tools headers: Update the copy of x86's mem{cpy,set}_64.S used in 'perf bench'

This is to get the changes from:

  68674f94 ("x86: don't use REP_GOOD or ERMS for small memory copies")
  20f3337d ("x86: don't use REP_GOOD or ERMS for small memory clearing")

This also make the 'perf bench mem' files stop referring to the erms
versions that gone away with the above patches.

That addresses these perf tools build warning:

  Warning: Kernel ABI header at 'tools/arch/x86/lib/memcpy_64.S' differs from latest version at 'arch/x86/lib/memcpy_64.S'
  diff -u tools/arch/x86/lib/memcpy_64.S arch/x86/lib/memcpy_64.S
  Warning: Kernel ABI header at 'tools/arch/x86/lib/memset_64.S' differs from latest version at 'arch/x86/lib/memset_64.S'
  diff -u tools/arch/x86/lib/memset_64.S arch/x86/lib/memset_64.S
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/arch/x86/lib/memcpy_64.S

@@ -9,13 +9,6 @@
 
 .section .noinstr.text, "ax"
 
-/*
- * We build a jump to memcpy_orig by default which gets NOPped out on
- * the majority of x86 CPUs which set REP_GOOD. In addition, CPUs which
- * have the enhanced REP MOVSB/STOSB feature (ERMS), change those NOPs
- * to a jmp to memcpy_erms which does the REP; MOVSB mem copy.
- */
-
 /*
  * memcpy - Copy a memory block.
  *
@@ -26,17 +19,21 @@
  *
  * Output:
  * rax original destination
+ *
+ * The FSRM alternative should be done inline (avoiding the call and
+ * the disgusting return handling), but that would require some help
+ * from the compiler for better calling conventions.
+ *
+ * The 'rep movsb' itself is small enough to replace the call, but the
+ * two register moves blow up the code. And one of them is "needed"
+ * only for the return value that is the same as the source input,
+ * which the compiler could/should do much better anyway.
  */
 SYM_TYPED_FUNC_START(__memcpy)
-	ALTERNATIVE_2 "jmp memcpy_orig", "", X86_FEATURE_REP_GOOD, \
-		      "jmp memcpy_erms", X86_FEATURE_ERMS
+	ALTERNATIVE "jmp memcpy_orig", "", X86_FEATURE_FSRM
 
 	movq %rdi, %rax
 	movq %rdx, %rcx
-	shrq $3, %rcx
-	andl $7, %edx
-	rep movsq
-	movl %edx, %ecx
 	rep movsb
 	RET
 SYM_FUNC_END(__memcpy)
@@ -45,17 +42,6 @@ EXPORT_SYMBOL(__memcpy)
 SYM_FUNC_ALIAS(memcpy, __memcpy)
 EXPORT_SYMBOL(memcpy)
 
-/*
- * memcpy_erms() - enhanced fast string memcpy. This is faster and
- * simpler than memcpy. Use memcpy_erms when possible.
- */
-SYM_FUNC_START_LOCAL(memcpy_erms)
-	movq %rdi, %rax
-	movq %rdx, %rcx
-	rep movsb
-	RET
-SYM_FUNC_END(memcpy_erms)
-
 SYM_FUNC_START_LOCAL(memcpy_orig)
 	movq %rdi, %rax
 

tools/arch/x86/lib/memset_64.S

@@ -18,27 +18,22 @@
  * rdx   count (bytes)
  *
  * rax   original destination
+ *
+ * The FSRS alternative should be done inline (avoiding the call and
+ * the disgusting return handling), but that would require some help
+ * from the compiler for better calling conventions.
+ *
+ * The 'rep stosb' itself is small enough to replace the call, but all
+ * the register moves blow up the code. And two of them are "needed"
+ * only for the return value that is the same as the source input,
+ * which the compiler could/should do much better anyway.
  */
 SYM_FUNC_START(__memset)
-	/*
-	 * Some CPUs support enhanced REP MOVSB/STOSB feature. It is recommended
-	 * to use it when possible. If not available, use fast string instructions.
-	 *
-	 * Otherwise, use original memset function.
-	 */
-	ALTERNATIVE_2 "jmp memset_orig", "", X86_FEATURE_REP_GOOD, \
-		      "jmp memset_erms", X86_FEATURE_ERMS
+	ALTERNATIVE "jmp memset_orig", "", X86_FEATURE_FSRS
 
 	movq %rdi,%r9
+	movb %sil,%al
 	movq %rdx,%rcx
-	andl $7,%edx
-	shrq $3,%rcx
-	/* expand byte value  */
-	movzbl %sil,%esi
-	movabs $0x0101010101010101,%rax
-	imulq %rsi,%rax
-	rep stosq
-	movl %edx,%ecx
 	rep stosb
 	movq %r9,%rax
 	RET
@@ -48,26 +43,6 @@ EXPORT_SYMBOL(__memset)
 SYM_FUNC_ALIAS(memset, __memset)
 EXPORT_SYMBOL(memset)
 
-/*
- * ISO C memset - set a memory block to a byte value. This function uses
- * enhanced rep stosb to override the fast string function.
- * The code is simpler and shorter than the fast string function as well.
- *
- * rdi   destination
- * rsi   value (char)
- * rdx   count (bytes)
- *
- * rax   original destination
- */
-SYM_FUNC_START_LOCAL(memset_erms)
-	movq %rdi,%r9
-	movb %sil,%al
-	movq %rdx,%rcx
-	rep stosb
-	movq %r9,%rax
-	RET
-SYM_FUNC_END(memset_erms)
-
 SYM_FUNC_START_LOCAL(memset_orig)
 	movq %rdi,%r10
 

tools/include/asm/alternative.h

@@ -4,7 +4,6 @@
 
 /* Just disable it so we can build arch/x86/lib/memcpy_64.S for perf bench: */
 
-#define altinstruction_entry #
-#define ALTERNATIVE_2 #
+#define ALTERNATIVE #
 
 #endif

tools/perf/bench/mem-memcpy-x86-64-asm-def.h

@@ -7,7 +7,3 @@ MEMCPY_FN(memcpy_orig,
 MEMCPY_FN(__memcpy,
 	"x86-64-movsq",
 	"movsq-based memcpy() in arch/x86/lib/memcpy_64.S")
-
-MEMCPY_FN(memcpy_erms,
-	"x86-64-movsb",
-	"movsb-based memcpy() in arch/x86/lib/memcpy_64.S")

tools/perf/bench/mem-memcpy-x86-64-asm.S

@@ -2,7 +2,7 @@
 
 /* Various wrappers to make the kernel .S file build in user-space: */
 
-// memcpy_orig and memcpy_erms are being defined as SYM_L_LOCAL but we need it
+// memcpy_orig is being defined as SYM_L_LOCAL but we need it
 #define SYM_FUNC_START_LOCAL(name)                      \
         SYM_START(name, SYM_L_GLOBAL, SYM_A_ALIGN)
 #define memcpy MEMCPY /* don't hide glibc's memcpy() */

tools/perf/bench/mem-memset-x86-64-asm-def.h

@@ -7,7 +7,3 @@ MEMSET_FN(memset_orig,
 MEMSET_FN(__memset,
 	"x86-64-stosq",
 	"movsq-based memset() in arch/x86/lib/memset_64.S")
-
-MEMSET_FN(memset_erms,
-	"x86-64-stosb",
-	"movsb-based memset() in arch/x86/lib/memset_64.S")

tools/perf/bench/mem-memset-x86-64-asm.S

 /* SPDX-License-Identifier: GPL-2.0 */
-// memset_orig and memset_erms are being defined as SYM_L_LOCAL but we need it
+// memset_orig is being defined as SYM_L_LOCAL but we need it
 #define SYM_FUNC_START_LOCAL(name)                      \
         SYM_START(name, SYM_L_GLOBAL, SYM_A_ALIGN)
 #define memset MEMSET /* don't hide glibc's memset() */