New ACPI PCI Link Routing code

[John Baldwin]

On Nov 19, 2004, at 5:58 PM, Ruslan Ermilov wrote:

> On Fri, Nov 19, 2004 at 03:59:46PM -0500, John Baldwin wrote:
>> On Thursday 18 November 2004 03:46 am, Ruslan Ermilov wrote:
>>> On Thu, Nov 18, 2004 at 02:19:47AM +0200, Ruslan Ermilov wrote:
>>>> On Wed, Nov 17, 2004 at 05:44:53PM -0500, John Baldwin wrote:
>>>> [...]
>>>>
>>>>> Erm, unless I've got a logic bug I don't run the KASSERT() for zero
>>>>> interrupts.  Oh darn, I left a bogus KASSERT() in in the function 
>>>>> that
>>>>> does the actual routing.  The assertion's on lines 497 and 511 can 
>>>>> be
>>>>> dropped. I'll update the patch in a second.
>>>>
>>>> With an updated patch, I no longer get panic on boot, and there
>>>> are no more interrupt storms, but the latter is probably at the
>>>> cost of old bug re-introduced.  My dc(4) PCCard doesn't get a
>>>> correct IRQ, "dc0: watchdog timeout".
>>>
>>> I've put a verbose boot output here, as requested:
>>>
>>> 	http://people.freebsd.org/~ru/dmesg.boot
>>
>> Looks like it hung everything off of IRQ 9 since the BIOS didn't 
>> preset any
>> device IRQs.  What was the behavior prior to this patch?
>>
>  	http://people.freebsd.org/~ru/dmesg.boot2
>
> The system is dead slow due to interrupt storm in IRQ 10,
> but at least dc(4) works.
>
> # vmstat -i
> interrupt                          total       rate
> irq0: clk                         353566        993
> irq1: atkbd0                        1134          3
> irq6: fdc0                            22          0
> irq8: rtc                          45394        127
> irq9: acpi0                          648          1
> irq10: cbb1 dc0                 28202293      79219
> irq11: cbb0 csa0                  689545       1936
> irq12: psm0                         1161          3
> irq13: npx0                            1          0
> irq14: ata0                         1939          5
> irq15: ata1                           52          0
> Total                           29295755      82291

Try setting 'hw.pci.link.LNKA.irq=11' in the loader with the patch and 
see if that helps dc(4) to work better.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org