Re: EPICS/RTEMS bug in osdPoolStatus.c

[Till Straumann <strauman@slac.stanford.edu>]

Kate Feng wrote:
> EPICS  OSI Release : 3.14.x   ( at least 3.14.6 and up )
> File : src/libCom/osi/os/RTEMS/osdPoolStatus.c
>
> Problem :
> "return (malloc_free_space() > 50000 + contiguousBlockSize);"
> will falsely return true when RTEMS function call
> malloc_free_space() returns -1, which actually indicates
> an internal RTEMS error such as corrupted heap,or illegal
> region id etc.  Since neither EPICS nor RTEMS prints out
> the error condition, this kind of error could be hard to be
> debugged.
>   
In most cases heap corruption results due to programming
errors (rather than internal errors) which emphasizes
Kate's point. osdPoolStatus should check
the return state of malloc_free_space(). However,
I'd suggest more drastic measures than just returning 0
such as printing a console message and aborting the
application.

Anyways, under RTEMS 4.8 the implementation has
changed and AFAIK malloc_free_space() can no longer
return -1.

T.
> FIX : src/libCom/osi/os/RTEMS/osdPoolStatus.c
> return ( ((int) malloc_free_space()) > 50000 + contiguousBlockSize);
>
> Or I actually preferred  something like :
>
> /*
>  * osiSufficentSpaceInPool ()
>  */
> epicsShareFunc int epicsShareAPI osiSufficentSpaceInPool ( size_t
> contiguousBlockSize )
> {
>   int free;
>
>   if ( (free=malloc_free_space())==-1)
>      return(0);
>   if ( free < 100000 + contiguousBlockSize) {
>     printf("Largest free space available : 0x%x,  requested
> contiguousBlockSize : 0x%x\n",
>     free, contiguousBlockSize);
>     return(0);
>   }
>   return(1);
> }
>
> Besides, RTEMS should print out what the internal error message is.
>
> Regards,
> Kate
>
>