Re: NFS vs jumbograms?

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

> I sometimes heard that wm had some problems with jumbo MTU, but I've
> never checked whether it really has a bug or not.  I suspect the bug
> is not in NFS stack but in wm.

Based on diffing tcpdump -x output, if there is such a bug it is in
udp4csum offload.  I consider this unlikely because I think I tried it
with all offloading turned off.  Of course, it's also possible tcpdump
isn't showing the full truth....

I should be able to create a test setup I can play with freely, so I
ought to be able to test this theory relatively painlessly - though if
it's specific to the details of the chip, my test setup may not
misbehave.  I also have access to some machines with bge instead of wm,
so I should be able to test with bge instead of wm in any of the four
combinations ({bge,wm} on NFS {server,client}).

> I don't use jumbo MTU, so usually don't test the function when I add
> the support of a new chip.  Sorry :(

In case it helps, the case I saw issues with has

wm1 at pci7 dev 8 function 0: Intel i82541GI 1000BASE-T Ethernet, rev. 5
wm1: interrupting at ioapic2 pin 1 (irq 3)
wm1: 32-bit 66MHz PCI bus
wm1: 65536 word (16 address bits) SPI EEPROM
wm1: Ethernet address 00:13:72:4d:83:03
igphy1 at wm1 phy 1: Intel IGP01E1000 Gigabit PHY, rev. 0
igphy1: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT, 
1000baseT-FDX, auto
ukphy1 at wm1 phy 2: Generic IEEE 802.3u media interface
ukphy1: OUI 0x000000, model 0x0000, rev. 0
ukphy1: no media present

on the NFS client and

wm0 at pci0 dev 10 function 0: Intel i82541PI 1000BASE-T Ethernet, rev. 5
wm0: interrupting at irq 12
wm0: 32-bit 33MHz PCI bus
wm0: 64 word (6 address bits) MicroWire EEPROM
wm0: Ethernet address 90:e2:ba:3b:f5:54
igphy0 at wm0 phy 1: Intel IGP01E1000 Gigabit PHY, rev. 0
igphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT, 
1000baseT-FDX, auto

on the NFS server.  I can show more details, up to and including full
boot-time autoconf messages, if that would help.  wm0 on the NFS server
is an add-on PCI card; wm1 on the NFS client is onboard.

I don't know whether the cable between them is straight or crossover.
(I also don't know whether that's relevant; I don't know enough about
how gigabit auto-x negotation works.)

/~\ The ASCII                             Mouse
\ / Ribbon Campaign
 X  Against HTML                mouse%rodents-montreal.org@localhost
/ \ Email!           7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B