kern/91408 : [irq] ata(4) failure: SETFEATURES SET TRANSFER MODE semaphore timeout !! DANGER Will Robinson !!

[John Baldwin]

The following reply was made to PR kern/91408; it has been noted by GNATS.

From: John Baldwin <jhb at freebsd.org>
To: Eugene Grosbein <eugen at grosbein.pp.ru>
Cc: bug-followup at freebsd.org, lightsquid at logvinov.com
Subject: Re: kern/91408 : [irq] ata(4) failure: SETFEATURES SET TRANSFER MODE semaphore timeout !! DANGER Will Robinson !!
Date: Mon, 1 May 2006 15:17:06 -0400

 On Sunday 30 April 2006 04:44, Eugene Grosbein wrote:
 > On Sat, Apr 29, 2006 at 10:54:09PM -0400, John Baldwin wrote:
 > 
 > > > With 6.1-RC of today I have the same sympthoms as with 5.4-RELEASE
 > > > (http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/80815):
 > > >
 > > > 1) machine suffers from ATA timeouts if ACPI is fully enabled;
 > > > 2) debug.acpi.disable="pci_link" in /boot/loader.conf eliminates the
 > > > problem.
 > > >
 > > > So, Soren seems to be right: this is interrupt routing problem
 > > > and not ATA problem.
 > > >
 > > > The question is: should I consider the workaround mentioned above as
 > > > solution? What will I miss if I keep debug.acpi.disable="pci_link" forever?
 > > 
 > > I think it's dangerous
 > 
 > What kind of trouble I am "asking for" while using debug.acpi.disable="pci_link"?
 
 Because ACPI is rather intertwined, so it is expecting to tell the OS how to
 route interrupts in a certain way, and if you enable ACPI the BIOS is expecting
 you to use it completely.
 
 > > and that you should just disable ACPI altogether if you wish to do that.
 > 
 > I think I do not wish that :-)
 
 Do so at your own risk then.
 
 > This machive has four OS now: FreeBSD 4.11-RELEASE (uses APM to turn power off),
 > Windows 98SE and Windows XP SP2 (use ACPI, no problems with it).
 > 
 > Would I be allowed to turn power off with 6.1 without ACPI enabled?
 > Would it be possible not only for 'shutdown -p', but for ACPI power button too? 
 
 No, the power button only works with ACPI.  shutdown -p can work using apm
 as on 4.x.
   
 > >  Can you provide verbose dmesg's for the
 > > case with pci_link disabled and the case where it is not disabled?
 > 
 > Here comes dmesg.acpi (ACPI is fully enabled):
 
 Well, all the IRQs are the same and none of the interrupts were changed to
 be edge triggered or anything like that, so it's not a problem with
 interrupt routing.  If the interrupt routing were busted, the IRQ numbers
 would be different.  All the pci_link devices do is help the OS figure out
 which IRQ number a device uses.  If those numbers are all the same, then
 interrupt routing is not the issue.
 
 -- 
 John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
 "Power Users Use the Power to Serve"  =  http://www.FreeBSD.org