3.0.2 PATCH: Fix Solaris 2/Intel support in boehm-gc
Boehm, Hans
hans_boehm@hp.com
Tue Sep 25 09:56:00 GMT 2001
Thanks.
Looks fine to me.
I applied it to my gc6.1 tree as well. The patch didn't apply cleanly
there, so you may have similar problems with the gcc mainline. (Gcconfig.h
changes rapidly. Things moved, and small pieces of the patch were already
there.) But if you adjust things in the obvious way they should work.
Hans
> -----Original Message-----
> From: Rainer Orth [ mailto:ro@TechFak.Uni-Bielefeld.DE ]
> Sent: Tuesday, September 25, 2001 9:00 AM
> To: gcc-patches@gcc.gnu.org; java-patches@gcc.gnu.org
> Cc: Hans Boehm; Richard Henderson
> Subject: 3.0.2 PATCH: Fix Solaris 2/Intel support in boehm-gc
>
>
> This patch is a companion to
>
> http://gcc.gnu.org/ml/gcc-patches/2001-09/msg01071.html
>
> and makes boehm-gc work on Solaris 2/Intel
> (i386-pc-solaris2.9). With this
> patch, gctest works both in the statically and dynamically
> linked cases,
> and the libjava testsuite passes (provided
> http://gcc.gnu.org/ml/java/2001-06/msg00247.html or an
> equivalent patch to
> make the libjava testsuite to work on Solaris 2 is applied):
>
> === libjava Summary ===
>
> # of expected passes 1660
> # of unexpected failures 8
> # of unexpected successes 10
> # of expected failures 14
> # of untested testcases 24
>
> The patch contains three parts:
>
> * With lots of debugging help from Hans Boehm, it turned out that the
> DATASTART and DATAEND definitions were wrong: neither etext
> nor end do
> exist, but _etext and _end do. The STACKBOTTOM change is
> already on the
> trunk.
>
> * Using PROC_VDB reduces gctest runtime from 27.8 to 6.9
> seconds, so it is
> very desirable performance-wise. I can confirm that it
> works in Solaris
> 8 and 9, so I statically enable it for those releases. If
> anyone can
> confirm that it works in 2.6 and/or 7, or knows a Sun Bug
> Id or testcase,
> this could be extended to other releases.
>
> * Solaris 2/Intel uses a default stack size limit slighly
> larger than 8 MB,
> thus triggering a warning for any program linked with libgcjgc.
> Originally, this caused a large part of the libjava
> testsuite to fail, so
> I enable the warning only if a non-default limit is used.
>
> Ok for branch and mainline?
>
> Rainer
>
> --------------------------------------------------------------
> ---------------
> Rainer Orth, Faculty of Technology, Bielefeld University
>
> Email: ro@TechFak.Uni-Bielefeld.DE
>
>
> Tue Sep 25 15:09:03 2001 Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>
>
> * solaris_threads.c (MAX_ORIG_STACK_SIZE) [I386]:
> Provide special
> Solaris 2/Intel definition.
> (GC_get_orig_stack_size): NL-terminate warning message.
>
> * configure.in (i?86-*-solaris2.[89]*): Define
> SOLARIS25_PROC_VDB_BUG_FIXED.
> * gcconfig.h [I386 && SUNOS5]: Use it.
> * configure: Regenerate.
>
> * gcconfig.h [I386 && SUNOS5] (DATASTART): Use _etext.
> (DATAEND): Define using _end.
> (STACKBOTTOM): Define.
>
> Index: configure.in
> ===================================================================
> RCS file: /cvs/gcc/gcc/boehm-gc/configure.in,v
> retrieving revision 1.21.4.2
> diff -u -p -r1.21.4.2 configure.in
> --- configure.in 2001/06/08 17:59:15 1.21.4.2
> +++ configure.in 2001/09/25 13:42:25
> @@ -105,6 +105,9 @@ case "$host" in
> # alpha*-*-*)
> # machdep="alpha_mach_dep.lo"
> # ;;
> + i?86-*-solaris2.[[89]]*)
> + AC_DEFINE(SOLARIS25_PROC_VDB_BUG_FIXED)
> + ;;
> mipstx39-*-elf*)
> machdep="mips_ultrix_mach_dep.lo"
> AC_DEFINE(STACKBASE, __stackbase)
> Index: gcconfig.h
> ===================================================================
> RCS file: /cvs/gcc/gcc/boehm-gc/Attic/gcconfig.h,v
> retrieving revision 1.15.4.2
> diff -u -p -r1.15.4.2 gcconfig.h
> --- gcconfig.h 2001/04/05 00:13:13 1.15.4.2
> +++ gcconfig.h 2001/09/25 13:42:25
> @@ -722,12 +722,16 @@
> # endif
> # ifdef SUNOS5
> # define OS_TYPE "SUNOS5"
> - extern int etext, _start;
> + extern int _etext, _end;
> extern char * GC_SysVGetDataStart();
> -# define DATASTART GC_SysVGetDataStart(0x1000, &etext)
> -# define STACKBOTTOM ((ptr_t)(&_start))
> +# define DATASTART GC_SysVGetDataStart(0x1000, &_etext)
> +# define DATAEND (&_end)
> +# include <sys/vm.h>
> +# define STACKBOTTOM USRSTACK
> /** At least in Solaris 2.5, PROC_VDB gives wrong values for
> dirty bits. */
> -/*# define PROC_VDB*/
> +# ifdef SOLARIS25_PROC_VDB_BUG_FIXED
> +# define PROC_VDB
> +# endif
> # define DYNAMIC_LOADING
> # ifndef USE_MMAP
> # define USE_MMAP
> Index: solaris_threads.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/boehm-gc/solaris_threads.c,v
> retrieving revision 1.4
> diff -u -p -r1.4 solaris_threads.c
> --- solaris_threads.c 2000/04/19 10:09:58 1.4
> +++ solaris_threads.c 2001/09/25 13:42:25
> @@ -616,7 +616,18 @@ GC_thread GC_lookup_thread(thread_t id)
> return(p);
> }
>
> +/* Solaris 2/Intel uses an initial stack size limit slightly
> bigger than the
> + SPARC default of 8 MB. Account for this to warn only if
> the user has
> + raised the limit beyond the default.
> +
> + This is identical to DFLSSIZ defined in
> <sys/vm_machparam.h>. This file
> + is installed in /usr/platform/`uname -m`/include, which
> is not in the
> + default include directory list, so copy the definition here. */
> +#ifdef I386
> +# define MAX_ORIG_STACK_SIZE (8 * 1024 * 1024 + ((USRSTACK)
> & 0x3FFFFF))
> +#else
> # define MAX_ORIG_STACK_SIZE (8 * 1024 * 1024)
> +#endif
>
> word GC_get_orig_stack_size() {
> struct rlimit rl;
> @@ -627,7 +638,7 @@ word GC_get_orig_stack_size() {
> result = (word)rl.rlim_cur & ~(HBLKSIZE-1);
> if (result > MAX_ORIG_STACK_SIZE) {
> if (!warned) {
> - WARN("Large stack limit(%ld): only scanning 8 MB", result);
> + WARN("Large stack limit(%ld): only scanning 8
> MB\n", result);
> warned = 1;
> }
> result = MAX_ORIG_STACK_SIZE;
>
More information about the Gcc-patches
mailing list