Re: [PATCH 2/9] thunderblot: Don't pass a bitfield to FIELD_GET

[David Laight <david.laight.linux@gmail.com>]

On Wed, 10 Dec 2025 13:13:55 -0500
Yury Norov <yury.norov@gmail.com> wrote:

> On Wed, Dec 10, 2025 at 10:18:42AM +0000, David Laight wrote:
> > On Wed, 10 Dec 2025 10:41:02 +0100
> > Mika Westerberg <mika.westerberg@linux.intel.com> wrote:
> >   
> > > On Wed, Dec 10, 2025 at 09:34:03AM +0000, David Laight wrote:  
> > > > On Wed, 10 Dec 2025 06:56:17 +0100
> > > > Mika Westerberg <mika.westerberg@linux.intel.com> wrote:
> > > >     
> > > > > $subject has typo: thunderblot -> thunderbolt ;-)
> > > > > 
> > > > > On Tue, Dec 09, 2025 at 10:03:06AM +0000, david.laight.linux@gmail.com wrote:    
> > > > > > From: David Laight <david.laight.linux@gmail.com>
> > > > > > 
> > > > > > FIELD_GET needs to use __auto_type to get the value of the 'reg'
> > > > > > parameter, this can't be used with bifields.
> > > > > > 
> > > > > > FIELD_GET also want to verify the size of 'reg' so can't add zero
> > > > > > to force the type to int.
> > > > > > 
> > > > > > So add a zero here.
> > > > > > 
> > > > > > Signed-off-by: David Laight <david.laight.linux@gmail.com>
> > > > > > ---
> > > > > >  drivers/thunderbolt/tb.h | 2 +-
> > > > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/drivers/thunderbolt/tb.h b/drivers/thunderbolt/tb.h
> > > > > > index e96474f17067..7ca2b5a0f01e 100644
> > > > > > --- a/drivers/thunderbolt/tb.h
> > > > > > +++ b/drivers/thunderbolt/tb.h
> > > > > > @@ -1307,7 +1307,7 @@ static inline struct tb_retimer *tb_to_retimer(struct device *dev)
> > > > > >   */
> > > > > >  static inline unsigned int usb4_switch_version(const struct tb_switch *sw)
> > > > > >  {
> > > > > > -	return FIELD_GET(USB4_VERSION_MAJOR_MASK, sw->config.thunderbolt_version);
> > > > > > +	return FIELD_GET(USB4_VERSION_MAJOR_MASK, sw->config.thunderbolt_version + 0);      
> > > > > 
> > > > > Can't this use a cast instead? If not then can you also add a comment here
> > > > > because next someone will send a patch "fixing" the unnecessary addition.    
> > > > 
> > > > A cast can do other (possibly incorrect) conversions, adding zero is never going
> > > > to so any 'damage' - even if it looks a bit odd.
> > > > 
> > > > Actually, I suspect the best thing here is to delete USB4_VERSION_MAJOR_MASK and
> > > > just do:
> > > > 	/* The major version is in the top 3 bits */
> > > > 	return sw->config.thunderbolt_version > 5;    
> > > 
> > > You mean 
> > > 
> > > 	return sw->config.thunderbolt_version >> 5;
> > > 
> > > ?
> > > 
> > > Yes that works but I prefer then:
> > > 
> > > 	return sw->config.thunderbolt_version >> USB4_VERSION_MAJOR_SHIFT;  
> > 
> > I've put that in for the next version (without the comment line).  
> 
> FIELD_GET() is here exactly to let people to not opencode this
> error-prone bit manipulation. So, let's continue using it.
> 
> David, can you explain in details why this code needs to be fixed? Why
> and when typecast wouldn't work so that you have to use an ugly '+0'
> hack, or even drop the FIELD_GET().
> 
> My current understanding is that the existing FIELD_GET()
> implementation works well with any data types, including bitfields,
> and what you suggested in this series - does not.

The underlying issue is that FIELD_GET() does a check that the supplied 'reg'
field is large enough for the mask.
In this case the 'reg' is a bitfield and you can't use sizeof(), typeof() or
__auto_type on a bitfield.

I don't want to (for example) add zero inside FIELD_GET() (as is done
for 'val') because that would promote u8/u16 to 32 bits - making the
size check less useful.

Ok, the current version relies on how the compiler happens to treat:
	_Generic(foo->bitfield, ...)
clang seems to treat a bitfield as 'int' (so sizeof() the result is 4)
regardless of the width.
OTOH gcc treats 'u32 bits:8' as 'unsigned char', but bits:7 selects
the 'default' and then __unsigned_scaler_typeof() fails.

I don't know what the standard says - but at least one of the compilers
is buggy.

So the current code doesn't really work with bitfields at all.

An alternate change would be to make the 4 fields in DWORD (sic) 4
just u8 instead of u32 xxx:8.
This would be similar to the two u16 at the top.

Perhaps that is the best fix.

	David

> 
> If it's correct, I don't think that switching to your version is
> well-justified.
> 
> Thanks,
> Yury