pc386dx exception 6 was Re: 4.6.0pre2 and docs released
Joel Sherrill
joel.sherrill at OARcorp.com
Wed Apr 23 13:12:45 UTC 2003
Angelo Fraietta wrote:
>
> Joel Sherrill wrote:
>
> > Angelo Fraietta wrote:
> >
> >> I have built and modified the ticker example to show how including
> >> iostream causes the program to crash.
> >> If you comment out line 23 (the include statement), the program
> >> runs.
> >> This was built using gcc3.2.2. and newlib 1.11.0 And the
> >> pre-release 2.
> >> I am curently only able to successfully run the first pre-release
> >> using
> >> gcc3.2 and newlib 1.10. As soon as I try to use anything later
> >> than
> >> this, it fails.
> >> I am running the pc386dx BSP.
> >> The program fails with exception 6 immediatly after the device
> >> drivers
> >> are initialised.
> >>
> > This is an illegal instruction exception. Where does the address
> > reported
> > correspond in your application? Can you do an objdump on the ticker
> > file
> > and see what is actually at that address?
> >
> The register values I get are
> eax = 1 ecx = 150d56 edx = 150d38
> esi= 150d60 edi=150e18 ebp = 165ae4 esp=165ab8
>
> Objdump shows me addresses ranging from
> 100000 to 11de51 -- Which one address I look at?
:( It should have helped if the dump had show eip so we would
have known the offending instruction. Can you run this in the
debugger and look at the stack address there?
> I am not actually using anything in the application. I included the
> iostream header but didi not use anything to make it fail. If I remove
> the include directive, the app runs and is reduced by about 1050K.
I suspect it is just calling some constructor. It doesn't surprise me
that the applciation shrinks.
> > I see PR381 in the ChangeLog for 4.6.0pre2 and it appears to be
> > committed
> > to src/newlibc.c. So that shouldn't be it.
> > What type of PC are you on? What CPU is REALLY in it? There are
> > some
> > legitimate instruction variations between ix86 CPU models. It is
> > possible that the updated tool combination somehow contains or
> > generates an instruction that is not valid for an i386DX. It is
> > also
> > possible that your CPU model is just enough not an i386dx where
> > something
> > has finally shown up.
> >
> I first tested on the DIMMPC, which is a 386sx. I then tested on 386dx
> with no-coprocessor and it failed.
> I tested on Pentium and it passed. The think that std cpp library
> getting linked has co processor instructions in it. I can't find my
> co-processor to plug into the dx computer right now so I can't confirm
> that.
>
> I want to re-iterate that the prerelease 1 that works is using gcc 3.2
> and not 3.2.2. This also happened with 3.2.1, when I first tried to
> build the pre-release 1, however, the task switching problem seemed to
> take precendence.
I am really suspicious that if we find the address of the offending
instruction
we will see an instruction that is not supported on i386 class CPUs.
There
are instruction differences between even the i486 and i386. On the
Pentium,
RTEMS even uses a different implementation of the benchmark timer since
you
have a cycle counter available.
Once we find the offensive instruction, the next step will be to
identify
whether it is hand-coded assembly language or misgenerated code. With
the ia64 mods to the i386 gcc backend, it is VERY possible that an edge
case broke with vanilla i386 code generation.
> > --joel
> >
> >> Joel Sherrill wrote:
> >>
> >> > Hi,
> >> > RTEMS 4.6.0pre2 is now available at:
> >> > ftp://ftp.oarcorp.com/pub/rtems/cd-working
> >> > It closes the following PRs:
> >> > PR 338/filesystem PR 348/bsps PR 349/bsps
> >> > PR 351/networking PR 352/rtems_misc PR 353/bsps
> >> > PR 354/networking PR 356/bsps PR
> >> > 357/rtems_misc
> >> > PR 360/bsps PR 367/filesystem PR
> >> > 368/filesystem
> >> > PR 368/filesystems PR 374/networking PR 378/bsps
> >> > PR 379/bsps PR 380/bsps PR
> >> > 381/core
> >> > PR 383/bsps PR 385/filesystem PR
> >> > 387/rtems_misc
> >> > PR 388/networking PR 389/rtems
> >> > [NOTE: Hopefully next time I will know how to automate the GNATS
> >> > query so we will see the PR synopsis with that list. :)
> >> > Corresponding documentation is at:
> >> > http://www.oarcorp.com/rtemsdoc-4.6.0-pre/
> >> > I hope this is the final preview before the release. But
> >> > feedback is
> >> > needed.
> >> > Thanks
> >> > --joel
> >> >
> >> --
> >> Angelo Fraietta
> >> PO Box 859
> >> Hamilton NSW 2303
> >> Home Page
> >> http://www.users.bigpond.com/angelo_f/
> >> There are those who seek knowledge for the sake of knowledge -
> >> that is CURIOSITY
> >> There are those who seek knowledge to be known by others - that is
> >> VANITY
> >> There are those who seek knowledge in order to serve - that is
> >> LOVE
> >> Bernard of Clairvaux (1090 - 1153)
> >> ------------------------------------------------------------------------
> >> #
> >> # Makefile
> >> #
> >> #
> >> # RTEMS_MAKEFILE_PATH is typically set in an environment variable
> >> #
> >> EXEC=ticker.exe
> >> PGM=${ARCH}/$(EXEC)
> >> # optional managers required
> >> MANAGERS=all
> >> # C source names
> >> CSRCS = init.c
> >> COBJS_ = $(CSRCS:.c=.o)
> >> COBJS = $(COBJS_:%=${ARCH}/%)
> >> # C++ source names
> >> CXXSRCS = tasks.cpp
> >> CXXOBJS_ = $(CXXSRCS:.cpp=.o)
> >> CXX
> >> OBJS = $(CXXOBJS_:%=${ARCH}/%)
> >> # AS source names
> >> ASSRCS =
> >> ASOBJS_ = $(ASSRCS:.s=.o)
> >> ASOBJS = $(ASOBJS_:%=${ARCH}/%)
> >> # Libraries
> >> LIBS = -lrtemsall -lc
> >> include $(RTEMS_MAKEFILE_PATH)/Makefile.inc
> >> include $(RTEMS_CUSTOM)
> >> include $(PROJECT_ROOT)/make/leaf.cfg
> >> OBJS= $(COBJS) $(CXXOBJS) $(ASOBJS)
> >> all: ${ARCH} $(PGM)
> >> $(PGM): $(OBJS)
> >> $(make-cxx-exe)
> >> j:
> >> @echo $(CFLAGS_DEBUG_V)
> >> @echo $(RTEMS_USE_GCC272)
> >> ------------------------------------------------------------------------
> >> /* Init
> >> *
> >> * This routine is the initialization task for this test program.
> >> * It is called from init_exec and has the responsibility for
> >> creating
> >> * and starting the tasks that make up the test. If the time of
> >> day
> >> * clock is required for the test, it should also be set to a
> >> known
> >> * value by this function.
> >> *
> >> * Input parameters: NONE
> >> *
> >> * Output paramete
> >> rs: NONE
> >> *
> >> * COPYRIGHT (c) 1989-1999.
> >> * On-Line Applications Research Corporation (OAR).
> >> *
> >> * The license and distribution terms for this file may be
> >> * found in the file LICENSE in this distribution or at
> >> * http://www.OARcorp.com/rtems/license.html.
> >> *
> >> * init.c,v 1.11 2000/06/12 15:00:12 joel Exp
> >> */
> >> #define CONFIGURE_INIT
> >> #include "system.h"
> >> #include <stdio.h>
> >> rtems_task Test_task(
> >> rtems_task_argument argument
> >> );
> >> /*
> >> * Keep the names and IDs in global variables so another task can
> >> use them.
> >> */
> >> rtems_id Task_id[ 4 ]; /* array of task ids */
> >> rtems_name Task_name[ 4 ]; /* array of task names */
> >> rtems_task Init(
> >> rtems_task_argument argument
> >> )
> >> {
> >> rtems_status_code status;
> >> rtems_time_of_day time;
> >> puts( "\n\n*** CLOCK TICK TEST ***" );
> >>
> >> time.year = 1988;
> >> time.month = 12;
> >> time.day = 31;
> >> time.hour = 9;
> >> time.minute = 0;
> >> time.second = 0;
> >> time.ticks = 0;
> >> status = rtems_clock_set( &time );
> >> Task_name[ 1 ] = rtems_build_name( 'T', 'A', '1', ' ' );
> >> Task_name[ 2 ] = rtems_build_name( 'T', 'A', '2', ' ' );
> >> Task_name[ 3 ] = rtems_build_name( 'T', 'A', '3', ' ' );
> >> status = rtems_task_create(
> >> Task_name[ 1 ], 1, RTEMS_MINIMUM_STACK_SIZE * 8,
> >> RTEMS_DEFAULT_MODES,
> >> RTEMS_DEFAULT_ATTRIBUTES, &Task_id[ 1 ]
> >> );
> >> /*
> >> status = rtems_task_create(
> >> Task_name[ 2 ], 1, RTEMS_MINIMUM_STACK_SIZE * 2,
> >> RTEMS_DEFAULT_MODES,
> >> RTEMS_DEFAULT_ATTRIBUTES, &Task_id[ 2 ]
> >> );
> >> status = rtems_task_create(
> >> Task_name[ 3 ], 1, RTEMS_MINIMUM_STACK_SIZE * 2,
> >> RTEMS_DEFAULT_MODES,
> >> RTEMS_DEFAULT_ATTRIBUTES, &Task_id[ 3 ]
> >> );
> >> */
> >> if (status == RTEMS_SUCCESSFUL)
> >>
> >> {
> >> printf ("Stack Size = %u\r\n",
> >> RTEMS_MINIMUM_STACK_SIZE);
> >> puts("Starting Task");
> >> status = rtems_task_start( Task_id[ 1 ],
> >> Test_task, 1 );
> >> }
> >> else
> >> {
> >> puts("Unable to create task");
> >> }
> >> /*
> >> status = rtems_task_start( Task_id[ 2 ], Test_task, 2 );
> >> status = rtems_task_start( Task_id[ 3 ], Test_task, 3 );
> >> */
> >> status = rtems_task_delete( RTEMS_SELF );
> >> }
> >> ------------------------------------------------------------------------
> >> /* system.h
> >> *
> >> * This include file contains information that is included in
> >> every
> >> * function in the test set.
> >> *
> >> * COPYRIGHT (c) 1989-1999.
> >> * On-Line Applications Research Corporation (OAR).
> >> *
> >> * The license and distribution terms for this file may be
> >> * found in the file LICENSE in this distribution or at
> >> * http://www.OARcorp.com/rtems/license.html.
> >> *
> >> * system.h,v 1.13 2000/06/12 15:00:12 joel Exp
> >> */
> >> #include <rtems.h>
> >> /* functions */
> >> rtems_task Init(
> >> rtems_task_argument argument
> >> );
> >> /* global variables */
> >> /*
> >> * Keep the names and IDs in global variables so another task can
> >> use them.
> >> */
> >> extern rtems_id Task_id[ 4 ]; /* array of task ids */
> >> extern rtems_name Task_name[ 4 ]; /* array of task names */
> >> /* configuration information */
> >> #include <bsp.h> /* for device driver prototypes */
> >> #define CONFIGURE_APPLICATION_NEEDS_CONSOLE_DRIVER
> >> #define CONFIGURE_APPLICATION_NEEDS_CLOCK_DRIVER
> >> #define CONFIGURE_MAXIMUM_TASKS 4
> >> #define CONFIGURE_RTEMS_INIT_TASKS_TABLE
> >> #define CONFIGURE_EXTRA_TASK_STACKS (10 *
> >> RTEMS_MINIMUM_STACK_SIZE * CONFIGURE_MAXIMUM_TASKS)<
> >> br>
> >> #include <confdefs.h>
> >> /*
> >> * Handy macros and static inline functions
> >> */
> >> /*
> >> * Macro to hide the ugliness of printing the time.
> >> */
> >> #define print_time(_s1, _tb, _s2) \
> >> do { \
> >> printf( "%s%02d:%02d:%02d %02d/%02d/%04d%s", \
> >> _s1, (_tb)->hour, (_tb)->minute, (_tb)->second, \
> >> (_tb)->month, (_tb)->day, (_tb)->year, _s2 ); \
> >> fflush(stdout); \
> >> } while ( 0 )
> >> /*
> >> * Macro to print an task name that is composed of ASCII
> >> characters.
> >> *
> >> */
> >> #define put_name( _name, _crlf ) \
> >> do { \
> >> rtems_unsigned32 c0, c1, c2, c3; \
> >> \
> >> c0 = ((_name) >> 24) & 0xff; \
> >> c1 = ((_name) >> 16) & 0xff; \
> >> c2 = ((_name) >> 8) & 0xff; \
> >> c3 = (_name) & 0xff; \
> >> putchar( (char)c0 ); \
> >> if ( c1 ) putchar( (char)c1 ); \
> >> if ( c2 ) putchar( (char)c2 ); \
> >> if ( c3 ) putchar( (char)c3 );
> >> \
> >> if ( (_crlf) ) \
> >> putchar( '\n' ); \
> >> } while (0)
> >> /*
> >> * static inline routine to make obtaining ticks per second
> >> easier.
> >> */
> >> static inline rtems_unsigned32 get_ticks_per_second( void )
> >> {
> >> rtems_interval ticks_per_second;
> >> (void) rtems_clock_get( RTEMS_CLOCK_GET_TICKS_PER_SECOND,
> >> &ticks_per_second ); return ticks_per_second;
> >> }
> >> /*
> >> * This allows us to view the "Test_task" instantiations as a set
> >> * of numbered tasks by eliminating the number of application
> >> * tasks created.
> >> *
> >> * In reality, this is too complex for the purposes of this
> >> * example. It would have been easier to pass a task argument.
> >> :)
> >> * But it shows how rtems_id's can sometimes be used.
> >> */
> >> #define task_number( tid ) \
> >> ( rtems_get_index( tid ) - \
> >> rtems_configuration_get_rtems_api_configuration()->number_of_initialization_tasks
> >> )
> >> /* end of include file */
> >> --------------------
> >> ----------------------------------------------------
> >> /* Test_task
> >> *
> >> * This routine serves as a test task. It verifies the basic
> >> task
> >> * switching capabilities of the executive.
> >> *
> >> * Input parameters: NONE
> >> *
> >> * Output parameters: NONE
> >> *
> >> * COPYRIGHT (c) 1989-1999.
> >> * On-Line Applications Research Corporation (OAR).
> >> *
> >> * The license and distribution terms for this file may be
> >> * found in the file LICENSE in this distribution or at
> >> * http://www.OARcorp.com/rtems/license.html.
> >> *
> >> * tasks.c,v 1.9 2002/04/17 14:01:40 joel Exp
> >> */
> >> #include "system.h"
> >> #include <stdio.h>
> >> #include <stdlib.h>
> >> #include <iostream>
> >> extern "C" rtems_task Test_task(
> >> rtems_task_argument unused
> >> )
> >> {
> >> rtems_id tid;
> >> rtems_time_of_day time;
> >> rtems_unsigned32 task_index;
> >> rt
> >> ems_status_code status;
> >> status = rtems_task_ident( RTEMS_SELF, RTEMS_SEARCH_ALL_NODES,
> >> &tid );
> >> task_index = task_number( tid );
> >> for ( ; ; ) {
> >> printf ("tick\r\n");
> >> status = rtems_clock_get( RTEMS_CLOCK_GET_TOD, &time );
> >> if ( time.second >= 35 ) {
> >> puts( "*** END OF CLOCK TICK TEST ***" );
> >> exit( 0 );
> >> }
> >> print_time( " - rtems_clock_get - ", &time, "\n" );
> >> status = rtems_task_wake_after( task_index * 5 *
> >> get_ticks_per_second() );
> >> }
> >> }
> >>
>
> --
> Angelo Fraietta
>
> PO Box 859
> Hamilton NSW 2303
>
> Home Page
>
> http://www.users.bigpond.com/angelo_f/
>
> There are those who seek knowledge for the sake of knowledge - that is CURIOSITY
> There are those who seek knowledge to be known by others - that is VANITY
> There are those who seek knowledge in order to serve - that is LOVE
> Bernard of Clairvaux (1090 - 1153)
--
Joel Sherrill, Ph.D. Director of Research & Development
joel at OARcorp.com On-Line Applications Research
Ask me about RTEMS: a free RTOS Huntsville AL 35805
Support Available (256) 722-9985
More information about the users
mailing list