This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix IPA crash in libgccjit
- From: David Malcolm <dmalcolm at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: jit at gcc dot gnu dot org, David Malcolm <dmalcolm at redhat dot com>
- Date: Wed, 20 Jun 2018 19:50:19 -0400
- Subject: [PATCH] Fix IPA crash in libgccjit
- References: <1b1d2bd3-3536-5525-1660-09a7b18a9ecd@gmail.com>
All/most of the jit.dg testcases are segfaulting on cleanup of
the 2nd in-process iteration:
PATH=.:$PATH LD_LIBRARY_PATH=. LIBRARY_PATH=. \
gdb --args \
testsuite/jit/test-factorial.c.exe
Starting program: /home/david/coding-3/gcc-git-static-analysis/build/gcc/testsuite/jit/test-factorial.c.exe
PASSED: test-factorial.c.exe iteration 1 of 5: set_up_logging: logfile is non-null
NOTE: test-factorial.c.exe iteration 1 of 5: writing reproducer to /home/david/coding-3/gcc-git-static-analysis/build/gcc/testsuite/jit/test-factorial.c.exe.reproducer.c
Detaching after fork from child process 35787.
Detaching after fork from child process 35789.
PASSED: test-factorial.c.exe iteration 1 of 5: verify_code: result is non-null
PASSED: test-factorial.c.exe iteration 1 of 5: verify_code: my_factorial is non-null
NOTE: my_factorial returned: 3628800
PASSED: test-factorial.c.exe iteration 1 of 5: verify_code: actual: val == expected: 3628800
PASSED: test-factorial.c.exe iteration 2 of 5: set_up_logging: logfile is non-null
NOTE: test-factorial.c.exe iteration 2 of 5: writing reproducer to /home/david/coding-3/gcc-git-static-analysis/build/gcc/testsuite/jit/test-factorial.c.exe.reproducer.c
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff71abc75 in ipcp_driver () at ../../src/gcc/ipa-cp.c:5091
5091 delete edge_clone_summaries;
This appears to be due to recent(?) IPA changes that appear to assume
that the IPA code is only initialized and cleaned up once.
This patch fixes the crashes:
Changes to jit.sum
------------------
FAIL: 65->0 (-65)
PASS: 3186->10290 (+7104)
UNRESOLVED: 1->0 (-1)
Successfully bootstrapped & regrtested on x86_64-pc-linux-gnu.
OK for trunk?
gcc/ChangeLog:
* ipa-cp.c (ipcp_driver): Set edge_clone_summaries to NULL after
deleting it.
* ipa-reference.c (ipa_reference_c_finalize): Delete
ipa_ref_opt_sum_summaries and set it to NULL.
---
gcc/ipa-cp.c | 1 +
gcc/ipa-reference.c | 6 ++++++
2 files changed, 7 insertions(+)
diff --git a/gcc/ipa-cp.c b/gcc/ipa-cp.c
index c192e84..42dd4cc 100644
--- a/gcc/ipa-cp.c
+++ b/gcc/ipa-cp.c
@@ -5089,6 +5089,7 @@ ipcp_driver (void)
/* Free all IPCP structures. */
free_toporder_info (&topo);
delete edge_clone_summaries;
+ edge_clone_summaries = NULL;
ipa_free_all_structures_after_ipa_cp ();
if (dump_file)
fprintf (dump_file, "\nIPA constant propagation end\n");
diff --git a/gcc/ipa-reference.c b/gcc/ipa-reference.c
index 9a9e94c..43bbdae 100644
--- a/gcc/ipa-reference.c
+++ b/gcc/ipa-reference.c
@@ -1230,6 +1230,12 @@ make_pass_ipa_reference (gcc::context *ctxt)
void
ipa_reference_c_finalize (void)
{
+ if (ipa_ref_opt_sum_summaries != NULL)
+ {
+ delete ipa_ref_opt_sum_summaries;
+ ipa_ref_opt_sum_summaries = NULL;
+ }
+
if (ipa_init_p)
{
bitmap_obstack_release (&optimization_summary_obstack);
--
1.8.5.3