Added support for the new CRISv32 architecture.

3 files changed, 116 insertions, 31 deletions

diff --git a/ldso/ldso/cris/dl-startup.h b/ldso/ldso/cris/dl-startup.h
index 3274b41a5..417cb5096 100644
--- a/ldso/ldso/cris/dl-startup.h
+++ b/ldso/ldso/cris/dl-startup.h
@@ -4,22 +4,43 @@
 
 /* This code fixes the stack pointer so that the dynamic linker
  * can find argc, argv and auxvt (Auxillary Vector Table).  */
+#ifdef __arch_v32
+
+asm(""					\
+"	.text\n"			\
+"	.globl _start\n"		\
+"	.type _start,@function\n"	\
+"_start:\n"				\
+"	move.d	$sp,$r10\n"		\
+"	lapc	_dl_start,$r9\n"	\
+"	jsr	$r9\n"			\
+"	nop\n"				\
+"	moveq	0,$r8\n"		\
+"	jump	$r10\n"			\
+"	move	$r8,$srp\n"		\
+"	.size _start,.-_start\n"	\
+"	.previous\n"			\
+);
+
+#else
+
 asm(""					\
 "	.text\n"			\
 "	.globl _start\n"		\
 "	.type _start,@function\n"	\
 "_start:\n"				\
-"	move.d $sp,$r10\n"		\
-"	move.d $pc,$r9\n"		\
-"	add.d _dl_start - ., $r9\n"	\
-"	jsr $r9\n"			\
-"	moveq 0,$r8\n"			\
-"	move $r8,$srp\n"		\
-"	jump $r10\n"			\
+"	move.d	$sp,$r10\n"		\
+"	move.d	$pc,$r9\n"		\
+"	add.d	_dl_start - ., $r9\n"	\
+"	jsr	$r9\n"			\
+"	moveq	0,$r8\n"		\
+"	move	$r8,$srp\n"		\
+"	jump	$r10\n"			\
 "	.size _start,.-_start\n"	\
 "	.previous\n"			\
 );
 
+#endif /* __arch_v32 */
 
 /* Get a pointer to the argv array.  On many platforms this can be just
  * the address if the first argument, on other platforms we need to
diff --git a/ldso/ldso/cris/dl-sysdep.h b/ldso/ldso/cris/dl-sysdep.h
index b58f5e0a3..b5fa89035 100644
--- a/ldso/ldso/cris/dl-sysdep.h
+++ b/ldso/ldso/cris/dl-sysdep.h
@@ -68,8 +68,32 @@ elf_machine_load_address(void)
 {
 	Elf32_Addr gotaddr_diff;
 
+#ifdef __arch_v32
+	extern char ___CRISv32_dummy[] __asm__ ("_dl_start");
+
+	__asm__ ("addo.w _dl_start:GOT16,$r0,$acr\n\t"
+	         "lapc _dl_start,%0\n\t"
+	         "sub.d [$acr],%0"
+	         /* For v32, we need to force GCC to have R0 loaded with
+	            _GLOBAL_OFFSET_TABLE_ at this point, which might not
+	            otherwise have happened in the caller.  (For v10, it's
+	            loaded for non-global variables too, so we don't need
+	            anything special there.)  We accomplish this by faking the
+	            address of a global variable (as seen by GCC) as input to
+	            the asm; that address calculation goes through the GOT.
+	            Use of this function happens before we've filled in the
+	            GOT, so the address itself will not be correctly
+	            calculated, therefore we don't use any symbol whose
+	            address may be re-used later on.  Let's just reuse the
+	            _dl_start symbol, faking it as a global by renaming it as
+	            another variable through an asm.  */
+	         : "=r" (gotaddr_diff)
+	         : "g" (___CRISv32_dummy)
+	         : "acr");
+#else
 	__asm__ ("sub.d [$r0+_dl_start:GOT16],$r0,%0\n\t"
 	         "add.d _dl_start:GOTOFF,%0" : "=r" (gotaddr_diff));
+#endif
 	return gotaddr_diff;
 }
 
diff --git a/ldso/ldso/cris/resolve.S b/ldso/ldso/cris/resolve.S
index 9d2f079fe..9ca27b5ef 100644
--- a/ldso/ldso/cris/resolve.S
+++ b/ldso/ldso/cris/resolve.S
@@ -17,33 +17,73 @@
 .globl _dl_linux_resolve
 .type _dl_linux_resolve,@function
 
+#ifdef __arch_v32
+
+_dl_linux_resolve:
+	subq	4,$sp
+	move.d	$r0,[$sp]
+	subq	4,$sp
+	move.d	$r13,[$sp]
+	subq	4,$sp
+	move.d	$r12,[$sp]
+	subq	4,$sp
+	move.d	$r11,[$sp]
+	subq	4,$sp
+	addoq	5*4,$sp,$acr
+	move.d	$r10,[$sp]
+	subq	4,$sp
+	move	$mof,$r10
+	move.d	$r9,[$sp]
+	subq	4,$sp
+	move.d	[$acr],$r11
+	move	$srp,[$sp]
+	lapc	_GLOBAL_OFFSET_TABLE_,$r0
+	move.d	_dl_linux_resolver:PLTG,$r9
+	add.d	$r0,$r9
+	jsr	$r9
+	nop
+	move.d	$r10,$acr
+	move	[$sp+],$srp
+	move.d	[$sp+],$r9
+	move.d	[$sp+],$r10
+	move.d	[$sp+],$r11
+	move.d	[$sp+],$r12
+	move.d	[$sp+],$r13
+	move.d	[$sp+],$r0
+	jump	$acr
+	addq	4,$sp
+
+#else
+
 _dl_linux_resolve:
-	push $r13
-	push $r12
-	push $r11
-	push $r10
-	push $r9
-	push $r0
-	push $srp
-	move.d [$sp+7*4],$r11
-	move $mof,$r10
+	push	$r13
+	push	$r12
+	push	$r11
+	push	$r10
+	push	$r9
+	push	$r0
+	push	$srp
+	move.d	[$sp+7*4],$r11
+	move	$mof,$r10
 #ifdef __PIC__
-	move.d $pc,$r0
-	sub.d .:GOTOFF,$r0
-	move.d _dl_linux_resolver:PLTG,$r9
-	add.d $r0,$r9
-	jsr $r9
+	move.d	$pc,$r0
+	sub.d	.:GOTOFF,$r0
+	move.d	_dl_linux_resolver:PLTG,$r9
+	add.d	$r0,$r9
+	jsr	$r9
 #else
-	jsr _dl_linux_resolver
+	jsr	_dl_linux_resolver
 #endif
-	move.d $r10,[$sp+7*4]
-	pop $srp
-	pop $r0
-	pop $r9
-	pop $r10
-	pop $r11
-	pop $r12
-	pop $r13
-	jump [$sp+]
+	move.d	$r10,[$sp+7*4]
+	pop	$srp
+	pop	$r0
+	pop	$r9
+	pop	$r10
+	pop	$r11
+	pop	$r12
+	pop	$r13
+	jump	[$sp+]
+
+#endif /* __arch_v32 */
 
 	.size _dl_linux_resolve, . - _dl_linux_resolve