[llvm-branch-commits] compiler-rt: Introduce runtime functions for emulated PAC. (PR #133530)

Jon Roelofs via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Mon Mar 31 09:11:31 PDT 2025


================
@@ -0,0 +1,115 @@
+#include <stdint.h>
+
+#define XXH_INLINE_ALL
+#define XXH_NO_STDLIB
+#define XXH_memcpy __builtin_memcpy
+#define XXH_memset __builtin_memset
+#define XXH_memcmp __builtin_memcmp
+#include "../xxhash.h"
+
+// EmuPAC implements runtime emulation of PAC instructions. If the current
+// CPU supports PAC, EmuPAC uses real PAC instructions. Otherwise, it uses the
+// emulation, which is effectively an implementation of PAC with an IMPDEF
+// hashing scheme based on XXH128.
+//
+// The purpose of the emulation is to allow programs to be built to be portable
+// to machines without PAC support, with some performance loss and increased
+// probability of false positives (due to not being able to portably determine
+// the VA size), while being functionally almost equivalent to running on a
+// machine with PAC support. One example of a use case is if PAC is used in
+// production as a security mitigation, but the testing environment is
+// heterogeneous (i.e. some machines lack PAC support). In this case we would
+// like the testing machines to be able to detect issues resulting
+// from the use of PAC instructions that would affect production by running
+// tests. This can be achieved by building test binaries with EmuPAC and
+// production binaries with real PAC.
+//
+// The emulation assumes that the VA size is at most 48 bits. The architecture
+// as of ARMv8.2, which was the last architecture version in which PAC was not
+// mandatory, permitted VA size up to 52 bits via ARMv8.2-LVA, but we are
+// unaware of an ARMv8.2 CPU that implemented ARMv8.2-LVA.
+
+const uint64_t kMaxVASize = 48;
+const uint64_t kPACMask = ((1ULL << 55) - 1) & ~((1ULL << kMaxVASize) - 1);
+const uint64_t kTTBR1Mask = 1ULL << 55;
+
+// Determine whether PAC is supported without accessing memory. This utilizes
+// the XPACLRI instruction which will copy bit 55 of x30 into at least bit 54 if
+// PAC is supported and acts as a NOP if PAC is not supported.
+static _Bool pac_supported() {
+  register uintptr_t x30 __asm__("x30") = 1ULL << 55;
+  __asm__ __volatile__("xpaclri" : "+r"(x30));
+  return x30 & (1ULL << 54);
+}
+
+// This asm snippet is used to force the creation of a frame record when
+// calling the EmuPAC functions. This is important because the EmuPAC functions
+// may crash if an auth failure is detected and may be unwound past using a
+// frame pointer based unwinder.
+#ifdef __GCC_HAVE_DWARF2_CFI_ASM
+#define frame_pointer_wrap(sym) \
+  "stp x29, x30, [sp, #-16]!\n" \
+  ".cfi_def_cfa_offset 16\n" \
+  "mov x29, sp\n" \
+  ".cfi_def_cfa w29, 16\n" \
+  ".cfi_offset w30, -8\n" \
+  ".cfi_offset w29, -16\n" \
+  "bl " #sym "\n" \
+  ".cfi_def_cfa wsp, 16\n" \
+  "ldp x29, x30, [sp], #16\n" \
+  ".cfi_def_cfa_offset 0\n" \
+  ".cfi_restore w30\n" \
+  ".cfi_restore w29\n" \
+  "ret"
+#else
+#define frame_pointer_wrap(sym) \
+  "stp x29, x30, [sp, #-16]!\n" \
+  "mov x29, sp\n" \
+  "bl " #sym "\n" \
+  "ldp x29, x30, [sp], #16\n" \
+  "ret"
+#endif
+
+uint64_t __emupac_pacda_impl(uint64_t ptr, uint64_t disc) {
+  if (pac_supported()) {
+    __asm__ __volatile__(".arch_extension pauth\npacda %0, %1"
+                         : "+r"(ptr)
+                         : "r"(disc));
+    return ptr;
+  }
+  if (ptr & kTTBR1Mask) {
+    if ((ptr & kPACMask) != kPACMask) {
+      return ptr | kPACMask;
+    }
+  } else {
+    if (ptr & kPACMask) {
+      return ptr & ~kPACMask;
+    }
+  }
+  uint64_t hash = XXH3_64bits_withSeed(&ptr, 8, disc);
+  return (ptr & ~kPACMask) | (hash & kPACMask);
+}
+
+__attribute__((naked)) uint64_t __emupac_pacda(uint64_t ptr, uint64_t disc) {
+  __asm__(frame_pointer_wrap(__emupac_pacda_impl));
+}
+
+uint64_t __emupac_autda_impl(uint64_t ptr, uint64_t disc) {
+  if (pac_supported()) {
----------------
jroelofs wrote:

Might be worth writing a lazy ifunc-like resolver for this so it doesn't have to be checked on every call. Something like:


```
typedef BOOL (*fptr_ty)(void);

static BOOL resolver(void);

// Should be signed with address diversity
fptr_ty __emupac_autda_impl = resolver;

static BOOL resolver(void) {
  if (pac_supported())
    __emupac_autda_impl = __emupac_autda_pac;
  else
    __emupac_autda_impl = __emupac_autda_nopac;

  return __emupac_autda_impl();
}
```

If we had FMV support for PAC I'd suggest `__attribute__((target_version("pac")))`. cc @labrinea


You might also want to consider giving `__emupac_autda` a preserves-none calling convention, so you don't have to save/restore around them.

https://github.com/llvm/llvm-project/pull/133530


More information about the llvm-branch-commits mailing list