7186778: MachO decoder implementation for MacOSX

Summary: Implementation of decoder for Apple's MacOSX. The implementation is based on the patch provided by Kevin Walls.
Reviewed-by: coleenp, kamg, kevinw

src/os/bsd/vm/decoder_machO.cpp

@@ -26,6 +26,139 @@
 
 #ifdef __APPLE__
 #include "decoder_machO.hpp"
+
+#include <cxxabi.h>
+#include <mach-o/loader.h>
+#include <mach-o/nlist.h>
+
+
+bool MachODecoder::demangle(const char* symbol, char *buf, int buflen) {
+  int   status;
+  char* result;
+  size_t size = (size_t)buflen;
+  // Don't pass buf to __cxa_demangle. In case of the 'buf' is too small,
+  // __cxa_demangle will call system "realloc" for additional memory, which
+  // may use different malloc/realloc mechanism that allocates 'buf'.
+  if ((result = abi::__cxa_demangle(symbol, NULL, NULL, &status)) != NULL) {
+    jio_snprintf(buf, buflen, "%s", result);
+      // call c library's free
+      ::free(result);
+      return true;
+  }
+  return false;
+}
+
+bool MachODecoder::decode(address addr, char *buf,
+      int buflen, int *offset, const void *mach_base) {
+  struct symtab_command * symt = (struct symtab_command *)
+    mach_find_command((struct mach_header_64 *)mach_base, LC_SYMTAB);
+  if (symt == NULL) {
+    DEBUG_ONLY(tty->print_cr("no symtab in mach file at 0x%lx", mach_base));
+    return false;
+  }
+  uint32_t off = symt->symoff;          /* symbol table offset (within this mach file) */
+  uint32_t nsyms = symt->nsyms;         /* number of symbol table entries */
+  uint32_t stroff = symt->stroff;       /* string table offset */
+  uint32_t strsize = symt->strsize;     /* string table size in bytes */
+
+  // iterate through symbol table trying to match our offset
+
+  uint32_t addr_relative = (uintptr_t) mach_base - (uintptr_t) addr; // offset we seek in the symtab
+  void * symtab_addr = (void*) ((uintptr_t) mach_base + off);
+  struct nlist_64 *cur_nlist = (struct nlist_64 *) symtab_addr;
+  struct nlist_64 *last_nlist = cur_nlist;  // no size stored in an entry, so keep previously seen nlist
+
+  int32_t found_strx = 0;
+  int32_t found_symval = 0;
+
+  for (uint32_t i=0; i < nsyms; i++) {
+    uint32_t this_value = cur_nlist->n_value;
+
+    if (addr_relative == this_value) {
+      found_strx =  cur_nlist->n_un.n_strx;
+      found_symval = this_value;
+      break;
+    } else if (addr_relative > this_value) {
+      // gone past it, use previously seen nlist:
+      found_strx = last_nlist->n_un.n_strx;
+      found_symval = last_nlist->n_value;
+      break;
+    }
+    last_nlist = cur_nlist;
+    cur_nlist = cur_nlist + sizeof(struct nlist_64);
+  }
+  if (found_strx == 0) {
+    return false;
+  }
+  // write the offset:
+  *offset = addr_relative - found_symval;
+
+  // lookup found_strx in the string table
+  char * symname = mach_find_in_stringtable((char*) ((uintptr_t)mach_base + stroff), strsize, found_strx);
+  if (symname) {
+      strncpy(buf, symname, buflen);
+      return true;
+  }
+  DEBUG_ONLY(tty->print_cr("no string or null string found."));
+  return false;
+}
+
+void* MachODecoder::mach_find_command(struct mach_header_64 * mach_base, uint32_t command_wanted) {
+  // possibly verify it is a mach_header, use magic number.
+  // commands begin immediately after the header.
+  struct load_command *pos = (struct load_command *) mach_base + sizeof(struct mach_header_64);
+  for (uint32_t i = 0; i < mach_base->ncmds; i++) {
+    struct load_command *this_cmd = (struct load_command *) pos;
+    if (this_cmd->cmd == command_wanted) {
+       return pos;
+    }
+    int cmdsize = this_cmd->cmdsize;
+    pos += cmdsize;
+  }
+  return NULL;
+}
+
+char* MachODecoder::mach_find_in_stringtable(char *strtab, uint32_t tablesize, int strx_wanted) {
+
+  if (strx_wanted == 0) {
+    return NULL;
+  }
+  char *strtab_end = strtab + tablesize;
+
+  // find the first string, skip over the space char
+  // (or the four zero bytes we see e.g. in libclient)
+  if (*strtab == ' ') {
+      strtab++;
+      if (*strtab != 0) {
+          DEBUG_ONLY(tty->print_cr("string table has leading space but no following zero."));
+          return NULL;
+      }
+      strtab++;
+  } else {
+      if ((uint32_t) *strtab != 0) {
+          DEBUG_ONLY(tty->print_cr("string table without leading space or leading int of zero."));
+          return NULL;
+      }
+      strtab+=4;
+  }
+  // read the real strings starting at index 1
+  int cur_strx = 1;
+  while (strtab < strtab_end) {
+    if (cur_strx == strx_wanted) {
+        return strtab;
+    }
+    // find start of next string
+    while (*strtab != 0) {
+        strtab++;
+    }
+    strtab++; // skip the terminating zero
+    cur_strx++;
+  }
+  DEBUG_ONLY(tty->print_cr("string number %d not found.", strx_wanted));
+  return NULL;
+}
+
+
 #endif
 
 

src/os/bsd/vm/decoder_machO.hpp

@@ -31,10 +31,25 @@
 
 // Just a placehold for now, a real implementation should derive
 // from AbstractDecoder
-class MachODecoder : public NullDecoder {
-public:
+class MachODecoder : public AbstractDecoder {
+ public:
   MachODecoder() { }
   ~MachODecoder() { }
+  virtual bool can_decode_C_frame_in_vm() const {
+    return true;
+  }
+  virtual bool demangle(const char* symbol, char* buf, int buflen);
+  virtual bool decode(address pc, char* buf, int buflen, int* offset,
+                      const void* base);
+  virtual bool decode(address pc, char* buf, int buflen, int* offset,
+                      const char* module_path = NULL) {
+    ShouldNotReachHere();
+    return false;
+  }
+
+ private:
+  void * mach_find_command(struct mach_header_64 * mach_base, uint32_t command_wanted);
+  char * mach_find_in_stringtable(char *strtab, uint32_t tablesize, int strx_wanted);
 };
 
 #endif

src/os/bsd/vm/os_bsd.cpp

@@ -1946,10 +1946,16 @@ bool os::address_is_in_vm(address addr) {
   return false;
 }
 
+
+#define MACH_MAXSYMLEN 256
+
 bool os::dll_address_to_function_name(address addr, char *buf,
                                       int buflen, int *offset) {
   Dl_info dlinfo;
+  char localbuf[MACH_MAXSYMLEN];
 
+  // dladdr will find names of dynamic functions only, but does
+  // it set dli_fbase with mach_header address when it "fails" ?
   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
     if (buf != NULL) {
       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
@@ -1965,6 +1971,14 @@ bool os::dll_address_to_function_name(address addr, char *buf,
     }
   }
 
+  // Handle non-dymanic manually:
+  if (dlinfo.dli_fbase != NULL &&
+      Decoder::decode(addr, localbuf, MACH_MAXSYMLEN, offset, dlinfo.dli_fbase)) {
+    if(!Decoder::demangle(localbuf, buf, buflen)) {
+      jio_snprintf(buf, buflen, "%s", localbuf);
+    }
+    return true;
+  }
   if (buf != NULL) buf[0] = '\0';
   if (offset != NULL) *offset = -1;
   return false;

src/os/windows/vm/decoder_windows.cpp

@@ -72,10 +72,10 @@ void WindowsDecoder::initialize() {
 
      // find out if jvm.dll contains private symbols, by decoding
      // current function and comparing the result
-     address addr = (address)Decoder::decode;
+     address addr = (address)Decoder::demangle;
      char buf[MAX_PATH];
      if (decode(addr, buf, sizeof(buf), NULL)) {
-       _can_decode_in_vm = !strcmp(buf, "Decoder::decode");
+       _can_decode_in_vm = !strcmp(buf, "Decoder::demangle");
      }
   }
 }

src/os/windows/vm/decoder_windows.hpp

@@ -45,6 +45,10 @@ public:
   bool can_decode_C_frame_in_vm() const;
   bool demangle(const char* symbol, char *buf, int buflen);
   bool decode(address addr, char *buf, int buflen, int* offset, const char* modulepath = NULL);
+  bool decode(address addr, char *buf, int buflen, int* offset, const void* base) {
+    ShouldNotReachHere();
+    return false;
+  }
 
 private:
   void initialize();

src/share/vm/utilities/decoder.cpp

@@ -91,6 +91,18 @@ bool Decoder::decode(address addr, char* buf, int buflen, int* offset, const cha
   return decoder->decode(addr, buf, buflen, offset, modulepath);
 }
 
+bool Decoder::decode(address addr, char* buf, int buflen, int* offset, const void* base) {
+  assert(_shared_decoder_lock != NULL, "Just check");
+  bool error_handling_thread = os::current_thread_id() == VMError::first_error_tid;
+  MutexLockerEx locker(error_handling_thread ? NULL : _shared_decoder_lock, true);
+  AbstractDecoder* decoder = error_handling_thread ?
+    get_error_handler_instance(): get_shared_instance();
+  assert(decoder != NULL, "null decoder");
+
+  return decoder->decode(addr, buf, buflen, offset, base);
+}
+
+
 bool Decoder::demangle(const char* symbol, char* buf, int buflen) {
   assert(_shared_decoder_lock != NULL, "Just check");
   bool error_handling_thread = os::current_thread_id() == VMError::first_error_tid;

src/share/vm/utilities/decoder.hpp

@@ -47,6 +47,8 @@ public:
   // the function
   virtual bool decode(address pc, char* buf, int buflen, int* offset,
     const char* modulepath = NULL) = 0;
+  virtual bool decode(address pc, char* buf, int buflen, int* offset, const void* base) = 0;
+
   // demangle a C++ symbol
   virtual bool demangle(const char* symbol, char* buf, int buflen) = 0;
   // if the decoder can decode symbols in vm
@@ -82,6 +84,10 @@ public:
     return false;
   }
 
+  virtual bool decode(address pc, char* buf, int buflen, int* offset, const void* base) {
+    return false;
+  }
+
   virtual bool demangle(const char* symbol, char* buf, int buflen) {
     return false;
   }
@@ -95,6 +101,7 @@ public:
 class Decoder : AllStatic {
 public:
   static bool decode(address pc, char* buf, int buflen, int* offset, const char* modulepath = NULL);
+  static bool decode(address pc, char* buf, int buflen, int* offset, const void* base);
   static bool demangle(const char* symbol, char* buf, int buflen);
   static bool can_decode_C_frame_in_vm();
 

src/share/vm/utilities/decoder_elf.hpp

@@ -43,6 +43,10 @@ public:
 
   bool demangle(const char* symbol, char *buf, int buflen);
   bool decode(address addr, char *buf, int buflen, int* offset, const char* filepath = NULL);
+  bool decode(address addr, char *buf, int buflen, int* offset, const void *base) {
+    ShouldNotReachHere();
+    return false;
+  }
 
 private:
   ElfFile*         get_elf_file(const char* filepath);