powerpc/64s: Use .machine power4 around dcbt

There are multiple decodings for the "dcbt" mnemonic, so the assembler
has to pick one.

That requires passing -many to the assembler, which is not recommended.

Without -many the clang 14 / binutils 2.38 build fails with:

  arch/powerpc/kernel/exceptions-64s.S:2976: Error: junk at end of line: `0b01010'
  clang: error: assembler command failed with exit code 1 (use -v to see invocation)

Fix it by adding .machine directives around the use of dcbt to specify
which encoding is desired.

Acked-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://msgid.link/20240229122521.762431-2-mpe@ellerman.id.au

arch/powerpc/include/asm/ppc_asm.h

@@ -508,11 +508,16 @@ END_FTR_SECTION_NESTED(CPU_FTR_CELL_TB_BUG, CPU_FTR_CELL_TB_BUG, 96)
  */
 #define DCBT_BOOK3S_STOP_ALL_STREAM_IDS(scratch)	\
        lis     scratch,0x60000000@h;			\
-       dcbt    0,scratch,0b01010
+       .machine push;					\
+       .machine power4;					\
+       dcbt    0,scratch,0b01010;			\
+       .machine pop;
 
 #define DCBT_SETUP_STREAMS(from, from_parms, to, to_parms, scratch)	\
 	lis	scratch,0x8000;	/* GO=1 */				\
 	clrldi	scratch,scratch,32;					\
+	.machine push;							\
+	.machine power4;						\
 	/* setup read stream 0 */					\
 	dcbt	0,from,0b01000;		/* addr from */			\
 	dcbt	0,from_parms,0b01010;	/* length and depth from */	\
@@ -520,7 +525,8 @@ END_FTR_SECTION_NESTED(CPU_FTR_CELL_TB_BUG, CPU_FTR_CELL_TB_BUG, 96)
 	dcbtst	0,to,0b01000;		/* addr to */			\
 	dcbtst	0,to_parms,0b01010;	/* length and depth to */	\
 	eieio;								\
-	dcbt	0,scratch,0b01010;	/* all streams GO */
+	dcbt	0,scratch,0b01010;	/* all streams GO */		\
+	.machine pop;
 
 /*
  * toreal/fromreal/tophys/tovirt macros. 32-bit BookE makes them