This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix part of PR58712


On Tue, Oct 15, 2013 at 12:31 PM, Markus Trippelsdorf
<markus@trippelsdorf.de> wrote:
> Valgrind complains:
> ==27870== Conditional jump or move depends on uninitialised value(s)
> ==27870==    at 0x557CDC: cgraph_create_edge_1(cgraph_node*, cgraph_node*, gimple_statement_d*, long, int) (cgraph.c:695)
> ==27870==    by 0x55882E: cgraph_create_edge(cgraph_node*, cgraph_node*, gimple_statement_d*, long, int) (cgraph.c:890)
> ==27870==    by 0x560891: cgraph_clone_edge(cgraph_edge*, cgraph_node*, gimple_statement_d*, unsigned int, long, int, bool) (cgraphclones.c:135)
> ==27870==    by 0x7F1F14: copy_body(copy_body_data*, long, int, basic_block_def*, basic_block_def*, basic_block_def*) (tree-inline.c:1741)
> ...
> ==27870==  Uninitialised value was created by a client request
> ==27870==    at 0x50BBEE: ggc_internal_alloc_stat(unsigned long) (ggc-page.c:1339)
> ==27870==    by 0x557D92: cgraph_create_edge_1(cgraph_node*, cgraph_node*, gimple_statement_d*, long, int) (cgraph.c:842)
> ==27870==    by 0x55882E: cgraph_create_edge(cgraph_node*, cgraph_node*, gimple_statement_d*, long, int) (cgraph.c:890)
> ==27870==    by 0x560891: cgraph_clone_edge(cgraph_edge*, cgraph_node*, gimple_statement_d*, unsigned int, long, int, bool) (cgraphclones.c:135)
> ,,,
>
> This happens because e->indirect_unknown_callee may be uninitialized in
> cgraph_add_edge_to_call_site_hash. Fixed by initializing it earlier.
>
> LTO bootstrapped and tested on x86_64-unknown-linux-gnu.
>
> Please apply, if this looks reasonable.

As we also have

struct cgraph_edge *
cgraph_create_indirect_edge (struct cgraph_node *caller, gimple call_stmt,
                             int ecf_flags,
                             gcov_type count, int freq)
{
  struct cgraph_edge *edge = cgraph_create_edge_1 (caller, NULL, call_stmt,
                                                   count, freq);
  tree target;

  edge->indirect_unknown_callee = 1;

I'd rather change the cgraph_create_edge_1 interface to get an additional
argument (the value to use for indirect_unknown_callee).  Or maybe
we can statically compute it from the current arguments already?

Honza?

Thanks,
Richard.

> Thanks.
>
> 2013-10-15  Markus Trippelsdorf  <markus@trippelsdorf.de>
>
>         PR ipa/58712
>         * cgraph.c (cgraph_create_edge_1): Initialize
>         indirect_unknown_callee earlier.
>         (cgraph_create_edge): Likewise.
>
> diff --git a/gcc/cgraph.c b/gcc/cgraph.c
> index 124ee0adf855..c5c4e13ba145 100644
> --- a/gcc/cgraph.c
> +++ b/gcc/cgraph.c
> @@ -874,6 +874,7 @@ cgraph_create_edge_1 (struct cgraph_node *caller, struct cgraph_node *callee,
>    edge->indirect_info = NULL;
>    edge->indirect_inlining_edge = 0;
>    edge->speculative = false;
> +  edge->indirect_unknown_callee = 0;
>    if (call_stmt && caller->call_site_hash)
>      cgraph_add_edge_to_call_site_hash (edge);
>
> @@ -889,7 +890,6 @@ cgraph_create_edge (struct cgraph_node *caller, struct cgraph_node *callee,
>    struct cgraph_edge *edge = cgraph_create_edge_1 (caller, callee, call_stmt,
>                                                    count, freq);
>
> -  edge->indirect_unknown_callee = 0;
>    initialize_inline_failed (edge);
>
>    edge->next_caller = callee->callers;
> --
> Markus


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]