Re: patch for at1700.c

[Linus Torvalds]

In article <[e-mail]>,
Brian S. Julin <[e-mail]> wrote:
>
>After much distilling, I've found the critical thing that Hiroaki
>Nagoya's at1700 driver does differently from the one in the mainstream
>tree.  This patch prevents the card from locking up completely during
>a transmit timeout.  I have no manuals and I don't know why it works,
>but it does, and the change seems harmless enough...
>
>--- linux/drivers/net/at1700.c.orig	Tue Jul 18 17:44:11 2000
>+++ linux/drivers/net/at1700.c	Thu Jul 27 19:43:01 2000
>@@ -557,7 +557,8 @@
> 	/* ToDo: We should try to restart the adaptor... */
> 	outw (0xffff, ioaddr + 24);
> 	outw (0xffff, ioaddr + TX_STATUS);
>-	outw (0xe85a, ioaddr + CONFIG_0);
>+	outb (0x5a, ioaddr + CONFIG_0);
>+	outb (0xe8, ioaddr + CONFIG_1);
> 	outw (0x8182, ioaddr + TX_INTR);
> 	outb (0x00, ioaddr + TX_START);
> 	outb (0x03, ioaddr + COL16CNTL);

If this patch makes such a big difference, I would suggest looking at
the "saved_bank" code in set_rx_mode() too.  That code also reads and
writes CONFIG_0/CONFIG_1 as a 16-bit word rather than as two 8-bit
values. 

		Linus