svn commit: r194844 - in head/sys: conf dev/ata

[Alexander Motin]

Rafal Jaworowski wrote:
> On 2009-06-25, at 12:19, Alexander Motin wrote:
>> Rafal Jaworowski wrote:
>>> =====================================================================
>>> --- head/sys/conf/files    Wed Jun 24 15:33:33 2009    (r194843)
>>> +++ head/sys/conf/files    Wed Jun 24 15:38:17 2009    (r194844)
>>> @@ -491,12 +491,12 @@ dev/ata/ata_if.m        optional ata | atacore
>>> dev/ata/ata-all.c        optional ata | atacore
>>> dev/ata/ata-lowlevel.c        optional ata | atacore
>>> dev/ata/ata-queue.c        optional ata | atacore
>>> +dev/ata/ata-dma.c        optional ata | atadma
>>> +dev/ata/ata-sata.c        optional ata | atasata
>>
>> What is atadma and atasata here, kernel options? What for are they 
>> needed? You will not be able to build most of drivers without them, 
>> while enabling them for others will not give you any benefit, just 
>> bigger code size. I think dependency must be reviewed there.
> 
> This was supposed to follow the fine grained kernel options route for 
> various ata subsystems. Both ata-dma.c and ata-sata.c seem orthogonal to 
> the rest of the ata framework (think ata controller without DMA, which 
> is often seen in embedded). They could also be made mandatory under 
> atacore, I have no problem with this approach too.

There is move for fine-grained PCI drivers modularization. But ata-dma.c 
and ata-sata.c are not a drivers and are not a kernel modules. They are 
not orthogonal, but mandatory requisites of some drivers (all PCI, plus 
may be some others).

All kernel build dependencies must be tracked without user influence. So 
please, or, as you said, add them both to the atacore, or, as I would 
prefer, ata-dma.c to atapci and any other requiring drivers, and 
ata-sata.c to atacore.

-- 
Alexander Motin