1 //===-- ExternalFunctions.cpp - Implement External Functions --------------===//
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 contains both code to deal with invoking "external" functions, but
10 //  also contains code that implements "exported" external functions.
11 //
12 //  There are currently two mechanisms for handling external functions in the
13 //  Interpreter.  The first is to implement lle_* wrapper functions that are
14 //  specific to well-known library functions which manually translate the
15 //  arguments from GenericValues and make the call.  If such a wrapper does
16 //  not exist, and libffi is available, then the Interpreter will attempt to
17 //  invoke the function using libffi, after finding its address.
18 //
19 //===----------------------------------------------------------------------===//
20 
21 #include "Interpreter.h"
22 #include "llvm/ADT/APInt.h"
23 #include "llvm/ADT/ArrayRef.h"
24 #include "llvm/Config/config.h" // Detect libffi
25 #include "llvm/ExecutionEngine/GenericValue.h"
26 #include "llvm/IR/DataLayout.h"
27 #include "llvm/IR/DerivedTypes.h"
28 #include "llvm/IR/Function.h"
29 #include "llvm/IR/Type.h"
30 #include "llvm/Support/Casting.h"
31 #include "llvm/Support/DynamicLibrary.h"
32 #include "llvm/Support/ErrorHandling.h"
33 #include "llvm/Support/ManagedStatic.h"
34 #include "llvm/Support/Mutex.h"
35 #include "llvm/Support/raw_ostream.h"
36 #include <cassert>
37 #include <cmath>
38 #include <csignal>
39 #include <cstdint>
40 #include <cstdio>
41 #include <cstring>
42 #include <map>
43 #include <mutex>
44 #include <string>
45 #include <utility>
46 #include <vector>
47 
48 #ifdef HAVE_FFI_CALL
49 #ifdef HAVE_FFI_H
50 #include <ffi.h>
51 #define USE_LIBFFI
52 #elif HAVE_FFI_FFI_H
53 #include <ffi/ffi.h>
54 #define USE_LIBFFI
55 #endif
56 #endif
57 
58 using namespace llvm;
59 
60 static ManagedStatic<sys::Mutex> FunctionsLock;
61 
62 typedef GenericValue (*ExFunc)(FunctionType *, ArrayRef<GenericValue>);
63 static ManagedStatic<std::map<const Function *, ExFunc> > ExportedFunctions;
64 static ManagedStatic<std::map<std::string, ExFunc> > FuncNames;
65 
66 #ifdef USE_LIBFFI
67 typedef void (*RawFunc)();
68 static ManagedStatic<std::map<const Function *, RawFunc> > RawFunctions;
69 #endif
70 
71 static Interpreter *TheInterpreter;
72 
73 static char getTypeID(Type *Ty) {
74   switch (Ty->getTypeID()) {
75   case Type::VoidTyID:    return 'V';
76   case Type::IntegerTyID:
77     switch (cast<IntegerType>(Ty)->getBitWidth()) {
78       case 1:  return 'o';
79       case 8:  return 'B';
80       case 16: return 'S';
81       case 32: return 'I';
82       case 64: return 'L';
83       default: return 'N';
84     }
85   case Type::FloatTyID:   return 'F';
86   case Type::DoubleTyID:  return 'D';
87   case Type::PointerTyID: return 'P';
88   case Type::FunctionTyID:return 'M';
89   case Type::StructTyID:  return 'T';
90   case Type::ArrayTyID:   return 'A';
91   default: return 'U';
92   }
93 }
94 
95 // Try to find address of external function given a Function object.
96 // Please note, that interpreter doesn't know how to assemble a
97 // real call in general case (this is JIT job), that's why it assumes,
98 // that all external functions has the same (and pretty "general") signature.
99 // The typical example of such functions are "lle_X_" ones.
100 static ExFunc lookupFunction(const Function *F) {
101   // Function not found, look it up... start by figuring out what the
102   // composite function name should be.
103   std::string ExtName = "lle_";
104   FunctionType *FT = F->getFunctionType();
105   ExtName += getTypeID(FT->getReturnType());
106   for (Type *T : FT->params())
107     ExtName += getTypeID(T);
108   ExtName += ("_" + F->getName()).str();
109 
110   sys::ScopedLock Writer(*FunctionsLock);
111   ExFunc FnPtr = (*FuncNames)[ExtName];
112   if (!FnPtr)
113     FnPtr = (*FuncNames)[("lle_X_" + F->getName()).str()];
114   if (!FnPtr)  // Try calling a generic function... if it exists...
115     FnPtr = (ExFunc)(intptr_t)sys::DynamicLibrary::SearchForAddressOfSymbol(
116         ("lle_X_" + F->getName()).str());
117   if (FnPtr)
118     ExportedFunctions->insert(std::make_pair(F, FnPtr));  // Cache for later
119   return FnPtr;
120 }
121 
122 #ifdef USE_LIBFFI
123 static ffi_type *ffiTypeFor(Type *Ty) {
124   switch (Ty->getTypeID()) {
125     case Type::VoidTyID: return &ffi_type_void;
126     case Type::IntegerTyID:
127       switch (cast<IntegerType>(Ty)->getBitWidth()) {
128         case 8:  return &ffi_type_sint8;
129         case 16: return &ffi_type_sint16;
130         case 32: return &ffi_type_sint32;
131         case 64: return &ffi_type_sint64;
132       }
133     case Type::FloatTyID:   return &ffi_type_float;
134     case Type::DoubleTyID:  return &ffi_type_double;
135     case Type::PointerTyID: return &ffi_type_pointer;
136     default: break;
137   }
138   // TODO: Support other types such as StructTyID, ArrayTyID, OpaqueTyID, etc.
139   report_fatal_error("Type could not be mapped for use with libffi.");
140   return NULL;
141 }
142 
143 static void *ffiValueFor(Type *Ty, const GenericValue &AV,
144                          void *ArgDataPtr) {
145   switch (Ty->getTypeID()) {
146     case Type::IntegerTyID:
147       switch (cast<IntegerType>(Ty)->getBitWidth()) {
148         case 8: {
149           int8_t *I8Ptr = (int8_t *) ArgDataPtr;
150           *I8Ptr = (int8_t) AV.IntVal.getZExtValue();
151           return ArgDataPtr;
152         }
153         case 16: {
154           int16_t *I16Ptr = (int16_t *) ArgDataPtr;
155           *I16Ptr = (int16_t) AV.IntVal.getZExtValue();
156           return ArgDataPtr;
157         }
158         case 32: {
159           int32_t *I32Ptr = (int32_t *) ArgDataPtr;
160           *I32Ptr = (int32_t) AV.IntVal.getZExtValue();
161           return ArgDataPtr;
162         }
163         case 64: {
164           int64_t *I64Ptr = (int64_t *) ArgDataPtr;
165           *I64Ptr = (int64_t) AV.IntVal.getZExtValue();
166           return ArgDataPtr;
167         }
168       }
169     case Type::FloatTyID: {
170       float *FloatPtr = (float *) ArgDataPtr;
171       *FloatPtr = AV.FloatVal;
172       return ArgDataPtr;
173     }
174     case Type::DoubleTyID: {
175       double *DoublePtr = (double *) ArgDataPtr;
176       *DoublePtr = AV.DoubleVal;
177       return ArgDataPtr;
178     }
179     case Type::PointerTyID: {
180       void **PtrPtr = (void **) ArgDataPtr;
181       *PtrPtr = GVTOP(AV);
182       return ArgDataPtr;
183     }
184     default: break;
185   }
186   // TODO: Support other types such as StructTyID, ArrayTyID, OpaqueTyID, etc.
187   report_fatal_error("Type value could not be mapped for use with libffi.");
188   return NULL;
189 }
190 
191 static bool ffiInvoke(RawFunc Fn, Function *F, ArrayRef<GenericValue> ArgVals,
192                       const DataLayout &TD, GenericValue &Result) {
193   ffi_cif cif;
194   FunctionType *FTy = F->getFunctionType();
195   const unsigned NumArgs = F->arg_size();
196 
197   // TODO: We don't have type information about the remaining arguments, because
198   // this information is never passed into ExecutionEngine::runFunction().
199   if (ArgVals.size() > NumArgs && F->isVarArg()) {
200     report_fatal_error("Calling external var arg function '" + F->getName()
201                       + "' is not supported by the Interpreter.");
202   }
203 
204   unsigned ArgBytes = 0;
205 
206   std::vector<ffi_type*> args(NumArgs);
207   for (Function::const_arg_iterator A = F->arg_begin(), E = F->arg_end();
208        A != E; ++A) {
209     const unsigned ArgNo = A->getArgNo();
210     Type *ArgTy = FTy->getParamType(ArgNo);
211     args[ArgNo] = ffiTypeFor(ArgTy);
212     ArgBytes += TD.getTypeStoreSize(ArgTy);
213   }
214 
215   SmallVector<uint8_t, 128> ArgData;
216   ArgData.resize(ArgBytes);
217   uint8_t *ArgDataPtr = ArgData.data();
218   SmallVector<void*, 16> values(NumArgs);
219   for (Function::const_arg_iterator A = F->arg_begin(), E = F->arg_end();
220        A != E; ++A) {
221     const unsigned ArgNo = A->getArgNo();
222     Type *ArgTy = FTy->getParamType(ArgNo);
223     values[ArgNo] = ffiValueFor(ArgTy, ArgVals[ArgNo], ArgDataPtr);
224     ArgDataPtr += TD.getTypeStoreSize(ArgTy);
225   }
226 
227   Type *RetTy = FTy->getReturnType();
228   ffi_type *rtype = ffiTypeFor(RetTy);
229 
230   if (ffi_prep_cif(&cif, FFI_DEFAULT_ABI, NumArgs, rtype, args.data()) ==
231       FFI_OK) {
232     SmallVector<uint8_t, 128> ret;
233     if (RetTy->getTypeID() != Type::VoidTyID)
234       ret.resize(TD.getTypeStoreSize(RetTy));
235     ffi_call(&cif, Fn, ret.data(), values.data());
236     switch (RetTy->getTypeID()) {
237       case Type::IntegerTyID:
238         switch (cast<IntegerType>(RetTy)->getBitWidth()) {
239           case 8:  Result.IntVal = APInt(8 , *(int8_t *) ret.data()); break;
240           case 16: Result.IntVal = APInt(16, *(int16_t*) ret.data()); break;
241           case 32: Result.IntVal = APInt(32, *(int32_t*) ret.data()); break;
242           case 64: Result.IntVal = APInt(64, *(int64_t*) ret.data()); break;
243         }
244         break;
245       case Type::FloatTyID:   Result.FloatVal   = *(float *) ret.data(); break;
246       case Type::DoubleTyID:  Result.DoubleVal  = *(double*) ret.data(); break;
247       case Type::PointerTyID: Result.PointerVal = *(void **) ret.data(); break;
248       default: break;
249     }
250     return true;
251   }
252 
253   return false;
254 }
255 #endif // USE_LIBFFI
256 
257 GenericValue Interpreter::callExternalFunction(Function *F,
258                                                ArrayRef<GenericValue> ArgVals) {
259   TheInterpreter = this;
260 
261   std::unique_lock<sys::Mutex> Guard(*FunctionsLock);
262 
263   // Do a lookup to see if the function is in our cache... this should just be a
264   // deferred annotation!
265   std::map<const Function *, ExFunc>::iterator FI = ExportedFunctions->find(F);
266   if (ExFunc Fn = (FI == ExportedFunctions->end()) ? lookupFunction(F)
267                                                    : FI->second) {
268     Guard.unlock();
269     return Fn(F->getFunctionType(), ArgVals);
270   }
271 
272 #ifdef USE_LIBFFI
273   std::map<const Function *, RawFunc>::iterator RF = RawFunctions->find(F);
274   RawFunc RawFn;
275   if (RF == RawFunctions->end()) {
276     RawFn = (RawFunc)(intptr_t)
277       sys::DynamicLibrary::SearchForAddressOfSymbol(std::string(F->getName()));
278     if (!RawFn)
279       RawFn = (RawFunc)(intptr_t)getPointerToGlobalIfAvailable(F);
280     if (RawFn != 0)
281       RawFunctions->insert(std::make_pair(F, RawFn));  // Cache for later
282   } else {
283     RawFn = RF->second;
284   }
285 
286   Guard.unlock();
287 
288   GenericValue Result;
289   if (RawFn != 0 && ffiInvoke(RawFn, F, ArgVals, getDataLayout(), Result))
290     return Result;
291 #endif // USE_LIBFFI
292 
293   if (F->getName() == "__main")
294     errs() << "Tried to execute an unknown external function: "
295       << *F->getType() << " __main\n";
296   else
297     report_fatal_error("Tried to execute an unknown external function: " +
298                        F->getName());
299 #ifndef USE_LIBFFI
300   errs() << "Recompiling LLVM with --enable-libffi might help.\n";
301 #endif
302   return GenericValue();
303 }
304 
305 //===----------------------------------------------------------------------===//
306 //  Functions "exported" to the running application...
307 //
308 
309 // void atexit(Function*)
310 static GenericValue lle_X_atexit(FunctionType *FT,
311                                  ArrayRef<GenericValue> Args) {
312   assert(Args.size() == 1);
313   TheInterpreter->addAtExitHandler((Function*)GVTOP(Args[0]));
314   GenericValue GV;
315   GV.IntVal = 0;
316   return GV;
317 }
318 
319 // void exit(int)
320 static GenericValue lle_X_exit(FunctionType *FT, ArrayRef<GenericValue> Args) {
321   TheInterpreter->exitCalled(Args[0]);
322   return GenericValue();
323 }
324 
325 // void abort(void)
326 static GenericValue lle_X_abort(FunctionType *FT, ArrayRef<GenericValue> Args) {
327   //FIXME: should we report or raise here?
328   //report_fatal_error("Interpreted program raised SIGABRT");
329   raise (SIGABRT);
330   return GenericValue();
331 }
332 
333 // int sprintf(char *, const char *, ...) - a very rough implementation to make
334 // output useful.
335 static GenericValue lle_X_sprintf(FunctionType *FT,
336                                   ArrayRef<GenericValue> Args) {
337   char *OutputBuffer = (char *)GVTOP(Args[0]);
338   const char *FmtStr = (const char *)GVTOP(Args[1]);
339   unsigned ArgNo = 2;
340 
341   // printf should return # chars printed.  This is completely incorrect, but
342   // close enough for now.
343   GenericValue GV;
344   GV.IntVal = APInt(32, strlen(FmtStr));
345   while (true) {
346     switch (*FmtStr) {
347     case 0: return GV;             // Null terminator...
348     default:                       // Normal nonspecial character
349       sprintf(OutputBuffer++, "%c", *FmtStr++);
350       break;
351     case '\\': {                   // Handle escape codes
352       sprintf(OutputBuffer, "%c%c", *FmtStr, *(FmtStr+1));
353       FmtStr += 2; OutputBuffer += 2;
354       break;
355     }
356     case '%': {                    // Handle format specifiers
357       char FmtBuf[100] = "", Buffer[1000] = "";
358       char *FB = FmtBuf;
359       *FB++ = *FmtStr++;
360       char Last = *FB++ = *FmtStr++;
361       unsigned HowLong = 0;
362       while (Last != 'c' && Last != 'd' && Last != 'i' && Last != 'u' &&
363              Last != 'o' && Last != 'x' && Last != 'X' && Last != 'e' &&
364              Last != 'E' && Last != 'g' && Last != 'G' && Last != 'f' &&
365              Last != 'p' && Last != 's' && Last != '%') {
366         if (Last == 'l' || Last == 'L') HowLong++;  // Keep track of l's
367         Last = *FB++ = *FmtStr++;
368       }
369       *FB = 0;
370 
371       switch (Last) {
372       case '%':
373         memcpy(Buffer, "%", 2); break;
374       case 'c':
375         sprintf(Buffer, FmtBuf, uint32_t(Args[ArgNo++].IntVal.getZExtValue()));
376         break;
377       case 'd': case 'i':
378       case 'u': case 'o':
379       case 'x': case 'X':
380         if (HowLong >= 1) {
381           if (HowLong == 1 &&
382               TheInterpreter->getDataLayout().getPointerSizeInBits() == 64 &&
383               sizeof(long) < sizeof(int64_t)) {
384             // Make sure we use %lld with a 64 bit argument because we might be
385             // compiling LLI on a 32 bit compiler.
386             unsigned Size = strlen(FmtBuf);
387             FmtBuf[Size] = FmtBuf[Size-1];
388             FmtBuf[Size+1] = 0;
389             FmtBuf[Size-1] = 'l';
390           }
391           sprintf(Buffer, FmtBuf, Args[ArgNo++].IntVal.getZExtValue());
392         } else
393           sprintf(Buffer, FmtBuf,uint32_t(Args[ArgNo++].IntVal.getZExtValue()));
394         break;
395       case 'e': case 'E': case 'g': case 'G': case 'f':
396         sprintf(Buffer, FmtBuf, Args[ArgNo++].DoubleVal); break;
397       case 'p':
398         sprintf(Buffer, FmtBuf, (void*)GVTOP(Args[ArgNo++])); break;
399       case 's':
400         sprintf(Buffer, FmtBuf, (char*)GVTOP(Args[ArgNo++])); break;
401       default:
402         errs() << "<unknown printf code '" << *FmtStr << "'!>";
403         ArgNo++; break;
404       }
405       size_t Len = strlen(Buffer);
406       memcpy(OutputBuffer, Buffer, Len + 1);
407       OutputBuffer += Len;
408       }
409       break;
410     }
411   }
412   return GV;
413 }
414 
415 // int printf(const char *, ...) - a very rough implementation to make output
416 // useful.
417 static GenericValue lle_X_printf(FunctionType *FT,
418                                  ArrayRef<GenericValue> Args) {
419   char Buffer[10000];
420   std::vector<GenericValue> NewArgs;
421   NewArgs.push_back(PTOGV((void*)&Buffer[0]));
422   llvm::append_range(NewArgs, Args);
423   GenericValue GV = lle_X_sprintf(FT, NewArgs);
424   outs() << Buffer;
425   return GV;
426 }
427 
428 // int sscanf(const char *format, ...);
429 static GenericValue lle_X_sscanf(FunctionType *FT,
430                                  ArrayRef<GenericValue> args) {
431   assert(args.size() < 10 && "Only handle up to 10 args to sscanf right now!");
432 
433   char *Args[10];
434   for (unsigned i = 0; i < args.size(); ++i)
435     Args[i] = (char*)GVTOP(args[i]);
436 
437   GenericValue GV;
438   GV.IntVal = APInt(32, sscanf(Args[0], Args[1], Args[2], Args[3], Args[4],
439                     Args[5], Args[6], Args[7], Args[8], Args[9]));
440   return GV;
441 }
442 
443 // int scanf(const char *format, ...);
444 static GenericValue lle_X_scanf(FunctionType *FT, ArrayRef<GenericValue> args) {
445   assert(args.size() < 10 && "Only handle up to 10 args to scanf right now!");
446 
447   char *Args[10];
448   for (unsigned i = 0; i < args.size(); ++i)
449     Args[i] = (char*)GVTOP(args[i]);
450 
451   GenericValue GV;
452   GV.IntVal = APInt(32, scanf( Args[0], Args[1], Args[2], Args[3], Args[4],
453                     Args[5], Args[6], Args[7], Args[8], Args[9]));
454   return GV;
455 }
456 
457 // int fprintf(FILE *, const char *, ...) - a very rough implementation to make
458 // output useful.
459 static GenericValue lle_X_fprintf(FunctionType *FT,
460                                   ArrayRef<GenericValue> Args) {
461   assert(Args.size() >= 2);
462   char Buffer[10000];
463   std::vector<GenericValue> NewArgs;
464   NewArgs.push_back(PTOGV(Buffer));
465   NewArgs.insert(NewArgs.end(), Args.begin()+1, Args.end());
466   GenericValue GV = lle_X_sprintf(FT, NewArgs);
467 
468   fputs(Buffer, (FILE *) GVTOP(Args[0]));
469   return GV;
470 }
471 
472 static GenericValue lle_X_memset(FunctionType *FT,
473                                  ArrayRef<GenericValue> Args) {
474   int val = (int)Args[1].IntVal.getSExtValue();
475   size_t len = (size_t)Args[2].IntVal.getZExtValue();
476   memset((void *)GVTOP(Args[0]), val, len);
477   // llvm.memset.* returns void, lle_X_* returns GenericValue,
478   // so here we return GenericValue with IntVal set to zero
479   GenericValue GV;
480   GV.IntVal = 0;
481   return GV;
482 }
483 
484 static GenericValue lle_X_memcpy(FunctionType *FT,
485                                  ArrayRef<GenericValue> Args) {
486   memcpy(GVTOP(Args[0]), GVTOP(Args[1]),
487          (size_t)(Args[2].IntVal.getLimitedValue()));
488 
489   // llvm.memcpy* returns void, lle_X_* returns GenericValue,
490   // so here we return GenericValue with IntVal set to zero
491   GenericValue GV;
492   GV.IntVal = 0;
493   return GV;
494 }
495 
496 void Interpreter::initializeExternalFunctions() {
497   sys::ScopedLock Writer(*FunctionsLock);
498   (*FuncNames)["lle_X_atexit"]       = lle_X_atexit;
499   (*FuncNames)["lle_X_exit"]         = lle_X_exit;
500   (*FuncNames)["lle_X_abort"]        = lle_X_abort;
501 
502   (*FuncNames)["lle_X_printf"]       = lle_X_printf;
503   (*FuncNames)["lle_X_sprintf"]      = lle_X_sprintf;
504   (*FuncNames)["lle_X_sscanf"]       = lle_X_sscanf;
505   (*FuncNames)["lle_X_scanf"]        = lle_X_scanf;
506   (*FuncNames)["lle_X_fprintf"]      = lle_X_fprintf;
507   (*FuncNames)["lle_X_memset"]       = lle_X_memset;
508   (*FuncNames)["lle_X_memcpy"]       = lle_X_memcpy;
509 }
510