From be1e5d311342e08ae1f8013342df27b7ded2c156 Mon Sep 17 00:00:00 2001 From: Anton Blanchard Date: Sat, 17 Aug 2013 18:34:40 +0930 Subject: 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. --- sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S | 8 +-- sysdeps/powerpc/powerpc32/fpu/setjmp-common.S | 73 ++++++++++++------------ 2 files changed, 40 insertions(+), 41 deletions(-) (limited to 'sysdeps/powerpc/powerpc32/fpu') 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): -- cgit 1.4.1