interrupt routing
dima
_pppp at mail.ru
Wed Feb 9 00:53:38 PST 2005
>>>> I am preparing a new server for production use.
>>>> It contains 2 1000BaseTX NICs and 2 SCSI controllers.
>>>> The interrupt assignment performed by ACPI looks kinda strange:
>>>> irq24: bge0 ahd0
>>>> irq25: bge1 ahd1
>>>> How can I affect it? I mean I want all the devices use different IRQ lines.
>>>
>>> What hardware, curiously? Are all of these parts onboard? Can you post
>>> the ouptut of 'devconf'? This will show the bus associations for these
>>> devices.
>> Its Tyan S2882 and all the devices are onboard ones.
>> I dont know about devconf ($ locate devconf produces empty output as well)
>
> Oops, soory, that should be 'devinfo'. But pciconf might tell me what I
> want to know.
>
>> but pciconf results are as follows:
>> ahd0 at pci2:6:0: class=0x010000 card=0x005e9005 chip=0x801d9005 rev=0x10 hdr=0x00
>> vendor = 'Adaptec Inc'
>> device = 'AIC-7902B Ultra320 SCSI Controller'
>> class = mass storage
>> subclass = SCSI
>> ahd1 at pci2:6:1: class=0x010000 card=0x005e9005 chip=0x801d9005 rev=0x10 hdr=0x00
>> vendor = 'Adaptec Inc'
>> device = 'AIC-7902B Ultra320 SCSI Controller'
>> class = mass storage
>> subclass = SCSI
>> bge0 at pci2:9:0: class=0x020000 card=0x164414e4 chip=0x164814e4 rev=0x03 hdr=0x00
>> vendor = 'Broadcom Corporation'
>> device = 'BCM5704 NetXtreme Dual Gigabit Adapter'
>> class = network
>> subclass = ethernet
>> bge1 at pci2:9:1: class=0x020000 card=0x164414e4 chip=0x164814e4 rev=0x03 hdr=0x00
>> vendor = 'Broadcom Corporation'
>> device = 'BCM5704 NetXtreme Dual Gigabit Adapter'
>> class = network
>> subclass = ethernet
>
> Ah, so they are all on the same bus. Yuck, performance is going to be
> sucky. Bad Tyan, no cookie. That'll also explain the limited number of
> interrupts available. I don't think there's anything we can do to help
> the situation, sadly.
I cannot affect the company equipment purchase policy either :/
2 more servers on Tyan motherboards perform pretty bad also.
>> The server runs i386 version of FreeBSD (5.3-RELEASE-p5) since
>> I experienced some problems building ports on amd64 version.
>
> You probably need to use the hw.physmem="2G" loader tunable to get 5.3-R
> installed. Once installed you can upgrade to 5-STABLE which fixes the
> problem.
Well, I dont experience any problems with the base system
(the server has 4G of physical RAM btw).
The ports collection isnt amd64-ready though. I compiled some ports patching their makefiles
but some of them dont compile at all. Say, I failed to build vnc server from ports
(I needed it to install Oracle) the only one I managed to build was an ancient realvnc (3.3.7),
but I couldnt connect to it. I tried to compile realvnc 4.x from sources but ran into namespace issues
(they were discussed on another thread here regarding some software package; seems to be a buggy gcc).
So, Ive given up and happily installed an i386 version.
>>> In many cases there are not other IRQs available to route, due to poor
>>> BIOS programming, ccorners cut in the physical board layout, etc.
>> I think this is the case :/
>
>> The BIOS assigned all those devices IRQ10 and there is no way to change
>> the settings...
More information about the freebsd-stable
mailing list