diff mbox

[Pointer,Bounds,Checker,14/x] Passes [2/n] IPA passes

Message ID 20141008185506.GB13454@msticlxl57.ims.intel.com
State New
Headers show

Commit Message

Ilya Enkovich Oct. 8, 2014, 6:55 p.m. UTC
Hi,

This patch introduces two IPA passes used by Pointer Bounds Checker.  One pass creates clones for instrumentation.  The other one transforms unneeded functions into thunks.

Thanks,
Ilya
--
2014-10-08  Ilya Enkovich  <ilya.enkovich@intel.com>

	* ipa-chkp.c: New.
	* ipa-chkp.h: New.
	* Makefile.in (OBJS): Add ipa-chkp.o.

Comments

Jan Hubicka Oct. 8, 2014, 7:41 p.m. UTC | #1
> +/*  Pointer Bounds Checker has two IPA passes to support code instrumentation.
> +
> +    In instrumented code each pointer is provided with bounds.  For input
> +    pointer parameters it means we also have bounds passed.  For calls it
> +    means we have additional bounds arguments for pointer arguments.
> +
> +    To have all IPA optimizations working correctly we have to express
> +    dataflow between passed and received bounds explicitly via additional
> +    entries in function declaration arguments list and in function type.
> +    Since we may have both instrumented and not instrumented code at the
> +    same time, we cannot replace all original functions with their
> +    instrumented variants.  Therefore we create clones (versions) instead.
> +
> +    Instrumentation clones creation is a separate IPA pass which is a part
> +    of early local passes.  Clones are created after SSA is built (because
> +    instrumentation pass works on SSA) and before any transformations
> +    which may change pointer flow and therefore lead to incorrect code
> +    instrumentation (possibly causing false bounds check failures).
> +
> +    Instrumentation clones have pointer bounds arguments added right after
> +    pointer arguments.  Clones have assembler name of the original
> +    function with suffix added.  New assembler name is in transparent
> +    alias chain with the original name.  Thus we expect all calls to the
> +    original and instrumented functions look similar in assembler.

Why new alias is created when calling conventions are the same?
> +
> +    During instrumentation versioning pass we create instrumented versions
> +    of all function with body and also for all their aliases and thunks.
> +    Clones for functions with no body are created on demand (usually
> +    during call instrumentation).
> +
> +    Original and instrumented function nodes are connected with IPA
> +    reference IPA_REF_CHKP.  It is mostly done to have reachability
> +    analysis working correctly.  We may have no references to the

So you need IPA_REF_CHKP basically to prevent ipa.c from killing the function body.
One alternative I can think of is to simply represent it as direct call from clone
to original (becuase the functions are the same modulo bounds checking
and thus this is correct for all side effects analysis)

> +    instrumented function in the code but it still should be counted
> +    as reachable if the original function is reachable.
> +
> +    When original function bodies are not needed anymore we release
> +    them and transform functions into a special kind of thunks.  Each
> +    thunk has a call edge to the instrumented version.  These thunks
> +    help to keep externally visible instrumented functions visible
> +    when linker reolution files are used.  Linker has no info about
> +    connection between original and instrumented function and
> +    therefore we may wrongly decide (due to difference in assember
> +    names) that instrumented function version is local and can be
> +    removed.  */

Can you, please, give me an example what happens here?
Also how this relates to LTO?

Honza
> +
> +#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
> +
> +/* Build a clone of FNDECL with a modified name.  */
> +
> +static tree
> +chkp_build_instrumented_fndecl (tree fndecl)
> +{
> +  tree new_decl = copy_node (fndecl);
> +  tree new_name;
> +  std::string s;
> +
> +  /* We want called_as_built_in recall instrumented calls
> +     to instrumented built-in functions.  Therefore use
> +     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
> +  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
> +  s += ".chkp";
> +  DECL_NAME (new_decl) = get_identifier (s.c_str ());
> +
> +  /* References to the original and to the instrumented version
> +     should look the same in the output assembly.  And we cannot
> +     use the same assembler name for the instrumented version
> +     because it conflicts with decl merging algorithms in LTO.
> +     Achieve the result by using transparent alias name for the
> +     instrumented version.  */
> +  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
> +  s += ".chkp";
> +  new_name = get_identifier (s.c_str ());
> +  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
> +  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
> +  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
> +
> +  /* For functions with body versioning will make a copy of arguments.
> +     For functions with no body we need to do it here.  */
> +  if (!gimple_has_body_p (fndecl))
> +    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
> +
> +  /* We are going to modify attributes list and therefore should
> +     make own copy.  */
> +  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
> +
> +  return new_decl;
> +}
> +
> +
> +/* Fix operands of attribute from ATTRS list named ATTR_NAME.
> +   Integer operands are replaced with values according to
> +   INDEXES map having LEN elements.  For operands out of len
> +   we just add DELTA.  */
> +
> +static void
> +chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
> +			   unsigned *indexes, int len, int delta)
> +{
> +  tree attr = lookup_attribute (attr_name, attrs);
> +  tree op;
> +
> +  if (!attr)
> +    return;
> +
> +  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
> +  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
> +    {
> +      int idx;
> +
> +      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
> +	continue;
> +
> +      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
> +
> +      /* If idx exceeds indexes length then we just
> +	 keep it at the same distance from the last
> +	 known arg.  */
> +      if (idx > len)
> +	idx += delta;
> +      else
> +	idx = indexes[idx - 1] + 1;
> +      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
> +    }
> +}
> +
> +/* Make a copy of function type ORIG_TYPE adding pointer
> +   bounds as additional arguments.  */
> +
> +tree
> +chkp_copy_function_type_adding_bounds (tree orig_type)
> +{
> +  tree type;
> +  tree arg_type, attrs, t;
> +  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
> +  unsigned *indexes = XALLOCAVEC (unsigned, len);
> +  unsigned idx = 0, new_idx = 0;
> +
> +  for (arg_type = TYPE_ARG_TYPES (orig_type);
> +       arg_type;
> +       arg_type = TREE_CHAIN (arg_type))
> +    if (TREE_VALUE (arg_type) == void_type_node)
> +      continue;
> +    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> +	     || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> +				   TREE_VALUE (arg_type), true)
> +	     || chkp_type_has_pointer (TREE_VALUE (arg_type)))
> +      break;
> +
> +  /* We may use original type if there are no bounds passed.  */
> +  if (!arg_type)
> +    return orig_type;
> +
> +  type = copy_node (orig_type);
> +  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
> +
> +  for (arg_type = TYPE_ARG_TYPES (type);
> +       arg_type;
> +       arg_type = TREE_CHAIN (arg_type))
> +    {
> +      indexes[idx++] = new_idx++;
> +
> +      /* pass_by_reference returns 1 for void type,
> +	 so check for it first.  */
> +      if (TREE_VALUE (arg_type) == void_type_node)
> +	continue;
> +      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> +	       || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> +				     TREE_VALUE (arg_type), true))
> +	{
> +	  tree new_type = build_tree_list (NULL_TREE,
> +					   pointer_bounds_type_node);
> +	  TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> +	  TREE_CHAIN (arg_type) = new_type;
> +
> +	  arg_type = TREE_CHAIN (arg_type);
> +	  new_idx++;
> +	}
> +      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
> +	{
> +	  bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
> +	  bitmap_iterator bi;
> +	  unsigned bnd_no;
> +
> +	  EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> +	    {
> +	      tree new_type = build_tree_list (NULL_TREE,
> +					       pointer_bounds_type_node);
> +	      TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> +	      TREE_CHAIN (arg_type) = new_type;
> +
> +	      arg_type = TREE_CHAIN (arg_type);
> +	      new_idx++;
> +	    }
> +	  BITMAP_FREE (slots);
> +	}
> +    }
> +
> +  /* If function type has attribute with arg indexes then
> +     we have to copy it fixing attribute ops.  Map for
> +     fixing is in indexes array.  */
> +  attrs = TYPE_ATTRIBUTES (type);
> +  if (lookup_attribute ("nonnull", attrs)
> +      || lookup_attribute ("format", attrs)
> +      || lookup_attribute ("format_arg", attrs))
> +    {
> +      int delta = new_idx - len;
> +      attrs = copy_list (TYPE_ATTRIBUTES (type));
> +      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
> +      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
> +      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
> +      TYPE_ATTRIBUTES (type) = attrs;
> +    }
> +
> +  t = TYPE_MAIN_VARIANT (orig_type);
> +  if (orig_type != t)
> +    {
> +      TYPE_MAIN_VARIANT (type) = t;
> +      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
> +      TYPE_NEXT_VARIANT (t) = type;
> +    }
> +  else
> +    {
> +      TYPE_MAIN_VARIANT (type) = type;
> +      TYPE_NEXT_VARIANT (type) = NULL;
> +    }
> +
> +
> +  return type;
> +}
> +
> +/* For given function FNDECL add bounds arguments to arguments
> +   list.  */
> +
> +static void
> +chkp_add_bounds_params_to_function (tree fndecl)
> +{
> +  tree arg;
> +
> +  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
> +    if (BOUNDED_P (arg))
> +      {
> +	std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> +	tree new_arg;
> +
> +	if (DECL_NAME (arg))
> +	  new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
> +	else
> +	  {
> +	    char uid[25];
> +	    snprintf (uid, 25, "D.%u", DECL_UID (arg));
> +	    new_name += uid;
> +	  }
> +
> +	new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
> +			      get_identifier (new_name.c_str ()),
> +			      pointer_bounds_type_node);
> +	DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> +	DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
> +	DECL_ARTIFICIAL (new_arg) = 1;
> +	DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> +	DECL_CHAIN (arg) = new_arg;
> +
> +	arg = DECL_CHAIN (arg);
> +
> +      }
> +    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
> +      {
> +	tree orig_arg = arg;
> +	bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
> +	bitmap_iterator bi;
> +	unsigned bnd_no;
> +
> +	EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> +	  {
> +	    std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> +	    tree new_arg;
> +	    char offs[25];
> +
> +	    if (DECL_NAME (orig_arg))
> +	      new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
> +	    else
> +	      {
> +		snprintf (offs, 25, "D.%u", DECL_UID (arg));
> +		new_name += offs;
> +	      }
> +	    snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
> +
> +	    new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
> +				  PARM_DECL,
> +				  get_identifier (new_name.c_str ()),
> +				  pointer_bounds_type_node);
> +	    DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> +	    DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
> +	    DECL_ARTIFICIAL (new_arg) = 1;
> +	    DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> +	    DECL_CHAIN (arg) = new_arg;
> +
> +	    arg = DECL_CHAIN (arg);
> +	  }
> +	BITMAP_FREE (slots);
> +      }
> +
> +  TREE_TYPE (fndecl) =
> +    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
> +}
> +
> +/* Return clone created for instrumentation of NODE or NULL.  */
> +
> +cgraph_node *
> +chkp_maybe_create_clone (tree fndecl)
> +{
> +  cgraph_node *node = cgraph_node::get_create (fndecl);
> +  cgraph_node *clone = node->instrumented_version;
> +
> +  gcc_assert (!node->instrumentation_clone);
> +
> +  if (!clone)
> +    {
> +      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
> +      struct cgraph_edge *e;
> +      struct ipa_ref *ref;
> +      int i;
> +
> +      clone = node->create_version_clone (new_decl, vNULL, NULL);
> +      clone->externally_visible = node->externally_visible;
> +      clone->local = node->local;
> +      clone->address_taken = node->address_taken;
> +      clone->thunk = node->thunk;
> +      clone->alias = node->alias;
> +      clone->weakref = node->weakref;
> +      clone->cpp_implicit_alias = node->cpp_implicit_alias;
> +      clone->instrumented_version = node;
> +      clone->orig_decl = fndecl;
> +      clone->instrumentation_clone = true;
> +      node->instrumented_version = clone;
> +
> +      if (gimple_has_body_p (fndecl))
> +	{
> +	  /* If function will not be instrumented, then it's instrumented
> +	     version is a thunk for the original.  */
> +	  if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
> +	      || (flag_chkp_instrument_marked_only
> +		  && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
> +	    {
> +	      clone->thunk.thunk_p = true;
> +	      clone->thunk.add_pointer_bounds_args = true;
> +	      clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
> +	    }
> +	  else
> +	    {
> +	      tree_function_versioning (fndecl, new_decl, NULL, false,
> +					NULL, false, NULL, NULL);
> +	      clone->lowered = true;
> +	    }
> +	}
> +
> +      /* New params are inserted after versioning because it
> +	 actually copies args list from the original decl.  */
> +      chkp_add_bounds_params_to_function (new_decl);
> +
> +      /* Clones have the same comdat group as originals.  */
> +      if (node->same_comdat_group
> +	  || DECL_ONE_ONLY (node->decl))
> +	clone->add_to_same_comdat_group (node);
> +
> +      if (gimple_has_body_p (fndecl))
> +	symtab->call_cgraph_insertion_hooks (clone);
> +
> +      /* Clone all aliases.  */
> +      for (i = 0; node->iterate_referring (i, ref); i++)
> +	if (ref->use == IPA_REF_ALIAS)
> +	  {
> +	    struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
> +	    struct cgraph_node *chkp_alias
> +	      = chkp_maybe_create_clone (alias->decl);
> +	    chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
> +	  }
> +
> +      /* Clone all thunks.  */
> +      for (e = node->callers; e; e = e->next_caller)
> +	if (e->caller->thunk.thunk_p)
> +	  {
> +	    struct cgraph_node *thunk
> +	      = chkp_maybe_create_clone (e->caller->decl);
> +	    /* Redirect thunk clone edge to the node clone.  */
> +	    thunk->callees->redirect_callee (clone);
> +	  }
> +
> +      /* For aliases and thunks we should make sure target is cloned
> +	 to have proper references and edges.  */
> +      if (node->thunk.thunk_p)
> +	chkp_maybe_create_clone (node->callees->callee->decl);
> +      else if (node->alias)
> +	{
> +	  struct cgraph_node *target;
> +
> +	  ref = node->ref_list.first_reference ();
> +	  if (ref)
> +	    chkp_maybe_create_clone (ref->referred->decl);
> +
> +	  if (node->alias_target)
> +	    {
> +	      if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
> +		{
> +		  target = chkp_maybe_create_clone (node->alias_target);
> +		  clone->alias_target = target->decl;
> +		}
> +	      else
> +		clone->alias_target = node->alias_target;
> +	    }
> +	}
> +
> +      /* Add IPA reference.  It's main role is to keep instrumented
> +	 version reachable while original node is reachable.  */
> +      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
> +    }
> +
> +  return clone;
> +}
> +
> +/* Create clone for all functions to be instrumented.  */
> +
> +static unsigned int
> +chkp_versioning (void)
> +{
> +  struct cgraph_node *node;
> +
> +  FOR_EACH_DEFINED_FUNCTION (node)
> +    {
> +      if (!node->instrumentation_clone
> +	  && !node->instrumented_version
> +	  && !node->alias
> +	  && !node->thunk.thunk_p
> +	  && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
> +	  && (!flag_chkp_instrument_marked_only
> +	      || lookup_attribute ("bnd_instrument",
> +				   DECL_ATTRIBUTES (node->decl)))
> +	  /* No builtins instrumentation for now.  */
> +	  && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
> +	chkp_maybe_create_clone (node->decl);
> +    }
> +
> +  /* Mark all aliases and thunks of functions with no instrumented
> +     version as legacy function.  */
> +  FOR_EACH_DEFINED_FUNCTION (node)
> +    {
> +      if (!node->instrumentation_clone
> +	  && !node->instrumented_version
> +	  && (node->alias || node->thunk.thunk_p)
> +	  && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
> +	DECL_ATTRIBUTES (node->decl)
> +	  = tree_cons (get_identifier ("bnd_legacy"), NULL,
> +		       DECL_ATTRIBUTES (node->decl));
> +    }
> +
> +  return 0;
> +}
> +
> +/* In this pass we remove bodies of functions having
> +   instrumented version.  Functions with removed bodies
> +   become a special kind of thunks to provide a connection
> +   between calls to the original version and instrumented
> +   function.  */
> +
> +static unsigned int
> +chkp_produce_thunks (void)
> +{
> +  struct cgraph_node *node;
> +
> +  FOR_EACH_DEFINED_FUNCTION (node)
> +    {
> +      if (!node->instrumentation_clone
> +	  && node->instrumented_version
> +	  && gimple_has_body_p (node->decl)
> +	  && gimple_has_body_p (node->instrumented_version->decl))
> +	{
> +	  node->release_body ();
> +	  node->remove_callees ();
> +	  node->remove_all_references ();
> +
> +	  node->thunk.thunk_p = true;
> +	  node->thunk.add_pointer_bounds_args = true;
> +	  node->create_edge (node->instrumented_version, NULL,
> +			     0, CGRAPH_FREQ_BASE);
> +	  node->create_reference (node->instrumented_version,
> +			       IPA_REF_CHKP, NULL);
> +	}
> +    }
> +
> +  /* Mark instrumentation clones created for aliases and thunks
> +     as insttrumented so they could be removed as unreachable
> +     now.  */
> +  FOR_EACH_DEFINED_FUNCTION (node)
> +    {
> +      if (node->instrumentation_clone
> +	  && (node->alias || node->thunk.thunk_p)
> +	  && !chkp_function_instrumented_p (node->decl))
> +	chkp_function_mark_instrumented (node->decl);
> +    }
> +
> +  symtab->remove_unreachable_nodes (true, dump_file);
> +
> +  return 0;
> +}
> +
> +const pass_data pass_data_ipa_chkp_versioning =
> +{
> +  SIMPLE_IPA_PASS, /* type */
> +  "chkp_versioning", /* name */
> +  OPTGROUP_NONE, /* optinfo_flags */
> +  TV_NONE, /* tv_id */
> +  0, /* properties_required */
> +  0, /* properties_provided */
> +  0, /* properties_destroyed */
> +  0, /* todo_flags_start */
> +  0 /* todo_flags_finish */
> +};
> +
> +const pass_data pass_data_ipa_chkp_produce_thunks =
> +{
> +  SIMPLE_IPA_PASS, /* type */
> +  "chkp_cleanup", /* name */
> +  OPTGROUP_NONE, /* optinfo_flags */
> +  TV_NONE, /* tv_id */
> +  0, /* properties_required */
> +  0, /* properties_provided */
> +  0, /* properties_destroyed */
> +  0, /* todo_flags_start */
> +  0 /* todo_flags_finish */
> +};
> +
> +class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
> +{
> +public:
> +  pass_ipa_chkp_versioning (gcc::context *ctxt)
> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
> +  {}
> +
> +  /* opt_pass methods: */
> +  virtual opt_pass * clone ()
> +    {
> +      return new pass_ipa_chkp_versioning (m_ctxt);
> +    }
> +
> +  virtual bool gate (function *)
> +    {
> +      return flag_check_pointer_bounds;
> +    }
> +
> +  virtual unsigned int execute (function *)
> +    {
> +      return chkp_versioning ();
> +    }
> +
> +}; // class pass_ipa_chkp_versioning
> +
> +class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
> +{
> +public:
> +  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
> +  {}
> +
> +  /* opt_pass methods: */
> +  virtual opt_pass * clone ()
> +    {
> +      return new pass_ipa_chkp_produce_thunks (m_ctxt);
> +    }
> +
> +  virtual unsigned int execute (function *)
> +    {
> +      return chkp_produce_thunks ();
> +    }
> +
> +}; // class pass_chkp_produce_thunks
> +
> +simple_ipa_opt_pass *
> +make_pass_ipa_chkp_versioning (gcc::context *ctxt)
> +{
> +  return new pass_ipa_chkp_versioning (ctxt);
> +}
> +
> +simple_ipa_opt_pass *
> +make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> +{
> +  return new pass_ipa_chkp_produce_thunks (ctxt);
> +}
> diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
> new file mode 100644
> index 0000000..9c92c04
> --- /dev/null
> +++ b/gcc/ipa-chkp.h
> @@ -0,0 +1,29 @@
> +/* Declaration of interface functions of Pointer Bounds Checker.
> +   Copyright (C) 2014 Free Software Foundation, Inc.
> +
> +This file is part of GCC.
> +
> +GCC is free software; you can redistribute it and/or modify it under
> +the terms of the GNU General Public License as published by the Free
> +Software Foundation; either version 3, or (at your option) any later
> +version.
> +
> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
> +for more details.
> +
> +You should have received a copy of the GNU General Public License
> +along with GCC; see the file COPYING3.  If not see
> +<http://www.gnu.org/licenses/>.  */
> +
> +#ifndef GCC_IPA_CHKP_H
> +#define GCC_IPA_CHKP_H
> +
> +#include "tree.h"
> +#include "cgraph.h"
> +
> +extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
> +extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
> +
> +#endif /* GCC_IPA_CHKP_H */
Ilya Enkovich Oct. 8, 2014, 9:18 p.m. UTC | #2
2014-10-08 23:41 GMT+04:00 Jan Hubicka <hubicka@ucw.cz>:
>> +/*  Pointer Bounds Checker has two IPA passes to support code instrumentation.
>> +
>> +    In instrumented code each pointer is provided with bounds.  For input
>> +    pointer parameters it means we also have bounds passed.  For calls it
>> +    means we have additional bounds arguments for pointer arguments.
>> +
>> +    To have all IPA optimizations working correctly we have to express
>> +    dataflow between passed and received bounds explicitly via additional
>> +    entries in function declaration arguments list and in function type.
>> +    Since we may have both instrumented and not instrumented code at the
>> +    same time, we cannot replace all original functions with their
>> +    instrumented variants.  Therefore we create clones (versions) instead.
>> +
>> +    Instrumentation clones creation is a separate IPA pass which is a part
>> +    of early local passes.  Clones are created after SSA is built (because
>> +    instrumentation pass works on SSA) and before any transformations
>> +    which may change pointer flow and therefore lead to incorrect code
>> +    instrumentation (possibly causing false bounds check failures).
>> +
>> +    Instrumentation clones have pointer bounds arguments added right after
>> +    pointer arguments.  Clones have assembler name of the original
>> +    function with suffix added.  New assembler name is in transparent
>> +    alias chain with the original name.  Thus we expect all calls to the
>> +    original and instrumented functions look similar in assembler.
>
> Why new alias is created when calling conventions are the same?

Instrumented versions have additional input arguments.

>> +
>> +    During instrumentation versioning pass we create instrumented versions
>> +    of all function with body and also for all their aliases and thunks.
>> +    Clones for functions with no body are created on demand (usually
>> +    during call instrumentation).
>> +
>> +    Original and instrumented function nodes are connected with IPA
>> +    reference IPA_REF_CHKP.  It is mostly done to have reachability
>> +    analysis working correctly.  We may have no references to the
>
> So you need IPA_REF_CHKP basically to prevent ipa.c from killing the function body.
> One alternative I can think of is to simply represent it as direct call from clone
> to original (becuase the functions are the same modulo bounds checking
> and thus this is correct for all side effects analysis)

I prevent clone's body from removal and therefore original should call
clone (otherwise clone may have no callers and be removed).  I think
call edge may work (need to recall other cases when reference do its
work) but is it OK to have call with no stmt for non thunk nodes?

>
>> +    instrumented function in the code but it still should be counted
>> +    as reachable if the original function is reachable.
>> +
>> +    When original function bodies are not needed anymore we release
>> +    them and transform functions into a special kind of thunks.  Each
>> +    thunk has a call edge to the instrumented version.  These thunks
>> +    help to keep externally visible instrumented functions visible
>> +    when linker reolution files are used.  Linker has no info about
>> +    connection between original and instrumented function and
>> +    therefore we may wrongly decide (due to difference in assember
>> +    names) that instrumented function version is local and can be
>> +    removed.  */
>
> Can you, please, give me an example what happens here?
> Also how this relates to LTO?

When we compile with LTO linker provides a resolution file with
functions usage information.  Consider I have external function test
and its clone test.chkp.  Linker says I have calls to test and no
calls to test.chkp.  Compiler just removes test.chkp then.  It doesn't
happen when test is trunsformed into a thunk of test.chkp.

Ilya

>
> Honza
>> +
>> +#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
>> +
>> +/* Build a clone of FNDECL with a modified name.  */
>> +
>> +static tree
>> +chkp_build_instrumented_fndecl (tree fndecl)
>> +{
>> +  tree new_decl = copy_node (fndecl);
>> +  tree new_name;
>> +  std::string s;
>> +
>> +  /* We want called_as_built_in recall instrumented calls
>> +     to instrumented built-in functions.  Therefore use
>> +     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
>> +  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
>> +  s += ".chkp";
>> +  DECL_NAME (new_decl) = get_identifier (s.c_str ());
>> +
>> +  /* References to the original and to the instrumented version
>> +     should look the same in the output assembly.  And we cannot
>> +     use the same assembler name for the instrumented version
>> +     because it conflicts with decl merging algorithms in LTO.
>> +     Achieve the result by using transparent alias name for the
>> +     instrumented version.  */
>> +  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
>> +  s += ".chkp";
>> +  new_name = get_identifier (s.c_str ());
>> +  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
>> +  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
>> +  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
>> +
>> +  /* For functions with body versioning will make a copy of arguments.
>> +     For functions with no body we need to do it here.  */
>> +  if (!gimple_has_body_p (fndecl))
>> +    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
>> +
>> +  /* We are going to modify attributes list and therefore should
>> +     make own copy.  */
>> +  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
>> +
>> +  return new_decl;
>> +}
>> +
>> +
>> +/* Fix operands of attribute from ATTRS list named ATTR_NAME.
>> +   Integer operands are replaced with values according to
>> +   INDEXES map having LEN elements.  For operands out of len
>> +   we just add DELTA.  */
>> +
>> +static void
>> +chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
>> +                        unsigned *indexes, int len, int delta)
>> +{
>> +  tree attr = lookup_attribute (attr_name, attrs);
>> +  tree op;
>> +
>> +  if (!attr)
>> +    return;
>> +
>> +  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
>> +  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
>> +    {
>> +      int idx;
>> +
>> +      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
>> +     continue;
>> +
>> +      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
>> +
>> +      /* If idx exceeds indexes length then we just
>> +      keep it at the same distance from the last
>> +      known arg.  */
>> +      if (idx > len)
>> +     idx += delta;
>> +      else
>> +     idx = indexes[idx - 1] + 1;
>> +      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
>> +    }
>> +}
>> +
>> +/* Make a copy of function type ORIG_TYPE adding pointer
>> +   bounds as additional arguments.  */
>> +
>> +tree
>> +chkp_copy_function_type_adding_bounds (tree orig_type)
>> +{
>> +  tree type;
>> +  tree arg_type, attrs, t;
>> +  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
>> +  unsigned *indexes = XALLOCAVEC (unsigned, len);
>> +  unsigned idx = 0, new_idx = 0;
>> +
>> +  for (arg_type = TYPE_ARG_TYPES (orig_type);
>> +       arg_type;
>> +       arg_type = TREE_CHAIN (arg_type))
>> +    if (TREE_VALUE (arg_type) == void_type_node)
>> +      continue;
>> +    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
>> +          || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
>> +                                TREE_VALUE (arg_type), true)
>> +          || chkp_type_has_pointer (TREE_VALUE (arg_type)))
>> +      break;
>> +
>> +  /* We may use original type if there are no bounds passed.  */
>> +  if (!arg_type)
>> +    return orig_type;
>> +
>> +  type = copy_node (orig_type);
>> +  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
>> +
>> +  for (arg_type = TYPE_ARG_TYPES (type);
>> +       arg_type;
>> +       arg_type = TREE_CHAIN (arg_type))
>> +    {
>> +      indexes[idx++] = new_idx++;
>> +
>> +      /* pass_by_reference returns 1 for void type,
>> +      so check for it first.  */
>> +      if (TREE_VALUE (arg_type) == void_type_node)
>> +     continue;
>> +      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
>> +            || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
>> +                                  TREE_VALUE (arg_type), true))
>> +     {
>> +       tree new_type = build_tree_list (NULL_TREE,
>> +                                        pointer_bounds_type_node);
>> +       TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
>> +       TREE_CHAIN (arg_type) = new_type;
>> +
>> +       arg_type = TREE_CHAIN (arg_type);
>> +       new_idx++;
>> +     }
>> +      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
>> +     {
>> +       bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
>> +       bitmap_iterator bi;
>> +       unsigned bnd_no;
>> +
>> +       EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
>> +         {
>> +           tree new_type = build_tree_list (NULL_TREE,
>> +                                            pointer_bounds_type_node);
>> +           TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
>> +           TREE_CHAIN (arg_type) = new_type;
>> +
>> +           arg_type = TREE_CHAIN (arg_type);
>> +           new_idx++;
>> +         }
>> +       BITMAP_FREE (slots);
>> +     }
>> +    }
>> +
>> +  /* If function type has attribute with arg indexes then
>> +     we have to copy it fixing attribute ops.  Map for
>> +     fixing is in indexes array.  */
>> +  attrs = TYPE_ATTRIBUTES (type);
>> +  if (lookup_attribute ("nonnull", attrs)
>> +      || lookup_attribute ("format", attrs)
>> +      || lookup_attribute ("format_arg", attrs))
>> +    {
>> +      int delta = new_idx - len;
>> +      attrs = copy_list (TYPE_ATTRIBUTES (type));
>> +      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
>> +      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
>> +      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
>> +      TYPE_ATTRIBUTES (type) = attrs;
>> +    }
>> +
>> +  t = TYPE_MAIN_VARIANT (orig_type);
>> +  if (orig_type != t)
>> +    {
>> +      TYPE_MAIN_VARIANT (type) = t;
>> +      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
>> +      TYPE_NEXT_VARIANT (t) = type;
>> +    }
>> +  else
>> +    {
>> +      TYPE_MAIN_VARIANT (type) = type;
>> +      TYPE_NEXT_VARIANT (type) = NULL;
>> +    }
>> +
>> +
>> +  return type;
>> +}
>> +
>> +/* For given function FNDECL add bounds arguments to arguments
>> +   list.  */
>> +
>> +static void
>> +chkp_add_bounds_params_to_function (tree fndecl)
>> +{
>> +  tree arg;
>> +
>> +  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
>> +    if (BOUNDED_P (arg))
>> +      {
>> +     std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
>> +     tree new_arg;
>> +
>> +     if (DECL_NAME (arg))
>> +       new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
>> +     else
>> +       {
>> +         char uid[25];
>> +         snprintf (uid, 25, "D.%u", DECL_UID (arg));
>> +         new_name += uid;
>> +       }
>> +
>> +     new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
>> +                           get_identifier (new_name.c_str ()),
>> +                           pointer_bounds_type_node);
>> +     DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
>> +     DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
>> +     DECL_ARTIFICIAL (new_arg) = 1;
>> +     DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
>> +     DECL_CHAIN (arg) = new_arg;
>> +
>> +     arg = DECL_CHAIN (arg);
>> +
>> +      }
>> +    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
>> +      {
>> +     tree orig_arg = arg;
>> +     bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
>> +     bitmap_iterator bi;
>> +     unsigned bnd_no;
>> +
>> +     EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
>> +       {
>> +         std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
>> +         tree new_arg;
>> +         char offs[25];
>> +
>> +         if (DECL_NAME (orig_arg))
>> +           new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
>> +         else
>> +           {
>> +             snprintf (offs, 25, "D.%u", DECL_UID (arg));
>> +             new_name += offs;
>> +           }
>> +         snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
>> +
>> +         new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
>> +                               PARM_DECL,
>> +                               get_identifier (new_name.c_str ()),
>> +                               pointer_bounds_type_node);
>> +         DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
>> +         DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
>> +         DECL_ARTIFICIAL (new_arg) = 1;
>> +         DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
>> +         DECL_CHAIN (arg) = new_arg;
>> +
>> +         arg = DECL_CHAIN (arg);
>> +       }
>> +     BITMAP_FREE (slots);
>> +      }
>> +
>> +  TREE_TYPE (fndecl) =
>> +    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
>> +}
>> +
>> +/* Return clone created for instrumentation of NODE or NULL.  */
>> +
>> +cgraph_node *
>> +chkp_maybe_create_clone (tree fndecl)
>> +{
>> +  cgraph_node *node = cgraph_node::get_create (fndecl);
>> +  cgraph_node *clone = node->instrumented_version;
>> +
>> +  gcc_assert (!node->instrumentation_clone);
>> +
>> +  if (!clone)
>> +    {
>> +      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
>> +      struct cgraph_edge *e;
>> +      struct ipa_ref *ref;
>> +      int i;
>> +
>> +      clone = node->create_version_clone (new_decl, vNULL, NULL);
>> +      clone->externally_visible = node->externally_visible;
>> +      clone->local = node->local;
>> +      clone->address_taken = node->address_taken;
>> +      clone->thunk = node->thunk;
>> +      clone->alias = node->alias;
>> +      clone->weakref = node->weakref;
>> +      clone->cpp_implicit_alias = node->cpp_implicit_alias;
>> +      clone->instrumented_version = node;
>> +      clone->orig_decl = fndecl;
>> +      clone->instrumentation_clone = true;
>> +      node->instrumented_version = clone;
>> +
>> +      if (gimple_has_body_p (fndecl))
>> +     {
>> +       /* If function will not be instrumented, then it's instrumented
>> +          version is a thunk for the original.  */
>> +       if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
>> +           || (flag_chkp_instrument_marked_only
>> +               && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
>> +         {
>> +           clone->thunk.thunk_p = true;
>> +           clone->thunk.add_pointer_bounds_args = true;
>> +           clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
>> +         }
>> +       else
>> +         {
>> +           tree_function_versioning (fndecl, new_decl, NULL, false,
>> +                                     NULL, false, NULL, NULL);
>> +           clone->lowered = true;
>> +         }
>> +     }
>> +
>> +      /* New params are inserted after versioning because it
>> +      actually copies args list from the original decl.  */
>> +      chkp_add_bounds_params_to_function (new_decl);
>> +
>> +      /* Clones have the same comdat group as originals.  */
>> +      if (node->same_comdat_group
>> +       || DECL_ONE_ONLY (node->decl))
>> +     clone->add_to_same_comdat_group (node);
>> +
>> +      if (gimple_has_body_p (fndecl))
>> +     symtab->call_cgraph_insertion_hooks (clone);
>> +
>> +      /* Clone all aliases.  */
>> +      for (i = 0; node->iterate_referring (i, ref); i++)
>> +     if (ref->use == IPA_REF_ALIAS)
>> +       {
>> +         struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
>> +         struct cgraph_node *chkp_alias
>> +           = chkp_maybe_create_clone (alias->decl);
>> +         chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
>> +       }
>> +
>> +      /* Clone all thunks.  */
>> +      for (e = node->callers; e; e = e->next_caller)
>> +     if (e->caller->thunk.thunk_p)
>> +       {
>> +         struct cgraph_node *thunk
>> +           = chkp_maybe_create_clone (e->caller->decl);
>> +         /* Redirect thunk clone edge to the node clone.  */
>> +         thunk->callees->redirect_callee (clone);
>> +       }
>> +
>> +      /* For aliases and thunks we should make sure target is cloned
>> +      to have proper references and edges.  */
>> +      if (node->thunk.thunk_p)
>> +     chkp_maybe_create_clone (node->callees->callee->decl);
>> +      else if (node->alias)
>> +     {
>> +       struct cgraph_node *target;
>> +
>> +       ref = node->ref_list.first_reference ();
>> +       if (ref)
>> +         chkp_maybe_create_clone (ref->referred->decl);
>> +
>> +       if (node->alias_target)
>> +         {
>> +           if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
>> +             {
>> +               target = chkp_maybe_create_clone (node->alias_target);
>> +               clone->alias_target = target->decl;
>> +             }
>> +           else
>> +             clone->alias_target = node->alias_target;
>> +         }
>> +     }
>> +
>> +      /* Add IPA reference.  It's main role is to keep instrumented
>> +      version reachable while original node is reachable.  */
>> +      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
>> +    }
>> +
>> +  return clone;
>> +}
>> +
>> +/* Create clone for all functions to be instrumented.  */
>> +
>> +static unsigned int
>> +chkp_versioning (void)
>> +{
>> +  struct cgraph_node *node;
>> +
>> +  FOR_EACH_DEFINED_FUNCTION (node)
>> +    {
>> +      if (!node->instrumentation_clone
>> +       && !node->instrumented_version
>> +       && !node->alias
>> +       && !node->thunk.thunk_p
>> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
>> +       && (!flag_chkp_instrument_marked_only
>> +           || lookup_attribute ("bnd_instrument",
>> +                                DECL_ATTRIBUTES (node->decl)))
>> +       /* No builtins instrumentation for now.  */
>> +       && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
>> +     chkp_maybe_create_clone (node->decl);
>> +    }
>> +
>> +  /* Mark all aliases and thunks of functions with no instrumented
>> +     version as legacy function.  */
>> +  FOR_EACH_DEFINED_FUNCTION (node)
>> +    {
>> +      if (!node->instrumentation_clone
>> +       && !node->instrumented_version
>> +       && (node->alias || node->thunk.thunk_p)
>> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
>> +     DECL_ATTRIBUTES (node->decl)
>> +       = tree_cons (get_identifier ("bnd_legacy"), NULL,
>> +                    DECL_ATTRIBUTES (node->decl));
>> +    }
>> +
>> +  return 0;
>> +}
>> +
>> +/* In this pass we remove bodies of functions having
>> +   instrumented version.  Functions with removed bodies
>> +   become a special kind of thunks to provide a connection
>> +   between calls to the original version and instrumented
>> +   function.  */
>> +
>> +static unsigned int
>> +chkp_produce_thunks (void)
>> +{
>> +  struct cgraph_node *node;
>> +
>> +  FOR_EACH_DEFINED_FUNCTION (node)
>> +    {
>> +      if (!node->instrumentation_clone
>> +       && node->instrumented_version
>> +       && gimple_has_body_p (node->decl)
>> +       && gimple_has_body_p (node->instrumented_version->decl))
>> +     {
>> +       node->release_body ();
>> +       node->remove_callees ();
>> +       node->remove_all_references ();
>> +
>> +       node->thunk.thunk_p = true;
>> +       node->thunk.add_pointer_bounds_args = true;
>> +       node->create_edge (node->instrumented_version, NULL,
>> +                          0, CGRAPH_FREQ_BASE);
>> +       node->create_reference (node->instrumented_version,
>> +                            IPA_REF_CHKP, NULL);
>> +     }
>> +    }
>> +
>> +  /* Mark instrumentation clones created for aliases and thunks
>> +     as insttrumented so they could be removed as unreachable
>> +     now.  */
>> +  FOR_EACH_DEFINED_FUNCTION (node)
>> +    {
>> +      if (node->instrumentation_clone
>> +       && (node->alias || node->thunk.thunk_p)
>> +       && !chkp_function_instrumented_p (node->decl))
>> +     chkp_function_mark_instrumented (node->decl);
>> +    }
>> +
>> +  symtab->remove_unreachable_nodes (true, dump_file);
>> +
>> +  return 0;
>> +}
>> +
>> +const pass_data pass_data_ipa_chkp_versioning =
>> +{
>> +  SIMPLE_IPA_PASS, /* type */
>> +  "chkp_versioning", /* name */
>> +  OPTGROUP_NONE, /* optinfo_flags */
>> +  TV_NONE, /* tv_id */
>> +  0, /* properties_required */
>> +  0, /* properties_provided */
>> +  0, /* properties_destroyed */
>> +  0, /* todo_flags_start */
>> +  0 /* todo_flags_finish */
>> +};
>> +
>> +const pass_data pass_data_ipa_chkp_produce_thunks =
>> +{
>> +  SIMPLE_IPA_PASS, /* type */
>> +  "chkp_cleanup", /* name */
>> +  OPTGROUP_NONE, /* optinfo_flags */
>> +  TV_NONE, /* tv_id */
>> +  0, /* properties_required */
>> +  0, /* properties_provided */
>> +  0, /* properties_destroyed */
>> +  0, /* todo_flags_start */
>> +  0 /* todo_flags_finish */
>> +};
>> +
>> +class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
>> +{
>> +public:
>> +  pass_ipa_chkp_versioning (gcc::context *ctxt)
>> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
>> +  {}
>> +
>> +  /* opt_pass methods: */
>> +  virtual opt_pass * clone ()
>> +    {
>> +      return new pass_ipa_chkp_versioning (m_ctxt);
>> +    }
>> +
>> +  virtual bool gate (function *)
>> +    {
>> +      return flag_check_pointer_bounds;
>> +    }
>> +
>> +  virtual unsigned int execute (function *)
>> +    {
>> +      return chkp_versioning ();
>> +    }
>> +
>> +}; // class pass_ipa_chkp_versioning
>> +
>> +class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
>> +{
>> +public:
>> +  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
>> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
>> +  {}
>> +
>> +  /* opt_pass methods: */
>> +  virtual opt_pass * clone ()
>> +    {
>> +      return new pass_ipa_chkp_produce_thunks (m_ctxt);
>> +    }
>> +
>> +  virtual unsigned int execute (function *)
>> +    {
>> +      return chkp_produce_thunks ();
>> +    }
>> +
>> +}; // class pass_chkp_produce_thunks
>> +
>> +simple_ipa_opt_pass *
>> +make_pass_ipa_chkp_versioning (gcc::context *ctxt)
>> +{
>> +  return new pass_ipa_chkp_versioning (ctxt);
>> +}
>> +
>> +simple_ipa_opt_pass *
>> +make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
>> +{
>> +  return new pass_ipa_chkp_produce_thunks (ctxt);
>> +}
>> diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
>> new file mode 100644
>> index 0000000..9c92c04
>> --- /dev/null
>> +++ b/gcc/ipa-chkp.h
>> @@ -0,0 +1,29 @@
>> +/* Declaration of interface functions of Pointer Bounds Checker.
>> +   Copyright (C) 2014 Free Software Foundation, Inc.
>> +
>> +This file is part of GCC.
>> +
>> +GCC is free software; you can redistribute it and/or modify it under
>> +the terms of the GNU General Public License as published by the Free
>> +Software Foundation; either version 3, or (at your option) any later
>> +version.
>> +
>> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
>> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
>> +for more details.
>> +
>> +You should have received a copy of the GNU General Public License
>> +along with GCC; see the file COPYING3.  If not see
>> +<http://www.gnu.org/licenses/>.  */
>> +
>> +#ifndef GCC_IPA_CHKP_H
>> +#define GCC_IPA_CHKP_H
>> +
>> +#include "tree.h"
>> +#include "cgraph.h"
>> +
>> +extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
>> +extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
>> +
>> +#endif /* GCC_IPA_CHKP_H */
Jan Hubicka Oct. 8, 2014, 10:36 p.m. UTC | #3
> 2014-10-08 23:41 GMT+04:00 Jan Hubicka <hubicka@ucw.cz>:
> >> +/*  Pointer Bounds Checker has two IPA passes to support code instrumentation.
> >> +
> >> +    In instrumented code each pointer is provided with bounds.  For input
> >> +    pointer parameters it means we also have bounds passed.  For calls it
> >> +    means we have additional bounds arguments for pointer arguments.
> >> +
> >> +    To have all IPA optimizations working correctly we have to express
> >> +    dataflow between passed and received bounds explicitly via additional
> >> +    entries in function declaration arguments list and in function type.
> >> +    Since we may have both instrumented and not instrumented code at the
> >> +    same time, we cannot replace all original functions with their
> >> +    instrumented variants.  Therefore we create clones (versions) instead.
> >> +
> >> +    Instrumentation clones creation is a separate IPA pass which is a part
> >> +    of early local passes.  Clones are created after SSA is built (because
> >> +    instrumentation pass works on SSA) and before any transformations
> >> +    which may change pointer flow and therefore lead to incorrect code
> >> +    instrumentation (possibly causing false bounds check failures).
> >> +
> >> +    Instrumentation clones have pointer bounds arguments added right after
> >> +    pointer arguments.  Clones have assembler name of the original
> >> +    function with suffix added.  New assembler name is in transparent
> >> +    alias chain with the original name.  Thus we expect all calls to the
> >> +    original and instrumented functions look similar in assembler.
> >
> > Why new alias is created when calling conventions are the same?
> 
> Instrumented versions have additional input arguments.

OK, can you explain be abit the lifetime of clones you create?  So if I undertand it
right, for every foo(ptr) you also create foo.chkp(ptr,bounds) clone that is produced
during early optimization.

Then you transfer all direct calls in the code compiled with bounds checking into foo.chkp
and eventually depending if you have any need for both foo and foo.chkp, you may turn
foo into a thunk calling foo.chkp with empty bound information.
If either of the variants disappears, you keep either the original funcion or the clone
alone?

Honza
> 
> >> +
> >> +    During instrumentation versioning pass we create instrumented versions
> >> +    of all function with body and also for all their aliases and thunks.
> >> +    Clones for functions with no body are created on demand (usually
> >> +    during call instrumentation).
> >> +
> >> +    Original and instrumented function nodes are connected with IPA
> >> +    reference IPA_REF_CHKP.  It is mostly done to have reachability
> >> +    analysis working correctly.  We may have no references to the
> >
> > So you need IPA_REF_CHKP basically to prevent ipa.c from killing the function body.
> > One alternative I can think of is to simply represent it as direct call from clone
> > to original (becuase the functions are the same modulo bounds checking
> > and thus this is correct for all side effects analysis)
> 
> I prevent clone's body from removal and therefore original should call
> clone (otherwise clone may have no callers and be removed).  I think
> call edge may work (need to recall other cases when reference do its
> work) but is it OK to have call with no stmt for non thunk nodes?
> 
> >
> >> +    instrumented function in the code but it still should be counted
> >> +    as reachable if the original function is reachable.
> >> +
> >> +    When original function bodies are not needed anymore we release
> >> +    them and transform functions into a special kind of thunks.  Each
> >> +    thunk has a call edge to the instrumented version.  These thunks
> >> +    help to keep externally visible instrumented functions visible
> >> +    when linker reolution files are used.  Linker has no info about
> >> +    connection between original and instrumented function and
> >> +    therefore we may wrongly decide (due to difference in assember
> >> +    names) that instrumented function version is local and can be
> >> +    removed.  */
> >
> > Can you, please, give me an example what happens here?
> > Also how this relates to LTO?
> 
> When we compile with LTO linker provides a resolution file with
> functions usage information.  Consider I have external function test
> and its clone test.chkp.  Linker says I have calls to test and no
> calls to test.chkp.  Compiler just removes test.chkp then.  It doesn't
> happen when test is trunsformed into a thunk of test.chkp.
> 
> Ilya
> 
> >
> > Honza
> >> +
> >> +#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
> >> +
> >> +/* Build a clone of FNDECL with a modified name.  */
> >> +
> >> +static tree
> >> +chkp_build_instrumented_fndecl (tree fndecl)
> >> +{
> >> +  tree new_decl = copy_node (fndecl);
> >> +  tree new_name;
> >> +  std::string s;
> >> +
> >> +  /* We want called_as_built_in recall instrumented calls
> >> +     to instrumented built-in functions.  Therefore use
> >> +     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
> >> +  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
> >> +  s += ".chkp";
> >> +  DECL_NAME (new_decl) = get_identifier (s.c_str ());
> >> +
> >> +  /* References to the original and to the instrumented version
> >> +     should look the same in the output assembly.  And we cannot
> >> +     use the same assembler name for the instrumented version
> >> +     because it conflicts with decl merging algorithms in LTO.
> >> +     Achieve the result by using transparent alias name for the
> >> +     instrumented version.  */
> >> +  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
> >> +  s += ".chkp";
> >> +  new_name = get_identifier (s.c_str ());
> >> +  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
> >> +  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
> >> +  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
> >> +
> >> +  /* For functions with body versioning will make a copy of arguments.
> >> +     For functions with no body we need to do it here.  */
> >> +  if (!gimple_has_body_p (fndecl))
> >> +    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
> >> +
> >> +  /* We are going to modify attributes list and therefore should
> >> +     make own copy.  */
> >> +  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
> >> +
> >> +  return new_decl;
> >> +}
> >> +
> >> +
> >> +/* Fix operands of attribute from ATTRS list named ATTR_NAME.
> >> +   Integer operands are replaced with values according to
> >> +   INDEXES map having LEN elements.  For operands out of len
> >> +   we just add DELTA.  */
> >> +
> >> +static void
> >> +chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
> >> +                        unsigned *indexes, int len, int delta)
> >> +{
> >> +  tree attr = lookup_attribute (attr_name, attrs);
> >> +  tree op;
> >> +
> >> +  if (!attr)
> >> +    return;
> >> +
> >> +  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
> >> +  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
> >> +    {
> >> +      int idx;
> >> +
> >> +      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
> >> +     continue;
> >> +
> >> +      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
> >> +
> >> +      /* If idx exceeds indexes length then we just
> >> +      keep it at the same distance from the last
> >> +      known arg.  */
> >> +      if (idx > len)
> >> +     idx += delta;
> >> +      else
> >> +     idx = indexes[idx - 1] + 1;
> >> +      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
> >> +    }
> >> +}
> >> +
> >> +/* Make a copy of function type ORIG_TYPE adding pointer
> >> +   bounds as additional arguments.  */
> >> +
> >> +tree
> >> +chkp_copy_function_type_adding_bounds (tree orig_type)
> >> +{
> >> +  tree type;
> >> +  tree arg_type, attrs, t;
> >> +  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
> >> +  unsigned *indexes = XALLOCAVEC (unsigned, len);
> >> +  unsigned idx = 0, new_idx = 0;
> >> +
> >> +  for (arg_type = TYPE_ARG_TYPES (orig_type);
> >> +       arg_type;
> >> +       arg_type = TREE_CHAIN (arg_type))
> >> +    if (TREE_VALUE (arg_type) == void_type_node)
> >> +      continue;
> >> +    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> >> +          || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> >> +                                TREE_VALUE (arg_type), true)
> >> +          || chkp_type_has_pointer (TREE_VALUE (arg_type)))
> >> +      break;
> >> +
> >> +  /* We may use original type if there are no bounds passed.  */
> >> +  if (!arg_type)
> >> +    return orig_type;
> >> +
> >> +  type = copy_node (orig_type);
> >> +  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
> >> +
> >> +  for (arg_type = TYPE_ARG_TYPES (type);
> >> +       arg_type;
> >> +       arg_type = TREE_CHAIN (arg_type))
> >> +    {
> >> +      indexes[idx++] = new_idx++;
> >> +
> >> +      /* pass_by_reference returns 1 for void type,
> >> +      so check for it first.  */
> >> +      if (TREE_VALUE (arg_type) == void_type_node)
> >> +     continue;
> >> +      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> >> +            || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> >> +                                  TREE_VALUE (arg_type), true))
> >> +     {
> >> +       tree new_type = build_tree_list (NULL_TREE,
> >> +                                        pointer_bounds_type_node);
> >> +       TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> >> +       TREE_CHAIN (arg_type) = new_type;
> >> +
> >> +       arg_type = TREE_CHAIN (arg_type);
> >> +       new_idx++;
> >> +     }
> >> +      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
> >> +     {
> >> +       bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
> >> +       bitmap_iterator bi;
> >> +       unsigned bnd_no;
> >> +
> >> +       EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> >> +         {
> >> +           tree new_type = build_tree_list (NULL_TREE,
> >> +                                            pointer_bounds_type_node);
> >> +           TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> >> +           TREE_CHAIN (arg_type) = new_type;
> >> +
> >> +           arg_type = TREE_CHAIN (arg_type);
> >> +           new_idx++;
> >> +         }
> >> +       BITMAP_FREE (slots);
> >> +     }
> >> +    }
> >> +
> >> +  /* If function type has attribute with arg indexes then
> >> +     we have to copy it fixing attribute ops.  Map for
> >> +     fixing is in indexes array.  */
> >> +  attrs = TYPE_ATTRIBUTES (type);
> >> +  if (lookup_attribute ("nonnull", attrs)
> >> +      || lookup_attribute ("format", attrs)
> >> +      || lookup_attribute ("format_arg", attrs))
> >> +    {
> >> +      int delta = new_idx - len;
> >> +      attrs = copy_list (TYPE_ATTRIBUTES (type));
> >> +      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
> >> +      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
> >> +      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
> >> +      TYPE_ATTRIBUTES (type) = attrs;
> >> +    }
> >> +
> >> +  t = TYPE_MAIN_VARIANT (orig_type);
> >> +  if (orig_type != t)
> >> +    {
> >> +      TYPE_MAIN_VARIANT (type) = t;
> >> +      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
> >> +      TYPE_NEXT_VARIANT (t) = type;
> >> +    }
> >> +  else
> >> +    {
> >> +      TYPE_MAIN_VARIANT (type) = type;
> >> +      TYPE_NEXT_VARIANT (type) = NULL;
> >> +    }
> >> +
> >> +
> >> +  return type;
> >> +}
> >> +
> >> +/* For given function FNDECL add bounds arguments to arguments
> >> +   list.  */
> >> +
> >> +static void
> >> +chkp_add_bounds_params_to_function (tree fndecl)
> >> +{
> >> +  tree arg;
> >> +
> >> +  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
> >> +    if (BOUNDED_P (arg))
> >> +      {
> >> +     std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> >> +     tree new_arg;
> >> +
> >> +     if (DECL_NAME (arg))
> >> +       new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
> >> +     else
> >> +       {
> >> +         char uid[25];
> >> +         snprintf (uid, 25, "D.%u", DECL_UID (arg));
> >> +         new_name += uid;
> >> +       }
> >> +
> >> +     new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
> >> +                           get_identifier (new_name.c_str ()),
> >> +                           pointer_bounds_type_node);
> >> +     DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> >> +     DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
> >> +     DECL_ARTIFICIAL (new_arg) = 1;
> >> +     DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> >> +     DECL_CHAIN (arg) = new_arg;
> >> +
> >> +     arg = DECL_CHAIN (arg);
> >> +
> >> +      }
> >> +    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
> >> +      {
> >> +     tree orig_arg = arg;
> >> +     bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
> >> +     bitmap_iterator bi;
> >> +     unsigned bnd_no;
> >> +
> >> +     EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> >> +       {
> >> +         std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> >> +         tree new_arg;
> >> +         char offs[25];
> >> +
> >> +         if (DECL_NAME (orig_arg))
> >> +           new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
> >> +         else
> >> +           {
> >> +             snprintf (offs, 25, "D.%u", DECL_UID (arg));
> >> +             new_name += offs;
> >> +           }
> >> +         snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
> >> +
> >> +         new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
> >> +                               PARM_DECL,
> >> +                               get_identifier (new_name.c_str ()),
> >> +                               pointer_bounds_type_node);
> >> +         DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> >> +         DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
> >> +         DECL_ARTIFICIAL (new_arg) = 1;
> >> +         DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> >> +         DECL_CHAIN (arg) = new_arg;
> >> +
> >> +         arg = DECL_CHAIN (arg);
> >> +       }
> >> +     BITMAP_FREE (slots);
> >> +      }
> >> +
> >> +  TREE_TYPE (fndecl) =
> >> +    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
> >> +}
> >> +
> >> +/* Return clone created for instrumentation of NODE or NULL.  */
> >> +
> >> +cgraph_node *
> >> +chkp_maybe_create_clone (tree fndecl)
> >> +{
> >> +  cgraph_node *node = cgraph_node::get_create (fndecl);
> >> +  cgraph_node *clone = node->instrumented_version;
> >> +
> >> +  gcc_assert (!node->instrumentation_clone);
> >> +
> >> +  if (!clone)
> >> +    {
> >> +      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
> >> +      struct cgraph_edge *e;
> >> +      struct ipa_ref *ref;
> >> +      int i;
> >> +
> >> +      clone = node->create_version_clone (new_decl, vNULL, NULL);
> >> +      clone->externally_visible = node->externally_visible;
> >> +      clone->local = node->local;
> >> +      clone->address_taken = node->address_taken;
> >> +      clone->thunk = node->thunk;
> >> +      clone->alias = node->alias;
> >> +      clone->weakref = node->weakref;
> >> +      clone->cpp_implicit_alias = node->cpp_implicit_alias;
> >> +      clone->instrumented_version = node;
> >> +      clone->orig_decl = fndecl;
> >> +      clone->instrumentation_clone = true;
> >> +      node->instrumented_version = clone;
> >> +
> >> +      if (gimple_has_body_p (fndecl))
> >> +     {
> >> +       /* If function will not be instrumented, then it's instrumented
> >> +          version is a thunk for the original.  */
> >> +       if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
> >> +           || (flag_chkp_instrument_marked_only
> >> +               && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
> >> +         {
> >> +           clone->thunk.thunk_p = true;
> >> +           clone->thunk.add_pointer_bounds_args = true;
> >> +           clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
> >> +         }
> >> +       else
> >> +         {
> >> +           tree_function_versioning (fndecl, new_decl, NULL, false,
> >> +                                     NULL, false, NULL, NULL);
> >> +           clone->lowered = true;
> >> +         }
> >> +     }
> >> +
> >> +      /* New params are inserted after versioning because it
> >> +      actually copies args list from the original decl.  */
> >> +      chkp_add_bounds_params_to_function (new_decl);
> >> +
> >> +      /* Clones have the same comdat group as originals.  */
> >> +      if (node->same_comdat_group
> >> +       || DECL_ONE_ONLY (node->decl))
> >> +     clone->add_to_same_comdat_group (node);
> >> +
> >> +      if (gimple_has_body_p (fndecl))
> >> +     symtab->call_cgraph_insertion_hooks (clone);
> >> +
> >> +      /* Clone all aliases.  */
> >> +      for (i = 0; node->iterate_referring (i, ref); i++)
> >> +     if (ref->use == IPA_REF_ALIAS)
> >> +       {
> >> +         struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
> >> +         struct cgraph_node *chkp_alias
> >> +           = chkp_maybe_create_clone (alias->decl);
> >> +         chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
> >> +       }
> >> +
> >> +      /* Clone all thunks.  */
> >> +      for (e = node->callers; e; e = e->next_caller)
> >> +     if (e->caller->thunk.thunk_p)
> >> +       {
> >> +         struct cgraph_node *thunk
> >> +           = chkp_maybe_create_clone (e->caller->decl);
> >> +         /* Redirect thunk clone edge to the node clone.  */
> >> +         thunk->callees->redirect_callee (clone);
> >> +       }
> >> +
> >> +      /* For aliases and thunks we should make sure target is cloned
> >> +      to have proper references and edges.  */
> >> +      if (node->thunk.thunk_p)
> >> +     chkp_maybe_create_clone (node->callees->callee->decl);
> >> +      else if (node->alias)
> >> +     {
> >> +       struct cgraph_node *target;
> >> +
> >> +       ref = node->ref_list.first_reference ();
> >> +       if (ref)
> >> +         chkp_maybe_create_clone (ref->referred->decl);
> >> +
> >> +       if (node->alias_target)
> >> +         {
> >> +           if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
> >> +             {
> >> +               target = chkp_maybe_create_clone (node->alias_target);
> >> +               clone->alias_target = target->decl;
> >> +             }
> >> +           else
> >> +             clone->alias_target = node->alias_target;
> >> +         }
> >> +     }
> >> +
> >> +      /* Add IPA reference.  It's main role is to keep instrumented
> >> +      version reachable while original node is reachable.  */
> >> +      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
> >> +    }
> >> +
> >> +  return clone;
> >> +}
> >> +
> >> +/* Create clone for all functions to be instrumented.  */
> >> +
> >> +static unsigned int
> >> +chkp_versioning (void)
> >> +{
> >> +  struct cgraph_node *node;
> >> +
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && !node->instrumented_version
> >> +       && !node->alias
> >> +       && !node->thunk.thunk_p
> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
> >> +       && (!flag_chkp_instrument_marked_only
> >> +           || lookup_attribute ("bnd_instrument",
> >> +                                DECL_ATTRIBUTES (node->decl)))
> >> +       /* No builtins instrumentation for now.  */
> >> +       && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
> >> +     chkp_maybe_create_clone (node->decl);
> >> +    }
> >> +
> >> +  /* Mark all aliases and thunks of functions with no instrumented
> >> +     version as legacy function.  */
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && !node->instrumented_version
> >> +       && (node->alias || node->thunk.thunk_p)
> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
> >> +     DECL_ATTRIBUTES (node->decl)
> >> +       = tree_cons (get_identifier ("bnd_legacy"), NULL,
> >> +                    DECL_ATTRIBUTES (node->decl));
> >> +    }
> >> +
> >> +  return 0;
> >> +}
> >> +
> >> +/* In this pass we remove bodies of functions having
> >> +   instrumented version.  Functions with removed bodies
> >> +   become a special kind of thunks to provide a connection
> >> +   between calls to the original version and instrumented
> >> +   function.  */
> >> +
> >> +static unsigned int
> >> +chkp_produce_thunks (void)
> >> +{
> >> +  struct cgraph_node *node;
> >> +
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && node->instrumented_version
> >> +       && gimple_has_body_p (node->decl)
> >> +       && gimple_has_body_p (node->instrumented_version->decl))
> >> +     {
> >> +       node->release_body ();
> >> +       node->remove_callees ();
> >> +       node->remove_all_references ();
> >> +
> >> +       node->thunk.thunk_p = true;
> >> +       node->thunk.add_pointer_bounds_args = true;
> >> +       node->create_edge (node->instrumented_version, NULL,
> >> +                          0, CGRAPH_FREQ_BASE);
> >> +       node->create_reference (node->instrumented_version,
> >> +                            IPA_REF_CHKP, NULL);
> >> +     }
> >> +    }
> >> +
> >> +  /* Mark instrumentation clones created for aliases and thunks
> >> +     as insttrumented so they could be removed as unreachable
> >> +     now.  */
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (node->instrumentation_clone
> >> +       && (node->alias || node->thunk.thunk_p)
> >> +       && !chkp_function_instrumented_p (node->decl))
> >> +     chkp_function_mark_instrumented (node->decl);
> >> +    }
> >> +
> >> +  symtab->remove_unreachable_nodes (true, dump_file);
> >> +
> >> +  return 0;
> >> +}
> >> +
> >> +const pass_data pass_data_ipa_chkp_versioning =
> >> +{
> >> +  SIMPLE_IPA_PASS, /* type */
> >> +  "chkp_versioning", /* name */
> >> +  OPTGROUP_NONE, /* optinfo_flags */
> >> +  TV_NONE, /* tv_id */
> >> +  0, /* properties_required */
> >> +  0, /* properties_provided */
> >> +  0, /* properties_destroyed */
> >> +  0, /* todo_flags_start */
> >> +  0 /* todo_flags_finish */
> >> +};
> >> +
> >> +const pass_data pass_data_ipa_chkp_produce_thunks =
> >> +{
> >> +  SIMPLE_IPA_PASS, /* type */
> >> +  "chkp_cleanup", /* name */
> >> +  OPTGROUP_NONE, /* optinfo_flags */
> >> +  TV_NONE, /* tv_id */
> >> +  0, /* properties_required */
> >> +  0, /* properties_provided */
> >> +  0, /* properties_destroyed */
> >> +  0, /* todo_flags_start */
> >> +  0 /* todo_flags_finish */
> >> +};
> >> +
> >> +class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
> >> +{
> >> +public:
> >> +  pass_ipa_chkp_versioning (gcc::context *ctxt)
> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
> >> +  {}
> >> +
> >> +  /* opt_pass methods: */
> >> +  virtual opt_pass * clone ()
> >> +    {
> >> +      return new pass_ipa_chkp_versioning (m_ctxt);
> >> +    }
> >> +
> >> +  virtual bool gate (function *)
> >> +    {
> >> +      return flag_check_pointer_bounds;
> >> +    }
> >> +
> >> +  virtual unsigned int execute (function *)
> >> +    {
> >> +      return chkp_versioning ();
> >> +    }
> >> +
> >> +}; // class pass_ipa_chkp_versioning
> >> +
> >> +class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
> >> +{
> >> +public:
> >> +  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
> >> +  {}
> >> +
> >> +  /* opt_pass methods: */
> >> +  virtual opt_pass * clone ()
> >> +    {
> >> +      return new pass_ipa_chkp_produce_thunks (m_ctxt);
> >> +    }
> >> +
> >> +  virtual unsigned int execute (function *)
> >> +    {
> >> +      return chkp_produce_thunks ();
> >> +    }
> >> +
> >> +}; // class pass_chkp_produce_thunks
> >> +
> >> +simple_ipa_opt_pass *
> >> +make_pass_ipa_chkp_versioning (gcc::context *ctxt)
> >> +{
> >> +  return new pass_ipa_chkp_versioning (ctxt);
> >> +}
> >> +
> >> +simple_ipa_opt_pass *
> >> +make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> >> +{
> >> +  return new pass_ipa_chkp_produce_thunks (ctxt);
> >> +}
> >> diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
> >> new file mode 100644
> >> index 0000000..9c92c04
> >> --- /dev/null
> >> +++ b/gcc/ipa-chkp.h
> >> @@ -0,0 +1,29 @@
> >> +/* Declaration of interface functions of Pointer Bounds Checker.
> >> +   Copyright (C) 2014 Free Software Foundation, Inc.
> >> +
> >> +This file is part of GCC.
> >> +
> >> +GCC is free software; you can redistribute it and/or modify it under
> >> +the terms of the GNU General Public License as published by the Free
> >> +Software Foundation; either version 3, or (at your option) any later
> >> +version.
> >> +
> >> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
> >> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
> >> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
> >> +for more details.
> >> +
> >> +You should have received a copy of the GNU General Public License
> >> +along with GCC; see the file COPYING3.  If not see
> >> +<http://www.gnu.org/licenses/>.  */
> >> +
> >> +#ifndef GCC_IPA_CHKP_H
> >> +#define GCC_IPA_CHKP_H
> >> +
> >> +#include "tree.h"
> >> +#include "cgraph.h"
> >> +
> >> +extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
> >> +extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
> >> +
> >> +#endif /* GCC_IPA_CHKP_H */
Jan Hubicka Oct. 8, 2014, 10:40 p.m. UTC | #4
> 
> I prevent clone's body from removal and therefore original should call
> clone (otherwise clone may have no callers and be removed).  I think
> call edge may work (need to recall other cases when reference do its
> work) but is it OK to have call with no stmt for non thunk nodes?

OK, I guess I understand what makes your situation different from what we have
already.  Basically you have two functions with bodies already produced (by
early optimization) but you want to keep body of the second around
to later turn the first into thunk.  But I do not see why this can not happen early?

I suppose we could either go with fake call edge or the new reference you have.
I basically wonder what would be least intrussive (and easiest to maintain) way
for doing so - adding special kinds of functions always produce extra maintenance
overhead in long term and we already have very many of them...
> 
> >
> >> +    instrumented function in the code but it still should be counted
> >> +    as reachable if the original function is reachable.
> >> +
> >> +    When original function bodies are not needed anymore we release
> >> +    them and transform functions into a special kind of thunks.  Each
> >> +    thunk has a call edge to the instrumented version.  These thunks
> >> +    help to keep externally visible instrumented functions visible
> >> +    when linker reolution files are used.  Linker has no info about
> >> +    connection between original and instrumented function and
> >> +    therefore we may wrongly decide (due to difference in assember
> >> +    names) that instrumented function version is local and can be
> >> +    removed.  */
> >
> > Can you, please, give me an example what happens here?
> > Also how this relates to LTO?
> 
> When we compile with LTO linker provides a resolution file with
> functions usage information.  Consider I have external function test
> and its clone test.chkp.  Linker says I have calls to test and no
> calls to test.chkp.  Compiler just removes test.chkp then.  It doesn't
> happen when test is trunsformed into a thunk of test.chkp.

Does the bounds checking work across translation unit boundaries somehow?
Would it make sense to do the redirection and instrumentation as an IPA pass?

Honza
> 
> Ilya
> 
> >
> > Honza
> >> +
> >> +#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
> >> +
> >> +/* Build a clone of FNDECL with a modified name.  */
> >> +
> >> +static tree
> >> +chkp_build_instrumented_fndecl (tree fndecl)
> >> +{
> >> +  tree new_decl = copy_node (fndecl);
> >> +  tree new_name;
> >> +  std::string s;
> >> +
> >> +  /* We want called_as_built_in recall instrumented calls
> >> +     to instrumented built-in functions.  Therefore use
> >> +     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
> >> +  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
> >> +  s += ".chkp";
> >> +  DECL_NAME (new_decl) = get_identifier (s.c_str ());
> >> +
> >> +  /* References to the original and to the instrumented version
> >> +     should look the same in the output assembly.  And we cannot
> >> +     use the same assembler name for the instrumented version
> >> +     because it conflicts with decl merging algorithms in LTO.
> >> +     Achieve the result by using transparent alias name for the
> >> +     instrumented version.  */
> >> +  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
> >> +  s += ".chkp";
> >> +  new_name = get_identifier (s.c_str ());
> >> +  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
> >> +  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
> >> +  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
> >> +
> >> +  /* For functions with body versioning will make a copy of arguments.
> >> +     For functions with no body we need to do it here.  */
> >> +  if (!gimple_has_body_p (fndecl))
> >> +    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
> >> +
> >> +  /* We are going to modify attributes list and therefore should
> >> +     make own copy.  */
> >> +  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
> >> +
> >> +  return new_decl;
> >> +}
> >> +
> >> +
> >> +/* Fix operands of attribute from ATTRS list named ATTR_NAME.
> >> +   Integer operands are replaced with values according to
> >> +   INDEXES map having LEN elements.  For operands out of len
> >> +   we just add DELTA.  */
> >> +
> >> +static void
> >> +chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
> >> +                        unsigned *indexes, int len, int delta)
> >> +{
> >> +  tree attr = lookup_attribute (attr_name, attrs);
> >> +  tree op;
> >> +
> >> +  if (!attr)
> >> +    return;
> >> +
> >> +  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
> >> +  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
> >> +    {
> >> +      int idx;
> >> +
> >> +      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
> >> +     continue;
> >> +
> >> +      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
> >> +
> >> +      /* If idx exceeds indexes length then we just
> >> +      keep it at the same distance from the last
> >> +      known arg.  */
> >> +      if (idx > len)
> >> +     idx += delta;
> >> +      else
> >> +     idx = indexes[idx - 1] + 1;
> >> +      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
> >> +    }
> >> +}
> >> +
> >> +/* Make a copy of function type ORIG_TYPE adding pointer
> >> +   bounds as additional arguments.  */
> >> +
> >> +tree
> >> +chkp_copy_function_type_adding_bounds (tree orig_type)
> >> +{
> >> +  tree type;
> >> +  tree arg_type, attrs, t;
> >> +  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
> >> +  unsigned *indexes = XALLOCAVEC (unsigned, len);
> >> +  unsigned idx = 0, new_idx = 0;
> >> +
> >> +  for (arg_type = TYPE_ARG_TYPES (orig_type);
> >> +       arg_type;
> >> +       arg_type = TREE_CHAIN (arg_type))
> >> +    if (TREE_VALUE (arg_type) == void_type_node)
> >> +      continue;
> >> +    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> >> +          || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> >> +                                TREE_VALUE (arg_type), true)
> >> +          || chkp_type_has_pointer (TREE_VALUE (arg_type)))
> >> +      break;
> >> +
> >> +  /* We may use original type if there are no bounds passed.  */
> >> +  if (!arg_type)
> >> +    return orig_type;
> >> +
> >> +  type = copy_node (orig_type);
> >> +  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
> >> +
> >> +  for (arg_type = TYPE_ARG_TYPES (type);
> >> +       arg_type;
> >> +       arg_type = TREE_CHAIN (arg_type))
> >> +    {
> >> +      indexes[idx++] = new_idx++;
> >> +
> >> +      /* pass_by_reference returns 1 for void type,
> >> +      so check for it first.  */
> >> +      if (TREE_VALUE (arg_type) == void_type_node)
> >> +     continue;
> >> +      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
> >> +            || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
> >> +                                  TREE_VALUE (arg_type), true))
> >> +     {
> >> +       tree new_type = build_tree_list (NULL_TREE,
> >> +                                        pointer_bounds_type_node);
> >> +       TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> >> +       TREE_CHAIN (arg_type) = new_type;
> >> +
> >> +       arg_type = TREE_CHAIN (arg_type);
> >> +       new_idx++;
> >> +     }
> >> +      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
> >> +     {
> >> +       bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
> >> +       bitmap_iterator bi;
> >> +       unsigned bnd_no;
> >> +
> >> +       EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> >> +         {
> >> +           tree new_type = build_tree_list (NULL_TREE,
> >> +                                            pointer_bounds_type_node);
> >> +           TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
> >> +           TREE_CHAIN (arg_type) = new_type;
> >> +
> >> +           arg_type = TREE_CHAIN (arg_type);
> >> +           new_idx++;
> >> +         }
> >> +       BITMAP_FREE (slots);
> >> +     }
> >> +    }
> >> +
> >> +  /* If function type has attribute with arg indexes then
> >> +     we have to copy it fixing attribute ops.  Map for
> >> +     fixing is in indexes array.  */
> >> +  attrs = TYPE_ATTRIBUTES (type);
> >> +  if (lookup_attribute ("nonnull", attrs)
> >> +      || lookup_attribute ("format", attrs)
> >> +      || lookup_attribute ("format_arg", attrs))
> >> +    {
> >> +      int delta = new_idx - len;
> >> +      attrs = copy_list (TYPE_ATTRIBUTES (type));
> >> +      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
> >> +      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
> >> +      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
> >> +      TYPE_ATTRIBUTES (type) = attrs;
> >> +    }
> >> +
> >> +  t = TYPE_MAIN_VARIANT (orig_type);
> >> +  if (orig_type != t)
> >> +    {
> >> +      TYPE_MAIN_VARIANT (type) = t;
> >> +      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
> >> +      TYPE_NEXT_VARIANT (t) = type;
> >> +    }
> >> +  else
> >> +    {
> >> +      TYPE_MAIN_VARIANT (type) = type;
> >> +      TYPE_NEXT_VARIANT (type) = NULL;
> >> +    }
> >> +
> >> +
> >> +  return type;
> >> +}
> >> +
> >> +/* For given function FNDECL add bounds arguments to arguments
> >> +   list.  */
> >> +
> >> +static void
> >> +chkp_add_bounds_params_to_function (tree fndecl)
> >> +{
> >> +  tree arg;
> >> +
> >> +  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
> >> +    if (BOUNDED_P (arg))
> >> +      {
> >> +     std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> >> +     tree new_arg;
> >> +
> >> +     if (DECL_NAME (arg))
> >> +       new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
> >> +     else
> >> +       {
> >> +         char uid[25];
> >> +         snprintf (uid, 25, "D.%u", DECL_UID (arg));
> >> +         new_name += uid;
> >> +       }
> >> +
> >> +     new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
> >> +                           get_identifier (new_name.c_str ()),
> >> +                           pointer_bounds_type_node);
> >> +     DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> >> +     DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
> >> +     DECL_ARTIFICIAL (new_arg) = 1;
> >> +     DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> >> +     DECL_CHAIN (arg) = new_arg;
> >> +
> >> +     arg = DECL_CHAIN (arg);
> >> +
> >> +      }
> >> +    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
> >> +      {
> >> +     tree orig_arg = arg;
> >> +     bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
> >> +     bitmap_iterator bi;
> >> +     unsigned bnd_no;
> >> +
> >> +     EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
> >> +       {
> >> +         std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
> >> +         tree new_arg;
> >> +         char offs[25];
> >> +
> >> +         if (DECL_NAME (orig_arg))
> >> +           new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
> >> +         else
> >> +           {
> >> +             snprintf (offs, 25, "D.%u", DECL_UID (arg));
> >> +             new_name += offs;
> >> +           }
> >> +         snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
> >> +
> >> +         new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
> >> +                               PARM_DECL,
> >> +                               get_identifier (new_name.c_str ()),
> >> +                               pointer_bounds_type_node);
> >> +         DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
> >> +         DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
> >> +         DECL_ARTIFICIAL (new_arg) = 1;
> >> +         DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
> >> +         DECL_CHAIN (arg) = new_arg;
> >> +
> >> +         arg = DECL_CHAIN (arg);
> >> +       }
> >> +     BITMAP_FREE (slots);
> >> +      }
> >> +
> >> +  TREE_TYPE (fndecl) =
> >> +    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
> >> +}
> >> +
> >> +/* Return clone created for instrumentation of NODE or NULL.  */
> >> +
> >> +cgraph_node *
> >> +chkp_maybe_create_clone (tree fndecl)
> >> +{
> >> +  cgraph_node *node = cgraph_node::get_create (fndecl);
> >> +  cgraph_node *clone = node->instrumented_version;
> >> +
> >> +  gcc_assert (!node->instrumentation_clone);
> >> +
> >> +  if (!clone)
> >> +    {
> >> +      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
> >> +      struct cgraph_edge *e;
> >> +      struct ipa_ref *ref;
> >> +      int i;
> >> +
> >> +      clone = node->create_version_clone (new_decl, vNULL, NULL);
> >> +      clone->externally_visible = node->externally_visible;
> >> +      clone->local = node->local;
> >> +      clone->address_taken = node->address_taken;
> >> +      clone->thunk = node->thunk;
> >> +      clone->alias = node->alias;
> >> +      clone->weakref = node->weakref;
> >> +      clone->cpp_implicit_alias = node->cpp_implicit_alias;
> >> +      clone->instrumented_version = node;
> >> +      clone->orig_decl = fndecl;
> >> +      clone->instrumentation_clone = true;
> >> +      node->instrumented_version = clone;
> >> +
> >> +      if (gimple_has_body_p (fndecl))
> >> +     {
> >> +       /* If function will not be instrumented, then it's instrumented
> >> +          version is a thunk for the original.  */
> >> +       if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
> >> +           || (flag_chkp_instrument_marked_only
> >> +               && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
> >> +         {
> >> +           clone->thunk.thunk_p = true;
> >> +           clone->thunk.add_pointer_bounds_args = true;
> >> +           clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
> >> +         }
> >> +       else
> >> +         {
> >> +           tree_function_versioning (fndecl, new_decl, NULL, false,
> >> +                                     NULL, false, NULL, NULL);
> >> +           clone->lowered = true;
> >> +         }
> >> +     }
> >> +
> >> +      /* New params are inserted after versioning because it
> >> +      actually copies args list from the original decl.  */
> >> +      chkp_add_bounds_params_to_function (new_decl);
> >> +
> >> +      /* Clones have the same comdat group as originals.  */
> >> +      if (node->same_comdat_group
> >> +       || DECL_ONE_ONLY (node->decl))
> >> +     clone->add_to_same_comdat_group (node);
> >> +
> >> +      if (gimple_has_body_p (fndecl))
> >> +     symtab->call_cgraph_insertion_hooks (clone);
> >> +
> >> +      /* Clone all aliases.  */
> >> +      for (i = 0; node->iterate_referring (i, ref); i++)
> >> +     if (ref->use == IPA_REF_ALIAS)
> >> +       {
> >> +         struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
> >> +         struct cgraph_node *chkp_alias
> >> +           = chkp_maybe_create_clone (alias->decl);
> >> +         chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
> >> +       }
> >> +
> >> +      /* Clone all thunks.  */
> >> +      for (e = node->callers; e; e = e->next_caller)
> >> +     if (e->caller->thunk.thunk_p)
> >> +       {
> >> +         struct cgraph_node *thunk
> >> +           = chkp_maybe_create_clone (e->caller->decl);
> >> +         /* Redirect thunk clone edge to the node clone.  */
> >> +         thunk->callees->redirect_callee (clone);
> >> +       }
> >> +
> >> +      /* For aliases and thunks we should make sure target is cloned
> >> +      to have proper references and edges.  */
> >> +      if (node->thunk.thunk_p)
> >> +     chkp_maybe_create_clone (node->callees->callee->decl);
> >> +      else if (node->alias)
> >> +     {
> >> +       struct cgraph_node *target;
> >> +
> >> +       ref = node->ref_list.first_reference ();
> >> +       if (ref)
> >> +         chkp_maybe_create_clone (ref->referred->decl);
> >> +
> >> +       if (node->alias_target)
> >> +         {
> >> +           if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
> >> +             {
> >> +               target = chkp_maybe_create_clone (node->alias_target);
> >> +               clone->alias_target = target->decl;
> >> +             }
> >> +           else
> >> +             clone->alias_target = node->alias_target;
> >> +         }
> >> +     }
> >> +
> >> +      /* Add IPA reference.  It's main role is to keep instrumented
> >> +      version reachable while original node is reachable.  */
> >> +      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
> >> +    }
> >> +
> >> +  return clone;
> >> +}
> >> +
> >> +/* Create clone for all functions to be instrumented.  */
> >> +
> >> +static unsigned int
> >> +chkp_versioning (void)
> >> +{
> >> +  struct cgraph_node *node;
> >> +
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && !node->instrumented_version
> >> +       && !node->alias
> >> +       && !node->thunk.thunk_p
> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
> >> +       && (!flag_chkp_instrument_marked_only
> >> +           || lookup_attribute ("bnd_instrument",
> >> +                                DECL_ATTRIBUTES (node->decl)))
> >> +       /* No builtins instrumentation for now.  */
> >> +       && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
> >> +     chkp_maybe_create_clone (node->decl);
> >> +    }
> >> +
> >> +  /* Mark all aliases and thunks of functions with no instrumented
> >> +     version as legacy function.  */
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && !node->instrumented_version
> >> +       && (node->alias || node->thunk.thunk_p)
> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
> >> +     DECL_ATTRIBUTES (node->decl)
> >> +       = tree_cons (get_identifier ("bnd_legacy"), NULL,
> >> +                    DECL_ATTRIBUTES (node->decl));
> >> +    }
> >> +
> >> +  return 0;
> >> +}
> >> +
> >> +/* In this pass we remove bodies of functions having
> >> +   instrumented version.  Functions with removed bodies
> >> +   become a special kind of thunks to provide a connection
> >> +   between calls to the original version and instrumented
> >> +   function.  */
> >> +
> >> +static unsigned int
> >> +chkp_produce_thunks (void)
> >> +{
> >> +  struct cgraph_node *node;
> >> +
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (!node->instrumentation_clone
> >> +       && node->instrumented_version
> >> +       && gimple_has_body_p (node->decl)
> >> +       && gimple_has_body_p (node->instrumented_version->decl))
> >> +     {
> >> +       node->release_body ();
> >> +       node->remove_callees ();
> >> +       node->remove_all_references ();
> >> +
> >> +       node->thunk.thunk_p = true;
> >> +       node->thunk.add_pointer_bounds_args = true;
> >> +       node->create_edge (node->instrumented_version, NULL,
> >> +                          0, CGRAPH_FREQ_BASE);
> >> +       node->create_reference (node->instrumented_version,
> >> +                            IPA_REF_CHKP, NULL);
> >> +     }
> >> +    }
> >> +
> >> +  /* Mark instrumentation clones created for aliases and thunks
> >> +     as insttrumented so they could be removed as unreachable
> >> +     now.  */
> >> +  FOR_EACH_DEFINED_FUNCTION (node)
> >> +    {
> >> +      if (node->instrumentation_clone
> >> +       && (node->alias || node->thunk.thunk_p)
> >> +       && !chkp_function_instrumented_p (node->decl))
> >> +     chkp_function_mark_instrumented (node->decl);
> >> +    }
> >> +
> >> +  symtab->remove_unreachable_nodes (true, dump_file);
> >> +
> >> +  return 0;
> >> +}
> >> +
> >> +const pass_data pass_data_ipa_chkp_versioning =
> >> +{
> >> +  SIMPLE_IPA_PASS, /* type */
> >> +  "chkp_versioning", /* name */
> >> +  OPTGROUP_NONE, /* optinfo_flags */
> >> +  TV_NONE, /* tv_id */
> >> +  0, /* properties_required */
> >> +  0, /* properties_provided */
> >> +  0, /* properties_destroyed */
> >> +  0, /* todo_flags_start */
> >> +  0 /* todo_flags_finish */
> >> +};
> >> +
> >> +const pass_data pass_data_ipa_chkp_produce_thunks =
> >> +{
> >> +  SIMPLE_IPA_PASS, /* type */
> >> +  "chkp_cleanup", /* name */
> >> +  OPTGROUP_NONE, /* optinfo_flags */
> >> +  TV_NONE, /* tv_id */
> >> +  0, /* properties_required */
> >> +  0, /* properties_provided */
> >> +  0, /* properties_destroyed */
> >> +  0, /* todo_flags_start */
> >> +  0 /* todo_flags_finish */
> >> +};
> >> +
> >> +class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
> >> +{
> >> +public:
> >> +  pass_ipa_chkp_versioning (gcc::context *ctxt)
> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
> >> +  {}
> >> +
> >> +  /* opt_pass methods: */
> >> +  virtual opt_pass * clone ()
> >> +    {
> >> +      return new pass_ipa_chkp_versioning (m_ctxt);
> >> +    }
> >> +
> >> +  virtual bool gate (function *)
> >> +    {
> >> +      return flag_check_pointer_bounds;
> >> +    }
> >> +
> >> +  virtual unsigned int execute (function *)
> >> +    {
> >> +      return chkp_versioning ();
> >> +    }
> >> +
> >> +}; // class pass_ipa_chkp_versioning
> >> +
> >> +class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
> >> +{
> >> +public:
> >> +  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
> >> +  {}
> >> +
> >> +  /* opt_pass methods: */
> >> +  virtual opt_pass * clone ()
> >> +    {
> >> +      return new pass_ipa_chkp_produce_thunks (m_ctxt);
> >> +    }
> >> +
> >> +  virtual unsigned int execute (function *)
> >> +    {
> >> +      return chkp_produce_thunks ();
> >> +    }
> >> +
> >> +}; // class pass_chkp_produce_thunks
> >> +
> >> +simple_ipa_opt_pass *
> >> +make_pass_ipa_chkp_versioning (gcc::context *ctxt)
> >> +{
> >> +  return new pass_ipa_chkp_versioning (ctxt);
> >> +}
> >> +
> >> +simple_ipa_opt_pass *
> >> +make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
> >> +{
> >> +  return new pass_ipa_chkp_produce_thunks (ctxt);
> >> +}
> >> diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
> >> new file mode 100644
> >> index 0000000..9c92c04
> >> --- /dev/null
> >> +++ b/gcc/ipa-chkp.h
> >> @@ -0,0 +1,29 @@
> >> +/* Declaration of interface functions of Pointer Bounds Checker.
> >> +   Copyright (C) 2014 Free Software Foundation, Inc.
> >> +
> >> +This file is part of GCC.
> >> +
> >> +GCC is free software; you can redistribute it and/or modify it under
> >> +the terms of the GNU General Public License as published by the Free
> >> +Software Foundation; either version 3, or (at your option) any later
> >> +version.
> >> +
> >> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
> >> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
> >> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
> >> +for more details.
> >> +
> >> +You should have received a copy of the GNU General Public License
> >> +along with GCC; see the file COPYING3.  If not see
> >> +<http://www.gnu.org/licenses/>.  */
> >> +
> >> +#ifndef GCC_IPA_CHKP_H
> >> +#define GCC_IPA_CHKP_H
> >> +
> >> +#include "tree.h"
> >> +#include "cgraph.h"
> >> +
> >> +extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
> >> +extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
> >> +
> >> +#endif /* GCC_IPA_CHKP_H */
Ilya Enkovich Oct. 9, 2014, 7:46 a.m. UTC | #5
2014-10-09 2:40 GMT+04:00 Jan Hubicka <hubicka@ucw.cz>:
>>
>> I prevent clone's body from removal and therefore original should call
>> clone (otherwise clone may have no callers and be removed).  I think
>> call edge may work (need to recall other cases when reference do its
>> work) but is it OK to have call with no stmt for non thunk nodes?
>
> OK, I guess I understand what makes your situation different from what we have
> already.  Basically you have two functions with bodies already produced (by
> early optimization) but you want to keep body of the second around
> to later turn the first into thunk.  But I do not see why this can not happen early?

Clones are created not only for functions with bodies.  Aliases and
thunks may also have clones.  Having call edges for aliases and more
than one edge for thunks seemed more intrusive to me than having a new
reference.  Regarding early transformation into thunks - I need to try
it.  It sounds reasonable and might work.

>
> I suppose we could either go with fake call edge or the new reference you have.
> I basically wonder what would be least intrussive (and easiest to maintain) way
> for doing so - adding special kinds of functions always produce extra maintenance
> overhead in long term and we already have very many of them...
>>
>> >
>> >> +    instrumented function in the code but it still should be counted
>> >> +    as reachable if the original function is reachable.
>> >> +
>> >> +    When original function bodies are not needed anymore we release
>> >> +    them and transform functions into a special kind of thunks.  Each
>> >> +    thunk has a call edge to the instrumented version.  These thunks
>> >> +    help to keep externally visible instrumented functions visible
>> >> +    when linker reolution files are used.  Linker has no info about
>> >> +    connection between original and instrumented function and
>> >> +    therefore we may wrongly decide (due to difference in assember
>> >> +    names) that instrumented function version is local and can be
>> >> +    removed.  */
>> >
>> > Can you, please, give me an example what happens here?
>> > Also how this relates to LTO?
>>
>> When we compile with LTO linker provides a resolution file with
>> functions usage information.  Consider I have external function test
>> and its clone test.chkp.  Linker says I have calls to test and no
>> calls to test.chkp.  Compiler just removes test.chkp then.  It doesn't
>> happen when test is trunsformed into a thunk of test.chkp.
>
> Does the bounds checking work across translation unit boundaries somehow?
> Would it make sense to do the redirection and instrumentation as an IPA pass?

Instrumentation is function local and I don't see benefit from making
it an IPA pass.

Ilya

>
> Honza
>>
>> Ilya
>>
>> >
>> > Honza
>> >> +
>> >> +#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
>> >> +
>> >> +/* Build a clone of FNDECL with a modified name.  */
>> >> +
>> >> +static tree
>> >> +chkp_build_instrumented_fndecl (tree fndecl)
>> >> +{
>> >> +  tree new_decl = copy_node (fndecl);
>> >> +  tree new_name;
>> >> +  std::string s;
>> >> +
>> >> +  /* We want called_as_built_in recall instrumented calls
>> >> +     to instrumented built-in functions.  Therefore use
>> >> +     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
>> >> +  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
>> >> +  s += ".chkp";
>> >> +  DECL_NAME (new_decl) = get_identifier (s.c_str ());
>> >> +
>> >> +  /* References to the original and to the instrumented version
>> >> +     should look the same in the output assembly.  And we cannot
>> >> +     use the same assembler name for the instrumented version
>> >> +     because it conflicts with decl merging algorithms in LTO.
>> >> +     Achieve the result by using transparent alias name for the
>> >> +     instrumented version.  */
>> >> +  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
>> >> +  s += ".chkp";
>> >> +  new_name = get_identifier (s.c_str ());
>> >> +  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
>> >> +  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
>> >> +  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
>> >> +
>> >> +  /* For functions with body versioning will make a copy of arguments.
>> >> +     For functions with no body we need to do it here.  */
>> >> +  if (!gimple_has_body_p (fndecl))
>> >> +    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
>> >> +
>> >> +  /* We are going to modify attributes list and therefore should
>> >> +     make own copy.  */
>> >> +  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
>> >> +
>> >> +  return new_decl;
>> >> +}
>> >> +
>> >> +
>> >> +/* Fix operands of attribute from ATTRS list named ATTR_NAME.
>> >> +   Integer operands are replaced with values according to
>> >> +   INDEXES map having LEN elements.  For operands out of len
>> >> +   we just add DELTA.  */
>> >> +
>> >> +static void
>> >> +chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
>> >> +                        unsigned *indexes, int len, int delta)
>> >> +{
>> >> +  tree attr = lookup_attribute (attr_name, attrs);
>> >> +  tree op;
>> >> +
>> >> +  if (!attr)
>> >> +    return;
>> >> +
>> >> +  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
>> >> +  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
>> >> +    {
>> >> +      int idx;
>> >> +
>> >> +      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
>> >> +     continue;
>> >> +
>> >> +      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
>> >> +
>> >> +      /* If idx exceeds indexes length then we just
>> >> +      keep it at the same distance from the last
>> >> +      known arg.  */
>> >> +      if (idx > len)
>> >> +     idx += delta;
>> >> +      else
>> >> +     idx = indexes[idx - 1] + 1;
>> >> +      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
>> >> +    }
>> >> +}
>> >> +
>> >> +/* Make a copy of function type ORIG_TYPE adding pointer
>> >> +   bounds as additional arguments.  */
>> >> +
>> >> +tree
>> >> +chkp_copy_function_type_adding_bounds (tree orig_type)
>> >> +{
>> >> +  tree type;
>> >> +  tree arg_type, attrs, t;
>> >> +  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
>> >> +  unsigned *indexes = XALLOCAVEC (unsigned, len);
>> >> +  unsigned idx = 0, new_idx = 0;
>> >> +
>> >> +  for (arg_type = TYPE_ARG_TYPES (orig_type);
>> >> +       arg_type;
>> >> +       arg_type = TREE_CHAIN (arg_type))
>> >> +    if (TREE_VALUE (arg_type) == void_type_node)
>> >> +      continue;
>> >> +    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
>> >> +          || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
>> >> +                                TREE_VALUE (arg_type), true)
>> >> +          || chkp_type_has_pointer (TREE_VALUE (arg_type)))
>> >> +      break;
>> >> +
>> >> +  /* We may use original type if there are no bounds passed.  */
>> >> +  if (!arg_type)
>> >> +    return orig_type;
>> >> +
>> >> +  type = copy_node (orig_type);
>> >> +  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
>> >> +
>> >> +  for (arg_type = TYPE_ARG_TYPES (type);
>> >> +       arg_type;
>> >> +       arg_type = TREE_CHAIN (arg_type))
>> >> +    {
>> >> +      indexes[idx++] = new_idx++;
>> >> +
>> >> +      /* pass_by_reference returns 1 for void type,
>> >> +      so check for it first.  */
>> >> +      if (TREE_VALUE (arg_type) == void_type_node)
>> >> +     continue;
>> >> +      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
>> >> +            || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
>> >> +                                  TREE_VALUE (arg_type), true))
>> >> +     {
>> >> +       tree new_type = build_tree_list (NULL_TREE,
>> >> +                                        pointer_bounds_type_node);
>> >> +       TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
>> >> +       TREE_CHAIN (arg_type) = new_type;
>> >> +
>> >> +       arg_type = TREE_CHAIN (arg_type);
>> >> +       new_idx++;
>> >> +     }
>> >> +      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
>> >> +     {
>> >> +       bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
>> >> +       bitmap_iterator bi;
>> >> +       unsigned bnd_no;
>> >> +
>> >> +       EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
>> >> +         {
>> >> +           tree new_type = build_tree_list (NULL_TREE,
>> >> +                                            pointer_bounds_type_node);
>> >> +           TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
>> >> +           TREE_CHAIN (arg_type) = new_type;
>> >> +
>> >> +           arg_type = TREE_CHAIN (arg_type);
>> >> +           new_idx++;
>> >> +         }
>> >> +       BITMAP_FREE (slots);
>> >> +     }
>> >> +    }
>> >> +
>> >> +  /* If function type has attribute with arg indexes then
>> >> +     we have to copy it fixing attribute ops.  Map for
>> >> +     fixing is in indexes array.  */
>> >> +  attrs = TYPE_ATTRIBUTES (type);
>> >> +  if (lookup_attribute ("nonnull", attrs)
>> >> +      || lookup_attribute ("format", attrs)
>> >> +      || lookup_attribute ("format_arg", attrs))
>> >> +    {
>> >> +      int delta = new_idx - len;
>> >> +      attrs = copy_list (TYPE_ATTRIBUTES (type));
>> >> +      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
>> >> +      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
>> >> +      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
>> >> +      TYPE_ATTRIBUTES (type) = attrs;
>> >> +    }
>> >> +
>> >> +  t = TYPE_MAIN_VARIANT (orig_type);
>> >> +  if (orig_type != t)
>> >> +    {
>> >> +      TYPE_MAIN_VARIANT (type) = t;
>> >> +      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
>> >> +      TYPE_NEXT_VARIANT (t) = type;
>> >> +    }
>> >> +  else
>> >> +    {
>> >> +      TYPE_MAIN_VARIANT (type) = type;
>> >> +      TYPE_NEXT_VARIANT (type) = NULL;
>> >> +    }
>> >> +
>> >> +
>> >> +  return type;
>> >> +}
>> >> +
>> >> +/* For given function FNDECL add bounds arguments to arguments
>> >> +   list.  */
>> >> +
>> >> +static void
>> >> +chkp_add_bounds_params_to_function (tree fndecl)
>> >> +{
>> >> +  tree arg;
>> >> +
>> >> +  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
>> >> +    if (BOUNDED_P (arg))
>> >> +      {
>> >> +     std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
>> >> +     tree new_arg;
>> >> +
>> >> +     if (DECL_NAME (arg))
>> >> +       new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
>> >> +     else
>> >> +       {
>> >> +         char uid[25];
>> >> +         snprintf (uid, 25, "D.%u", DECL_UID (arg));
>> >> +         new_name += uid;
>> >> +       }
>> >> +
>> >> +     new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
>> >> +                           get_identifier (new_name.c_str ()),
>> >> +                           pointer_bounds_type_node);
>> >> +     DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
>> >> +     DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
>> >> +     DECL_ARTIFICIAL (new_arg) = 1;
>> >> +     DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
>> >> +     DECL_CHAIN (arg) = new_arg;
>> >> +
>> >> +     arg = DECL_CHAIN (arg);
>> >> +
>> >> +      }
>> >> +    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
>> >> +      {
>> >> +     tree orig_arg = arg;
>> >> +     bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
>> >> +     bitmap_iterator bi;
>> >> +     unsigned bnd_no;
>> >> +
>> >> +     EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
>> >> +       {
>> >> +         std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
>> >> +         tree new_arg;
>> >> +         char offs[25];
>> >> +
>> >> +         if (DECL_NAME (orig_arg))
>> >> +           new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
>> >> +         else
>> >> +           {
>> >> +             snprintf (offs, 25, "D.%u", DECL_UID (arg));
>> >> +             new_name += offs;
>> >> +           }
>> >> +         snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
>> >> +
>> >> +         new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
>> >> +                               PARM_DECL,
>> >> +                               get_identifier (new_name.c_str ()),
>> >> +                               pointer_bounds_type_node);
>> >> +         DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
>> >> +         DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
>> >> +         DECL_ARTIFICIAL (new_arg) = 1;
>> >> +         DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
>> >> +         DECL_CHAIN (arg) = new_arg;
>> >> +
>> >> +         arg = DECL_CHAIN (arg);
>> >> +       }
>> >> +     BITMAP_FREE (slots);
>> >> +      }
>> >> +
>> >> +  TREE_TYPE (fndecl) =
>> >> +    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
>> >> +}
>> >> +
>> >> +/* Return clone created for instrumentation of NODE or NULL.  */
>> >> +
>> >> +cgraph_node *
>> >> +chkp_maybe_create_clone (tree fndecl)
>> >> +{
>> >> +  cgraph_node *node = cgraph_node::get_create (fndecl);
>> >> +  cgraph_node *clone = node->instrumented_version;
>> >> +
>> >> +  gcc_assert (!node->instrumentation_clone);
>> >> +
>> >> +  if (!clone)
>> >> +    {
>> >> +      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
>> >> +      struct cgraph_edge *e;
>> >> +      struct ipa_ref *ref;
>> >> +      int i;
>> >> +
>> >> +      clone = node->create_version_clone (new_decl, vNULL, NULL);
>> >> +      clone->externally_visible = node->externally_visible;
>> >> +      clone->local = node->local;
>> >> +      clone->address_taken = node->address_taken;
>> >> +      clone->thunk = node->thunk;
>> >> +      clone->alias = node->alias;
>> >> +      clone->weakref = node->weakref;
>> >> +      clone->cpp_implicit_alias = node->cpp_implicit_alias;
>> >> +      clone->instrumented_version = node;
>> >> +      clone->orig_decl = fndecl;
>> >> +      clone->instrumentation_clone = true;
>> >> +      node->instrumented_version = clone;
>> >> +
>> >> +      if (gimple_has_body_p (fndecl))
>> >> +     {
>> >> +       /* If function will not be instrumented, then it's instrumented
>> >> +          version is a thunk for the original.  */
>> >> +       if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
>> >> +           || (flag_chkp_instrument_marked_only
>> >> +               && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
>> >> +         {
>> >> +           clone->thunk.thunk_p = true;
>> >> +           clone->thunk.add_pointer_bounds_args = true;
>> >> +           clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
>> >> +         }
>> >> +       else
>> >> +         {
>> >> +           tree_function_versioning (fndecl, new_decl, NULL, false,
>> >> +                                     NULL, false, NULL, NULL);
>> >> +           clone->lowered = true;
>> >> +         }
>> >> +     }
>> >> +
>> >> +      /* New params are inserted after versioning because it
>> >> +      actually copies args list from the original decl.  */
>> >> +      chkp_add_bounds_params_to_function (new_decl);
>> >> +
>> >> +      /* Clones have the same comdat group as originals.  */
>> >> +      if (node->same_comdat_group
>> >> +       || DECL_ONE_ONLY (node->decl))
>> >> +     clone->add_to_same_comdat_group (node);
>> >> +
>> >> +      if (gimple_has_body_p (fndecl))
>> >> +     symtab->call_cgraph_insertion_hooks (clone);
>> >> +
>> >> +      /* Clone all aliases.  */
>> >> +      for (i = 0; node->iterate_referring (i, ref); i++)
>> >> +     if (ref->use == IPA_REF_ALIAS)
>> >> +       {
>> >> +         struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
>> >> +         struct cgraph_node *chkp_alias
>> >> +           = chkp_maybe_create_clone (alias->decl);
>> >> +         chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
>> >> +       }
>> >> +
>> >> +      /* Clone all thunks.  */
>> >> +      for (e = node->callers; e; e = e->next_caller)
>> >> +     if (e->caller->thunk.thunk_p)
>> >> +       {
>> >> +         struct cgraph_node *thunk
>> >> +           = chkp_maybe_create_clone (e->caller->decl);
>> >> +         /* Redirect thunk clone edge to the node clone.  */
>> >> +         thunk->callees->redirect_callee (clone);
>> >> +       }
>> >> +
>> >> +      /* For aliases and thunks we should make sure target is cloned
>> >> +      to have proper references and edges.  */
>> >> +      if (node->thunk.thunk_p)
>> >> +     chkp_maybe_create_clone (node->callees->callee->decl);
>> >> +      else if (node->alias)
>> >> +     {
>> >> +       struct cgraph_node *target;
>> >> +
>> >> +       ref = node->ref_list.first_reference ();
>> >> +       if (ref)
>> >> +         chkp_maybe_create_clone (ref->referred->decl);
>> >> +
>> >> +       if (node->alias_target)
>> >> +         {
>> >> +           if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
>> >> +             {
>> >> +               target = chkp_maybe_create_clone (node->alias_target);
>> >> +               clone->alias_target = target->decl;
>> >> +             }
>> >> +           else
>> >> +             clone->alias_target = node->alias_target;
>> >> +         }
>> >> +     }
>> >> +
>> >> +      /* Add IPA reference.  It's main role is to keep instrumented
>> >> +      version reachable while original node is reachable.  */
>> >> +      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
>> >> +    }
>> >> +
>> >> +  return clone;
>> >> +}
>> >> +
>> >> +/* Create clone for all functions to be instrumented.  */
>> >> +
>> >> +static unsigned int
>> >> +chkp_versioning (void)
>> >> +{
>> >> +  struct cgraph_node *node;
>> >> +
>> >> +  FOR_EACH_DEFINED_FUNCTION (node)
>> >> +    {
>> >> +      if (!node->instrumentation_clone
>> >> +       && !node->instrumented_version
>> >> +       && !node->alias
>> >> +       && !node->thunk.thunk_p
>> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
>> >> +       && (!flag_chkp_instrument_marked_only
>> >> +           || lookup_attribute ("bnd_instrument",
>> >> +                                DECL_ATTRIBUTES (node->decl)))
>> >> +       /* No builtins instrumentation for now.  */
>> >> +       && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
>> >> +     chkp_maybe_create_clone (node->decl);
>> >> +    }
>> >> +
>> >> +  /* Mark all aliases and thunks of functions with no instrumented
>> >> +     version as legacy function.  */
>> >> +  FOR_EACH_DEFINED_FUNCTION (node)
>> >> +    {
>> >> +      if (!node->instrumentation_clone
>> >> +       && !node->instrumented_version
>> >> +       && (node->alias || node->thunk.thunk_p)
>> >> +       && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
>> >> +     DECL_ATTRIBUTES (node->decl)
>> >> +       = tree_cons (get_identifier ("bnd_legacy"), NULL,
>> >> +                    DECL_ATTRIBUTES (node->decl));
>> >> +    }
>> >> +
>> >> +  return 0;
>> >> +}
>> >> +
>> >> +/* In this pass we remove bodies of functions having
>> >> +   instrumented version.  Functions with removed bodies
>> >> +   become a special kind of thunks to provide a connection
>> >> +   between calls to the original version and instrumented
>> >> +   function.  */
>> >> +
>> >> +static unsigned int
>> >> +chkp_produce_thunks (void)
>> >> +{
>> >> +  struct cgraph_node *node;
>> >> +
>> >> +  FOR_EACH_DEFINED_FUNCTION (node)
>> >> +    {
>> >> +      if (!node->instrumentation_clone
>> >> +       && node->instrumented_version
>> >> +       && gimple_has_body_p (node->decl)
>> >> +       && gimple_has_body_p (node->instrumented_version->decl))
>> >> +     {
>> >> +       node->release_body ();
>> >> +       node->remove_callees ();
>> >> +       node->remove_all_references ();
>> >> +
>> >> +       node->thunk.thunk_p = true;
>> >> +       node->thunk.add_pointer_bounds_args = true;
>> >> +       node->create_edge (node->instrumented_version, NULL,
>> >> +                          0, CGRAPH_FREQ_BASE);
>> >> +       node->create_reference (node->instrumented_version,
>> >> +                            IPA_REF_CHKP, NULL);
>> >> +     }
>> >> +    }
>> >> +
>> >> +  /* Mark instrumentation clones created for aliases and thunks
>> >> +     as insttrumented so they could be removed as unreachable
>> >> +     now.  */
>> >> +  FOR_EACH_DEFINED_FUNCTION (node)
>> >> +    {
>> >> +      if (node->instrumentation_clone
>> >> +       && (node->alias || node->thunk.thunk_p)
>> >> +       && !chkp_function_instrumented_p (node->decl))
>> >> +     chkp_function_mark_instrumented (node->decl);
>> >> +    }
>> >> +
>> >> +  symtab->remove_unreachable_nodes (true, dump_file);
>> >> +
>> >> +  return 0;
>> >> +}
>> >> +
>> >> +const pass_data pass_data_ipa_chkp_versioning =
>> >> +{
>> >> +  SIMPLE_IPA_PASS, /* type */
>> >> +  "chkp_versioning", /* name */
>> >> +  OPTGROUP_NONE, /* optinfo_flags */
>> >> +  TV_NONE, /* tv_id */
>> >> +  0, /* properties_required */
>> >> +  0, /* properties_provided */
>> >> +  0, /* properties_destroyed */
>> >> +  0, /* todo_flags_start */
>> >> +  0 /* todo_flags_finish */
>> >> +};
>> >> +
>> >> +const pass_data pass_data_ipa_chkp_produce_thunks =
>> >> +{
>> >> +  SIMPLE_IPA_PASS, /* type */
>> >> +  "chkp_cleanup", /* name */
>> >> +  OPTGROUP_NONE, /* optinfo_flags */
>> >> +  TV_NONE, /* tv_id */
>> >> +  0, /* properties_required */
>> >> +  0, /* properties_provided */
>> >> +  0, /* properties_destroyed */
>> >> +  0, /* todo_flags_start */
>> >> +  0 /* todo_flags_finish */
>> >> +};
>> >> +
>> >> +class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
>> >> +{
>> >> +public:
>> >> +  pass_ipa_chkp_versioning (gcc::context *ctxt)
>> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
>> >> +  {}
>> >> +
>> >> +  /* opt_pass methods: */
>> >> +  virtual opt_pass * clone ()
>> >> +    {
>> >> +      return new pass_ipa_chkp_versioning (m_ctxt);
>> >> +    }
>> >> +
>> >> +  virtual bool gate (function *)
>> >> +    {
>> >> +      return flag_check_pointer_bounds;
>> >> +    }
>> >> +
>> >> +  virtual unsigned int execute (function *)
>> >> +    {
>> >> +      return chkp_versioning ();
>> >> +    }
>> >> +
>> >> +}; // class pass_ipa_chkp_versioning
>> >> +
>> >> +class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
>> >> +{
>> >> +public:
>> >> +  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
>> >> +    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
>> >> +  {}
>> >> +
>> >> +  /* opt_pass methods: */
>> >> +  virtual opt_pass * clone ()
>> >> +    {
>> >> +      return new pass_ipa_chkp_produce_thunks (m_ctxt);
>> >> +    }
>> >> +
>> >> +  virtual unsigned int execute (function *)
>> >> +    {
>> >> +      return chkp_produce_thunks ();
>> >> +    }
>> >> +
>> >> +}; // class pass_chkp_produce_thunks
>> >> +
>> >> +simple_ipa_opt_pass *
>> >> +make_pass_ipa_chkp_versioning (gcc::context *ctxt)
>> >> +{
>> >> +  return new pass_ipa_chkp_versioning (ctxt);
>> >> +}
>> >> +
>> >> +simple_ipa_opt_pass *
>> >> +make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
>> >> +{
>> >> +  return new pass_ipa_chkp_produce_thunks (ctxt);
>> >> +}
>> >> diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
>> >> new file mode 100644
>> >> index 0000000..9c92c04
>> >> --- /dev/null
>> >> +++ b/gcc/ipa-chkp.h
>> >> @@ -0,0 +1,29 @@
>> >> +/* Declaration of interface functions of Pointer Bounds Checker.
>> >> +   Copyright (C) 2014 Free Software Foundation, Inc.
>> >> +
>> >> +This file is part of GCC.
>> >> +
>> >> +GCC is free software; you can redistribute it and/or modify it under
>> >> +the terms of the GNU General Public License as published by the Free
>> >> +Software Foundation; either version 3, or (at your option) any later
>> >> +version.
>> >> +
>> >> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
>> >> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> >> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
>> >> +for more details.
>> >> +
>> >> +You should have received a copy of the GNU General Public License
>> >> +along with GCC; see the file COPYING3.  If not see
>> >> +<http://www.gnu.org/licenses/>.  */
>> >> +
>> >> +#ifndef GCC_IPA_CHKP_H
>> >> +#define GCC_IPA_CHKP_H
>> >> +
>> >> +#include "tree.h"
>> >> +#include "cgraph.h"
>> >> +
>> >> +extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
>> >> +extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
>> >> +
>> >> +#endif /* GCC_IPA_CHKP_H */
Jeff Law Oct. 10, 2014, 7:14 p.m. UTC | #6
On 10/08/14 12:55, Ilya Enkovich wrote:
> Hi,
>
> This patch introduces two IPA passes used by Pointer Bounds Checker.  One pass creates clones for instrumentation.  The other one transforms unneeded functions into thunks.
>
> Thanks,
> Ilya
> --
2014-10-08  Ilya Enkovich  <ilya.enkovich@intel.com>

	* ipa-chkp.c: New.
	* ipa-chkp.h: New.
	* Makefile.in (OBJS): Add ipa-chkp.o.


> +    when linker reolution files are used.  Linker has no info about
s/reolution/???


> +    connection between original and instrumented function and
> +    therefore we may wrongly decide (due to difference in assember
s/assember/assembler/


> +
> +  /* We want called_as_built_in recall instrumented calls
> +     to instrumented built-in functions.
Sorry, I simply can't parse that sentence...  Can you rewrite/revise it 
to be clearer?




> +
> +  /* If function type has attribute with arg indexes then
> +     we have to copy it fixing attribute ops.  Map for
> +     fixing is in indexes array.  */
> +  attrs = TYPE_ATTRIBUTES (type);
> +  if (lookup_attribute ("nonnull", attrs)
> +      || lookup_attribute ("format", attrs)
> +      || lookup_attribute ("format_arg", attrs))
So you need to adjust the indices in those attributes so that they refer 
to the right argument.  I get that.  My concern is how do we keep the 
list of attributes that must be fixed up-to-date.

I can easily see someone adding a new argument attribute and folks 
forgetting to update this code.

Seems to me we need to document the attribute code with a reference back 
to these ipa-chkp bits.  It's not foolproof, but it's better than 
depending on my memory to catch this stuff during reviews.

OK with the nits above fixed and a suitable comment that folks adding 
new attributes with parameter indices are likely to see to refer them 
back to the index adjustment code in ipa-chkp.c.

Thanks,
Jeff
diff mbox

Patch

diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index 3113a9f..d8c8488 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -1253,6 +1253,7 @@  OBJS = \
 	incpath.o \
 	init-regs.o \
 	internal-fn.o \
+	ipa-chkp.o \
 	ipa-cp.o \
 	ipa-devirt.o \
 	ipa-split.o \
diff --git a/gcc/ipa-chkp.c b/gcc/ipa-chkp.c
new file mode 100644
index 0000000..f7ac713
--- /dev/null
+++ b/gcc/ipa-chkp.c
@@ -0,0 +1,622 @@ 
+/* Pointer Bounds Checker IPA passes.
+   Copyright (C) 2014 Free Software Foundation, Inc.
+   Contributed by Ilya Enkovich (ilya.enkovich@intel.com)
+
+This file is part of GCC.
+
+GCC is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 3, or (at your option) any later
+version.
+
+GCC is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with GCC; see the file COPYING3.  If not see
+<http://www.gnu.org/licenses/>.  */
+
+#include "config.h"
+#include "system.h"
+#include "coretypes.h"
+#include "tree-core.h"
+#include "stor-layout.h"
+#include "tree.h"
+#include "tree-pass.h"
+#include "stringpool.h"
+#include "bitmap.h"
+#include "gimple-expr.h"
+#include "function.h"
+#include "tree-chkp.h"
+#include <string>
+
+/*  Pointer Bounds Checker has two IPA passes to support code instrumentation.
+
+    In instrumented code each pointer is provided with bounds.  For input
+    pointer parameters it means we also have bounds passed.  For calls it
+    means we have additional bounds arguments for pointer arguments.
+
+    To have all IPA optimizations working correctly we have to express
+    dataflow between passed and received bounds explicitly via additional
+    entries in function declaration arguments list and in function type.
+    Since we may have both instrumented and not instrumented code at the
+    same time, we cannot replace all original functions with their
+    instrumented variants.  Therefore we create clones (versions) instead.
+
+    Instrumentation clones creation is a separate IPA pass which is a part
+    of early local passes.  Clones are created after SSA is built (because
+    instrumentation pass works on SSA) and before any transformations
+    which may change pointer flow and therefore lead to incorrect code
+    instrumentation (possibly causing false bounds check failures).
+
+    Instrumentation clones have pointer bounds arguments added right after
+    pointer arguments.  Clones have assembler name of the original
+    function with suffix added.  New assembler name is in transparent
+    alias chain with the original name.  Thus we expect all calls to the
+    original and instrumented functions look similar in assembler.
+
+    During instrumentation versioning pass we create instrumented versions
+    of all function with body and also for all their aliases and thunks.
+    Clones for functions with no body are created on demand (usually
+    during call instrumentation).
+
+    Original and instrumented function nodes are connected with IPA
+    reference IPA_REF_CHKP.  It is mostly done to have reachability
+    analysis working correctly.  We may have no references to the
+    instrumented function in the code but it still should be counted
+    as reachable if the original function is reachable.
+
+    When original function bodies are not needed anymore we release
+    them and transform functions into a special kind of thunks.  Each
+    thunk has a call edge to the instrumented version.  These thunks
+    help to keep externally visible instrumented functions visible
+    when linker reolution files are used.  Linker has no info about
+    connection between original and instrumented function and
+    therefore we may wrongly decide (due to difference in assember
+    names) that instrumented function version is local and can be
+    removed.  */
+
+#define CHKP_BOUNDS_OF_SYMBOL_PREFIX "__chkp_bounds_of_"
+
+/* Build a clone of FNDECL with a modified name.  */
+
+static tree
+chkp_build_instrumented_fndecl (tree fndecl)
+{
+  tree new_decl = copy_node (fndecl);
+  tree new_name;
+  std::string s;
+
+  /* We want called_as_built_in recall instrumented calls
+     to instrumented built-in functions.  Therefore use
+     DECL_NAME for cloning instead of DECL_ASSEMBLER_NAME.  */
+  s = IDENTIFIER_POINTER (DECL_NAME (fndecl));
+  s += ".chkp";
+  DECL_NAME (new_decl) = get_identifier (s.c_str ());
+
+  /* References to the original and to the instrumented version
+     should look the same in the output assembly.  And we cannot
+     use the same assembler name for the instrumented version
+     because it conflicts with decl merging algorithms in LTO.
+     Achieve the result by using transparent alias name for the
+     instrumented version.  */
+  s = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (fndecl));
+  s += ".chkp";
+  new_name = get_identifier (s.c_str ());
+  IDENTIFIER_TRANSPARENT_ALIAS (new_name) = 1;
+  TREE_CHAIN (new_name) = DECL_ASSEMBLER_NAME (fndecl);
+  SET_DECL_ASSEMBLER_NAME (new_decl, new_name);
+
+  /* For functions with body versioning will make a copy of arguments.
+     For functions with no body we need to do it here.  */
+  if (!gimple_has_body_p (fndecl))
+    DECL_ARGUMENTS (new_decl) = copy_list (DECL_ARGUMENTS (fndecl));
+
+  /* We are going to modify attributes list and therefore should
+     make own copy.  */
+  DECL_ATTRIBUTES (new_decl) = copy_list (DECL_ATTRIBUTES (fndecl));
+
+  return new_decl;
+}
+
+
+/* Fix operands of attribute from ATTRS list named ATTR_NAME.
+   Integer operands are replaced with values according to
+   INDEXES map having LEN elements.  For operands out of len
+   we just add DELTA.  */
+
+static void
+chkp_map_attr_arg_indexes (tree attrs, const char *attr_name,
+			   unsigned *indexes, int len, int delta)
+{
+  tree attr = lookup_attribute (attr_name, attrs);
+  tree op;
+
+  if (!attr)
+    return;
+
+  TREE_VALUE (attr) = copy_list (TREE_VALUE (attr));
+  for (op = TREE_VALUE (attr); op; op = TREE_CHAIN (op))
+    {
+      int idx;
+
+      if (TREE_CODE (TREE_VALUE (op)) != INTEGER_CST)
+	continue;
+
+      idx = TREE_INT_CST_LOW (TREE_VALUE (op));
+
+      /* If idx exceeds indexes length then we just
+	 keep it at the same distance from the last
+	 known arg.  */
+      if (idx > len)
+	idx += delta;
+      else
+	idx = indexes[idx - 1] + 1;
+      TREE_VALUE (op) = build_int_cst (TREE_TYPE (TREE_VALUE (op)), idx);
+    }
+}
+
+/* Make a copy of function type ORIG_TYPE adding pointer
+   bounds as additional arguments.  */
+
+tree
+chkp_copy_function_type_adding_bounds (tree orig_type)
+{
+  tree type;
+  tree arg_type, attrs, t;
+  unsigned len = list_length (TYPE_ARG_TYPES (orig_type));
+  unsigned *indexes = XALLOCAVEC (unsigned, len);
+  unsigned idx = 0, new_idx = 0;
+
+  for (arg_type = TYPE_ARG_TYPES (orig_type);
+       arg_type;
+       arg_type = TREE_CHAIN (arg_type))
+    if (TREE_VALUE (arg_type) == void_type_node)
+      continue;
+    else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
+	     || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
+				   TREE_VALUE (arg_type), true)
+	     || chkp_type_has_pointer (TREE_VALUE (arg_type)))
+      break;
+
+  /* We may use original type if there are no bounds passed.  */
+  if (!arg_type)
+    return orig_type;
+
+  type = copy_node (orig_type);
+  TYPE_ARG_TYPES (type) = copy_list (TYPE_ARG_TYPES (type));
+
+  for (arg_type = TYPE_ARG_TYPES (type);
+       arg_type;
+       arg_type = TREE_CHAIN (arg_type))
+    {
+      indexes[idx++] = new_idx++;
+
+      /* pass_by_reference returns 1 for void type,
+	 so check for it first.  */
+      if (TREE_VALUE (arg_type) == void_type_node)
+	continue;
+      else if (BOUNDED_TYPE_P (TREE_VALUE (arg_type))
+	       || pass_by_reference (NULL, TYPE_MODE (TREE_VALUE (arg_type)),
+				     TREE_VALUE (arg_type), true))
+	{
+	  tree new_type = build_tree_list (NULL_TREE,
+					   pointer_bounds_type_node);
+	  TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
+	  TREE_CHAIN (arg_type) = new_type;
+
+	  arg_type = TREE_CHAIN (arg_type);
+	  new_idx++;
+	}
+      else if (chkp_type_has_pointer (TREE_VALUE (arg_type)))
+	{
+	  bitmap slots = chkp_find_bound_slots (TREE_VALUE (arg_type));
+	  bitmap_iterator bi;
+	  unsigned bnd_no;
+
+	  EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
+	    {
+	      tree new_type = build_tree_list (NULL_TREE,
+					       pointer_bounds_type_node);
+	      TREE_CHAIN (new_type) = TREE_CHAIN (arg_type);
+	      TREE_CHAIN (arg_type) = new_type;
+
+	      arg_type = TREE_CHAIN (arg_type);
+	      new_idx++;
+	    }
+	  BITMAP_FREE (slots);
+	}
+    }
+
+  /* If function type has attribute with arg indexes then
+     we have to copy it fixing attribute ops.  Map for
+     fixing is in indexes array.  */
+  attrs = TYPE_ATTRIBUTES (type);
+  if (lookup_attribute ("nonnull", attrs)
+      || lookup_attribute ("format", attrs)
+      || lookup_attribute ("format_arg", attrs))
+    {
+      int delta = new_idx - len;
+      attrs = copy_list (TYPE_ATTRIBUTES (type));
+      chkp_map_attr_arg_indexes (attrs, "nonnull", indexes, len, delta);
+      chkp_map_attr_arg_indexes (attrs, "format", indexes, len, delta);
+      chkp_map_attr_arg_indexes (attrs, "format_arg", indexes, len, delta);
+      TYPE_ATTRIBUTES (type) = attrs;
+    }
+
+  t = TYPE_MAIN_VARIANT (orig_type);
+  if (orig_type != t)
+    {
+      TYPE_MAIN_VARIANT (type) = t;
+      TYPE_NEXT_VARIANT (type) = TYPE_NEXT_VARIANT (t);
+      TYPE_NEXT_VARIANT (t) = type;
+    }
+  else
+    {
+      TYPE_MAIN_VARIANT (type) = type;
+      TYPE_NEXT_VARIANT (type) = NULL;
+    }
+
+
+  return type;
+}
+
+/* For given function FNDECL add bounds arguments to arguments
+   list.  */
+
+static void
+chkp_add_bounds_params_to_function (tree fndecl)
+{
+  tree arg;
+
+  for (arg = DECL_ARGUMENTS (fndecl); arg; arg = DECL_CHAIN (arg))
+    if (BOUNDED_P (arg))
+      {
+	std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
+	tree new_arg;
+
+	if (DECL_NAME (arg))
+	  new_name += IDENTIFIER_POINTER (DECL_NAME (arg));
+	else
+	  {
+	    char uid[25];
+	    snprintf (uid, 25, "D.%u", DECL_UID (arg));
+	    new_name += uid;
+	  }
+
+	new_arg = build_decl (DECL_SOURCE_LOCATION (arg), PARM_DECL,
+			      get_identifier (new_name.c_str ()),
+			      pointer_bounds_type_node);
+	DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
+	DECL_CONTEXT (new_arg) = DECL_CONTEXT (arg);
+	DECL_ARTIFICIAL (new_arg) = 1;
+	DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
+	DECL_CHAIN (arg) = new_arg;
+
+	arg = DECL_CHAIN (arg);
+
+      }
+    else if (chkp_type_has_pointer (TREE_TYPE (arg)))
+      {
+	tree orig_arg = arg;
+	bitmap slots = chkp_find_bound_slots (TREE_TYPE (arg));
+	bitmap_iterator bi;
+	unsigned bnd_no;
+
+	EXECUTE_IF_SET_IN_BITMAP (slots, 0, bnd_no, bi)
+	  {
+	    std::string new_name = CHKP_BOUNDS_OF_SYMBOL_PREFIX;
+	    tree new_arg;
+	    char offs[25];
+
+	    if (DECL_NAME (orig_arg))
+	      new_name += IDENTIFIER_POINTER (DECL_NAME (orig_arg));
+	    else
+	      {
+		snprintf (offs, 25, "D.%u", DECL_UID (arg));
+		new_name += offs;
+	      }
+	    snprintf (offs, 25, "__%u", bnd_no * POINTER_SIZE / BITS_PER_UNIT);
+
+	    new_arg = build_decl (DECL_SOURCE_LOCATION (orig_arg),
+				  PARM_DECL,
+				  get_identifier (new_name.c_str ()),
+				  pointer_bounds_type_node);
+	    DECL_ARG_TYPE (new_arg) = pointer_bounds_type_node;
+	    DECL_CONTEXT (new_arg) = DECL_CONTEXT (orig_arg);
+	    DECL_ARTIFICIAL (new_arg) = 1;
+	    DECL_CHAIN (new_arg) = DECL_CHAIN (arg);
+	    DECL_CHAIN (arg) = new_arg;
+
+	    arg = DECL_CHAIN (arg);
+	  }
+	BITMAP_FREE (slots);
+      }
+
+  TREE_TYPE (fndecl) =
+    chkp_copy_function_type_adding_bounds (TREE_TYPE (fndecl));
+}
+
+/* Return clone created for instrumentation of NODE or NULL.  */
+
+cgraph_node *
+chkp_maybe_create_clone (tree fndecl)
+{
+  cgraph_node *node = cgraph_node::get_create (fndecl);
+  cgraph_node *clone = node->instrumented_version;
+
+  gcc_assert (!node->instrumentation_clone);
+
+  if (!clone)
+    {
+      tree new_decl = chkp_build_instrumented_fndecl (fndecl);
+      struct cgraph_edge *e;
+      struct ipa_ref *ref;
+      int i;
+
+      clone = node->create_version_clone (new_decl, vNULL, NULL);
+      clone->externally_visible = node->externally_visible;
+      clone->local = node->local;
+      clone->address_taken = node->address_taken;
+      clone->thunk = node->thunk;
+      clone->alias = node->alias;
+      clone->weakref = node->weakref;
+      clone->cpp_implicit_alias = node->cpp_implicit_alias;
+      clone->instrumented_version = node;
+      clone->orig_decl = fndecl;
+      clone->instrumentation_clone = true;
+      node->instrumented_version = clone;
+
+      if (gimple_has_body_p (fndecl))
+	{
+	  /* If function will not be instrumented, then it's instrumented
+	     version is a thunk for the original.  */
+	  if (lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (fndecl))
+	      || (flag_chkp_instrument_marked_only
+		  && !lookup_attribute ("bnd_instrument", DECL_ATTRIBUTES (fndecl))))
+	    {
+	      clone->thunk.thunk_p = true;
+	      clone->thunk.add_pointer_bounds_args = true;
+	      clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
+	    }
+	  else
+	    {
+	      tree_function_versioning (fndecl, new_decl, NULL, false,
+					NULL, false, NULL, NULL);
+	      clone->lowered = true;
+	    }
+	}
+
+      /* New params are inserted after versioning because it
+	 actually copies args list from the original decl.  */
+      chkp_add_bounds_params_to_function (new_decl);
+
+      /* Clones have the same comdat group as originals.  */
+      if (node->same_comdat_group
+	  || DECL_ONE_ONLY (node->decl))
+	clone->add_to_same_comdat_group (node);
+
+      if (gimple_has_body_p (fndecl))
+	symtab->call_cgraph_insertion_hooks (clone);
+
+      /* Clone all aliases.  */
+      for (i = 0; node->iterate_referring (i, ref); i++)
+	if (ref->use == IPA_REF_ALIAS)
+	  {
+	    struct cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring);
+	    struct cgraph_node *chkp_alias
+	      = chkp_maybe_create_clone (alias->decl);
+	    chkp_alias->create_reference (clone, IPA_REF_ALIAS, NULL);
+	  }
+
+      /* Clone all thunks.  */
+      for (e = node->callers; e; e = e->next_caller)
+	if (e->caller->thunk.thunk_p)
+	  {
+	    struct cgraph_node *thunk
+	      = chkp_maybe_create_clone (e->caller->decl);
+	    /* Redirect thunk clone edge to the node clone.  */
+	    thunk->callees->redirect_callee (clone);
+	  }
+
+      /* For aliases and thunks we should make sure target is cloned
+	 to have proper references and edges.  */
+      if (node->thunk.thunk_p)
+	chkp_maybe_create_clone (node->callees->callee->decl);
+      else if (node->alias)
+	{
+	  struct cgraph_node *target;
+
+	  ref = node->ref_list.first_reference ();
+	  if (ref)
+	    chkp_maybe_create_clone (ref->referred->decl);
+
+	  if (node->alias_target)
+	    {
+	      if (TREE_CODE (node->alias_target) == FUNCTION_DECL)
+		{
+		  target = chkp_maybe_create_clone (node->alias_target);
+		  clone->alias_target = target->decl;
+		}
+	      else
+		clone->alias_target = node->alias_target;
+	    }
+	}
+
+      /* Add IPA reference.  It's main role is to keep instrumented
+	 version reachable while original node is reachable.  */
+      ref = node->create_reference (clone, IPA_REF_CHKP, NULL);
+    }
+
+  return clone;
+}
+
+/* Create clone for all functions to be instrumented.  */
+
+static unsigned int
+chkp_versioning (void)
+{
+  struct cgraph_node *node;
+
+  FOR_EACH_DEFINED_FUNCTION (node)
+    {
+      if (!node->instrumentation_clone
+	  && !node->instrumented_version
+	  && !node->alias
+	  && !node->thunk.thunk_p
+	  && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl))
+	  && (!flag_chkp_instrument_marked_only
+	      || lookup_attribute ("bnd_instrument",
+				   DECL_ATTRIBUTES (node->decl)))
+	  /* No builtins instrumentation for now.  */
+	  && DECL_BUILT_IN_CLASS (node->decl) == NOT_BUILT_IN)
+	chkp_maybe_create_clone (node->decl);
+    }
+
+  /* Mark all aliases and thunks of functions with no instrumented
+     version as legacy function.  */
+  FOR_EACH_DEFINED_FUNCTION (node)
+    {
+      if (!node->instrumentation_clone
+	  && !node->instrumented_version
+	  && (node->alias || node->thunk.thunk_p)
+	  && !lookup_attribute ("bnd_legacy", DECL_ATTRIBUTES (node->decl)))
+	DECL_ATTRIBUTES (node->decl)
+	  = tree_cons (get_identifier ("bnd_legacy"), NULL,
+		       DECL_ATTRIBUTES (node->decl));
+    }
+
+  return 0;
+}
+
+/* In this pass we remove bodies of functions having
+   instrumented version.  Functions with removed bodies
+   become a special kind of thunks to provide a connection
+   between calls to the original version and instrumented
+   function.  */
+
+static unsigned int
+chkp_produce_thunks (void)
+{
+  struct cgraph_node *node;
+
+  FOR_EACH_DEFINED_FUNCTION (node)
+    {
+      if (!node->instrumentation_clone
+	  && node->instrumented_version
+	  && gimple_has_body_p (node->decl)
+	  && gimple_has_body_p (node->instrumented_version->decl))
+	{
+	  node->release_body ();
+	  node->remove_callees ();
+	  node->remove_all_references ();
+
+	  node->thunk.thunk_p = true;
+	  node->thunk.add_pointer_bounds_args = true;
+	  node->create_edge (node->instrumented_version, NULL,
+			     0, CGRAPH_FREQ_BASE);
+	  node->create_reference (node->instrumented_version,
+			       IPA_REF_CHKP, NULL);
+	}
+    }
+
+  /* Mark instrumentation clones created for aliases and thunks
+     as insttrumented so they could be removed as unreachable
+     now.  */
+  FOR_EACH_DEFINED_FUNCTION (node)
+    {
+      if (node->instrumentation_clone
+	  && (node->alias || node->thunk.thunk_p)
+	  && !chkp_function_instrumented_p (node->decl))
+	chkp_function_mark_instrumented (node->decl);
+    }
+
+  symtab->remove_unreachable_nodes (true, dump_file);
+
+  return 0;
+}
+
+const pass_data pass_data_ipa_chkp_versioning =
+{
+  SIMPLE_IPA_PASS, /* type */
+  "chkp_versioning", /* name */
+  OPTGROUP_NONE, /* optinfo_flags */
+  TV_NONE, /* tv_id */
+  0, /* properties_required */
+  0, /* properties_provided */
+  0, /* properties_destroyed */
+  0, /* todo_flags_start */
+  0 /* todo_flags_finish */
+};
+
+const pass_data pass_data_ipa_chkp_produce_thunks =
+{
+  SIMPLE_IPA_PASS, /* type */
+  "chkp_cleanup", /* name */
+  OPTGROUP_NONE, /* optinfo_flags */
+  TV_NONE, /* tv_id */
+  0, /* properties_required */
+  0, /* properties_provided */
+  0, /* properties_destroyed */
+  0, /* todo_flags_start */
+  0 /* todo_flags_finish */
+};
+
+class pass_ipa_chkp_versioning : public simple_ipa_opt_pass
+{
+public:
+  pass_ipa_chkp_versioning (gcc::context *ctxt)
+    : simple_ipa_opt_pass (pass_data_ipa_chkp_versioning, ctxt)
+  {}
+
+  /* opt_pass methods: */
+  virtual opt_pass * clone ()
+    {
+      return new pass_ipa_chkp_versioning (m_ctxt);
+    }
+
+  virtual bool gate (function *)
+    {
+      return flag_check_pointer_bounds;
+    }
+
+  virtual unsigned int execute (function *)
+    {
+      return chkp_versioning ();
+    }
+
+}; // class pass_ipa_chkp_versioning
+
+class pass_ipa_chkp_produce_thunks : public simple_ipa_opt_pass
+{
+public:
+  pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
+    : simple_ipa_opt_pass (pass_data_ipa_chkp_produce_thunks, ctxt)
+  {}
+
+  /* opt_pass methods: */
+  virtual opt_pass * clone ()
+    {
+      return new pass_ipa_chkp_produce_thunks (m_ctxt);
+    }
+
+  virtual unsigned int execute (function *)
+    {
+      return chkp_produce_thunks ();
+    }
+
+}; // class pass_chkp_produce_thunks
+
+simple_ipa_opt_pass *
+make_pass_ipa_chkp_versioning (gcc::context *ctxt)
+{
+  return new pass_ipa_chkp_versioning (ctxt);
+}
+
+simple_ipa_opt_pass *
+make_pass_ipa_chkp_produce_thunks (gcc::context *ctxt)
+{
+  return new pass_ipa_chkp_produce_thunks (ctxt);
+}
diff --git a/gcc/ipa-chkp.h b/gcc/ipa-chkp.h
new file mode 100644
index 0000000..9c92c04
--- /dev/null
+++ b/gcc/ipa-chkp.h
@@ -0,0 +1,29 @@ 
+/* Declaration of interface functions of Pointer Bounds Checker.
+   Copyright (C) 2014 Free Software Foundation, Inc.
+
+This file is part of GCC.
+
+GCC is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 3, or (at your option) any later
+version.
+
+GCC is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with GCC; see the file COPYING3.  If not see
+<http://www.gnu.org/licenses/>.  */
+
+#ifndef GCC_IPA_CHKP_H
+#define GCC_IPA_CHKP_H
+
+#include "tree.h"
+#include "cgraph.h"
+
+extern tree chkp_copy_function_type_adding_bounds (tree orig_type);
+extern cgraph_node *chkp_maybe_create_clone (tree fndecl);
+
+#endif /* GCC_IPA_CHKP_H */