On Sun, 1 Jan 1995, Sanjay Lal wrote:

> Hmm, I must have forgotten to submit this fix for the fxp driver. Here
> it goes (sorry I dont have the diffs.
> 
> In -current the file in question is .../dev/ice/i82557.c In both cases
> the bug is a endian issue. The driver issued a command to the chip and
> polls a bit waiting for it to complete. however due to the endian
> swap, the wrong bit is polled, and the kernel hangs an in infinite
> while loop.  I have added the htole16() call whithin the while loop.

That _seems_ to have worked.  There was a third occurrence of the problem
that you'd forgotten about, though.  Here's the patch (beware tab->space
cut/paste lossage):

--------8<--------
--- i82557.c.orig       Thu Mar 30 13:22:53 2000
+++ i82557.c    Wed May  3 23:48:55 2000
@@ -1373,7 +1373,7 @@
        do {
                FXP_CDCONFIGSYNC(sc,
                    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
-       } while ((cbp->cb_status & FXP_CB_STATUS_C) == 0);
+       } while ((htole16(cbp->cb_status) & FXP_CB_STATUS_C) == 0);
 
        /*
         * Initialize the station address.
@@ -1398,7 +1398,7 @@
        do {
                FXP_CDIASSYNC(sc,
                    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
-       } while ((cb_ias->cb_status & FXP_CB_STATUS_C) == 0);
+       } while ((htole16(cb_ias->cb_status) & FXP_CB_STATUS_C) == 0);
 
        /*
         * Initialize the transmit descriptor ring.  txlast is initialized
@@ -1848,7 +1848,7 @@
        do {
                FXP_CDMCSSYNC(sc,
                    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
-       } while ((mcsp->cb_status & FXP_CB_STATUS_C) == 0);
+       } while ((htole16(mcsp->cb_status) & FXP_CB_STATUS_C) == 0);
 }
 
 int
-------->8--------
This was to revision 1.25 of the file.  Incidentally, wouldn't le16toh be
slightly more appropriate here?

FWIW, here's the dmesg fragment:
--------8<--------
fxp0 at pci0 dev 13 function 0: Intel i82557 Ethernet, rev 1
fxp0: interrupting at irq 23
fxp0: Ethernet address a0:00:10:c9:53:93, 10 Mb/s
fxp0: Seeq 80c24 AutoDUPLEX media interface present
-------->8--------
The card is an Intel EtherExpress PRO/10+ PCI LAN Adapter, according to
the cover of the manual.

I've not tested the interface under load yet, but I can at least log in
over it now.

-- 
Ben Harris
Unix Support, University of Cambridge Computing Service.