[eepro100] Re: No Resources / RX Buffers bug

Andrey Savochkin saw@saw.sw.com.sg
Wed, 30 Aug 2000 12:04:38 +0800


On Tue, Aug 29, 2000 at 02:00:21PM -0700, Jim Reese wrote:
> In my opinion it's clearly a bug. Google is seeing the same problem at

Yes, it's a bug.
Actually, I believe that the whole order of initialization steps in the
driver is completely wrong, and need to be rewritten for the reliable

> the same frequency (~10%). For the time being we've written a few init

It appears in not 100% of cases because it depends on the relative times of
initialization of the chip components submitted to be performed in parallel.

> scripts: one runs before the network is started and forks a process to
> see that a script that runs AFTER the network has been started
> actually ran. If not, the first script reboots the machine.
> It's really a hack, but it does work.

The better work-around is to slow down the initialization.
It may be done by #define USE_IO in the beginning of the driver (and has a
drawback of slowing a bit all operations), or inserting 'udelay(1)' before
each wait_for_cmd_done call from speedo_resume.
I've explained the problem and the workarounds on linux-kernel mailing list
several times.

> Nonetheless, we would love to have this problem solved given that we
> have > 6000 machines with eepro NICs.
> One note: We seem to have this problem ONLY with the eepro100 NICs
> built into the 810E motherboards and not with the standalone NIC.

Best regards
					Andrey V.