Properly support PIC code in x86_64 code with jump tables.

7 files changed, 75 insertions, 39 deletions

diff --git a/ChangeLog b/ChangeLog
index 7fc57b39b..bc0af108c 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,7 +1,16 @@
-2012-12-11  Torbjorn Granlund  <tege@gmplib.org>
+2012-12-12  Torbjorn Granlund  <tege@gmplib.org>
 
 	* Version 5.1.0 released.
 
+	* mpn/x86_64/x86_64-defs.m4 (JMPENT): New macro.
+	* mpn/x86_64/dos64.m4: Likewise.
+	* mpn/x86_64/darwin.m4: Likewise.
+	* mpn/x86_64/mod_34lsub1.asm: Use JMPENT to properly support PIC.
+	* mpn/x86_64/mullo_basecase.asm: Likewise.
+	* mpn/x86_64/sqr_basecase.asm: Likewise.
+
+2012-12-11  Torbjorn Granlund  <tege@gmplib.org>
+
 	* mpn/x86_64/mod_34lsub1.asm: Try different jump table for the benefit
 	of broken Apple linkers.
 
diff --git a/mpn/x86_64/darwin.m4 b/mpn/x86_64/darwin.m4
index c327631e7..d14009694 100644
--- a/mpn/x86_64/darwin.m4
+++ b/mpn/x86_64/darwin.m4
@@ -32,8 +32,13 @@ dnl  Simply override the definition in x86_64-defs.m4.
 define(`CALL',`call	GSYM_PREFIX`'$1')
 
 
-define(`JUMPTABSECT', `DATA')
+define(`JUMPTABSECT', `RODATA')
 
+ifdef(`PIC',
+  `define(`JMPENT',
+	`.set	$1_tmp, $1-$2
+	.long	$1_tmp')',
+  `define(`JMPENT', `.quad	$1')')
 
 dnl  Target ABI macros.  For Darwin we override IFELF (and leave default for
 dnl  IFDOS and IFSTD).
diff --git a/mpn/x86_64/dos64.m4 b/mpn/x86_64/dos64.m4
index 74ca5489f..c1ae46c04 100644
--- a/mpn/x86_64/dos64.m4
+++ b/mpn/x86_64/dos64.m4
@@ -24,6 +24,10 @@ define(`LEA',`
 
 define(`JUMPTABSECT', `RODATA')
 
+ifdef(`PIC',
+  `define(`JMPENT', `.long	$1-$2')',
+  `define(`JMPENT', `.quad	$1')')
+
 
 dnl  Usage: FUNC_ENTRY(nregparmas)
 dnl  Usage: FUNC_EXIT()
diff --git a/mpn/x86_64/mod_34lsub1.asm b/mpn/x86_64/mod_34lsub1.asm
index 2e69be430..d088381bb 100644
--- a/mpn/x86_64/mod_34lsub1.asm
+++ b/mpn/x86_64/mod_34lsub1.asm
@@ -103,24 +103,27 @@ forloop(i,1,UNROLL-1,`dnl
 	jnc	L(top)
 
 L(end):
-C Handle 0-8 remaining limbs.  The switch code here is zany in an attempt to
-C fit the tiny envelope of code that triggers no Apple bugs
 	lea	L(tab)(%rip), %r8
+ifdef(`PIC',
+`	movslq	36(%r8,%rsi,4), %r10
+	add	%r10, %r8
+	jmp	*%r8
+',`
 	jmp	*72(%r8,%rsi,8)
-	
+')
 	JUMPTABSECT
 	ALIGN(8)
-L(tab):	.quad	L(0)
-	.quad	L(1)
-	.quad	L(2)
-	.quad	L(3)
-	.quad	L(4)
-	.quad	L(5)
-	.quad	L(6)
-	.quad	L(7)
-	.quad	L(8)
-
+L(tab):	JMPENT(	L(0), L(tab))
+	JMPENT(	L(1), L(tab))
+	JMPENT(	L(2), L(tab))
+	JMPENT(	L(3), L(tab))
+	JMPENT(	L(4), L(tab))
+	JMPENT(	L(5), L(tab))
+	JMPENT(	L(6), L(tab))
+	JMPENT(	L(7), L(tab))
+	JMPENT(	L(8), L(tab))
 	TEXT
+
 L(6):	add	(ap), %rax
 	adc	8(ap), %rcx
 	adc	16(ap), %rdx
diff --git a/mpn/x86_64/mullo_basecase.asm b/mpn/x86_64/mullo_basecase.asm
index 95ccdb5fc..a066b8256 100644
--- a/mpn/x86_64/mullo_basecase.asm
+++ b/mpn/x86_64/mullo_basecase.asm
@@ -64,23 +64,28 @@ PROLOGUE(mpn_mullo_basecase)
 	mov	(up), %rax		C u0
 	mov	(vp_param), %r8		C v0
 
-	lea	L(jmptab)(%rip), %r9
+	lea	L(tab)(%rip), %r9
+ifdef(`PIC',
+`	movslq	(%r9,%rcx,4), %r10
+	add	%r10, %r9
+	jmp	*%r9
+',`
 	jmp	*(%r9,n,8)
+')
 	JUMPTABSECT
 	ALIGN(8)
-L(jmptab):
-	.quad	0			C not allowed
-	.quad	L(1)			C 1
-	.quad	L(2)			C 2
-	.quad	L(3)			C 3
-C	.quad	L(0m4)			C 4
-C	.quad	L(1m4)			C 5
-C	.quad	L(2m4)			C 6
-C	.quad	L(3m4)			C 7
-C	.quad	L(0m4)			C 8
-C	.quad	L(1m4)			C 9
-C	.quad	L(2m4)			C 10
-C	.quad	L(3m4)			C 11
+L(tab):	JMPENT(	L(tab), L(tab))			C not allowed
+	JMPENT(	L(1), L(tab))			C 1
+	JMPENT(	L(2), L(tab))			C 2
+	JMPENT(	L(3), L(tab))			C 3
+dnl	JMPENT(	L(0m4), L(tab))			C 4
+dnl	JMPENT(	L(1m4), L(tab))			C 5
+dnl	JMPENT(	L(2m4), L(tab))			C 6
+dnl	JMPENT(	L(3m4), L(tab))			C 7
+dnl	JMPENT(	L(0m4), L(tab))			C 8
+dnl	JMPENT(	L(1m4), L(tab))			C 9
+dnl	JMPENT(	L(2m4), L(tab))			C 10
+dnl	JMPENT(	L(3m4), L(tab))			C 11
 	TEXT
 
 L(1):	imul	%r8, %rax
diff --git a/mpn/x86_64/sqr_basecase.asm b/mpn/x86_64/sqr_basecase.asm
index 412288131..9c3dd7667 100644
--- a/mpn/x86_64/sqr_basecase.asm
+++ b/mpn/x86_64/sqr_basecase.asm
@@ -99,19 +99,25 @@ PROLOGUE(mpn_sqr_basecase)
 	mov	%r14, (%rsp)
 
 	cmovg	%r8, %rcx
-	lea	L(jmptab)(%rip), %rax
+
+	lea	L(tab)(%rip), %rax
+ifdef(`PIC',
+`	movslq	(%rax,%rcx,4), %r10
+	add	%r10, %rax
+	jmp	*%rax
+',`
 	jmp	*(%rax,%rcx,8)
+')
 	JUMPTABSECT
 	ALIGN(8)
-L(jmptab):
-	.quad	L(4)
-	.quad	L(1)
-	.quad	L(2)
-	.quad	L(3)
-	.quad	L(0m4)
-	.quad	L(1m4)
-	.quad	L(2m4)
-	.quad	L(3m4)
+L(tab):	JMPENT(	L(4), L(tab))
+	JMPENT(	L(1), L(tab))
+	JMPENT(	L(2), L(tab))
+	JMPENT(	L(3), L(tab))
+	JMPENT(	L(0m4), L(tab))
+	JMPENT(	L(1m4), L(tab))
+	JMPENT(	L(2m4), L(tab))
+	JMPENT(	L(3m4), L(tab))
 	TEXT
 
 L(1):	mov	(up), %rax
diff --git a/mpn/x86_64/x86_64-defs.m4 b/mpn/x86_64/x86_64-defs.m4
index 5180ad5d8..870abb6b7 100644
--- a/mpn/x86_64/x86_64-defs.m4
+++ b/mpn/x86_64/x86_64-defs.m4
@@ -187,6 +187,10 @@ ifdef(`PIC',
 
 define(`JUMPTABSECT', `.section	.data.rel.ro.local,"aw",@progbits')
 
+ifdef(`PIC',
+  `define(`JMPENT', `.long	$1-$2')',
+  `define(`JMPENT', `.quad	$1')')
+
 
 dnl  These macros are defined just for DOS64, where they provide calling
 dnl  sequence glue code.