Re: [RESEND PATCH 1/3] x86: Adding structs to reflect cpuid fields

["Radim Krčmář" <rkrcmar@redhat.com>]

2014-09-18 09:19+0200, Borislav Petkov:
> On Thu, Sep 18, 2014 at 02:29:54AM +0200, Radim Krčmář wrote:
> > I think you proposed to use magic constant in place of of MASK_FAM_X, so
> 
> Huh, what?

Your example.  It cannot be verbatim MASK_FAM_X in real code.

I interpreted it to be a placeholder for 0x1ffff (first 17 bits) and
said why I think it is not a good thing.
The other interpretation (well named macro) was against your goal.

> > Second problem:  Most elements don't begin at offset 0, so the usual
> > retrieval would add a shift, (repurposing max_monitor_line_size)
> 
> So what?

Your goal was easy parsing (last sentence of the mail).

My argument is that bitfields are easier to read.  (With examples.)

> >  max_monitor_line_size = (cpuid & MASK_FAM_X) >> OFFSET_FAM_X;
> > 
> > and it's not better when we write it back after doing stuff.
> 
> Writing back CPUID on baremetal? You can't be serious.

We are not talking just about baremetal, see patch [3/3].

There are, and will be new, use cases for modifying and storing cpuid.

> Ok, this is getting ridiculous so I'll stop wasting time. I stand by my
> initial statement - this is a bad idea.

Acknowledged.