AddressSanitizer: 64-bit SPARC/Linux port and various SPARC fixes

[ebotcazou]

Summary

This contains the required changes to make the AddressSanitizer work on 64-bit SPARC/Linux (SPARC-T4 and later) as well as general fixes for the Sanitizer on the SPARC architecture, both Linux (where the sigaction interfacing was totally broken) and Solaris. This was tested with GCC on SPARC/Solaris and 64-bit SPARC/Linux.

[vitalybuka]

compiler-rt/lib/sanitizer_common/sanitizer_allocator_primary32.h

44

Could you please split this into smaller interdependent patches?
e.g.
asan should be in the different patch.
kSignExtendedAddresses related stuff is also looks like a separate patch
maybe something else

[ebotcazou]

OK, I'm going to split it into 4 parts.

compiler-rt/lib/asan/asan_allocator.h

 const uptr kAllocatorSize  =  0x2000000000ULL;  // 128G.
 typedef VeryCompactSizeClassMap SizeClassMap;
 # elif defined(__aarch64__)
-// AArch64/SANITIZER_CAN_USER_ALLOCATOR64 is only for 42-bit VMA
+// AArch64/SANITIZER_CAN_USE_ALLOCATOR64 is only for 42-bit VMA
 // so no need to different values for different VMA.
 const uptr kAllocatorSpace =  0x10000000000ULL;
 const uptr kAllocatorSize  =  0x10000000000ULL;  // 3T.
 typedef DefaultSizeClassMap SizeClassMap;
+# elif defined(__sparc__)
+const uptr kAllocatorSpace = ~(uptr)0;
+const uptr kAllocatorSize  =  0x20000000000ULL;  // 2T.
+typedef DefaultSizeClassMap SizeClassMap;
 # elif SANITIZER_WINDOWS
 const uptr kAllocatorSpace = ~(uptr)0;
 const uptr kAllocatorSize  =  0x8000000000ULL;  // 500G

compiler-rt/lib/asan/asan_globals.cc

     if (flags()->report_globals >= 2)
       ReportGlobal(g, "Search");
     if (IsAddressNearGlobal(addr, g)) {
+#if defined(__GNUC__) && defined(__sparc__)
+      internal_memcpy (&globals[res], &g, sizeof (Global));
+#else
       globals[res] = g;
+#endif
       if (reg_sites)
         reg_sites[res] = FindRegistrationSite(&g);
       res++;

compiler-rt/lib/asan/asan_mapping.h

 // || `[0x10000000000000, 0x11ffffffffffff]` || LowShadow  ||
 // || `[0x00000000000000, 0x0fffffffffffff]` || LowMem     ||
 //
+// Default Linux/SPARC64 (52-bit VMA) mapping:
+// || `[0x8000000000000, 0xfffffffffffff]` || HighMem    ||
+// || `[0x1080000000000, 0x207ffffffffff]` || HighShadow ||
+// || `[0x0090000000000, 0x107ffffffffff]` || ShadowGap  ||
+// || `[0x0080000000000, 0x008ffffffffff]` || LowShadow  ||
+// || `[0x0000000000000, 0x007ffffffffff]` || LowMem     ||
+//
 // Shadow mapping on FreeBSD/x86-64 with SHADOW_OFFSET == 0x400000000000:
 // || `[0x500000000000, 0x7fffffffffff]` || HighMem    ||
 // || `[0x4a0000000000, 0x4fffffffffff]` || HighShadow ||

 static const u64 kMIPS64_ShadowOffset64 = 1ULL << 37;
 static const u64 kPPC64_ShadowOffset64 = 1ULL << 44;
 static const u64 kSystemZ_ShadowOffset64 = 1ULL << 52;
+static const u64 kSPARC64_ShadowOffset64 = 1ULL << 43;  // 0x80000000000
 static const u64 kFreeBSD_ShadowOffset32 = 1ULL << 30;  // 0x40000000
 static const u64 kFreeBSD_ShadowOffset64 = 1ULL << 46;  // 0x400000000000
 static const u64 kNetBSD_ShadowOffset32 = 1ULL << 30;  // 0x40000000

 #   define SHADOW_OFFSET kDefaultShadowOffset64
 #  elif defined(__mips64)
 #   define SHADOW_OFFSET kMIPS64_ShadowOffset64
+#  elif defined(__sparc__)
+#   define SHADOW_OFFSET kSPARC64_ShadowOffset64
 #  elif SANITIZER_WINDOWS64
 #   define SHADOW_OFFSET __asan_shadow_memory_dynamic_address
 #  else

 
 #if SANITIZER_MYRIAD2
 #include "asan_mapping_myriad.h"
+#elif defined(__sparc__) && SANITIZER_WORDSIZE == 64
+#include "asan_mapping_sparc64.h"
 #else
 #define MEM_TO_SHADOW(mem) (((mem) >> SHADOW_SCALE) + (SHADOW_OFFSET))
 

compiler-rt/lib/asan/asan_mapping_sparc64.h

+//===-- asan_mapping_sparc64.h -----------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of AddressSanitizer, an address sanity checker.
+//
+// SPARC64-specific definitions for ASan memory mapping.
+//===----------------------------------------------------------------------===//
+#ifndef ASAN_MAPPING_SPARC64_H
+#define ASAN_MAPPING_SPARC64_H
+
+// This is tailored to the 52-bit VM layout on SPARC-T4 and later.
+// The VM space is split into two 51-bit halves at both ends: the low part
+// has all the bits above the 51st cleared, while the high part has them set.
+//   0xfff8000000000000 - 0xffffffffffffffff
+//   0x0000000000000000 - 0x0007ffffffffffff
+
+#define VMA_BITS  52
+#define HIGH_BITS (64 - VMA_BITS)
+
+// The idea is to chop the high bits before doing the scaling, so the two
+// parts become contiguous again and the usual scheme can be applied.
+
+#define MEM_TO_SHADOW(mem) \
+  ((((mem) << HIGH_BITS) >> (HIGH_BITS + (SHADOW_SCALE))) + (SHADOW_OFFSET))
+
+#define kLowMemBeg     0
+#define kLowMemEnd     (SHADOW_OFFSET - 1)
+
+#define kLowShadowBeg  SHADOW_OFFSET
+#define kLowShadowEnd  MEM_TO_SHADOW(kLowMemEnd)
+
+// But of course there is the huge hole between the high shadow memory,
+// which is in the low part, and the beginning of the high part.
+
+#define kHighMemBeg    (-(1ULL << (VMA_BITS - 1)))
+
+#define kHighShadowBeg MEM_TO_SHADOW(kHighMemBeg)
+#define kHighShadowEnd MEM_TO_SHADOW(kHighMemEnd)
+
+#define kMidShadowBeg  0
+#define kMidShadowEnd  0
+
+// With the zero shadow base we can not actually map pages starting from 0.
+// This constant is somewhat arbitrary.
+#define kZeroBaseShadowStart 0
+#define kZeroBaseMaxShadowStart (1 << 18)
+
+#define kShadowGapBeg  (kLowShadowEnd + 1)
+#define kShadowGapEnd  (kHighShadowBeg - 1)
+
+#define kShadowGap2Beg 0
+#define kShadowGap2End 0
+
+#define kShadowGap3Beg 0
+#define kShadowGap3End 0
+
+namespace __asan {
+
+static inline bool AddrIsInLowMem(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return a <= kLowMemEnd;
+}
+
+static inline bool AddrIsInLowShadow(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return a >= kLowShadowBeg && a <= kLowShadowEnd;
+}
+
+static inline bool AddrIsInMidMem(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return false;
+}
+
+static inline bool AddrIsInMidShadow(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return false;
+}
+
+static inline bool AddrIsInHighMem(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return kHighMemBeg && a >= kHighMemBeg && a <= kHighMemEnd;
+}
+
+static inline bool AddrIsInHighShadow(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return kHighMemBeg && a >= kHighShadowBeg && a <= kHighShadowEnd;
+}
+
+static inline bool AddrIsInShadowGap(uptr a) {
+  PROFILE_ASAN_MAPPING();
+  return a >= kShadowGapBeg && a <= kShadowGapEnd;
+}
+
+}  // namespace __asan
+
+#endif  // ASAN_MAPPING_SPARC64_H

compiler-rt/lib/sanitizer_common/sanitizer_allocator_internal.h

   using AddressSpaceView = LocalAddressSpaceView;
   using ByteMap = __sanitizer::ByteMap;
   typedef NoOpMapUnmapCallback MapUnmapCallback;
-  static const uptr kFlags = 0;
+  static const uptr kFlags =
+    SANITIZER_SIGN_EXTENDED_ADDRESSES ?
+    SizeClassAllocator32FlagMasks::kSignExtendedAddresses : 0;
 };
 typedef SizeClassAllocator32<AP32> PrimaryInternalAllocator;
 

compiler-rt/lib/sanitizer_common/sanitizer_allocator_primary32.h

   enum {
     kRandomShuffleChunks = 1,
     kUseSeparateSizeClassForBatch = 2,
+    kSignExtendedAddresses = 4,
   };
 };
 

       SizeClassAllocator32FlagMasks::kRandomShuffleChunks;
   static const bool kUseSeparateSizeClassForBatch = Params::kFlags &
       SizeClassAllocator32FlagMasks::kUseSeparateSizeClassForBatch;
+  static const bool kSignExtendedAddresses = Params::kFlags &
+      SizeClassAllocator32FlagMasks::kSignExtendedAddresses;
+  COMPILER_CHECK(!(kSignExtendedAddresses && (kSpaceSize & (kSpaceSize - 1))));
 
   struct TransferBatch {
     static const uptr kMaxNumCached = SizeClassMap::kMaxNumCachedHint - 2;

 
   bool PointerIsMine(const void *p) {
     uptr mem = reinterpret_cast<uptr>(p);
+    if (kSignExtendedAddresses)
+      mem &= (kSpaceSize - 1);
     if (mem < kSpaceBeg || mem >= kSpaceBeg + kSpaceSize)
       return false;
     return GetSizeClass(p) != 0;

   COMPILER_CHECK(sizeof(SizeClassInfo) % kCacheLineSize == 0);
 
   uptr ComputeRegionId(uptr mem) {
+    if (kSignExtendedAddresses)
+      mem &= (kSpaceSize - 1);
     const uptr res = mem >> kRegionSizeLog;
     CHECK_LT(res, kNumPossibleRegions);
     return res;

compiler-rt/lib/sanitizer_common/sanitizer_linux.cc

   return (1ULL << 40) - 1;  // 0x000000ffffffffffUL;
 # elif defined(__s390x__)
   return (1ULL << 53) - 1;  // 0x001fffffffffffffUL;
+# elif defined(__sparc__)
+  return ~(uptr)0;
 # else
   return (1ULL << 47) - 1;  // 0x00007fffffffffffUL;
 # endif

   u64 esr;
   if (!Aarch64GetESR(ucontext, &esr)) return UNKNOWN;
   return esr & ESR_ELx_WNR ? WRITE : READ;
-#elif SANITIZER_SOLARIS && defined(__sparc__)
+#elif defined(__sparc__)
   // Decode the instruction to determine the access type.
   // From OpenSolaris $SRC/uts/sun4/os/trap.c (get_accesstype).
+# if SANITIZER_SOLARIS
   uptr pc = ucontext->uc_mcontext.gregs[REG_PC];
+# else
+  // Historical BSDism here.
+  struct sigcontext *scontext = (struct sigcontext *)context;
+#  if defined(__arch64__)
+  uptr pc = scontext->sigc_regs.tpc;
+#  else
+  uptr pc = scontext->si_regs.pc;
+#  endif
+# endif
   u32 instr = *(u32 *)pc;
   return (instr >> 21) & 1 ? WRITE: READ;
 #else

   // pointer, but GCC always uses r31 when we need a frame pointer.
   *bp = ucontext->uc_mcontext.regs->gpr[PT_R31];
 #elif defined(__sparc__)
-  ucontext_t *ucontext = (ucontext_t*)context;
-  uptr *stk_ptr;
-# if defined(__sparcv9) || defined (__arch64__)
-# ifndef MC_PC
-#  define MC_PC REG_PC
-# endif
-# ifndef MC_O6
-#  define MC_O6 REG_O6
+# if defined(__arch64__) || defined(__sparcv9)
+#  define STACK_BIAS 2047
+# else
+#  define STACK_BIAS 0
 # endif
 # if SANITIZER_SOLARIS
-#  define mc_gregs gregs
-# endif
-  *pc = ucontext->uc_mcontext.mc_gregs[MC_PC];
-  *sp = ucontext->uc_mcontext.mc_gregs[MC_O6];
-  stk_ptr = (uptr *) (*sp + 2047);
-  *bp = stk_ptr[15];
-# else
+  ucontext_t *ucontext = (ucontext_t*)context;
   *pc = ucontext->uc_mcontext.gregs[REG_PC];
-  *sp = ucontext->uc_mcontext.gregs[REG_O6];
-  stk_ptr = (uptr *) *sp;
-  *bp = stk_ptr[15];
+  *sp = ucontext->uc_mcontext.gregs[REG_O6] + STACK_BIAS;
+# else
+  // Historical BSDism here.
+  struct sigcontext *scontext = (struct sigcontext *)context;
+#  if defined(__arch64__)
+  *pc = scontext->sigc_regs.tpc;
+  *sp = scontext->sigc_regs.u_regs[14] + STACK_BIAS;
+#  else
+  *pc = scontext->si_regs.pc;
+  *sp = scontext->si_regs.u_regs[14];
+#  endif
 # endif
+  *bp = (uptr) ((uhwptr *) *sp)[14] + STACK_BIAS;
 #elif defined(__mips__)
   ucontext_t *ucontext = (ucontext_t*)context;
   *pc = ucontext->uc_mcontext.pc;

compiler-rt/lib/sanitizer_common/sanitizer_platform.h

 # else
 #  define SANITIZER_MMAP_RANGE_SIZE FIRST_32_SECOND_64(1ULL << 32, 1ULL << 48)
 # endif
+#elif defined(__sparc__)
+# define SANITIZER_MMAP_RANGE_SIZE FIRST_32_SECOND_64(1ULL << 32, 1ULL << 52)
 #else
 # define SANITIZER_MMAP_RANGE_SIZE FIRST_32_SECOND_64(1ULL << 32, 1ULL << 47)
 #endif
 
+// Whether the addresses are sign-extended from the VMA range to the word.
+// The SPARC64 Linux port implements this to split the VMA space into two
+// non-contiguous halves with a huge hole in the middle.
+#if defined(__sparc__) && SANITIZER_WORDSIZE == 64
+# define SANITIZER_SIGN_EXTENDED_ADDRESSES 1
+#else
+# define SANITIZER_SIGN_EXTENDED_ADDRESSES 0
+#endif
+
 // The AArch64 linux port uses the canonical syscall set as mandated by
 // the upstream linux community for all new ports. Other ports may still
 // use legacy syscalls.

compiler-rt/lib/sanitizer_common/sanitizer_stacktrace.h

 
 static const u32 kStackTraceMax = 256;
 
-#if defined(__sparc__) || (SANITIZER_LINUX && defined(__mips__))
+#if SANITIZER_LINUX && defined(__mips__)
 # define SANITIZER_CAN_FAST_UNWIND 0
 #elif SANITIZER_WINDOWS
 # define SANITIZER_CAN_FAST_UNWIND 0

compiler-rt/lib/sanitizer_common/sanitizer_stacktrace.cc

 namespace __sanitizer {
 
 uptr StackTrace::GetNextInstructionPc(uptr pc) {
-#if defined(__mips__)
+#if defined(__sparc__) || defined(__mips__)
   return pc + 8;
-#elif defined(__powerpc__) || defined(__sparc__) || defined(__arm__) || \
-    defined(__aarch64__)
+#elif defined(__powerpc__) || defined(__arm__) || defined(__aarch64__)
   return pc + 4;
 #else
   return pc + 1;

compiler-rt/lib/sanitizer_common/sanitizer_stacktrace_sparc.cc

 // Implemention of fast stack unwinding for Sparc.
 //===----------------------------------------------------------------------===//
 
-// This file is ported to Sparc v8, but it should be easy to port to
-// Sparc v9.
-#if defined(__sparcv8__) || defined(__sparcv8) || defined(__sparc_v8__)
+#if defined(__sparc__)
+
+#if defined(__arch64__) || defined(__sparcv9)
+#define STACK_BIAS 2047
+#else
+#define STACK_BIAS 0
+#endif
 
 #include "sanitizer_common.h"
 #include "sanitizer_stacktrace.h"

                                          uptr stack_bottom, u32 max_depth) {
   const uptr kPageSize = GetPageSizeCached();
   CHECK_GE(max_depth, 2);
+#if defined(__GNUC__)
+  // __builtin_return_address returns the address of the call instruction
+  // on the SPARC and not the return address, so we need to compensate.
+  trace_buffer[0] = GetNextInstructionPc(pc);
+#else
   trace_buffer[0] = pc;
+#endif
   size = 1;
   if (stack_top < 4096) return;  // Sanity check for stack top.
   // Flush register windows to memory
+#if defined(__sparc_v9__) || defined(__sparcv9__) || defined(__sparcv9)
+  asm volatile("flushw" ::: "memory");
+#else
   asm volatile("ta 3" ::: "memory");
-  uhwptr *frame = (uhwptr*)bp;
+#endif
+  // On the SPARC, the return address is not in the frame, it is in a
+  // register.  There is no way to access it off of the current frame
+  // pointer, but it can be accessed off the previous frame pointer by
+  // reading the value from the register window save area.
+  uptr prev_bp = GET_CURRENT_FRAME();
+  uptr next_bp = prev_bp;
+  unsigned int i = 0;
+  while (next_bp != bp &&
+   IsAligned(next_bp, sizeof(uhwptr)) &&
+   i++ < 8) {
+    prev_bp = next_bp;
+    next_bp = (uptr) ((uhwptr *) next_bp)[14] + STACK_BIAS;
+  }
+  if (next_bp == bp)
+    bp = prev_bp;
   // Lowest possible address that makes sense as the next frame pointer.
   // Goes up as we walk the stack.
   uptr bottom = stack_bottom;
   // Avoid infinite loop when frame == frame[0] by using frame > prev_frame.
-  while (IsValidFrame((uptr)frame, stack_top, bottom) &&
-         IsAligned((uptr)frame, sizeof(*frame)) &&
+  while (IsValidFrame(bp, stack_top, bottom) &&
+         IsAligned(bp, sizeof(uhwptr)) &&
          size < max_depth) {
-    uhwptr pc1 = frame[15];
+    uhwptr pc1 = ((uhwptr *)bp)[15];
     // Let's assume that any pointer in the 0th page is invalid and
     // stop unwinding here.  If we're adding support for a platform
     // where this isn't true, we need to reconsider this check.
     if (pc1 < kPageSize)
       break;
     if (pc1 != pc) {
-      trace_buffer[size++] = (uptr) pc1;
+      // %o7 contains the address of the call instruction and not the
+      // return address, so we need to compensate.
+      trace_buffer[size++] = GetNextInstructionPc((uptr) pc1);
     }
-    bottom = (uptr)frame;
-    frame = (uhwptr*)frame[14];
+    bottom = bp;
+    bp = (uptr) ((uhwptr *) bp)[14] + STACK_BIAS;
   }
 }
 
 }  // namespace __sanitizer
 
-#endif  // !defined(__sparcv8__) && !defined(__sparcv8) &&
-        // !defined(__sparc_v8__)
+#endif  // !defined(__sparc__)

compiler-rt/lib/sanitizer_common/sanitizer_unwind_linux_libcdep.cc

   if (to_pop == 0 && size > 1)
     to_pop = 1;
   PopStackFrames(to_pop);
+#if defined(__GNUC__) && defined(__sparc__)
+  // __builtin_return_address returns the address of the call instruction
+  // on the SPARC and not the return address, so we need to compensate.
+  trace_buffer[0] = GetNextInstructionPc(pc);
+#else
   trace_buffer[0] = pc;
+#endif
 }
 
 void BufferedStackTrace::SlowUnwindStackWithContext(uptr pc, void *context,