Re: [PATCH] libata queue updated

[Tejun Heo <htejun@gmail.com>]

Ingo Oeser wrote:
> Hi Jeff,
> 
> 
> On Saturday 28 January 2006 19:25, Jeff Garzik wrote:
>> Testing and merge point in Tejun's flood of patches :)  The patch
>> below is against current linux-2.6.git.
> 
> These "function(unsigned int *classes)" style functions in 
> "libata-core.c" worry me somewhat.  Esp. that sometimes you have one class,
> sometimes two.
> This looks like a bug waiting to happen for me.
> 
> Could we somehow get a 
> 
> struct ata_classes {
> 	unsigned int master;
> 	unsigned int slave;
> }
> 
> here (or similiar), before this is in used everywhere?
> 
> Usage would be function(struct ata_classes *classes) then.
> 

Hello,

I object.  Using array is intentional.  Slave aware controllers (PATA / 
ata_piix) will use [0..1], most SATA controllers will use only [0], and 
PM aware ones will use [0..15].  The intention was requiring low level 
drivers of only what they know and normalize them in the core layer.

eg. Current std SATA reset routines consider the argument as *class (a 
single class value) and it's intentional.  As long as a lldd is aware of 
only one device per port, it's allowed/recommeded to consider the passed 
classes argument as a pointer to single class value.  The rest is upto 
the core libata layer.

-- 
tejun