arm-wince-pe-gcj: Finalizers not getting called

["Craig A. Vanderborgh" <craigv at voxware dot com>]

Hello All:

After an excruciating week we have finally tracked down a very bad
problem with our arm-wince-pe port of gcj/boehm/libgcj.

We are leaking synchronization objects big-time.  In fact, every call to
"new" appears to leak a M$ CRITICAL_SECTION.  The resulting leak is not
in the process address space but CE kernel pages are leaked.

The problem turns out to be rather simple.  InitializeCriticalSection,
the M$ call that creates the CRITICAL_SECTION is being called, and
DeleteCriticalSection is NOT being called.  Not even once.  The
finalizer appears to be getting defined and registered, vis:

00297508 <__Z18finalize_sync_infoPN4java4lang6ObjectE>:
  297508:	e92d4030 	stmdb	sp!, {r4, r5, lr}
  29750c:	e1a04000 	mov	r4, r0
  297510:	e3a05000 	mov	r5, #0	; 0x0
  297514:	e280002c 	add	r0, r0, #44	; 0x2c
  297518:	e5845024 	str	r5, [r4, #36]
  29751c:	e5845028 	str	r5, [r4, #40]
  297520:	eb03bf34 	bl	3871f8 <_DeleteCriticalSection>
  297524:	e5c45000 	strb	r5, [r4]
  297528:	e8bd8030 	ldmia	sp!, {r4, r5, pc}

0029752c <__ZN4java4lang6Object9sync_initEv>:
  29752c:	e92d4070 	stmdb	sp!, {r4, r5, r6, lr}
  297530:	e59f50ac 	ldr	r5, [pc, #172]	; 2975e4 <__ZN4java4lang6Object9sync_initEv+0xb8>
  297534:	e1a06000 	mov	r6, r0
  297538:	e2850008 	add	r0, r5, #8	; 0x8
  29753c:	eb040902 	bl	39994c <_EnterCriticalSection>
  297540:	e59f20a0 	ldr	r2, [pc, #160]	; 2975e8 <__ZN4java4lang6Object9sync_initEv+0xbc>
  297544:	e5953004 	ldr	r3, [r5, #4]
  297548:	e596c004 	ldr	ip, [r6, #4]
  29754c:	e5921000 	ldr	r1, [r2]
  297550:	e2833001 	add	r3, r3, #1	; 0x1
  297554:	e35c0000 	cmp	ip, #0	; 0x0
  297558:	e3a00040 	mov	r0, #64	; 0x40
  29755c:	e885000a 	stmia	r5, {r1, r3}
  297560:	0a000002 	beq	297570 <__ZN4java4lang6Object9sync_initEv+0x44>
  297564:	e5dc3000 	ldrb	r3, [ip]
  297568:	e3530000 	cmp	r3, #0	; 0x0
  29756c:	1a00000e 	bne	2975ac <__ZN4java4lang6Object9sync_initEv+0x80>
  297570:	eb014373 	bl	2e8344 <__Z14_Jv_AllocBytesi>
  297574:	e3a03000 	mov	r3, #0	; 0x0
  297578:	e1a04000 	mov	r4, r0
  29757c:	e280002c 	add	r0, r0, #44	; 0x2c
  297580:	e5843028 	str	r3, [r4, #40]
  297584:	e5843024 	str	r3, [r4, #36]
  297588:	eb03bf07 	bl	3871ac <_InitializeCriticalSection>
  29758c:	e2840004 	add	r0, r4, #4	; 0x4
  297590:	eb014638 	bl	2e8e78 <__Z12_Jv_CondInitP23_Jv_ConditionVariable_t>
  297594:	e3a03001 	mov	r3, #1	; 0x1
  297598:	e1a00004 	mov	r0, r4
  29759c:	e59f1048 	ldr	r1, [pc, #72]	; 2975ec <__ZN4java4lang6Object9sync_initEv+0xc0>
  2975a0:	e5c43000 	strb	r3, [r4]
  2975a4:	eb014382 	bl	2e83b4 <__Z21_Jv_RegisterFinalizerPvPFvPN4java4lang6ObjectEE>
  2975a8:	e5864004 	str	r4, [r6, #4]
  2975ac:	e59f2034 	ldr	r2, [pc, #52]	; 2975e8 <__ZN4java4lang6Object9sync_initEv+0xbc>
  2975b0:	e5953000 	ldr	r3, [r5]
  2975b4:	e5921000 	ldr	r1, [r2]
  2975b8:	e59f0030 	ldr	r0, [pc, #48]	; 2975f0 <__ZN4java4lang6Object9sync_initEv+0xc4>
  2975bc:	e1530001 	cmp	r3, r1
  2975c0:	0a000000 	beq	2975c8 <__ZN4java4lang6Object9sync_initEv+0x9c>
  2975c4:	e8bd8070 	ldmia	sp!, {r4, r5, r6, pc}
  2975c8:	e5953004 	ldr	r3, [r5, #4]
  2975cc:	e2433001 	sub	r3, r3, #1	; 0x1
  2975d0:	e3530000 	cmp	r3, #0	; 0x0
  2975d4:	e5853004 	str	r3, [r5, #4]
  2975d8:	05853000 	streq	r3, [r5]
  2975dc:	eb0408dd 	bl	399958 <_LeaveCriticalSection>
  2975e0:	eafffff7 	b	2975c4 <__ZN4java4lang6Object9sync_initEv+0x98>
  2975e4:	00594a90 	ldreqb	r4, [r9], -#160
  2975e8:	ffffc808 	swinv	0x00ffc808
  2975ec:	00297508 	eoreq	r7, r9, r8, lsl #10
  2975f0:	00594a98 	ldreqb	r4, [r9], -#168

Thus it appears that:
1. The configuration is done so that the finalizer, finalize_sync_info()
is there as it should be.
2. finalize_sync_info() contains the needed DCS
3. The finalizer is apparently "registered" via the RegisterFinalizer
call.

Thus all indications are that the finalizer is simply not being called. 
Where do finalizers get called?  Presumably in the GC, right?

If possible please advise us on what we ought to try to do next.

TIA,
craig vanderborgh
voxware incorporated