2006-09-07 Carlos O'Donell <carlos@systemhalted.org>

	* sysdeps/hppa/dl-machine.h (elf_machine_fixup_plt): Remove
	lvalue cast.
	* sysdeps/hppa/dl-trampoline.S (_dl_fixup): Correct stack usage.
	(_dl_runtime_profile): LA fixups.
	* sysdeps/unix/sysv/linux/hppa/clone.S: Correct stack usage. Return
	-1 on error. Use branch and link for error handler funciton.
	* sysdeps/unix/sysv/linux/hppa/sysdep.h: Correct stack usage.
	Avoid register shuffling.
	* sysdeps/unix/sysv/linux/hppa/bits/atomic.h (ASM_EAGAIN): Define
	as -EAGAIN.
	* sysdeps/unix/sysv/linux/hppa/bits/mman.h: Adjust definitions to
	match required standards.
	* sysdeps/unix/sysv/linux/hppa/nptl/lowlevellock.h
	(lll_futex_wait): Return __ret.
	(lll_futex_timed_wait): Likewise.
	(lll_futex_wake): Likewise.
	(lll_futex_requeue): Likewise.
	* sysdeps/unix/sysv/linux/hppa/nptl/pt-vfork.S: Correct stack
	usage and adjust error return.
	* sysdeps/unix/sysv/linux/hppa/nptl/sysdep-cancel.h: Adjust
	stack usage for gdb, and avoid extra register loads.
	* sysdeps/unix/sysv/linux/hppa/nptl/unwind-forcedunwind.c: Copy
	nptl/sysdeps/pthread/unwind-forcedunwind.c.
	(LIBGCC_SO): Define and use.
	* sysdeps/unix/sysv/linux/hppa/nptl/unwind-resume.c: Copy
	nptl/sysdeps/pthread/unwind-resume.c.
	(LIBGCC_SO): Define and use.

2 files changed, 170 insertions, 78 deletions

diff --git a/sysdeps/hppa/dl-machine.h b/sysdeps/hppa/dl-machine.h
index dd5a28190b..67536c37d7 100644
--- a/sysdeps/hppa/dl-machine.h
+++ b/sysdeps/hppa/dl-machine.h
@@ -123,12 +123,13 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t t,
 		       const Elf32_Rela *reloc,
 		       Elf32_Addr *reloc_addr, struct fdesc value)
 {
+  volatile Elf32_Addr *rfdesc = reloc_addr;
   /* map is the link_map for the caller, t is the link_map for the object
      being called */
-  reloc_addr[1] = value.gp;
+  rfdesc[1] = value.gp;
   /* Need to ensure that the gp is visible before the code
      entry point is updated */
-  ((volatile Elf32_Addr *) reloc_addr)[0] = value.ip;
+  rfdesc[0] = value.ip;
   return value;
 }
 
diff --git a/sysdeps/hppa/dl-trampoline.S b/sysdeps/hppa/dl-trampoline.S
index c476138ebe..e0d3b9b108 100644
--- a/sysdeps/hppa/dl-trampoline.S
+++ b/sysdeps/hppa/dl-trampoline.S
@@ -1,5 +1,5 @@
 /* PLT trampolines. hppa version.
-   Copyright (C) 2005 Free Software Foundation, Inc.
+   Copyright (C) 2005, 2006 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -21,9 +21,9 @@
 
 /* This code gets called via the .plt stub, and is used in
    dl-runtime.c to call the `_dl_fixup' function and then redirect 
-   to the    address it returns. `_dl_fixup' takes two
-   arguments, however `_dl_profile_fixup' takes a number of 
-   parameters for use with library auditing (LA).
+   to the address it returns. `_dl_fixup' takes two arguments, however 
+   `_dl_profile_fixup' takes a number of parameters for use with 
+   library auditing (LA).
    
    WARNING: This template is also used by gcc's __cffc, and expects
    that the "bl" for _dl_runtime_resolve exist at a particular offset.
@@ -34,43 +34,46 @@
    
    Enter with r19 = reloc offset, r20 = got-8, r21 = fixup ltp.  */
 
-	/* FAKE bl to provide gcc's __cffc with fixup loc. */
+	/* RELOCATION MARKER: bl to provide gcc's __cffc with fixup loc. */
 	.text
+	/* THIS CODE DOES NOT EXECUTE */
 	bl	_dl_fixup, %r2
         .text
-        .align 4
         .global _dl_runtime_resolve
         .type _dl_runtime_resolve,@function
+	cfi_startproc
+        .align 4
 _dl_runtime_resolve:
         .PROC
         .CALLINFO FRAME=128,CALLS,SAVE_RP,ENTRY_GR=3
         .ENTRY
         /* SAVE_RP says we do */
-        stw %rp, -20(%sp)
+        stw	%rp, -20(%sp)
 
 	/* Save static link register */
 	stw	%r29,-16(%sp)
- 	/* Save argument registers in the call stack frame. */
+ 	/* Save argument registers */
 	stw	%r26,-36(%sp)
 	stw	%r25,-40(%sp)
 	stw	%r24,-44(%sp)
 	stw	%r23,-48(%sp)
 
 	/* Build a call frame, and save structure pointer. */
-	copy	%sp, %r26	/* Copy previous sp */
+	copy	%sp, %r1	/* Copy previous sp */
 	/* Save function result address (on entry) */
 	stwm	%r28,128(%sp)
-
-	/* Save floating point argument registers */
-	ldo	-56(%sp),%r26	
-	fstd,ma	%fr4,-8(%r26)
-	fstd,ma	%fr5,-8(%r26)
-	fstd,ma	%fr6,-8(%r26)
-	fstd	%fr7,0(%r26)
-
 	/* Fillin some frame info to follow ABI */
+	stw	%r1,-4(%sp)	/* Previous sp */
 	stw	%r21,-32(%sp)	/* PIC register value */
-	stw	%r26,-4(%sp)	/* Previous sp */
+
+	/* Save input floating point registers. This must be done
+	   in the new frame since the previous frame doesn't have
+	   enough space */
+	ldo	-56(%sp),%r1
+	fstd,ma	%fr4,-8(%r1)
+	fstd,ma	%fr5,-8(%r1)
+	fstd,ma	%fr6,-8(%r1)
+	fstd,ma	%fr7,-8(%r1)
 
  	/* Set up args to fixup func, needs only two arguments  */
 	ldw	8+4(%r20),%r26		/* (1) got[1] == struct link_map */
@@ -81,15 +84,15 @@ _dl_runtime_resolve:
 	copy	%r21,%r19		/* set fixup func ltp */
 
 	/* Load up the returned func descriptor */
-	copy	%ret0, %r22
-	copy	%ret1, %r19
+	copy	%r28, %r22
+	copy	%r29, %r19
 
 	/* Reload arguments fp args */
-	ldo	-80(%sp),%r26
-	fldd,ma	8(%r26),%fr7
-	fldd,ma	8(%r26),%fr6
-	fldd,ma	8(%r26),%fr5
-	fldd	0(%r26),%fr4
+	ldo	-56(%sp),%r1
+	fldd,ma	-8(%r1),%fr4
+	fldd,ma	-8(%r1),%fr5
+	fldd,ma	-8(%r1),%fr6
+	fldd,ma	-8(%r1),%fr7
 
 	/* Adjust sp, and restore function result address*/
 	ldwm	-128(%sp),%r28
@@ -107,91 +110,179 @@ _dl_runtime_resolve:
 	ldw	-20(%sp),%rp
         .EXIT
         .PROCEND
+	cfi_endproc
 	.size   _dl_runtime_resolve, . - _dl_runtime_resolve
 
-
-	/* FIXME:
-		Need to largely rewrite the bottom half of
-		this code in order to save and restore the
-		LA struct from the stack along with
-		interpreted parameters.
-	*/
         .text
-        .align 4
         .global _dl_runtime_profile
         .type _dl_runtime_profile,@function
+	cfi_startproc
+        .align 4
 _dl_runtime_profile:
         .PROC
-        .CALLINFO FRAME=128,CALLS,SAVE_RP,ENTRY_GR=3
+        .CALLINFO FRAME=192,CALLS,SAVE_RP,ENTRY_GR=3
         .ENTRY
 
         /* SAVE_RP says we do */
-        stw %rp, -20(%sp)
-
+        stw	%rp, -20(%sp)
 	/* Save static link register */
 	stw	%r29,-16(%sp)
- 	/* Save argument registers in the call stack frame. */
-	stw	%r26,-36(%sp)
-	stw	%r25,-40(%sp)
-	stw	%r24,-44(%sp)
-	stw	%r23,-48(%sp)
 
 	/* Build a call frame, and save structure pointer. */
-	copy	%sp, %r26	/* Copy previous sp */
+	copy	%sp, %r1	/* Copy previous sp */
 	/* Save function result address (on entry) */
-	stwm	%r28,128(%sp)
-
-	/* Save floating point argument registers */
-	ldo	-56(%sp),%r26	
-	fstd,ma	%fr4,-8(%r26)
-	fstd,ma	%fr5,-8(%r26)
-	fstd,ma	%fr6,-8(%r26)
-	fstd	%fr7,0(%r26)
-
+	stwm	%r28,192(%sp)
 	/* Fillin some frame info to follow ABI */
+	stw	%r1,-4(%sp)	/* Previous sp */
 	stw	%r21,-32(%sp)	/* PIC register value */
-	stw	%r26,-4(%sp)	/* Previous sp */
+
+	/* Create La_hppa_retval */
+	/* -140, lrv_r28 
+           -136, lrv_r29
+           -132, 4 byte pad 
+           -128, lr_fr4 (8 bytes) */
+
+	/* Create save space for _dl_profile_fixup arguments
+	   -120, Saved reloc offset 
+	   -116, Saved struct link_map 
+	   -112, *framesizep */
+
+	/* Create La_hppa_regs */
+	/* 32-bit registers */
+	stw	%r26,-108(%sp)
+	stw	%r25,-104(%sp)
+	stw	%r24,-100(%sp)
+	stw	%r23,-96(%sp)
+	/* -92, 4 byte pad */
+	/* 64-bit floating point registers */
+	ldo	-88(%sp),%r1
+	fstd,ma	%fr4,8(%r1)
+	fstd,ma	%fr5,8(%r1)
+	fstd,ma	%fr6,8(%r1)
+	fstd,ma	%fr7,8(%r1)
+	/* 32-bit stack pointer and return register */
+	stw	%sp,-56(%sp)
+	stw	%r2,-52(%sp)
+
 
  	/* Set up args to fixup func, needs five arguments  */
 	ldw	8+4(%r20),%r26		/* (1) got[1] == struct link_map */
+	stw	%r26,-116(%sp)		/* Save struct link_map */
 	copy	%r19,%r25		/* (2) reloc offset  */
+	stw	%r25,-120(%sp)		/* Save reloc offset */
 	copy    %rp,%r24		/* (3) profile_fixup needs rp */
-	copy	%r0,%r23		/* (4) regs */
-	ldo	-56(%sp), %r1
+	ldo	-56(%sp),%r23		/* (4) La_hppa_regs */
+	ldo	-112(%sp), %r1
 	stw	%r1, -52(%sp)		/* (5) long int *framesizep */
 
  	/* Call the real address resolver. */
 	bl	_dl_profile_fixup,%rp
 	copy	%r21,%r19		/* set fixup func ltp */
 
-	/* Load up the returned func descriptor */
-	copy	%ret0, %r22
-	copy	%ret1, %r19
-
-	/* Reload arguments fp args */
-	ldo	-80(%sp),%r26
-	fldd,ma	8(%r26),%fr7
-	fldd,ma	8(%r26),%fr6
-	fldd,ma	8(%r26),%fr5
-	fldd	0(%r26),%fr4
+	/* Load up the returned function descriptor */
+	copy	%r28, %r22
+	copy	%r29, %r19
+
+	/* Restore gr/fr/sp/rp */
+	ldw	-108(%sp),%r26
+	ldw	-104(%sp),%r25
+	ldw	-100(%sp),%r24
+	ldw	-96(%sp),%r23
+	/* -92, 4 byte pad, skip */
+	ldo	-88(%sp),%r1
+	fldd,ma	8(%r1),%fr4
+	fldd,ma	8(%r1),%fr5
+	fldd,ma	8(%r1),%fr6
+	fldd,ma	8(%r1),%fr7
+	ldw	-52(%sp),%rp
+
+	/* Reload static link register -(192+16) without adjusting stack */
+	ldw	-208(%sp),%r29
+
+	/* *framesizep is >= 0 if we have to run pltexit */
+	ldw	-112(%sp),%r28
+	cmpb,>>=,N %r0,%r28,L(cpe)
 
 	/* Adjust sp, and restore function result address*/
-	ldwm	-128(%sp),%r28
-
-	/* Reload static link register */
-	ldw	-16(%sp),%r29
-	/* Reload general args */
-	ldw	-36(%sp),%r26
-	ldw	-40(%sp),%r25
-	ldw	-44(%sp),%r24
-	ldw	-48(%sp),%r23
-
+	ldwm	-192(%sp),%r28
 	/* Jump to new function, but return to previous function */
 	bv	%r0(%r22)
 	ldw	-20(%sp),%rp
+	/* NO RETURN */
+
+L(nf):
+	/* Call the returned function descriptor */
+	bv	%r0(%r22)
+	nop
+	b,n	L(cont)
+
+L(cpe):
+	/* We are going to call the resolved function, but we have a 
+	   stack frame in the middle. We use the value of framesize to
+	   guess how much extra frame we need, and how much frame to
+	   copy forward. */
+
+	/* Round to nearest multiple of 64 */
+	addi	63, %r28, %r28
+	depi	0, 27, 6, %r28
+
+	/* Calcualte start of stack copy */
+	ldo	-192(%sp),%r2
+
+	/* Increate the stack by *framesizep */
+	copy	%sp, %r1
+	add	%sp, %r28, %sp
+	/* Save stack pointer */
+	stw	%r1, -4(%sp)
+
+	/* Single byte copy of prevous stack onto newly allocated stack */
+1:	ldb	%r28(%r2), %r1
+	add	%r28, %sp, %r26
+	stb	%r1, 0(%r26)
+	addi,<	-1,%r28,%r28
+	b,n	1b
+
+	/* Retore r28 and r27 and r2 already points at -192(%sp) */
+	ldw	0(%r2),%r28
+	ldw	84(%r2),%r26
+
+	/* Calculate address of L(cont) */
+	b,l	L(nf),%r2
+	depwi 0,31,2,%r2
+L(cont):
+	/* Undo fake stack */
+	ldw	-4(%sp),%r1
+	copy	%r1, %sp
+
+	/* Arguments to _dl_call_pltexit */
+	ldw	-116(%sp), %r26		/* (1) got[1] == struct link_map */ 
+	ldw 	-120(%sp), %r25		/* (2) reloc offsets */
+	ldo	-56(%sp), %r24		/* (3) *La_hppa_regs */
+	ldo	-124(%sp), %r23		/* (4) *La_hppa_retval */
+
+	/* Fill *La_hppa_retval */
+	stw	%r28,-140(%sp)
+	stw	%r29,-136(%sp)
+	ldo	-128(%sp), %r1
+	fstd	%fr4,0(%r1)
+
+	/* Call _dl_call_pltexit */
+	bl	_dl_call_pltexit,%rp
+	nop
+
+	/* Restore *La_hppa_retval */
+	ldw	-140(%sp), %r28
+	ldw	-136(%sp), %r29
+	ldo	-128(%sp), %r1
+	fldd	0(%r1), %fr4
+
+	/* Unwind the stack */
+	ldo	192(%sp),%sp
+	/* Retore callers rp */
+        ldw -20(%sp),%rp
+	/* Return */
+	bv,n	0(%r2)
         .EXIT
         .PROCEND
 	.size   _dl_runtime_profile, . - _dl_runtime_profile
 
-
-