3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Sven Köhler]

Hi,

when using acpi=off in the kernel or xen command line, the system won't
boot. On real hardware, I saw a few interrupt related warnings from the
usb drivers. The system then seemed to lock up when trying to do I/O via
AHCI. Same in virtualbox. System won't come up.


Regards,
  Sven

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Konrad Rzeszutek Wilk]

On Sun, Sep 11, 2011 at 12:43:18AM +0200, Sven Köhler wrote:
> Hi,
> 
> when using acpi=off in the kernel or xen command line, the system won't
> boot. On real hardware, I saw a few interrupt related warnings from the
> usb drivers. The system then seemed to lock up when trying to do I/O via
> AHCI. Same in virtualbox. System won't come up.

Not surprised. Without the ACPI we can't find parse the interrupt table,
so you don't get any interrupts.

This is result of the reboot issue you have been seeing with your box?
You might want to try some parameters on the Xen line to alter how
it is suppose to reboot.

/*
 * reboot=b[ios] | t[riple] | k[bd] | n[o] [, [w]arm | [c]old]
 * warm   Don't set the cold reboot flag
 * cold   Set the cold reboot flag
 * bios   Reboot by jumping through the BIOS (only for X86_32)
 * triple Force a triple fault (init)
 * kbd    Use the keyboard controller. cold reset (default)
 * acpi   Use the RESET_REG in the FADT
 */

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Sven Köhler]

[-- Attachment #1.1: Type: text/plain, Size: 1980 bytes --]

Am 11.09.2011 02:28, schrieb Konrad Rzeszutek Wilk:
> On Sun, Sep 11, 2011 at 12:43:18AM +0200, Sven Köhler wrote:
>> Hi,
>>
>> when using acpi=off in the kernel or xen command line, the system won't
>> boot. On real hardware, I saw a few interrupt related warnings from the
>> usb drivers. The system then seemed to lock up when trying to do I/O via
>> AHCI. Same in virtualbox. System won't come up.
> 
> Not surprised. Without the ACPI we can't find parse the interrupt table,
> so you don't get any interrupts.

Thanks for explaining. Now what about the future? Will there be some
solution for the acpi=off case?

I'm a bit confused, since your words don't sound like there is a way to
boot with acpi=off. But other dom0 kernels actually boot with acpi=off.
So after all, some other way for setting up interrupts seems to exist.

> This is result of the reboot issue you have been seeing with your box?

Yes.

> You might want to try some parameters on the Xen line to alter how
> it is suppose to reboot.
> 
> /*
>  * reboot=b[ios] | t[riple] | k[bd] | n[o] [, [w]arm | [c]old]

Thanks for the list.
I guess, both reboot=bios and reboot=b is accepted?
BTW: "no" is missing in the list below. acpi is missing in the list
above. And actually what's the source for list?
(I never find any documentation about the hypervisor options, which is
pretty frustrating sometimes)

>  * warm   Don't set the cold reboot flag
>  * cold   Set the cold reboot flag
>  * bios   Reboot by jumping through the BIOS (only for X86_32)
>  * triple Force a triple fault (init)
>  * kbd    Use the keyboard controller. cold reset (default)
>  * acpi   Use the RESET_REG in the FADT
>  */

So in fact, xen is doing the reboot, and not the dom0 kernel, right?
(Some people have claimed otherwise)

Could you imagine to adapt xen's reboot code to the one of linux 3.0
(which was tweaked quite a lot for maximum compatibility)



Regards,
  Sven


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Ian Campbell]

On Sat, 2011-09-10 at 20:44 -0400, Sven Köhler wrote:
> Am 11.09.2011 02:28, schrieb Konrad Rzeszutek Wilk:
> > On Sun, Sep 11, 2011 at 12:43:18AM +0200, Sven Köhler wrote:
> >> Hi,
> >>
> >> when using acpi=off in the kernel or xen command line, the system won't
> >> boot. On real hardware, I saw a few interrupt related warnings from the
> >> usb drivers. The system then seemed to lock up when trying to do I/O via
> >> AHCI. Same in virtualbox. System won't come up.
> > 
> > Not surprised. Without the ACPI we can't find parse the interrupt table,
> > so you don't get any interrupts.
> 
> Thanks for explaining. Now what about the future? Will there be some
> solution for the acpi=off case?
> 
> I'm a bit confused, since your words don't sound like there is a way to
> boot with acpi=off. But other dom0 kernels actually boot with acpi=off.
> So after all, some other way for setting up interrupts seems to exist.

Using acpi=off is sometimes a useful way to diagnose an issue, and once
upon a time it may have even done more good than harm and been a useful
solution to get a machine to work (on native as well under Xen) but on a
modern system disabling ACPI is likely to do more harm than good, it's
simply too ingrained into the way things work these days (again, on
native as much as under Xen).

> So in fact, xen is doing the reboot, and not the dom0 kernel, right?
> (Some people have claimed otherwise)

The dom0 kernel reboot method is always "via Xen", which is probably
where the confusion arises, in many cases dom0 _initiates_ the reboot
but doesn't actually do the reboot itself, Xen does the actual rebooting
since it is the only entity with access to the required mechanisms in
most cases.

> Could you imagine to adapt xen's reboot code to the one of linux 3.0
> (which was tweaked quite a lot for maximum compatibility)

It's probably worth revisiting and resyncing with what Linux does, IIRC
Matthew Garret did a bunch of work on this recently. In particular 
        
        commit 660e34cebf0a11d54f2d5dd8838607452355f321
        Author: Matthew Garrett <mjg@redhat.com>
        Date:   Mon Apr 4 13:55:05 2011 -0400
        
            x86: Reorder reboot method preferences
            
but also all the stuff in the quirks file I suppose.

Ian.

> 
> 
> Regards,
>   Sven
> 

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Stefano Stabellini]

[-- Attachment #1: Type: text/plain, Size: 571 bytes --]

On Sun, 11 Sep 2011, Sven Köhler wrote:
> Thanks for explaining. Now what about the future? Will there be some
> solution for the acpi=off case?

Not likely.


> I'm a bit confused, since your words don't sound like there is a way to
> boot with acpi=off. But other dom0 kernels actually boot with acpi=off.
> So after all, some other way for setting up interrupts seems to exist.

It exists in xenlinux kernels, where xen support is maintained in an out
of tree patch queue by distros.
Currently Linux mainline only offers acpi and msi registration hooks to
subsystems.

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Konrad Rzeszutek Wilk]

On Sun, Sep 11, 2011 at 02:44:35AM +0200, Sven Köhler wrote:
> Am 11.09.2011 02:28, schrieb Konrad Rzeszutek Wilk:
> > On Sun, Sep 11, 2011 at 12:43:18AM +0200, Sven Köhler wrote:
> >> Hi,
> >>
> >> when using acpi=off in the kernel or xen command line, the system won't
> >> boot. On real hardware, I saw a few interrupt related warnings from the
> >> usb drivers. The system then seemed to lock up when trying to do I/O via
> >> AHCI. Same in virtualbox. System won't come up.
> > 
> > Not surprised. Without the ACPI we can't find parse the interrupt table,
> > so you don't get any interrupts.
> 
> Thanks for explaining. Now what about the future? Will there be some
> solution for the acpi=off case?

No.
> 
> I'm a bit confused, since your words don't sound like there is a way to
> boot with acpi=off. But other dom0 kernels actually boot with acpi=off.

Sure.
> So after all, some other way for setting up interrupts seems to exist.

The older kernels (XenOLinux) made it possible by copying a lot of the
generic code in its own and making it work. That is not possible with the
upstream kernel.

Well, maybe it is possible, but I am not sure if it is worth the effort.

> 
> > This is result of the reboot issue you have been seeing with your box?
> 
> Yes.
> 
> > You might want to try some parameters on the Xen line to alter how
> > it is suppose to reboot.
> > 
> > /*
> >  * reboot=b[ios] | t[riple] | k[bd] | n[o] [, [w]arm | [c]old]
> 
> Thanks for the list.
> I guess, both reboot=bios and reboot=b is accepted?
> BTW: "no" is missing in the list below. acpi is missing in the list
> above. And actually what's the source for list?

Xen hypervisor source. I just did a quick search for 'reboot='

> (I never find any documentation about the hypervisor options, which is
> pretty frustrating sometimes)
> 
> >  * warm   Don't set the cold reboot flag
> >  * cold   Set the cold reboot flag
> >  * bios   Reboot by jumping through the BIOS (only for X86_32)
> >  * triple Force a triple fault (init)
> >  * kbd    Use the keyboard controller. cold reset (default)
> >  * acpi   Use the RESET_REG in the FADT
> >  */
> 
> So in fact, xen is doing the reboot, and not the dom0 kernel, right?

Yes. Dom0 triggers it though (by invoking an hypercall that tells
Xen to reboot/shutdown the machine).
> (Some people have claimed otherwise)
> 
> Could you imagine to adapt xen's reboot code to the one of linux 3.0
> (which was tweaked quite a lot for maximum compatibility)

I can imagine it.. but without any ideas of why your machine is not rebooting
it is a bit .. difficult.

You could also try on the Xen hypervisor line (Ctlr-A three times) try the 'R'
and see if it does anything.

> 
> 
> 
> Regards,
>   Sven
> 

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Sven Köhler]

[-- Attachment #1.1: Type: text/plain, Size: 570 bytes --]

> You could also try on the Xen hypervisor line (Ctlr-A three times) try the 'R'
> and see if it does anything.

As long as I have been using Xen, I have never managed to make this work
(getting to the hypervisor line by pressing Ctrl-A I mean). I have never
seen the prompt, if there is any.

Right now, I'm pressing Ctrl-A like a mad man. Nothing happens :-(

The key stroke is valid at any time? I'm pressing Ctrl-A-A-A (without
releasing the Ctrl) when dom0 has booted. Also tried before dom0 kernel
started. Nothing so far :-(


Regards,
  Sven



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Ian Campbell]

.On Mon, 2011-09-12 at 17:40 -0400, Sven Köhler wrote:
> > You could also try on the Xen hypervisor line (Ctlr-A three times) try the 'R'
> > and see if it does anything.
> 
> As long as I have been using Xen, I have never managed to make this work
> (getting to the hypervisor line by pressing Ctrl-A I mean). I have never
> seen the prompt, if there is any.
> 
> Right now, I'm pressing Ctrl-A like a mad man. Nothing happens :-(
> 
> The key stroke is valid at any time? I'm pressing Ctrl-A-A-A (without
> releasing the Ctrl) when dom0 has booted. Also tried before dom0 kernel
> started. Nothing so far :-(

This sequence is only valid on serial console not on VGA console if
that's what you are doing.

Ian.

Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Sven Köhler]

Am 12.09.2011 17:06, schrieb Konrad Rzeszutek Wilk:
> On Sun, Sep 11, 2011 at 02:44:35AM +0200, Sven Köhler wrote:
>> Am 11.09.2011 02:28, schrieb Konrad Rzeszutek Wilk:
>>> You might want to try some parameters on the Xen line to alter how
>>> it is suppose to reboot.
>>>
>>> /*
>>>  * reboot=b[ios] | t[riple] | k[bd] | n[o] [, [w]arm | [c]old]
>>
>> Thanks for the list.
>> I guess, both reboot=bios and reboot=b is accepted?
>> BTW: "no" is missing in the list below. acpi is missing in the list
>> above. And actually what's the source for list?
> 
> Xen hypervisor source. I just did a quick search for 'reboot='

I checked the sources of Linux 3.1rc4 and he xen hypervisor (4.1.1).
The code for reboot is almost the same. One tiny difference is that the
code of xen sets the reset flag of the kbd controller 100 times, while
Linux does that only 10 times:

Xen:
>             for ( i = 0; i < 100; i++ )
>             {
>                 kb_wait();
>                 udelay(50);
>                 outb(0xfe,0x64); /* pulse reset low */
>                 udelay(50);
>             }

Linux:
>                         for (i = 0; i < 10; i++) {
>                                 kb_wait();
>                                 udelay(50);
>                                 outb(0xfe, 0x64); /* pulse reset low */
>                                 udelay(50);
>                         }


Summing up, both Linux 3.1 and Xen 4.1 both do the following sequence by
default:

ACPI, KBD, ACPI, KBD, TRIPLE, KBD, TRIPLE, KBD, ...

While each KBD stands for 10 (Linux) or 100 (Xen) times setting the kbd
controller reset flag. I wonder why Xen does the kbd controller reset a
hundred times. Maybe it's a left over from xen 3.x?

Would you mind changing it from 100 to 10?


Now taking a look at Xen 3.4.2, the default reboot sequence is a bit
different. It's

ACPI, KBD, ACPI, KBD, ACPI, KBD, ACPI, KBD, ....

No triple fault reset attempts.

Re: Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Konrad Rzeszutek Wilk]

On Sun, Sep 18, 2011 at 09:27:03PM +0200, Sven Köhler wrote:
> Am 12.09.2011 17:06, schrieb Konrad Rzeszutek Wilk:
> > On Sun, Sep 11, 2011 at 02:44:35AM +0200, Sven Köhler wrote:
> >> Am 11.09.2011 02:28, schrieb Konrad Rzeszutek Wilk:
> >>> You might want to try some parameters on the Xen line to alter how
> >>> it is suppose to reboot.
> >>>
> >>> /*
> >>>  * reboot=b[ios] | t[riple] | k[bd] | n[o] [, [w]arm | [c]old]
> >>
> >> Thanks for the list.
> >> I guess, both reboot=bios and reboot=b is accepted?
> >> BTW: "no" is missing in the list below. acpi is missing in the list
> >> above. And actually what's the source for list?
> > 
> > Xen hypervisor source. I just did a quick search for 'reboot='
> 
> I checked the sources of Linux 3.1rc4 and he xen hypervisor (4.1.1).
> The code for reboot is almost the same. One tiny difference is that the
> code of xen sets the reset flag of the kbd controller 100 times, while
> Linux does that only 10 times:
> 
> Xen:
> >             for ( i = 0; i < 100; i++ )
> >             {
> >                 kb_wait();
> >                 udelay(50);
> >                 outb(0xfe,0x64); /* pulse reset low */
> >                 udelay(50);
> >             }
> 
> Linux:
> >                         for (i = 0; i < 10; i++) {
> >                                 kb_wait();
> >                                 udelay(50);
> >                                 outb(0xfe, 0x64); /* pulse reset low */
> >                                 udelay(50);
> >                         }
> 
> 
> Summing up, both Linux 3.1 and Xen 4.1 both do the following sequence by
> default:
> 
> ACPI, KBD, ACPI, KBD, TRIPLE, KBD, TRIPLE, KBD, ...
> 
> While each KBD stands for 10 (Linux) or 100 (Xen) times setting the kbd
> controller reset flag. I wonder why Xen does the kbd controller reset a
> hundred times. Maybe it's a left over from xen 3.x?

Could be.
> 
> Would you mind changing it from 100 to 10?

Does it fix the problem with this particular board? If so, then that sounds
like we should do.
> 
> 
> Now taking a look at Xen 3.4.2, the default reboot sequence is a bit
> different. It's
> 
> ACPI, KBD, ACPI, KBD, ACPI, KBD, ACPI, KBD, ....
> 
> No triple fault reset attempts.

Ah, could be that we just never had it implemented then.
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: Re: 3.0.4 and 3.1-rc4 based dom0 won't boot with acpi=off

[Sven Köhler]

[-- Attachment #1.1: Type: text/plain, Size: 810 bytes --]

Am 21.09.2011 20:06, schrieb Konrad Rzeszutek Wilk:
> On Sun, Sep 18, 2011 at 09:27:03PM +0200, Sven Köhler wrote:
>>
>> Would you mind changing it from 100 to 10?
> 
> Does it fix the problem with this particular board? If so, then that sounds
> like we should do.

I still have to check that. I will let you know.
(I was hoping, that being more similar to what Linux 3.x does, would be
a sufficient reason.)

>> Now taking a look at Xen 3.4.2, the default reboot sequence is a bit
>> different. It's
>>
>> ACPI, KBD, ACPI, KBD, ACPI, KBD, ACPI, KBD, ....
>>
>> No triple fault reset attempts.
> 
> Ah, could be that we just never had it implemented then.

You never had what implemented?
TRIPLE reset was implemented. However, by the default it was never used.


Regards,
  Sven


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel