typeof and operands in named address spaces

[Peter Zijlstra]

( restoring at least linux-toolchains@vger.kernel.org, since that seems
  to have gone missing )

On Mon, Nov 16, 2020 at 10:11:50AM +0100, Richard Biener wrote:
> On Sun, Nov 15, 2020 at 11:53 AM Uecker, Martin
> <Martin.Uecker@med.uni-goettingen.de> wrote:
> > > On Wed, Nov 04, 2020 at 07:31:42PM +0100, Uros Bizjak wrote:
> > > > Hello!
> > > >
> > > > I was looking at the recent linux patch series [1] where segment
> > > > qualifiers (named address spaces) were introduced to handle percpu
> > > > variables. In the patch [2], the author mentions that:
> > > >
> > > > --q--
> > > > Unfortunately, gcc does not provide a way to remove segment
> > > > qualifiers, which is needed to use typeof() to create local instances
> > > > of the per-cpu variable. For this reason, do not use the segment
> > > > qualifier for per-cpu variables, and do casting using the segment
> > > > qualifier instead.
> > > > --/q--
> > >
> > > C in general does not provide means to strip qualifiers. We recently had
> > > a _lot_ of 'fun' trying to strip volatile from a type, see here:
> > >
> > >   https://lore.kernel.org/lkml/875zimp0ay.fsf@mpe.ellerman.id.au
> > >
> > > which resulted in the current __unqual_scalar_typeof() hack.
> > >
> > > If we're going to do compiler extentions here, can we pretty please have
> > > a sane means of modifying qualifiers in general?
> >
> > Another way to drop qualifiers is using a cast. So you
> > can use typeof twice:
> >
> > typeof((typeof(_var))_var) tmp__;
> >
> > This also works for non-scalars but this is a GCC extension.
> >
> >
> > WG14 plans to standardize typeof. I would like to hear opinion
> > whether we should have typeof drop qualifiers or not.
> >
> > Currently, it does not do this on all compilers I tested
> > (except _Atomic on GCC) and there are also use cases for
> > keeping qualifiers. This is an argument for keeping qualifiers
> > should we standardize it, but then we need a way to drop
> > qualifiers.
> >
> >
> > lvalue conversion drops qualifers in C.  In GCC, this is not
> > implemented correctly as it is unobvervable in standard C
> > (but it using typeof).
> >
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97702
> >
> > A have a working patch in preparation to change this. Then you
> > could use
> >
> > typeof( ((void)0, x) )

Neat, that actually already works with clang. And I suppose we can use
the above GCC extention until such time as that GCC is fixed.

See below..

> > to drop qualifiers. But this would then
> > also do array-to-pointer conversion. I am not sure
> > whether this is a problem.

I don't _think_ so, but..

> > Of course, we could also introduce a new feature for
> > dropping qualifiers. Thoughts?

> Just add a new qualifier that un-qualifies?
> 
> _Unqual volatile T x;
> 
> is T with volatile (evenually) removed.  Or just a way to drop
> all using _Unqual?
> 
> _Unqual T x;
> 
> removing all qualifiers from T.  Or add a special _Unqual_all
> to achieve that.  I think removing a specific qualification is
> useful.  Leaves cases like
> 
> _Unqual volatile volatile T x;
> 
> to be specified (that is ordering and cancellation of the
> unqual and qual variants of qualifiers).

I rather like this, however I think I'd prefer the syntax be something
like:

	_Unqual T x;

for removing all qualifiers, and:

	_Unqual(volatile) volatile T X;

for stripping specific qualifiers. The syntax as proposed above seems
very error prone to me.


---
Subject: compiler: Improve __unqual_typeof()

Improve our __unqual_scalar_typeof() implementation by relying on C
dropping qualifiers for lvalue convesions. There is one small catch in
that GCC is currently known broken in this respect, however it happens
to have a C language extention that achieves the very same, it drops
qualifiers on casts.

This gets rid of the _Generic() usage and should improve compile times
(less preprocessor output) as well as increases the capabilities of the
macros.

XXX: I've only verified the below actually compiles, I've not verified
     the generated code is actually 'correct'.

Suggested-by: "Uecker, Martin" <Martin.Uecker@med.uni-goettingen.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
diff --git a/include/linux/compiler-gcc.h b/include/linux/compiler-gcc.h
index 74c6c0486eed..3c5cb52c12f9 100644
--- a/include/linux/compiler-gcc.h
+++ b/include/linux/compiler-gcc.h
@@ -156,3 +156,11 @@
 #else
 #define __diag_GCC_8(s)
 #endif
+
+/*
+ * GCC has a bug where lvalue conversion doesn't drop qualifiers, use a GCC
+ * extention instead. GCC drops qualifiers on a cast, so use a double typeof().
+ *
+ *  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97702
+ */
+#define __unqual_typeof(type)	typeof( (typeof(type))type )
diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
index ac3fa37a84f9..4a6e2caab17b 100644
--- a/include/linux/compiler_types.h
+++ b/include/linux/compiler_types.h
@@ -250,27 +250,14 @@ struct ftrace_likely_data {
 /* Are two types/vars the same type (ignoring qualifiers)? */
 #define __same_type(a, b) __builtin_types_compatible_p(typeof(a), typeof(b))
 
+#ifndef __unqual_typeof
 /*
- * __unqual_scalar_typeof(x) - Declare an unqualified scalar type, leaving
- *			       non-scalar types unchanged.
+ * lvalue conversion drops qualifiers
  */
-/*
- * Prefer C11 _Generic for better compile-times and simpler code. Note: 'char'
- * is not type-compatible with 'signed char', and we define a separate case.
- */
-#define __scalar_type_to_expr_cases(type)				\
-		unsigned type:	(unsigned type)0,			\
-		signed type:	(signed type)0
-
-#define __unqual_scalar_typeof(x) typeof(				\
-		_Generic((x),						\
-			 char:	(char)0,				\
-			 __scalar_type_to_expr_cases(char),		\
-			 __scalar_type_to_expr_cases(short),		\
-			 __scalar_type_to_expr_cases(int),		\
-			 __scalar_type_to_expr_cases(long),		\
-			 __scalar_type_to_expr_cases(long long),	\
-			 default: (x)))
+#define __unqual_typeof(type)	typeof( ((void)0, type) )
+#endif
+
+#define __unqual_scalar_typeof(type) __unqual_typeof(type)
 
 /* Is this type a native word size -- useful for atomic operations */
 #define __native_word(t) \