public inbox for jit@gcc.gnu.org
 help / color / mirror / Atom feed
From: Vladimir Makarov <vmakarov@redhat.com>
To: David Malcolm <dmalcolm@redhat.com>
Cc: Andrea Corallo <andrea.corallo@arm.com>,
	gcc-patches@gcc.gnu.org, jit@gcc.gnu.org, nd <nd@arm.com>
Subject: Re: [PATCH] lra: set insn_code_data to NULL when freeing
Date: Mon, 30 Mar 2020 18:51:50 -0400	[thread overview]
Message-ID: <adac7764-7bd6-02e6-a2ed-7b5460f2cbf7@redhat.com> (raw)
In-Reply-To: <20200330160608.10383-1-dmalcolm@redhat.com>


On 2020-03-30 12:06 p.m., David Malcolm wrote:
> It's a double-free bug in lra.c, albeit one that requires being used
> in a multithreaded way from libgccjit to be triggered.
>
> libgccjit's test-threads.c repeatedly compiles and runs numerous tests,
> each in a separate thread.
>
> Attempting to add an empty test that generates no code leads to a
> double-free ICE within that thread, within lra.c's
> finish_insn_code_data_once.
>
> The root cause is that the insn_code_data array is cleared in
> init_insn_code_data_once, but this is only called the first time
> a cgraph_node is expanded [1], whereas the "loop-over-all-elements
> and free them" is unconditionally called in finalize [2].  Hence
> if there are no functions:
> * the array is not re-initialized for the empty context
> * when finish_insn_code_data_once is called for the empty context
> it still contains the freed pointers from the previous context
> that held the jit mutex, and hence the free is a double-free.
>
> This patch sets the pointers to NULL after freeing them, fixing
> the ICE.  The calls to free are still guarded by a check for NULL,
> which is redundant, but maybe there's a reason for not wanting to
> call "free" on a possibly-NULL value many times on process exit?
> (it makes the diff cleaner, at least)
>
> Fixes the issue in jit.dg.
>
> Full bootstrap & regression test in progress.
>
> Is it OK for master if it passes?

Sure, David.  Thank you for the patch.

> gcc/ChangeLog:
> 	* lra.c (finish_insn_code_data_once): Set the array elements
> 	to NULL after freeing them.
>
> gcc/testsuite/ChangeLog:
> 	* jit.dg/all-non-failing-tests.h: Add test-empty.c
> ---
>   gcc/lra.c                                    |  5 ++++-
>   gcc/testsuite/jit.dg/all-non-failing-tests.h | 10 ++++++++++
>   2 files changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/lra.c b/gcc/lra.c
> index d5ea3622686..5e8b75b1fda 100644
> --- a/gcc/lra.c
> +++ b/gcc/lra.c
> @@ -653,7 +653,10 @@ finish_insn_code_data_once (void)
>     for (unsigned int i = 0; i < NUM_INSN_CODES; i++)
>       {
>         if (insn_code_data[i] != NULL)
> -	free (insn_code_data[i]);
> +	{
> +	  free (insn_code_data[i]);
> +	  insn_code_data[i] = NULL;
> +	}
>       }
>   }
>   
> diff --git a/gcc/testsuite/jit.dg/all-non-failing-tests.h b/gcc/testsuite/jit.dg/all-non-failing-tests.h
> index b2acc74ae95..af744192a73 100644
> --- a/gcc/testsuite/jit.dg/all-non-failing-tests.h
> +++ b/gcc/testsuite/jit.dg/all-non-failing-tests.h
> @@ -116,6 +116,13 @@
>   #undef create_code
>   #undef verify_code
>   
> +/* test-empty.c */
> +#define create_code create_code_empty
> +#define verify_code verify_code_empty
> +#include "test-empty.c"
> +#undef create_code
> +#undef verify_code
> +
>   /* test-error-*.c: We don't use these test cases, since they deliberately
>      introduce errors, which we don't want here.  */
>   
> @@ -328,6 +335,9 @@ const struct testcase testcases[] = {
>     {"expressions",
>      create_code_expressions,
>      verify_code_expressions},
> +  {"empty",
> +   create_code_empty,
> +   verify_code_empty},
>     {"factorial",
>      create_code_factorial,
>      verify_code_factorial},


      reply	other threads:[~2020-03-30 22:51 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-30 16:06 David Malcolm
2020-03-30 22:51 ` Vladimir Makarov [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=adac7764-7bd6-02e6-a2ed-7b5460f2cbf7@redhat.com \
    --to=vmakarov@redhat.com \
    --cc=andrea.corallo@arm.com \
    --cc=dmalcolm@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=jit@gcc.gnu.org \
    --cc=nd@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).