Merge commit 'origin/davem/sparc'

35 files changed, 777 insertions, 1854 deletions

diff --git a/ChangeLog b/ChangeLog
index 27b4e35e4a..b8118dd8f0 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -2,6 +2,63 @@
 
 	* elf/elf.h (NT_X86_XSTATE): New macro.
 
+2010-02-25  David S. Miller  <davem@davemloft.net>
+
+	* sysdeps/sparc/sparc64/Makefile: Add align-cpy rule.
+	* sysdeps/sparc/sparc64/align-cpy.S: New.
+	* sysdeps/sparc/sparc64/memcpy.S (__align_cpy_1, __align_cpy_2,
+	__align_cpy_4, __align_cpy_8, __align_cpy_16): Remove.
+	* sysdeps/sparc/sparc64/sparcv9b/memcpy.S (__align_cpy_1,
+	__align_cpy_2, __align_cpy_4, __align_cpy_8, __align_cpy_16):
+	Remove.
+	* sysdeps/sparc/sparc64/sparcv9v/memcpy.S (__align_cpy_1,
+	__align_cpy_2, __align_cpy_4, __align_cpy_8, __align_cpy_16):
+	Remove.
+	* sysdeps/sparc/sparc64/sparcv9v2/memcpy.S (__align_cpy_1,
+	__align_cpy_2, __align_cpy_4, __align_cpy_8, __align_cpy_16):
+	Remove.
+
+	* sysdeps/sparc/sparc32/bcopy.c: Delete.
+	* sysdeps/sparc/sparc32/memmove.c: Delete.
+	* sysdeps/sparc/sparc32/sparcv9/bcopy.c: Delete.
+	* sysdeps/sparc/sparc32/sparcv9/memmove.c: Delete.
+	* sysdeps/sparc/sparc64/bcopy.c: Delete.
+	* sysdeps/sparc/sparc64/memmove.c: Delete.
+	* sysdeps/sparc/sparc64/memcopy.h: New.
+	* sysdeps/sparc/sparc32/memcpy.S (bcopy, memmove): Remove.
+	* sysdeps/sparc/sparc64/memcpy.S (bcopy, memmove): Likewise.
+	* sysdeps/sparc/sparc64/sparcv9b/memcpy.S (bcopy, memmove): Likewise.
+	* sysdeps/sparc/sparc64/sparcv9v/memcpy.S (bcopy, memmove): Likewise.
+	* sysdeps/sparc/sparc64/sparcv9v2/memcpy.S (bcopy, memmove): Likewise.
+
+	* sysdeps/sparc/elf/rtld-global-offsets.sym: New file.
+	* sysdeps/sparc/Makefile (csu): Add rtld-global-offsets.sym to
+	gen-as-const-headers.
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/Makefile: New file.
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memcpy.S: New file.
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memset.S: New file.
+	* sysdeps/sparc/sparc32/sparcv9/sparcv9b/memcpy.S: Move to...
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memcpy-ultra3.S: ...here.
+	* sysdeps/sparc/sparc32/sparcv9/sparcv9v/memcpy.S: Move to...
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memcpy-niagara1.S: ...here.
+	* sysdeps/sparc/sparc32/sparcv9/sparcv9v/memset.S: Move to...
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memset-niagara1.S: ...here.
+	* sysdeps/sparc/sparc32/sparcv9/sparcv9v2/memcpy.S: Move to...
+	* sysdeps/sparc/sparc32/sparcv9/multiarch/memcpy-niagara2.S: ...here.
+	* sysdeps/sparc/sparc32/sparcv9/sparcv9v2/memset.S: Removed.
+	* sysdeps/sparc/sparc64/multiarch/Makefile: New file.
+	* sysdeps/sparc/sparc64/sparcv9v/memcpy.S: Move to...
+	* sysdeps/sparc/sparc64/multiarch/memcpy-niagara1.S: ...here.
+	* sysdeps/sparc/sparc64/sparcv9v2/memcpy.S: Move to...
+	* sysdeps/sparc/sparc64/multiarch/memcpy-niagara2.S: ...here.
+	* sysdeps/sparc/sparc64/sparcv9b/memcpy.S: Move to...
+	* sysdeps/sparc/sparc64/multiarch/memcpy-ultra3.S: ...here.
+	* sysdeps/sparc/sparc64/sparcv9v/memset.S: Move to...
+	* sysdeps/sparc/sparc64/multiarch/memset-niagara1.S: ...here.
+	* sysdeps/sparc/sparc64/sparcv9v2/memset.S: Removed.
+	* sysdeps/sparc/sparc64/multiarch/memcpy.S: New file.
+	* sysdeps/sparc/sparc64/multiarch/memset.S: New file.
+
 2010-02-20  H.J. Lu  <hongjiu.lu@intel.com>
 
 	[BZ #11332]
diff --git a/sysdeps/sparc/Makefile b/sysdeps/sparc/Makefile
index 73b926554e..735e4a40db 100644
--- a/sysdeps/sparc/Makefile
+++ b/sysdeps/sparc/Makefile
@@ -10,3 +10,8 @@ endif
 ifeq ($(subdir),db2)
 CPPFLAGS += -DHAVE_SPINLOCKS=1 -DHAVE_ASSEM_SPARC_GCC=1
 endif
+
+ifeq ($(subdir),csu)
+# get offset to rtld_global._dl_hwcap
+gen-as-const-headers += rtld-global-offsets.sym
+endif
diff --git a/sysdeps/sparc/elf/rtld-global-offsets.sym b/sysdeps/sparc/elf/rtld-global-offsets.sym
new file mode 100644
index 0000000000..ff4e97f2a6
--- /dev/null
+++ b/sysdeps/sparc/elf/rtld-global-offsets.sym
@@ -0,0 +1,7 @@
+#define SHARED 1
+
+#include <ldsodefs.h>
+
+#define rtld_global_ro_offsetof(mem) offsetof (struct rtld_global_ro, mem)
+
+RTLD_GLOBAL_RO_DL_HWCAP_OFFSET	rtld_global_ro_offsetof (_dl_hwcap)
diff --git a/sysdeps/sparc/sparc32/bcopy.c b/sysdeps/sparc/sparc32/bcopy.c
deleted file mode 100644
index 9a455f33c4..0000000000
--- a/sysdeps/sparc/sparc32/bcopy.c
+++ /dev/null
@@ -1 +0,0 @@
-/* bcopy is in memcpy.S */
diff --git a/sysdeps/sparc/sparc32/memcpy.S b/sysdeps/sparc/sparc32/memcpy.S
index 6bd55c06a1..c9c7c40e81 100644
--- a/sysdeps/sparc/sparc32/memcpy.S
+++ b/sysdeps/sparc/sparc32/memcpy.S
@@ -68,45 +68,6 @@
 	stb	%t0, [%dst - offset - 0x02]; 								\
 	stb	%t1, [%dst - offset - 0x01];
 
-/* Both these macros have to start with exactly the same insn */
-#define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) 				\
-	ldd	[%src - offset - 0x20], %t0; 								\
-	ldd	[%src - offset - 0x18], %t2; 								\
-	ldd	[%src - offset - 0x10], %t4; 								\
-	ldd	[%src - offset - 0x08], %t6; 								\
-	st	%t0, [%dst - offset - 0x20]; 								\
-	st	%t1, [%dst - offset - 0x1c]; 								\
-	st	%t2, [%dst - offset - 0x18]; 								\
-	st	%t3, [%dst - offset - 0x14]; 								\
-	st	%t4, [%dst - offset - 0x10]; 								\
-	st	%t5, [%dst - offset - 0x0c]; 								\
-	st	%t6, [%dst - offset - 0x08]; 								\
-	st	%t7, [%dst - offset - 0x04];
-
-#define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) 				\
-	ldd	[%src - offset - 0x20], %t0; 								\
-	ldd	[%src - offset - 0x18], %t2; 								\
-	ldd	[%src - offset - 0x10], %t4; 								\
-	ldd	[%src - offset - 0x08], %t6; 								\
-	std	%t0, [%dst - offset - 0x20]; 								\
-	std	%t2, [%dst - offset - 0x18]; 								\
-	std	%t4, [%dst - offset - 0x10]; 								\
-	std	%t6, [%dst - offset - 0x08];
-
-#define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) 						\
-	ldd	[%src + offset + 0x00], %t0; 								\
-	ldd	[%src + offset + 0x08], %t2; 								\
-	st	%t0, [%dst + offset + 0x00]; 								\
-	st	%t1, [%dst + offset + 0x04]; 								\
-	st	%t2, [%dst + offset + 0x08]; 								\
-	st	%t3, [%dst + offset + 0x0c];
-
-#define RMOVE_SHORTCHUNK(src, dst, offset, t0, t1) 							\
-	ldub	[%src + offset + 0x00], %t0; 								\
-	ldub	[%src + offset + 0x01], %t1; 								\
-	stb	%t0, [%dst + offset + 0x00]; 								\
-	stb	%t1, [%dst + offset + 0x01];
-
 #define SMOVE_CHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, prev, shil, shir, offset2) 		\
 	ldd	[%src + offset + 0x00], %t0; 								\
 	ldd	[%src + offset + 0x08], %t2; 								\
@@ -146,312 +107,6 @@
 	.text
 	.align	4
 
-ENTRY(bcopy)
-	mov		%o0, %o3
-	mov		%o1, %o0
-	mov		%o3, %o1
-END(bcopy)
-
-ENTRY(memmove)
-	cmp		%o0, %o1
-	st		%o0, [%sp + 64]
-	bleu		9f
-	 sub		%o0, %o1, %o4
-
-	add		%o1, %o2, %o3
-	cmp		%o3, %o0
-	bleu		0f
-	 andcc		%o4, 3, %o5
-
-	add		%o1, %o2, %o1
-	add		%o0, %o2, %o0
-	bne		77f
-	 cmp		%o2, 15
-	bleu		91f
-	 andcc		%o1, 3, %g0
-	be		3f
-	 nop
-
-	andcc		%o1, 1, %g0
-	be		4f
-	 andcc		%o1, 2, %g0
-
-	ldub		[%o1 - 1], %g2
-	sub		%o1, 1, %o1
-	stb		%g2, [%o0 - 1]
-	sub		%o2, 1, %o2
-	be		3f
-	 sub		%o0, 1, %o0
-4:	lduh		[%o1 - 2], %g2
-	sub		%o1, 2, %o1
-	sth		%g2, [%o0 - 2]
-	sub		%o2, 2, %o2
-	sub		%o0, 2, %o0
-
-3:	andcc		%o1, 4, %g0
-
-	be		2f
-	 mov		%o2, %g1
-
-	ld		[%o1 - 4], %o4
-	sub		%g1, 4, %g1
-	st		%o4, [%o0 - 4]
-	sub		%o1, 4, %o1
-	sub		%o0, 4, %o0
-2:	andcc		%g1, 0xffffff80, %g6
-	be		3f
-	 andcc		%o0, 4, %g0
-
-	be		74f + 4
-5:	RMOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
-	subcc		%g6, 128, %g6
-	sub		%o1, 128, %o1
-	bne		5b
-	 sub		%o0, 128, %o0
-
-3:	andcc		%g1, 0x70, %g6
-	be		72f
-	 andcc		%g1, 8, %g0
-
-	srl		%g6, 1, %o4
-	mov		%o7, %g2
-	add		%g6, %o4, %o4
-101:	call		100f
-	 sub		%o1, %g6, %o1
-	mov		%g2, %o7
-	jmpl		%o5 + (72f - 101b), %g0
-	 sub		%o0, %g6, %o0
-
-71:	RMOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5)
-	RMOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5)
-72:	be		73f
-	 andcc		%g1, 4, %g0
-
-	ldd		[%o1 - 0x08], %g2
-	sub		%o0, 8, %o0
-	sub		%o1, 8, %o1
-	st		%g2, [%o0]
-	st		%g3, [%o0 + 0x04]
-73:	be		1f
-	 andcc		%g1, 2, %g0
-
-	ld		[%o1 - 4], %g2
-	sub		%o1, 4, %o1
-	st		%g2, [%o0 - 4]
-	sub		%o0, 4, %o0
-1:	be		1f
-	 andcc		%g1, 1, %g0
-
-	lduh		[%o1 - 2], %g2
-	sub		%o1, 2, %o1
-	sth		%g2, [%o0 - 2]
-	sub		%o0, 2, %o0
-1:	be		1f
-	 nop
-
-	ldub		[%o1 - 1], %g2
-	stb		%g2, [%o0 - 1]
-1:	retl
-	 ld		[%sp + 64], %o0
-
-74:	RMOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
-	RMOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
-	subcc		%g6, 128, %g6
-	sub		%o1, 128, %o1
-	bne		74b
-	 sub		%o0, 128, %o0
-
-	andcc		%g1, 0x70, %g6
-	be		72b
-	 andcc		%g1, 8, %g0
-
-	srl		%g6, 1, %o4
-	mov		%o7, %g2
-	add		%g6, %o4, %o4
-102:	call		100f
-	 sub		%o1, %g6, %o1
-	mov		%g2, %o7
-	jmpl		%o5 + (72b - 102b), %g0
-	 sub		%o0, %g6, %o0
-
-75:	and		%o2, 0xe, %o3
-	mov		%o7, %g2
-	sll		%o3, 3, %o4
-	sub		%o0, %o3, %o0
-103:	call		100f
-	 sub		%o1, %o3, %o1
-	mov		%g2, %o7
-	jmpl		%o5 + (76f - 103b), %g0
-	 andcc		%o2, 1, %g0
-
-	RMOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3)
-	RMOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3)
-
-76:	be		1f
-	 nop
-	ldub		[%o1 - 1], %g2
-	stb		%g2, [%o0 - 1]
-1:	retl
-	 ld		[%sp + 64], %o0
-
-91:	bne		75b
-	 andcc		%o2, 8, %g0
-
-	be		1f
-	 andcc		%o2, 4, %g0
-
-	ld		[%o1 - 0x08], %g2
-	ld		[%o1 - 0x04], %g3
-	sub		%o1, 8, %o1
-	st		%g2, [%o0 - 0x08]
-	st		%g3, [%o0 - 0x04]
-	sub		%o0, 8, %o0
-1:	b		73b
-	 mov		%o2, %g1
-
-77:	cmp		%o2, 15
-	bleu		75b
-	 andcc		%o0, 3, %g0
-	be		64f
-	 andcc		%o0, 1, %g0
-	be		63f
-	 andcc		%o0, 2, %g0
-	ldub		[%o1 - 1], %g5
-	sub		%o1, 1, %o1
-	stb		%g5, [%o0 - 1]
-	sub		%o0, 1, %o0
-	be		64f
-	 sub		%o2, 1, %o2
-
-63:	ldub		[%o1 - 1], %g5
-	sub		%o1, 2, %o1
-	stb		%g5, [%o0 - 1]
-	sub		%o0, 2, %o0
-	ldub		[%o1], %g5
-	sub		%o2, 2, %o2
-	stb		%g5, [%o0]
-64:	and		%o1, 3, %g2
-	and		%o1, -4, %o1
-	and		%o2, 0xc, %g3
-	add		%o1, 4, %o1
-	cmp		%g3, 4
-	sll		%g2, 3, %g4
-	mov		32, %g2
-	be		4f
-	 sub		%g2, %g4, %g6
-
-	blu		3f
-	 cmp		%g3, 8
-
-	be		2f
-	 srl		%o2, 2, %g3
-
-	ld		[%o1 - 4], %o3
-	add		%o0, -8, %o0
-	ld		[%o1 - 8], %o4
-	add		%o1, -16, %o1
-	b		7f
-	 add		%g3, 1, %g3
-2:	ld		[%o1 - 4], %o4
-	add		%o0, -4, %o0
-	ld		[%o1 - 8], %g1
-	add		%o1, -12, %o1
-	b		8f
-	 add		%g3, 2, %g3
-3:	ld		[%o1 - 4], %o5
-	add		%o0, -12, %o0
-	ld		[%o1 - 8], %o3
-	add		%o1, -20, %o1
-	b		6f
-	 srl		%o2, 2, %g3
-4:	ld		[%o1 - 4], %g1
-	srl		%o2, 2, %g3
-	ld		[%o1 - 8], %o5
-	add		%o1, -24, %o1
-	add		%o0, -16, %o0
-	add		%g3, -1, %g3
-
-	ld		[%o1 + 12], %o3
-5:	sll		%o5, %g4, %g2
-	srl		%g1, %g6, %g5
-	or		%g2, %g5, %g2
-	st		%g2, [%o0 + 12]
-6:	ld		[%o1 + 8], %o4
-	sll		%o3, %g4, %g2
-	srl		%o5, %g6, %g5
-	or		%g2, %g5, %g2
-	st		%g2, [%o0 + 8]
-7:	ld		[%o1 + 4], %g1
-	sll		%o4, %g4, %g2
-	srl		%o3, %g6, %g5
-	or		%g2, %g5, %g2
-	st		%g2, [%o0 + 4]
-8:	ld		[%o1], %o5
-	sll		%g1, %g4, %g2
-	srl		%o4, %g6, %g5
-	addcc		%g3, -4, %g3
-	or		%g2, %g5, %g2
-	add		%o1, -16, %o1
-	st		%g2, [%o0]
-	add		%o0, -16, %o0
-	bne,a		5b	
-	 ld		[%o1 + 12], %o3
-	sll		%o5, %g4, %g2
-	srl		%g1, %g6, %g5
-	srl		%g4, 3, %g3
-	or		%g2, %g5, %g2
-	add		%o1, %g3, %o1
-	andcc		%o2, 2, %g0
-	st		%g2, [%o0 + 12]
-	be		1f
-	 andcc		%o2, 1, %g0
-	
-	ldub		[%o1 + 15], %g5
-	add		%o1, -2, %o1
-	stb		%g5, [%o0 + 11]
-	add		%o0, -2, %o0
-	ldub		[%o1 + 16], %g5
-	stb		%g5, [%o0 + 12]
-1:	be		1f
-	 nop
-	ldub		[%o1 + 15], %g5
-	stb		%g5, [%o0 + 11]
-1:	retl
-	 ld		[%sp + 64], %o0
-
-78:	andcc		%o1, 1, %g0
-	be		4f
-	 andcc		%o1, 2, %g0
-
-	ldub		[%o1], %g2
-	add		%o1, 1, %o1
-	stb		%g2, [%o0]
-	sub		%o2, 1, %o2
-	bne		3f
-	 add		%o0, 1, %o0
-4:	lduh		[%o1], %g2
-	add		%o1, 2, %o1
-	sth		%g2, [%o0]
-	sub		%o2, 2, %o2
-	b		3f
-	 add		%o0, 2, %o0
-END(memmove)
-
 ENTRY(memcpy)		/* %o0=dst %o1=src %o2=len */
 	sub		%o0, %o1, %o4
 	st		%o0, [%sp + 64]
@@ -968,5 +623,5 @@ ENTRY(memcpy)		/* %o0=dst %o1=src %o2=len */
 110:	retl
 	 sub		%o7, %g6, %o5
 END(memcpy)
+
 libc_hidden_builtin_def (memcpy)
-libc_hidden_builtin_def (memmove)
diff --git a/sysdeps/sparc/sparc32/memmove.c b/sysdeps/sparc/sparc32/memmove.c
deleted file mode 100644
index a8d2d49948..0000000000
--- a/sysdeps/sparc/sparc32/memmove.c
+++ /dev/null
@@ -1 +0,0 @@
-/* memmove is in memcpy.S */
diff --git a/sysdeps/sparc/sparc32/sparcv9/bcopy.c b/sysdeps/sparc/sparc32/sparcv9/b