Interrupt storm on acpi0 occurred after madt.c change

[John Baldwin]

On Thursday 05 February 2004 06:19 pm, Nate Lawson wrote:
> It should work fine in Linux.  mp_config_ioapic_for_sci() in
> arch/i386/kernel/mpparse.c has an exception for Tyans:
>
> /*
>  * Although the ACPI spec says that the SCI should be level/low
>  * don't reprogram it unless there is an explicit MADT OVR entry
>  * instructing us to do so -- otherwise we break Tyan boards which
>  * have the SCI wired edge/high but no MADT OVR.
>  */
>
> This is for the case where there is no source override in the MADT for the
> interrupt defined as ACPI_FADT->sci_int.

The problem is that these motherboards _do_ have an MADT override entry.  Even 
worse, I have seen different motherboards (some Tyan, some non-Tyan) that 
both have the exact same MADT override entry that maps IRQ 9 to interrupt 20 
with active-hi polarity and level trigger.  Now, on the !Tyan motherboards, 
using those params results in an interrupt storm, so I force the polarity to 
active-lo (on these motherboards, another PCI device such as fxp0 also uses 
IRQ 20).  On the Tyan motherboards, the MADT entry is _identical_, but it 
storms if I set the polarity to low but works if it uses the bogus setting of 
active-hi (but with level trigger).  The Tyan motherboards do not seem to 
share IRQ 20 with a PCI device, but I can't tell if a PCI device uses IRQ 20 
in the MADT parser.  It's a big mess, as I have two different vendors using 
the same exact MADT entry both with a bogus polarity, and one wants hte bogus 
polarity and one wants the normal polarity (i.e. use active-lo with level 
trigger).

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