Re: High TLB shootdown I interrupts with wm gigabit NIC?

[Mark Cullen <mark.r.cullen%gmail.com@localhost>]

Matthias Scheler wrote:
> On Tue, Jul 29, 2008 at 09:21:46PM +0100, Mark Cullen wrote:
> > Firstly, transfer rates are "only" about 30MB/s, both with Samba and by 
> > FTP. This is in both directions. This seems a bit low?
>
> Both aren't good network benchmarks because they are limited by the
> application performance, disk speed, etc..
>
> Please use the "netio" or "ttcp" package from "pkgsrc" for network
> benchmarking.
>
> > Secondly, I have noticed that whilst idle "TLB shootdown I" (there are 
> > two of these, dual core, I believe this is the one for CPU #1) is next 
> > to nothing. It'll go up to 5 every so often and then just not register 
> > for the rest of the time. Whilst doing an FTP... it  goes through the roof:
> [...]
>
> What version of NetBSD are you using?
>
>         Kind regards

Sorry, I forgot to mention the more important stuff.


(markie@homeserver:ttyp0)-(~)$ uname -a
NetBSD homeserver.bone.servebeer.com 4.0_STABLE NetBSD 4.0_STABLE 
(HOMESERVER.MP) #2: Sat Jul 12 14:54:54 BST 2008 
markie%homeserver.bone.servebeer.com@localhost:/usr/obj/sys/arch/amd64/compile/HOMESERVER.MP 
amd64

(markie@homeserver:ttyp0)-(~)$ dmesg
[...]
ppb0 at pci0 dev 28 function 0: Intel 82801GB/GR PCI Express Port #1 
(rev. 0x01)
pci1 at ppb0 bus 1
pci1: i/o space, memory space enabled, rd/line, wr/inv ok
wm0 at pci1 dev 0 function 0: Intel i82572EI 1000baseT Ethernet, rev. 6
wm0: interrupting at ioapic0 pin 16 (irq 10)
wm0: PCI-Express bus
wm0: 65536 word (16 address bits) SPI EEPROM
wm0: Ethernet address 00:1b:21:1e:77:98
igphy0 at wm0 phy 1: Intel IGP01E1000 Gigabit PHY, rev. 0
igphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT, 
1000baseT-FDX, auto
[...]

(markie@homeserver:ttyp0)-(~)$ dmesg | grep irq
pckbc1: io 0x60,0x64 irq 1
pckbc1: using irq 1 for kbd slot
wm0: interrupting at ioapic0 pin 16 (irq 10)
uhci0: interrupting at ioapic0 pin 23 (irq 12)
uhci1: interrupting at ioapic0 pin 19 (irq 11)
uhci2: interrupting at ioapic0 pin 18 (irq 5)
uhci3: interrupting at ioapic0 pin 16 (irq 10)
ehci0: interrupting at ioapic0 pin 23 (irq 12)
fxp0: interrupting at ioapic0 pin 19 (irq 11)
piixide0: primary channel interrupting at ioapic0 pin 14 (irq 14)
piixide0: secondary channel interrupting at ioapic0 pin 15 (irq 15)
piixide1: using ioapic0 pin 19 (irq 11) for native-PCI interrupt
lpt0 at isa0 port 0x378-0x37b irq 7
com0 at isa0 port 0x3f8-0x3ff irq 4: ns16550a, working fifo


I think I mentioned it was a PCI-E card, and it's on a Gigabyte G31M-S2L 
board with a dual core E2180. It does have an onboard RealTek gigabit 
NIC, but it appears to be a bit dodgy... hence buying the Intel one!

(markie@homeserver:ttyp0)-(~)$ ifconfig wm0
wm0: flags=9843<UP,BROADCAST,RUNNING,SIMPLEX,LINK0,MULTICAST> mtu 4470

capabilities=7ff80<TSO4,IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx,TCP6CSUM_Rx,TCP6CSUM_Tx,UDP6CSUM_Rx,UDP6CSUM_Tx,TSO6>

enabled=3f00<IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx>
        address: 00:1b:21:1e:77:98
        media: Ethernet autoselect (1000baseT 
full-duplex,flowcontrol,rxpause,txpause)
        status: active

It's attached to a brand new 3com OfficeConnect 8 port 10/100/1000 
switch (3C1670800B-UK)

Cheers,
--
Mark Cullen <mark.r.cullen%gmail.com@localhost>