Re: [RFC PATCH 14/23] thermal: intel: hfi: Update the class of the current task

[Ricardo Neri]

On Tue, Sep 27, 2022 at 01:46:59PM +0200, Peter Zijlstra wrote:
> On Fri, Sep 09, 2022 at 04:11:56PM -0700, Ricardo Neri wrote:
> > +union hfi_thread_feedback_char_msr {
> > +	struct {
> > +		u8	classid;
> > +		u64	__reserved:55;
> > +		u8	valid:1;
> > +	} split;
> > +	u64 full;
> > +};
> 
> Urgh, did you perhaps mean:
> 
> 	struct {
> 		u64	classid    :8;
> 		u64	__reserved :55;
> 		u64	valid      :1
> 	};
> 
> ?
> 
> Because yes, GCC does fold that into a single u64, but that's
> implementation defined behaviour; the C spec doesn't require one to pack
> adjacent bitfields of different types together.
> 
> I layout of:
> 
> 	u8	class;			// offset 0
> 	u64	__reserver : 55;	// offset 8
> 	u8	valid : 1;		// offset 16
> 
> with a total size of 24 bytes is, AFAIU, a valid result of what you
> wrote.

I checked the C99 and C11 specs and, IIUC, it does prescribe how to handle
adjacent bit-fields:

	"An implementation may allocate any addressable storage unit large
	 enough to hold a bitfield. If enough space remains, a bit-field
	 that immediately follows another bit-field in a structure shall be
	 packed into adjacent bits of the same unit."

Hence, __reserved and valid should be packed. classid, however, it is not
guaranteed to be adjacent to __reserved.

I will implement the struct are you have described.

Thanks and BR,
Ricardo