Re: [PATCH v2 04/11] PCI/IDE: Enumerate Selective Stream IDE capabilities

[Dan Williams <dan.j.williams@intel.com>]

Alexey Kardashevskiy wrote:
> On 11/3/25 16:46, Aneesh Kumar K.V wrote:
> > Dan Williams <dan.j.williams@intel.com> writes:
[..]
> >> +		nr_ide_mem = nr_assoc;
> >>
> > 
> > What is the purpose of the loop?
> 
> It is to be able to implement sel_ide_offset() as simple as:
> 
> offset = PCI_IDE_LINK_STREAM_0 + nr_link_ide * PCI_IDE_LINK_BLOCK_SIZE +
> stream_index * PCI_IDE_SEL_BLOCK_SIZE(nr_ide_mem);
> 
> as each stream can potentially have different "Number of Address 
> Association Register Blocks" (which we do not want for some reason).

The cost to spec writers of allowing more degrees of freedom is zero.
The cost of supporting it in system-software is non-zero.

It is useful to test that implementations have a practical need to
maximize their complexity. That testing serves 2 purposes, it either
stops the complexity from being realized, or otherwise puts the burden
on the vendor who thinks they need that to come back and fixup the
implementation.

Meanwhile we get the relief of living in a slightly less complicated
world, if only for a moment.