* Re: Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-19 21:04 Mohammad Sadegh Sadri
2007-06-21 14:53 ` Andrei Konovalov
0 siblings, 1 reply; 9+ messages in thread
From: Mohammad Sadegh Sadri @ 2007-06-19 21:04 UTC (permalink / raw)
To: linuxppc-embedded, kashiwagi, glenn.g.hart
Dear Glenn,
Dear Yoshio
I'm wondering where are the original developers of TEMAC driver these days,=
I have not seen any posts from any of andrei nor grant in the recent weeks=
. may be they are involved in other projects.
checksum off loading can be used only with SGDMA, it can also remain disabl=
ed in this mode and the CPU will compute the CSUM itself.=20
Grant's kernel source tree is 2.6.21 and the latest release of EDK is 9.1SP=
2. these two make a good set of dev tools for building the needed systems.=
=20
as far as I know there are some bugs in the SP1 version of edk 9.1 which ma=
kes the ethernet link unstable. In our tests the systems implemented with E=
DK 9.1SP1 were not able to bring up the linux root file system on NFS corre=
ctly and we always encoutered "segmentation fault" messages even when we li=
mited the packet sizes with rsize and wsize kernel parameters.
Using EDK 9.1SP2 however ( and also the latest virtex-dev tree from grant's=
git tree ) we were able to bring up the root file system via NFS with exac=
tly no problem over both of 100mbits and 1000mbits links.
However unfortunately our band width tests for the gigabit ethernet link of=
ML403 has not resulted any good values till now ( all of the obtained valu=
es are smaller than 40mbits/s for both of 100 and 1000 links ) we have not =
understood the reason for this behavior yet.
any how, the main disscussion is about avnet mini module and availability o=
f good examples and good base systems for it.=20
Basically the main question can be stated as follows:
- Is there any reference design, or example available for avnet mini-module=
.?=20
---------------------------------------------------------------------------=
-----
>Message: 9
>Date: Tue, 19 Jun 2007 14:06:18 -0400
>From: Glenn.G.Hart@us.westinghouse.com
>Subject: Re: Virtex 4 Avnet Mini-Module Networking.
>To: kashiwagi@co-nss.co.jp, linuxppc-embedded@ozlabs.org
>Message-ID:
>=09
>Content-Type: text/plain; charset=3D"us-ascii"
>=20
>=20
>Yoshio Kashiwagi,
>=20
>Yes the checksum off loading is turned off. Should it be turned on? My
>documentation says I need to use DMA mode to use checksum offloading.
>=20
>Glenn
_________________________________________________________________
Connect to the next generation of MSN Messenger=A0
http://imagine-msn.com/messenger/launch80/default.aspx?locale=3Den-us&sourc=
e=3Dwlmailtagline=
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Virtex 4 Avnet Mini-Module Networking.
2007-06-19 21:04 Virtex 4 Avnet Mini-Module Networking Mohammad Sadegh Sadri
@ 2007-06-21 14:53 ` Andrei Konovalov
0 siblings, 0 replies; 9+ messages in thread
From: Andrei Konovalov @ 2007-06-21 14:53 UTC (permalink / raw)
To: Mohammad Sadegh Sadri; +Cc: linuxppc-embedded
Mohammad Sadegh Sadri wrote:
> I'm wondering where are the original developers of TEMAC driver these days, I have not seen any posts from any of andrei
> nor grant in the recent weeks. may be they are involved in other projects.
As for me - no, not until the end of this week..
I am puzzled with the following:
I've got ML405 recently. There is a prebuilt bitstream with TEMAC in SGDMA mode for this board made by Xilinx (XAPP941).
With the TEMAC driver from EDK 9.1SP2 and the XAPP941 bitstream I've got (no jumbo frames):
XTemac: using sgDMA mode.
XTemac: using TxDRE mode
XTemac: using RxDRE mode
XTemac: buffer descriptor size: 32768 (0x8000)
XTemac: (buffer_descriptor_init) phy: 0x3cf8000, virt: 0xff100000, size: 0x8000
eth%d: XTemac: PHY detected at address 7.
eth0: Xilinx TEMAC #0 at 0x80000000 mapped to 0xC5060000, irq=0
eth0: XTemac id 1.0f, block id 5, type 8
<snip>
eth0: XTemac: Options: 0xb8f2
eth0: XTemac: We renegotiated the speed to: 1000
eth0: XTemac: speed set to 1000Mb/s
eth0: XTemac: Send Threshold = 16, Receive Threshold = 2
eth0: XTemac: Send Wait bound = 1, Receive Wait bound = 1
<snip>
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10,
2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.11
9.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # 10^6bits/sec
217088 32768 30.00 26391 0 230.60
107520 30.00 26377 230.48
Not that bad for this setup.
With xps (EDK 9.1 without any SPs) I've tried creating exact the same design.
So that exact the same kernel etc could be used for the both bitstreams.
And with the self-made one I see exact the same output with the only difference:
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10,
2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.11
9.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # 10^6bits/sec
217088 32768 30.00 9593 0 83.82
107520 30.00 9586 83.76
:o(
Rebuilding the XAPP941 project with EDK 9.1.00 still leaves 230Mb/sec.
I also pulled the additional timing constraints from XAPP941's data/system.ucf
into the self-made data/system.ucf, but this haven't change anything.
I've run out of ideas, and switched from TEMAC to other Xilinx stuff for a while.
Thanks,
Andrei
^ permalink raw reply [flat|nested] 9+ messages in thread
* RE: Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-22 21:21 Mohammad Sadegh Sadri
0 siblings, 0 replies; 9+ messages in thread
From: Mohammad Sadegh Sadri @ 2007-06-22 21:21 UTC (permalink / raw)
To: Andrei Konovalov; +Cc: linuxppc-embedded
Andrei,
well you are correct, using EDK9.1SP2 I created a custome design for ML405 =
and implemented it. My design was exactly the same as what xapp941 presents=
.=20
I have not access to ML405, but examining implementation results for two de=
signs it is obvious that they are different. logic resource usage for xapp9=
41 is over 6000 slices but for the custome design it is near 5700 slices.=20
In addition the fan out value for PLB TEMAC clock is near 5000 for xapp941 =
and near 3000 for our custome design. this means that plb temac module in x=
app941 is much larger than our plb temac module.=20
I examined the MHS files for both designs carefully, they are really the sa=
me. Of course it seems that the mhs file for xapp941 is edited by hand or a=
t least is not created by the usual method we use. This is because signal n=
ames and module names are completely different from an automatic EDK9.1 gen=
erated design.=20
Finally why not ask the two authors of this app note about what they have d=
one in their desing? is it some kind of secret?=20
and have you ever tried using ngc file for plb temac of xapp941 in a custom=
e design? .... should not be really neccessary of course.
----------------------------------------
> Date: Thu, 21 Jun 2007 22:18:52 +0400
> From: akonovalov@ru.mvista.com
> To: mamsadegh@hotmail.com
> CC: linuxppc-embedded@ozlabs.org
> Subject: Re: Virtex 4 Avnet Mini-Module Networking.
>=20
> Mohammad Sadegh Sadri wrote:
> >=20
> > very interesting
> >=20
> > the provided xps project from xapp941 results 230Mbits but your custome=
system is 83.0Mbits, yes?
>=20
> yes
>=20
> > just, are you sure the ppc core has the same core clock frequency in bo=
th designs? I mean may be the xapp941 is running at 300MHz but yours at 100=
MHz.=20
>=20
> both run at 300MHz
>=20
> > And what about a custom system with 9.1SP2? does that gives the same pe=
rformance?
>=20
> haven't tried this yet
>=20
> > Finally If possible please release the test results for TCP streams. ( =
9.1 no SP and 9.1 SP2 )
>=20
> The TEMAC driver is from EDK 9.1.02 (IOW, SP2).
>=20
> bitstream made by myself with EDK 9.1.00:
> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>=20
> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1=
-i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1 (1=
92.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
>=20
> 87380 217088 32768 30.00 41.66
>=20
> bitstream from XAPP941:
> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>=20
> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1=
-i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1 (1=
92.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
>=20
> 87380 217088 32768 30.00 111.86
>=20
> And I didn't test ML405 receive performance.
>=20
> Thanks,
> Andrei
>=20
> > thanks
> >=20
> >=20
> >=20
> > ----------------------------------------
> >> Date: Thu, 21 Jun 2007 18:53:49 +0400
> >> From: akonovalov@ru.mvista.com
> >> To: mamsadegh@hotmail.com
> >> CC: linuxppc-embedded@ozlabs.org; kashiwagi@co-nss.co.jp; glenn.g.hart=
@us.westinghouse.com
> >> Subject: Re: Virtex 4 Avnet Mini-Module Networking.
> >>
> >> Mohammad Sadegh Sadri wrote:
> >>> I'm wondering where are the original developers of TEMAC driver these=
days, I have not seen any posts from any of andrei
> >>> nor grant in the recent weeks. may be they are involved in other proj=
ects.
> >> As for me - no, not until the end of this week..
> >>
> >> I am puzzled with the following:
> >>
> >> I've got ML405 recently. There is a prebuilt bitstream with TEMAC in S=
GDMA mode for this board made by Xilinx (XAPP941).
> >> With the TEMAC driver from EDK 9.1SP2 and the XAPP941 bitstream I've g=
ot (no jumbo frames):
> >>
> >> XTemac: using sgDMA mode.
> >> XTemac: using TxDRE mode
> >> XTemac: using RxDRE mode
> >> XTemac: buffer descriptor size: 32768 (0x8000)
> >> XTemac: (buffer_descriptor_init) phy: 0x3cf8000, virt: 0xff100000, =
size: 0x8000
> >> eth%d: XTemac: PHY detected at address 7.
> >> eth0: Xilinx TEMAC #0 at 0x80000000 mapped to 0xC5060000, irq=3D0
> >> eth0: XTemac id 1.0f, block id 5, type 8
> >>
> >> eth0: XTemac: Options: 0xb8f2
> >> eth0: XTemac: We renegotiated the speed to: 1000
> >> eth0: XTemac: speed set to 1000Mb/s
> >> eth0: XTemac: Send Threshold =3D 16, Receive Threshold =3D 2
> >> eth0: XTemac: Send Wait bound =3D 1, Receive Wait bound =3D 1
> >>
> >> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168=
.119.1 -i 10,
> >> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
> >> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET =
to 192.168.11
> >> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> >> Socket Message Elapsed Messages
> >> Size Size Time Okay Errors Throughput
> >> bytes bytes secs # # 10^6bits/sec
> >>
> >> 217088 32768 30.00 26391 0 230.60
> >> 107520 30.00 26377 230.48
> >>
> >> Not that bad for this setup.
> >>
> >> With xps (EDK 9.1 without any SPs) I've tried creating exact the same =
design.
> >> So that exact the same kernel etc could be used for the both bitstream=
s.
> >> And with the self-made one I see exact the same output with the only d=
ifference:
> >>
> >> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168=
.119.1 -i 10,
> >> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
> >> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET =
to 192.168.11
> >> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> >> Socket Message Elapsed Messages
> >> Size Size Time Okay Errors Throughput
> >> bytes bytes secs # # 10^6bits/sec
> >>
> >> 217088 32768 30.00 9593 0 83.82
> >> 107520 30.00 9586 83.76
> >>
> >> :o(
> >>
> >> Rebuilding the XAPP941 project with EDK 9.1.00 still leaves 230Mb/sec.
> >> I also pulled the additional timing constraints from XAPP941's data/sy=
stem.ucf
> >> into the self-made data/system.ucf, but this haven't change anything.
> >>
> >> I've run out of ideas, and switched from TEMAC to other Xilinx stuff f=
or a while.
> >>
> >>
> >> Thanks,
> >> Andrei
> >=20
> > _________________________________________________________________
> > Connect to the next generation of MSN Messenger=20
> > http://imagine-msn.com/messenger/launch80/default.aspx?locale=3Den-us&s=
ource=3Dwlmailtagline
> > _______________________________________________
> > Linuxppc-embedded mailing list
> > Linuxppc-embedded@ozlabs.org
> > https://ozlabs.org/mailman/listinfo/linuxppc-embedded
>=20
_________________________________________________________________
Connect to the next generation of MSN Messenger=A0
http://imagine-msn.com/messenger/launch80/default.aspx?locale=3Den-us&sourc=
e=3Dwlmailtagline=
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-22 19:02 Glenn.G.Hart
0 siblings, 0 replies; 9+ messages in thread
From: Glenn.G.Hart @ 2007-06-22 19:02 UTC (permalink / raw)
To: linuxppc-embedded
Has anybody or know of anybody that has got the ethernet to work on the=
Avnet V4 Mini-Module with Linux?
My closest came with EDK 8.2iSP3 and Linux 2.6.19. I was able to ping =
the
computers on the network, but I could not fully trasnfer data. Capturn=
ing
the packets using Ethereal showed the TCP checksum was incorrect while =
the
IP checksum was correct. I has a lot of trouble get the DMA to attempt=
to
run with thei statement.
I have now tried upgrading to EDK 9.1iSP2 and Linux-2.6.21 from Grant's=
git
server. Now I do not see any packets using thernet evrything else seem=
s
the same. When I try to ping another computer, I see activity on the
switch, but no packets are transmitted. Samething happens using FIFO o=
r
DMA Below is the kernel output when booting.
[ 4.186714] XTemac: using sgDMA mode.
[ 4.230800] XTemac: using TxDRE mode
[ 4.273829] XTemac: using RxDRE mode
[ 4.316805] XTemac: buffer descriptor size: 32768 (0x8000)
[ 4.383980] XTemac: (buffer_descriptor_init) phy: 0x5c8000, virt:
0xff100000, size: 0x8000
[ 4.494600] eth%d: XTemac: PHY detected at address 4.
[ 4.557994] eth0: Xilinx TEMAC #0 at 0x81200000 mapped to 0xC5000000=
,
irq=3D0
[ 4.641808] eth0: XTemac id 1.0f, block id 5, type 8
[ 4.703065] mice: PS/2 mouse device common for all mice
[ 4.766425] TCP cubic registered
[ 4.805581] NET: Registered protocol family 1
[ 4.858089] NET: Registered protocol family 17
[ 4.914484] RAMDISK: Compressed image found at block 0
[ 7.190462] VFS: Mounted root (ext2 filesystem).
[ 7.246742] Freeing unused kernel memory: 92k init
[ 7.445413] eth0: XTemac: Options: 0xb8f2
[ 12.986126] eth0: XTemac: We renegotiated the speed to: 1000
[ 13.054001] eth0: XTemac: speed set to 1000Mb/s
[ 13.120469] eth0: XTemac: Send Threshold =3D 16, Receive Threshold =3D=
2
[ 13.212064] eth0: XTemac: Send Wait bound =3D 1, Receive Wait bound =
=3D 1
Any ideas/comments. It would be great to hear from somebody that this =
has
a chance of working because you got it wokrking.
Thanks,
Glenn
Mohammad Sadegh Sadri wrote:
>
> very interesting
>
> the provided xps project from xapp941 results 230Mbits but your custo=
me
system is 83.0Mbits, yes?
yes
> just, are you sure the ppc core has the same core clock frequency in =
both
designs? I mean may be the xapp941 is running at 300MHz but yours at
100MHz.
both run at 300MHz
> And what about a custom system with 9.1SP2? does that gives the same
performance?
haven't tried this yet
> Finally If possible please release the test results for TCP streams. =
(
9.1 no SP and 9.1 SP2 )
The TEMAC driver is from EDK 9.1.02 (IOW, SP2).
bitstream made by myself with EDK 9.1.00:
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119=
.1
-i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1
(192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 217088 32768 30.00 41.66
bitstream from XAPP941:
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119=
.1
-i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1
(192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 217088 32768 30.00 111.86
And I didn't test ML405 receive performance.
Thanks,
Andrei
> thanks
>
>
>
> ----------------------------------------
>> Date: Thu, 21 Jun 2007 18:53:49 +0400
>> From: akonovalov@ru.mvista.com
>> To: mamsadegh@hotmail.com
>> CC: linuxppc-embedded@ozlabs.org; kashiwagi@co-nss.co.jp;
glenn.g.hart@us.westinghouse.com
>> Subject: Re: Virtex 4 Avnet Mini-Module Networking.
>>
>> Mohammad Sadegh Sadri wrote:
>>> I'm wondering where are the original developers of TEMAC driver the=
se
days, I have not seen any posts from any of andrei
>>> nor grant in the recent weeks. may be they are involved in other
projects.
>> As for me - no, not until the end of this week..
>>
>> I am puzzled with the following:
>>
>> I've got ML405 recently. There is a prebuilt bitstream with TEMAC in=
SGDMA mode for this board made by Xilinx (XAPP941).
>> With the TEMAC driver from EDK 9.1SP2 and the XAPP941 bitstream I've=
got
(no jumbo frames):
>>
>> XTemac: using sgDMA mode.
>> XTemac: using TxDRE mode
>> XTemac: using RxDRE mode
>> XTemac: buffer descriptor size: 32768 (0x8000)
>> XTemac: (buffer_descriptor_init) phy: 0x3cf8000, virt: 0xff100000=
,
size: 0x8000
>> eth%d: XTemac: PHY detected at address 7.
>> eth0: Xilinx TEMAC #0 at 0x80000000 mapped to 0xC5060000, irq=3D0=
>> eth0: XTemac id 1.0f, block id 5, type 8
>>
>> eth0: XTemac: Options: 0xb8f2
>> eth0: XTemac: We renegotiated the speed to: 1000
>> eth0: XTemac: speed set to 1000Mb/s
>> eth0: XTemac: Send Threshold =3D 16, Receive Threshold =3D 2
>> eth0: XTemac: Send Wait bound =3D 1, Receive Wait bound =3D 1
>>
>> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H
192.168.119.1 -i 10,
>> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
>> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INE=
T to
192.168.11
>> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
>> Socket Message Elapsed Messages
>> Size Size Time Okay Errors Throughput
>> bytes bytes secs # # 10^6bits/sec
>>
>> 217088 32768 30.00 26391 0 230.60
>> 107520 30.00 26377 230.48
>>
>> Not that bad for this setup.
>>
>> With xps (EDK 9.1 without any SPs) I've tried creating exact the sam=
e
design.
>> So that exact the same kernel etc could be used for the both bitstre=
ams.
>> And with the self-made one I see exact the same output with the only=
difference:
>>
>> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H
192.168.119.1 -i 10,
>> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
>> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INE=
T to
192.168.11
>> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
>> Socket Message Elapsed Messages
>> Size Size Time Okay Errors Throughput
>> bytes bytes secs # # 10^6bits/sec
>>
>> 217088 32768 30.00 9593 0 83.82
>> 107520 30.00 9586 83.76
>>
>> :o(
>>
>> Rebuilding the XAPP941 project with EDK 9.1.00 still leaves 230Mb/se=
c.
>> I also pulled the additional timing constraints from XAPP941's
data/system.ucf
>> into the self-made data/system.ucf, but this haven't change anything=
.
>>
>> I've run out of ideas, and switched from TEMAC to other Xilinx stuff=
for
a while.
>>
>>
>> Thanks,
>> Andrei
>
> _________________________________________________________________
> Connect to the next generation of MSN Messenger
>
http://imagine-msn.com/messenger/launch80/default.aspx?locale=3Den-us&s=
ource=3Dwlmailtagline
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded
=
^ permalink raw reply [flat|nested] 9+ messages in thread
* RE: Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-21 16:05 Mohammad Sadegh Sadri
2007-06-21 18:18 ` Andrei Konovalov
0 siblings, 1 reply; 9+ messages in thread
From: Mohammad Sadegh Sadri @ 2007-06-21 16:05 UTC (permalink / raw)
To: Andrei Konovalov; +Cc: linuxppc-embedded
very interesting
the provided xps project from xapp941 results 230Mbits but your custome sys=
tem is 83.0Mbits, yes?
just, are you sure the ppc core has the same core clock frequency in both d=
esigns? I mean may be the xapp941 is running at 300MHz but yours at 100MHz.=
=20
And what about a custom system with 9.1SP2? does that gives the same perfor=
mance?
Finally If possible please release the test results for TCP streams. ( 9.1 =
no SP and 9.1 SP2 )
thanks
----------------------------------------
> Date: Thu, 21 Jun 2007 18:53:49 +0400
> From: akonovalov@ru.mvista.com
> To: mamsadegh@hotmail.com
> CC: linuxppc-embedded@ozlabs.org; kashiwagi@co-nss.co.jp; glenn.g.hart@us=
.westinghouse.com
> Subject: Re: Virtex 4 Avnet Mini-Module Networking.
>=20
> Mohammad Sadegh Sadri wrote:
> > I'm wondering where are the original developers of TEMAC driver these d=
ays, I have not seen any posts from any of andrei
> > nor grant in the recent weeks. may be they are involved in other projec=
ts.
>=20
> As for me - no, not until the end of this week..
>=20
> I am puzzled with the following:
>=20
> I've got ML405 recently. There is a prebuilt bitstream with TEMAC in SGDM=
A mode for this board made by Xilinx (XAPP941).
> With the TEMAC driver from EDK 9.1SP2 and the XAPP941 bitstream I've got =
(no jumbo frames):
>=20
> XTemac: using sgDMA mode.
> XTemac: using TxDRE mode
> XTemac: using RxDRE mode
> XTemac: buffer descriptor size: 32768 (0x8000)
> XTemac: (buffer_descriptor_init) phy: 0x3cf8000, virt: 0xff100000, siz=
e: 0x8000
> eth%d: XTemac: PHY detected at address 7.
> eth0: Xilinx TEMAC #0 at 0x80000000 mapped to 0xC5060000, irq=3D0
> eth0: XTemac id 1.0f, block id 5, type 8
>=20
> eth0: XTemac: Options: 0xb8f2
> eth0: XTemac: We renegotiated the speed to: 1000
> eth0: XTemac: speed set to 1000Mb/s
> eth0: XTemac: Send Threshold =3D 16, Receive Threshold =3D 2
> eth0: XTemac: Send Wait bound =3D 1, Receive Wait bound =3D 1
>=20
> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.11=
9.1 -i 10,
> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to =
192.168.11
> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> Socket Message Elapsed Messages
> Size Size Time Okay Errors Throughput
> bytes bytes secs # # 10^6bits/sec
>=20
> 217088 32768 30.00 26391 0 230.60
> 107520 30.00 26377 230.48
>=20
> Not that bad for this setup.
>=20
> With xps (EDK 9.1 without any SPs) I've tried creating exact the same des=
ign.
> So that exact the same kernel etc could be used for the both bitstreams.
> And with the self-made one I see exact the same output with the only diff=
erence:
>=20
> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.11=
9.1 -i 10,
> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to =
192.168.11
> 9.1 (192.168.119.1) port 0 AF_INET : =B15.0% @ 99% conf.
> Socket Message Elapsed Messages
> Size Size Time Okay Errors Throughput
> bytes bytes secs # # 10^6bits/sec
>=20
> 217088 32768 30.00 9593 0 83.82
> 107520 30.00 9586 83.76
>=20
> :o(
>=20
> Rebuilding the XAPP941 project with EDK 9.1.00 still leaves 230Mb/sec.
> I also pulled the additional timing constraints from XAPP941's data/syste=
m.ucf
> into the self-made data/system.ucf, but this haven't change anything.
>=20
> I've run out of ideas, and switched from TEMAC to other Xilinx stuff for =
a while.
>=20
>=20
> Thanks,
> Andrei
_________________________________________________________________
Connect to the next generation of MSN Messenger=A0
http://imagine-msn.com/messenger/launch80/default.aspx?locale=3Den-us&sourc=
e=3Dwlmailtagline=
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Virtex 4 Avnet Mini-Module Networking.
2007-06-21 16:05 Mohammad Sadegh Sadri
@ 2007-06-21 18:18 ` Andrei Konovalov
0 siblings, 0 replies; 9+ messages in thread
From: Andrei Konovalov @ 2007-06-21 18:18 UTC (permalink / raw)
To: Mohammad Sadegh Sadri; +Cc: linuxppc-embedded
Mohammad Sadegh Sadri wrote:
>
> very interesting
>
> the provided xps project from xapp941 results 230Mbits but your custome system is 83.0Mbits, yes?
yes
> just, are you sure the ppc core has the same core clock frequency in both designs? I mean may be the xapp941 is running at 300MHz but yours at 100MHz.
both run at 300MHz
> And what about a custom system with 9.1SP2? does that gives the same performance?
haven't tried this yet
> Finally If possible please release the test results for TCP streams. ( 9.1 no SP and 9.1 SP2 )
The TEMAC driver is from EDK 9.1.02 (IOW, SP2).
bitstream made by myself with EDK 9.1.00:
=========================================
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 217088 32768 30.00 41.66
bitstream from XAPP941:
=======================
root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10, 2 -I 99,10 -t TCP_STREAM -- -m 32768 -s 262144
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.119.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 217088 32768 30.00 111.86
And I didn't test ML405 receive performance.
Thanks,
Andrei
> thanks
>
>
>
> ----------------------------------------
>> Date: Thu, 21 Jun 2007 18:53:49 +0400
>> From: akonovalov@ru.mvista.com
>> To: mamsadegh@hotmail.com
>> CC: linuxppc-embedded@ozlabs.org; kashiwagi@co-nss.co.jp; glenn.g.hart@us.westinghouse.com
>> Subject: Re: Virtex 4 Avnet Mini-Module Networking.
>>
>> Mohammad Sadegh Sadri wrote:
>>> I'm wondering where are the original developers of TEMAC driver these days, I have not seen any posts from any of andrei
>>> nor grant in the recent weeks. may be they are involved in other projects.
>> As for me - no, not until the end of this week..
>>
>> I am puzzled with the following:
>>
>> I've got ML405 recently. There is a prebuilt bitstream with TEMAC in SGDMA mode for this board made by Xilinx (XAPP941).
>> With the TEMAC driver from EDK 9.1SP2 and the XAPP941 bitstream I've got (no jumbo frames):
>>
>> XTemac: using sgDMA mode.
>> XTemac: using TxDRE mode
>> XTemac: using RxDRE mode
>> XTemac: buffer descriptor size: 32768 (0x8000)
>> XTemac: (buffer_descriptor_init) phy: 0x3cf8000, virt: 0xff100000, size: 0x8000
>> eth%d: XTemac: PHY detected at address 7.
>> eth0: Xilinx TEMAC #0 at 0x80000000 mapped to 0xC5060000, irq=0
>> eth0: XTemac id 1.0f, block id 5, type 8
>>
>> eth0: XTemac: Options: 0xb8f2
>> eth0: XTemac: We renegotiated the speed to: 1000
>> eth0: XTemac: speed set to 1000Mb/s
>> eth0: XTemac: Send Threshold = 16, Receive Threshold = 2
>> eth0: XTemac: Send Wait bound = 1, Receive Wait bound = 1
>>
>> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10,
>> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
>> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.11
>> 9.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
>> Socket Message Elapsed Messages
>> Size Size Time Okay Errors Throughput
>> bytes bytes secs # # 10^6bits/sec
>>
>> 217088 32768 30.00 26391 0 230.60
>> 107520 30.00 26377 230.48
>>
>> Not that bad for this setup.
>>
>> With xps (EDK 9.1 without any SPs) I've tried creating exact the same design.
>> So that exact the same kernel etc could be used for the both bitstreams.
>> And with the self-made one I see exact the same output with the only difference:
>>
>> root@192.168.119.11:~/netperf-2.4.1/src# ./netperf -l 30 -H 192.168.119.1 -i 10,
>> 2 -I 99,10 -t UDP_STREAM -- -m 32768 -s 262144
>> UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.11
>> 9.1 (192.168.119.1) port 0 AF_INET : ±5.0% @ 99% conf.
>> Socket Message Elapsed Messages
>> Size Size Time Okay Errors Throughput
>> bytes bytes secs # # 10^6bits/sec
>>
>> 217088 32768 30.00 9593 0 83.82
>> 107520 30.00 9586 83.76
>>
>> :o(
>>
>> Rebuilding the XAPP941 project with EDK 9.1.00 still leaves 230Mb/sec.
>> I also pulled the additional timing constraints from XAPP941's data/system.ucf
>> into the self-made data/system.ucf, but this haven't change anything.
>>
>> I've run out of ideas, and switched from TEMAC to other Xilinx stuff for a while.
>>
>>
>> Thanks,
>> Andrei
>
> _________________________________________________________________
> Connect to the next generation of MSN Messenger
> http://imagine-msn.com/messenger/launch80/default.aspx?locale=en-us&source=wlmailtagline
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-19 18:06 Glenn.G.Hart
0 siblings, 0 replies; 9+ messages in thread
From: Glenn.G.Hart @ 2007-06-19 18:06 UTC (permalink / raw)
To: kashiwagi, linuxppc-embedded
[-- Attachment #1: Type: text/plain, Size: 3000 bytes --]
Yoshio Kashiwagi,
Yes the checksum off loading is turned off. Should it be turned on? My
documentation says I need to use DMA mode to use checksum offloading.
Glenn
(Embedded Yoshio Kashiwagi <kashiwagi@co-nss.co.jp>@ozlabs.org
image moved 06/19/2007 11:51 AM
to file:
pic31426.pcx)
Sent by:
linuxppc-embedded-bounces+glenn.g.hart=us.westinghouse.com@ozlabs.org
To: Glenn.G.Hart@us.westinghouse.com, linuxppc-embedded@ozlabs.org
cc:
Subject: Re: Virtex 4 Avnet Mini-Module Networking.
Security Level:? Internal
Glenn-san,
The PLB TEMAC driver is supporting checksum off loading (checksum
calculation by hardware).
If the same kernel source is used, isn't the hardware checksum
of PLB TEMAC disabled by the MHS file of Mini-Module?
BEGIN plb_temac
PARAMETER INSTANCE = plb_temac_0
PARAMETER HW_VER = 3.00.a
PARAMETER C_PLB_CLK_PERIOD_PS = 10000
PARAMETER C_TXFIFO_DEPTH = 131072
PARAMETER C_RXFIFO_DEPTH = 131072
PARAMETER C_MAC_FIFO_DEPTH = 64
PARAMETER C_BASEADDR = 0x80000000
PARAMETER C_HIGHADDR = 0x8000ffff
PARAMETER C_INCLUDE_RX_CSUM = 1 <--- H/W checksum enable
PARAMETER C_INCLUDE_TX_CSUM = 1 <--- H/W checksum enable
PARAMETER C_RX_DRE_TYPE = 2
PARAMETER C_TX_DRE_TYPE = 2
Yoshio Kashiwagi - Nissin Systems
> I am having trouble using the network on the Avent Mini-Module. I had
the
> network working on the ML403 board with Linux 2.6.19.
>
> The kernel comes up fine with the Mini-Module and the network seems to
work
> until I try to use it. I can configured the ethernet with ifconfig
and can
> even ping another machine on the network.
>
> The problem is with the TCP layer. It seems almost everytime the
> Mini-Module sends TCP information the checksum is reported an wrong
(via
> ethereal). The only thing that comes to mind is the ML403 uses a
Marvell
> PHY and the Mini-Module has a Broadcom PHY. I do not know enough
about
> low-level networking to know where the problem is. Is the checksum
> generated in the PHY? Could something be wrong in Linux programming
the
> Broadcom PHY and not the Marvell PHY? Why is it only the TCP
information
> (the IP information has the correct checksum)?
>
> Thanks for any information you can offer.
>
> Glenn
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded
[-- Attachment #2: pic31426.pcx --]
[-- Type: application/octet-stream, Size: 128 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread* Virtex 4 Avnet Mini-Module Networking.
@ 2007-06-19 15:21 Glenn.G.Hart
2007-06-19 15:51 ` Yoshio Kashiwagi
0 siblings, 1 reply; 9+ messages in thread
From: Glenn.G.Hart @ 2007-06-19 15:21 UTC (permalink / raw)
To: linuxppc-embedded
I am having trouble using the network on the Avent Mini-Module. I had the
network working on the ML403 board with Linux 2.6.19.
The kernel comes up fine with the Mini-Module and the network seems to work
until I try to use it. I can configured the ethernet with ifconfig and can
even ping another machine on the network.
The problem is with the TCP layer. It seems almost everytime the
Mini-Module sends TCP information the checksum is reported an wrong (via
ethereal). The only thing that comes to mind is the ML403 uses a Marvell
PHY and the Mini-Module has a Broadcom PHY. I do not know enough about
low-level networking to know where the problem is. Is the checksum
generated in the PHY? Could something be wrong in Linux programming the
Broadcom PHY and not the Marvell PHY? Why is it only the TCP information
(the IP information has the correct checksum)?
Thanks for any information you can offer.
Glenn
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Virtex 4 Avnet Mini-Module Networking.
2007-06-19 15:21 Glenn.G.Hart
@ 2007-06-19 15:51 ` Yoshio Kashiwagi
0 siblings, 0 replies; 9+ messages in thread
From: Yoshio Kashiwagi @ 2007-06-19 15:51 UTC (permalink / raw)
To: Glenn.G.Hart, linuxppc-embedded
Glenn-san,
The PLB TEMAC driver is supporting checksum off loading (checksum
calculation by hardware).
If the same kernel source is used, isn't the hardware checksum
of PLB TEMAC disabled by the MHS file of Mini-Module?
BEGIN plb_temac
PARAMETER INSTANCE = plb_temac_0
PARAMETER HW_VER = 3.00.a
PARAMETER C_PLB_CLK_PERIOD_PS = 10000
PARAMETER C_TXFIFO_DEPTH = 131072
PARAMETER C_RXFIFO_DEPTH = 131072
PARAMETER C_MAC_FIFO_DEPTH = 64
PARAMETER C_BASEADDR = 0x80000000
PARAMETER C_HIGHADDR = 0x8000ffff
PARAMETER C_INCLUDE_RX_CSUM = 1 <--- H/W checksum enable
PARAMETER C_INCLUDE_TX_CSUM = 1 <--- H/W checksum enable
PARAMETER C_RX_DRE_TYPE = 2
PARAMETER C_TX_DRE_TYPE = 2
Yoshio Kashiwagi - Nissin Systems
> I am having trouble using the network on the Avent Mini-Module. I had
the
> network working on the ML403 board with Linux 2.6.19.
>
> The kernel comes up fine with the Mini-Module and the network seems to
work
> until I try to use it. I can configured the ethernet with ifconfig
and can
> even ping another machine on the network.
>
> The problem is with the TCP layer. It seems almost everytime the
> Mini-Module sends TCP information the checksum is reported an wrong
(via
> ethereal). The only thing that comes to mind is the ML403 uses a
Marvell
> PHY and the Mini-Module has a Broadcom PHY. I do not know enough
about
> low-level networking to know where the problem is. Is the checksum
> generated in the PHY? Could something be wrong in Linux programming
the
> Broadcom PHY and not the Marvell PHY? Why is it only the TCP
information
> (the IP information has the correct checksum)?
>
> Thanks for any information you can offer.
>
> Glenn
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2007-06-22 21:21 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-06-19 21:04 Virtex 4 Avnet Mini-Module Networking Mohammad Sadegh Sadri
2007-06-21 14:53 ` Andrei Konovalov
-- strict thread matches above, loose matches on Subject: below --
2007-06-22 21:21 Mohammad Sadegh Sadri
2007-06-22 19:02 Glenn.G.Hart
2007-06-21 16:05 Mohammad Sadegh Sadri
2007-06-21 18:18 ` Andrei Konovalov
2007-06-19 18:06 Glenn.G.Hart
2007-06-19 15:21 Glenn.G.Hart
2007-06-19 15:51 ` Yoshio Kashiwagi
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).