PowerPC LE setjmp/longjmp

http://sourceware.org/ml/libc-alpha/2013-08/msg00089.html

Little-endian fixes for setjmp/longjmp.  When writing these I noticed
the setjmp code corrupts the non volatile VMX registers when using an
unaligned buffer.  Anton fixed this, and also simplified it quite a
bit.

The current code uses boilerplate for the case where we want to store
16 bytes to an unaligned address.  For that we have to do a
read/modify/write of two aligned 16 byte quantities.  In our case we
are storing a bunch of back to back data (consective VMX registers),
and only the start and end of the region need the read/modify/write.

	[BZ #15723]
	* sysdeps/powerpc/jmpbuf-offsets.h: Comment fix.
	* sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S: Correct
	_dl_hwcap access for little-endian.
	* sysdeps/powerpc/powerpc32/fpu/setjmp-common.S: Likewise.  Don't
	destroy vmx regs when saving unaligned.
	* sysdeps/powerpc/powerpc64/__longjmp-common.S: Correct CR load.
	* sysdeps/powerpc/powerpc64/setjmp-common.S: Likewise CR save.  Don't
	destroy vmx regs when saving unaligned.

5 files changed, 78 insertions, 85 deletions

diff --git a/sysdeps/powerpc/jmpbuf-offsets.h b/sysdeps/powerpc/jmpbuf-offsets.h
index 64c658a588..f2116bd703 100644
--- a/sysdeps/powerpc/jmpbuf-offsets.h
+++ b/sysdeps/powerpc/jmpbuf-offsets.h
@@ -21,12 +21,10 @@
 #define JB_LR     2  /* The address we will return to */
 #if __WORDSIZE == 64
 # define JB_GPRS   3  /* GPRs 14 through 31 are saved, 18*2 words total.  */
-# define JB_CR     21 /* Condition code registers with the VRSAVE at */
-                       /* offset 172 (low half of the double word.  */
+# define JB_CR     21 /* Shared dword with VRSAVE.  CR word at offset 172.  */
 # define JB_FPRS   22 /* FPRs 14 through 31 are saved, 18*2 words total.  */
 # define JB_SIZE   (64 * 8) /* As per PPC64-VMX ABI.  */
-# define JB_VRSAVE 21 /* VRSAVE shares a double word with the CR at offset */
-                       /* 168 (high half of the double word).  */
+# define JB_VRSAVE 21 /* Shared dword with CR.  VRSAVE word at offset 168.  */
 # define JB_VRS    40 /* VRs 20 through 31 are saved, 12*4 words total.  */
 #else
 # define JB_GPRS   3  /* GPRs 14 through 31 are saved, 18 in total.  */
diff --git a/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S b/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
index 9d34cd9165..d02aa57549 100644
--- a/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
+++ b/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
@@ -43,16 +43,16 @@ ENTRY (__longjmp)
 #   endif
 	mtlr    r6
 	cfi_same_value (lr)
-	lwz     r5,RTLD_GLOBAL_RO_DL_HWCAP_OFFSET+4(r5)
+	lwz     r5,RTLD_GLOBAL_RO_DL_HWCAP_OFFSET+LOWORD(r5)
 #  else
 	lwz     r5,_dl_hwcap@got(r5)
 	mtlr    r6
 	cfi_same_value (lr)
-	lwz     r5,4(r5)
+	lwz     r5,LOWORD(r5)
 #  endif
 # else
-	lis	r5,(_dl_hwcap+4)@ha
-	lwz     r5,(_dl_hwcap+4)@l(r5)
+	lis	r5,(_dl_hwcap+LOWORD)@ha
+	lwz     r5,(_dl_hwcap+LOWORD)@l(r5)
 # endif
 	andis.	r5,r5,(PPC_FEATURE_HAS_ALTIVEC >> 16)
 	beq	L(no_vmx)
diff --git a/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S b/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
index 46ea2b00f9..f3244060e0 100644
--- a/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
+++ b/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
@@ -94,14 +94,14 @@ ENTRY (__sigsetjmp)
 #   else
 	lwz     r5,_rtld_global_ro@got(r5)
 #   endif
-	lwz     r5,RTLD_GLOBAL_RO_DL_HWCAP_OFFSET+4(r5)
+	lwz     r5,RTLD_GLOBAL_RO_DL_HWCAP_OFFSET+LOWORD(r5)
 #  else
 	lwz     r5,_dl_hwcap@got(r5)
-	lwz     r5,4(r5)
+	lwz     r5,LOWORD(r5)
 #  endif
 # else
-	lis	r6,(_dl_hwcap+4)@ha
-	lwz     r5,(_dl_hwcap+4)@l(r6)
+	lis	r6,(_dl_hwcap+LOWORD)@ha
+	lwz     r5,(_dl_hwcap+LOWORD)@l(r6)
 # endif
 	andis.	r5,r5,(PPC_FEATURE_HAS_ALTIVEC >> 16)
 	beq	L(no_vmx)
@@ -111,44 +111,43 @@ ENTRY (__sigsetjmp)
 	stw	r0,((JB_VRSAVE)*4)(3)
 	addi	r6,r5,16
 	beq+	L(aligned_save_vmx)
-	lvsr	v0,0,r5
-	vspltisb v1,-1         /* set v1 to all 1's */
-	vspltisb v2,0          /* set v2 to all 0's */
-	vperm   v3,v2,v1,v0   /* v3 contains shift mask with num all 1 bytes on left = misalignment  */
 
+	lvsr	v0,0,r5
+	lvsl	v1,0,r5
+	addi	r6,r5,-16
 
-	/* Special case for v20 we need to preserve what is in save area below v20 before obliterating it */
-	lvx     v5,0,r5
-	vperm   v20,v20,v20,v0
-	vsel    v5,v5,v20,v3
-	vsel    v20,v20,v2,v3
-	stvx    v5,0,r5
+# define save_misaligned_vmx(savevr,prevvr,shiftvr,tmpvr,savegpr,addgpr) \
+	addi	addgpr,addgpr,32;					 \
+	vperm	tmpvr,prevvr,savevr,shiftvr;				 \
+	stvx	tmpvr,0,savegpr
 
-#define save_2vmx_partial(savevr,prev_savevr,hivr,shiftvr,maskvr,savegpr,addgpr) \
-	addi    addgpr,addgpr,32; \
-	vperm   savevr,savevr,savevr,shiftvr; \
-	vsel    hivr,prev_savevr,savevr,maskvr; \
-	stvx    hivr,0,savegpr;
+	/*
+	 * We have to be careful not to corrupt the data below v20 and
+	 * above v31. To keep things simple we just rotate both ends in
+	 * the opposite direction to our main permute so we can use
+	 * the common macro.
+	 */
 
-	save_2vmx_partial(v21,v20,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v22,v21,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v23,v22,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v24,v23,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v25,v24,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v26,v25,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v27,v26,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v28,v27,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v29,v28,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v30,v29,v5,v0,v3,r5,r6)
+	/* load and rotate data below v20 */
+	lvx	v2,0,r5
+	vperm	v2,v2,v2,v1
+	save_misaligned_vmx(v20,v2,v0,v3,r5,r6)
+	save_misaligned_vmx(v21,v20,v0,v3,r6,r5)
+	save_misaligned_vmx(v22,v21,v0,v3,r5,r6)
+	save_misaligned_vmx(v23,v22,v0,v3,r6,r5)
+	save_misaligned_vmx(v24,v23,v0,v3,r5,r6)
+	save_misaligned_vmx(v25,v24,v0,v3,r6,r5)
+	save_misaligned_vmx(v26,v25,v0,v3,r5,r6)
+	save_misaligned_vmx(v27,v26,v0,v3,r6,r5)
+	save_misaligned_vmx(v28,v27,v0,v3,r5,r6)
+	save_misaligned_vmx(v29,v28,v0,v3,r6,r5)
+	save_misaligned_vmx(v30,v29,v0,v3,r5,r6)
+	save_misaligned_vmx(v31,v30,v0,v3,r6,r5)
+	/* load and rotate data above v31 */
+	lvx	v2,0,r6
+	vperm	v2,v2,v2,v1
+	save_misaligned_vmx(v2,v31,v0,v3,r5,r6)
 
-	/* Special case for r31 we need to preserve what is in save area above v31 before obliterating it */
-	addi    r5,r5,32
-	vperm   v31,v31,v31,v0
-	lvx     v4,0,r5
-	vsel    v5,v30,v31,v3
-	stvx    v5,0,r6
-	vsel    v4,v31,v4,v3
-	stvx    v4,0,r5
 	b	L(no_vmx)
 
 L(aligned_save_vmx):
diff --git a/sysdeps/powerpc/powerpc64/__longjmp-common.S b/sysdeps/powerpc/powerpc64/__longjmp-common.S
index 70c370439f..4f1e3c88d3 100644
--- a/sysdeps/powerpc/powerpc64/__longjmp-common.S
+++ b/sysdeps/powerpc/powerpc64/__longjmp-common.S
@@ -57,7 +57,7 @@ ENTRY (__longjmp)
 	beq	L(no_vmx)
 	la	r5,((JB_VRS)*8)(3)
 	andi.	r6,r5,0xf
-	lwz	r0,((JB_VRSAVE)*8)(3)
+	lwz	r0,((JB_VRSAVE)*8)(3)	/* 32-bit VRSAVE.  */
 	mtspr	VRSAVE,r0
 	beq+	L(aligned_restore_vmx)
 	addi    r6,r5,16
@@ -153,7 +153,7 @@ L(no_vmx):
 	lfd fp21,((JB_FPRS+7)*8)(r3)
 	ld r22,((JB_GPRS+8)*8)(r3)
 	lfd fp22,((JB_FPRS+8)*8)(r3)
-	ld r0,(JB_CR*8)(r3)
+	lwz r0,((JB_CR*8)+4)(r3)	/* 32-bit CR.  */
 	ld r23,((JB_GPRS+9)*8)(r3)
 	lfd fp23,((JB_FPRS+9)*8)(r3)
 	ld r24,((JB_GPRS+10)*8)(r3)
diff --git a/sysdeps/powerpc/powerpc64/setjmp-common.S b/sysdeps/powerpc/powerpc64/setjmp-common.S
index 58ec610620..1829b9ab65 100644
--- a/sysdeps/powerpc/powerpc64/setjmp-common.S
+++ b/sysdeps/powerpc/powerpc64/setjmp-common.S
@@ -95,7 +95,7 @@ JUMPTARGET(GLUE(__sigsetjmp,_ent)):
 	mfcr r0
 	std  r16,((JB_GPRS+2)*8)(3)
 	stfd fp16,((JB_FPRS+2)*8)(3)
-	std  r0,(JB_CR*8)(3)
+	stw  r0,((JB_CR*8)+4)(3)	/* 32-bit CR.  */
 	std  r17,((JB_GPRS+3)*8)(3)
 	stfd fp17,((JB_FPRS+3)*8)(3)
 	std  r18,((JB_GPRS+4)*8)(3)
@@ -139,50 +139,46 @@ JUMPTARGET(GLUE(__sigsetjmp,_ent)):
 	la	r5,((JB_VRS)*8)(3)
 	andi.	r6,r5,0xf
 	mfspr	r0,VRSAVE
-	stw	r0,((JB_VRSAVE)*8)(3)
+	stw	r0,((JB_VRSAVE)*8)(3)	/* 32-bit VRSAVE.  */
 	addi	r6,r5,16
 	beq+	L(aligned_save_vmx)
-	lvsr	v0,0,r5
-	vspltisb v1,-1         /* set v1 to all 1's */
-	vspltisb v2,0          /* set v2 to all 0's */
-	vperm   v3,v2,v1,v0   /* v3 contains shift mask with num all 1 bytes
-				 on left = misalignment  */
 
+	lvsr	v0,0,r5
+	lvsl	v1,0,r5
+	addi	r6,r5,-16
 
-	/* Special case for v20 we need to preserve what is in save area
-	   below v20 before obliterating it */
-	lvx     v5,0,r5
-	vperm   v20,v20,v20,v0
-	vsel    v5,v5,v20,v3
-	vsel    v20,v20,v2,v3
-	stvx    v5,0,r5
+# define save_misaligned_vmx(savevr,prevvr,shiftvr,tmpvr,savegpr,addgpr) \
+	addi	addgpr,addgpr,32;					 \
+	vperm	tmpvr,prevvr,savevr,shiftvr;				 \
+	stvx	tmpvr,0,savegpr
 
-# define save_2vmx_partial(savevr,prev_savevr,hivr,shiftvr,maskvr,savegpr,addgpr) \
-	addi    addgpr,addgpr,32; \
-	vperm   savevr,savevr,savevr,shiftvr; \
-	vsel    hivr,prev_savevr,savevr,maskvr; \
-	stvx    hivr,0,savegpr;
+	/*
+	 * We have to be careful not to corrupt the data below v20 and
+	 * above v31. To keep things simple we just rotate both ends in
+	 * the opposite direction to our main permute so we can use
+	 * the common macro.
+	 */
 
-	save_2vmx_partial(v21,v20,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v22,v21,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v23,v22,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v24,v23,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v25,v24,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v26,v25,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v27,v26,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v28,v27,v5,v0,v3,r5,r6)
-	save_2vmx_partial(v29,v28,v5,v0,v3,r6,r5)
-	save_2vmx_partial(v30,v29,v5,v0,v3,r5,r6)
+	/* load and rotate data below v20 */
+	lvx	v2,0,r5
+	vperm	v2,v2,v2,v1
+	save_misaligned_vmx(v20,v2,v0,v3,r5,r6)
+	save_misaligned_vmx(v21,v20,v0,v3,r6,r5)
+	save_misaligned_vmx(v22,v21,v0,v3,r5,r6)
+	save_misaligned_vmx(v23,v22,v0,v3,r6,r5)
+	save_misaligned_vmx(v24,v23,v0,v3,r5,r6)
+	save_misaligned_vmx(v25,v24,v0,v3,r6,r5)
+	save_misaligned_vmx(v26,v25,v0,v3,r5,r6)
+	save_misaligned_vmx(v27,v26,v0,v3,r6,r5)
+	save_misaligned_vmx(v28,v27,v0,v3,r5,r6)
+	save_misaligned_vmx(v29,v28,v0,v3,r6,r5)
+	save_misaligned_vmx(v30,v29,v0,v3,r5,r6)
+	save_misaligned_vmx(v31,v30,v0,v3,r6,r5)
+	/* load and rotate data above v31 */
+	lvx	v2,0,r6
+	vperm	v2,v2,v2,v1
+	save_misaligned_vmx(v2,v31,v0,v3,r5,r6)
 
-	/* Special case for r31 we need to preserve what is in save area
-	   above v31 before obliterating it */
-	addi    r5,r5,32
-	vperm   v31,v31,v31,v0
-	lvx     v4,0,r5
-	vsel    v5,v30,v31,v3
-	stvx    v5,0,r6
-	vsel    v4,v31,v4,v3
-	stvx    v4,0,r5
 	b	L(no_vmx)
 
 L(aligned_save_vmx):