1# x86/x86_64 support for -fsplit-stack.
2# Copyright (C) 2009-2013 Free Software Foundation, Inc.
3# Contributed by Ian Lance Taylor <iant@google.com>.
4
5# This file is part of GCC.
6
7# GCC is free software; you can redistribute it and/or modify it under
8# the terms of the GNU General Public License as published by the Free
9# Software Foundation; either version 3, or (at your option) any later
10# version.
11
12# GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13# WARRANTY; without even the implied warranty of MERCHANTABILITY or
14# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
15# for more details.
16
17# Under Section 7 of GPL version 3, you are granted additional
18# permissions described in the GCC Runtime Library Exception, version
19# 3.1, as published by the Free Software Foundation.
20
21# You should have received a copy of the GNU General Public License and
22# a copy of the GCC Runtime Library Exception along with this program;
23# see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
24# <http://www.gnu.org/licenses/>.
25
26
27# Support for allocating more stack space when using -fsplit-stack.
28# When a function discovers that it needs more stack space, it will
29# call __morestack with the size of the stack frame and the size of
30# the parameters to copy from the old stack frame to the new one.
31# The __morestack function preserves the parameter registers and
32# calls __generic_morestack to actually allocate the stack space.
33
34# When this is called stack space is very low, but we ensure that
35# there is enough space to push the parameter registers and to call
36# __generic_morestack.
37
38# When calling __generic_morestack, FRAME_SIZE points to the size of
39# the desired frame when the function is called, and the function
40# sets it to the size of the allocated stack.  OLD_STACK points to
41# the parameters on the old stack and PARAM_SIZE is the number of
42# bytes of parameters to copy to the new stack.  These are the
43# parameters of the function that called __morestack.  The
44# __generic_morestack function returns the new stack pointer,
45# pointing to the address of the first copied parameter.  The return
46# value minus the returned *FRAME_SIZE will be the first address on
47# the stack which we should not use.
48
49# void *__generic_morestack (size_t *frame_size, void *old_stack,
50#			     size_t param_size);
51
52# The __morestack routine has to arrange for the caller to return to a
53# stub on the new stack.  The stub is responsible for restoring the
54# old stack pointer and returning to the caller's caller.  This calls
55# __generic_releasestack to retrieve the old stack pointer and release
56# the newly allocated stack.
57
58# void *__generic_releasestack (size_t *available);
59
60# We do a little dance so that the processor's call/return return
61# address prediction works out.  The compiler arranges for the caller
62# to look like this:
63#   call __generic_morestack
64#   ret
65#  L:
66#   // carry on with function
67# After we allocate more stack, we call L, which is in our caller.
68# When that returns (to the predicted instruction), we release the
69# stack segment and reset the stack pointer.  We then return to the
70# predicted instruction, namely the ret instruction immediately after
71# the call to __generic_morestack.  That then returns to the caller of
72# the original caller.
73
74
75# The amount of extra space we ask for.  In general this has to be
76# enough for the dynamic loader to find a symbol and for a signal
77# handler to run.
78
79#ifndef __x86_64__
80#define BACKOFF (1024)
81#else
82#define BACKOFF (1536)
83#endif
84
85
86# The amount of space we ask for when calling non-split-stack code.
87#define NON_SPLIT_STACK 0x100000
88
89# This entry point is for split-stack code which calls non-split-stack
90# code.  When the linker sees this case, it converts the call to
91# __morestack to call __morestack_non_split instead.  We just bump the
92# requested stack space by 16K.
93
94	.global __morestack_non_split
95	.hidden	__morestack_non_split
96
97#ifdef __ELF__
98       .type	__morestack_non_split,@function
99#endif
100
101__morestack_non_split:
102	.cfi_startproc
103
104#ifndef __x86_64__
105
106	# See below for an extended explanation of this.
107	.cfi_def_cfa %esp,16
108
109	pushl	%eax			# Save %eax in case it is a parameter.
110
111	.cfi_adjust_cfa_offset 4	# Account for pushed register.
112
113	movl	%esp,%eax		# Current stack,
114	subl	8(%esp),%eax		# less required stack frame size,
115	subl	$NON_SPLIT_STACK,%eax	# less space for non-split code.
116	cmpl	%gs:0x30,%eax		# See if we have enough space.
117	jb	2f			# Get more space if we need it.
118
119	# Here the stack is
120	#	%esp + 20:	stack pointer after two returns
121	#	%esp + 16:	return address of morestack caller's caller
122	#	%esp + 12:	size of parameters
123	#	%esp + 8:	new stack frame size
124	#	%esp + 4:	return address of this function
125	#	%esp:		saved %eax
126	#
127	# Since we aren't doing a full split stack, we don't need to
128	# do anything when our caller returns.  So we return to our
129	# caller rather than calling it, and let it return as usual.
130	# To make that work we adjust the return address.
131
132	# This breaks call/return address prediction for the call to
133	# this function.  I can't figure out a way to make it work
134	# short of copying the parameters down the stack, which will
135	# probably take more clock cycles than we will lose breaking
136	# call/return address prediction.  We will only break
137	# prediction for this call, not for our caller.
138
139	movl	4(%esp),%eax		# Increment the return address
140	cmpb	$0xc3,(%eax)		# to skip the ret instruction;
141	je	1f			# see above.
142	addl	$2,%eax
1431:	inc	%eax
144
145	# If the instruction that we return to is
146	#   leal  20(%ebp),{%eax,%ecx,%edx}
147	# then we have been called by a varargs function that expects
148	# %ebp to hold a real value.  That can only work if we do the
149	# full stack split routine.  FIXME: This is fragile.
150	cmpb	$0x8d,(%eax)
151	jne	3f
152	cmpb	$0x14,2(%eax)
153	jne	3f
154	cmpb	$0x45,1(%eax)
155	je	2f
156	cmpb	$0x4d,1(%eax)
157	je	2f
158	cmpb	$0x55,1(%eax)
159	je	2f
160
1613:
162	movl	%eax,4(%esp)		# Update return address.
163
164	popl	%eax			# Restore %eax and stack.
165
166	.cfi_adjust_cfa_offset -4	# Account for popped register.
167
168	ret	$8			# Return to caller, popping args.
169
1702:
171	.cfi_adjust_cfa_offset 4	# Back to where we were.
172
173	popl	%eax			# Restore %eax and stack.
174
175	.cfi_adjust_cfa_offset -4	# Account for popped register.
176
177	# Increment space we request.
178	addl	$NON_SPLIT_STACK+0x1000+BACKOFF,4(%esp)
179
180	# Fall through into morestack.
181
182#else
183
184	# See below for an extended explanation of this.
185	.cfi_def_cfa %rsp,16
186
187	pushq	%rax			# Save %rax in case caller is using
188					# it to preserve original %r10.
189	.cfi_adjust_cfa_offset 8	# Adjust for pushed register.
190
191	movq	%rsp,%rax		# Current stack,
192	subq	%r10,%rax		# less required stack frame size,
193	subq	$NON_SPLIT_STACK,%rax	# less space for non-split code.
194
195#ifdef __LP64__
196	cmpq	%fs:0x70,%rax		# See if we have enough space.
197#else
198	cmpl	%fs:0x40,%eax
199#endif
200
201	jb	2f			# Get more space if we need it.
202
203	# If the instruction that we return to is
204	#   leaq  24(%rbp), %r11n
205	# then we have been called by a varargs function that expects
206	# %ebp to hold a real value.  That can only work if we do the
207	# full stack split routine.  FIXME: This is fragile.
208	movq	8(%rsp),%rax
209	incq	%rax			# Skip ret instruction in caller.
210	cmpl	$0x185d8d4c,(%rax)
211	je	2f
212
213	# This breaks call/return prediction, as described above.
214	incq	8(%rsp)			# Increment the return address.
215
216	popq	%rax			# Restore register.
217
218	.cfi_adjust_cfa_offset -8	# Adjust for popped register.
219
220	ret				# Return to caller.
221
2222:
223	popq	%rax			# Restore register.
224
225	.cfi_adjust_cfa_offset -8	# Adjust for popped register.
226
227	# Increment space we request.
228	addq	$NON_SPLIT_STACK+0x1000+BACKOFF,%r10
229
230	# Fall through into morestack.
231
232#endif
233
234	.cfi_endproc
235#ifdef __ELF__
236	.size	__morestack_non_split, . - __morestack_non_split
237#endif
238
239# __morestack_non_split falls through into __morestack.
240
241
242# The __morestack function.
243
244	.global	__morestack
245	.hidden	__morestack
246
247#ifdef __ELF__
248	.type	__morestack,@function
249#endif
250
251__morestack:
252.LFB1:
253	.cfi_startproc
254
255
256#ifndef __x86_64__
257
258
259# The 32-bit __morestack function.
260
261	# We use a cleanup to restore the stack guard if an exception
262	# is thrown through this code.
263#ifndef __PIC__
264	.cfi_personality 0,__gcc_personality_v0
265	.cfi_lsda 0,.LLSDA1
266#else
267	.cfi_personality 0x9b,DW.ref.__gcc_personality_v0
268	.cfi_lsda 0x1b,.LLSDA1
269#endif
270
271	# We return below with a ret $8.  We will return to a single
272	# return instruction, which will return to the caller of our
273	# caller.  We let the unwinder skip that single return
274	# instruction, and just return to the real caller.
275
276	# Here CFA points just past the return address on the stack,
277	# e.g., on function entry it is %esp + 4.  The stack looks
278	# like this:
279	#	CFA + 12:	stack pointer after two returns
280	#	CFA + 8:	return address of morestack caller's caller
281	#	CFA + 4:	size of parameters
282	#	CFA:		new stack frame size
283	#	CFA - 4:	return address of this function
284	#	CFA - 8:	previous value of %ebp; %ebp points here
285	# Setting the new CFA to be the current CFA + 12 (i.e., %esp +
286	# 16) will make the unwinder pick up the right return address.
287
288	.cfi_def_cfa %esp,16
289
290	pushl	%ebp
291	.cfi_adjust_cfa_offset 4
292	.cfi_offset %ebp, -20
293	movl	%esp,%ebp
294	.cfi_def_cfa_register %ebp
295
296	# In 32-bit mode the parameters are pushed on the stack.  The
297	# argument size is pushed then the new stack frame size is
298	# pushed.
299
300	# In the body of a non-leaf function, the stack pointer will
301	# be aligned to a 16-byte boundary.  That is CFA + 12 in the
302	# stack picture above: (CFA + 12) % 16 == 0.  At this point we
303	# have %esp == CFA - 8, so %esp % 16 == 12.  We need some
304	# space for saving registers and passing parameters, and we
305	# need to wind up with %esp % 16 == 0.
306	subl	$44,%esp
307
308	# Because our cleanup code may need to clobber %ebx, we need
309	# to save it here so the unwinder can restore the value used
310	# by the caller.  Note that we don't have to restore the
311	# register, since we don't change it, we just have to save it
312	# for the unwinder.
313	movl	%ebx,-4(%ebp)
314	.cfi_offset %ebx, -24
315
316	# In 32-bit mode the registers %eax, %edx, and %ecx may be
317	# used for parameters, depending on the regparm and fastcall
318	# attributes.
319
320	movl	%eax,-8(%ebp)
321	movl	%edx,-12(%ebp)
322	movl	%ecx,-16(%ebp)
323
324	call	__morestack_block_signals
325
326	movl	12(%ebp),%eax		# The size of the parameters.
327	movl	%eax,8(%esp)
328	leal	20(%ebp),%eax		# Address of caller's parameters.
329	movl	%eax,4(%esp)
330	addl	$BACKOFF,8(%ebp)	# Ask for backoff bytes.
331	leal	8(%ebp),%eax		# The address of the new frame size.
332	movl	%eax,(%esp)
333
334	call	__generic_morestack
335
336	movl	%eax,%esp		# Switch to the new stack.
337	subl	8(%ebp),%eax		# The end of the stack space.
338	addl	$BACKOFF,%eax		# Back off 512 bytes.
339
340.LEHB0:
341	# FIXME: The offset must match
342	# TARGET_THREAD_SPLIT_STACK_OFFSET in
343	# gcc/config/i386/linux.h.
344	movl	%eax,%gs:0x30		# Save the new stack boundary.
345
346	call	__morestack_unblock_signals
347
348	movl	-12(%ebp),%edx		# Restore registers.
349	movl	-16(%ebp),%ecx
350
351	movl	4(%ebp),%eax		# Increment the return address
352	cmpb	$0xc3,(%eax)		# to skip the ret instruction;
353	je	1f			# see above.
354	addl	$2,%eax
3551:	inc	%eax
356
357	movl	%eax,-12(%ebp)		# Store return address in an
358					# unused slot.
359
360	movl	-8(%ebp),%eax		# Restore the last register.
361
362	call	*-12(%ebp)		# Call our caller!
363
364	# The caller will return here, as predicted.
365
366	# Save the registers which may hold a return value.  We
367	# assume that __generic_releasestack does not touch any
368	# floating point or vector registers.
369	pushl	%eax
370	pushl	%edx
371
372	# Push the arguments to __generic_releasestack now so that the
373	# stack is at a 16-byte boundary for
374	# __morestack_block_signals.
375	pushl	$0			# Where the available space is returned.
376	leal	0(%esp),%eax		# Push its address.
377	push	%eax
378
379	call	__morestack_block_signals
380
381	call	__generic_releasestack
382
383	subl	4(%esp),%eax		# Subtract available space.
384	addl	$BACKOFF,%eax		# Back off 512 bytes.
385.LEHE0:
386	movl	%eax,%gs:0x30		# Save the new stack boundary.
387
388	addl	$8,%esp			# Remove values from stack.
389
390	# We need to restore the old stack pointer, which is in %rbp,
391	# before we unblock signals.  We also need to restore %eax and
392	# %edx after we unblock signals but before we return.  Do this
393	# by moving %eax and %edx from the current stack to the old
394	# stack.
395
396	popl	%edx			# Pop return value from current stack.
397	popl	%eax
398
399	movl	%ebp,%esp		# Restore stack pointer.
400
401	# As before, we now have %esp % 16 == 12.
402
403	pushl	%eax			# Push return value on old stack.
404	pushl	%edx
405	subl	$4,%esp			# Align stack to 16-byte boundary.
406
407	call	__morestack_unblock_signals
408
409	addl	$4,%esp
410	popl	%edx			# Restore return value.
411	popl	%eax
412
413	.cfi_remember_state
414
415	# We never changed %ebx, so we don't have to actually restore it.
416	.cfi_restore %ebx
417
418	popl	%ebp
419	.cfi_restore %ebp
420	.cfi_def_cfa %esp, 16
421	ret	$8			# Return to caller, which will
422					# immediately return.  Pop
423					# arguments as we go.
424
425# This is the cleanup code called by the stack unwinder when unwinding
426# through the code between .LEHB0 and .LEHE0 above.
427
428.L1:
429	.cfi_restore_state
430	subl	$16,%esp		# Maintain 16 byte alignment.
431	movl	%eax,4(%esp)		# Save exception header.
432	movl	%ebp,(%esp)		# Stack pointer after resume.
433	call	__generic_findstack
434	movl	%ebp,%ecx		# Get the stack pointer.
435	subl	%eax,%ecx		# Subtract available space.
436	addl	$BACKOFF,%ecx		# Back off 512 bytes.
437	movl	%ecx,%gs:0x30		# Save new stack boundary.
438	movl	4(%esp),%eax		# Function argument.
439	movl	%eax,(%esp)
440#ifdef __PIC__
441	call	__x86.get_pc_thunk.bx	# %ebx may not be set up for us.
442	addl	$_GLOBAL_OFFSET_TABLE_, %ebx
443	call	_Unwind_Resume@PLT	# Resume unwinding.
444#else
445	call	_Unwind_Resume
446#endif
447
448#else /* defined(__x86_64__) */
449
450
451# The 64-bit __morestack function.
452
453	# We use a cleanup to restore the stack guard if an exception
454	# is thrown through this code.
455#ifndef __PIC__
456	.cfi_personality 0x3,__gcc_personality_v0
457	.cfi_lsda 0x3,.LLSDA1
458#else
459	.cfi_personality 0x9b,DW.ref.__gcc_personality_v0
460	.cfi_lsda 0x1b,.LLSDA1
461#endif
462
463	# We will return a single return instruction, which will
464	# return to the caller of our caller.  Let the unwinder skip
465	# that single return instruction, and just return to the real
466	# caller.
467	.cfi_def_cfa %rsp,16
468
469	# Set up a normal backtrace.
470	pushq	%rbp
471	.cfi_adjust_cfa_offset 8
472	.cfi_offset %rbp, -24
473	movq	%rsp, %rbp
474	.cfi_def_cfa_register %rbp
475
476	# In 64-bit mode the new stack frame size is passed in r10
477        # and the argument size is passed in r11.
478
479	addq	$BACKOFF,%r10		# Ask for backoff bytes.
480	pushq	%r10			# Save new frame size.
481
482	# In 64-bit mode the registers %rdi, %rsi, %rdx, %rcx, %r8,
483	# and %r9 may be used for parameters.  We also preserve %rax
484	# which the caller may use to hold %r10.
485
486	pushq	%rax
487	pushq	%rdi
488	pushq	%rsi
489	pushq	%rdx
490	pushq	%rcx
491	pushq	%r8
492	pushq	%r9
493
494	pushq	%r11
495
496	# We entered morestack with the stack pointer aligned to a
497	# 16-byte boundary (the call to morestack's caller used 8
498	# bytes, and the call to morestack used 8 bytes).  We have now
499	# pushed 10 registers, so we are still aligned to a 16-byte
500	# boundary.
501
502	call	__morestack_block_signals
503
504	leaq	-8(%rbp),%rdi		# Address of new frame size.
505	leaq	24(%rbp),%rsi		# The caller's parameters.
506	popq	%rdx			# The size of the parameters.
507
508	subq	$8,%rsp			# Align stack.
509
510	call	__generic_morestack
511
512	movq	-8(%rbp),%r10		# Reload modified frame size
513	movq	%rax,%rsp		# Switch to the new stack.
514	subq	%r10,%rax		# The end of the stack space.
515	addq	$BACKOFF,%rax		# Back off 1024 bytes.
516
517.LEHB0:
518	# FIXME: The offset must match
519	# TARGET_THREAD_SPLIT_STACK_OFFSET in
520	# gcc/config/i386/linux64.h.
521	# Macro to save the new stack boundary.
522#ifdef __LP64__
523#define X86_64_SAVE_NEW_STACK_BOUNDARY(reg)	movq	%r##reg,%fs:0x70
524#else
525#define X86_64_SAVE_NEW_STACK_BOUNDARY(reg)	movl	%e##reg,%fs:0x40
526#endif
527	X86_64_SAVE_NEW_STACK_BOUNDARY (ax)
528
529	call	__morestack_unblock_signals
530
531	movq	-24(%rbp),%rdi		# Restore registers.
532	movq	-32(%rbp),%rsi
533	movq	-40(%rbp),%rdx
534	movq	-48(%rbp),%rcx
535	movq	-56(%rbp),%r8
536	movq	-64(%rbp),%r9
537
538	movq	8(%rbp),%r10		# Increment the return address
539	incq	%r10			# to skip the ret instruction;
540					# see above.
541
542	movq	-16(%rbp),%rax		# Restore caller's %rax.
543
544	call	*%r10			# Call our caller!
545
546	# The caller will return here, as predicted.
547
548	# Save the registers which may hold a return value.  We
549	# assume that __generic_releasestack does not touch any
550	# floating point or vector registers.
551	pushq	%rax
552	pushq	%rdx
553
554	call	__morestack_block_signals
555
556	pushq	$0			# For alignment.
557	pushq	$0			# Where the available space is returned.
558	leaq	0(%rsp),%rdi		# Pass its address.
559
560	call	__generic_releasestack
561
562	subq	0(%rsp),%rax		# Subtract available space.
563	addq	$BACKOFF,%rax		# Back off 1024 bytes.
564.LEHE0:
565	X86_64_SAVE_NEW_STACK_BOUNDARY (ax)
566
567	addq	$16,%rsp		# Remove values from stack.
568
569	# We need to restore the old stack pointer, which is in %rbp,
570	# before we unblock signals.  We also need to restore %rax and
571	# %rdx after we unblock signals but before we return.  Do this
572	# by moving %rax and %rdx from the current stack to the old
573	# stack.
574
575	popq	%rdx			# Pop return value from current stack.
576	popq	%rax
577
578	movq	%rbp,%rsp		# Restore stack pointer.
579
580	# Now (%rsp & 16) == 8.
581
582	subq	$8,%rsp			# For alignment.
583	pushq	%rax			# Push return value on old stack.
584	pushq	%rdx
585
586	call	__morestack_unblock_signals
587
588	popq	%rdx			# Restore return value.
589	popq	%rax
590	addq	$8,%rsp
591
592	.cfi_remember_state
593	popq	%rbp
594	.cfi_restore %rbp
595	.cfi_def_cfa %rsp, 16
596	ret				# Return to caller, which will
597					# immediately return.
598
599# This is the cleanup code called by the stack unwinder when unwinding
600# through the code between .LEHB0 and .LEHE0 above.
601
602.L1:
603	.cfi_restore_state
604	subq	$16,%rsp		# Maintain 16 byte alignment.
605	movq	%rax,(%rsp)		# Save exception header.
606	movq	%rbp,%rdi		# Stack pointer after resume.
607	call	__generic_findstack
608	movq	%rbp,%rcx		# Get the stack pointer.
609	subq	%rax,%rcx		# Subtract available space.
610	addq	$BACKOFF,%rcx		# Back off 1024 bytes.
611	X86_64_SAVE_NEW_STACK_BOUNDARY (cx)
612	movq	(%rsp),%rdi		# Restore exception data for call.
613#ifdef __PIC__
614	call	_Unwind_Resume@PLT	# Resume unwinding.
615#else
616	call	_Unwind_Resume		# Resume unwinding.
617#endif
618
619#endif /* defined(__x86_64__) */
620
621	.cfi_endproc
622#ifdef __ELF__
623	.size	__morestack, . - __morestack
624#endif
625
626#if !defined(__x86_64__) && defined(__PIC__)
627# Output the thunk to get PC into bx, since we use it above.
628	.section	.text.__x86.get_pc_thunk.bx,"axG",@progbits,__x86.get_pc_thunk.bx,comdat
629	.globl	__x86.get_pc_thunk.bx
630	.hidden	__x86.get_pc_thunk.bx
631#ifdef __ELF__
632	.type	__x86.get_pc_thunk.bx, @function
633#endif
634__x86.get_pc_thunk.bx:
635	.cfi_startproc
636	movl	(%esp), %ebx
637	ret
638	.cfi_endproc
639#ifdef __ELF__
640	.size	__x86.get_pc_thunk.bx, . - __x86.get_pc_thunk.bx
641#endif
642#endif
643
644# The exception table.  This tells the personality routine to execute
645# the exception handler.
646
647	.section	.gcc_except_table,"a",@progbits
648	.align	4
649.LLSDA1:
650	.byte	0xff	# @LPStart format (omit)
651	.byte	0xff	# @TType format (omit)
652	.byte	0x1	# call-site format (uleb128)
653	.uleb128 .LLSDACSE1-.LLSDACSB1	# Call-site table length
654.LLSDACSB1:
655	.uleb128 .LEHB0-.LFB1	# region 0 start
656	.uleb128 .LEHE0-.LEHB0	# length
657	.uleb128 .L1-.LFB1	# landing pad
658	.uleb128 0		# action
659.LLSDACSE1:
660
661
662	.global __gcc_personality_v0
663#ifdef __PIC__
664	# Build a position independent reference to the basic
665        # personality function.
666	.hidden DW.ref.__gcc_personality_v0
667	.weak   DW.ref.__gcc_personality_v0
668	.section .data.DW.ref.__gcc_personality_v0,"awG",@progbits,DW.ref.__gcc_personality_v0,comdat
669	.type	DW.ref.__gcc_personality_v0, @object
670DW.ref.__gcc_personality_v0:
671#ifndef __LP64__
672	.align 4
673	.size	DW.ref.__gcc_personality_v0, 4
674	.long	__gcc_personality_v0
675#else
676	.align 8
677	.size	DW.ref.__gcc_personality_v0, 8
678	.quad	__gcc_personality_v0
679#endif
680#endif
681
682#if defined __x86_64__ && defined __LP64__
683
684# This entry point is used for the large model.  With this entry point
685# the upper 32 bits of %r10 hold the argument size and the lower 32
686# bits hold the new stack frame size.  There doesn't seem to be a way
687# to know in the assembler code that we are assembling for the large
688# model, and there doesn't seem to be a large model multilib anyhow.
689# If one is developed, then the non-PIC code is probably OK since we
690# will probably be close to the morestack code, but the PIC code
691# almost certainly needs to be changed.  FIXME.
692
693	.text
694	.global	__morestack_large_model
695	.hidden	__morestack_large_model
696
697#ifdef __ELF__
698	.type	__morestack_large_model,@function
699#endif
700
701__morestack_large_model:
702
703	.cfi_startproc
704
705	movq	%r10, %r11
706	andl	$0xffffffff, %r10d
707	sarq	$32, %r11
708	jmp	__morestack
709
710	.cfi_endproc
711#ifdef __ELF__
712       .size	__morestack_large_model, . - __morestack_large_model
713#endif
714
715#endif /* __x86_64__ && __LP64__ */
716
717# Initialize the stack test value when the program starts or when a
718# new thread starts.  We don't know how large the main stack is, so we
719# guess conservatively.  We might be able to use getrlimit here.
720
721	.text
722	.global	__stack_split_initialize
723	.hidden	__stack_split_initialize
724
725#ifdef __ELF__
726	.type	__stack_split_initialize, @function
727#endif
728
729__stack_split_initialize:
730
731#ifndef __x86_64__
732
733	leal	-16000(%esp),%eax	# We should have at least 16K.
734	movl	%eax,%gs:0x30
735	pushl	$16000
736	pushl	%esp
737#ifdef __PIC__
738	call	__generic_morestack_set_initial_sp@PLT
739#else
740	call	__generic_morestack_set_initial_sp
741#endif
742	addl	$8,%esp
743	ret
744
745#else /* defined(__x86_64__) */
746
747	leaq	-16000(%rsp),%rax	# We should have at least 16K.
748	X86_64_SAVE_NEW_STACK_BOUNDARY (ax)
749	movq	%rsp,%rdi
750	movq	$16000,%rsi
751#ifdef __PIC__
752	call	__generic_morestack_set_initial_sp@PLT
753#else
754	call	__generic_morestack_set_initial_sp
755#endif
756	ret
757
758#endif /* defined(__x86_64__) */
759
760#ifdef __ELF__
761	.size	__stack_split_initialize, . - __stack_split_initialize
762#endif
763
764# Routines to get and set the guard, for __splitstack_getcontext,
765# __splitstack_setcontext, and __splitstack_makecontext.
766
767# void *__morestack_get_guard (void) returns the current stack guard.
768	.text
769	.global	__morestack_get_guard
770	.hidden	__morestack_get_guard
771
772#ifdef __ELF__
773	.type	__morestack_get_guard,@function
774#endif
775
776__morestack_get_guard:
777
778#ifndef __x86_64__
779	movl	%gs:0x30,%eax
780#else
781#ifdef __LP64__
782	movq	%fs:0x70,%rax
783#else
784	movl	%fs:0x40,%eax
785#endif
786#endif
787	ret
788
789#ifdef __ELF__
790	.size	__morestack_get_guard, . - __morestack_get_guard
791#endif
792
793# void __morestack_set_guard (void *) sets the stack guard.
794	.global	__morestack_set_guard
795	.hidden	__morestack_set_guard
796
797#ifdef __ELF__
798	.type	__morestack_set_guard,@function
799#endif
800
801__morestack_set_guard:
802
803#ifndef __x86_64__
804	movl	4(%esp),%eax
805	movl	%eax,%gs:0x30
806#else
807	X86_64_SAVE_NEW_STACK_BOUNDARY (di)
808#endif
809	ret
810
811#ifdef __ELF__
812	.size	__morestack_set_guard, . - __morestack_set_guard
813#endif
814
815# void *__morestack_make_guard (void *, size_t) returns the stack
816# guard value for a stack.
817	.global	__morestack_make_guard
818	.hidden	__morestack_make_guard
819
820#ifdef __ELF__
821	.type	__morestack_make_guard,@function
822#endif
823
824__morestack_make_guard:
825
826#ifndef __x86_64__
827	movl	4(%esp),%eax
828	subl	8(%esp),%eax
829	addl	$BACKOFF,%eax
830#else
831	subq	%rsi,%rdi
832	addq	$BACKOFF,%rdi
833	movq	%rdi,%rax
834#endif
835	ret
836
837#ifdef __ELF__
838	.size	__morestack_make_guard, . - __morestack_make_guard
839#endif
840
841# Make __stack_split_initialize a high priority constructor.  FIXME:
842# This is ELF specific.
843
844	.section	.ctors.65535,"aw",@progbits
845
846#ifndef __LP64__
847	.align	4
848	.long	__stack_split_initialize
849	.long	__morestack_load_mmap
850#else
851	.align	8
852	.quad	__stack_split_initialize
853	.quad	__morestack_load_mmap
854#endif
855
856#ifdef __ELF__
857	.section	.note.GNU-stack,"",@progbits
858	.section	.note.GNU-split-stack,"",@progbits
859	.section	.note.GNU-no-split-stack,"",@progbits
860#endif
861