[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[G.David]

It is interesting that you bring this up...

>Hello 9fans!
>
>FAQ says in comp.os.plan9:
>>Ethernet Adapters
>>	3Com 3C509, 3C509B are recommended. The EISA 3C579 works, but
>>	isn't worth the extra cost. The PCMCIA 3C589, PCI 3C590 and PCI
>>!	3C595 (fast ethernet) also work. AMD 79C970 based adapters seem to
>>	work fine. SMC (WD) series up to the Elite (and the Elite Ultra),
>>	some NE2000 compatibles (including an NE4100 PCMCIA card) and one
>>	Eagle NE3210 EISA card. The 3Com 3C503 does not work at all under
>>!	load. The 3Com 3C595 is not supported.
>Does 3C595 work or not?
>
>According to 3COME Home Page
>the Fast EtherLink that 3COM currentlly sails are:
>PCI: 3C905-TX, 3C905-T4
>EISA: 3C597-TX
>ISA: 3C515-TX

I have a 3c509, 3c579, 3c597 and 3c905.  The ISA card works, except
under heavy receive load with an adaptec 1542 bus-master the small
receive FIFO will drive you crazy!

But the 32bit cards don't work worth a sh*t!  After spending too
much time going through Plan9 and Net/FreeBSD source I have determined
these cards go into stupid mode on transmit when presented with
bursts of receive data.  In other words, if you rip a lot of data
off the card or force a lot into the card, it works great.  But just
try a single compile session, which does full 8k reads and writes,
and the cards go stupid.  A cp of a ~large file in a directory will
do it to.

They will not accept any more data into the transmit FIFO, will not
transmit what is in the FIFO and will not give you a TxAvail interrupt
when the FIFO finally does drain!

I don't have any documentation to the cards, so I'm in the dark.
Any body want to help?

David Butler
gdb@dbSystems.com

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[jim]

since everyone else stepped back leaving him in front, i've given
"G. David Butler" <dbSystems.com!gdb> the current driver for retrofitting
to plan9.

--jim

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[G.David]

>no, that's an errata for the 3c59x cards. the important erratum for the
>3c905 concerns it hanging if you do busmaster receives at 10Mb.

I'm using 10Mb.  humm...

>you do not need to set anything in the pci config space for the 3c905,
>the card and bios should have set it correctly (they do on every system
>i've tried).

Good to know.

>if there's interest and a volunteer i can give them the current brazil
>driver provided they promise to give back a plan9 version for inclusion
>in the website updates. it will be a bit of work:
>    % ls -l /sys/lib/pcdist/src/9/pc/ether509.c /sys/src/brazil/pc/etherelnk3.c
>    --rw-rw-r-- M 4 jmk sys 16762 Feb 27  1996 /sys/lib/pcdist/src/9/pc/ether509.c
>    --rw-rw-r-- M 4 jmk sys 41136 Dec  7 00:22 /sys/src/brazil/pc/etherelnk3.c
>    % 

I'm game!  My first priority is the fileserver kernel, but the
cpu/terminal kernel will be close behind...

David

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[jim]

	>                                           in busmastering mode there
	>are chip errata which have to be dealt with or it can hang up too.

	You mean like you have to set the latency timer to 0xFF?

no, that's an errata for the 3c59x cards. the important erratum for the
3c905 concerns it hanging if you do busmaster receives at 10Mb.

you do not need to set anything in the pci config space for the 3c905,
the card and bios should have set it correctly (they do on every system
i've tried).

if there's interest and a volunteer i can give them the current brazil
driver provided they promise to give back a plan9 version for inclusion
in the website updates. it will be a bit of work:
    % ls -l /sys/lib/pcdist/src/9/pc/ether509.c /sys/src/brazil/pc/etherelnk3.c
    --rw-rw-r-- M 4 jmk sys 16762 Feb 27  1996 /sys/lib/pcdist/src/9/pc/ether509.c
    --rw-rw-r-- M 4 jmk sys 41136 Dec  7 00:22 /sys/src/brazil/pc/etherelnk3.c
    % 

--jim

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[G.David]

>>as i mentioned in a previous message i still have problems with the
>>905 so i'm interested in any other experiences with it.
>
>I'm trying to get it to work reliably too.  I don't have the
>right documentation for PCI to know how to set the busmaster bit
>or to set the latency timer as specified by the errata.
>Can you help here?

I think I have that part down now.  After GlobalReset I do a
pcicfgr, set busMaster in command, set latency to 0xff and
do pcicfgw to 0x04 for a length of 12.

Now I get MasterStatus flag masterAbort at every transfer.

For the fun of it, I tried to use the DMA at transmit as well
as receive.  First I PIOd enough data to keep the transmitter
from underrunning then turned on the DMA.  Next, to make sure
the DMA code was sane, I memmoved the packet to a local buffer
with the preamble in front and DMAd the whole thing.  I always
get masterAbort!  I think I would only use the DMA on transmit
for gigabit ethernet though.  Is it possible to get underflow
with PIO at 100mb with current hardware?

So, any ideas about the abort?

David Butler
gdb@dbSystems.com

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[G.David]

>which of the above cards are you classifying as 32-bit?
>the 579 and 597 are 32-bit (eisa).  we have 579s and they work fine.
>the 597 probably won't work without dealing with various chip errata.

I didn't have the piece of information that "Large packet support
affects bit-field widths in the following registers: RxStatus,
MasterStatus, TxStartThresh, TxAvailableThresh and RxEarlyThresh".
My 579s now work fine, too.

>the 905 is problematic: although it has compatibility with the older
>cards the fifos are tiny and the card really has to be run in
>busmastering mode or it gums up a lot (it works well enough in fifo
>mode to boot a kernel with b.com though).

Mine seems to work in PIO mode.  I don't see any receive overruns
when pushed and I can easily keep the transmit FIFO full.

>                                           in busmastering mode there
>are chip errata which have to be dealt with or it can hang up too.

You mean like you have to set the latency timer to 0xFF?

>as i mentioned in a previous message i still have problems with the
>905 so i'm interested in any other experiences with it.

I'm trying to get it to work reliably too.  I don't have the
right documentation for PCI to know how to set the busmaster bit
or to set the latency timer as specified by the errata.
Can you help here?

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[forsyth]

>>to fetch the boddles to check.  i've a feeling that the kernel's
>>driver has got the change but b.com's hasn't.

unless i picked the wrong boddles, neither ether509 driver has
got the txashift change.

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[forsyth]

oh.  when i said that my 3com XL `works fine', i forgot to add
that it's in dull, antique 10m/bit mode, not 100 m/bit mode (we've only
just got the 100m/bit equipment, and i'm not yet connected
to it).  i won't even think about g/bit.

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[forsyth]

i've been using a 3com XL quite happily for months now.
it will hang up if you haven't got the txashift change posted
here some time ago.  the newer card has got a bigger fifo,
and the representation of the TxAvailable value differs
(it's in words, not bytes, or some such thing).

b.com needs a similar change.

i wasn't sure whether the plan9.bell-labs.com updates had
the change, but i thought they had.  i haven't had a chance
to fetch the boddles to check.  i've a feeling that the kernel's
driver has got the change but b.com's hasn't.

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[jim]

	>It is interesting that you bring this up...
	>
	>>Hello 9fans!
	>>
	>>FAQ says in comp.os.plan9:
	>>>Ethernet Adapters
	>>>	3Com 3C509, 3C509B are recommended. The EISA 3C579 works, but
	>>>	isn't worth the extra cost. The PCMCIA 3C589, PCI 3C590 and PCI
	>>>!	3C595 (fast ethernet) also work. AMD 79C970 based adapters seem to
	>>>	work fine. SMC (WD) series up to the Elite (and the Elite Ultra),
	>>>	some NE2000 compatibles (including an NE4100 PCMCIA card) and one
	>>>	Eagle NE3210 EISA card. The 3Com 3C503 does not work at all under
	>>>!	load. The 3Com 3C595 is not supported.
	>>Does 3C595 work or not?
	>>
	>>According to 3COME Home Page
	>>the Fast EtherLink that 3COM currentlly sails are:
	>>PCI: 3C905-TX, 3C905-T4
	>>EISA: 3C597-TX
	>>ISA: 3C515-TX
	>
	>I have a 3c509, 3c579, 3c597 and 3c905.  The ISA card works, except
	>under heavy receive load with an adaptec 1542 bus-master the small
	>receive FIFO will drive you crazy!
	>
	>But the 32bit cards don't work worth a sh*t!  After spending too
	>much time going through Plan9 and Net/FreeBSD source I have determined
	>these cards go into stupid mode on transmit when presented with
	>bursts of receive data.  In other words, if you rip a lot of data
	>off the card or force a lot into the card, it works great.  But just
	>try a single compile session, which does full 8k reads and writes,
	>and the cards go stupid.  A cp of a ~large file in a directory will
	>do it to.
	>
	>They will not accept any more data into the transmit FIFO, will not
	>transmit what is in the FIFO and will not give you a TxAvail interrupt
	>when the FIFO finally does drain!
	>
	>I don't have any documentation to the cards, so I'm in the dark.
	>Any body want to help?
	>
	>David Butler
	>gdb@dbSystems.com

which of the above cards are you classifying as 32-bit? the 579 and 597 are 
32-bit (eisa). we have 579s and they work fine. the 597 probably won't work without
dealing with various chip errata.

the 905 is problematic: although it has compatibility with the older cards the
fifos are tiny and the card really has to be run in busmastering mode or it gums
up a lot (it works well enough in fifo mode to boot a kernel with b.com though).
in busmastering mode there are chip errata which have to be dealt with or it can
hang up too.

as i mentioned in a previous message i still have problems with the 905 so i'm
interested in any other experiences with it.

--jim

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[jim]

	> Does 3C595 work or not?

the 3c595 works if all the updates are applied.
of course, it's an old and discontinued card, replaced by the 3c905.
however, you should be able to find one without too much difficulty.

the 3c905 is a full busmastering card and has some backwards compatibility
with the older 3c5xx series, e.g. it will boot with the current drivers,
but the performance is abysmal because it's really designed to be used
as a busmaster. i have a driver but the there are problems under load which
i can't seem to resolve and i can't get the problems to happen in a test
environment.

--jim

[9fans] Re: Plan 9 from Bell Labs - Frequently Asked

[Kenji]

Hello 9fans!

FAQ says in comp.os.plan9:
>Ethernet Adapters
>	3Com 3C509, 3C509B are recommended. The EISA 3C579 works, but
>	isn't worth the extra cost. The PCMCIA 3C589, PCI 3C590 and PCI
>!	3C595 (fast ethernet) also work. AMD 79C970 based adapters seem to
>	work fine. SMC (WD) series up to the Elite (and the Elite Ultra),
>	some NE2000 compatibles (including an NE4100 PCMCIA card) and one
>	Eagle NE3210 EISA card. The 3Com 3C503 does not work at all under
>!	load. The 3Com 3C595 is not supported.
Does 3C595 work or not?

According to 3COME Home Page
the Fast EtherLink that 3COM currentlly sails are:
PCI: 3C905-TX, 3C905-T4
EISA: 3C597-TX
ISA: 3C515-TX

Misprint is not ?

Kenji Arisawa
E-mail: arisawa@aichi-u.ac.jp