Subversion Repositories openrisc

#
#  $Id: PROBLEMS,v 1.2 2001-09-27 11:59:11 chris Exp $
#

This is the list of outstanding problems in this release.

+   AMD 29k port is based on a non-GNU toolset.
 
+   The test spfatal is out of date and as a result will NOT execute
    correctly.  The addition of POSIX and consequent ongoing initialization
    reorganization makes it pointless to fix this until the POSIX support
    is completely in place.

+   The m68k family has become quite large and an understanding of the 
    compatibility of the peripherals on the various members of the 683xx 
    family would allow someone to designate some of the drivers submitted
    for the gen683xx BSPs as useful on other members.

+   The only supported i960 family member is the CA.  No support for the
    floating point support found in other family members is present.
    This also implies that RTEMS may "think" of something as generic
    across the i960 family when in fact it is specific to the CA.
    To make matters worse, the i960 target board owned by the RTEMS Project
    is now broken and as a result even the i960CA is a "compile only" port.

+   Some of the BSPs still define RAM_START and RAM_END in the bsp.h file.
    It is better to define these in the linkcmds file.  It is also nice
    to use the linkcmds file to place overlays for on-board hardware.

+   Not all of the BSP console drivers have been converted to termios.
    Look at the m68k/gen68360, sparc/erc32, and powerpc/psim BSPs for
    examples.

+   UNIX port notes:

       + sometimes a stray SIGALRM is reported as spfatal completes.

       + There are conflicts between the names of native library routines
         which MUST be used and those in the POSIX support.  This must
         be addressed.  The POSIX API cannot be used with this port as a
         result of this.

       + Someone suggested writing a mini-system call interface to
         include with RTEMS which would eliminate name conflicts.  This
         would allow the RTEMS POSIX API to be tested in this configuration.

+   Some of the tests may execute correctly and not produce the exact
    ordering of lines in the screen file.  This appears to be a combination
    of a number of factors including buffering, processor speed, IO
    device overhead, and clock interrupt rate.  The biggest problem is that
    some tests depend on polled IO with no unexpected context switches.
    These may not be resolvable while maintaining the spirit of the test.

+   The clock device drivers should really avoid doing the division
    by 1000 in the clock tick ISR to convert microseconds into
    milliseconds.  This only applies to clock drivers which generate
    an ISR each millisecond and only call rtems_clock_tick every
    so many ISRs.

+  Cross-check configure --enable-* flags.
   + warn/refuse to configure when --enable-libcdir and
     --enable-gcc28 are given.
   + force --enable-libcdir when --disable-gcc28 is given

+  make profile does not currently work for a variety of reasons.  Few
   BSPs include profile versions of the libraries in their bsp_specs
   file.  There is no mechanism to sample data for gperf to process.
   All of this will need to be addressed before "make profile" is truly
   useful.

+  Bare BSP does not compile for all configurations yet.  This is 
   primarily due to libcpu support code assuming that the BSP has
   made something available which is not with a bare BSP.