[PATCH] Re: Queries concerning RTOS protection in newllib

[Antony KING]

Hi yet again,

Please find attached some additional patches to newlib which hopefully 
fix the following problems I have discovered when using newlib built 
with non-stub versions of the locking API (in sys/lock.h) and in a 
multi-threaded environment:

* Missing initialisation of the _lock field in FILE objects declared 
locally on the stack in the implementations of the sprintf/sscanf class 
of functions.

Only files in libc/stdio affected: asprintf.c, siprintf.c, snprintf.c, 
sprintf.c, sscanf.c, vasprintf.c, vsnprintf.c, vsprintf.c, vsscanf.c.

* _GLOBAL_REENT structure "walked" (_fwalk) multiple times when cleaning 
up the FILE objects (via the _cleanup_r function) in a multi-threading 
environment where _reclaim_reent is called on the death (and deletion) 
of each task. This is a problem now that the function being used to walk 
the FILE objects is fclose and not as previously fflush. As a result all 
open FILE objects are closed on the death of the first task, which is 
not very clever :-(. Solution is to remove the additional walking of 
_GLOBAL_REENT in _fwalk/_fwalk_reent until explicitly invoked via the 
call to _cleanup_r of _GLOBAL_REENT in the exit function.

File affected: libc/stdio/fwalk.c

* I have also removed, what I believe to be, an out of date check for 
_REENT_ONLY in libc/stdlib/exit.c since there does not exist anywhere in 
the newlib sources implementations of _exit_r.

Please note that this patch assumes a previous patch I have submitted 
(http://sources.redhat.com/ml/newlib/2004/msg00293.html) has been 
applied (really only the change to newlib/ChangeLog).

Cheers,

Antony.

Antony King wrote:
> Hi again,
> 
> Please find attached a patch for some changes which were overlooked in 
> my original patch; tightening up the definition of the lock macros in 
> libc/include/sys/stdio.h (missing parenthesis around macro arguments) 
> and a missed file (libc/stdio64/freopen64.c) in the removal the fp lock 
> casting. Hope thats it for the time being.
> 
> Thanks,
> 
> Antony.
> 
> Jeff Johnston wrote:
> 
>> Thanks Antony.  Patch checked in.  A patch was also required to get 
>> x86-linux building with these changes so I added that as well.  In the 
>> future, it would be helpful to me if you could add a ChangeLog entry 
>> for any changes you make.
>>
>> -- Jeff J.

-- 
-----------------------------------------------------------------
Antony King                        | Email: antony.king@st.com
STMicroelectronics (R&D) Ltd.      |
Bristol, BS32 4SQ. United Kingdom. | Tel: +44 (0)1454 462646
-------------- next part --------------
A non-text attachment was scrubbed...
Name: newlib.patch.gz
Type: application/x-gzip
Size: 1691 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/newlib/attachments/20040616/c26bd8e0/attachment.bin>