Re: Re: Move softcursor out of fbdev to fbcon

["Antonino A. Daplas" <adaplas@gmail.com>]

James Simmons wrote:
>>>>>> And the next step is to eliminate all fbcon-specific fields from
>>>>>> fb_info to another structure, such as fb_imageblit, fb_fillrect,
>>>>>> fb_cursor, fb_copyarea.  We'll have a smaller kernel size for
>>>>>> if fbcon is not enabled.
>>>>>>             
>>> Yipes!!! That will be really hard for hardware accelerated drivers. You 
>>> have to make fbdv drivers be aware of when fbcon is loaded. Then check to 
>>> see if the device maps to a range of VCs. Then register with the fbcon 
>>> layer. Really nasty. More bloat in the long run.
>>>       
>> You're making it more complicated than it sounds.  As an initial step,
>> fields in struct fb_info that are fbcon-specific are placed in another
>> struct, say struct fbcon_info. 
>>     
>
> That is what struct display os for in fbcon.
>   
Exactly.  But the drivers  do not know anything about fbcon, so it 
cannot be used.

>   
>> In register_framebuffer, both fb_info
>> and fbcon_info are registered. If CONFIG_FRAMEBUFFER_CONSOLE = n, struct
>> fbcon_info will either be null or just end up containing NULL/dummy
>> entries. The end result will be a smaller and cleaner fb_info.
>>
>> There is really no need for fbdev drivers to detect that fbcon is loaded
>> or not, since they do not know anything about it. They just need to check
>> if CONFIG_FRAMEBUFFER_CONSOLE is set or not.
>>     
>
> This would work for the case of everything being statically built into the 
> kernel. With modular fbdev drivers and in the future a modular fbcon then 
> it would get messy.
>
>   
I don't see the problem.

Tony



-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO September
19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf