1// Copyright 2014 The Go Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style
3// license that can be found in the LICENSE file.
4
5// +build ppc64 ppc64le
6
7#include "go_asm.h"
8#include "go_tls.h"
9#include "funcdata.h"
10#include "textflag.h"
11#include "asm_ppc64x.h"
12
13TEXT runtime·rt0_go(SB),NOSPLIT,$0
14	// R1 = stack; R3 = argc; R4 = argv; R13 = C TLS base pointer
15
16	// initialize essential registers
17	BL	runtime·reginit(SB)
18
19	SUB	$(FIXED_FRAME+16), R1
20	MOVD	R2, 24(R1)		// stash the TOC pointer away again now we've created a new frame
21	MOVW	R3, FIXED_FRAME+0(R1)	// argc
22	MOVD	R4, FIXED_FRAME+8(R1)	// argv
23
24	// create istack out of the given (operating system) stack.
25	// _cgo_init may update stackguard.
26	MOVD	$runtime·g0(SB), g
27	MOVD	$(-64*1024), R31
28	ADD	R31, R1, R3
29	MOVD	R3, g_stackguard0(g)
30	MOVD	R3, g_stackguard1(g)
31	MOVD	R3, (g_stack+stack_lo)(g)
32	MOVD	R1, (g_stack+stack_hi)(g)
33
34	// if there is a _cgo_init, call it using the gcc ABI.
35	MOVD	_cgo_init(SB), R12
36	CMP	R0, R12
37	BEQ	nocgo
38	MOVD	R12, CTR		// r12 = "global function entry point"
39	MOVD	R13, R5			// arg 2: TLS base pointer
40	MOVD	$setg_gcc<>(SB), R4 	// arg 1: setg
41	MOVD	g, R3			// arg 0: G
42	// C functions expect 32 bytes of space on caller stack frame
43	// and a 16-byte aligned R1
44	MOVD	R1, R14			// save current stack
45	SUB	$32, R1			// reserve 32 bytes
46	RLDCR	$0, R1, $~15, R1	// 16-byte align
47	BL	(CTR)			// may clobber R0, R3-R12
48	MOVD	R14, R1			// restore stack
49	MOVD	24(R1), R2
50	XOR	R0, R0			// fix R0
51
52nocgo:
53	// update stackguard after _cgo_init
54	MOVD	(g_stack+stack_lo)(g), R3
55	ADD	$const__StackGuard, R3
56	MOVD	R3, g_stackguard0(g)
57	MOVD	R3, g_stackguard1(g)
58
59	// set the per-goroutine and per-mach "registers"
60	MOVD	$runtime·m0(SB), R3
61
62	// save m->g0 = g0
63	MOVD	g, m_g0(R3)
64	// save m0 to g0->m
65	MOVD	R3, g_m(g)
66
67	BL	runtime·check(SB)
68
69	// args are already prepared
70	BL	runtime·args(SB)
71	BL	runtime·osinit(SB)
72	BL	runtime·schedinit(SB)
73
74	// create a new goroutine to start program
75	MOVD	$runtime·mainPC(SB), R3		// entry
76	MOVDU	R3, -8(R1)
77	MOVDU	R0, -8(R1)
78	MOVDU	R0, -8(R1)
79	MOVDU	R0, -8(R1)
80	MOVDU	R0, -8(R1)
81	MOVDU	R0, -8(R1)
82	BL	runtime·newproc(SB)
83	ADD	$(16+FIXED_FRAME), R1
84
85	// start this M
86	BL	runtime·mstart(SB)
87
88	MOVD	R0, 1(R0)
89	RET
90
91DATA	runtime·mainPC+0(SB)/8,$runtime·main(SB)
92GLOBL	runtime·mainPC(SB),RODATA,$8
93
94TEXT runtime·breakpoint(SB),NOSPLIT|NOFRAME,$0-0
95	MOVD	R0, 2(R0) // TODO: TD
96	RET
97
98TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
99	RET
100
101TEXT _cgo_reginit(SB),NOSPLIT|NOFRAME,$0-0
102	// crosscall_ppc64 and crosscall2 need to reginit, but can't
103	// get at the 'runtime.reginit' symbol.
104	BR	runtime·reginit(SB)
105
106TEXT runtime·reginit(SB),NOSPLIT|NOFRAME,$0-0
107	// set R0 to zero, it's expected by the toolchain
108	XOR R0, R0
109	RET
110
111/*
112 *  go-routine
113 */
114
115// void gosave(Gobuf*)
116// save state in Gobuf; setjmp
117TEXT runtime·gosave(SB), NOSPLIT|NOFRAME, $0-8
118	MOVD	buf+0(FP), R3
119	MOVD	R1, gobuf_sp(R3)
120	MOVD	LR, R31
121	MOVD	R31, gobuf_pc(R3)
122	MOVD	g, gobuf_g(R3)
123	MOVD	R0, gobuf_lr(R3)
124	MOVD	R0, gobuf_ret(R3)
125	// Assert ctxt is zero. See func save.
126	MOVD	gobuf_ctxt(R3), R3
127	CMP	R0, R3
128	BEQ	2(PC)
129	BL	runtime·badctxt(SB)
130	RET
131
132// void gogo(Gobuf*)
133// restore state from Gobuf; longjmp
134TEXT runtime·gogo(SB), NOSPLIT, $16-8
135	MOVD	buf+0(FP), R5
136
137	// If ctxt is not nil, invoke deletion barrier before overwriting.
138	MOVD	gobuf_ctxt(R5), R3
139	CMP	R0, R3
140	BEQ	nilctxt
141	MOVD	$gobuf_ctxt(R5), R3
142	MOVD	R3, FIXED_FRAME+0(R1)
143	MOVD	R0, FIXED_FRAME+8(R1)
144	BL	runtime·writebarrierptr_prewrite(SB)
145	MOVD	buf+0(FP), R5
146
147nilctxt:
148	MOVD	gobuf_g(R5), g	// make sure g is not nil
149	BL	runtime·save_g(SB)
150
151	MOVD	0(g), R4
152	MOVD	gobuf_sp(R5), R1
153	MOVD	gobuf_lr(R5), R31
154	MOVD	R31, LR
155	MOVD	gobuf_ret(R5), R3
156	MOVD	gobuf_ctxt(R5), R11
157	MOVD	R0, gobuf_sp(R5)
158	MOVD	R0, gobuf_ret(R5)
159	MOVD	R0, gobuf_lr(R5)
160	MOVD	R0, gobuf_ctxt(R5)
161	CMP	R0, R0 // set condition codes for == test, needed by stack split
162	MOVD	gobuf_pc(R5), R12
163	MOVD	R12, CTR
164	BR	(CTR)
165
166// void mcall(fn func(*g))
167// Switch to m->g0's stack, call fn(g).
168// Fn must never return. It should gogo(&g->sched)
169// to keep running g.
170TEXT runtime·mcall(SB), NOSPLIT|NOFRAME, $0-8
171	// Save caller state in g->sched
172	MOVD	R1, (g_sched+gobuf_sp)(g)
173	MOVD	LR, R31
174	MOVD	R31, (g_sched+gobuf_pc)(g)
175	MOVD	R0, (g_sched+gobuf_lr)(g)
176	MOVD	g, (g_sched+gobuf_g)(g)
177
178	// Switch to m->g0 & its stack, call fn.
179	MOVD	g, R3
180	MOVD	g_m(g), R8
181	MOVD	m_g0(R8), g
182	BL	runtime·save_g(SB)
183	CMP	g, R3
184	BNE	2(PC)
185	BR	runtime·badmcall(SB)
186	MOVD	fn+0(FP), R11			// context
187	MOVD	0(R11), R12			// code pointer
188	MOVD	R12, CTR
189	MOVD	(g_sched+gobuf_sp)(g), R1	// sp = m->g0->sched.sp
190	MOVDU	R3, -8(R1)
191	MOVDU	R0, -8(R1)
192	MOVDU	R0, -8(R1)
193	MOVDU	R0, -8(R1)
194	MOVDU	R0, -8(R1)
195	BL	(CTR)
196	MOVD	24(R1), R2
197	BR	runtime·badmcall2(SB)
198
199// systemstack_switch is a dummy routine that systemstack leaves at the bottom
200// of the G stack. We need to distinguish the routine that
201// lives at the bottom of the G stack from the one that lives
202// at the top of the system stack because the one at the top of
203// the system stack terminates the stack walk (see topofstack()).
204TEXT runtime·systemstack_switch(SB), NOSPLIT, $0-0
205	// We have several undefs here so that 16 bytes past
206	// $runtime·systemstack_switch lies within them whether or not the
207        // instructions that derive r2 from r12 are there.
208	UNDEF
209	UNDEF
210	UNDEF
211	BL	(LR)	// make sure this function is not leaf
212	RET
213
214// func systemstack(fn func())
215TEXT runtime·systemstack(SB), NOSPLIT, $0-8
216	MOVD	fn+0(FP), R3	// R3 = fn
217	MOVD	R3, R11		// context
218	MOVD	g_m(g), R4	// R4 = m
219
220	MOVD	m_gsignal(R4), R5	// R5 = gsignal
221	CMP	g, R5
222	BEQ	noswitch
223
224	MOVD	m_g0(R4), R5	// R5 = g0
225	CMP	g, R5
226	BEQ	noswitch
227
228	MOVD	m_curg(R4), R6
229	CMP	g, R6
230	BEQ	switch
231
232	// Bad: g is not gsignal, not g0, not curg. What is it?
233	// Hide call from linker nosplit analysis.
234	MOVD	$runtime·badsystemstack(SB), R12
235	MOVD	R12, CTR
236	BL	(CTR)
237
238switch:
239	// save our state in g->sched. Pretend to
240	// be systemstack_switch if the G stack is scanned.
241	MOVD	$runtime·systemstack_switch(SB), R6
242	ADD     $16, R6 // get past prologue (including r2-setting instructions when they're there)
243	MOVD	R6, (g_sched+gobuf_pc)(g)
244	MOVD	R1, (g_sched+gobuf_sp)(g)
245	MOVD	R0, (g_sched+gobuf_lr)(g)
246	MOVD	g, (g_sched+gobuf_g)(g)
247
248	// switch to g0
249	MOVD	R5, g
250	BL	runtime·save_g(SB)
251	MOVD	(g_sched+gobuf_sp)(g), R3
252	// make it look like mstart called systemstack on g0, to stop traceback
253	SUB	$FIXED_FRAME, R3
254	MOVD	$runtime·mstart(SB), R4
255	MOVD	R4, 0(R3)
256	MOVD	R3, R1
257
258	// call target function
259	MOVD	0(R11), R12	// code pointer
260	MOVD	R12, CTR
261	BL	(CTR)
262
263	// restore TOC pointer. It seems unlikely that we will use systemstack
264	// to call a function defined in another module, but the results of
265	// doing so would be so confusing that it's worth doing this.
266	MOVD	g_m(g), R3
267	MOVD	m_curg(R3), g
268	MOVD	(g_sched+gobuf_sp)(g), R3
269	MOVD	24(R3), R2
270	// switch back to g
271	MOVD	g_m(g), R3
272	MOVD	m_curg(R3), g
273	BL	runtime·save_g(SB)
274	MOVD	(g_sched+gobuf_sp)(g), R1
275	MOVD	R0, (g_sched+gobuf_sp)(g)
276	RET
277
278noswitch:
279	// already on m stack, just call directly
280	MOVD	0(R11), R12	// code pointer
281	MOVD	R12, CTR
282	BL	(CTR)
283	MOVD	24(R1), R2
284	RET
285
286/*
287 * support for morestack
288 */
289
290// Called during function prolog when more stack is needed.
291// Caller has already loaded:
292// R3: framesize, R4: argsize, R5: LR
293//
294// The traceback routines see morestack on a g0 as being
295// the top of a stack (for example, morestack calling newstack
296// calling the scheduler calling newm calling gc), so we must
297// record an argument size. For that purpose, it has no arguments.
298TEXT runtime·morestack(SB),NOSPLIT|NOFRAME,$0-0
299	// Cannot grow scheduler stack (m->g0).
300	MOVD	g_m(g), R7
301	MOVD	m_g0(R7), R8
302	CMP	g, R8
303	BNE	3(PC)
304	BL	runtime·badmorestackg0(SB)
305	BL	runtime·abort(SB)
306
307	// Cannot grow signal stack (m->gsignal).
308	MOVD	m_gsignal(R7), R8
309	CMP	g, R8
310	BNE	3(PC)
311	BL	runtime·badmorestackgsignal(SB)
312	BL	runtime·abort(SB)
313
314	// Called from f.
315	// Set g->sched to context in f.
316	MOVD	R1, (g_sched+gobuf_sp)(g)
317	MOVD	LR, R8
318	MOVD	R8, (g_sched+gobuf_pc)(g)
319	MOVD	R5, (g_sched+gobuf_lr)(g)
320	// newstack will fill gobuf.ctxt.
321
322	// Called from f.
323	// Set m->morebuf to f's caller.
324	MOVD	R5, (m_morebuf+gobuf_pc)(R7)	// f's caller's PC
325	MOVD	R1, (m_morebuf+gobuf_sp)(R7)	// f's caller's SP
326	MOVD	g, (m_morebuf+gobuf_g)(R7)
327
328	// Call newstack on m->g0's stack.
329	MOVD	m_g0(R7), g
330	BL	runtime·save_g(SB)
331	MOVD	(g_sched+gobuf_sp)(g), R1
332	MOVDU   R0, -(FIXED_FRAME+8)(R1)	// create a call frame on g0
333	MOVD	R11, FIXED_FRAME+0(R1)	// ctxt argument
334	BL	runtime·newstack(SB)
335
336	// Not reached, but make sure the return PC from the call to newstack
337	// is still in this function, and not the beginning of the next.
338	UNDEF
339
340TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
341	MOVD	R0, R11
342	BR	runtime·morestack(SB)
343
344TEXT runtime·stackBarrier(SB),NOSPLIT,$0
345	// We came here via a RET to an overwritten LR.
346	// R3 may be live. Other registers are available.
347
348	// Get the original return PC, g.stkbar[g.stkbarPos].savedLRVal.
349	MOVD	(g_stkbar+slice_array)(g), R4
350	MOVD	g_stkbarPos(g), R5
351	MOVD	$stkbar__size, R6
352	MULLD	R5, R6
353	ADD	R4, R6
354	MOVD	stkbar_savedLRVal(R6), R6
355	// Record that this stack barrier was hit.
356	ADD	$1, R5
357	MOVD	R5, g_stkbarPos(g)
358	// Jump to the original return PC.
359	MOVD	R6, CTR
360	BR	(CTR)
361
362// reflectcall: call a function with the given argument list
363// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
364// we don't have variable-sized frames, so we use a small number
365// of constant-sized-frame functions to encode a few bits of size in the pc.
366// Caution: ugly multiline assembly macros in your future!
367
368#define DISPATCH(NAME,MAXSIZE)		\
369	MOVD	$MAXSIZE, R31;		\
370	CMP	R3, R31;		\
371	BGT	4(PC);			\
372	MOVD	$NAME(SB), R12;		\
373	MOVD	R12, CTR;		\
374	BR	(CTR)
375// Note: can't just "BR NAME(SB)" - bad inlining results.
376
377TEXT reflect·call(SB), NOSPLIT, $0-0
378	BR	·reflectcall(SB)
379
380TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
381	MOVWZ argsize+24(FP), R3
382	DISPATCH(runtime·call32, 32)
383	DISPATCH(runtime·call64, 64)
384	DISPATCH(runtime·call128, 128)
385	DISPATCH(runtime·call256, 256)
386	DISPATCH(runtime·call512, 512)
387	DISPATCH(runtime·call1024, 1024)
388	DISPATCH(runtime·call2048, 2048)
389	DISPATCH(runtime·call4096, 4096)
390	DISPATCH(runtime·call8192, 8192)
391	DISPATCH(runtime·call16384, 16384)
392	DISPATCH(runtime·call32768, 32768)
393	DISPATCH(runtime·call65536, 65536)
394	DISPATCH(runtime·call131072, 131072)
395	DISPATCH(runtime·call262144, 262144)
396	DISPATCH(runtime·call524288, 524288)
397	DISPATCH(runtime·call1048576, 1048576)
398	DISPATCH(runtime·call2097152, 2097152)
399	DISPATCH(runtime·call4194304, 4194304)
400	DISPATCH(runtime·call8388608, 8388608)
401	DISPATCH(runtime·call16777216, 16777216)
402	DISPATCH(runtime·call33554432, 33554432)
403	DISPATCH(runtime·call67108864, 67108864)
404	DISPATCH(runtime·call134217728, 134217728)
405	DISPATCH(runtime·call268435456, 268435456)
406	DISPATCH(runtime·call536870912, 536870912)
407	DISPATCH(runtime·call1073741824, 1073741824)
408	MOVD	$runtime·badreflectcall(SB), R12
409	MOVD	R12, CTR
410	BR	(CTR)
411
412#define CALLFN(NAME,MAXSIZE)			\
413TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
414	NO_LOCAL_POINTERS;			\
415	/* copy arguments to stack */		\
416	MOVD	arg+16(FP), R3;			\
417	MOVWZ	argsize+24(FP), R4;			\
418	MOVD	R1, R5;				\
419	ADD	$(FIXED_FRAME-1), R5;			\
420	SUB	$1, R3;				\
421	ADD	R5, R4;				\
422	CMP	R5, R4;				\
423	BEQ	4(PC);				\
424	MOVBZU	1(R3), R6;			\
425	MOVBZU	R6, 1(R5);			\
426	BR	-4(PC);				\
427	/* call function */			\
428	MOVD	f+8(FP), R11;			\
429	MOVD	(R11), R12;			\
430	MOVD	R12, CTR;			\
431	PCDATA  $PCDATA_StackMapIndex, $0;	\
432	BL	(CTR);				\
433	MOVD	24(R1), R2;			\
434	/* copy return values back */		\
435	MOVD	argtype+0(FP), R7;		\
436	MOVD	arg+16(FP), R3;			\
437	MOVWZ	n+24(FP), R4;			\
438	MOVWZ	retoffset+28(FP), R6;		\
439	ADD	$FIXED_FRAME, R1, R5;		\
440	ADD	R6, R5; 			\
441	ADD	R6, R3;				\
442	SUB	R6, R4;				\
443	BL	callRet<>(SB);			\
444	RET
445
446// callRet copies return values back at the end of call*. This is a
447// separate function so it can allocate stack space for the arguments
448// to reflectcallmove. It does not follow the Go ABI; it expects its
449// arguments in registers.
450TEXT callRet<>(SB), NOSPLIT, $32-0
451	MOVD	R7, FIXED_FRAME+0(R1)
452	MOVD	R3, FIXED_FRAME+8(R1)
453	MOVD	R5, FIXED_FRAME+16(R1)
454	MOVD	R4, FIXED_FRAME+24(R1)
455	BL	runtime·reflectcallmove(SB)
456	RET
457
458CALLFNcall32, 32)
459CALLFNcall64, 64)
460CALLFNcall128, 128)
461CALLFNcall256, 256)
462CALLFNcall512, 512)
463CALLFNcall1024, 1024)
464CALLFNcall2048, 2048)
465CALLFNcall4096, 4096)
466CALLFNcall8192, 8192)
467CALLFNcall16384, 16384)
468CALLFNcall32768, 32768)
469CALLFNcall65536, 65536)
470CALLFNcall131072, 131072)
471CALLFNcall262144, 262144)
472CALLFNcall524288, 524288)
473CALLFNcall1048576, 1048576)
474CALLFNcall2097152, 2097152)
475CALLFNcall4194304, 4194304)
476CALLFNcall8388608, 8388608)
477CALLFNcall16777216, 16777216)
478CALLFNcall33554432, 33554432)
479CALLFNcall67108864, 67108864)
480CALLFNcall134217728, 134217728)
481CALLFNcall268435456, 268435456)
482CALLFNcall536870912, 536870912)
483CALLFNcall1073741824, 1073741824)
484
485TEXT runtime·procyield(SB),NOSPLIT,$0-0
486	RET
487
488// void jmpdefer(fv, sp);
489// called from deferreturn.
490// 1. grab stored LR for caller
491// 2. sub 8 bytes to get back to either nop or toc reload before deferreturn
492// 3. BR to fn
493// When dynamically linking Go, it is not sufficient to rewind to the BL
494// deferreturn -- we might be jumping between modules and so we need to reset
495// the TOC pointer in r2. To do this, codegen inserts MOVD 24(R1), R2 *before*
496// the BL deferreturn and jmpdefer rewinds to that.
497TEXT runtime·jmpdefer(SB), NOSPLIT|NOFRAME, $0-16
498	MOVD	0(R1), R31
499	SUB     $8, R31
500	MOVD	R31, LR
501
502	MOVD	fv+0(FP), R11
503	MOVD	argp+8(FP), R1
504	SUB	$FIXED_FRAME, R1
505	MOVD	0(R11), R12
506	MOVD	R12, CTR
507	BR	(CTR)
508
509// Save state of caller into g->sched. Smashes R31.
510TEXT gosave<>(SB),NOSPLIT|NOFRAME,$0
511	MOVD	LR, R31
512	MOVD	R31, (g_sched+gobuf_pc)(g)
513	MOVD	R1, (g_sched+gobuf_sp)(g)
514	MOVD	R0, (g_sched+gobuf_lr)(g)
515	MOVD	R0, (g_sched+gobuf_ret)(g)
516	// Assert ctxt is zero. See func save.
517	MOVD	(g_sched+gobuf_ctxt)(g), R31
518	CMP	R0, R31
519	BEQ	2(PC)
520	BL	runtime·badctxt(SB)
521	RET
522
523// func asmcgocall(fn, arg unsafe.Pointer) int32
524// Call fn(arg) on the scheduler stack,
525// aligned appropriately for the gcc ABI.
526// See cgocall.go for more details.
527TEXT ·asmcgocall(SB),NOSPLIT,$0-20
528	MOVD	fn+0(FP), R3
529	MOVD	arg+8(FP), R4
530
531	MOVD	R1, R7		// save original stack pointer
532	MOVD	g, R5
533
534	// Figure out if we need to switch to m->g0 stack.
535	// We get called to create new OS threads too, and those
536	// come in on the m->g0 stack already.
537	MOVD	g_m(g), R6
538	MOVD	m_g0(R6), R6
539	CMP	R6, g
540	BEQ	g0
541	BL	gosave<>(SB)
542	MOVD	R6, g
543	BL	runtime·save_g(SB)
544	MOVD	(g_sched+gobuf_sp)(g), R1
545
546	// Now on a scheduling stack (a pthread-created stack).
547g0:
548	// Save room for two of our pointers, plus 32 bytes of callee
549	// save area that lives on the caller stack.
550	SUB	$48, R1
551	RLDCR	$0, R1, $~15, R1	// 16-byte alignment for gcc ABI
552	MOVD	R5, 40(R1)	// save old g on stack
553	MOVD	(g_stack+stack_hi)(R5), R5
554	SUB	R7, R5
555	MOVD	R5, 32(R1)	// save depth in old g stack (can't just save SP, as stack might be copied during a callback)
556	MOVD	R0, 0(R1)	// clear back chain pointer (TODO can we give it real back trace information?)
557	// This is a "global call", so put the global entry point in r12
558	MOVD	R3, R12
559	MOVD	R12, CTR
560	MOVD	R4, R3		// arg in r3
561	BL	(CTR)
562
563	// C code can clobber R0, so set it back to 0.  F27-F31 are
564	// callee save, so we don't need to recover those.
565	XOR	R0, R0
566	// Restore g, stack pointer, toc pointer.
567	// R3 is errno, so don't touch it
568	MOVD	40(R1), g
569	MOVD    (g_stack+stack_hi)(g), R5
570	MOVD    32(R1), R6
571	SUB     R6, R5
572	MOVD    24(R5), R2
573	BL	runtime·save_g(SB)
574	MOVD	(g_stack+stack_hi)(g), R5
575	MOVD	32(R1), R6
576	SUB	R6, R5
577	MOVD	R5, R1
578
579	MOVW	R3, ret+16(FP)
580	RET
581
582// cgocallback(void (*fn)(void*), void *frame, uintptr framesize, uintptr ctxt)
583// Turn the fn into a Go func (by taking its address) and call
584// cgocallback_gofunc.
585TEXT runtime·cgocallback(SB),NOSPLIT,$32-32
586	MOVD	$fn+0(FP), R3
587	MOVD	R3, FIXED_FRAME+0(R1)
588	MOVD	frame+8(FP), R3
589	MOVD	R3, FIXED_FRAME+8(R1)
590	MOVD	framesize+16(FP), R3
591	MOVD	R3, FIXED_FRAME+16(R1)
592	MOVD	ctxt+24(FP), R3
593	MOVD	R3, FIXED_FRAME+24(R1)
594	MOVD	$runtime·cgocallback_gofunc(SB), R12
595	MOVD	R12, CTR
596	BL	(CTR)
597	RET
598
599// cgocallback_gofunc(FuncVal*, void *frame, uintptr framesize, uintptr ctxt)
600// See cgocall.go for more details.
601TEXT ·cgocallback_gofunc(SB),NOSPLIT,$16-32
602	NO_LOCAL_POINTERS
603
604	// Load m and g from thread-local storage.
605	MOVB	runtime·iscgo(SB), R3
606	CMP	R3, $0
607	BEQ	nocgo
608	BL	runtime·load_g(SB)
609nocgo:
610
611	// If g is nil, Go did not create the current thread.
612	// Call needm to obtain one for temporary use.
613	// In this case, we're running on the thread stack, so there's
614	// lots of space, but the linker doesn't know. Hide the call from
615	// the linker analysis by using an indirect call.
616	CMP	g, $0
617	BEQ	needm
618
619	MOVD	g_m(g), R8
620	MOVD	R8, savedm-8(SP)
621	BR	havem
622
623needm:
624	MOVD	g, savedm-8(SP) // g is zero, so is m.
625	MOVD	$runtime·needm(SB), R12
626	MOVD	R12, CTR
627	BL	(CTR)
628
629	// Set m->sched.sp = SP, so that if a panic happens
630	// during the function we are about to execute, it will
631	// have a valid SP to run on the g0 stack.
632	// The next few lines (after the havem label)
633	// will save this SP onto the stack and then write
634	// the same SP back to m->sched.sp. That seems redundant,
635	// but if an unrecovered panic happens, unwindm will
636	// restore the g->sched.sp from the stack location
637	// and then systemstack will try to use it. If we don't set it here,
638	// that restored SP will be uninitialized (typically 0) and
639	// will not be usable.
640	MOVD	g_m(g), R8
641	MOVD	m_g0(R8), R3
642	MOVD	R1, (g_sched+gobuf_sp)(R3)
643
644havem:
645	// Now there's a valid m, and we're running on its m->g0.
646	// Save current m->g0->sched.sp on stack and then set it to SP.
647	// Save current sp in m->g0->sched.sp in preparation for
648	// switch back to m->curg stack.
649	// NOTE: unwindm knows that the saved g->sched.sp is at 8(R1) aka savedsp-16(SP).
650	MOVD	m_g0(R8), R3
651	MOVD	(g_sched+gobuf_sp)(R3), R4
652	MOVD	R4, savedsp-16(SP)
653	MOVD	R1, (g_sched+gobuf_sp)(R3)
654
655	// Switch to m->curg stack and call runtime.cgocallbackg.
656	// Because we are taking over the execution of m->curg
657	// but *not* resuming what had been running, we need to
658	// save that information (m->curg->sched) so we can restore it.
659	// We can restore m->curg->sched.sp easily, because calling
660	// runtime.cgocallbackg leaves SP unchanged upon return.
661	// To save m->curg->sched.pc, we push it onto the stack.
662	// This has the added benefit that it looks to the traceback
663	// routine like cgocallbackg is going to return to that
664	// PC (because the frame we allocate below has the same
665	// size as cgocallback_gofunc's frame declared above)
666	// so that the traceback will seamlessly trace back into
667	// the earlier calls.
668	//
669	// In the new goroutine, -8(SP) is unused (where SP refers to
670	// m->curg's SP while we're setting it up, before we've adjusted it).
671	MOVD	m_curg(R8), g
672	BL	runtime·save_g(SB)
673	MOVD	(g_sched+gobuf_sp)(g), R4 // prepare stack as R4
674	MOVD	(g_sched+gobuf_pc)(g), R5
675	MOVD	R5, -(FIXED_FRAME+16)(R4)
676	MOVD	ctxt+24(FP), R3
677	MOVD	R3, -16(R4)
678	MOVD	$-(FIXED_FRAME+16)(R4), R1
679	BL	runtime·cgocallbackg(SB)
680
681	// Restore g->sched (== m->curg->sched) from saved values.
682	MOVD	0(R1), R5
683	MOVD	R5, (g_sched+gobuf_pc)(g)
684	MOVD	$(FIXED_FRAME+16)(R1), R4
685	MOVD	R4, (g_sched+gobuf_sp)(g)
686
687	// Switch back to m->g0's stack and restore m->g0->sched.sp.
688	// (Unlike m->curg, the g0 goroutine never uses sched.pc,
689	// so we do not have to restore it.)
690	MOVD	g_m(g), R8
691	MOVD	m_g0(R8), g
692	BL	runtime·save_g(SB)
693	MOVD	(g_sched+gobuf_sp)(g), R1
694	MOVD	savedsp-16(SP), R4
695	MOVD	R4, (g_sched+gobuf_sp)(g)
696
697	// If the m on entry was nil, we called needm above to borrow an m
698	// for the duration of the call. Since the call is over, return it with dropm.
699	MOVD	savedm-8(SP), R6
700	CMP	R6, $0
701	BNE	droppedm
702	MOVD	$runtime·dropm(SB), R12
703	MOVD	R12, CTR
704	BL	(CTR)
705droppedm:
706
707	// Done!
708	RET
709
710// void setg(G*); set g. for use by needm.
711TEXT runtime·setg(SB), NOSPLIT, $0-8
712	MOVD	gg+0(FP), g
713	// This only happens if iscgo, so jump straight to save_g
714	BL	runtime·save_g(SB)
715	RET
716
717// void setg_gcc(G*); set g in C TLS.
718// Must obey the gcc calling convention.
719TEXT setg_gcc<>(SB),NOSPLIT|NOFRAME,$0-0
720	// The standard prologue clobbers R31, which is callee-save in
721	// the C ABI, so we have to use $-8-0 and save LR ourselves.
722	MOVD	LR, R4
723	// Also save g and R31, since they're callee-save in C ABI
724	MOVD	R31, R5
725	MOVD	g, R6
726
727	MOVD	R3, g
728	BL	runtime·save_g(SB)
729
730	MOVD	R6, g
731	MOVD	R5, R31
732	MOVD	R4, LR
733	RET
734
735TEXT runtime·getcallerpc(SB),NOSPLIT,$8-16
736	MOVD	FIXED_FRAME+8(R1), R3		// LR saved by caller
737	MOVD	runtime·stackBarrierPC(SB), R4
738	CMP	R3, R4
739	BNE	nobar
740	// Get original return PC.
741	BL	runtime·nextBarrierPC(SB)
742	MOVD	FIXED_FRAME+0(R1), R3
743nobar:
744	MOVD	R3, ret+8(FP)
745	RET
746
747TEXT runtime·setcallerpc(SB),NOSPLIT,$8-16
748	MOVD	pc+8(FP), R3
749	MOVD	FIXED_FRAME+8(R1), R4
750	MOVD	runtime·stackBarrierPC(SB), R5
751	CMP	R4, R5
752	BEQ	setbar
753	MOVD	R3, FIXED_FRAME+8(R1)		// set LR in caller
754	RET
755setbar:
756	// Set the stack barrier return PC.
757	MOVD	R3, FIXED_FRAME+0(R1)
758	BL	runtime·setNextBarrierPC(SB)
759	RET
760
761TEXT runtime·abort(SB),NOSPLIT|NOFRAME,$0-0
762	MOVW	(R0), R0
763	UNDEF
764
765#define	TBRL	268
766#define	TBRU	269		/* Time base Upper/Lower */
767
768// int64 runtime·cputicks(void)
769TEXT runtime·cputicks(SB),NOSPLIT,$0-8
770	MOVW	SPR(TBRU), R4
771	MOVW	SPR(TBRL), R3
772	MOVW	SPR(TBRU), R5
773	CMPW	R4, R5
774	BNE	-4(PC)
775	SLD	$32, R5
776	OR	R5, R3
777	MOVD	R3, ret+0(FP)
778	RET
779
780// memhash_varlen(p unsafe.Pointer, h seed) uintptr
781// redirects to memhash(p, h, size) using the size
782// stored in the closure.
783TEXT runtime·memhash_varlen(SB),NOSPLIT,$40-24
784	GO_ARGS
785	NO_LOCAL_POINTERS
786	MOVD	p+0(FP), R3
787	MOVD	h+8(FP), R4
788	MOVD	8(R11), R5
789	MOVD	R3, FIXED_FRAME+0(R1)
790	MOVD	R4, FIXED_FRAME+8(R1)
791	MOVD	R5, FIXED_FRAME+16(R1)
792	BL	runtime·memhash(SB)
793	MOVD	FIXED_FRAME+24(R1), R3
794	MOVD	R3, ret+16(FP)
795	RET
796
797// AES hashing not implemented for ppc64
798TEXT runtime·aeshash(SB),NOSPLIT|NOFRAME,$0-0
799	MOVW	(R0), R1
800TEXT runtime·aeshash32(SB),NOSPLIT|NOFRAME,$0-0
801	MOVW	(R0), R1
802TEXT runtime·aeshash64(SB),NOSPLIT|NOFRAME,$0-0
803	MOVW	(R0), R1
804TEXT runtime·aeshashstr(SB),NOSPLIT|NOFRAME,$0-0
805	MOVW	(R0), R1
806
807TEXT runtime·memequal(SB),NOSPLIT,$0-25
808	MOVD    a+0(FP), R3
809	MOVD    b+8(FP), R4
810	MOVD    size+16(FP), R5
811
812	BL	runtime·memeqbody(SB)
813	MOVB    R9, ret+24(FP)
814	RET
815
816// memequal_varlen(a, b unsafe.Pointer) bool
817TEXT runtime·memequal_varlen(SB),NOSPLIT,$40-17
818	MOVD	a+0(FP), R3
819	MOVD	b+8(FP), R4
820	CMP	R3, R4
821	BEQ	eq
822	MOVD	8(R11), R5    // compiler stores size at offset 8 in the closure
823	BL	runtime·memeqbody(SB)
824	MOVB	R9, ret+16(FP)
825	RET
826eq:
827	MOVD	$1, R3
828	MOVB	R3, ret+16(FP)
829	RET
830
831// Do an efficient memcmp for ppc64le
832// R3 = s1 len
833// R4 = s2 len
834// R5 = s1 addr
835// R6 = s2 addr
836// R7 = addr of return value
837TEXT cmpbodyLE<>(SB),NOSPLIT|NOFRAME,$0-0
838	MOVD	R3,R8		// set up length
839	CMP	R3,R4,CR2	// unequal?
840	BC	12,8,setuplen	// BLT CR2
841	MOVD	R4,R8		// use R4 for comparison len
842setuplen:
843	MOVD	R8,CTR		// set up loop counter
844	CMP	R8,$8		// only optimize >=8
845	BLT	simplecheck
846	DCBT	(R5)		// cache hint
847	DCBT	(R6)
848	CMP	R8,$32		// optimize >= 32
849	MOVD	R8,R9
850	BLT	setup8a		// 8 byte moves only
851setup32a:
852	SRADCC	$5,R8,R9	// number of 32 byte chunks
853	MOVD	R9,CTR
854
855        // Special processing for 32 bytes or longer.
856        // Loading this way is faster and correct as long as the
857	// doublewords being compared are equal. Once they
858	// are found unequal, reload them in proper byte order
859	// to determine greater or less than.
860loop32a:
861	MOVD	0(R5),R9	// doublewords to compare
862	MOVD	0(R6),R10	// get 4 doublewords
863	MOVD	8(R5),R14
864	MOVD	8(R6),R15
865	CMPU	R9,R10		// bytes equal?
866	MOVD	$0,R16		// set up for cmpne
867	BNE	cmpne		// further compare for LT or GT
868	MOVD	16(R5),R9	// get next pair of doublewords
869	MOVD	16(R6),R10
870	CMPU	R14,R15		// bytes match?
871	MOVD	$8,R16		// set up for cmpne
872	BNE	cmpne		// further compare for LT or GT
873	MOVD	24(R5),R14	// get next pair of doublewords
874	MOVD    24(R6),R15
875	CMPU	R9,R10		// bytes match?
876	MOVD	$16,R16		// set up for cmpne
877	BNE	cmpne		// further compare for LT or GT
878	MOVD	$-8,R16		// for cmpne, R5,R6 already inc by 32
879	ADD	$32,R5		// bump up to next 32
880	ADD	$32,R6
881	CMPU    R14,R15		// bytes match?
882	BC	8,2,loop32a	// br ctr and cr
883	BNE	cmpne
884	ANDCC	$24,R8,R9	// Any 8 byte chunks?
885	BEQ	leftover	// and result is 0
886setup8a:
887	SRADCC	$3,R9,R9	// get the 8 byte count
888	BEQ	leftover	// shifted value is 0
889	MOVD	R9,CTR		// loop count for doublewords
890loop8:
891	MOVDBR	(R5+R0),R9	// doublewords to compare
892	MOVDBR	(R6+R0),R10	// LE compare order
893	ADD	$8,R5
894	ADD	$8,R6
895	CMPU	R9,R10		// match?
896	BC	8,2,loop8	// bt ctr <> 0 && cr
897	BGT	greater
898	BLT	less
899leftover:
900	ANDCC	$7,R8,R9	// check for leftover bytes
901	MOVD	R9,CTR		// save the ctr
902	BNE	simple		// leftover bytes
903	BC	12,10,equal	// test CR2 for length comparison
904	BC	12,8,less
905	BR	greater
906simplecheck:
907	CMP	R8,$0		// remaining compare length 0
908	BNE	simple		// do simple compare
909	BC	12,10,equal	// test CR2 for length comparison
910	BC	12,8,less	// 1st len < 2nd len, result less
911	BR	greater		// 1st len > 2nd len must be greater
912simple:
913	MOVBZ	0(R5), R9	// get byte from 1st operand
914	ADD	$1,R5
915	MOVBZ	0(R6), R10	// get byte from 2nd operand
916	ADD	$1,R6
917	CMPU	R9, R10
918	BC	8,2,simple	// bc ctr <> 0 && cr
919	BGT	greater		// 1st > 2nd
920	BLT	less		// 1st < 2nd
921	BC	12,10,equal	// test CR2 for length comparison
922	BC	12,9,greater	// 2nd len > 1st len
923	BR	less		// must be less
924cmpne:				// only here is not equal
925	MOVDBR	(R5+R16),R8	// reload in reverse order
926	MOVDBR	(R6+R16),R9
927	CMPU	R8,R9		// compare correct endianness
928	BGT	greater		// here only if NE
929less:
930	MOVD	$-1,R3
931	MOVD	R3,(R7)		// return value if A < B
932	RET
933equal:
934	MOVD	$0,(R7)		// return value if A == B
935	RET
936greater:
937	MOVD	$1,R3
938	MOVD	R3,(R7)		// return value if A > B
939	RET
940
941// Do an efficient memcmp for ppc64 (BE)
942// R3 = s1 len
943// R4 = s2 len
944// R5 = s1 addr
945// R6 = s2 addr
946// R7 = addr of return value
947TEXT cmpbodyBE<>(SB),NOSPLIT|NOFRAME,$0-0
948	MOVD	R3,R8		// set up length
949	CMP	R3,R4,CR2	// unequal?
950	BC	12,8,setuplen	// BLT CR2
951	MOVD	R4,R8		// use R4 for comparison len
952setuplen:
953	MOVD	R8,CTR		// set up loop counter
954	CMP	R8,$8		// only optimize >=8
955	BLT	simplecheck
956	DCBT	(R5)		// cache hint
957	DCBT	(R6)
958	CMP	R8,$32		// optimize >= 32
959	MOVD	R8,R9
960	BLT	setup8a		// 8 byte moves only
961
962setup32a:
963	SRADCC	$5,R8,R9	// number of 32 byte chunks
964	MOVD	R9,CTR
965loop32a:
966	MOVD	0(R5),R9	// doublewords to compare
967	MOVD	0(R6),R10	// get 4 doublewords
968	MOVD	8(R5),R14
969	MOVD	8(R6),R15
970	CMPU	R9,R10		// bytes equal?
971	BLT	less		// found to be less
972	BGT	greater		// found to be greater
973	MOVD	16(R5),R9	// get next pair of doublewords
974	MOVD	16(R6),R10
975	CMPU	R14,R15		// bytes match?
976	BLT	less		// found less
977	BGT	greater		// found greater
978	MOVD	24(R5),R14	// get next pair of doublewords
979	MOVD	24(R6),R15
980	CMPU	R9,R10		// bytes match?
981	BLT	less		// found to be less
982	BGT	greater		// found to be greater
983	ADD	$32,R5		// bump up to next 32
984	ADD	$32,R6
985	CMPU	R14,R15		// bytes match?
986	BC	8,2,loop32a	// br ctr and cr
987	BLT	less		// with BE, byte ordering is
988	BGT	greater		// good for compare
989	ANDCC	$24,R8,R9	// Any 8 byte chunks?
990	BEQ	leftover	// and result is 0
991setup8a:
992	SRADCC	$3,R9,R9	// get the 8 byte count
993	BEQ	leftover	// shifted value is 0
994	MOVD	R9,CTR		// loop count for doublewords
995loop8:
996	MOVD	(R5),R9
997	MOVD	(R6),R10
998	ADD	$8,R5
999	ADD	$8,R6
1000	CMPU	R9,R10		// match?
1001	BC	8,2,loop8	// bt ctr <> 0 && cr
1002	BGT	greater
1003	BLT	less
1004leftover:
1005	ANDCC	$7,R8,R9	// check for leftover bytes
1006	MOVD	R9,CTR		// save the ctr
1007	BNE	simple		// leftover bytes
1008	BC	12,10,equal	// test CR2 for length comparison
1009	BC	12,8,less
1010	BR	greater
1011simplecheck:
1012	CMP	R8,$0		// remaining compare length 0
1013	BNE	simple		// do simple compare
1014	BC	12,10,equal	// test CR2 for length comparison
1015	BC 	12,8,less	// 1st len < 2nd len, result less
1016	BR	greater		// same len, must be equal
1017simple:
1018	MOVBZ	0(R5),R9	// get byte from 1st operand
1019	ADD	$1,R5
1020	MOVBZ	0(R6),R10	// get byte from 2nd operand
1021	ADD	$1,R6
1022	CMPU	R9,R10
1023	BC	8,2,simple	// bc ctr <> 0 && cr
1024	BGT	greater		// 1st > 2nd
1025	BLT	less		// 1st < 2nd
1026	BC	12,10,equal	// test CR2 for length comparison
1027	BC	12,9,greater	// 2nd len > 1st len
1028less:
1029	MOVD	$-1,R3
1030	MOVD    R3,(R7)		// return value if A < B
1031	RET
1032equal:
1033	MOVD    $0,(R7)		// return value if A == B
1034	RET
1035greater:
1036	MOVD	$1,R3
1037	MOVD	R3,(R7)		// return value if A > B
1038	RET
1039
1040// Do an efficient memequal for ppc64
1041// R3 = s1
1042// R4 = s2
1043// R5 = len
1044// R9 = return value
1045TEXT runtime·memeqbody(SB),NOSPLIT|NOFRAME,$0-0
1046	MOVD    R5,CTR
1047	CMP     R5,$8		// only optimize >=8
1048	BLT     simplecheck
1049	DCBT	(R3)		// cache hint
1050	DCBT	(R4)
1051	CMP	R5,$32		// optimize >= 32
1052	MOVD	R5,R6		// needed if setup8a branch
1053	BLT	setup8a		// 8 byte moves only
1054setup32a:                       // 8 byte aligned, >= 32 bytes
1055	SRADCC  $5,R5,R6        // number of 32 byte chunks to compare
1056	MOVD	R6,CTR
1057loop32a:
1058	MOVD    0(R3),R6        // doublewords to compare
1059	MOVD    0(R4),R7
1060	MOVD	8(R3),R8	//
1061	MOVD	8(R4),R9
1062	CMP     R6,R7           // bytes batch?
1063	BNE     noteq
1064	MOVD	16(R3),R6
1065	MOVD	16(R4),R7
1066	CMP     R8,R9		// bytes match?
1067	MOVD	24(R3),R8
1068	MOVD	24(R4),R9
1069	BNE     noteq
1070	CMP     R6,R7           // bytes match?
1071	BNE	noteq
1072	ADD     $32,R3		// bump up to next 32
1073	ADD     $32,R4
1074	CMP     R8,R9           // bytes match?
1075	BC      8,2,loop32a	// br ctr and cr
1076	BNE	noteq
1077	ANDCC	$24,R5,R6       // Any 8 byte chunks?
1078	BEQ	leftover	// and result is 0
1079setup8a:
1080	SRADCC  $3,R6,R6        // get the 8 byte count
1081	BEQ	leftover	// shifted value is 0
1082	MOVD    R6,CTR
1083loop8:
1084	MOVD    0(R3),R6        // doublewords to compare
1085	ADD	$8,R3
1086	MOVD    0(R4),R7
1087	ADD     $8,R4
1088	CMP     R6,R7           // match?
1089	BC	8,2,loop8	// bt ctr <> 0 && cr
1090	BNE     noteq
1091leftover:
1092	ANDCC   $7,R5,R6        // check for leftover bytes
1093	BEQ     equal
1094	MOVD    R6,CTR
1095	BR	simple
1096simplecheck:
1097	CMP	R5,$0
1098	BEQ	equal
1099simple:
1100	MOVBZ   0(R3), R6
1101	ADD	$1,R3
1102	MOVBZ   0(R4), R7
1103	ADD     $1,R4
1104	CMP     R6, R7
1105	BNE     noteq
1106	BC      8,2,simple
1107	BNE	noteq
1108	BR	equal
1109noteq:
1110	MOVD    $0, R9
1111	RET
1112equal:
1113	MOVD    $1, R9
1114	RET
1115
1116// eqstring tests whether two strings are equal.
1117// The compiler guarantees that strings passed
1118// to eqstring have equal length.
1119// See runtime_test.go:eqstring_generic for
1120// equivalent Go code.
1121TEXT runtime·eqstring(SB),NOSPLIT,$0-33
1122	MOVD    s1_base+0(FP), R3
1123	MOVD    s2_base+16(FP), R4
1124	MOVD    $1, R5
1125	MOVB    R5, ret+32(FP)
1126	CMP     R3, R4
1127	BNE     2(PC)
1128	RET
1129	MOVD    s1_len+8(FP), R5
1130	BL      runtime·memeqbody(SB)
1131	MOVB    R9, ret+32(FP)
1132	RET
1133
1134TEXT bytes·Equal(SB),NOSPLIT,$0-49
1135	MOVD	a_len+8(FP), R4
1136	MOVD	b_len+32(FP), R5
1137	CMP	R5, R4		// unequal lengths are not equal
1138	BNE	noteq
1139	MOVD	a+0(FP), R3
1140	MOVD	b+24(FP), R4
1141	BL	runtime·memeqbody(SB)
1142
1143	MOVBZ	R9,ret+48(FP)
1144	RET
1145
1146noteq:
1147	MOVBZ	$0,ret+48(FP)
1148	RET
1149
1150equal:
1151	MOVD	$1,R3
1152	MOVBZ	R3,ret+48(FP)
1153	RET
1154
1155TEXT bytes·IndexByte(SB),NOSPLIT,$0-40
1156	MOVD	s+0(FP), R3
1157	MOVD	s_len+8(FP), R4
1158	MOVBZ	c+24(FP), R5	// byte to find
1159	MOVD	R3, R6		// store base for later
1160	SUB	$1, R3
1161	ADD	R3, R4		// end-1
1162
1163loop:
1164	CMP	R3, R4
1165	BEQ	notfound
1166	MOVBZU	1(R3), R7
1167	CMP	R7, R5
1168	BNE	loop
1169
1170	SUB	R6, R3		// remove base
1171	MOVD	R3, ret+32(FP)
1172	RET
1173
1174notfound:
1175	MOVD	$-1, R3
1176	MOVD	R3, ret+32(FP)
1177	RET
1178
1179TEXT strings·IndexByte(SB),NOSPLIT,$0-32
1180	MOVD	p+0(FP), R3
1181	MOVD	b_len+8(FP), R4
1182	MOVBZ	c+16(FP), R5	// byte to find
1183	MOVD	R3, R6		// store base for later
1184	SUB	$1, R3
1185	ADD	R3, R4		// end-1
1186
1187loop:
1188	CMP	R3, R4
1189	BEQ	notfound
1190	MOVBZU	1(R3), R7
1191	CMP	R7, R5
1192	BNE	loop
1193
1194	SUB	R6, R3		// remove base
1195	MOVD	R3, ret+24(FP)
1196	RET
1197
1198notfound:
1199	MOVD	$-1, R3
1200	MOVD	R3, ret+24(FP)
1201	RET
1202
1203TEXT runtime·cmpstring(SB),NOSPLIT|NOFRAME,$0-40
1204	MOVD	s1_base+0(FP), R5
1205	MOVD	s1_len+8(FP), R3
1206	MOVD	s2_base+16(FP), R6
1207	MOVD	s2_len+24(FP), R4
1208	MOVD	$ret+32(FP), R7
1209#ifdef	GOARCH_ppc64le
1210	BR	cmpbodyLE<>(SB)
1211#else
1212	BR      cmpbodyBE<>(SB)
1213#endif
1214
1215TEXT bytes·Compare(SB),NOSPLIT|NOFRAME,$0-56
1216	MOVD	s1+0(FP), R5
1217	MOVD	s1+8(FP), R3
1218	MOVD	s2+24(FP), R6
1219	MOVD	s2+32(FP), R4
1220	MOVD	$ret+48(FP), R7
1221#ifdef	GOARCH_ppc64le
1222	BR	cmpbodyLE<>(SB)
1223#else
1224	BR      cmpbodyBE<>(SB)
1225#endif
1226
1227TEXT runtime·fastrand(SB), NOSPLIT, $0-4
1228	MOVD	g_m(g), R4
1229	MOVWZ	m_fastrand(R4), R3
1230	ADD	R3, R3
1231	CMPW	R3, $0
1232	BGE	2(PC)
1233	XOR	$0x88888eef, R3
1234	MOVW	R3, m_fastrand(R4)
1235	MOVW	R3, ret+0(FP)
1236	RET
1237
1238TEXT runtime·return0(SB), NOSPLIT, $0
1239	MOVW	$0, R3
1240	RET
1241
1242// Called from cgo wrappers, this function returns g->m->curg.stack.hi.
1243// Must obey the gcc calling convention.
1244TEXT _cgo_topofstack(SB),NOSPLIT|NOFRAME,$0
1245	// g (R30) and R31 are callee-save in the C ABI, so save them
1246	MOVD	g, R4
1247	MOVD	R31, R5
1248	MOVD	LR, R6
1249
1250	BL	runtime·load_g(SB)	// clobbers g (R30), R31
1251	MOVD	g_m(g), R3
1252	MOVD	m_curg(R3), R3
1253	MOVD	(g_stack+stack_hi)(R3), R3
1254
1255	MOVD	R4, g
1256	MOVD	R5, R31
1257	MOVD	R6, LR
1258	RET
1259
1260// The top-most function running on a goroutine
1261// returns to goexit+PCQuantum.
1262//
1263// When dynamically linking Go, it can be returned to from a function
1264// implemented in a different module and so needs to reload the TOC pointer
1265// from the stack (although this function declares that it does not set up x-a
1266// frame, newproc1 does in fact allocate one for goexit and saves the TOC
1267// pointer in the correct place).
1268// goexit+_PCQuantum is halfway through the usual global entry point prologue
1269// that derives r2 from r12 which is a bit silly, but not harmful.
1270TEXT runtime·goexit(SB),NOSPLIT|NOFRAME,$0-0
1271	MOVD	24(R1), R2
1272	BL	runtime·goexit1(SB)	// does not return
1273	// traceback from goexit1 must hit code range of goexit
1274	MOVD	R0, R0	// NOP
1275
1276TEXT runtime·prefetcht0(SB),NOSPLIT,$0-8
1277	RET
1278
1279TEXT runtime·prefetcht1(SB),NOSPLIT,$0-8
1280	RET
1281
1282TEXT runtime·prefetcht2(SB),NOSPLIT,$0-8
1283	RET
1284
1285TEXT runtime·prefetchnta(SB),NOSPLIT,$0-8
1286	RET
1287
1288TEXT runtime·sigreturn(SB),NOSPLIT,$0-0
1289	RET
1290
1291// prepGoExitFrame saves the current TOC pointer (i.e. the TOC pointer for the
1292// module containing runtime) to the frame that goexit will execute in when
1293// the goroutine exits. It's implemented in assembly mainly because that's the
1294// easiest way to get access to R2.
1295TEXT runtime·prepGoExitFrame(SB),NOSPLIT,$0-8
1296      MOVD    sp+0(FP), R3
1297      MOVD    R2, 24(R3)
1298      RET
1299
1300TEXT runtime·addmoduledata(SB),NOSPLIT|NOFRAME,$0-0
1301	ADD	$-8, R1
1302	MOVD	R31, 0(R1)
1303	MOVD	runtime·lastmoduledatap(SB), R4
1304	MOVD	R3, moduledata_next(R4)
1305	MOVD	R3, runtime·lastmoduledatap(SB)
1306	MOVD	0(R1), R31
1307	ADD	$8, R1
1308	RET
1309
1310TEXT ·checkASM(SB),NOSPLIT,$0-1
1311	MOVW	$1, R3
1312	MOVB	R3, ret+0(FP)
1313	RET
1314