EFI Boot Stub Early Console

EFI Boot Stub Early Console

[Shea Levy]

Hello,

How do I get the early boot console to show up on my screen when booting 
with the EFI boot stub? I can see it through the serial port if I set 
console=tty0, and I see the second half of the boot sequence after my 
initrd loads fbcon, but the screen remains unchanging until then. I'm 
using 3.3-rc7 on Virtualbox's EFI implementation.

Thanks,
Shea

Re: EFI Boot Stub Early Console

[Shea Levy]

On 3/12/12 2:56 PM, Shea Levy wrote:
> console=tty0

Sorry, that should have been console=ttyS0

Re: EFI Boot Stub Early Console

[H. Peter Anvin]

On 03/12/2012 12:00 PM, Shea Levy wrote:
> On 3/12/12 2:56 PM, Shea Levy wrote:
>> console=tty0
> 
> Sorry, that should have been console=ttyS0

Have you tried earlyprintk=serial,ttyS0?

	-hpa

Re: EFI Boot Stub Early Console

[Shea Levy]

On 3/12/12 6:11 PM, H. Peter Anvin wrote:
> On 03/12/2012 12:00 PM, Shea Levy wrote:
>> On 3/12/12 2:56 PM, Shea Levy wrote:
>>> console=tty0
>> Sorry, that should have been console=ttyS0
> Have you tried earlyprintk=serial,ttyS0?
>
> 	-hpa
>

Just tried. Still nothing in the VirtualBox window, serial port showed 
boot output up until "[    0.000000] console [tty0] enabled, bootconsole 
disabled", when the rest of the boot process happened in VirtualBox.

~Shea

Re: EFI Boot Stub Early Console

[H. Peter Anvin]

On 03/12/2012 03:18 PM, Shea Levy wrote:
> On 3/12/12 6:11 PM, H. Peter Anvin wrote:
>> On 03/12/2012 12:00 PM, Shea Levy wrote:
>>> On 3/12/12 2:56 PM, Shea Levy wrote:
>>>> console=tty0
>>> Sorry, that should have been console=ttyS0
>> Have you tried earlyprintk=serial,ttyS0?
>>
>>     -hpa
>>
> 
> Just tried. Still nothing in the VirtualBox window, serial port showed
> boot output up until "[    0.000000] console [tty0] enabled, bootconsole
> disabled", when the rest of the boot process happened in VirtualBox.
> 

That sounds like one would expect, no?

I don't think anyone has implemented an EFI earlyconsole yet, and you
can't rely on EFI handing over the screen in a mode the kernel can
handle with earlyconsole=vga, so there aren't really any other alternatives.

	-hpa

Re: EFI Boot Stub Early Console

[Shea Levy]

On 3/12/12 6:23 PM, H. Peter Anvin wrote:
> I don't think anyone has implemented an EFI earlyconsole yet, and you
> can't rely on EFI handing over the screen in a mode the kernel can
> handle with earlyconsole=vga, so there aren't really any other alternatives.

Ah, ok, so what I'm looking for doesn't exist (yet?). Thanks.

~Shea

Re: EFI Boot Stub Early Console

[Shea Levy]

On 3/12/12 6:23 PM, H. Peter Anvin wrote:
> I don't think anyone has implemented an EFI earlyconsole yet

I'm considering taking a shot at this. Do you have any pointers about where to look or how to approach it? Will I have to use UGA/GOP or is there some way to take advantage of the text console after exit_boot_services?

Cheers,
Shea

Re: EFI Boot Stub Early Console

[Shea Levy]

> Would keep_bootcon make any difference to your setup?
>

Just tried this, didn't help. Thanks though.

> Thanks,
> Nish
>

~Shea

Re: EFI Boot Stub Early Console

[Matt Fleming]

On Sat, 2012-03-17 at 09:12 -0400, Shea Levy wrote:
> On 3/12/12 6:23 PM, H. Peter Anvin wrote:
> > I don't think anyone has implemented an EFI earlyconsole yet
> 
> I'm considering taking a shot at this. Do you have any pointers about
> where to look or how to approach it? Will I have to use UGA/GOP or is
> there some way to take advantage of the text console after
> exit_boot_services?

I don't think you'll need to use the graphics protocols, you should just
be able to put something together using the 'output_string' function
from the EFI simple text output protocol, e.g. something like this,

	https://lkml.org/lkml/2012/3/16/193

Re: EFI Boot Stub Early Console

[Shea Levy]

Hi Matt,

On 3/18/12 7:21 AM, Matt Fleming wrote:
> I don't think you'll need to use the graphics protocols, you should 
> just be able to put something together using the 'output_string' 
> function from the EFI simple text output protocol, e.g. something like 
> this, https://lkml.org/lkml/2012/3/16/193 

 From the UEFI 2.3.1 spec:

> Prior to a call to ExitBootServices(), all of the fields of the EFI 
> System Table are valid. After an operating system has taken control of 
> the platform with a call to ExitBootServices(), only the Hdr, 
> FirmwareVendor, FirmwareRevision, RuntimeServices, 
> NumberOfTableEntries, and ConfigurationTable fields are valid.

Doesn't this mean that ConOut can't be used for the earlyprintk, since 
any bootloader (including the bootstub) will call ExitBootServices() 
before any printk'ing actually happens?? Or am I misunderstanding the 
spec somehow? I suppose I could just try it and see what happens...

Cheers,
Shea

Re: EFI Boot Stub Early Console

[Matt Fleming]

On Sun, 2012-03-18 at 08:10 -0400, Shea Levy wrote:
> Hi Matt,
> 
> On 3/18/12 7:21 AM, Matt Fleming wrote:
> > I don't think you'll need to use the graphics protocols, you should 
> > just be able to put something together using the 'output_string' 
> > function from the EFI simple text output protocol, e.g. something like 
> > this, https://lkml.org/lkml/2012/3/16/193 
> 
>  From the UEFI 2.3.1 spec:
> 
> > Prior to a call to ExitBootServices(), all of the fields of the EFI 
> > System Table are valid. After an operating system has taken control of 
> > the platform with a call to ExitBootServices(), only the Hdr, 
> > FirmwareVendor, FirmwareRevision, RuntimeServices, 
> > NumberOfTableEntries, and ConfigurationTable fields are valid.
> 
> Doesn't this mean that ConOut can't be used for the earlyprintk, since 
> any bootloader (including the bootstub) will call ExitBootServices() 
> before any printk'ing actually happens?? Or am I misunderstanding the 
> spec somehow? I suppose I could just try it and see what happens...

Oops. Good point, that isn't going to work.

Re: EFI Boot Stub Early Console

[Alan Cox]

> > any bootloader (including the bootstub) will call ExitBootServices() 
> > before any printk'ing actually happens?? Or am I misunderstanding the 
> > spec somehow? I suppose I could just try it and see what happens...
> 
> Oops. Good point, that isn't going to work.

Surely it works fine if you are using bootstub directly from EFI ?

Re: EFI Boot Stub Early Console

[Shea Levy]

On 3/19/12 11:56 AM, Alan Cox wrote:
>>> any bootloader (including the bootstub) will call ExitBootServices()
>>> before any printk'ing actually happens?? Or am I misunderstanding the
>>> spec somehow? I suppose I could just try it and see what happens...
>> Oops. Good point, that isn't going to work.
> Surely it works fine if you are using bootstub directly from EFI ?

ConOut will work before the call to ExitBootServices(), but any 
bootloader (including the boot stub) calls that before starting the 
normal kernel boot process. I'm working on a generic earlyprintk that 
provides the same functionality as the existing earlyprintks that will 
work for any EFI boot. Matt already submitted a patch for an efi_printk 
that will be sufficient for the boot stub itself before the call to 
ExitBootServices().

~Shea

Re: EFI Boot Stub Early Console

[Matt Fleming]

On Mon, 2012-03-19 at 15:56 +0000, Alan Cox wrote:
> > > any bootloader (including the bootstub) will call ExitBootServices() 
> > > before any printk'ing actually happens?? Or am I misunderstanding the 
> > > spec somehow? I suppose I could just try it and see what happens...
> > 
> > Oops. Good point, that isn't going to work.
> 
> Surely it works fine if you are using bootstub directly from EFI ?

Yeah, up until the point you call ExitBootServices(), which is very
early in the boot process. After that you can't rely on the validity of
the console pointers in the system table.

Re: EFI Boot Stub Early Console

[Maarten Lankhorst]

Hey Matt,

Op 19-03-12 17:03, Matt Fleming schreef:
> On Mon, 2012-03-19 at 15:56 +0000, Alan Cox wrote:
>>>> any bootloader (including the bootstub) will call ExitBootServices() 
>>>> before any printk'ing actually happens?? Or am I misunderstanding the 
>>>> spec somehow? I suppose I could just try it and see what happens...
>>> Oops. Good point, that isn't going to work.
>> Surely it works fine if you are using bootstub directly from EFI ?
> Yeah, up until the point you call ExitBootServices(), which is very
> early in the boot process. After that you can't rely on the validity of
> the console pointers in the system table.
>
I think my system spontaneously rebooted when I tried to
do some output right after ExitBootServices when debugging
the efi stub, so not only is this unreliable, it will fail to boot for
at least some systems out there.

~Maarten

Re: EFI Boot Stub Early Console

[H. Peter Anvin]

On 03/20/2012 08:45 AM, Maarten Lankhorst wrote:
> Hey Matt,
> 
> Op 19-03-12 17:03, Matt Fleming schreef:
>> On Mon, 2012-03-19 at 15:56 +0000, Alan Cox wrote:
>>>>> any bootloader (including the bootstub) will call ExitBootServices() 
>>>>> before any printk'ing actually happens?? Or am I misunderstanding the 
>>>>> spec somehow? I suppose I could just try it and see what happens...
>>>> Oops. Good point, that isn't going to work.
>>> Surely it works fine if you are using bootstub directly from EFI ?
>> Yeah, up until the point you call ExitBootServices(), which is very
>> early in the boot process. After that you can't rely on the validity of
>> the console pointers in the system table.
>>
> I think my system spontaneously rebooted when I tried to
> do some output right after ExitBootServices when debugging
> the efi stub, so not only is this unreliable, it will fail to boot for
> at least some systems out there.
> 

I would expect it to fail for just about any system.

	-hpa


-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.