This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] infcmd.c: Don't attempt to record a NULL value after a finish command
- From: Pedro Alves <palves at redhat dot com>
- To: Kevin Buettner <kevinb at redhat dot com>
- Cc: gdb-patches at sourceware dot org
- Date: Wed, 30 Sep 2015 13:21:34 +0100
- Subject: Re: [PATCH] infcmd.c: Don't attempt to record a NULL value after a finish command
- Authentication-results: sourceware.org; auth=none
- References: <20150929214350 dot 44b076c7 at pinnacle dot lan> <560BB90D dot 70905 at redhat dot com> <20150930051736 dot 457e072c at pinnacle dot lan>
On 09/30/2015 01:17 PM, Kevin Buettner wrote:
> On Wed, 30 Sep 2015 11:27:25 +0100
> Pedro Alves <palves@redhat.com> wrote:
>
>> On 09/30/2015 05:43 AM, Kevin Buettner wrote:
>>> Architectures which use RETURN_VALUE_STRUCT_CONVENTION will have a
>>> NULL return value after executing a finish command. See get_return_value()
>>> in infcmd.c.
>>>
>>> This patch avoids an eventual SIGSEV (caused by attempting to
>>> derefrence a NULL pointer) by adding a suitable test to
>>> finish_command_fsm_should_stop().
>>>
>>> I encountered this problem while testing msp430:
>>>
>>> (gdb) PASS: gdb.base/structs.exp: zed L<n> for finish; return 1 structs-tc
>>> finish
>>> Run till exit from #0 fun1 () at /ironwood1/sourceware-git/msp430-elf/../binutils-gdb/gdb/testsuite/gdb.base/structs.c:125
>>> ERROR: Process no longer exists
>>>
>>> gdb/ChangeLog:
>>> * infcmd.c (finish_command_fsm_should_stop): Don't attempt to
>>> record a NULL value.
>>> ---
>>> gdb/infcmd.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/gdb/infcmd.c b/gdb/infcmd.c
>>> index c4d7d8b..6be95e4 100644
>>> --- a/gdb/infcmd.c
>>> +++ b/gdb/infcmd.c
>>> @@ -1788,7 +1788,7 @@ finish_command_fsm_should_stop (struct thread_fsm *self)
>>> internal_error (__FILE__, __LINE__,
>>> _("finish_command: function has no target type"));
>>>
>>> - if (TYPE_CODE (rv->type) != TYPE_CODE_VOID)
>>> + if (rv->value != NULL && TYPE_CODE (rv->type) != TYPE_CODE_VOID)
>>> {
>>> struct value *func;
>>
>> It's this else block that below sets rv->value in the first place:
>>
>> if (TYPE_CODE (rv->type) != TYPE_CODE_VOID)
>> {
>> struct value *func;
>>
>> func = read_var_value (f->function, NULL, get_current_frame ());
>> rv->value = get_return_value (func, rv->type);
>> rv->value_history_index = record_latest_value (rv->value);
>> }
>>
>> So seems like that patch would break !RETURN_VALUE_STRUCT_CONVENTION
>> targets?
>>
>> I assume that it's record_latest_value that crashes? Isn't the right
>> fix instead:
>>
>> rv->value = get_return_value (func, rv->type);
>> - rv->value_history_index = record_latest_value (rv->value);
>> + if (rv->value != NULL)
>> + rv->value_history_index = record_latest_value (rv->value);
>>
>
> You're right. (I honestly don't know what I was thinking.)
>
> Please commit your patch instead.
I don't have a patch -- I wrote that hunk by hand
in the mail client. :-) Please fix up yours and push it.
Thanks,
Pedro Alves