The fcc code in 8260_io has been obsoleted by fs_enet driver (drivers/net/fs_enet).
It is known to work with 85xx at least in recent kernels. If it does not for some reason,
please letus know.
Thanks,
-Vitaly
Hi Vitaly,
The problem seems to be similar in recent kernels
(/drivers/net/fs_enet/mac-fcc.c). The transmit clock is inverted (TCI
bit set). The packets are well received, transmitted packets go out (tx
counter is ok), unfortunately as the clock is inverted (TCI) the data
are not correctly received by the other side. The FCC works in 100Mbps
(luck?) but doesn't in 10Mbps. I solved this problem by removing the TCI
bit set as described in the patch. Actually, this problem depends also
of the Ethernet phy used on board.
Cedric
-----Original Message-----
From: Vitaly Bordug [mailto:vbordug@ru.mvista.com]=20
Sent: mardi 6 mars 2007 10:42
To: Li Yang-r58472
Cc: Pontois, Cedric UMTS (CEZ:8Z10); dmalek@jlc.net;
linux-kernel@vger.kernel.org; linuxppc-embedded@ozlabs.org
Subject: Re: [patch 1/1] ppc: Possible bug fix for FCC driver
On Tue, 6 Mar 2007 10:23:43 +0800
Li Yang-r58472 wrote:
Hi Cedric,
=20
For ppc embedded related patches, please also cc:=20
linuxppc-embedded@ozlabs.org.
=20
quoted
We use a kernel 2.6.14 on PPC platform (MPC 8555). The FCC driver
works
=20
To submit a kernel patch upstream, the patch should be against the=20
latest kernel version which is 2.6.21-rc now.
quoted
well with a 100Mbps link. But it doesn't with a 10Mbps link. To=20
solve it, I modified the GFMR register init: removed TCI bit and set
quoted
CRC32
bit
quoted
instead of.
=20
I don't know how these bits caused the 10M link issue. Do you have=20
any reasoning?
=20
The fcc code in 8260_io has been obsoleted by fs_enet driver
(drivers/net/fs_enet). It is known to work with 85xx at least in recent
kernels. If it does not for some reason, please letus know.
Thanks,
-Vitaly
On Tue, 6 Mar 2007 12:03:57 +0100
Cedric Pontois wrote:
Hi Vitaly,
The problem seems to be similar in recent kernels
(/drivers/net/fs_enet/mac-fcc.c). The transmit clock is inverted (TCI
bit set). The packets are well received, transmitted packets go out
(tx counter is ok), unfortunately as the clock is inverted (TCI) the
data are not correctly received by the other side. The FCC works in
100Mbps (luck?) but doesn't in 10Mbps. I solved this problem by
removing the TCI bit set as described in the patch. Actually, this
problem depends also of the Ethernet phy used on board.
IIRC it worked for us with Davicom PHY on 10 MBps on one of 8260-like...
Well I give it a try once more with the recent kernel, but anyway if you think
you've located and fixed root cause, do (and test) a patch against recent kernel,
send it to me cc'ing netdev and linuxppc-embedded, and I'll push it along
if it is not breaking existing boards.
-Vitaly
Cedric
-----Original Message-----
From: Vitaly Bordug [mailto:vbordug@ru.mvista.com]
Sent: mardi 6 mars 2007 10:42
To: Li Yang-r58472
Cc: Pontois, Cedric UMTS (CEZ:8Z10); dmalek@jlc.net;
linux-kernel@vger.kernel.org; linuxppc-embedded@ozlabs.org
Subject: Re: [patch 1/1] ppc: Possible bug fix for FCC driver
On Tue, 6 Mar 2007 10:23:43 +0800
Li Yang-r58472 wrote:
quoted
Hi Cedric,
For ppc embedded related patches, please also cc:
linuxppc-embedded@ozlabs.org.
quoted
We use a kernel 2.6.14 on PPC platform (MPC 8555). The FCC driver
works
To submit a kernel patch upstream, the patch should be against the
latest kernel version which is 2.6.21-rc now.
quoted
well with a 100Mbps link. But it doesn't with a 10Mbps link. To
solve it, I modified the GFMR register init: removed TCI bit and
set
quoted
quoted
CRC32
bit
quoted
instead of.
I don't know how these bits caused the 10M link issue. Do you have
any reasoning?
The fcc code in 8260_io has been obsoleted by fs_enet driver
(drivers/net/fs_enet). It is known to work with 85xx at least in
recent kernels. If it does not for some reason, please letus know.
Thanks,
-Vitaly