* via-rhine "reset did not complete" errors
@ 2002-10-28 15:45 Larry Sendlosky
2002-10-28 19:38 ` Donald Becker
0 siblings, 1 reply; 6+ messages in thread
From: Larry Sendlosky @ 2002-10-28 15:45 UTC (permalink / raw)
To: linux-net, netdev; +Cc: rl
Hi,
We're using VIA EPIA mini-ITX with 800Mhz C3 and the
VT6103 PHY. (via-rhine driver says VT6102). We have made sure
power supply is "big enough". Our kernel is 2.4.18 with
via-rhine.c patches to fix TX timeout.
Our TX timeout issues seem to have gone away with the recent
patches. However we are still plagued with the "reset did not
complete in 10ms" errors. Once it this state, a warm restart of
the system is necessary (and we have seen this problem at
boot time, which is more confusing).
Anyway, we can make this problem occur easily by doing
'ifconfig down eth0' followed by 'ifconfig up eth0'. The
error happens on the 'open' of the device. However, when
we do this with 'no link' (cable not plugged in), the
problem does not happen. It only seems to happen when
there is activity on the wire and plugged in. When the
reset does not happen, the CR0/CR1 state after the reset
is always 0x8008 now matter what the current state of
those regs going into the reset.
So it appears the problem is actually in the 'close' when
possibly in the process of receiving a packet. Especially
given that via_rhine_close puts the device in loopback mode
to void hardware races. Does via_rhine_close need to be
a little smarter/paranoid? :)) Besides register definitions,
the docs are pretty useless.
Are there hardware issues that are still not fixed? eg.
a reset does not reset the chip regardless of state.
Any help or ideas would be appreciated.
thanks
larry
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: via-rhine "reset did not complete" errors
2002-10-28 15:45 via-rhine "reset did not complete" errors Larry Sendlosky
@ 2002-10-28 19:38 ` Donald Becker
2002-10-29 16:35 ` Roger Luethi
0 siblings, 1 reply; 6+ messages in thread
From: Donald Becker @ 2002-10-28 19:38 UTC (permalink / raw)
To: Larry Sendlosky; +Cc: linux-net, netdev, rl
On Mon, 28 Oct 2002, Larry Sendlosky wrote:
> We're using VIA EPIA mini-ITX with 800Mhz C3 and the
> VT6103 PHY. (via-rhine driver says VT6102). We have made sure
> power supply is "big enough". Our kernel is 2.4.18 with
> via-rhine.c patches to fix TX timeout.
Those are evil patches...
> Our TX timeout issues seem to have gone away with the recent
> patches. However we are still plagued with the "reset did not
> complete in 10ms" errors.
A driver _really_, _really_ shouldn't be busy-waiting for link
negotiation to complete. That happened a lot with MS-DOS drivers, but
it wasn't even reasonable there.
Yet it's far easier for someone to get a horribly flawed patch like that
accepted, while my patches went completely ignored.
> Once it this state, a warm restart of
> the system is necessary (and we have seen this problem at
> boot time, which is more confusing).
Look at the what the code is doing. It is easier to write drivers when
you make the rest of the kernel single threaded on your code...
--
Donald Becker becker@scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Scyld Beowulf cluster system
Annapolis MD 21403 410-990-9993
^ permalink raw reply [flat|nested] 6+ messages in thread
* RE: via-rhine "reset did not complete" errors
@ 2002-10-28 21:03 Larry Sendlosky
0 siblings, 0 replies; 6+ messages in thread
From: Larry Sendlosky @ 2002-10-28 21:03 UTC (permalink / raw)
To: Donald Becker; +Cc: linux-net, netdev, rl
Hi Donald,
Thanks for the reply.
The "reset timeout" happens on the initial reset in the 'open'
routine. The reset bit (bit 7 of CR1) is not clearing after
being written. This usually happens after the 'close' routine
(on a previous 'ifconfig eth0 down') writes the stop cmd to CR0
and the "start" bit is still on after writing the stop cmd to CR0.
What is the proper way to "stop" this little beast?
thanks again,
larry
-----Original Message-----
From: Donald Becker [mailto:becker@scyld.com]
Sent: Monday, October 28, 2002 2:38 PM
To: Larry Sendlosky
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: Re: via-rhine "reset did not complete" errors
On Mon, 28 Oct 2002, Larry Sendlosky wrote:
> We're using VIA EPIA mini-ITX with 800Mhz C3 and the
> VT6103 PHY. (via-rhine driver says VT6102). We have made sure
> power supply is "big enough". Our kernel is 2.4.18 with
> via-rhine.c patches to fix TX timeout.
Those are evil patches...
> Our TX timeout issues seem to have gone away with the recent
> patches. However we are still plagued with the "reset did not
> complete in 10ms" errors.
A driver _really_, _really_ shouldn't be busy-waiting for link
negotiation to complete. That happened a lot with MS-DOS drivers, but
it wasn't even reasonable there.
Yet it's far easier for someone to get a horribly flawed patch like that
accepted, while my patches went completely ignored.
> Once it this state, a warm restart of
> the system is necessary (and we have seen this problem at
> boot time, which is more confusing).
Look at the what the code is doing. It is easier to write drivers when
you make the rest of the kernel single threaded on your code...
--
Donald Becker becker@scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Scyld Beowulf cluster system
Annapolis MD 21403 410-990-9993
^ permalink raw reply [flat|nested] 6+ messages in thread
* RE: via-rhine "reset did not complete" errors
@ 2002-10-28 21:20 Larry Sendlosky
0 siblings, 0 replies; 6+ messages in thread
From: Larry Sendlosky @ 2002-10-28 21:20 UTC (permalink / raw)
To: Larry Sendlosky, Donald Becker; +Cc: linux-net, netdev, rl
Oop, litte red-faced here....basically, I can't count.
When the 'stop' routine does issue a stop cmd,
the "bad" status of CR0 is 0xc (not 0xE as my mind
read it). The "start" bit is not on after the "stop".
Just the recvr and stop bits. A subsequent reset issued by
the 'start' routine will not complete when in this state.
sorry for the mind fart...
larry
-----Original Message-----
From: Larry Sendlosky
Sent: Monday, October 28, 2002 4:03 PM
To: Donald Becker
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: RE: via-rhine "reset did not complete" errors
Hi Donald,
Thanks for the reply.
The "reset timeout" happens on the initial reset in the 'open'
routine. The reset bit (bit 7 of CR1) is not clearing after
being written. This usually happens after the 'close' routine
(on a previous 'ifconfig eth0 down') writes the stop cmd to CR0
and the "start" bit is still on after writing the stop cmd to CR0.
What is the proper way to "stop" this little beast?
thanks again,
larry
-----Original Message-----
From: Donald Becker [mailto:becker@scyld.com]
Sent: Monday, October 28, 2002 2:38 PM
To: Larry Sendlosky
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: Re: via-rhine "reset did not complete" errors
On Mon, 28 Oct 2002, Larry Sendlosky wrote:
> We're using VIA EPIA mini-ITX with 800Mhz C3 and the
> VT6103 PHY. (via-rhine driver says VT6102). We have made sure
> power supply is "big enough". Our kernel is 2.4.18 with
> via-rhine.c patches to fix TX timeout.
Those are evil patches...
> Our TX timeout issues seem to have gone away with the recent
> patches. However we are still plagued with the "reset did not
> complete in 10ms" errors.
A driver _really_, _really_ shouldn't be busy-waiting for link
negotiation to complete. That happened a lot with MS-DOS drivers, but
it wasn't even reasonable there.
Yet it's far easier for someone to get a horribly flawed patch like that
accepted, while my patches went completely ignored.
> Once it this state, a warm restart of
> the system is necessary (and we have seen this problem at
> boot time, which is more confusing).
Look at the what the code is doing. It is easier to write drivers when
you make the rest of the kernel single threaded on your code...
--
Donald Becker becker@scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Scyld Beowulf cluster system
Annapolis MD 21403 410-990-9993
^ permalink raw reply [flat|nested] 6+ messages in thread
* RE: via-rhine "reset did not complete" errors
@ 2002-10-28 22:38 Larry Sendlosky
0 siblings, 0 replies; 6+ messages in thread
From: Larry Sendlosky @ 2002-10-28 22:38 UTC (permalink / raw)
To: Larry Sendlosky, Donald Becker; +Cc: linux-net, netdev, rl
More info/possible solution:
I remembered the Via linuxfet driver and dug it up.
The linuxfet close routine does acknowledge some
problems and does some special casing and action when
the recvr does not turn off in a certain amount of time.
I put that code in the our via-rhine driver and my
'ifconfig eth0 down; ifconfig eth0 up; ping -c2 xxxx; sleep 2;'
loop has been running for about 20 minutes. Without changes
it usually took less than a minute for "fail".
larry
-----Original Message-----
From: Larry Sendlosky
Sent: Monday, October 28, 2002 4:20 PM
To: Larry Sendlosky; Donald Becker
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: RE: via-rhine "reset did not complete" errors
Oop, litte red-faced here....basically, I can't count.
When the 'stop' routine does issue a stop cmd,
the "bad" status of CR0 is 0xc (not 0xE as my mind
read it). The "start" bit is not on after the "stop".
Just the recvr and stop bits. A subsequent reset issued by
the 'start' routine will not complete when in this state.
sorry for the mind fart...
larry
-----Original Message-----
From: Larry Sendlosky
Sent: Monday, October 28, 2002 4:03 PM
To: Donald Becker
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: RE: via-rhine "reset did not complete" errors
Hi Donald,
Thanks for the reply.
The "reset timeout" happens on the initial reset in the 'open'
routine. The reset bit (bit 7 of CR1) is not clearing after
being written. This usually happens after the 'close' routine
(on a previous 'ifconfig eth0 down') writes the stop cmd to CR0
and the "start" bit is still on after writing the stop cmd to CR0.
What is the proper way to "stop" this little beast?
thanks again,
larry
-----Original Message-----
From: Donald Becker [mailto:becker@scyld.com]
Sent: Monday, October 28, 2002 2:38 PM
To: Larry Sendlosky
Cc: linux-net@vger.kernel.org; netdev@oss.sgi.com; rl@hellgate.ch
Subject: Re: via-rhine "reset did not complete" errors
On Mon, 28 Oct 2002, Larry Sendlosky wrote:
> We're using VIA EPIA mini-ITX with 800Mhz C3 and the
> VT6103 PHY. (via-rhine driver says VT6102). We have made sure
> power supply is "big enough". Our kernel is 2.4.18 with
> via-rhine.c patches to fix TX timeout.
Those are evil patches...
> Our TX timeout issues seem to have gone away with the recent
> patches. However we are still plagued with the "reset did not
> complete in 10ms" errors.
A driver _really_, _really_ shouldn't be busy-waiting for link
negotiation to complete. That happened a lot with MS-DOS drivers, but
it wasn't even reasonable there.
Yet it's far easier for someone to get a horribly flawed patch like that
accepted, while my patches went completely ignored.
> Once it this state, a warm restart of
> the system is necessary (and we have seen this problem at
> boot time, which is more confusing).
Look at the what the code is doing. It is easier to write drivers when
you make the rest of the kernel single threaded on your code...
--
Donald Becker becker@scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Scyld Beowulf cluster system
Annapolis MD 21403 410-990-9993
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: via-rhine "reset did not complete" errors
2002-10-28 19:38 ` Donald Becker
@ 2002-10-29 16:35 ` Roger Luethi
0 siblings, 0 replies; 6+ messages in thread
From: Roger Luethi @ 2002-10-29 16:35 UTC (permalink / raw)
To: Donald Becker; +Cc: Larry Sendlosky, linux-net, netdev
On Mon, 28 Oct 2002 14:38:18 -0500, Donald Becker wrote:
> On Mon, 28 Oct 2002, Larry Sendlosky wrote:
>
> > We're using VIA EPIA mini-ITX with 800Mhz C3 and the
> > VT6103 PHY. (via-rhine driver says VT6102). We have made sure
> > power supply is "big enough". Our kernel is 2.4.18 with
> > via-rhine.c patches to fix TX timeout.
>
> Those are evil patches...
Care to elaborate? Since I wrote the patch to fix the Tx timeout issue, I'd
be very interested to learn how exactly it is evil.
Roger
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2002-10-29 16:35 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-10-28 15:45 via-rhine "reset did not complete" errors Larry Sendlosky
2002-10-28 19:38 ` Donald Becker
2002-10-29 16:35 ` Roger Luethi
-- strict thread matches above, loose matches on Subject: below --
2002-10-28 21:03 Larry Sendlosky
2002-10-28 21:20 Larry Sendlosky
2002-10-28 22:38 Larry Sendlosky
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).