[compiler-rt][ARM] Add missing PACBTI support to assembly aeabi functions (#142400)

vhscampos · web-flow · commit a59a8ae1a9de · 2025-06-10T10:20:42.000+01:00
Some of the aeabi functions were missing PACBTI support. The lack of it
resulted in exceptions at runtime if the running environment had PAC
and/or BTI enabled.

This patch adds this support. This involves the addition of PACBTI
instructions, depending on whether each of these features is enabled,
plus the saving and restoring of the PAC code that resides in r12. Some
of the common code has been put in preprocessor macros to reduce
duplication, but not all, especially since some register saving and
restoring is very specific to each context.
diff --git a/compiler-rt/lib/builtins/arm/aeabi_cdcmp.S b/compiler-rt/lib/builtins/arm/aeabi_cdcmp.S
@@ -22,7 +22,12 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push {r0-r3, r12, lr}
+#else
         push {r0-r3, lr}
+#endif
         bl __aeabi_cdcmpeq_check_nan
         cmp r0, #1
 #if defined(USE_THUMB_1)
@@ -37,8 +42,13 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
         movs r0, #0xF
         lsls r0, r0, #31
         pop {r0-r3, pc}
+#else
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop {r0-r3, r12, lr}
+        aut r12, lr, sp
 #else
         pop {r0-r3, lr}
+#endif
 
         // NaN has been ruled out, so __aeabi_cdcmple can't trap
         // Use "it ne" + unconditional branch to guarantee a supported relocation if
@@ -47,8 +57,8 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
         bne __aeabi_cdcmple
 
 #if defined(USE_THUMB_2)
-        mov ip, #APSR_C
-        msr APSR_nzcvq, ip
+        mov r12, #APSR_C
+        msr APSR_nzcvq, r12
 #else
         msr APSR_nzcvq, #APSR_C
 #endif
@@ -70,9 +80,14 @@ END_COMPILERRT_FUNCTION(__aeabi_cdcmpeq)
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmple)
+        PACBTI_LANDING
         // Per the RTABI, this function must preserve r0-r11.
         // Save lr in the same instruction for compactness
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push {r0-r3, r12, lr}
+#else
         push {r0-r3, lr}
+#endif
 
         bl __aeabi_dcmplt
         cmp r0, #1
@@ -111,9 +126,14 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cdcmple)
 
 1:
         msr APSR_nzcvq, ip
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop {r0-r3, r12, lr}
+        bxaut r12, lr, sp
+#else
         pop {r0-r3}
         POP_PC()
 #endif
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cdcmple)
 
 // int __aeabi_cdrcmple(double a, double b) {
@@ -123,6 +143,9 @@ END_COMPILERRT_FUNCTION(__aeabi_cdcmple)
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cdrcmple)
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         // Swap r0 and r2
         mov ip, r0
         mov r0, r2
diff --git a/compiler-rt/lib/builtins/arm/aeabi_cfcmp.S b/compiler-rt/lib/builtins/arm/aeabi_cfcmp.S
@@ -22,7 +22,12 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push {r0-r3, r12, lr}
+#else
         push {r0-r3, lr}
+#endif
         bl __aeabi_cfcmpeq_check_nan
         cmp r0, #1
 #if defined(USE_THUMB_1)
@@ -37,8 +42,13 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
         movs r0, #0xF
         lsls r0, r0, #31
         pop {r0-r3, pc}
+#else
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop {r0-r3, r12, lr}
+        aut r12, lr, sp
 #else
         pop {r0-r3, lr}
+#endif
 
         // NaN has been ruled out, so __aeabi_cfcmple can't trap
         // Use "it ne" + unconditional branch to guarantee a supported relocation if
@@ -47,8 +57,8 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
         bne __aeabi_cfcmple
 
 #if defined(USE_THUMB_2)
-        mov ip, #APSR_C
-        msr APSR_nzcvq, ip
+        mov r12, #APSR_C
+        msr APSR_nzcvq, r12
 #else
         msr APSR_nzcvq, #APSR_C
 #endif
@@ -70,9 +80,14 @@ END_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmple)
+        PACBTI_LANDING
         // Per the RTABI, this function must preserve r0-r11.
         // Save lr in the same instruction for compactness
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push {r0-r3, r12, lr}
+#else
         push {r0-r3, lr}
+#endif
 
         bl __aeabi_fcmplt
         cmp r0, #1
@@ -111,9 +126,14 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmple)
 
 1:
         msr APSR_nzcvq, ip
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop {r0-r3, r12, lr}
+        bxaut r12, lr, sp
+#else
         pop {r0-r3}
         POP_PC()
 #endif
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_cfcmple)
 
 // int __aeabi_cfrcmple(float a, float b) {
@@ -123,6 +143,9 @@ END_COMPILERRT_FUNCTION(__aeabi_cfcmple)
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_cfrcmple)
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         // Swap r0 and r1
         mov ip, r0
         mov r0, r1
diff --git a/compiler-rt/lib/builtins/arm/aeabi_dcmp.S b/compiler-rt/lib/builtins/arm/aeabi_dcmp.S
@@ -25,20 +25,34 @@
 #  define CONVERT_DCMP_ARGS_TO_DF2_ARGS
 #endif
 
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+#  define PROLOGUE PACBTI_LANDING        SEPARATOR \
+                   push      { r12, lr }
+#  define EPILOGUE pop       { r12, lr } SEPARATOR \
+                   bxaut     r12, lr, sp
+#elif defined(__ARM_FEATURE_BTI_DEFAULT)
+#  define PROLOGUE PACBTI_LANDING        SEPARATOR \
+                   push      { r4, lr }
+#  define EPILOGUE pop       { r4, pc }
+#else
+#  define PROLOGUE push      { r4, lr }
+#  define EPILOGUE pop       { r4, pc }
+#endif
+
 #define DEFINE_AEABI_DCMP(cond)                            \
         .syntax unified                          SEPARATOR \
         .p2align 2                               SEPARATOR \
 DEFINE_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)           \
-        push      { r4, lr }                     SEPARATOR \
+        PROLOGUE                                 SEPARATOR \
         CONVERT_DCMP_ARGS_TO_DF2_ARGS            SEPARATOR \
         bl        SYMBOL_NAME(__ ## cond ## df2) SEPARATOR \
         cmp       r0, #0                         SEPARATOR \
         b ## cond 1f                             SEPARATOR \
         movs      r0, #0                         SEPARATOR \
-        pop       { r4, pc }                     SEPARATOR \
+        EPILOGUE                                 SEPARATOR \
 1:                                               SEPARATOR \
         movs      r0, #1                         SEPARATOR \
-        pop       { r4, pc }                     SEPARATOR \
+        EPILOGUE                                 SEPARATOR \
 END_COMPILERRT_FUNCTION(__aeabi_dcmp ## cond)
 
 DEFINE_AEABI_DCMP(eq)
diff --git a/compiler-rt/lib/builtins/arm/aeabi_fcmp.S b/compiler-rt/lib/builtins/arm/aeabi_fcmp.S
@@ -25,20 +25,34 @@
 #  define CONVERT_FCMP_ARGS_TO_SF2_ARGS
 #endif
 
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+#  define PROLOGUE PACBTI_LANDING        SEPARATOR \
+                   push      { r12, lr }
+#  define EPILOGUE pop       { r12, lr } SEPARATOR \
+                   bxaut     r12, lr, sp
+#elif defined(__ARM_FEATURE_BTI_DEFAULT)
+#  define PROLOGUE PACBTI_LANDING        SEPARATOR \
+                   push      { r4, lr }
+#  define EPILOGUE pop       { r4, pc }
+#else
+#  define PROLOGUE push      { r4, lr }
+#  define EPILOGUE pop       { r4, pc }
+#endif
+
 #define DEFINE_AEABI_FCMP(cond)                            \
         .syntax unified                          SEPARATOR \
         .p2align 2                               SEPARATOR \
 DEFINE_COMPILERRT_FUNCTION(__aeabi_fcmp ## cond)           \
-        push      { r4, lr }                     SEPARATOR \
+        PROLOGUE                                 SEPARATOR \
         CONVERT_FCMP_ARGS_TO_SF2_ARGS            SEPARATOR \
         bl        SYMBOL_NAME(__ ## cond ## sf2) SEPARATOR \
         cmp       r0, #0                         SEPARATOR \
         b ## cond 1f                             SEPARATOR \
         movs      r0, #0                         SEPARATOR \
-        pop       { r4, pc }                     SEPARATOR \
+        EPILOGUE                                 SEPARATOR \
 1:                                               SEPARATOR \
         movs      r0, #1                         SEPARATOR \
-        pop       { r4, pc }                     SEPARATOR \
+        EPILOGUE                                 SEPARATOR \
 END_COMPILERRT_FUNCTION(__aeabi_fcmp ## cond)
 
 DEFINE_AEABI_FCMP(eq)
diff --git a/compiler-rt/lib/builtins/arm/aeabi_idivmod.S b/compiler-rt/lib/builtins/arm/aeabi_idivmod.S
@@ -31,7 +31,12 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_idivmod)
         subs    r1, r1, r2
         JMP     (r3)
 #else  // defined(USE_THUMB_1)
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push    { r12, lr }
+#else
         push    { lr }
+#endif
         sub     sp, sp, #4
         mov     r2, sp
 #if defined(__MINGW32__)
@@ -42,7 +47,12 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_idivmod)
         bl      SYMBOL_NAME(__divmodsi4)
         ldr     r1, [sp]
         add     sp, sp, #4
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop     { r12, lr }
+        bxaut   r12, lr, sp
+#else
         pop     { pc }
+#endif
 #endif //  defined(USE_THUMB_1)
 END_COMPILERRT_FUNCTION(__aeabi_idivmod)
 
diff --git a/compiler-rt/lib/builtins/arm/aeabi_ldivmod.S b/compiler-rt/lib/builtins/arm/aeabi_ldivmod.S
@@ -22,7 +22,12 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_ldivmod)
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push    {r6, r12, lr}
+#else
         push    {r6, lr}
+#endif
         sub     sp, sp, #16
         add     r6, sp, #8
         str     r6, [sp]
@@ -38,7 +43,12 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_ldivmod)
         ldr     r2, [sp, #8]
         ldr     r3, [sp, #12]
         add     sp, sp, #16
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop     {r6, r12, lr}
+        bxaut   r12, lr, sp
+#else
         pop     {r6, pc}
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_ldivmod)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/compiler-rt/lib/builtins/arm/aeabi_memcpy.S b/compiler-rt/lib/builtins/arm/aeabi_memcpy.S
@@ -18,6 +18,9 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_memcpy)
         bl      memcpy
         pop     {r7, pc}
 #else
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         b       memcpy
 #endif
 END_COMPILERRT_FUNCTION(__aeabi_memcpy)
diff --git a/compiler-rt/lib/builtins/arm/aeabi_memmove.S b/compiler-rt/lib/builtins/arm/aeabi_memmove.S
@@ -17,6 +17,9 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_memmove)
         bl      memmove
         pop     {r7, pc}
 #else
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         b       memmove
 #endif
 END_COMPILERRT_FUNCTION(__aeabi_memmove)
diff --git a/compiler-rt/lib/builtins/arm/aeabi_memset.S b/compiler-rt/lib/builtins/arm/aeabi_memset.S
@@ -14,6 +14,9 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_memset)
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         mov     r3, r1
         mov     r1, r2
         mov     r2, r3
@@ -31,6 +34,9 @@ DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_memset8, __aeabi_memset)
 
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_memclr)
+#if defined(__ARM_FEATURE_BTI_DEFAULT)
+        bti
+#endif
         mov     r2, r1
         movs    r1, #0
 #ifdef USE_THUMB_1
diff --git a/compiler-rt/lib/builtins/arm/aeabi_uidivmod.S b/compiler-rt/lib/builtins/arm/aeabi_uidivmod.S
@@ -38,7 +38,12 @@ LOCAL_LABEL(case_denom_larger):
         movs    r0, #0
         JMP     (lr)
 #else // defined(USE_THUMB_1)
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push    { r12, lr }
+#else
         push    { lr }
+#endif
         sub     sp, sp, #4
         mov     r2, sp
 #if defined(__MINGW32__)
@@ -49,8 +54,13 @@ LOCAL_LABEL(case_denom_larger):
         bl      SYMBOL_NAME(__udivmodsi4)
         ldr     r1, [sp]
         add     sp, sp, #4
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop     { r12, lr }
+        bxaut   r12, lr, sp
+#else
         pop     { pc }
 #endif
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_uidivmod)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/compiler-rt/lib/builtins/arm/aeabi_uldivmod.S b/compiler-rt/lib/builtins/arm/aeabi_uldivmod.S
@@ -22,7 +22,12 @@
         .syntax unified
         .p2align 2
 DEFINE_COMPILERRT_FUNCTION(__aeabi_uldivmod)
-        push	{r6, lr}
+        PACBTI_LANDING
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        push    {r6, r12, lr}
+#else
+        push    {r6, lr}
+#endif
         sub	sp, sp, #16
         add	r6, sp, #8
         str	r6, [sp]
@@ -38,7 +43,12 @@ DEFINE_COMPILERRT_FUNCTION(__aeabi_uldivmod)
         ldr	r2, [sp, #8]
         ldr	r3, [sp, #12]
         add	sp, sp, #16
-        pop	{r6, pc}
+#if defined(__ARM_FEATURE_PAC_DEFAULT)
+        pop     {r6, r12, lr}
+        bxaut   r12, lr, sp
+#else
+        pop     {r6, pc}
+#endif
 END_COMPILERRT_FUNCTION(__aeabi_uldivmod)
 
 NO_EXEC_STACK_DIRECTIVE
diff --git a/compiler-rt/lib/builtins/assembly.h b/compiler-rt/lib/builtins/assembly.h
@@ -194,6 +194,17 @@
 #else
 #define WIDE(op) op
 #endif
+
+#if defined(__ARM_FEATURE_PAC_DEFAULT) && defined(__ARM_FEATURE_BTI_DEFAULT)
+#define PACBTI_LANDING pacbti r12, lr, sp
+#elif defined(__ARM_FEATURE_PAC_DEFAULT)
+#define PACBTI_LANDING pac r12, lr, sp
+#elif defined(__ARM_FEATURE_BTI_DEFAULT)
+#define PACBTI_LANDING bti
+#else
+#define PACBTI_LANDING
+#endif
+
 #else // !defined(__arm)
 #define DECLARE_FUNC_ENCODING
 #define DEFINE_CODE_STATE
