RTEMS5 _REENT

[Sebastian Huber]

Hello Ian,

On 07/11/2018 07:26, Ian Caddy wrote:
> Hi All,
>
> We are migrating a product from RTEMS4.10 to RTEMS5, and I have come 
> up with a problem that we have not experienced previously.
>
> We have some code that works before RTEMS mulit-tasking has been started.

yes, unfortunately some Newlib functions need now a valid executing 
thread. These are functions which use errno or ctype.h methods for 
example. If you use these functions before a executing thread exists, 
then you end up in a NULL pointer access.

>
> This code uses some string functions which are causing the system to 
> crash (bus error out), and I have worked it out to a call to:
>
> _REENT
>
> One example function is:
>
> snprintf
>
> using the newlib 3.0.0 the first few lines of the function are:
>
> int
> snprintf (char *__restrict str,
>        size_t size,
>        const char *__restrict fmt, ...)
> {
>   int ret;
>   va_list ap;
>   FILE f;
>   struct _reent *ptr = _REENT;
>
> The _REENT macro ends up calling:
>
> cpukit/include/rtems/confdefs.h:2273 __getreent
>
> which ends up returning the thread executing structure:
>
>  return _Thread_Get_executing()->libc_reent;
>
> Since this is before multi-tasking, _Thread_Get_executing() returns 
> NULL, and this is not valid memory.
>
> So, if I want to support this, do I need to configure:
>
> #define CONFIGURE_DISABLE_NEWLIB_REENTRANCY
>
> If I do this will it effect things later?

Yes, this is probably not what you want. The 
CONFIGURE_DISABLE_NEWLIB_REENTRANCY disables the per-thread reentrancy 
structure entirely. This is one of the configuration options which 
should be used with special care.

>
> Or should this be fixed to check whether multi-tasking is running 
> before checking the currently running thread, we do this in some of 
> our other code?

Code which runs before the idle threads are created should not use 
complex Newlib functions. Unfortunately snprintf() is a complex Newlib 
function. I think we need a low level replacement for snprintf() which 
could be used during BSP initialization for example. This could be a 
simple wrapper for _IO_Printf().

I had a similar issue here:

https://lists.rtems.org/pipermail/devel/2018-September/023013.html

-- 
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber at embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.