1 //===-- xray_AArch64.cpp ----------------------------------------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file is a part of XRay, a dynamic runtime instrumentation system. 10 // 11 // Implementation of AArch64-specific routines (64-bit). 12 // 13 //===----------------------------------------------------------------------===// 14 #include "sanitizer_common/sanitizer_common.h" 15 #include "xray_defs.h" 16 #include "xray_interface_internal.h" 17 #include <atomic> 18 #include <cassert> 19 20 extern "C" void __clear_cache(void *start, void *end); 21 22 namespace __xray { 23 24 // The machine codes for some instructions used in runtime patching. 25 enum class PatchOpcodes : uint32_t { 26 PO_StpX0X30SP_m16e = 0xA9BF7BE0, // STP X0, X30, [SP, #-16]! 27 PO_LdrX16_12 = 0x58000070, // LDR X16, #12 28 PO_BlrX16 = 0xD63F0200, // BLR X16 29 PO_LdpX0X30SP_16 = 0xA8C17BE0, // LDP X0, X30, [SP], #16 30 PO_B32 = 0x14000008 // B #32 31 }; 32 33 inline static bool patchSled(const bool Enable, const uint32_t FuncId, 34 const XRaySledEntry &Sled, 35 void (*TracingHook)()) XRAY_NEVER_INSTRUMENT { 36 // When |Enable| == true, 37 // We replace the following compile-time stub (sled): 38 // 39 // xray_sled_n: 40 // B #32 41 // 7 NOPs (24 bytes) 42 // 43 // With the following runtime patch: 44 // 45 // xray_sled_n: 46 // STP X0, X30, [SP, #-16]! ; PUSH {r0, lr} 47 // LDR W17, #12 ; W17 := function ID 48 // LDR X16,#12 ; X16 := address of the trampoline 49 // BLR X16 50 // ;DATA: 32 bits of function ID 51 // ;DATA: lower 32 bits of the address of the trampoline 52 // ;DATA: higher 32 bits of the address of the trampoline 53 // LDP X0, X30, [SP], #16 ; POP {r0, lr} 54 // 55 // Replacement of the first 4-byte instruction should be the last and atomic 56 // operation, so that the user code which reaches the sled concurrently 57 // either jumps over the whole sled, or executes the whole sled when the 58 // latter is ready. 59 // 60 // When |Enable|==false, we set back the first instruction in the sled to be 61 // B #32 62 63 uint32_t *FirstAddress = reinterpret_cast<uint32_t *>(Sled.address()); 64 uint32_t *CurAddress = FirstAddress + 1; 65 if (Enable) { 66 *CurAddress++ = 0x18000071; // ldr w17, #12 67 *CurAddress = uint32_t(PatchOpcodes::PO_LdrX16_12); 68 CurAddress++; 69 *CurAddress = uint32_t(PatchOpcodes::PO_BlrX16); 70 CurAddress++; 71 *CurAddress = FuncId; 72 CurAddress++; 73 *reinterpret_cast<void (**)()>(CurAddress) = TracingHook; 74 CurAddress += 2; 75 *CurAddress = uint32_t(PatchOpcodes::PO_LdpX0X30SP_16); 76 CurAddress++; 77 std::atomic_store_explicit( 78 reinterpret_cast<std::atomic<uint32_t> *>(FirstAddress), 79 uint32_t(PatchOpcodes::PO_StpX0X30SP_m16e), std::memory_order_release); 80 } else { 81 std::atomic_store_explicit( 82 reinterpret_cast<std::atomic<uint32_t> *>(FirstAddress), 83 uint32_t(PatchOpcodes::PO_B32), std::memory_order_release); 84 } 85 __clear_cache(reinterpret_cast<char *>(FirstAddress), 86 reinterpret_cast<char *>(CurAddress)); 87 return true; 88 } 89 90 bool patchFunctionEntry(const bool Enable, const uint32_t FuncId, 91 const XRaySledEntry &Sled, 92 const XRayTrampolines &Trampolines, 93 bool LogArgs) XRAY_NEVER_INSTRUMENT { 94 auto Trampoline = 95 LogArgs ? Trampolines.LogArgsTrampoline : Trampolines.EntryTrampoline; 96 return patchSled(Enable, FuncId, Sled, Trampoline); 97 } 98 99 bool patchFunctionExit( 100 const bool Enable, const uint32_t FuncId, const XRaySledEntry &Sled, 101 const XRayTrampolines &Trampolines) XRAY_NEVER_INSTRUMENT { 102 return patchSled(Enable, FuncId, Sled, Trampolines.ExitTrampoline); 103 } 104 105 bool patchFunctionTailExit( 106 const bool Enable, const uint32_t FuncId, const XRaySledEntry &Sled, 107 const XRayTrampolines &Trampolines) XRAY_NEVER_INSTRUMENT { 108 return patchSled(Enable, FuncId, Sled, Trampolines.TailExitTrampoline); 109 } 110 111 // AArch64AsmPrinter::LowerPATCHABLE_EVENT_CALL generates this code sequence: 112 // 113 // .Lxray_event_sled_N: 114 // b 1f 115 // save x0 and x1 (and also x2 for TYPED_EVENT_CALL) 116 // set up x0 and x1 (and also x2 for TYPED_EVENT_CALL) 117 // bl __xray_CustomEvent or __xray_TypedEvent 118 // restore x0 and x1 (and also x2 for TYPED_EVENT_CALL) 119 // 1f 120 // 121 // There are 6 instructions for EVENT_CALL and 9 for TYPED_EVENT_CALL. 122 // 123 // Enable: b .+24 => nop 124 // Disable: nop => b .+24 125 bool patchCustomEvent(const bool Enable, const uint32_t FuncId, 126 const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT { 127 uint32_t Inst = Enable ? 0xd503201f : 0x14000006; 128 std::atomic_store_explicit( 129 reinterpret_cast<std::atomic<uint32_t> *>(Sled.address()), Inst, 130 std::memory_order_release); 131 return false; 132 } 133 134 // Enable: b +36 => nop 135 // Disable: nop => b +36 136 bool patchTypedEvent(const bool Enable, const uint32_t FuncId, 137 const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT { 138 uint32_t Inst = Enable ? 0xd503201f : 0x14000009; 139 std::atomic_store_explicit( 140 reinterpret_cast<std::atomic<uint32_t> *>(Sled.address()), Inst, 141 std::memory_order_release); 142 return false; 143 } 144 145 // FIXME: Maybe implement this better? 146 bool probeRequiredCPUFeatures() XRAY_NEVER_INSTRUMENT { return true; } 147 148 } // namespace __xray 149