[boehm-gc] Use thr_stksegment to determine Solaris stack base

[boehm-gc] Use thr_stksegment to determine Solaris stack base

[Rainer Orth]

[-- Attachment #1: Type: text/plain, Size: 1663 bytes --]

Since Solaris 11.1/x86, a few 32-bit boehm-gc testcases started to FAIL:

FAIL: boehm-gc.c/gctest.c -O2 execution test
FAIL: boehm-gc.c/leak_test.c -O2 execution test
FAIL: boehm-gc.c/thread_leak_test.c -O2 execution test
FAIL: boehm-gc.lib/staticrootstest.c -O2 execution test

They all SEGV on the first call to GC_mark_from.  This turns out to be a
side effect of the virtual memory system rewrite (VM 2.0) in that
Solaris version: one of the very few user-visible changes in that
release was that the 32-bit x86 stack base changed, unfortunately
without updating the USRSTACK definition in <sys/vmparam.h>.

Unlike the boehm-gc 6.6-based version in the gcc tree, the failure
doesn't happen with gc 7.2c.  It turns out that this is case since that
version uses the thr_stksegment(3C) function instead of the macro.  This
function exists since at least Solaris 2.5.1 and has the additional
advantage of also working with ASLR also introduced with VM2, where a
fixed stack base is wrong.

This patch fixes the failures by backporting the use of thr_stksegment
from gc 7.2c.

Bootstrapped without regressions on i386-pc-solaris2.{9,10,11} and
sparc-sun-solaris2.{9,10,11}, fixes the i386-pc-solaris2.11 failures.

Ok for mainline?

	Rainer


2013-03-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

	* os_dep.c [SOLARIS_STACKBOTTOM] (GC_solaris_stack_base): New
	function.
	[!BEOS && !AMIGA && !MSWIN32 && !MSWINCE && !OS2 && !NOSYS &&
	!ECOS] (GC_get_stack_base): Use it.
	* include/private/gcconfig.h [SPARC && SUNOS5]
	(SOLARIS_STACKBOTTOM): Define.
	(STACKBOTTOM, HEURISTIC2): Remove.
	[I386 && SUNOS5]
	(SOLARIS_STACKBOTTOM): Define.
	(STACKBOTTOM): Remove.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: sol2-boehm-gc-stack.patch --]
[-- Type: text/x-patch, Size: 4413 bytes --]

# HG changeset patch
# Parent 3b82ec434332de00d9006fa3985dd9180c0cf499
Use thr_stksegment to determine Solaris stack base

diff --git a/boehm-gc/include/private/gcconfig.h b/boehm-gc/include/private/gcconfig.h
--- a/boehm-gc/include/private/gcconfig.h
+++ b/boehm-gc/include/private/gcconfig.h
@@ -927,18 +927,7 @@
 #	  define HEAP_START DATAEND
 #       endif
 #	define PROC_VDB
-/*	HEURISTIC1 reportedly no longer works under 2.7.  		*/
-/*  	HEURISTIC2 probably works, but this appears to be preferable.	*/
-/*	Apparently USRSTACK is defined to be USERLIMIT, but in some	*/
-/* 	installations that's undefined.  We work around this with a	*/
-/*	gross hack:							*/
-#       include <sys/vmparam.h>
-#	ifdef USERLIMIT
-	  /* This should work everywhere, but doesn't.	*/
-#	  define STACKBOTTOM USRSTACK
-#       else
-#	  define HEURISTIC2
-#       endif
+#	define SOLARIS_STACKBOTTOM
 #	include <unistd.h>
 #       define GETPAGESIZE()  sysconf(_SC_PAGESIZE)
 		/* getpagesize() appeared to be missing from at least one */
@@ -1067,13 +1056,7 @@
   	extern ptr_t GC_SysVGetDataStart();
 #       define DATASTART GC_SysVGetDataStart(0x1000, _etext)
 #	define DATAEND (_end)
-/*	# define STACKBOTTOM ((ptr_t)(_start)) worked through 2.7,  	*/
-/*      but reportedly breaks under 2.8.  It appears that the stack	*/
-/* 	base is a property of the executable, so this should not break	*/
-/* 	old executables.						*/
-/*  	HEURISTIC2 probably works, but this appears to be preferable.	*/
-#       include <sys/vm.h>
-#	define STACKBOTTOM USRSTACK
+#	define SOLARIS_STACKBOTTOM
 /* At least in Solaris 2.5, PROC_VDB gives wrong values for dirty bits. */
 /* It appears to be fixed in 2.8 and 2.9.				*/
 #	ifdef SOLARIS25_PROC_VDB_BUG_FIXED
diff --git a/boehm-gc/os_dep.c b/boehm-gc/os_dep.c
--- a/boehm-gc/os_dep.c
+++ b/boehm-gc/os_dep.c
@@ -1008,13 +1008,62 @@ ptr_t GC_get_stack_base()
 
 #endif /* FREEBSD_STACKBOTTOM */
 
+#ifdef SOLARIS_STACKBOTTOM
+
+# include <thread.h>
+# include <signal.h>
+# include <pthread.h>
+
+  /* These variables are used to cache ss_sp value for the primordial   */
+  /* thread (it's better not to call thr_stksegment() twice for this    */
+  /* thread - see JDK bug #4352906).                                    */
+  static pthread_t stackbase_main_self = 0;
+                        /* 0 means stackbase_main_ss_sp value is unset. */
+  static void *stackbase_main_ss_sp = NULL;
+
+  ptr_t GC_solaris_stack_base(void)
+  {
+    stack_t s;
+    pthread_t self = pthread_self();
+    if (self == stackbase_main_self)
+      {
+        /* If the client calls GC_get_stack_base() from the main thread */
+        /* then just return the cached value.                           */
+        GC_ASSERT(stackbase_main_ss_sp != NULL);
+        return stackbase_main_ss_sp;
+      }
+
+    if (thr_stksegment(&s)) {
+      /* According to the manual, the only failure error code returned  */
+      /* is EAGAIN meaning "the information is not available due to the */
+      /* thread is not yet completely initialized or it is an internal  */
+      /* thread" - this shouldn't happen here.                          */
+      ABORT("thr_stksegment failed");
+    }
+    /* s.ss_sp holds the pointer to the stack bottom. */
+    GC_ASSERT((void *)&s HOTTER_THAN s.ss_sp);
+
+    if (!stackbase_main_self)
+      {
+        /* Cache the stack base value for the primordial thread (this   */
+        /* is done during GC_init, so there is no race).                */
+        stackbase_main_ss_sp = s.ss_sp;
+        stackbase_main_self = self;
+      }
+
+    return s.ss_sp;
+  }
+
+#endif /* GC_SOLARIS_THREADS */
+
 #if !defined(BEOS) && !defined(AMIGA) && !defined(MSWIN32) \
     && !defined(MSWINCE) && !defined(OS2) && !defined(NOSYS) && !defined(ECOS)
 
 ptr_t GC_get_stack_base()
 {
 #   if defined(HEURISTIC1) || defined(HEURISTIC2) || \
-       defined(LINUX_STACKBOTTOM) || defined(FREEBSD_STACKBOTTOM)
+       defined(LINUX_STACKBOTTOM) || defined(FREEBSD_STACKBOTTOM) || \
+       defined(SOLARIS_STACKBOTTOM)       
     word dummy;
     ptr_t result;
 #   endif
@@ -1040,6 +1089,9 @@ ptr_t GC_get_stack_base()
 #	ifdef FREEBSD_STACKBOTTOM
 	   result = GC_freebsd_stack_base();
 #	endif
+#	ifdef SOLARIS_STACKBOTTOM
+	   result = GC_solaris_stack_base();
+#	endif
 #	ifdef HEURISTIC2
 #	    ifdef STACK_GROWS_DOWN
 		result = GC_find_limit((ptr_t)(&dummy), TRUE);

[-- Attachment #3: Type: text/plain, Size: 143 bytes --]


-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [boehm-gc] Use thr_stksegment to determine Solaris stack base

[Bryce McKinlay]

On Wed, Mar 6, 2013 at 11:31 AM, Rainer Orth
<ro@cebitec.uni-bielefeld.de> wrote:

> This patch fixes the failures by backporting the use of thr_stksegment
> from gc 7.2c.
>
> Bootstrapped without regressions on i386-pc-solaris2.{9,10,11} and
> sparc-sun-solaris2.{9,10,11}, fixes the i386-pc-solaris2.11 failures.
>
> Ok for mainline?

Yes, thanks. Though of course, we really ought to update the entire GC tree!

Bryce