It is simple to understand -- the driver assumes that it can directly
access the shared memory (memory-mapped i/o) region which has been
okay for i386 up to v2.0, but that breaks in v2.1. If you really want
to know the nitty-gritty details, then read the file:
linux/Documentation/IO-mapping.txt
> Is there something I can do to help find/erradicate this bug?
Yes, actually -- you can test the following patch against v2.1.30, seeing
as I don't even have an sk_g16 card. I fixed a bunch of the obvious
ones in v2.0.30, and it will now boot without going south, but a lot more
were hidden due to the way the driver was coded and it will explode
when you "ifconfig" it . Hopefully I've found them all now, and this
patch (against v2.1.30) will fix it.
Note that v2.1.30 has other problems in linux/arch/i386/kernel/irq.c
that may cause problems. I've included my fix for that here as well,
even though it has nothing to do with the sk_g16.
If it still fails, please send the kernel messages along with your
System.map file for that kernel, as otherwise the numbers in the kernel
message are useless.
Paul.
----------------------
--- linux/arch/i386/kernel/irq.c.orig Fri Mar 28 14:37:06 1997
+++ linux/arch/i386/kernel/irq.c Fri Mar 28 14:37:10 1997
@@ -485,8 +485,8 @@
#else
-#define irq_enter(cpu, irq) do { } while (0)
-#define irq_exit(cpu, irq) do { } while (0)
+#define irq_enter(cpu, irq) atomic_inc(&intr_count)
+#define irq_exit(cpu, irq) atomic_dec(&intr_count)
#endif
--- linux/drivers/net/sk_g16.c.orig Thu Mar 27 22:21:07 1997
+++ linux/drivers/net/sk_g16.c Fri Mar 28 19:35:41 1997
@@ -40,9 +40,6 @@
* M. Hipp (mhipp@student.uni-tuebingen.de)
* R. Bolz (Schneider & Koch, Germany)
*
- * See README.sk_g16 for details about limitations and bugs for the
- * current version.
- *
* To Do:
* - Support of SK_G8 and other SK Network Cards.
* - Autoset memory mapped RAM. Check for free memory and then
@@ -322,20 +319,6 @@
#define ETHERCARD_TOTAL_SIZE SK_POS_SIZE
/*
- * Portreserve is there to mark the Card I/O Port region as used.
- * Check_region is to check if the region at ioaddr with the size "size"
- * is free or not.
- * Snarf_region allocates the I/O Port region.
- */
-
-#ifndef HAVE_PORTRESERVE
-
-#define check_region(ioaddr, size) 0
-#define request_region(ioaddr, size,name) do ; while (0)
-
-#endif
-
-/*
* SK_DEBUG
*
* Here you can choose what level of debugging wanted.
@@ -1028,6 +1011,7 @@
static int SK_lance_init(struct device *dev, unsigned short mode)
{
int i;
+ unsigned long flags;
struct priv *p = (struct priv *) dev->priv;
struct tmd *tmdp;
struct rmd *rmdp;
@@ -1046,10 +1030,10 @@
{
tmdp = p->tmdhead + i;
- tmdp->u.buffer = (unsigned long) p->tmdbufs[i]; /* assign buffer */
+ writel((unsigned long) p->tmdbufs[i], tmdp->u.buffer); /* assign buffer */
/* Mark TMD as start and end of packet */
- tmdp->u.s.status = TX_STP | TX_ENP;
+ writeb(TX_STP | TX_ENP, tmdp->u.s.status);
}
@@ -1062,42 +1046,43 @@
rmdp = p->rmdhead + i;
- rmdp->u.buffer = (unsigned long) p->rmdbufs[i]; /* assign buffer */
+ writel((unsigned long) p->rmdbufs[i], rmdp->u.buffer); /* assign buffer */
/*
* LANCE must be owner at beginning so that he can fill in
* receiving packets, set status and release RMD
*/
- rmdp->u.s.status = RX_OWN;
+ writeb(RX_OWN, rmdp->u.s.status);
- rmdp->blen = -PKT_BUF_SZ; /* Buffer Size in a two's complement */
+ writew(-PKT_BUF_SZ, rmdp->blen); /* Buffer Size (two's complement) */
- rmdp->mlen = 0; /* init message length */
+ writeb(0, rmdp->mlen); /* init message length */
}
/* Fill LANCE Initialize Block */
- (p->ram)->ib.mode = mode; /* Set operation mode */
+ writew(mode, (p->ram)->ib.mode); /* Set operation mode */
for (i = 0; i < ETH_ALEN; i++) /* Set physical address */
{
- (p->ram)->ib.paddr[i] = dev->dev_addr[i];
+ writeb(dev->dev_addr[i], (p->ram)->ib.paddr[i]);
}
for (i = 0; i < 8; i++) /* Set multicast, logical address */
{
- (p->ram)->ib.laddr[i] = 0; /* We do not use logical addressing */
+ writeb(0, (p->ram)->ib.laddr[i]); /* We do not use logical addressing */
}
/* Set ring descriptor pointers and set number of descriptors */
- (p->ram)->ib.rdrp = (int) p->rmdhead | RMDNUMMASK;
- (p->ram)->ib.tdrp = (int) p->tmdhead | TMDNUMMASK;
+ writel((int)p->rmdhead | RMDNUMMASK, (p->ram)->ib.rdrp);
+ writel((int)p->tmdhead | TMDNUMMASK, (p->ram)->ib.tdrp);
/* Prepare LANCE Control and Status Registers */
+ save_flags(flags);
cli();
SK_write_reg(CSR3, CSR3_ACON); /* Ale Control !!!THIS MUST BE SET!!!! */
@@ -1133,7 +1118,7 @@
SK_write_reg(CSR0, CSR0_INIT);
- sti();
+ restore_flags(flags);
/* Wait until LANCE finished initialization */
@@ -1216,22 +1201,6 @@
}
- /*
- * If some upper Layer thinks we missed a transmit done interrupt
- * we are passed NULL.
- * (dev_queue_xmit net/inet/dev.c
- */
-
- if (skb == NULL)
- {
- /*
- * Dequeue packets from transmit queue and send them.
- */
- dev_tint(dev);
-
- return 0;
- }
-
PRINTK2(("## %s: SK_send_packet() called, CSR0 %#04x.\n",
SK_NAME, SK_read_reg(CSR0)));
@@ -1258,7 +1227,7 @@
memcpy_toio((tmdp->u.buffer & 0x00ffffff), skb->data, skb->len);
- tmdp->blen = -len; /* set length to transmit */
+ writew(-len, tmdp->blen); /* set length to transmit */
/*
* Packet start and end is always set because we use the maximum
@@ -1266,7 +1235,7 @@
* Relinquish ownership to LANCE
*/
- tmdp->u.s.status = TX_OWN | TX_STP | TX_ENP;
+ writeb(TX_OWN | TX_STP | TX_ENP, tmdp->u.s.status);
/* Start Demand Transmission */
SK_write_reg(CSR0, CSR0_TDMD | CSR0_INEA);
@@ -1278,7 +1247,7 @@
p->tmdnum &= TMDNUM-1;
/* Do we own the next transmit buffer ? */
- if (! ((p->tmdhead + p->tmdnum)->u.s.status & TX_OWN) )
+ if (! (readb((p->tmdhead + p->tmdnum)->u.s.status) & TX_OWN) )
{
/*
* We own next buffer and are ready to transmit, so
@@ -1400,7 +1369,7 @@
p->tmdlast++;
p->tmdlast &= TMDNUM-1;
- tmdstat = tmdp->u.s.status & 0xff00; /* filter out status bits 15:08 */
+ tmdstat = readb(tmdp->u.s.status);
/*
* We check status of transmitted packet.
@@ -1408,21 +1377,22 @@
*/
if (tmdstat & TX_ERR) /* Error occurred */
{
- printk("%s: TX error: %04x %04x\n", dev->name, (int) tmdstat,
- (int) tmdp->status2);
+ int stat2 = readw(tmdp->status2);
+
+ printk("%s: TX error: %04x %04x\n", dev->name, tmdstat, stat2);
- if (tmdp->status2 & TX_TDR) /* TDR problems? */
+ if (stat2 & TX_TDR) /* TDR problems? */
{
printk("%s: tdr-problems \n", dev->name);
}
- if (tmdp->status2 & TX_RTRY) /* Failed in 16 attempts to transmit ? */
+ if (stat2 & TX_RTRY) /* Failed in 16 attempts to transmit ? */
p->stats.tx_aborted_errors++;
- if (tmdp->status2 & TX_LCOL) /* Late collision ? */
+ if (stat2 & TX_LCOL) /* Late collision ? */
p->stats.tx_window_errors++;
- if (tmdp->status2 & TX_LCAR) /* Loss of Carrier ? */
+ if (stat2 & TX_LCAR) /* Loss of Carrier ? */
p->stats.tx_carrier_errors++;
- if (tmdp->status2 & TX_UFLO) /* Underflow error ? */
+ if (stat2 & TX_UFLO) /* Underflow error ? */
{
p->stats.tx_fifo_errors++;
@@ -1436,7 +1406,7 @@
p->stats.tx_errors++;
- tmdp->status2 = 0; /* Clear error flags */
+ writew(0, tmdp->status2); /* Clear error flags */
}
else if (tmdstat & TX_MORE) /* Collisions occurred ? */
{
@@ -1449,7 +1419,7 @@
* First I did not have this in but then I thought at minimum
* we see that something was not ok.
* If anyone knows something better than this to handle this
- * please report it. (see Email addresses in the README file)
+ * please report it.
*/
p->stats.collisions++;
@@ -1519,7 +1489,7 @@
* it up to higher layer
*/
- while (!( (rmdstat = rmdp->u.s.status) & RX_OWN))
+ while (!( (rmdstat = readb(rmdp->u.s.status)) & RX_OWN))
{
/*
* Start and end of packet must be set, because we use
@@ -1549,7 +1519,7 @@
* packets.
*/
- rmdp->u.s.status = RX_OWN; /* Relinquish ownership to LANCE */
+ writeb(RX_OWN, rmdp->u.s.status); /* Relinquish ownership to LANCE */
}
else if (rmdstat & RX_ERR) /* Receive Error ? */
@@ -1561,13 +1531,13 @@
if (rmdstat & RX_FRAM) p->stats.rx_frame_errors++;
if (rmdstat & RX_CRC) p->stats.rx_crc_errors++;
- rmdp->u.s.status = RX_OWN; /* Relinquish ownership to LANCE */
+ writeb(RX_OWN, rmdp->u.s.status); /* Relinquish ownership to LANCE */
}
else /* We have a packet which can be queued for the upper layers */
{
- int len = (rmdp->mlen & 0x0fff); /* extract message length from receive buffer */
+ int len = readw(rmdp->mlen) & 0x0fff; /* extract message length from receive buffer */
struct sk_buff *skb;
skb = dev_alloc_skb(len+2); /* allocate socket buffer */
@@ -1580,7 +1550,7 @@
* to Lance, update statistics and go ahead.
*/
- rmdp->u.s.status = RX_OWN; /* Relinquish ownership to LANCE */
+ writeb(RX_OWN, rmdp->u.s.status); /* Relinquish ownership to LANCE */
printk("%s: Couldn't allocate sk_buff, deferring packet.\n",
dev->name);
p->stats.rx_dropped++;
@@ -1618,7 +1588,7 @@
* free our descriptor and update statistics
*/
- rmdp->u.s.status = RX_OWN;
+ writeb(RX_OWN, rmdp->u.s.status);
p->stats.rx_packets++;