Enabling AHCI on ICH7 without BIOS support

Enabling AHCI on ICH7 without BIOS support

[Jonathan Bastien-Filiatrault]

Hello,


I am the owner of a Dell Latitude D620 notebook with an ICH7 SATA 
controller, unfortunately, the BIOS does not offer the option of using 
"native mode" instead of "compatibility mode" when it comes to SATA. For 
now, the ata_piix runs happily in compatibility mode. Looking at the 
ICH7 docs, I found that poking register 0x90 in PCI configuration space 
with value 0x40 should enable AHCI. However, there are some issues with 
poking that register with setpci in an early initramfs:


- The kernel does not seem to update the device id from PCI config 
space, which changes with the controller mode (it would seem the PCI 
subsystem keeps a copy)

- Poking PCI registers from userspace is far from ideal


It would be great if the AHCI driver (or otherwise) had a 
"force_ich7_ahci" to properly setup and initialize this device in full 
"native" AHCI mode. This option would be greatly welcome from users like 
me who have a stupid bios (eh, who doesn't).


Thank you,

Jonathan

Re: Enabling AHCI on ICH7 without BIOS support

[Tejun Heo]

Hello,

Jonathan Bastien-Filiatrault wrote:
> Hello,
> 
> I am the owner of a Dell Latitude D620 notebook with an ICH7 SATA
> controller, unfortunately, the BIOS does not offer the option of using
> "native mode" instead of "compatibility mode" when it comes to SATA. For
> now, the ata_piix runs happily in compatibility mode. Looking at the
> ICH7 docs, I found that poking register 0x90 in PCI configuration space
> with value 0x40 should enable AHCI. However, there are some issues with
> poking that register with setpci in an early initramfs:
> 
> 
> - The kernel does not seem to update the device id from PCI config
> space, which changes with the controller mode (it would seem the PCI
> subsystem keeps a copy)
> 
> - Poking PCI registers from userspace is far from ideal
> 
> 
> It would be great if the AHCI driver (or otherwise) had a
> "force_ich7_ahci" to properly setup and initialize this device in full
> "native" AHCI mode. This option would be greatly welcome from users like
> me who have a stupid bios (eh, who doesn't).

Sorry but most BIOSen don't map ABAR unless it supports AHCI and there
usually isn't a hole to fit in ABAR afterwards unless the kernel
completely reallocates IO resources, which, again unfortunately, can't
be done reliably cuz machines can be quirky and might puke if addresses
aren't laid out as their BIOSen would do.

So, you'll have to live with ata_piix.

-- 
tejun

Re: Enabling AHCI on ICH7 without BIOS support

[Jonathan Bastien-Filiatrault]

Tejun Heo wrote:

> Hello,
>
> Jonathan Bastien-Filiatrault wrote:
>   
>> Hello,
>>
>> I am the owner of a Dell Latitude D620 notebook with an ICH7 SATA
>> controller, unfortunately, the BIOS does not offer the option of using
>> "native mode" instead of "compatibility mode" when it comes to SATA. For
>> now, the ata_piix runs happily in compatibility mode. Looking at the
>> ICH7 docs, I found that poking register 0x90 in PCI configuration space
>> with value 0x40 should enable AHCI. However, there are some issues with
>> poking that register with setpci in an early initramfs:
>>
>>
>> - The kernel does not seem to update the device id from PCI config
>> space, which changes with the controller mode (it would seem the PCI
>> subsystem keeps a copy)
>>
>> - Poking PCI registers from userspace is far from ideal
>>
>>
>> It would be great if the AHCI driver (or otherwise) had a
>> "force_ich7_ahci" to properly setup and initialize this device in full
>> "native" AHCI mode. This option would be greatly welcome from users like
>> me who have a stupid bios (eh, who doesn't).
>>     
>
> Sorry but most BIOSen don't map ABAR unless it supports AHCI and there
> usually isn't a hole to fit in ABAR afterwards unless the kernel
> completely reallocates IO resources, which, again unfortunately, can't
> be done reliably cuz machines can be quirky and might puke if addresses
> aren't laid out as their BIOSen would do.
>
> So, you'll have to live with ata_piix.
>
>   
So far, I have managed to patch up the AHCI driver to initialize the 
ICH7 AHCI state machine as the BIOS would do (as per the AHCI spec 
instructions). This involves mapping ABAR and such. So far, I have 
gotten the AHCI driver to detect the chip properly, but it does not 
detect any drives. It looks like I will need to further initialize the 
chip. I think there might be a chance of success, but it will involve 
manually specifying which ports are present via a module parameter.


Thank you,
Jonathan

Re: Enabling AHCI on ICH7 without BIOS support

[Tejun Heo]

Jonathan Bastien-Filiatrault wrote:
> So far, I have managed to patch up the AHCI driver to initialize the
> ICH7 AHCI state machine as the BIOS would do (as per the AHCI spec
> instructions). This involves mapping ABAR and such. So far, I have
> gotten the AHCI driver to detect the chip properly, but it does not
> detect any drives. It looks like I will need to further initialize the
> chip. I think there might be a chance of success, but it will involve
> manually specifying which ports are present via a module parameter.

You probably need to program PCS.

-- 
tejun

Re: Enabling AHCI on ICH7 without BIOS support

[ISHIKAWA Mutsumi]

>>>>> In <46C44A05.9040302@dastyle.net> 
>>>>>	Jonathan Bastien-Filiatrault <joe@dastyle.net> wrote:
>> Tejun Heo wrote:
>>
>> > Hello,
>> >
>> > Jonathan Bastien-Filiatrault wrote:
>> >   
>> >> Hello,
>> >>
>> >> I am the owner of a Dell Latitude D620 notebook with an ICH7 SATA
>> >> controller, unfortunately, the BIOS does not offer the option of using
>> >> "native mode" instead of "compatibility mode" when it comes to SATA. For
>> >> now, the ata_piix runs happily in compatibility mode. Looking at the
>> >> ICH7 docs, I found that poking register 0x90 in PCI configuration space
>> >> with value 0x40 should enable AHCI. However, there are some issues with
>> >> poking that register with setpci in an early initramfs:
>> >>
>> >>
>> >> - The kernel does not seem to update the device id from PCI config
>> >> space, which changes with the controller mode (it would seem the PCI
>> >> subsystem keeps a copy)
>> >>
>> >> - Poking PCI registers from userspace is far from ideal
>> >>
>> >>
>> >> It would be great if the AHCI driver (or otherwise) had a
>> >> "force_ich7_ahci" to properly setup and initialize this device in full
>> >> "native" AHCI mode. This option would be greatly welcome from users like
>> >> me who have a stupid bios (eh, who doesn't).
>> >>     
>> >
>> > Sorry but most BIOSen don't map ABAR unless it supports AHCI and there
>> > usually isn't a hole to fit in ABAR afterwards unless the kernel
>> > completely reallocates IO resources, which, again unfortunately, can't
>> > be done reliably cuz machines can be quirky and might puke if addresses
>> > aren't laid out as their BIOSen would do.
>> >
>> > So, you'll have to live with ata_piix.
>> >
>> >   
>> So far, I have managed to patch up the AHCI driver to initialize the 
>> ICH7 AHCI state machine as the BIOS would do (as per the AHCI spec 
>> instructions). This involves mapping ABAR and such. So far, I have 
>> gotten the AHCI driver to detect the chip properly, but it does not 
>> detect any drives. It looks like I will need to further initialize the 
>> chip. I think there might be a chance of success, but it will involve 
>> manually specifying which ports are present via a module parameter.

 Do you know Edvin's patch?

http://www.bughost.org/pipermail/power/2007-June/000538.html

 I've tested it and works fine on my note
(Sony VAIO type S VGN-SZ92PS, ICH7-M)

-- 
ISHIKAWA Mutsumi
 <ishikawa@fonfun.co.jp>, <ishikawa@debian.org>

Re: Enabling AHCI on ICH7 without BIOS support

[Tejun Heo]

ISHIKAWA Mutsumi wrote:
>>>>>> In <46C44A05.9040302@dastyle.net> 
>>>>>> 	Jonathan Bastien-Filiatrault <joe@dastyle.net> wrote:
>>> Tejun Heo wrote:
>>>
>>>> Hello,
>>>>
>>>> Jonathan Bastien-Filiatrault wrote:
>>>>   
>>>>> Hello,
>>>>>
>>>>> I am the owner of a Dell Latitude D620 notebook with an ICH7 SATA
>>>>> controller, unfortunately, the BIOS does not offer the option of using
>>>>> "native mode" instead of "compatibility mode" when it comes to SATA. For
>>>>> now, the ata_piix runs happily in compatibility mode. Looking at the
>>>>> ICH7 docs, I found that poking register 0x90 in PCI configuration space
>>>>> with value 0x40 should enable AHCI. However, there are some issues with
>>>>> poking that register with setpci in an early initramfs:
>>>>>
>>>>>
>>>>> - The kernel does not seem to update the device id from PCI config
>>>>> space, which changes with the controller mode (it would seem the PCI
>>>>> subsystem keeps a copy)
>>>>>
>>>>> - Poking PCI registers from userspace is far from ideal
>>>>>
>>>>>
>>>>> It would be great if the AHCI driver (or otherwise) had a
>>>>> "force_ich7_ahci" to properly setup and initialize this device in full
>>>>> "native" AHCI mode. This option would be greatly welcome from users like
>>>>> me who have a stupid bios (eh, who doesn't).
>>>>>     
>>>> Sorry but most BIOSen don't map ABAR unless it supports AHCI and there
>>>> usually isn't a hole to fit in ABAR afterwards unless the kernel
>>>> completely reallocates IO resources, which, again unfortunately, can't
>>>> be done reliably cuz machines can be quirky and might puke if addresses
>>>> aren't laid out as their BIOSen would do.
>>>>
>>>> So, you'll have to live with ata_piix.
>>>>
>>>>   
>>> So far, I have managed to patch up the AHCI driver to initialize the 
>>> ICH7 AHCI state machine as the BIOS would do (as per the AHCI spec 
>>> instructions). This involves mapping ABAR and such. So far, I have 
>>> gotten the AHCI driver to detect the chip properly, but it does not 
>>> detect any drives. It looks like I will need to further initialize the 
>>> chip. I think there might be a chance of success, but it will involve 
>>> manually specifying which ports are present via a module parameter.
> 
>  Do you know Edvin's patch?
> 
> http://www.bughost.org/pipermail/power/2007-June/000538.html
> 
>  I've tested it and works fine on my note
> (Sony VAIO type S VGN-SZ92PS, ICH7-M)

Yeah, I've seen it but I'm not sure whether it would worth the risk.
For one, there are BIOSen which won't assign IO memory region to ABAR at
all.  There also is a patch to reallocate IO memory regions and make
room for ABAR but that can break things.

Thanks.

-- 
tejun

Re: Enabling AHCI on ICH7 without BIOS support

[Török Edwin]

Tejun Heo <htejun <at> gmail.com> writes:

> ISHIKAWA Mutsumi wrote:
> >  Do you know Edvin's patch?
> > 
> > http://www.bughost.org/pipermail/power/2007-June/000538.html
> > 
> >  I've tested it and works fine on my note
> > (Sony VAIO type S VGN-SZ92PS, ICH7-M)

Thanks for testing.

> 
> Yeah, I've seen it but I'm not sure whether it would worth the risk.
> For one, there are BIOSen which won't assign IO memory region to ABAR at
> all.  There also is a patch to reallocate IO memory regions and make
> room for ABAR but that can break things.
> 

It works fine on 2.6.22 (except that you loose the CD-ROM, and I found no way to
reliably reactivate it), however since 2.6.23-rc it breaks.
I get misrouted IRQs: IRQ 12 (keyboard) is being fired instead of IRQ 18 (ahci),
the kernel disables IRQ 18 because of spurious IRQs, and then IRQ 0 gets fired
instead of IRQ 18. It is an edge-triggered interrupt, and I get the feeling  the
interrupt status isn't cleared after handling it, since I continously get
messages about misrouted IRQs even if the system idle.
It works with irqpoll, but I don't have to tell you how slow that is.

When I'll be back from holiday I will try to compare lspci -xxx outputs, and
then maybe try a git-bisect.

Bottom line is that force-enabling AHCI like this can stop working at any time
with a new kernel version.
If there is a document describing _exactly_ what steps I need to follow to
properly enable the device, I'd be happy to read it, but I guess such
documentation is only available to BIOS developers.

--Edwin

Re: Enabling AHCI on ICH7 without BIOS support

[Tejun Heo]

Török Edwin wrote:
> If there is a document describing _exactly_ what steps I need to follow to
> properly enable the device, I'd be happy to read it, but I guess such
> documentation is only available to BIOS developers.

It all depends on how BIOS set up the controller / IO regions during
boot and even worse BIOS might expect certain configuration when ACPI
suspend/resume or shutdown transition is requested.  I don't really
think we can include AHCI force-enable in the mainline.  It's just too
risky without too much benefit.

-- 
tejun