NAPI 8139too.c for 2.4.23

NAPI 8139too.c for 2.4.23

[Octave]

Stephen,
I get your patch from http://lwn.net/Articles/54815/ for 2.6.X and 
I rewrote it for 2.4.23. Tested with 2.4.23 on high load servers. I 
have no more "Too much work at interrupt".                                     

I dropped it on ftp://ftp.ovh.net/made-in-ovh/8139too.c-2.4-0.9.27

Hope it helps.
Octave

before:  
-------  
# ps auxw
root       256  0.0  0.0     0    0 ?        SW   Nov28   0:00 [eth0]
# ifconfig
          RX packets:40940899 errors:250542 dropped:7052 overruns:250542 frame:0
          TX packets:33057049 errors:0 dropped:0 overruns:20 carrier:0
# dmesg
eth0: Setting 100mbps full-duplex based on auto-negotiated partner ability 41e1.
nfs: server X.X.X.X not responding, still trying
nfs: server X.X.X.X OK
eth0: Too much work at interrupt, IntrStatus=0x0040.

with NAPI
---------
          RX packets:428253 errors:0 dropped:0 overruns:0 frame:0
          TX packets:357949 errors:0 dropped:0 overruns:0 carrier:0
8139too Fast Ethernet driver 0.9.27
PCI: Found IRQ 11 for device 00:0b.0
eth0: RealTek RTL8139 at 0xec00, 00:e0:4c:91:03:b0, IRQ 11
eth0:  Identified 8139 chip type 'RTL-8100B/8139D'

Re: NAPI 8139too.c for 2.4.23

[Jeff Garzik]

On Mon, Dec 01, 2003 at 09:50:38PM +0100, Octave wrote:
> Stephen,
> I get your patch from http://lwn.net/Articles/54815/ for 2.6.X and 
> I rewrote it for 2.4.23. Tested with 2.4.23 on high load servers. I 
> have no more "Too much work at interrupt".                                     
> 
> I dropped it on ftp://ftp.ovh.net/made-in-ovh/8139too.c-2.4-0.9.27
> 
> Hope it helps.
> Octave

Very cool!  Thanks for testing.

Is there any chance you could do some benchmark runs with ttcp or
somesuch?

	Jeff

Re: NAPI 8139too.c for 2.4.23

[Marcello]

Octave ha scritto:

>Stephen,
>I get your patch from http://lwn.net/Articles/54815/ for 2.6.X and 
>I rewrote it for 2.4.23. Tested with 2.4.23 on high load servers. I 
>have no more "Too much work at interrupt".                                     
>
>I dropped it on ftp://ftp.ovh.net/made-in-ovh/8139too.c-2.4-0.9.27
>
>Hope it helps.
>Octave
>
>before:  
>-------  
># ps auxw
>root       256  0.0  0.0     0    0 ?        SW   Nov28   0:00 [eth0]
># ifconfig
>          RX packets:40940899 errors:250542 dropped:7052 overruns:250542 frame:0
>          TX packets:33057049 errors:0 dropped:0 overruns:20 carrier:0
># dmesg
>eth0: Setting 100mbps full-duplex based on auto-negotiated partner ability 41e1.
>nfs: server X.X.X.X not responding, still trying
>nfs: server X.X.X.X OK
>eth0: Too much work at interrupt, IntrStatus=0x0040.
>
>with NAPI
>---------
>          RX packets:428253 errors:0 dropped:0 overruns:0 frame:0
>          TX packets:357949 errors:0 dropped:0 overruns:0 carrier:0
>8139too Fast Ethernet driver 0.9.27
>PCI: Found IRQ 11 for device 00:0b.0
>eth0: RealTek RTL8139 at 0xec00, 00:e0:4c:91:03:b0, IRQ 11
>eth0:  Identified 8139 chip type 'RTL-8100B/8139D'
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-net" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>  
>
I try immediatly your variant of the driver :)

Bye

Marcello

Re: NAPI 8139too.c for 2.4.23

[Marcello]

Octave ha scritto:

>Stephen,
>I get your patch from http://lwn.net/Articles/54815/ for 2.6.X and 
>I rewrote it for 2.4.23. Tested with 2.4.23 on high load servers. I 
>have no more "Too much work at interrupt".                                     
>
>I dropped it on ftp://ftp.ovh.net/made-in-ovh/8139too.c-2.4-0.9.27
>
>Hope it helps.
>Octave
>
>before:  
>-------  
># ps auxw
>root       256  0.0  0.0     0    0 ?        SW   Nov28   0:00 [eth0]
># ifconfig
>          RX packets:40940899 errors:250542 dropped:7052 overruns:250542 frame:0
>          TX packets:33057049 errors:0 dropped:0 overruns:20 carrier:0
># dmesg
>eth0: Setting 100mbps full-duplex based on auto-negotiated partner ability 41e1.
>nfs: server X.X.X.X not responding, still trying
>nfs: server X.X.X.X OK
>eth0: Too much work at interrupt, IntrStatus=0x0040.
>
>with NAPI
>---------
>          RX packets:428253 errors:0 dropped:0 overruns:0 frame:0
>          TX packets:357949 errors:0 dropped:0 overruns:0 carrier:0
>8139too Fast Ethernet driver 0.9.27
>PCI: Found IRQ 11 for device 00:0b.0
>eth0: RealTek RTL8139 at 0xec00, 00:e0:4c:91:03:b0, IRQ 11
>eth0:  Identified 8139 chip type 'RTL-8100B/8139D'
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-net" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>  
>
OCTAVE!!

Your rewrite of the driver is great!! You have resolved all my problems 
with ethernet!! (collisions for now , but I think errors too :) )

You are my new personal HERO!! :D

Thanks

Marcello

Re: NAPI 8139too.c for 2.4.23

[Marcello]

Octave ha scritto:

>Stephen,
>I get your patch from http://lwn.net/Articles/54815/ for 2.6.X and 
>I rewrote it for 2.4.23. Tested with 2.4.23 on high load servers. I 
>have no more "Too much work at interrupt".                                     
>
>I dropped it on ftp://ftp.ovh.net/made-in-ovh/8139too.c-2.4-0.9.27
>
>Hope it helps.
>Octave
>
>before:  
>-------  
># ps auxw
>root       256  0.0  0.0     0    0 ?        SW   Nov28   0:00 [eth0]
># ifconfig
>          RX packets:40940899 errors:250542 dropped:7052 overruns:250542 frame:0
>          TX packets:33057049 errors:0 dropped:0 overruns:20 carrier:0
># dmesg
>eth0: Setting 100mbps full-duplex based on auto-negotiated partner ability 41e1.
>nfs: server X.X.X.X not responding, still trying
>nfs: server X.X.X.X OK
>eth0: Too much work at interrupt, IntrStatus=0x0040.
>
>with NAPI
>---------
>          RX packets:428253 errors:0 dropped:0 overruns:0 frame:0
>          TX packets:357949 errors:0 dropped:0 overruns:0 carrier:0
>8139too Fast Ethernet driver 0.9.27
>PCI: Found IRQ 11 for device 00:0b.0
>eth0: RealTek RTL8139 at 0xec00, 00:e0:4c:91:03:b0, IRQ 11
>eth0:  Identified 8139 chip type 'RTL-8100B/8139D'
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-net" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>  
>
Sorry , my happiness was a bit premature :D

This doesn't recognize with autonegotiation the Speed correctly .

My ethernet device is connected to an ethernet ADSL Router , speed 
10Mbit/HD , but the autonegotiation recongnize 10Mbit/FD , and this 
makes the connection very slow , because of the very much errors , but 
the strange thing is that ifconfig doesn't reports any error (but , of 
course, I know that there are as with the previous driver ) , for 
ifconfig all is OK , but my connection is veeeeery slow .

Forcing the 10MBps/HD with MII-TOOL makes my connection the same as ever .

Bye

Marcello

Re: NAPI 8139too.c for 2.4.23

[Octave]

> Is there any chance you could do some benchmark runs with ttcp or
> somesuch?

I tested on 6-7 servers running with 
eepro eth0: Intel Corp. 82557/8/9 [Ethernet Pro 100], 00:E0:18:01:78:6C, IRQ 10.
realtek 8139too with NAPI 8139too Fast Ethernet driver 0.9.27 
realtek 8139too with no NAPI (standard driver with soft polling)

If this quick test is correct, realtek 8139too's driver works as good as
eepro's driver. 

Octave 

>> from realtek (no NAPI) to realtek (no NAPI)
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-r: 327680000 bytes in 42.90 real seconds = 59671.99 Kbit/sec +++
ttcp-r: 224852 I/O calls, msec/call = 0.20, calls/sec = 5241.16
ttcp-r: 0.1user 1.5sys 0:42real 3% 0i+0d 0maxrss 0+2pf 0+0csw  

>> from eepro to eepro
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-r: 327680000 bytes in 28.33 real seconds = 90379.31 Kbit/sec +++
ttcp-r: 225058 I/O calls, msec/call = 0.13, calls/sec = 7945.54
ttcp-r: 0.2user 4.2sys 0:28real 15% 0i+0d 0maxrss 0+2pf 0+0csw 

>> from realtek (NAPI) to realtek (NAPI)
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-r: 327680000 bytes in 29.21 real seconds = 87644.11 Kbit/sec +++
ttcp-r: 225735 I/O calls, msec/call = 0.13, calls/sec = 7728.26
ttcp-r: 0.0user 1.7sys 0:29real 6% 0i+0d 0maxrss 0+2pf 0+0csw  

>> from eepro to realtek (no NAPI)
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-t: 327680000 bytes in 34.32 real seconds = 74594.99 Kbit/sec +++
ttcp-t: 40000 I/O calls, msec/call = 0.88, calls/sec = 1165.55
ttcp-t: 0.0user 1.2sys 0:34real 3% 0i+0d 0maxrss 0+2pf 0+0csw

>> from realtek (NAPI) to realtek (no NAPI)
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-r: 327680000 bytes in 32.60 real seconds = 78532.74 Kbit/sec +++
ttcp-r: 225544 I/O calls, msec/call = 0.15, calls/sec = 6918.98
ttcp-r: 0.1user 1.6sys 0:32real 5% 0i+0d 0maxrss 0+2pf 0+0csw

>> from realtek (NAPI) to eepro
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: socket
ttcp-r: accept from
ttcp-r: 327680000 bytes in 34.02 real seconds = 75250.05 Kbit/sec +++
ttcp-r: 225685 I/O calls, msec/call = 0.15, calls/sec = 6633.91
ttcp-r: 0.1user 3.7sys 0:34real 11% 0i+0d 0maxrss 0+2pf 0+0csw

Re: NAPI 8139too.c for 2.4.23

[Ben Greear]

Octave wrote:
>>Is there any chance you could do some benchmark runs with ttcp or
>>somesuch?
> 
> 
> I tested on 6-7 servers running with 
> eepro eth0: Intel Corp. 82557/8/9 [Ethernet Pro 100], 00:E0:18:01:78:6C, IRQ 10.
> realtek 8139too with NAPI 8139too Fast Ethernet driver 0.9.27 
> realtek 8139too with no NAPI (standard driver with soft polling)
> 
> If this quick test is correct, realtek 8139too's driver works as good as
> eepro's driver. 
> 
> Octave 

Those are some nice numbers!  I may have to bring some of my $5 realteks out of
retirement!  Anyone make a 4-port NIC with realteks on it?

Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: NAPI 8139too.c for 2.4.23

[Jeff Garzik]

Ben Greear wrote:
> Those are some nice numbers!  I may have to bring some of my $5 realteks 
> out of
> retirement!  Anyone make a 4-port NIC with realteks on it?

hah!  I hope not :)

(I've never heard of such a beast, but who knows...)

	Jeff