[PATCH 4/5 v2] SPARC: Fatal_halt use source and exit codes

Daniel Hellstrom daniel at gaisler.com
Fri Jul 4 20:07:03 UTC 2014


On 07/04/2014 04:20 PM, Joel Sherrill wrote:
>
>
> On Jul 4, 2014 3:47 AM, Daniel Hellstrom <daniel at gaisler.com> wrote:
> >
> > On 07/03/2014 04:55 PM, Joel Sherrill wrote:
> > > I am wondering if I don't have a vision for what this
> > > series of patches is trying to accomplish in whole.
> > * make exit-code propagate to debugger/hypervisor. I though this 
> would be nice for other architectures as well.
> > * make SMP fatal exit work
> > * instead of using BSP specific reset on LEON3, rely on RTEMS and 
> SPARC to stop execution
> > * fix bad behaviour of SPARC fatal halt, always spinning.
> >
> > >
> > > And I am hung up on not wanting BSP references in
> > > score. So I would like an elegant solution.
> > I understand that, however some part of me still thinks that halting 
> is very BSP specific. What makes me a bit confused is that there is 
> just one halt approach, there is no difference between reset,
> > reboot, halt and power-down. I remember Win95 or Linux, when you 
> type halt it does not reboot or power-down the machine, since you can 
> see the last terminal shutdown message and push the power-down
> > button yourself ;)
> >
> > > How does tweaking the bsp_reset() method not
> > > address all of this? It is invoked along the _Terminate()
> > > path already.
> > >
> > > As an aside, should the _Internal_errors_What_happened
> > > structure be filled in earlier in _Terminate? How about
> > > just after disabling interrupts?
> > That is one of the problems I faced.
> >   * bsp_reset() does not have any arguments of what gone wrong
> >   * _Internal_errors_What_happened is not modified until after 
> bsp_reset()
> >   * _Internal_errors_What_happened can not be updated before 
> bsp_reset() on SMP since it will be written by all CPUs
> >   * The exit codes of CPUs are different. If one CPU finds a 
> problem, signals to the others to shut down, their IPI handler will 
> call fatal handler with another exit code. Now the BSP in the fatal
> > extions can look at the exit codes to determine which CPU requested 
> the fatal exit and handle them differently. The BSP makes sure that 
> only one CPU ends up in the
> >
>
> Would it help if bsp_reset() took arguments?
>
No, I have implemented a bsp_fatal_handler instead with patch 5/5, that 
has both arguments. The same basic problem exists.

> I see the problem now.
>
I'm not sure how to proceed. Do you want me to update the series to:
   A. revert changes to the arguments to CPU_Fatal_halt, or
   B. remove all arguments to CPU_Fatal_halt() in all archs, and use 
_Internal_errors_What_happened from CPU_Fatal_halt to get the currently 
required information by the archs.

Then I could make SPARC BSPs totally similar by removing bsp_reset() in 
LEON2 and ERC32. And remove the now unused BSP_fatal_exit introduced.

Thanks,
DanielH


> > >
> > > On 7/3/2014 2:29 AM, Daniel Hellstrom wrote:
> > >> PATCHv2: BSP_fatal_halt renamed to _BSP_Fatal_halt
> > >>
> > >> The Fatal_halt handler now have two options, either halt
> > >> as before or enter system error state to return to
> > >> debugger or simulator. The exit-code is now also
> > >> propagated to the debugger which is very useful for
> > >> testing.
> > >>
> > >> The CPU_Fatal_halt handler was split up into two, since
> > >> the only the LEON3 support the CPU power down.
> > >>
> > >> The LEON3 halt now uses the power-down instruction to save
> > >> CPU power. This doesn't stop a potential watch-dog timer
> > >> from expiring.
> > >> ---
> > >>   c/src/lib/libbsp/sparc/erc32/Makefile.am           |    1 +
> > >>   c/src/lib/libbsp/sparc/erc32/configure.ac          |    5 +++
> > >>   c/src/lib/libbsp/sparc/leon2/Makefile.am           |    1 +
> > >>   c/src/lib/libbsp/sparc/leon2/configure.ac          |    5 +++
> > >>   c/src/lib/libbsp/sparc/leon3/Makefile.am           |    1 +
> > >>   c/src/lib/libbsp/sparc/leon3/configure.ac          |    5 +++
> > >>   .../libbsp/sparc/leon3/startup/bsp_fatal_halt.c    |   36 
> ++++++++++++++++++
> > >>   .../libbsp/sparc/shared/startup/bsp_fatal_halt.c   |   38 
> ++++++++++++++++++++
> > >>   cpukit/score/cpu/sparc/rtems/score/cpu.h           |   12 ++----
> > >>   9 files changed, 96 insertions(+), 8 deletions(-)
> > >>   create mode 100644 
> c/src/lib/libbsp/sparc/leon3/startup/bsp_fatal_halt.c
> > >>   create mode 100644 
> c/src/lib/libbsp/sparc/shared/startup/bsp_fatal_halt.c
> > >>
> > >> diff --git a/c/src/lib/libbsp/sparc/erc32/Makefile.am 
> b/c/src/lib/libbsp/sparc/erc32/Makefile.am
> > >> index b9f1c53..00e398b 100644
> > >> --- a/c/src/lib/libbsp/sparc/erc32/Makefile.am
> > >> +++ b/c/src/lib/libbsp/sparc/erc32/Makefile.am
> > >> @@ -46,6 +46,7 @@ libbsp_a_SOURCES += startup/bspidle.c
> > >>   libbsp_a_SOURCES += startup/bspdelay.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/startup/early_malloc.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/startup/bsp_fatal_exit.c
> > >> +libbsp_a_SOURCES += ../../sparc/shared/startup/bsp_fatal_halt.c
> > >>   # ISR Handler
> > >>   libbsp_a_SOURCES += ../../sparc/shared/cpu.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/irq_asm.S
> > >> diff --git a/c/src/lib/libbsp/sparc/erc32/configure.ac 
> b/c/src/lib/libbsp/sparc/erc32/configure.ac
> > >> index 4cbfb7e..bb5174a 100644
> > >> --- a/c/src/lib/libbsp/sparc/erc32/configure.ac
> > >> +++ b/c/src/lib/libbsp/sparc/erc32/configure.ac
> > >> @@ -43,6 +43,11 @@ RTEMS_BSPOPTS_HELP([ENABLE_SIS_QUIRKS],
> > >>    BSP will be enabled.  In particular, SIS requires special
> > >>    initialization not used on real ERC32 hardware.])
> > >>
> > >> +RTEMS_BSPOPTS_SET([BSP_HALT_AT_FATAL_EXIT],[*],[])
> > >> +RTEMS_BSPOPTS_HELP([BSP_HALT_AT_FATAL_EXIT],
> > >> +[If defined, CPU is spinning on fatal exit. Otherwise generate 
> system
> > >> + error which will hand over to debugger, simulator, etc.])
> > >> +
> > > Why would anyone want the CPU to go into a spin?
> > This was the previous behaviour of SPARC CPU_fatal_halt(), the only 
> reason for that as I could come up with would be that the watchdog 
> timer expires and causes a hard reset. I wanted to preserve the
> > current behaviour, I change quite much in the patch series anyway.
> >
> > >>   RTEMS_BSP_CLEANUP_OPTIONS(0,1,1)
> > >>
> > >>   # Explicitly list all Makefiles here
> > >> diff --git a/c/src/lib/libbsp/sparc/leon2/Makefile.am 
> b/c/src/lib/libbsp/sparc/leon2/Makefile.am
> > >> index 3f8edf4..ac6f327 100644
> > >> --- a/c/src/lib/libbsp/sparc/leon2/Makefile.am
> > >> +++ b/c/src/lib/libbsp/sparc/leon2/Makefile.am
> > >> @@ -61,6 +61,7 @@ libbsp_a_SOURCES += ../../shared/bspinit.c
> > >>   libbsp_a_SOURCES += startup/bspdelay.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/startup/early_malloc.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/startup/bsp_fatal_exit.c
> > >> +libbsp_a_SOURCES += ../../sparc/shared/startup/bsp_fatal_halt.c
> > >>
> > >>   # ISR Handler
> > >>   libbsp_a_SOURCES += ../../sparc/shared/cpu.c
> > >> diff --git a/c/src/lib/libbsp/sparc/leon2/configure.ac 
> b/c/src/lib/libbsp/sparc/leon2/configure.ac
> > >> index 610a049..1975a10 100644
> > >> --- a/c/src/lib/libbsp/sparc/leon2/configure.ac
> > >> +++ b/c/src/lib/libbsp/sparc/leon2/configure.ac
> > >> @@ -36,6 +36,11 @@ RTEMS_BSPOPTS_HELP([SIMSPARC_FAST_IDLE],
> > >>    time spent in the idle task is minimized.  This significantly 
> reduces
> > >>    the wall time required to execute the RTEMS test suites.])
> > >>
> > >> +RTEMS_BSPOPTS_SET([BSP_HALT_AT_FATAL_EXIT],[*],[])
> > >> +RTEMS_BSPOPTS_HELP([BSP_HALT_AT_FATAL_EXIT],
> > >> +[If defined, CPU is spinning on fatal exit. Otherwise generate 
> system
> > >> + error which will hand over to debugger, simulator, etc.])
> > >> +
> > >>   RTEMS_BSP_CLEANUP_OPTIONS(0, 1, 1)
> > >>
> > >>   # Explicitly list all Makefiles here
> > >> diff --git a/c/src/lib/libbsp/sparc/leon3/Makefile.am 
> b/c/src/lib/libbsp/sparc/leon3/Makefile.am
> > >> index 54c9f22..d1f07a7 100644
> > >> --- a/c/src/lib/libbsp/sparc/leon3/Makefile.am
> > >> +++ b/c/src/lib/libbsp/sparc/leon3/Makefile.am
> > >> @@ -42,6 +42,7 @@ libbsp_a_SOURCES += ../../shared/bspclean.c 
> ../../shared/bsplibc.c \
> > >>   libbsp_a_SOURCES += startup/bspreset.c
> > >>   libbsp_a_SOURCES += startup/cpucounter.c
> > >>   libbsp_a_SOURCES += ../../sparc/shared/startup/bsp_fatal_exit.c
> > >> +libbsp_a_SOURCES += startup/bsp_fatal_halt.c
> > >>
> > >>   # ISR Handler
> > >>   libbsp_a_SOURCES += ../../sparc/shared/cpu.c
> > >> diff --git a/c/src/lib/libbsp/sparc/leon3/configure.ac 
> b/c/src/lib/libbsp/sparc/leon3/configure.ac
> > >> index 3fe1e48..206a662 100644
> > >> --- a/c/src/lib/libbsp/sparc/leon3/configure.ac
> > >> +++ b/c/src/lib/libbsp/sparc/leon3/configure.ac
> > >> @@ -38,6 +38,11 @@ RTEMS_BSPOPTS_HELP([BSP_LEON3_SMP],
> > >>   [Always defined when on a LEON3 to enable the LEON3 support for
> > >>    determining the CPU core number in an SMP configuration.])
> > >>
> > >> +RTEMS_BSPOPTS_SET([BSP_HALT_AT_FATAL_EXIT],[*],[])
> > >> +RTEMS_BSPOPTS_HELP([BSP_HALT_AT_FATAL_EXIT],
> > >> +[If defined, CPU is powered down on fatal exit. Otherwise 
> generate system
> > >> + error which will hand over to debugger, simulator, etc.])
> > >> +
> > >>   RTEMS_BSP_CLEANUP_OPTIONS(0, 1, 1)
> > >>   RTEMS_BSP_LINKCMDS
> > >>
> > >> diff --git 
> a/c/src/lib/libbsp/sparc/leon3/startup/bsp_fatal_halt.c 
> b/c/src/lib/libbsp/sparc/leon3/startup/bsp_fatal_halt.c
> > >> new file mode 100644
> > >> index 0000000..4350faf
> > >> --- /dev/null
> > >> +++ b/c/src/lib/libbsp/sparc/leon3/startup/bsp_fatal_halt.c
> > >> @@ -0,0 +1,36 @@
> > >> +/**
> > >> + * @file
> > >> + * @ingroup sparc_leon3
> > >> + * @brief LEON3 BSP Fatal_halt handler.
> > >> + *
> > >> + *  COPYRIGHT (c) 2014.
> > >> + *  Aeroflex Gaisler AB.
> > >> + *
> > >> + *  The license and distribution terms for this file may be
> > >> + *  found in the file LICENSE in this distribution or at
> > >> + *  http://www.rtems.org/license/LICENSE.
> > >> + */
> > >> +
> > >> +#include <leon.h>
> > >> +#include <rtems/score/sparc.h>
> > >> +
> > >> +#ifdef BSP_HALT_AT_FATAL_EXIT
> > >> +
> > >> +/* Power down LEON CPU on fatal error exit */
> > >> +void _BSP_Fatal_halt(uint32_t source, uint32_t error)
> > >> +{
> > >> +  sparc_disable_interrupts();
> > >> +  leon3_power_down_loop();
> > >> +}
> > >> +
> > >> +#else
> > >> +
> > >> +/* return to debugger, simulator, hypervisor or similar by exiting
> > >> + * with an error code. g1=1, g2=FATAL_SOURCE, G3=error-code.
> > >> + */
> > >> +void _BSP_Fatal_halt(uint32_t source, uint32_t error)
> > >> +{
> > >> +  sparc_syscall_exit(source, error);
> > >> +}
> > >> +
> > >> +#endif
> > >> diff --git 
> a/c/src/lib/libbsp/sparc/shared/startup/bsp_fatal_halt.c 
> b/c/src/lib/libbsp/sparc/shared/startup/bsp_fatal_halt.c
> > >> new file mode 100644
> > >> index 0000000..349db20
> > >> --- /dev/null
> > >> +++ b/c/src/lib/libbsp/sparc/shared/startup/bsp_fatal_halt.c
> > >> @@ -0,0 +1,38 @@
> > >> +/**
> > >> + * @file
> > >> + * @ingroup sparc_bsp
> > >> + * @brief ERC32/LEON2 BSP Fatal_halt handler.
> > >> + *
> > >> + *  COPYRIGHT (c) 2014.
> > >> + *  Aeroflex Gaisler AB.
> > >> + *
> > >> + *  The license and distribution terms for this file may be
> > >> + *  found in the file LICENSE in this distribution or at
> > >> + *  http://www.rtems.org/license/LICENSE.
> > >> + */
> > >> +
> > >> +#include <rtems/score/sparc.h>
> > >> +
> > >> +#ifdef BSP_HALT_AT_FATAL_EXIT
> > >> +
> > >> +/* Spin CPU on fatal error exit */
> > >> +void _BSP_Fatal_halt(uint32_t source, uint32_t error)
> > >> +{
> > >> +  uint32_t level = sparc_disable_interrupts();
> > >> +
> > >> +  __asm__ volatile ( "mov  %0, %%g1 " : "=r" (level) : "0" 
> (level) );
> > >> +
> > >> +  while (1) ; /* loop forever */
> > >> +}
> > >> +
> > >> +#else
> > >> +
> > >> +/* return to debugger, simulator, hypervisor or similar by exiting
> > >> + * with an error code. g1=1, g2=FATAL_SOURCE, G3=error-code.
> > >> + */
> > >> +void _BSP_Fatal_halt(uint32_t source, uint32_t error)
> > >> +{
> > >> +  sparc_syscall_exit(source, error);
> > >> +}
> > >> +
> > >> +#endif
> > >> diff --git a/cpukit/score/cpu/sparc/rtems/score/cpu.h 
> b/cpukit/score/cpu/sparc/rtems/score/cpu.h
> > >> index 58c843a..0bb1bf5 100644
> > >> --- a/cpukit/score/cpu/sparc/rtems/score/cpu.h
> > >> +++ b/cpukit/score/cpu/sparc/rtems/score/cpu.h
> > >> @@ -1075,19 +1075,15 @@ void _CPU_Context_Initialize(
> > >>
> > >>   /* Fatal Error manager macros */
> > >>
> > >> +extern void _BSP_Fatal_halt(uint32_t source, uint32_t error)
> > >> +  RTEMS_COMPILER_NO_RETURN_ATTRIBUTE;
> > >> +
> > > I am really having trouble invoking a BSP specific routine
> > > out of score/cpu. Can't you do this in the bsp_reset()
> > > method?
> > No, the debugger/hypervisor typically reads registers as exit-codes 
> not memory so it has to be done last. I'm not sure how it works on 
> other architectures, can we always rely on memory for getting
> > exit-codes? If we would read the exit codes from memory from 
> loader/debugger/hypervisor it would require ELF capability or that the 
> BSP defines which address _Internal_errors_What_happened is to be
> > located at I guess
> >
> > > If not, the leon3 is now a multilib so you can just have a
> > > leon3 specific version of this macro.
> > I'm not sure I follow here.
> >
> >
> > Please see the code below where the SPARC implements an endless spin 
> after disabling interrupt. I substituted that with a power-down 
> instead, as the commit message suggests it does not stop a watchdog
> > to expire and reset.
> > >>   /**
> > >>    * This routine copies _error into a known place -- typically a 
> stack
> > >>    * location or a register, optionally disables interrupts, and
> > >>    * halts/stops the CPU.
> > >>    */
> > >> -#define _CPU_Fatal_halt( _source, _error ) \
> > >> -  do { \
> > >> -    uint32_t   level; \
> > >> -    \
> > >> -    level = sparc_disable_interrupts(); \
> > >> -    __asm__ volatile ( "mov  %0, %%g1 " : "=r" (level) : "0" 
> (level) ); \
> > >> -    while (1); /* loop forever */ \
> > >> -  } while (0)
> > >> +#define _CPU_Fatal_halt( _source, _error ) _BSP_Fatal_halt( 
> _source, _error )
> > >>
> > >>   /* end of Fatal Error manager macros */
> > >>
> >
>


More information about the devel mailing list