Re: [PATCH] libata queue updated

[Ingo Oeser <ioe-lkml@rameria.de>]

Hi,

On Monday 30 January 2006 09:44, Tejun Heo wrote:
> So, are you saying....
> 
> struct ata_classes {
> 	unsigned int classes[2];
> |;
> 
> is safer than
> 
> unsigned int *class;
> 
> ?
 
No, but with a little bit of additional code it CAN be safer.

Or maybe, we can store the classification in a different way.

What about putting the information directly into "ap->device[INDEX].class" 
in the sole caller (ata_drive_probe_reset) so far?

> > So please let the core layer pass a bounded array here or provide
> > a function from core layer to set that and check the index.
> > 
> 
> Can you show me what you have in mind as code?
 
/* Define this to 15, if you need to */
#define ATA_MAX_CLASSES 2
struct ata_set {
        unsigned int class[ATA_MAX_CLASSES];
};

void set_ata_class(struct ata_set *cls, unsigned int idx, unsigned int what)
{
        BUG_ON(idx >= ARRAY_SIZE(cls->class);
        cls->class[idx] = what;
}

set_ata_class(&myclass, 0, what);

You can enforce that even better by making "what" 
a typedef like we do it with pte/pmd/pud/pgd in the VM.

But I prefer not passing this class stuff around, which would even safe
arguments and thus reduce code size.

Maybe we should even have a classify ata port operation instead?


Regards

Ingo Oeser