]> git.saurik.com Git - apple/libc.git/blobdiff - ppc/sys/_longjmp.s
Libc-339.tar.gz
[apple/libc.git] / ppc / sys / _longjmp.s
index fdb851b05b68d3b35db260080fdc01269df07eb2..35633c552744137f50d6616446cd3f0e8daee4da 100644 (file)
@@ -3,6 +3,8 @@
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
+ * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
@@ -21,9 +23,6 @@
  * @APPLE_LICENSE_HEADER_END@
  */
 
-/*     void _longjmp(jmp_buf env, int val); */
-
-/*     int _longjmp(jmp_buf env); */
 /*
  * Copyright (c) 1998 Apple Computer, Inc. All rights reserved.
  *
  *             Created. Derived from longjmp.s
  */
 
-#include <architecture/ppc/asm_help.h>
-#include "_setjmp.h"
-
-#define        VRSave  256
+/* We use mode-independent "g" opcodes such as "lg", and/or
+ * mode-independent macros such as MI_CALL_EXTERNAL.  These expand
+ * into word operations when targeting __ppc__, and into doubleword
+ * operations when targeting __ppc64__.
+ */
+#include <architecture/ppc/mode_independent_asm.h>
 
-/* special flag bit definitions copied from /osfmk/ppc/thread_act.h */
+#include "_setjmp.h"
 
-#define floatUsedbit   1
-#define vectorUsedbit  2
+#define __APPLE_API_PRIVATE
+#include <machine/cpu_capabilities.h>
+#undef  __APPLE_API_PRIVATE
 
+#define        VRSave  256
 
-#if defined(__DYNAMIC__)
-        .data
-       .non_lazy_symbol_pointer
-       .align 2
-L_memmove$non_lazy_ptr:
-       .indirect_symbol _memmove
-       .long 0
-       .non_lazy_symbol_pointer
-       .align 2
-L__cpu_has_altivec$non_lazy_ptr:
-       .indirect_symbol __cpu_has_altivec
-       .long 0
-        .text
-#endif        
-        
-LEAF(__longjmp)
 
-        ; need to restore FPRs or VRs?
-        
-        lwz    r5,JMP_flags(r3)
-        lwz    r6,JMP_addr_at_setjmp(r3)
-        rlwinm r7,r5,0,vectorUsedbit,vectorUsedbit
-        rlwinm r8,r5,0,floatUsedbit,floatUsedbit
-        cmpw   cr1,r3,r6               ; jmp_buf still at same address?
-        cmpwi  cr3,r7,0                ; set cr3 iff VRs in use (non-volatile CR)
-        cmpwi  cr4,r8,0                ; set cr4 iff FPRs in use (non-volatile CR)
-        beq+   cr1,LRestoreVRs
+/*     int _longjmp(jmp_buf env, int val); */
+            
+MI_ENTRY_POINT(__longjmp)
+        lg      r6,JMP_addr_at_setjmp(r3)
+        lbz     r7, _COMM_PAGE_ALTIVEC(0)
+        cmpg   cr1,r3,r6           ; jmpbuf still at same address?
+        cmpwi   cr2,r7,0            ; Altivec available? (using non-volatile cr)
+        beq++   cr1,LRestoreVRs     ; jmpbuf has not moved
         
         ; jmp_buf was moved since setjmp (or is uninitialized.)
         ; We must move VRs and FPRs to be quadword aligned at present address.
         
-        stw    r3,JMP_addr_at_setjmp(r3) ; update, in case we longjmp to this again
-        mr     r31,r4                  ; save "val" arg across memmove
-        mr     r30,r3                  ; and jmp_buf ptr
+        st    r3,JMP_addr_at_setjmp(r3) ; update, in case we longjmp to this again
+        mr      r31,r4              ; save "val" arg across memmove
+        mr      r30,r3              ; and jmp_buf ptr
         addi   r3,r3,JMP_vr_base_addr
         addi   r4,r6,JMP_vr_base_addr
-        rlwinm r3,r3,0,0,27            ; r3 <- QW aligned addr where they should be
-        rlwinm r4,r4,0,0,27            ; r4 <- QW aligned addr where they originally were
-        sub    r7,r4,r6                ; r7 <- offset of VRs/FPRs within jmp_buf
-        add    r4,r30,r7               ; r4 <- where they are now
+        clrrgi  r3,r3,4             ; r3 <- QW aligned addr where they should be
+        clrrgi  r4,r4,4             ; r4 <- QW aligned addr where they originally were
+        sub     r7,r4,r6            ; r7 <- offset of VRs/FPRs within jmp_buf
+        add     r4,r30,r7           ; r4 <- where they are now
         li     r5,(JMP_buf_end - JMP_vr_base_addr)
-#if defined(__DYNAMIC__)
-        bcl     20,31,1f               ; Get pic-base
-1:      mflr    r12                    
-        addis   r12, r12, ha16(L_memmove$non_lazy_ptr - 1b)
-        lwz     r12, lo16(L_memmove$non_lazy_ptr - 1b)(r12)
-        mtctr   r12                    ; Get address left by dyld
-        bctrl
-#else
-       bl      _memmove
-#endif
-        mr     r3,r30
-        mr     r4,r31
+        
+        MI_CALL_EXTERNAL(_memmove)
+
+        mr      r3,r30              ; restore parameters
+        mr      r4,r31
         
         ; Restore VRs iff any
-        ;      cr3 - bne if VRs
-        ;      cr4 - bne if FPRs
+        ;      cr2 - beq if AltiVec not available
         
 LRestoreVRs:
-        beq+   cr3,LZeroVRSave         ; no VRs
-        lwz    r0,JMP_vrsave(r3)
+        lg      r0,JMP_vrsave(r3)   ; get VRSAVE at setjmp()
         addi   r6,r3,JMP_vr_base_addr
-        cmpwi  r0,0                    ; any live VRs?
-        mtspr  VRSave,r0
-        beq+   LRestoreFPRs
-        lvx    v20,0,r6
-        li     r7,16*1
-        lvx    v21,r7,r6
-        li     r7,16*2
-        lvx    v22,r7,r6
-        li     r7,16*3
-        lvx    v23,r7,r6
-        li     r7,16*4
-        lvx    v24,r7,r6
-        li     r7,16*5
-        lvx    v25,r7,r6
-        li     r7,16*6
-        lvx    v26,r7,r6
-        li     r7,16*7
-        lvx    v27,r7,r6
-        li     r7,16*8
-        lvx    v28,r7,r6
-        li     r7,16*9
-        lvx    v29,r7,r6
-        li     r7,16*10
-        lvx    v30,r7,r6
-        li     r7,16*11
-        lvx    v31,r7,r6
-        b      LRestoreFPRs            ; skip zeroing VRSave
-        
-        ; Zero VRSave iff Altivec is supported, but VRs were not in use
-        ; at setjmp time.  This covers the case where VRs are first used after
-        ; the setjmp but before the longjmp, and where VRSave is nonzero at
-        ; the longjmp.  We need to zero it now, or it will always remain
-        ; nonzero since they are sticky bits.
-
-LZeroVRSave:
-#if defined(__DYNAMIC__)
-        bcl    20,31,1f
-1:     mflr    r9                      ; get our address
-        addis  r6,r9,ha16(L__cpu_has_altivec$non_lazy_ptr - 1b)
-        lwz    r7,lo16(L__cpu_has_altivec$non_lazy_ptr - 1b)(r6)
-        lwz    r7,0(r7)                ; load the flag
-#else
-        lis    r7, ha16(__cpu_has_altivec)
-       lwz     r7, lo16(__cpu_has_altivec)(r7)
-#endif
-       cmpwi   r7,0
-        li     r8,0
-        beq    LRestoreFPRs            ; no Altivec, so skip
-        mtspr  VRSave,r8
+        beq--   cr2,LRestoreFPRs    ; AltiVec not available so skip
+        cmpwi  r0,0                ; any live VRs?
+        mtspr  VRSave,r0           ; update VRSAVE whether 0 or not
+        beq++  LRestoreFPRs        ; VRSAVE is 0 so no VRs to reload
+        lvx     v20,0,r6
+        li      r7,16*1
+        lvx     v21,r7,r6
+        li      r7,16*2
+        lvx     v22,r7,r6
+        li      r7,16*3
+        lvx     v23,r7,r6
+        li      r7,16*4
+        lvx     v24,r7,r6
+        li      r7,16*5
+        lvx     v25,r7,r6
+        li      r7,16*6
+        lvx     v26,r7,r6
+        li      r7,16*7
+        lvx     v27,r7,r6
+        li      r7,16*8
+        lvx     v28,r7,r6
+        li      r7,16*9
+        lvx     v29,r7,r6
+        li      r7,16*10
+        lvx     v30,r7,r6
+        li      r7,16*11
+        lvx     v31,r7,r6
         
-        ; Restore FPRs if any
-        ;      cr4 - bne iff FPRs
+        ; Restore FPRs
         
 LRestoreFPRs:
-        beq    cr4,LRestoreGPRs        ; FPRs not in use at setjmp
         addi   r6,r3,JMP_fp_base_addr
-        rlwinm r6,r6,0,0,27            ; mask off low 4 bits to qw align
-        lfd    f14,0*8(r6)
-        lfd    f15,1*8(r6)
-        lfd    f16,2*8(r6)
-        lfd    f17,3*8(r6)
-        lfd    f18,4*8(r6)
-        lfd    f19,5*8(r6)
-        lfd    f20,6*8(r6)
-        lfd    f21,7*8(r6)
-        lfd    f22,8*8(r6)
-        lfd    f23,9*8(r6)
-        lfd    f24,10*8(r6)
-        lfd    f25,11*8(r6)
-        lfd    f26,12*8(r6)
-        lfd    f27,13*8(r6)
-        lfd    f28,14*8(r6)
-        lfd    f29,15*8(r6)
-        lfd    f30,16*8(r6)
-        lfd    f31,17*8(r6)
+        lfd     f0,JMP_fpscr(r3)    ; get FPSCR from non-sliding section of jmpbuf
+        clrrgi r6,r6,4             ; mask off low 4 bits to qw align
+        lfd     f14,0*8(r6)
+        lfd     f15,1*8(r6)
+        lfd     f16,2*8(r6)
+        lfd     f17,3*8(r6)
+        lfd     f18,4*8(r6)
+        lfd     f19,5*8(r6)
+        lfd     f20,6*8(r6)
+        lfd     f21,7*8(r6)
+        lfd     f22,8*8(r6)
+        lfd     f23,9*8(r6)
+        lfd     f24,10*8(r6)
+        lfd     f25,11*8(r6)
+        lfd     f26,12*8(r6)
+        lfd     f27,13*8(r6)
+        lfd     f28,14*8(r6)
+        lfd     f29,15*8(r6)
+        lfd     f30,16*8(r6)
+        lfd     f31,17*8(r6)
+        mtfsf   0xFF,f0             ; restore entire FPSCR
         
         ; Restore GPRs
         
-LRestoreGPRs:
-       lwz r31, JMP_r31(r3)
-       /* r1, r14-r30 */
-       lwz r1,  JMP_r1 (r3)
-       lwz r2,  JMP_r2 (r3)
-       lwz r13, JMP_r13(r3)
-       lwz r14, JMP_r14(r3)
-       lwz r15, JMP_r15(r3)
-       lwz r16, JMP_r16(r3)
-       lwz r17, JMP_r17(r3)
-       lwz r18, JMP_r18(r3)
-       lwz r19, JMP_r19(r3)
-       lwz r20, JMP_r20(r3)
-       lwz r21, JMP_r21(r3)
-       lwz r22, JMP_r22(r3)
-       lwz r23, JMP_r23(r3)
-       lwz r24, JMP_r24(r3)
-       lwz r25, JMP_r25(r3)
-       lwz r26, JMP_r26(r3)
-       lwz r27, JMP_r27(r3)
-       lwz r28, JMP_r28(r3)
-       lwz r29, JMP_r29(r3)
-       lwz r30, JMP_r30(r3)
-       lwz r0, JMP_cr(r3)
-       mtcrf 0xff,r0
-       lwz r0, JMP_lr(r3)
-       mtlr r0
-       lwz r0, JMP_ctr(r3)                     ; XXX   ctr is volatile
-       mtctr r0
-       lwz r0, JMP_xer(r3)                     ; XXX   xer is volatile
-       mtxer r0
-       mr. r3, r4
-       bnelr
-       li  r3, 1
-       blr
+        lg      r5,  JMP_cr(r3)     ; r5 <- CR
+        lg      r6,  JMP_lr(r3)     ; r6 <- LR (ie, return addres)
+        cmplgi  r4,0                ; is return value 0? (not permitted)
+        lg      r1,  JMP_r1 (r3)
+        lg      r2,  JMP_r2 (r3)
+        lg      r13, JMP_r13(r3)
+        lg      r14, JMP_r14(r3)
+        lg      r15, JMP_r15(r3)
+        lg      r16, JMP_r16(r3)
+        lg      r17, JMP_r17(r3)
+        mtcrf   0x20,r5             ; restore cr2 (we only restore non-volatile CRs)
+        lg      r18, JMP_r18(r3)
+        lg      r19, JMP_r19(r3)
+        lg      r20, JMP_r20(r3)
+        lg      r21, JMP_r21(r3)
+        mtctr   r6                  ; set up return address, avoiding LR since it will mispredict
+        lg      r22, JMP_r22(r3)
+        lg      r23, JMP_r23(r3)
+        lg      r24, JMP_r24(r3)
+        lg      r25, JMP_r25(r3)
+        mtcrf   0x10,r5             ; restore cr3
+        lg      r26, JMP_r26(r3)
+        lg      r27, JMP_r27(r3)
+        lg      r28, JMP_r28(r3)
+        lg      r29, JMP_r29(r3)
+        mtcrf   0x08,r5             ; restore cr4
+        lg      r30, JMP_r30(r3)
+        lg      r31, JMP_r31(r3)
+        mr      r3,r4               ; move return code into position (cr0 is set on r4)
+        bnectr++                    ; return code was not 0
+        li      r3, 1               ; cannot return zero from longjmp(), so return 1 instead
+        bctr