Re: Call for testers: new pci code

[Sepherosa Ziehau <sepherosa@xxxxxxxxx>]

On Mon, Jul 6, 2009 at 3:42 AM, Matthew
Dillon<dillon@apollo.backplane.com> wrote:
> :Hi all,
> :
> :After some additional work, I integrated Polakov's new pci code ported
> :from FreeBSD 7.2:
> :http://gitweb.dragonflybsd.org/~sephe/dragonfly.git/shortlog/refs/heads/pci
> :
> :I have tested SMP+IOAPIC, SMP+ICU, UP (there is not improvement in
> :IOAPIC in this branch; my boxes just work with the current IOAPIC :)
> :CARDBUS and PCCARD is tested on an old Dell laptop.
> :
> :I am currently running it on all of my boxes :)
> :
> :Though it should be safe to be tested, but I still suggest you to keep
> :one copy of your working kernel and module!
> :I also suggest to test it with HPET enabled.
> :
> :If there are any problems, please post the boot verbose dmesg first :)
> :
> :Best Regards,
> :sephe
>
>    Here is the output for a SMP build on my new Phenom-based shuttle test
>    box, without APIC_IO and with APIC_IO.
>
>    With SMP, without APIC_IO (works normally):

Good to know :)

>
>        fetch http://apollo.backplane.com/DFlyMisc/dmesg01.boot
>
>    With SMP, With APIC_IO (did not work before, does not work now):

heh

>
>        fetch http://apollo.backplane.com/DFlyMisc/dmesg02.boot
>
>    With APIC_IO the AHCI device seems to route its interrupt ok.
>    However, just about everything else is still broken.  i.e. NFE,
>    EHCI, etc... all route to irq 2 and are broken.  I guess that
>    is to be expected at this stage?.  It never worked with APIC_IO
>    before and still doesn't :-)

Yep, there is no improvement in respect to APIC_IO in this patch.

>
>    There is an ACPI entry for the HPET but I don't see any recognition of
>    it as a clock source.

Ah, you need to set tunable:
debug.acpi.enabled="hpet"

>
>    In anycase, the machine boots without APIC_IO like it always has but
>    I'm not sure if there is some sort of loader.conf option I need to set
>    to test the new code or not.

Nope, there are no additional tunables (at least the irq overridden
tunable has no effect if APIC_IO is defined)

Best Regards,
sephe

-- 
Live Free or Die