x86/io: Make readq() / writeq() API consistent

Despite the following commit:

  93093d0 ("x86: provide readq()/writeq() on 32-bit too, complete")

which says:

  ...Also, map all the APIs to the strongest ordering variant. It's way
  too easy to mess such details up in drivers and the difference between
  "memory" and "" constrained asm() constructs is in the noise range.

... we have for now only one user of this API (i.e. writeq_relaxed() in
drivers/hwtracing/intel_th/sth.c) on x86 and it does care about
"relaxed" part of it.

Moreover 32-bit support has been removed from that header, though appeared
later in specific headers that emphasizes its non-atomic context.

The rest should keep in mind a consistent picture of the __raw_IO() vs. IO()
vs. IO_relaxed() API.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Baolin Wang <baolin.wang@spreadtrum.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mika Westerberg <mika.westerberg@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: intel-gfx@lists.freedesktop.org
Cc: linux-i2c@vger.kernel.org
Cc: wsa@the-dreams.de
Link: http://lkml.kernel.org/r/20170630170934.83028-6-andriy.shevchenko@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/include/asm/io.h

@@ -94,13 +94,15 @@ build_mmio_write(__writel, "l", unsigned int, "r", )
 #ifdef CONFIG_X86_64
 
 build_mmio_read(readq, "q", unsigned long, "=r", :"memory")
+build_mmio_read(__readq, "q", unsigned long, "=r", )
 build_mmio_write(writeq, "q", unsigned long, "r", :"memory")
+build_mmio_write(__writeq, "q", unsigned long, "r", )
 
-#define readq_relaxed(a)	readq(a)
-#define writeq_relaxed(v, a)	writeq(v, a)
+#define readq_relaxed(a)	__readq(a)
+#define writeq_relaxed(v, a)	__writeq(v, a)
 
-#define __raw_readq(a)		readq(a)
-#define __raw_writeq(val, addr)	writeq(val, addr)
+#define __raw_readq		__readq
+#define __raw_writeq		__writeq
 
 /* Let people know that we have them */
 #define readq			readq