do nothing here */
}
else
{
rbdp->count = count;
rbdp = (I596_RBD *) S2H(rbdp->next);
count = 0;
}
}
frame_done:
priv0->rbdp = rbdp;
if (dgrs_nicmode)
priv0->rfdp->dstchan = privN->chan;
priv0->rfdp->status = I596_RFD_C | I596_RFD_OK;
priv0->rfdp = (I596_RFD *) S2H(priv0->rfdp->next);
++privN->stats.tx_packets;
dev_kfree_skb (skb);
return (0);
no_resources:
priv0->scbp->status |= I596_SCB_RNR; /* simulate I82596 */
return (-EAGAIN);
}
/*
* Open the interface
*/
static int
dgrs_open( struct net_device *dev )
{
netif_start_queue(dev);
return (0);
}
/*
* Close the interface
*/
static int dgrs_close( struct net_device *dev )
{
netif_stop_queue(dev);
return (0);
}
/*
* Get statistics
*/
static struct net_device_stats *dgrs_get_stats( struct net_device *dev )
{
DGRS_PRIV *priv = (DGRS_PRIV *) dev->priv;
return (&priv->stats);
}
/*
* Set multicast list and/or promiscuous mode
*/
static void dgrs_set_multicast_list( struct net_device *dev)
{
DGRS_PRIV *priv = (DGRS_PRIV *) dev->priv;
priv->port->is_promisc = (dev->flags & IFF_PROMISC) ? 1 : 0;
}
/*
* Unique ioctl's
*/
static int dgrs_ioctl(struct net_device *devN, struct ifreq *ifr, int cmd)
{
DGRS_PRIV *privN = (DGRS_PRIV *) devN->priv;
DGRS_IOCTL ioc;
int i;
if (cmd != DGRSIOCTL)
return -EINVAL;
if(copy_from_user(&ioc, ifr->ifr_data, sizeof(DGRS_IOCTL)))
return -EFAULT;
switch (ioc.cmd)
{
case DGRS_GETMEM:
if (ioc.len != sizeof(ulong))
return -EINVAL;
if(copy_to_user(ioc.data, &devN->mem_start, ioc.len))
return -EFAULT;
return (0);
case DGRS_SETFILTER:
if (!capable(CAP_NET_ADMIN))
return -EPERM;
if (ioc.port > privN->bcomm->bc_nports)
return -EINVAL;
if (ioc.filter >= NFILTERS)
return -EINVAL;
if (ioc.len > privN->bcomm->bc_filter_area_len)
return -EINVAL;
/* Wait for old command to finish */
for (i = 0; i < 1000; ++i)
{
if ( (volatile long) privN->bcomm->bc_filter_cmd <= 0 )
break;
udelay(1);
}
if (i >= 1000)
return -EIO;
privN->bcomm->bc_filter_port = ioc.port;
privN->bcomm->bc_filter_num = ioc.filter;
privN->bcomm->bc_filter_len = ioc.len;
if (ioc.len)
{
if(copy_from_user(S2HN(privN->bcomm->bc_filter_area),
ioc.data, ioc.len))
return -EFAULT;
privN->bcomm->bc_filter_cmd = BC_FILTER_SET;
}
else
privN->bcomm->bc_filter_cmd = BC_FILTER_CLR;
return(0);
default:
return -EOPNOTSUPP;
}
}
/*
* Process interrupts
*
* dev, priv will always refer to the 0th device in Multi-NIC mode.
*/
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 06:55:46 -07:00
static irqreturn_t dgrs_intr(int irq, void *dev_id)
{
struct net_device *dev0 = dev_id;
DGRS_PRIV *priv0 = dev0->priv;
I596_CB *cbp;
int cmd;
int i;
++priv0->intrcnt;
if (1) ++priv0->bcomm->bc_cnt[4];
if (0)
{
static int cnt = 100;
if (--cnt > 0)
printk("%s: interrupt: irq %d\n", dev0->name, irq);
}
/*
* Get 596 command
*/
cmd = priv0->scbp->cmd;
/*
* See if RU has been restarted
*/
if ( (cmd & I596_SCB_RUC) == I596_SCB_RUC_START)
{
if (0) printk("%s: RUC start\n", dev0->name);
priv0->rfdp = (I596_RFD *) S2H(priv0->scbp->rfdp);
priv0->rbdp = (I596_RBD *) S2H(priv0->rfdp->rbdp);
priv0->scbp->status &= ~(I596_SCB_RNR|I596_SCB_RUS);
/*
* Tell upper half (halves)
*/
if (dgrs_nicmode)
{
for (i = 0; i < priv0->nports; ++i)
netif_wake_queue (priv0->devtbl[i]);
}
else
netif_wake_queue (dev0);
/* if (bd->flags & TX_QUEUED)
DL_sched(bd, bdd); */
}
/*
* See if any CU commands to process
*/
if ( (cmd & I596_SCB_CUC) != I596_SCB_CUC_START)
{
priv0->scbp->cmd = 0; /* Ignore all other commands */
goto ack_intr;
}
priv0->scbp->status &= ~(I596_SCB_CNA|I596_SCB_CUS);
/*
* Process a command
*/
cbp = (I596_CB *) S2H(priv0->scbp->cbp);
priv0->scbp->cmd = 0; /* Safe to clear the command */
for (;;)
{
switch (cbp->nop.cmd & I596_CB_CMD)
{
case I596_CB_CMD_XMIT:
dgrs_rcv_frame(dev0, priv0, cbp);
break;
default:
cbp->nop.status = I596_CB_STATUS_C | I596_CB_STATUS_OK;
break;
}
if (cbp->nop.cmd & I596_CB_CMD_EL)
break;
cbp = (I596_CB *) S2H(cbp->nop.next);
}
priv0->scbp->status |= I596_SCB_CNA;
/*
* Ack the interrupt
*/
ack_intr:
if (priv0->plxreg)
OUTL(dev0->base_addr + PLX_LCL2PCI_DOORBELL, 1);
return IRQ_HANDLED;
}
/*
* Download the board firmware
*/
static int __init
dgrs_download(struct net_device *dev0)
{
DGRS_PRIV *priv0 = (DGRS_PRIV *) dev0->priv;
int is;
unsigned long i;
2006-03-03 19:33:57 -07:00
static const int iv2is[16] = {
0, 0, 0, ES4H_IS_INT3,
0, ES4H_IS_INT5, 0, ES4H_IS_INT7,
0, 0, ES4H_IS_INT10, ES4H_IS_INT11,
ES4H_IS_INT12, 0, 0, ES4H_IS_INT15 };
/*
* Map in the dual port memory
*/
priv0->vmem = ioremap(dev0->mem_start, 2048*1024);
if (!priv0->vmem)
{
printk("%s: cannot map in board memory\n", dev0->name);
return -ENXIO;
}
/*
* Hold the processor and configure the board addresses
*/
if (priv0->plxreg)
{ /* PCI bus */
proc_reset(dev0, 1);
}
else
{ /* EISA bus */
is = iv2is[dev0->irq & 0x0f];
if (!is)
{
printk("%s: Illegal IRQ %d\n", dev0->name, dev0->irq);
iounmap(priv0->vmem);
priv0->vmem = NULL;
return -ENXIO;
}
OUTB(dev0->base_addr + ES4H_AS_31_24,
(uchar) (dev0->mem_start >> 24) );
OUTB(dev0->base_addr + ES4H_AS_23_16,
(uchar) (dev0->mem_start >> 16) );
priv0->is_reg = ES4H_IS_LINEAR | is |
((uchar) (dev0->mem_start >> 8) & ES4H_IS_AS15);
OUTB(dev0->base_addr + ES4H_IS, priv0->is_reg);
OUTB(dev0->base_addr + ES4H_EC, ES4H_EC_ENABLE);
OUTB(dev0->base_addr + ES4H_PC, ES4H_PC_RESET);
OUTB(dev0->base_addr + ES4H_MW, ES4H_MW_ENABLE | 0x00);
}
/*
* See if we can do DMA on the SE-6
*/
priv0->use_dma = check_board_dma(dev0);
if (priv0->use_dma)
printk("%s: Bus Master DMA is enabled.\n", dev0->name);
/*
* Load and verify the code at the desired address
*/
memcpy(priv0->vmem, dgrs_code, dgrs_ncode); /* Load code */
if (memcmp(priv0->vmem, dgrs_code, dgrs_ncode))
{
iounmap(priv0->vmem);
priv0->vmem = NULL;
printk("%s: download compare failed\n", dev0->name);
return -ENXIO;
}
/*
* Configurables
*/
priv0->bcomm = (struct bios_comm *) (priv0->vmem + 0x0100);
priv0->bcomm->bc_nowait = 1; /* Tell board to make printf not wait */
priv0->bcomm->bc_squelch = 0; /* Flag from Space.c */
priv0->bcomm->bc_150ohm = 0; /* Flag from Space.c */
priv0->bcomm->bc_spew = 0; /* Debug flag from Space.c */
priv0->bcomm->bc_maxrfd = 0; /* Debug flag from Space.c */
priv0->bcomm->bc_maxrbd = 0; /* Debug flag from Space.c */
/*
* Tell board we are operating in switch mode (1) or in
* multi-NIC mode (2).
*/
priv0->bcomm->bc_host = dgrs_nicmode ? BC_MULTINIC : BC_SWITCH;
/*
* Request memory space on board for DMA chains
*/
if (priv0->use_dma)
priv0->bcomm->bc_hostarea_len = (2048/64) * 16;
/*
* NVRAM configurables from Space.c
*/
priv0->bcomm->bc_spantree = dgrs_spantree;
priv0->bcomm->bc_hashexpire = dgrs_hashexpire;
memcpy(priv0->bcomm->bc_ipaddr, dgrs_ipaddr, 4);
memcpy(priv0->bcomm->bc_iptrap, dgrs_iptrap, 4);
memcpy(priv0->bcomm->bc_ipxnet, &dgrs_ipxnet, 4);
/*
* Release processor, wait 8 seconds for board to initialize
*/
proc_reset(dev0, 0);
for (i = jiffies + 8 * HZ; time_after(i, jiffies); )
{
barrier(); /* Gcc 2.95 needs this */
if (priv0->bcomm->bc_status >= BC_RUN)
break;
}
if (priv0->bcomm->bc_status < BC_RUN)
{
printk("%s: board not operating\n", dev0->name);
iounmap(priv0->vmem);
priv0->vmem = NULL;
return -ENXIO;
}
priv0->port = (PORT *) S2H(priv0->bcomm->bc_port);
priv0->scbp = (I596_SCB *) S2H(priv0->port->scbp);
priv0->rfdp = (I596_RFD *) S2H(priv0->scbp->rfdp);
priv0->rbdp = (I596_RBD *) S2H(priv0->rfdp->rbdp);
priv0->scbp->status = I596_SCB_CNA; /* CU is idle */
/*
* Get switch physical and host virtual pointers to DMA
* chaining area. NOTE: the MSB of the switch physical
* address *must* be turned off. Otherwise, the HW kludge
* that allows host access of the PLX DMA registers will
* erroneously select the PLX registers.
*/
priv0->dmadesc_s = (DMACHAIN *) S2DMA(priv0->bcomm->bc_hostarea);
if (priv0->dmadesc_s)
priv0->dmadesc_h = (DMACHAIN *) S2H(priv0->dmadesc_s);
else
priv0->dmadesc_h = NULL;
/*
* Enable board interrupts
*/
if (priv0->plxreg)
{ /* PCI bus */
OUTL(dev0->base_addr + PLX_INT_CSR,
inl(dev0->base_addr + PLX_INT_CSR)
| PLX_PCI_DOORBELL_IE); /* Enable intr to host */
OUTL(dev0->base_addr + PLX_LCL2PCI_DOORBELL, 1);
}
else
{ /* EISA bus */
}
return (0);
}
/*
* Probe (init) a board
*/
static int __init
dgrs_probe1(struct net_device *dev)
{
DGRS_PRIV *priv = (DGRS_PRIV *) dev->priv;
unsigned long i;
int rc;
printk("%s: Digi RightSwitch io=%lx mem=%lx irq=%d plx=%lx dma=%lx\n",
dev->name, dev->base_addr, dev->mem_start, dev->irq,
priv->plxreg, priv->plxdma);
/*
* Download the firmware and light the processor
*/
rc = dgrs_download(dev);
if (rc)
goto err_out;
/*
* Get ether address of board
*/
printk("%s: Ethernet address", dev->name);
memcpy(dev->dev_addr, priv->port->ethaddr, 6);
for (i = 0; i < 6; ++i)
printk("%c%2.2x", i ? ':' : ' ', dev->dev_addr[i]);
printk("\n");
if (dev->dev_addr[0] & 1)
{
printk("%s: Illegal Ethernet Address\n", dev->name);
rc = -ENXIO;
goto err_out;
}
/*
* ACK outstanding interrupts, hook the interrupt,
* and verify that we are getting interrupts from the board.
*/
if (priv->plxreg)
OUTL(dev->base_addr + PLX_LCL2PCI_DOORBELL, 1);
rc = request_irq(dev->irq, &dgrs_intr, IRQF_SHARED, "RightSwitch", dev);
if (rc)
goto err_out;
priv->intrcnt = 0;
for (i = jiffies + 2*HZ + HZ/2; time_after(i, jiffies); )
{
cpu_relax();
if (priv->intrcnt >= 2)
break;
}
if (priv->intrcnt < 2)
{
printk(KERN_ERR "%s: Not interrupting on IRQ %d (%d)\n",
dev->name, dev->irq, priv->intrcnt);
rc = -ENXIO;
goto err_free_irq;
}
/*
* Entry points...
*/
dev->open = &dgrs_open;
dev->stop = &dgrs_close;
dev->get_stats = &dgrs_get_stats;
dev->hard_start_xmit = &dgrs_start_xmit;
dev->set_multicast_list = &dgrs_set_multicast_list;
dev->do_ioctl = &dgrs_ioctl;
return rc;
err_free_irq:
free_irq(dev->irq, dev);
err_out:
return rc;
}
static int __init
dgrs_initclone(struct net_device *dev)
{
DGRS_PRIV *priv = (DGRS_PRIV *) dev->priv;
int i;
printk("%s: Digi RightSwitch port %d ",
dev->name, priv->chan);
for (i = 0; i < 6; ++i)
printk("%c%2.2x", i ? ':' : ' ', dev->dev_addr[i]);
printk("\n");
return (0);
}
static struct net_device * __init
dgrs_found_device(
int io,
ulong mem,
int irq,
ulong plxreg,
ulong plxdma,
struct device *pdev
)
{
DGRS_PRIV *priv;
struct net_device *dev;
int i, ret = -ENOMEM;
dev = alloc_etherdev(sizeof(DGRS_PRIV));
if (!dev)
goto err0;
priv = (DGRS_PRIV *)dev->priv;
dev->base_addr = io;
dev->mem_start = mem;
dev->mem_end = mem + 2048 * 1024 - 1;
dev->irq = irq;
priv->plxreg = plxreg;
priv->plxdma = plxdma;
priv->vplxdma = NULL;
priv->chan = 1;
priv->devtbl[0] = dev;
SET_MODULE_OWNER(dev);
SET_NETDEV_DEV(dev, pdev);
ret = dgrs_probe1(dev);
if (ret)
goto err1;
ret = register_netdev(dev);
if (ret)
goto err2;
if ( !dgrs_nicmode )
return dev; /* Switch mode, we are done */
/*
* Operating card as N separate NICs
*/
priv->nports = priv->bcomm->bc_nports;
for (i = 1; i < priv->nports; ++i)
{
struct net_device *devN;
DGRS_PRIV *privN;
/* Allocate new dev and priv structures */
devN = alloc_etherdev(sizeof(DGRS_PRIV));
ret = -ENOMEM;
if (!devN)
goto fail;
/* Don't copy the network device structure! */
/* copy the priv structure of dev[0] */
privN = (DGRS_PRIV *)devN->priv;
*privN = *priv;
/* ... and zero out VM areas */
privN->vmem = NULL;
privN->vplxdma = NULL;
/* ... and zero out IRQ */
devN->irq = 0;
/* ... and base MAC address off address of 1st port */
devN->dev_addr[5] += i;
ret = dgrs_initclone(devN);
if (ret)
goto fail;
SET_MODULE_OWNER(devN);
SET_NETDEV_DEV(dev, pdev);
ret = register_netdev(devN);
if (ret) {
free_netdev(devN);
goto fail;
}
privN->chan = i+1;
priv->devtbl[i] = devN;
}
return dev;
fail:
while (i >= 0) {
struct net_device *d = priv->devtbl[i--];
unregister_netdev(d);
free_netdev(d);
}
err2:
free_irq(dev->irq, dev);
err1:
free_netdev(dev);
err0:
return ERR_PTR(ret);
}
static void __devexit dgrs_remove(struct net_device *dev)
{
DGRS_PRIV *priv = dev->priv;
int i;
unregister_netdev(dev);
for (i = 1; i < priv->nports; ++i) {
struct net_device *d = priv->devtbl[i];
if (d) {
unregister_netdev(d);
free_netdev(d);
}
}
proc_reset(priv->devtbl[0], 1);
if (priv->vmem)
iounmap(priv->vmem);
if (priv->vplxdma)
iounmap((uchar *) priv->vplxdma);
if (dev->irq)
free_irq(dev->irq, dev);
for (i = 1; i < priv->nports; ++i) {
if (priv->devtbl[i])
unregister_netdev(priv->devtbl[i]);
}
}
#ifdef CONFIG_PCI
static int __init dgrs_pci_probe(struct pci_dev *pdev,
const struct pci_device_id *ent)
{
struct net_device *dev;
int err;
uint io;
uint mem;
uint irq;
uint plxreg;
uint plxdma;
/*
* Get and check the bus-master and latency values.
* Some PCI BIOSes fail to set the master-enable bit,
* and the latency timer must be set to the maximum
* value to avoid data corruption that occurs when the
* timer expires during a transfer. Yes, it's a bug.
*/
err = pci_enable_device(pdev);
if (err)
return err;
err = pci_request_regions(pdev, "RightSwitch");
if (err)
return err;
pci_set_master(pdev);
plxreg = pci_resource_start (pdev, 0);
io = pci_resource_start (pdev, 1);
mem = pci_resource_start (pdev, 2);
pci_read_config_dword(pdev, 0x30, &plxdma);
irq = pdev->irq;
plxdma &= ~15;
/*
* On some BIOSES, the PLX "expansion rom" (used for DMA)
* address comes up as "0". This is probably because
* the BIOS doesn't see a valid 55 AA ROM signature at
* the "ROM" start and zeroes the address. To get
* around this problem the SE-6 is configured to ask
* for 4 MB of space for the dual port memory. We then
* must set its range back to 2 MB, and use the upper
* half for DMA register access
*/
OUTL(io + PLX_SPACE0_RANGE, 0xFFE00000L);
if (plxdma == 0)
plxdma = mem + (2048L * 1024L);
pci_write_config_dword(pdev, 0x30, plxdma + 1);
pci_read_config_dword(pdev, 0x30, &plxdma);
plxdma &= ~15;
dev = dgrs_found_device(io, mem, irq, plxreg, plxdma, &pdev->dev);
if (IS_ERR(dev)) {
pci_release_regions(pdev);
return PTR_ERR(dev);
}
pci_set_drvdata(pdev, dev);
return 0;
}
static void __devexit dgrs_pci_remove(struct pci_dev *pdev)
{
struct net_device *dev = pci_get_drvdata(pdev);
dgrs_remove(dev);
pci_release_regions(pdev);
free_netdev(dev);
}
static struct pci_driver dgrs_pci_driver = {
.name = "dgrs",
.id_table = dgrs_pci_tbl,
.probe = dgrs_pci_probe,
.remove = __devexit_p(dgrs_pci_remove),
};
#else
static struct pci_driver dgrs_pci_driver = {};
#endif
#ifdef CONFIG_EISA
static int is2iv[8] __initdata = { 0, 3, 5, 7, 10, 11, 12, 15 };
static int __init dgrs_eisa_probe (struct device *gendev)
{
struct net_device *dev;
struct eisa_device *edev = to_eisa_device(gendev);
uint io = edev->base_addr;
uint mem;
uint irq;
int rc = -ENODEV; /* Not EISA configured */
if (!request_region(io, 256, "RightSwitch")) {
printk(KERN_ERR "dgrs: eisa io 0x%x, which is busy.\n", io);
return -EBUSY;
}
if ( ! (inb(io+ES4H_EC) & ES4H_EC_ENABLE) )
goto err_out;
mem = (inb(io+ES4H_AS_31_24) << 24)
+ (inb(io+ES4H_AS_23_16) << 16);
irq = is2iv[ inb(io+ES4H_IS) & ES4H_IS_INTMASK ];
dev = dgrs_found_device(io, mem, irq, 0L, 0L, gendev);
if (IS_ERR(dev)) {
rc = PTR_ERR(dev);
goto err_out;
}
gendev->driver_data = dev;
return 0;
err_out:
release_region(io, 256);
return rc;
}
static int __devexit dgrs_eisa_remove(struct device *gendev)
{
struct net_device *dev = gendev->driver_data;
dgrs_remove(dev);
release_region(dev->base_addr, 256);
free_netdev(dev);
return 0;
}
static struct eisa_driver dgrs_eisa_driver = {
.id_table = dgrs_eisa_tbl,
.driver = {
.name = "dgrs",
.probe = dgrs_eisa_probe,
.remove = __devexit_p(dgrs_eisa_remove),
}
};
#endif
/*
* Variables that can be overriden from module command line
*/
static int debug = -1;
static int dma = -1;
static int hashexpire = -1;
static int spantree = -1;
static int ipaddr[4] = { -1 };
static int iptrap[4] = { -1 };
static __u32 ipxnet = -1;
static int nicmode = -1;
module_param(debug, int, 0);
module_param(dma, int, 0);
module_param(hashexpire, int, 0);
module_param(spantree, int, 0);
module_param_array(ipaddr, int, NULL, 0);
module_param_array(iptrap, int, NULL, 0);
module_param(ipxnet, int, 0);
module_param(nicmode, int, 0);
MODULE_PARM_DESC(debug, "Digi RightSwitch enable debugging (0-1)");
MODULE_PARM_DESC(dma, "Digi RightSwitch enable BM DMA (0-1)");
MODULE_PARM_DESC(nicmode, "Digi RightSwitch operating mode (1: switch, 2: multi-NIC)");
static int __init dgrs_init_module (void)
{
int i;
int err;
/*
* Command line variable overrides
* debug=NNN
* dma=0/1
* spantree=0/1
* hashexpire=NNN
* ipaddr=A,B,C,D
* iptrap=A,B,C,D
* ipxnet=NNN
* nicmode=NNN
*/
if (debug >= 0)
dgrs_debug = debug;
if (dma >= 0)
dgrs_dma = dma;
if (nicmode >= 0)
dgrs_nicmode = nicmode;
if (hashexpire >= 0)
dgrs_hashexpire = hashexpire;
if (spantree >= 0)
dgrs_spantree = spantree;
if (ipaddr[0] != -1)
for (i = 0; i < 4; ++i)
dgrs_ipaddr[i] = ipaddr[i];
if (iptrap[0] != -1)
for (i = 0; i < 4; ++i)
dgrs_iptrap[i] = iptrap[i];
if (ipxnet != -1)
dgrs_ipxnet = htonl( ipxnet );
if (dgrs_debug)
{
printk(KERN_INFO "dgrs: SW=%s FW=Build %d %s\nFW Version=%s\n",
version, dgrs_firmnum, dgrs_firmdate, dgrs_firmver);
}
/*
* Find and configure all the cards
*/
#ifdef CONFIG_EISA
err = eisa_driver_register(&dgrs_eisa_driver);
if (err)
return err;
#endif
err = pci_register_driver(&dgrs_pci_driver);
if (err)
return err;
return 0;
}
static void __exit dgrs_cleanup_module (void)
{
#ifdef CONFIG_EISA
eisa_driver_unregister (&dgrs_eisa_driver);
#endif
#ifdef CONFIG_PCI
pci_unregister_driver (&dgrs_pci_driver);
#endif
}
module_init(dgrs_init_module);
module_exit(dgrs_cleanup_module);