efifb not detected on Intel DQ67SW

efifb not detected on Intel DQ67SW

[Andrew Lutomirski]

Hi-

For mostly masochistic reasons, I boot my DQ67SW box in UEFI mode.
efifb doesn't detect the anything, so I have no console until i915
loads.

Looking at the code, I'm a little confused how it's supposed to work.
AFAICT, unless there's a DMI match, then the driver will only load
ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.

Should there be something that looks at
EFI_GRAPHICS_OUTPUT_PROTOCOL_MODE to find the frame buffer?  (Does
that not work on Apple machines?)

--Andy

Re: efifb not detected on Intel DQ67SW

[Matthew Garrett]

On Wed, May 25, 2011 at 02:21:17PM -0400, Andrew Lutomirski wrote:

> Looking at the code, I'm a little confused how it's supposed to work.
> AFAICT, unless there's a DMI match, then the driver will only load
> ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.

grub should be setting that.

> Should there be something that looks at
> EFI_GRAPHICS_OUTPUT_PROTOCOL_MODE to find the frame buffer?  (Does
> that not work on Apple machines?)

By the time we get into the kernel we've called ExitBootServices(), so 
at that point we can't do anything with GOP. grub should have grabbed 
that information and stashed it in the appropriate field.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: efifb not detected on Intel DQ67SW

[Andrew Lutomirski]

On Wed, May 25, 2011 at 2:32 PM, Matthew Garrett <mjg@redhat.com> wrote:
> On Wed, May 25, 2011 at 02:21:17PM -0400, Andrew Lutomirski wrote:
>
>> Looking at the code, I'm a little confused how it's supposed to work.
>> AFAICT, unless there's a DMI match, then the driver will only load
>> ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.
>
> grub should be setting that.

It looks like grub-fedora (the git version, anyway) has a function
set_kernel_params that tries to do this.  Peter, the git tree claims
that you maintain it.  Are there any experiments you'd like me to do?
(Dumping boot_params, perhaps?)

--Andy

Re: efifb not detected on Intel DQ67SW

[Andrew Lutomirski]

On Wed, May 25, 2011 at 2:54 PM, Andrew Lutomirski <luto@mit.edu> wrote:
> On Wed, May 25, 2011 at 2:32 PM, Matthew Garrett <mjg@redhat.com> wrote:
>> On Wed, May 25, 2011 at 02:21:17PM -0400, Andrew Lutomirski wrote:
>>
>>> Looking at the code, I'm a little confused how it's supposed to work.
>>> AFAICT, unless there's a DMI match, then the driver will only load
>>> ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.
>>
>> grub should be setting that.
>
> It looks like grub-fedora (the git version, anyway) has a function
> set_kernel_params that tries to do this.  Peter, the git tree claims
> that you maintain it.  Are there any experiments you'd like me to do?
> (Dumping boot_params, perhaps?)

It's a bug in grub.  If I have a splashimage then efifb works.  If I
don't, then efifb doesn't.  If I don't hear anything back soon, I'll
file a bug against Fedora.

FWIW, efifb seems really slow.

--Andy

>
> --Andy
>

Re: efifb not detected on Intel DQ67SW

[Peter Jones]

On 05/26/2011 12:09 AM, Andrew Lutomirski wrote:
> On Wed, May 25, 2011 at 2:54 PM, Andrew Lutomirski <luto@mit.edu> wrote:
>> On Wed, May 25, 2011 at 2:32 PM, Matthew Garrett <mjg@redhat.com> wrote:
>>> On Wed, May 25, 2011 at 02:21:17PM -0400, Andrew Lutomirski wrote:
>>>
>>>> Looking at the code, I'm a little confused how it's supposed to work.
>>>> AFAICT, unless there's a DMI match, then the driver will only load
>>>> ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.
>>>
>>> grub should be setting that.
>>
>> It looks like grub-fedora (the git version, anyway) has a function
>> set_kernel_params that tries to do this.  Peter, the git tree claims
>> that you maintain it.  Are there any experiments you'd like me to do?
>> (Dumping boot_params, perhaps?)
> 
> It's a bug in grub.  If I have a splashimage then efifb works.  If I
> don't, then efifb doesn't.  If I don't hear anything back soon, I'll
> file a bug against Fedora.

That's expected, though I admit it's not the best thing ever. We're currently
working on not needing that bootloader at all any more, in favor of grub2,
which doesn't have this limitation.

> FWIW, efifb seems really slow.

It's a non-accelerated console.  Of course it's slow.

-- 
        Peter

Re: efifb not detected on Intel DQ67SW

[Andrew Lutomirski]

On Thu, May 26, 2011 at 9:56 AM, Peter Jones <pjones@redhat.com> wrote:
> On 05/26/2011 12:09 AM, Andrew Lutomirski wrote:
>> On Wed, May 25, 2011 at 2:54 PM, Andrew Lutomirski <luto@mit.edu> wrote:
>>> On Wed, May 25, 2011 at 2:32 PM, Matthew Garrett <mjg@redhat.com> wrote:
>>>> On Wed, May 25, 2011 at 02:21:17PM -0400, Andrew Lutomirski wrote:
>>>>
>>>>> Looking at the code, I'm a little confused how it's supposed to work.
>>>>> AFAICT, unless there's a DMI match, then the driver will only load
>>>>> ifthe boot code sets VIDEO_TYPE_EFI, but nothing sets that.
>>>>
>>>> grub should be setting that.
>>>
>>> It looks like grub-fedora (the git version, anyway) has a function
>>> set_kernel_params that tries to do this.  Peter, the git tree claims
>>> that you maintain it.  Are there any experiments you'd like me to do?
>>> (Dumping boot_params, perhaps?)
>>
>> It's a bug in grub.  If I have a splashimage then efifb works.  If I
>> don't, then efifb doesn't.  If I don't hear anything back soon, I'll
>> file a bug against Fedora.
>
> That's expected, though I admit it's not the best thing ever. We're currently
> working on not needing that bootloader at all any more, in favor of grub2,
> which doesn't have this limitation.
>
>> FWIW, efifb seems really slow.
>
> It's a non-accelerated console.  Of course it's slow.

Yes, but it's really really amazingly slow.  It's so slow that merely
enabling it doubles the time it takes to get i915 loaded.  A one-liner
to fix it is coming right now...

--Andy