diff mbox series

[committed] jit: fix ICE on BUILT_IN_TRAP [PR99126]

Message ID 20210219023144.1800203-1-dmalcolm@redhat.com
State New
Headers show
Series [committed] jit: fix ICE on BUILT_IN_TRAP [PR99126] | expand

Commit Message

David Malcolm Feb. 19, 2021, 2:31 a.m. UTC
I tried several approaches to fixing this; this seemed the
least invasive.

Successfully bootstrapped & regrtested on x86_64-pc-linux-gnu.
Pushed to trunk as r11-7288-gb258e263e0d74ca1f76aeaac5f4d1abef6b13707.

gcc/jit/ChangeLog:
	PR jit/99126
	* jit-builtins.c
	(gcc::jit::builtins_manager::get_builtin_function_by_id):
	Update assertion to reject BUILT_IN_NONE.
	(gcc::jit::builtins_manager::ensure_optimization_builtins_exist):
	New.
	* jit-builtins.h
	(gcc::jit::builtins_manager::ensure_optimization_builtins_exist):
	New decl.
	* jit-playback.c (gcc::jit::playback::context::replay): Call it.
	Remove redundant conditional on bm.

gcc/testsuite/ChangeLog:
	PR jit/99126
	* jit.dg/test-trap.c: New test.
---
 gcc/jit/jit-builtins.c           | 14 +++++++-
 gcc/jit/jit-builtins.h           |  3 ++
 gcc/jit/jit-playback.c           | 11 +++---
 gcc/testsuite/jit.dg/test-trap.c | 59 ++++++++++++++++++++++++++++++++
 4 files changed, 82 insertions(+), 5 deletions(-)
 create mode 100644 gcc/testsuite/jit.dg/test-trap.c

Comments

Andrea Corallo Feb. 19, 2021, 10:16 a.m. UTC | #1
David Malcolm via Gcc-patches <gcc-patches@gcc.gnu.org> writes:

> I tried several approaches to fixing this; this seemed the
> least invasive.

Hi Dave,

thanks for fixing this.

I do have a trivial question: are we guaranteed that the middle-end will
not try to add any build-in other than a trap?

Regards

  Andrea
Andrea Corallo Feb. 19, 2021, 10:32 a.m. UTC | #2
Andrea Corallo via Gcc-patches <gcc-patches@gcc.gnu.org> writes:

> David Malcolm via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>
>> I tried several approaches to fixing this; this seemed the
>> least invasive.
>
> Hi Dave,
>
> thanks for fixing this.
>
> I do have a trivial question: are we guaranteed that the middle-end will
> not try to add any build-in other than a trap?

Ah also, shouldn't the fix be also back-ported?

Thanks

  Andrea
David Malcolm Feb. 19, 2021, 9:13 p.m. UTC | #3
On Fri, 2021-02-19 at 11:16 +0100, Andrea Corallo wrote:
> David Malcolm via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> 
> > I tried several approaches to fixing this; this seemed the
> > least invasive.
> 
> Hi Dave,
> 
> thanks for fixing this.
> 
> I do have a trivial question: are we guaranteed that the middle-end
> will
> not try to add any build-in other than a trap?

No; the middle-end can add various other builtins.

The C/C++ FE have c_define_builtins, which calls def_builtin_1 on
everything with DEF_BUILTIN, then calls:
  targetm.init_builtins ();
  build_common_builtin_nodes ();

jit_langhook_init calls build_common_builtin_nodes.

The jit builtins_manager creates recording::functions (and their
supporting types) for any builtins that are explicitly used, and these
recording::functions get turned into trees during playback.

I looked at the doing the equivalent of DEF_BUILTIN for any builtins
that haven't been created yet, but when to do it?

(a) in the builtin manager... but not all builtin-types are supported
yet (e.g. BT_FLOAT16), so it would be a big patch to do it that way

(b) in jit_langhook_init... but then if the user creates a builtin that
wasn't referenced before in a 2nd in-memory compile, I think the types
can get out-of-sync.

So I went with the patch I did as it seems to be the least invasive way
of fixing the specific problem.

Dave
diff mbox series

Patch

diff --git a/gcc/jit/jit-builtins.c b/gcc/jit/jit-builtins.c
index 18e477cc907..1ea96f4e025 100644
--- a/gcc/jit/jit-builtins.c
+++ b/gcc/jit/jit-builtins.c
@@ -162,7 +162,7 @@  builtins_manager::get_builtin_function (const char *name)
 recording::function *
 builtins_manager::get_builtin_function_by_id (enum built_in_function builtin_id)
 {
-  gcc_assert (builtin_id >= 0);
+  gcc_assert (builtin_id > BUILT_IN_NONE);
   gcc_assert (builtin_id < END_BUILTINS);
 
   /* Lazily build the functions, caching them so that repeated calls for
@@ -600,6 +600,18 @@  builtins_manager::make_ptr_type (enum jit_builtin_type,
   return base_type->get_pointer ();
 }
 
+/* Ensure that builtins that could be needed during optimization
+   get created ahead of time.  */
+
+void
+builtins_manager::ensure_optimization_builtins_exist ()
+{
+  /* build_common_builtin_nodes does most of this, but not all.
+     We can't loop through all of the builtin_data array, we don't
+     support all types yet.  */
+  (void)get_builtin_function_by_id (BUILT_IN_TRAP);
+}
+
 /* Playback support.  */
 
 /* A builtins_manager is associated with a recording::context
diff --git a/gcc/jit/jit-builtins.h b/gcc/jit/jit-builtins.h
index b9f008dd4e2..c5e2b2dd600 100644
--- a/gcc/jit/jit-builtins.h
+++ b/gcc/jit/jit-builtins.h
@@ -127,6 +127,9 @@  public:
   tree
   get_attrs_tree (enum built_in_attribute attr);
 
+  void
+  ensure_optimization_builtins_exist ();
+
   void
   finish_playback (void);
 
diff --git a/gcc/jit/jit-playback.c b/gcc/jit/jit-playback.c
index 152ef250949..c6136301243 100644
--- a/gcc/jit/jit-playback.c
+++ b/gcc/jit/jit-playback.c
@@ -2949,6 +2949,11 @@  replay ()
   /* Replay the recorded events:  */
   timevar_push (TV_JIT_REPLAY);
 
+  /* Ensure that builtins that could be needed during optimization
+     get created ahead of time.  */
+  builtins_manager *bm = m_recording_ctxt->get_builtins_manager ();
+  bm->ensure_optimization_builtins_exist ();
+
   m_recording_ctxt->replay_into (this);
 
   /* Clean away the temporary references from recording objects
@@ -2957,13 +2962,11 @@  replay ()
      refs.  Hence we must stop using them before the GC can run.  */
   m_recording_ctxt->disassociate_from_playback ();
 
-  /* The builtins_manager, if any, is associated with the recording::context
+  /* The builtins_manager is associated with the recording::context
      and might be reused for future compiles on other playback::contexts,
      but its m_attributes array is not GTY-labeled and hence will become
      nonsense if the GC runs.  Purge this state.  */
-  builtins_manager *bm = get_builtins_manager ();
-  if (bm)
-    bm->finish_playback ();
+  bm->finish_playback ();
 
   timevar_pop (TV_JIT_REPLAY);
 
diff --git a/gcc/testsuite/jit.dg/test-trap.c b/gcc/testsuite/jit.dg/test-trap.c
new file mode 100644
index 00000000000..4eb65cd14c1
--- /dev/null
+++ b/gcc/testsuite/jit.dg/test-trap.c
@@ -0,0 +1,59 @@ 
+#include <stdlib.h>
+#include <stdio.h>
+#include <stddef.h>
+
+#include "libgccjit.h"
+
+#include "harness.h"
+
+void
+create_code (gcc_jit_context *ctxt, void *user_data)
+{
+  /* Let's try to inject the equivalent of:
+
+     void
+     test_trap (void)
+     {
+       *((int *)0) = 42;
+     }
+  */
+  gcc_jit_type *void_type
+    = gcc_jit_context_get_type (ctxt, GCC_JIT_TYPE_VOID);
+  gcc_jit_type *int_type
+    = gcc_jit_context_get_type (ctxt, GCC_JIT_TYPE_INT);
+  gcc_jit_type *int_ptr_type
+    = gcc_jit_type_get_pointer (int_type);
+
+  /* Build the test_fn.  */
+  gcc_jit_function *func
+    = gcc_jit_context_new_function (ctxt, NULL,
+				    GCC_JIT_FUNCTION_EXPORTED,
+				    void_type,
+				    "test_trap",
+				    0, NULL,
+				    0);
+
+  gcc_jit_block *initial = gcc_jit_function_new_block (func, "initial");
+
+  gcc_jit_rvalue *null_ptr
+    = gcc_jit_context_new_rvalue_from_ptr (ctxt, int_ptr_type, NULL);
+
+  /* "*((int *)0) = 42;" */
+  gcc_jit_block_add_assignment (
+    initial, NULL,
+    gcc_jit_rvalue_dereference (null_ptr, NULL),
+    gcc_jit_context_new_rvalue_from_int (ctxt, int_type, 42));
+
+  gcc_jit_block_end_with_void_return (initial, NULL);
+}
+
+void
+verify_code (gcc_jit_context *ctxt, gcc_jit_result *result)
+{
+  typedef void (*fn_type) (void);
+  CHECK_NON_NULL (result);
+  fn_type test_array =
+    (fn_type)gcc_jit_result_get_code (result, "test_trap");
+  CHECK_NON_NULL (test_array);
+  /* Don't attempt to call it.  */
+}