Plan for e100-e1000 in mainline

Plan for e100-e1000 in mainline

[J.A. Magallon]

Hi.

Well, subject says it all. Which is the status/plans for inclussion
of those drivers in mainline kernel ? AFAIR, e1000 had been licensed,
but e100 was not clear yet.

Any ideas ?

-- 
J.A. Magallon                           #  Let the source be with you...        
mailto:jamagallon@able.es
Mandrake Linux release 8.3 (Cooker) for i586
Linux werewolf 2.4.19-pre7-jam8 #2 SMP mié may 1 02:29:38 CEST 2002 i686

Re: Plan for e100-e1000 in mainline

[Jeff Garzik]

J.A. Magallon wrote:

>Hi.
>
>Well, subject says it all. Which is the status/plans for inclussion
>of those drivers in mainline kernel ? AFAIR, e1000 had been licensed,
>but e100 was not clear yet.
>

e100 has been in 2.5.x for quite a long time.  All license issues have 
similarly been resolved a long time ago.

I expect Intel's Q/A to green light their current driver.  With a few 
patches it should be ready for 2.4.x soon.

You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely 
compiles without modification.

    Jeff

Re: Plan for e100-e1000 in mainline

[Roy Sigurd Karlsbakk]

> I expect Intel's Q/A to green light their current driver.  With a few
> patches it should be ready for 2.4.x soon.
> 
> You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely
> compiles without modification.

Will this be done in 2.4, or is it a 2.[56]-only driver?

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: Plan for e100-e1000 in mainline

[Steffen Persvold]

On Wed, 1 May 2002, Jeff Garzik wrote:

> J.A. Magallon wrote:
>
> >Hi.
> >
> >Well, subject says it all. Which is the status/plans for inclussion
> >of those drivers in mainline kernel ? AFAIR, e1000 had been licensed,
> >but e100 was not clear yet.
> >
>
> e100 has been in 2.5.x for quite a long time.  All license issues have
> similarly been resolved a long time ago.
>
> I expect Intel's Q/A to green light their current driver.  With a few
> patches it should be ready for 2.4.x soon.
>
> You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely
> compiles without modification.
>

Has the latency issues with e100 compared to eepro100 been resolved too ?
Last time I checked the e100 driver (version 1.8.38) had terrible high
latency on small messages even in a back-to-back configuration (both
machines using the same driver). As a test I used netperf's TCP_RR
benchmark (TCP roundtrip) :

Message size  | eepro100 latency (usec) | e100 latency (usec)
--------------------------------------------------------------
     16                 108.985                 114.116
     24                 110.940                 116.679
     32                 114.129                 119.608
     48                 118.614                 125.507
     64                 124.341                1093.150
     96                 135.864                1102.190
    128                 147.177                1102.726
    192                 167.343                 838.799
    256                 188.410                 858.572
    384                 230.797                 900.941
    512                 271.281                 941.104
    768                 354.270                1024.753
   1024                 437.307                1108.231
   1536                 612.688                1293.912
   2048                 699.168                1334.828
   3072                 872.554                1670.787
   4096                1032.183                1556.056


As you can see, with a TCP payload of 64 bytes the latency is quite high
with the e100 driver compared to the eepro100 driver. As a side note: it
didn't work well with the bonding module either (neither did the e1000
module).

Another funny thing is that the latency for the gigabit adapter (e1000)
is also higher than fast ethernet (eepro100) with small messages (<256
bytes) :

Message size  | eepro100 latency (usec) | e1000 latency (usec)
--------------------------------------------------------------
     16                 108.985                 177.138
     24                 110.940                 177.370
     32                 114.129                 177.523
     48                 118.614                 178.189
     64                 124.341                 178.769
     96                 135.864                 179.669
    128                 147.177                 180.517
    192                 167.343                 183.339
    256                 188.410                 184.552
    384                 230.797                 188.929
    512                 271.281                 191.787
    768                 354.270                 198.745
   1024                 437.307                 205.793
   1536                 612.688                 240.772
   2048                 699.168                 249.540
   3072                 872.554                 282.348
   4096                1032.183                 299.220

Regards,
 --
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:sp@scali.com |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency

RE: Plan for e100-e1000 in mainline

[Leech, Christopher]

> Another funny thing is that the latency for the gigabit 
> adapter (e1000) is also higher than fast ethernet (eepro100) 
> with small messages (<256 bytes) :

You could try setting the RxIntDelay module parameter to 0, that should
improve round trip latency.  The balance between latency on the receive path
and interrupt rate can be difficult to manage, hopefully a dynamic method
like NAPI will result in Ethernet drivers that need less hand tuning.

--
Chris Leech <christopher.leech@intel.com>
Network Software Engineer
LAN Access Division, Intel Corporation

[PATCH] intel eths for 2.4 [was: Plan for e100-e1000 in mainline]

[J.A. Magallon]

On 2002.05.01 Jeff Garzik wrote:
>J.A. Magallon wrote:
>
>>Hi.
>>
>>Well, subject says it all. Which is the status/plans for inclussion
>>of those drivers in mainline kernel ? AFAIR, e1000 had been licensed,
>>but e100 was not clear yet.
>>
>
>e100 has been in 2.5.x for quite a long time.  All license issues have 
>similarly been resolved a long time ago.
>
>I expect Intel's Q/A to green light their current driver.  With a few 
>patches it should be ready for 2.4.x soon.
>
>You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely 
>compiles without modification.
>

I did it, taking drivers from 2.5.12, and at least it compiles.
I have to try in the real box, but I don't think there were any problems,
at least the same than 2.5....

Marcelo, is there any chance to get this in next -pre or in .19 ?
Patches are big, so I put them at:

http://giga.cps.unizar.es/~magallon/linux/kernel/intel/e100-2.0.27-pre3.bz2
http://giga.cps.unizar.es/~magallon/linux/kernel/intel/e1000-4.2.8.bz2

(note, second gives some offsets and a fail if applied without the first
one...)

Some notes/questions:

- e100 has no help text.
- Versions are far newer than those listen in intel web pages:
  * e100: Intel web is 1.8.35, the driver taken from 2.5.12 is 2.0.27-pre3
  * e1000: Intel web is 4.1.7, driver in 2.5 is 4.2.8

Thanks for all the info.

-- 
J.A. Magallon                           #  Let the source be with you...        
mailto:jamagallon@able.es
Mandrake Linux release 8.3 (Cooker) for i586
Linux werewolf 2.4.19-pre7-jam9 #2 SMP mié may 1 12:09:38 CEST 2002 i686

Re: [PATCH] intel eths for 2.4 [was: Plan for e100-e1000 in mainline]

[Jeff Garzik]

J.A. Magallon wrote:

>On 2002.05.01 Jeff Garzik wrote:
>
>>I expect Intel's Q/A to green light their current driver.  With a few 
>>patches it should be ready for 2.4.x soon.
>>

>I did it, taking drivers from 2.5.12, and at least it compiles.
>I have to try in the real box, but I don't think there were any problems,
>at least the same than 2.5....
>
>Marcelo, is there any chance to get this in next -pre or in .19 ?
>

When they are suitable for Marcelo, I'm going to send them to Marcelo.

As I wrote in the quoted message, they need some more patches, and I'm 
also interested in feedback from Intel Q/A (which is scheduled for 
sometime this week).

If you are interesting in maintaining 2.4.x patches for a short time, go 
for it.  But I would rather not have a almost-ready e100 go to Marcelo 
and get released in 2.4.19 in incomplete form.  It's out there, it's 
public, let's leave at that for a little while.

    Jeff

RE: Plan for e100-e1000 in mainline

[Steffen Persvold]

On Wed, 1 May 2002, Leech, Christopher wrote:

>
> > Another funny thing is that the latency for the gigabit
> > adapter (e1000) is also higher than fast ethernet (eepro100)
> > with small messages (<256 bytes) :
>
> You could try setting the RxIntDelay module parameter to 0, that should
> improve round trip latency.  The balance between latency on the receive path
> and interrupt rate can be difficult to manage, hopefully a dynamic method
> like NAPI will result in Ethernet drivers that need less hand tuning.
>

Hmm the docs say that RxIntDelay is by default 0 but when I tried it, it
helped for sure (although the numbers are still a bit higher that with
eepro100).

Thanks,
-- 
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:sp@scali.com |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency

Re: [PATCH] intel eths for 2.4 [was: Plan for e100-e1000 in mainline]

[J.A. Magallon]

On 2002.05.02 Jeff Garzik wrote:
>J.A. Magallon wrote:
>
>>On 2002.05.01 Jeff Garzik wrote:
>>
>>>I expect Intel's Q/A to green light their current driver.  With a few 
>>>patches it should be ready for 2.4.x soon.
>>>
>
>>I did it, taking drivers from 2.5.12, and at least it compiles.
>>I have to try in the real box, but I don't think there were any problems,
>>at least the same than 2.5....
>>
>>Marcelo, is there any chance to get this in next -pre or in .19 ?
>>
>
>When they are suitable for Marcelo, I'm going to send them to Marcelo.
>
>As I wrote in the quoted message, they need some more patches, and I'm 
>also interested in feedback from Intel Q/A (which is scheduled for 
>sometime this week).
>

Oops, sorry, I misunderstood something.

>If you are interesting in maintaining 2.4.x patches for a short time, go 
>for it.  But I would rather not have a almost-ready e100 go to Marcelo 
>and get released in 2.4.19 in incomplete form.  It's out there, it's 
>public, let's leave at that for a little while.
>

OK, just forget all about -pre and .19.

If somebody wants to try them, look at one other post about -jam9.
It is the same story, I need them for some new boxen (e1000) and I prefer
to have all in the same tree than apart, so took those from 2.5
instead of Intel's.

-- 
J.A. Magallon                           #  Let the source be with you...        
mailto:jamagallon@able.es
Mandrake Linux release 8.3 (Cooker) for i586
Linux werewolf 2.4.19-pre7-jam9 #2 SMP mié may 1 12:09:38 CEST 2002 i686

Re: Plan for e100-e1000 in mainline

[Jamie Lokier]

Jeff Garzik wrote:
> You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely 
> compiles without modification.

It does, except that you need to
  #define cpu_relax() rep_nop()
or something very similar.

Fwiw, the e100 driver from 2.5, running on 2.4, can forward at a
somewhat higher packet rate than the e100 which Red Hat 7.2 provides
with 2.4...

-- Jamie

Re: Plan for e100-e1000 in mainline

[J.A. Magallon]

On 2002.05.06 Jamie Lokier wrote:
>Jeff Garzik wrote:
>> You can easily copy drivers/net/e100[0] into a 2.4.x kernel, it likely 
>> compiles without modification.
>
>It does, except that you need to
>  #define cpu_relax() rep_nop()
>or something very similar.
>

It is already defined in processor.h (2.4.19-pre8). They build fine for me.

-- 
J.A. Magallon                           #  Let the source be with you...        
mailto:jamagallon@able.es
Mandrake Linux release 8.3 (Cooker) for i586
Linux werewolf 2.4.19-pre8-jam1 #1 SMP dom may 5 23:46:04 CEST 2002 i686