Re: [PATCH] UML: don't discard .text.exit section

[Dmitry Vyukov]

On Thu, Aug 18, 2016 at 3:08 AM, Andrey Ryabinin
<aryabinin@xxxxxxxxxxxxx> wrote:
>
>
> On 08/17/2016 08:11 PM, Dmitry Vyukov wrote:
>> On Wed, Aug 17, 2016 at 8:10 AM, Andrey Ryabinin
>> <aryabinin@xxxxxxxxxxxxx> wrote:
>>> Commit e41f501d3912 ("vmlinux.lds: account for destructor sections")
>>> added '.text.exit' to EXIT_TEXT which is discarded at link time by default.
>>> This breaks compilation of UML:
>>>      `.text.exit' referenced in section `.fini_array' of
>>>      /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libc.a(sdlerror.o):
>>>      defined in discarded section `.text.exit' of
>>>      /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libc.a(sdlerror.o)
>>>
>>> Apparently UML doesn't want to discard exit text, so let's place all EXIT_TEXT
>>> sections in .exit.text.
>>>
>>> Fixes: e41f501d3912 ("vmlinux.lds: account for destructor sections")
>>> Reported-by: Stefan Traby <stefan@xxxxxxxxxxxxxxxxx>
>>> Signed-off-by: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
>>> Cc: <stable@xxxxxxxxxxxxxxx> # 4.0+
>>> ---
>>>  arch/um/include/asm/common.lds.S | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/arch/um/include/asm/common.lds.S b/arch/um/include/asm/common.lds.S
>>> index 1dd5bd8..1330553 100644
>>> --- a/arch/um/include/asm/common.lds.S
>>> +++ b/arch/um/include/asm/common.lds.S
>>> @@ -81,7 +81,7 @@
>>>    .altinstr_replacement : { *(.altinstr_replacement) }
>>>    /* .exit.text is discard at runtime, not link time, to deal with references
>>>       from .altinstructions and .eh_frame */
>>> -  .exit.text : { *(.exit.text) }
>>> +  .exit.text : { EXIT_TEXT }
>>>    .exit.data : { *(.exit.data) }
>>>
>>>    .preinit_array : {
>>> --
>>> 2.7.3
>>>
>>
>>
>> Sorry for delays, I am travelling.
>> Do we need ".fini_array" section? It's also destructors that we don't
>> run. Or does UML use them? Does discarding ".fini_array" help?
>>
>
> libc has desctructors and use them for whatever purpose it needs.


Does UML actually gracefully exit running global destructors? That
would also require gracefully shutting down all threads/cpus. Doesn't
it just _exit (or syscall(SYS_exit_group))?