ASUS P5A broken by ACPI black-list

[Nate Lawson]

Kevin Oberman wrote:

>>Date: Mon, 04 Oct 2004 09:54:19 -0700
>>From: Nate Lawson <nate at root.org>
>>
>>Kevin Oberman wrote:
>>
>>>After building some old kernels, it became apparent that the network
>>>broke long before the disk problem showed up and was the result of the
>>>ACPI black-list implemented back on June 30. Unfortunately, this board
>>>seems to require ACPI to run. Without it I get continuous xl0: watchdog
>>>timeout messages from the system. 
>>>
>>>I have now re-enabled ACPI in the hints file and the network is
>>>fine. (I'm still looking for the source of the disk problem.) Why it
>>>fails without ACPI is another issue. This is a rather old board and has
>>>run just fine for years on V3 and V4 kernels.
>>>
>>>The quirk entry is:
>>># ASUS P5A 03/12/99
>>>name:		ASUS_P5A
>>>oem:		FADT "ASUS  " "P5A     "
>>>oem_rev:	FADT <= 0x58582e31
>>>quirks:		ACPI_Q_BROKEN
>>>
>>>My FADT is:
>>>  FADT: FACS=0x5fff000, DSDT=0x5ffc100
>>>        INT_MODEL=PIC
>>>        Preferred_PM_Profile=Unspecified (0)
>>>        SCI_INT=9
>>>        SMI_CMD=0xb1, ACPI_ENABLE=0xa1, ACPI_DISABLE=0xa0, S4BIOS_REQ=0x0
>>>        PSTATE_CNT=0x0
>>>        PM1a_EVT_BLK=0xec00-0xec03
>>>        PM1a_CNT_BLK=0xec04-0xec05
>>>        PM2_CNT_BLK=0xec30-0xec30
>>>        PM_TMR_BLK=0xec08-0xec0b
>>>        GPE0_BLK=0xec18-0xec1b
>>>        GPE1_BLK=0xec1c-0xec1f, GPE1_BASE=16
>>>        P_LVL2_LAT=90 us, P_LVL3_LAT=900 us
>>>        FLUSH_SIZE=0, FLUSH_STRIDE=0
>>>        DUTY_OFFSET=1, DUTY_WIDTH=0
>>>        DAY_ALRM=13, MON_ALRM=0, CENTURY=0
>>>        IAPC_BOOT_ARCH=
>>>        Flags={WBINVD,PROC_C1,SLP_BUTTON,RTC_S4}
>>
>>I need output from a newer version of acpidump.  The output above 
>>doesn't include the OEM header for the FADT.  June 30th was the commit 
>>that added this output.
> 
> 
> That does make a bit of a difference. I have 0x58582e31, so it is
> black-listed.

For the record, please send the full acpidump -t output (with a recent 
acpidump).

> Unfortunately, it does not work without ACPI. The only
> ACPI problem is that the clock runs at double speed.

That may be why it was blacklisted.  I don't know all the rationale for 
some of the blacklist entries, only that Windows won't install acpi.sys 
on machines that we blacklist.  In your case, it's likely a broken 
timer.  We can refine/fine-tune blacklist entries to only disable the 
problem parts (if possible).

> It looks like interrupts from the Ethernet are not delivered without
> ACPI, but that is hardly your problem. I have over-ridden the black-list
> and things are back to normal.

The reason this system works in Windows without ACPI is that irq routing 
in Windows uses multiple info sources including _PIR and $PIR.  John 
Baldwin has patches to do this for us too.

> Is it possible to only disable the use of the clock but leave the rest
> of ACPI running? (Not that using the sysctl to switch to TSC is much of
> a problem.)

Yes:
debug.acpi.disable="timer"

-- 
Nate