"Glove box" BIOS patch causes boot flicker

"Glove box" BIOS patch causes boot flicker

[Jan Engelhardt]

Hi,


some (long) time ago, I noticed a change whereby at the start of Linux 
(before any log messages are output) the monitor would do one reset on 
x86/VGA (the classic modes à la 80x25 - no fbdev stuff). It is only now 
that I found enough joy to revisit this.

Between 2.6.30 and 2.6.31, commit cf06de7b9cdd was introduced by you, 
which I have bisected the problem down to.

commit cf06de7b9cdd3efee7a59dced1977b3c21d43732
Author: H. Peter Anvin <hpa@linux.intel.com>
Date:   Wed Apr 1 18:20:11 2009 -0700

    x86, setup: "glove box" BIOS interrupts in the video code
    
    Impact: BIOS proofing
    
    "Glove box" off BIOS interrupts in the video code.

The exact artifacts caused by it are:

- a mode set from 80x25 to 80x25...
- cursor position is reset to the top-left corner, overwriting any 
logs from previous code such as BIOSes or bootloaders. Previously, the 
linux kernel log nicely followed these, now it is intersecting.

("glove box off" is a ventured description to me. Even the German 
translation gave no hint; an image search was required. Suggesting "seal 
off".)

Anyhow, the commit description lacks a bit of information to coders 
whose area of expertise is outside hardware (net/ is what I do), and why 
this commit is needed, given all I seemingly got from it was a 
degradation of my VGA console experience.

Could you elaborate on it, and/or what could be done to address the two 
issues?

thanks,
Jan

Re: "Glove box" BIOS patch causes boot flicker

[H. Peter Anvin]

A "glove box" is a device used to handle toxic stuff.  In this case the 
toxic stuff is the BIOS.

There were some minor bugs (typos) in the conversion to the glove box 
coding style.  The ones we have found have been fixed, so the first 
thing is to try the current mainline or the current tip tree.  Otherwise 
I would look for issues where either the wrong register is changed, or 
something like a variable is changed instead of a field.



	-hpa

Re: "Glove box" BIOS patch causes boot flicker

[Jan Engelhardt]

On Wednesday 2010-09-22 02:34, H. Peter Anvin wrote:

>There were some minor bugs (typos) in the conversion to the glove box 
>coding style. The ones we have found have been fixed, so the first 
>thing is to try the current mainline or the current tip tree.  
>Otherwise I would look for issues where either the wrong register is 
>changed, or something like a variable is changed instead of a field.

I tried v2.6.36-rc5 now, and it looks better. The cursor position is 
retained as desired. The mode switch is still there, albeit much 
shorter than it used to be.

Re: "Glove box" BIOS patch causes boot flicker

[H. Peter Anvin]

On 09/23/2010 11:54 AM, Jan Engelhardt wrote:
> On Wednesday 2010-09-22 02:34, H. Peter Anvin wrote:
> 
>> There were some minor bugs (typos) in the conversion to the glove box 
>> coding style. The ones we have found have been fixed, so the first 
>> thing is to try the current mainline or the current tip tree.  
>> Otherwise I would look for issues where either the wrong register is 
>> changed, or something like a variable is changed instead of a field.
> 
> I tried v2.6.36-rc5 now, and it looks better. The cursor position is 
> retained as desired. The mode switch is still there, albeit much 
> shorter than it used to be.

The mode switch was there in the original code.

	-hpa

Re: "Glove box" BIOS patch causes boot flicker

[Jan Engelhardt]

On Thursday 2010-09-23 22:00, H. Peter Anvin wrote:
>On 09/23/2010 11:54 AM, Jan Engelhardt wrote:
>> On Wednesday 2010-09-22 02:34, H. Peter Anvin wrote:
>> 
>>> There were some minor bugs (typos) in the conversion to the glove box 
>>> coding style. The ones we have found have been fixed, so the first 
>>> thing is to try the current mainline or the current tip tree.  
>>> Otherwise I would look for issues where either the wrong register is 
>>> changed, or something like a variable is changed instead of a field.
>> 
>> I tried v2.6.36-rc5 now, and it looks better. The cursor position is 
>> retained as desired. The mode switch is still there, albeit much 
>> shorter than it used to be.
>
>The mode switch was there in the original code.

That may very well be so, but there clearly is distinguished behavior 
w.r.t. that in 2.6.36 compared to .30.

Re: "Glove box" BIOS patch causes boot flicker

[H. Peter Anvin]

On 09/23/2010 01:46 PM, Jan Engelhardt wrote:
> 
> On Thursday 2010-09-23 22:00, H. Peter Anvin wrote:
>> On 09/23/2010 11:54 AM, Jan Engelhardt wrote:
>>> On Wednesday 2010-09-22 02:34, H. Peter Anvin wrote:
>>>
>>>> There were some minor bugs (typos) in the conversion to the glove box 
>>>> coding style. The ones we have found have been fixed, so the first 
>>>> thing is to try the current mainline or the current tip tree.  
>>>> Otherwise I would look for issues where either the wrong register is 
>>>> changed, or something like a variable is changed instead of a field.
>>>
>>> I tried v2.6.36-rc5 now, and it looks better. The cursor position is 
>>> retained as desired. The mode switch is still there, albeit much 
>>> shorter than it used to be.
>>
>> The mode switch was there in the original code.
> 
> That may very well be so, but there clearly is distinguished behavior 
> w.r.t. that in 2.6.36 compared to .30.

Yes, but that caused regressions from the old assembly code.

	-hpa

Re: "Glove box" BIOS patch causes boot flicker

[H. Peter Anvin]

On 09/23/2010 01:48 PM, H. Peter Anvin wrote:
> On 09/23/2010 01:46 PM, Jan Engelhardt wrote:
>>
>> On Thursday 2010-09-23 22:00, H. Peter Anvin wrote:
>>> On 09/23/2010 11:54 AM, Jan Engelhardt wrote:
>>>> On Wednesday 2010-09-22 02:34, H. Peter Anvin wrote:
>>>>
>>>>> There were some minor bugs (typos) in the conversion to the glove box 
>>>>> coding style. The ones we have found have been fixed, so the first 
>>>>> thing is to try the current mainline or the current tip tree.  
>>>>> Otherwise I would look for issues where either the wrong register is 
>>>>> changed, or something like a variable is changed instead of a field.
>>>>
>>>> I tried v2.6.36-rc5 now, and it looks better. The cursor position is 
>>>> retained as desired. The mode switch is still there, albeit much 
>>>> shorter than it used to be.
>>>
>>> The mode switch was there in the original code.
>>
>> That may very well be so, but there clearly is distinguished behavior 
>> w.r.t. that in 2.6.36 compared to .30.
> 
> Yes, but that caused regressions from the old assembly code.
> 

By the way, if you don't want a mode reset, use "vga=current".

	-hpa