Skip to contentSkip to conversation

Via-Rhine ethernet driver problem?

DISCUSSION walt 4 participants 5 messages Mar 29, 2004 – Jun 18, 2004

lore
Share & tools
lore

walt

author
#0 lore raw

Hi, Roger et al,

ECS K7VTA3 motherboard with built-in ethernet chip:

00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev 74)
Subsystem: VIA Technologies, Inc. VT6102 [Rhine II] Embeded Ethernet Controller on VT8235
Flags: bus master, medium devsel, latency 32, IRQ 11
I/O ports at ec00 [size=256]
Memory at e9041000 (32-bit, non-prefetchable) [size=256]
Capabilities: [40] Power Management version 2

The problem is terrible performance -- I noticed that NFS file transfers grind to a complete halt almost immediately on this machine.

I also discovered by using 'scp' to copy files between machines that the bad performance is assymetrical: copying a file *to* this machine runs at about half-speed (5 MB/sec) whereas copying a file *from* this machine runs at 45 KiloB/sec, about one percent of expected.

The reason I feel this is software and not hardware is that the same machine running any of the BSD's runs full-speed in both directions.

The 2.4.x and 2.6.x drivers appear to be the same and they behave the same, as you might expect.

I'd be happy to supply any other information that might be helpful.

Thanks!

#1 lore raw
Quoted 7 lines · walt

ECS K7VTA3 motherboard with built-in ethernet chip:

00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev
74)

[...]

Quoted 7 lines

I also discovered by using 'scp' to copy files between machines that the bad performance is assymetrical: copying a file *to* this machine runs at about half-speed (5 MB/sec) whereas copying a file *from* this machine runs at 45 KiloB/sec, about one percent of expected.

The reason I feel this is software and not hardware is that the same machine running any of the BSD's runs full-speed in both directions.

If you have ACPI and/or IO-APIC enabled, does the behavior change if you turn them off? Any info in the kernel log? If not, what if you change the driver's debug level to 3? Is the slow transfer rate the result of short, fast bursts or actual sustained throughput at 45 KB/sec?

Roger

#3 lore raw
In reply to walt
Quoted 14 lines · walt

Hi, Roger et al,

ECS K7VTA3 motherboard with built-in ethernet chip:

00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev 74)
Subsystem: VIA Technologies, Inc. VT6102 [Rhine II] Embeded Ethernet Controller on VT8235
Flags: bus master, medium devsel, latency 32, IRQ 11
I/O ports at ec00 [size=256]
Memory at e9041000 (32-bit, non-prefetchable) [size=256]
Capabilities: [40] Power Management version 2

The problem is terrible performance -- I noticed that NFS file transfers grind to a complete halt almost immediately on this machine.

Quoted 7 lines

I also discovered by using 'scp' to copy files between machines that the bad performance is assymetrical: copying a file *to* this machine runs at about half-speed (5 MB/sec) whereas copying a file *from* this machine runs at 45 KiloB/sec, about one percent of expected.

The reason I feel this is software and not hardware is that the same machine running any of the BSD's runs full-speed in both directions.

Hello,

I'm trying to take some load off from actual kernel developers and doing just a small check: is your NIC in full duplex mode while in Linux? Please check it with mii-tool or ethtool. This sounds like a half-duplex problem for me...

Best Regards,

Janne Pikkarainen

author
#2 lore raw
In reply to Roger Luethi
Quoted 12 lines · Roger Luethi
Quoted 8 lines · walt

ECS K7VTA3 motherboard with built-in ethernet chip:

00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev 74) [...]
I also discovered by using 'scp' to copy files between machines that the bad performance is assymetrical: copying a file *to* this machine runs at about half-speed (5 MB/sec) whereas copying a file *from* this machine runs at 45 KiloB/sec, about one percent of expected.

Quoted 2 lines

If you have ACPI and/or IO-APIC enabled, does the behavior change if you turn them off?

I did have both -- no difference when I turned them both off.

Quoted 2 lines

Any info in the kernel log? If not, what if you
change the driver's debug level to 3?

This is debug=4 since debug=3 didn't put out much:

Mar 30 04:27:58 k9 kernel: via-rhine.c:v1.10-LK1.1.19-2.5  July-12-2003  Written by Donald Becker
Mar 30 04:27:58 k9 kernel:   http://www.scyld.com/network/via-rhine.html
Mar 30 04:27:58 k9 kernel: via-rhine: Reset succeeded.
Mar 30 04:27:58 k9 kernel: eth0: VIA VT6102 Rhine-II at 0xe9041000, 00:0d:87:a2:ce:02, IRQ 11.
Mar 30 04:27:58 k9 kernel: eth0: MII PHY found at address 1, status 0x786d advertising 05e1 Link 45e1.
Mar 30 04:27:58 k9 kernel: eth0: MII PHY found at address 2, status 0x8000 advertising 0000 Link 8000.
Mar 30 04:27:58 k9 kernel: eth0: MII PHY found at address 3, status 0x8000 advertising 0000 Link 8000.
Mar 30 04:27:58 k9 kernel: eth0: MII PHY found at address 4, status 0x8000 advertising 0000 Link 8000.
Mar 30 04:28:37 k9 kernel: eth0: via_rhine_open() irq 11.
Mar 30 04:28:37 k9 kernel: eth0: Reset succeeded.
Mar 30 04:28:37 k9 kernel: eth0: Done via_rhine_open(), status 081a MII status: 786d.
Mar 30 04:28:37 k9 kernel: eth0: VIA Rhine monitor tick, status 0000.
Mar 30 04:29:17 k9 last message repeated 4 times
Mar 30 04:29:37 k9 last message repeated 2 times
Mar 30 04:29:44 k9 kernel: eth0: exiting interrupt, status=00000000.
Mar 30 04:29:44 k9 last message repeated 25 times
Mar 30 04:29:47 k9 kernel: eth0: VIA Rhine monitor tick, status 0000.
Mar 30 04:29:49 k9 kernel: eth0: exiting interrupt, status=00000000.
Mar 30 04:29:57 k9 last message repeated 543 times

[much more snipped]

Quoted 3 lines

Is the slow transfer rate
the result of short, fast bursts or actual sustained throughput at 45 KB/sec?

Just from watching the lights on the ethernet hub I would say that it is bursting. Is there a more accurate way to tell?

Thanks for the quick response!

BTW, this is from ethtool:
Supported ports: [ TP MII ]
Supported link modes: 10baseT/Half 10baseT/Full

                        100baseT/Half 100baseT/Full
Supports auto-negotiation: Yes
Advertised link modes:  10baseT/Half 10baseT/Full
                        100baseT/Half 100baseT/Full

Advertised auto-negotiation: Yes
Speed: 100Mb/s
Duplex: Full
Port: MII
PHYAD: 1
Transceiver: internal
Auto-negotiation: on
Current message level: 0x00000004 (4)
Link detected: yes

#4 lore raw
In reply to walt

walt wrote:

[...]

Quoted 4 lines

I also discovered by using 'scp' to copy files between machines that the bad performance is assymetrical: copying a file *to* this machine runs at about half-speed (5 MB/sec) whereas copying a file *from* this machine runs at 45 KiloB/sec, about one percent of expected.

I have exactly the same problem on a asrock k7vt2 with VIA's Rhine II:

megabyte:~# uname -a
Linux megabyte 2.6.7 #1 Fri Jun 18 20:37:37 CEST 2004 i686 GNU/Linux

=================================================================

megabyte:~# lspci -v
0000:00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev 74)
        Subsystem: VIA Technologies, Inc. VT6102 [Rhine-II]
        Flags: bus master, medium devsel, latency 64, IRQ 11
        I/O ports at dc00
        Memory at dffffe00 (32-bit, non-prefetchable) [size=256]
        Capabilities: [40] Power Management version 2

==================================================================

up down
100baseTx-FD: ~45 kB/s ~100 MBit/s
100baseTx: ~45 kB/s ~100 MBit/s
10baseT-FD: ~45 kB/s ~10 MBit/s
10baseT: ~10 MBit/s ~10 MBit/s

Signature · 2 lines
Greetings
Bjoern Schmidt