xref: /openbmc/linux/arch/ia64/kernel/head.S (revision 9b68f30b)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Here is where the ball gets rolling as far as the kernel is concerned.
4 * When control is transferred to _start, the bootload has already
5 * loaded us to the correct address.  All that's left to do here is
6 * to set up the kernel's global pointer and jump to the kernel
7 * entry point.
8 *
9 * Copyright (C) 1998-2001, 2003, 2005 Hewlett-Packard Co
10 *	David Mosberger-Tang <davidm@hpl.hp.com>
11 *	Stephane Eranian <eranian@hpl.hp.com>
12 * Copyright (C) 1999 VA Linux Systems
13 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
14 * Copyright (C) 1999 Intel Corp.
15 * Copyright (C) 1999 Asit Mallick <Asit.K.Mallick@intel.com>
16 * Copyright (C) 1999 Don Dugger <Don.Dugger@intel.com>
17 * Copyright (C) 2002 Fenghua Yu <fenghua.yu@intel.com>
18 *   -Optimize __ia64_save_fpu() and __ia64_load_fpu() for Itanium 2.
19 * Copyright (C) 2004 Ashok Raj <ashok.raj@intel.com>
20 *   Support for CPU Hotplug
21 */
22
23
24#include <linux/pgtable.h>
25#include <asm/asmmacro.h>
26#include <asm/fpu.h>
27#include <asm/kregs.h>
28#include <asm/mmu_context.h>
29#include <asm/asm-offsets.h>
30#include <asm/pal.h>
31#include <asm/processor.h>
32#include <asm/ptrace.h>
33#include <asm/mca_asm.h>
34#include <linux/init.h>
35#include <linux/linkage.h>
36#include <asm/export.h>
37
38#ifdef CONFIG_HOTPLUG_CPU
39#define SAL_PSR_BITS_TO_SET				\
40	(IA64_PSR_AC | IA64_PSR_BN | IA64_PSR_MFH | IA64_PSR_MFL)
41
42#define SAVE_FROM_REG(src, ptr, dest)	\
43	mov dest=src;;						\
44	st8 [ptr]=dest,0x08
45
46#define RESTORE_REG(reg, ptr, _tmp)		\
47	ld8 _tmp=[ptr],0x08;;				\
48	mov reg=_tmp
49
50#define SAVE_BREAK_REGS(ptr, _idx, _breg, _dest)\
51	mov ar.lc=IA64_NUM_DBG_REGS-1;; 			\
52	mov _idx=0;; 								\
531: 												\
54	SAVE_FROM_REG(_breg[_idx], ptr, _dest);;	\
55	add _idx=1,_idx;;							\
56	br.cloop.sptk.many 1b
57
58#define RESTORE_BREAK_REGS(ptr, _idx, _breg, _tmp, _lbl)\
59	mov ar.lc=IA64_NUM_DBG_REGS-1;;			\
60	mov _idx=0;;							\
61_lbl:  RESTORE_REG(_breg[_idx], ptr, _tmp);;	\
62	add _idx=1, _idx;;						\
63	br.cloop.sptk.many _lbl
64
65#define SAVE_ONE_RR(num, _reg, _tmp) \
66	movl _tmp=(num<<61);;	\
67	mov _reg=rr[_tmp]
68
69#define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
70	SAVE_ONE_RR(0,_r0, _tmp);; \
71	SAVE_ONE_RR(1,_r1, _tmp);; \
72	SAVE_ONE_RR(2,_r2, _tmp);; \
73	SAVE_ONE_RR(3,_r3, _tmp);; \
74	SAVE_ONE_RR(4,_r4, _tmp);; \
75	SAVE_ONE_RR(5,_r5, _tmp);; \
76	SAVE_ONE_RR(6,_r6, _tmp);; \
77	SAVE_ONE_RR(7,_r7, _tmp);;
78
79#define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
80	st8 [ptr]=_r0, 8;; \
81	st8 [ptr]=_r1, 8;; \
82	st8 [ptr]=_r2, 8;; \
83	st8 [ptr]=_r3, 8;; \
84	st8 [ptr]=_r4, 8;; \
85	st8 [ptr]=_r5, 8;; \
86	st8 [ptr]=_r6, 8;; \
87	st8 [ptr]=_r7, 8;;
88
89#define RESTORE_REGION_REGS(ptr, _idx1, _idx2, _tmp) \
90	mov		ar.lc=0x08-1;;						\
91	movl	_idx1=0x00;;						\
92RestRR:											\
93	dep.z	_idx2=_idx1,61,3;;					\
94	ld8		_tmp=[ptr],8;;						\
95	mov		rr[_idx2]=_tmp;;					\
96	srlz.d;;									\
97	add		_idx1=1,_idx1;;						\
98	br.cloop.sptk.few	RestRR
99
100#define SET_AREA_FOR_BOOTING_CPU(reg1, reg2) \
101	movl reg1=sal_state_for_booting_cpu;;	\
102	ld8 reg2=[reg1];;
103
104/*
105 * Adjust region registers saved before starting to save
106 * break regs and rest of the states that need to be preserved.
107 */
108#define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(_reg1,_reg2,_pred)  \
109	SAVE_FROM_REG(b0,_reg1,_reg2);;						\
110	SAVE_FROM_REG(b1,_reg1,_reg2);;						\
111	SAVE_FROM_REG(b2,_reg1,_reg2);;						\
112	SAVE_FROM_REG(b3,_reg1,_reg2);;						\
113	SAVE_FROM_REG(b4,_reg1,_reg2);;						\
114	SAVE_FROM_REG(b5,_reg1,_reg2);;						\
115	st8 [_reg1]=r1,0x08;;								\
116	st8 [_reg1]=r12,0x08;;								\
117	st8 [_reg1]=r13,0x08;;								\
118	SAVE_FROM_REG(ar.fpsr,_reg1,_reg2);;				\
119	SAVE_FROM_REG(ar.pfs,_reg1,_reg2);;					\
120	SAVE_FROM_REG(ar.rnat,_reg1,_reg2);;				\
121	SAVE_FROM_REG(ar.unat,_reg1,_reg2);;				\
122	SAVE_FROM_REG(ar.bspstore,_reg1,_reg2);;			\
123	SAVE_FROM_REG(cr.dcr,_reg1,_reg2);;					\
124	SAVE_FROM_REG(cr.iva,_reg1,_reg2);;					\
125	SAVE_FROM_REG(cr.pta,_reg1,_reg2);;					\
126	SAVE_FROM_REG(cr.itv,_reg1,_reg2);;					\
127	SAVE_FROM_REG(cr.pmv,_reg1,_reg2);;					\
128	SAVE_FROM_REG(cr.cmcv,_reg1,_reg2);;				\
129	SAVE_FROM_REG(cr.lrr0,_reg1,_reg2);;				\
130	SAVE_FROM_REG(cr.lrr1,_reg1,_reg2);;				\
131	st8 [_reg1]=r4,0x08;;								\
132	st8 [_reg1]=r5,0x08;;								\
133	st8 [_reg1]=r6,0x08;;								\
134	st8 [_reg1]=r7,0x08;;								\
135	st8 [_reg1]=_pred,0x08;;							\
136	SAVE_FROM_REG(ar.lc, _reg1, _reg2);;				\
137	stf.spill.nta [_reg1]=f2,16;;						\
138	stf.spill.nta [_reg1]=f3,16;;						\
139	stf.spill.nta [_reg1]=f4,16;;						\
140	stf.spill.nta [_reg1]=f5,16;;						\
141	stf.spill.nta [_reg1]=f16,16;;						\
142	stf.spill.nta [_reg1]=f17,16;;						\
143	stf.spill.nta [_reg1]=f18,16;;						\
144	stf.spill.nta [_reg1]=f19,16;;						\
145	stf.spill.nta [_reg1]=f20,16;;						\
146	stf.spill.nta [_reg1]=f21,16;;						\
147	stf.spill.nta [_reg1]=f22,16;;						\
148	stf.spill.nta [_reg1]=f23,16;;						\
149	stf.spill.nta [_reg1]=f24,16;;						\
150	stf.spill.nta [_reg1]=f25,16;;						\
151	stf.spill.nta [_reg1]=f26,16;;						\
152	stf.spill.nta [_reg1]=f27,16;;						\
153	stf.spill.nta [_reg1]=f28,16;;						\
154	stf.spill.nta [_reg1]=f29,16;;						\
155	stf.spill.nta [_reg1]=f30,16;;						\
156	stf.spill.nta [_reg1]=f31,16;;
157
158#else
159#define SET_AREA_FOR_BOOTING_CPU(a1, a2)
160#define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(a1,a2, a3)
161#define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
162#define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
163#endif
164
165#define SET_ONE_RR(num, pgsize, _tmp1, _tmp2, vhpt) \
166	movl _tmp1=(num << 61);;	\
167	mov _tmp2=((ia64_rid(IA64_REGION_ID_KERNEL, (num<<61)) << 8) | (pgsize << 2) | vhpt);; \
168	mov rr[_tmp1]=_tmp2
169
170	__PAGE_ALIGNED_DATA
171
172	.global empty_zero_page
173EXPORT_DATA_SYMBOL_GPL(empty_zero_page)
174empty_zero_page:
175	.skip PAGE_SIZE
176
177	.global swapper_pg_dir
178swapper_pg_dir:
179	.skip PAGE_SIZE
180
181	.rodata
182halt_msg:
183	stringz "Halting kernel\n"
184
185	__REF
186
187	.global start_ap
188
189	/*
190	 * Start the kernel.  When the bootloader passes control to _start(), r28
191	 * points to the address of the boot parameter area.  Execution reaches
192	 * here in physical mode.
193	 */
194GLOBAL_ENTRY(_start)
195start_ap:
196	.prologue
197	.save rp, r0		// terminate unwind chain with a NULL rp
198	.body
199
200	rsm psr.i | psr.ic
201	;;
202	srlz.i
203	;;
204 {
205	flushrs				// must be first insn in group
206	srlz.i
207 }
208	;;
209	/*
210	 * Save the region registers, predicate before they get clobbered
211	 */
212	SAVE_REGION_REGS(r2, r8,r9,r10,r11,r12,r13,r14,r15);
213	mov r25=pr;;
214
215	/*
216	 * Initialize kernel region registers:
217	 *	rr[0]: VHPT enabled, page size = PAGE_SHIFT
218	 *	rr[1]: VHPT enabled, page size = PAGE_SHIFT
219	 *	rr[2]: VHPT enabled, page size = PAGE_SHIFT
220	 *	rr[3]: VHPT enabled, page size = PAGE_SHIFT
221	 *	rr[4]: VHPT enabled, page size = PAGE_SHIFT
222	 *	rr[5]: VHPT enabled, page size = PAGE_SHIFT
223	 *	rr[6]: VHPT disabled, page size = IA64_GRANULE_SHIFT
224	 *	rr[7]: VHPT disabled, page size = IA64_GRANULE_SHIFT
225	 * We initialize all of them to prevent inadvertently assuming
226	 * something about the state of address translation early in boot.
227	 */
228	SET_ONE_RR(0, PAGE_SHIFT, r2, r16, 1);;
229	SET_ONE_RR(1, PAGE_SHIFT, r2, r16, 1);;
230	SET_ONE_RR(2, PAGE_SHIFT, r2, r16, 1);;
231	SET_ONE_RR(3, PAGE_SHIFT, r2, r16, 1);;
232	SET_ONE_RR(4, PAGE_SHIFT, r2, r16, 1);;
233	SET_ONE_RR(5, PAGE_SHIFT, r2, r16, 1);;
234	SET_ONE_RR(6, IA64_GRANULE_SHIFT, r2, r16, 0);;
235	SET_ONE_RR(7, IA64_GRANULE_SHIFT, r2, r16, 0);;
236	/*
237	 * Now pin mappings into the TLB for kernel text and data
238	 */
239	mov r18=KERNEL_TR_PAGE_SHIFT<<2
240	movl r17=KERNEL_START
241	;;
242	mov cr.itir=r18
243	mov cr.ifa=r17
244	mov r16=IA64_TR_KERNEL
245	mov r3=ip
246	movl r18=PAGE_KERNEL
247	;;
248	dep r2=0,r3,0,KERNEL_TR_PAGE_SHIFT
249	;;
250	or r18=r2,r18
251	;;
252	srlz.i
253	;;
254	itr.i itr[r16]=r18
255	;;
256	itr.d dtr[r16]=r18
257	;;
258	srlz.i
259
260	/*
261	 * Switch into virtual mode:
262	 */
263	movl r16=(IA64_PSR_IT|IA64_PSR_IC|IA64_PSR_DT|IA64_PSR_RT|IA64_PSR_DFH|IA64_PSR_BN \
264		  |IA64_PSR_DI)
265	;;
266	mov cr.ipsr=r16
267	movl r17=1f
268	;;
269	mov cr.iip=r17
270	mov cr.ifs=r0
271	;;
272	rfi
273	;;
2741:	// now we are in virtual mode
275
276	SET_AREA_FOR_BOOTING_CPU(r2, r16);
277
278	STORE_REGION_REGS(r16, r8,r9,r10,r11,r12,r13,r14,r15);
279	SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(r16,r17,r25)
280	;;
281
282	// set IVT entry point---can't access I/O ports without it
283	movl r3=ia64_ivt
284	;;
285	mov cr.iva=r3
286	movl r2=FPSR_DEFAULT
287	;;
288	srlz.i
289	movl gp=__gp
290
291	mov ar.fpsr=r2
292	;;
293
294#define isAP	p2	// are we an Application Processor?
295#define isBP	p3	// are we the Bootstrap Processor?
296
297#ifdef CONFIG_SMP
298	/*
299	 * Find the init_task for the currently booting CPU.  At poweron, and in
300	 * UP mode, task_for_booting_cpu is NULL.
301	 */
302	movl r3=task_for_booting_cpu
303 	;;
304	ld8 r3=[r3]
305	movl r2=init_task
306	;;
307	cmp.eq isBP,isAP=r3,r0
308	;;
309(isAP)	mov r2=r3
310#else
311	movl r2=init_task
312	cmp.eq isBP,isAP=r0,r0
313#endif
314	;;
315	tpa r3=r2		// r3 == phys addr of task struct
316	mov r16=-1
317(isBP)	br.cond.dpnt .load_current // BP stack is on region 5 --- no need to map it
318
319	// load mapping for stack (virtaddr in r2, physaddr in r3)
320	rsm psr.ic
321	movl r17=PAGE_KERNEL
322	;;
323	srlz.d
324	dep r18=0,r3,0,12
325	;;
326	or r18=r17,r18
327	dep r2=-1,r3,61,3	// IMVA of task
328	;;
329	mov r17=rr[r2]
330	shr.u r16=r3,IA64_GRANULE_SHIFT
331	;;
332	dep r17=0,r17,8,24
333	;;
334	mov cr.itir=r17
335	mov cr.ifa=r2
336
337	mov r19=IA64_TR_CURRENT_STACK
338	;;
339	itr.d dtr[r19]=r18
340	;;
341	ssm psr.ic
342	srlz.d
343  	;;
344
345.load_current:
346	// load the "current" pointer (r13) and ar.k6 with the current task
347	mov IA64_KR(CURRENT)=r2		// virtual address
348	mov IA64_KR(CURRENT_STACK)=r16
349	mov r13=r2
350	/*
351	 * Reserve space at the top of the stack for "struct pt_regs".  Kernel
352	 * threads don't store interesting values in that structure, but the space
353	 * still needs to be there because time-critical stuff such as the context
354	 * switching can be implemented more efficiently (for example, __switch_to()
355	 * always sets the psr.dfh bit of the task it is switching to).
356	 */
357
358	addl r12=IA64_STK_OFFSET-IA64_PT_REGS_SIZE-16,r2
359	addl r2=IA64_RBS_OFFSET,r2	// initialize the RSE
360	mov ar.rsc=0		// place RSE in enforced lazy mode
361	;;
362	loadrs			// clear the dirty partition
363	movl r19=__phys_per_cpu_start
364	mov r18=PERCPU_PAGE_SIZE
365	;;
366#ifndef CONFIG_SMP
367	add r19=r19,r18
368	;;
369#else
370(isAP)	br.few 2f
371	movl r20=__cpu0_per_cpu
372	;;
373	shr.u r18=r18,3
3741:
375	ld8 r21=[r19],8;;
376	st8[r20]=r21,8
377	adds r18=-1,r18;;
378	cmp4.lt p7,p6=0,r18
379(p7)	br.cond.dptk.few 1b
380	mov r19=r20
381	;;
3822:
383#endif
384	tpa r19=r19
385	;;
386	.pred.rel.mutex isBP,isAP
387(isBP)	mov IA64_KR(PER_CPU_DATA)=r19	// per-CPU base for cpu0
388(isAP)	mov IA64_KR(PER_CPU_DATA)=r0	// clear physical per-CPU base
389	;;
390	mov ar.bspstore=r2	// establish the new RSE stack
391	;;
392	mov ar.rsc=0x3		// place RSE in eager mode
393
394(isBP)	dep r28=-1,r28,61,3	// make address virtual
395(isBP)	movl r2=ia64_boot_param
396	;;
397(isBP)	st8 [r2]=r28		// save the address of the boot param area passed by the bootloader
398
399#ifdef CONFIG_SMP
400(isAP)	br.call.sptk.many rp=start_secondary
401.ret0:
402(isAP)	br.cond.sptk self
403#endif
404
405	// This is executed by the bootstrap processor (bsp) only:
406
407	br.call.sptk.many rp=start_kernel
408.ret2:	addl r3=@ltoff(halt_msg),gp
409	;;
410	alloc r2=ar.pfs,8,0,2,0
411	;;
412	ld8 out0=[r3]
413	br.call.sptk.many b0=console_print
414
415self:	hint @pause
416	br.sptk.many self		// endless loop
417END(_start)
418
419	.text
420
421GLOBAL_ENTRY(ia64_save_debug_regs)
422	alloc r16=ar.pfs,1,0,0,0
423	mov r20=ar.lc			// preserve ar.lc
424	mov ar.lc=IA64_NUM_DBG_REGS-1
425	mov r18=0
426	add r19=IA64_NUM_DBG_REGS*8,in0
427	;;
4281:	mov r16=dbr[r18]
429#ifdef CONFIG_ITANIUM
430	;;
431	srlz.d
432#endif
433	mov r17=ibr[r18]
434	add r18=1,r18
435	;;
436	st8.nta [in0]=r16,8
437	st8.nta [r19]=r17,8
438	br.cloop.sptk.many 1b
439	;;
440	mov ar.lc=r20			// restore ar.lc
441	br.ret.sptk.many rp
442END(ia64_save_debug_regs)
443
444GLOBAL_ENTRY(ia64_load_debug_regs)
445	alloc r16=ar.pfs,1,0,0,0
446	lfetch.nta [in0]
447	mov r20=ar.lc			// preserve ar.lc
448	add r19=IA64_NUM_DBG_REGS*8,in0
449	mov ar.lc=IA64_NUM_DBG_REGS-1
450	mov r18=-1
451	;;
4521:	ld8.nta r16=[in0],8
453	ld8.nta r17=[r19],8
454	add r18=1,r18
455	;;
456	mov dbr[r18]=r16
457#ifdef CONFIG_ITANIUM
458	;;
459	srlz.d				// Errata 132 (NoFix status)
460#endif
461	mov ibr[r18]=r17
462	br.cloop.sptk.many 1b
463	;;
464	mov ar.lc=r20			// restore ar.lc
465	br.ret.sptk.many rp
466END(ia64_load_debug_regs)
467
468GLOBAL_ENTRY(__ia64_save_fpu)
469	alloc r2=ar.pfs,1,4,0,0
470	adds loc0=96*16-16,in0
471	adds loc1=96*16-16-128,in0
472	;;
473	stf.spill.nta [loc0]=f127,-256
474	stf.spill.nta [loc1]=f119,-256
475	;;
476	stf.spill.nta [loc0]=f111,-256
477	stf.spill.nta [loc1]=f103,-256
478	;;
479	stf.spill.nta [loc0]=f95,-256
480	stf.spill.nta [loc1]=f87,-256
481	;;
482	stf.spill.nta [loc0]=f79,-256
483	stf.spill.nta [loc1]=f71,-256
484	;;
485	stf.spill.nta [loc0]=f63,-256
486	stf.spill.nta [loc1]=f55,-256
487	adds loc2=96*16-32,in0
488	;;
489	stf.spill.nta [loc0]=f47,-256
490	stf.spill.nta [loc1]=f39,-256
491	adds loc3=96*16-32-128,in0
492	;;
493	stf.spill.nta [loc2]=f126,-256
494	stf.spill.nta [loc3]=f118,-256
495	;;
496	stf.spill.nta [loc2]=f110,-256
497	stf.spill.nta [loc3]=f102,-256
498	;;
499	stf.spill.nta [loc2]=f94,-256
500	stf.spill.nta [loc3]=f86,-256
501	;;
502	stf.spill.nta [loc2]=f78,-256
503	stf.spill.nta [loc3]=f70,-256
504	;;
505	stf.spill.nta [loc2]=f62,-256
506	stf.spill.nta [loc3]=f54,-256
507	adds loc0=96*16-48,in0
508	;;
509	stf.spill.nta [loc2]=f46,-256
510	stf.spill.nta [loc3]=f38,-256
511	adds loc1=96*16-48-128,in0
512	;;
513	stf.spill.nta [loc0]=f125,-256
514	stf.spill.nta [loc1]=f117,-256
515	;;
516	stf.spill.nta [loc0]=f109,-256
517	stf.spill.nta [loc1]=f101,-256
518	;;
519	stf.spill.nta [loc0]=f93,-256
520	stf.spill.nta [loc1]=f85,-256
521	;;
522	stf.spill.nta [loc0]=f77,-256
523	stf.spill.nta [loc1]=f69,-256
524	;;
525	stf.spill.nta [loc0]=f61,-256
526	stf.spill.nta [loc1]=f53,-256
527	adds loc2=96*16-64,in0
528	;;
529	stf.spill.nta [loc0]=f45,-256
530	stf.spill.nta [loc1]=f37,-256
531	adds loc3=96*16-64-128,in0
532	;;
533	stf.spill.nta [loc2]=f124,-256
534	stf.spill.nta [loc3]=f116,-256
535	;;
536	stf.spill.nta [loc2]=f108,-256
537	stf.spill.nta [loc3]=f100,-256
538	;;
539	stf.spill.nta [loc2]=f92,-256
540	stf.spill.nta [loc3]=f84,-256
541	;;
542	stf.spill.nta [loc2]=f76,-256
543	stf.spill.nta [loc3]=f68,-256
544	;;
545	stf.spill.nta [loc2]=f60,-256
546	stf.spill.nta [loc3]=f52,-256
547	adds loc0=96*16-80,in0
548	;;
549	stf.spill.nta [loc2]=f44,-256
550	stf.spill.nta [loc3]=f36,-256
551	adds loc1=96*16-80-128,in0
552	;;
553	stf.spill.nta [loc0]=f123,-256
554	stf.spill.nta [loc1]=f115,-256
555	;;
556	stf.spill.nta [loc0]=f107,-256
557	stf.spill.nta [loc1]=f99,-256
558	;;
559	stf.spill.nta [loc0]=f91,-256
560	stf.spill.nta [loc1]=f83,-256
561	;;
562	stf.spill.nta [loc0]=f75,-256
563	stf.spill.nta [loc1]=f67,-256
564	;;
565	stf.spill.nta [loc0]=f59,-256
566	stf.spill.nta [loc1]=f51,-256
567	adds loc2=96*16-96,in0
568	;;
569	stf.spill.nta [loc0]=f43,-256
570	stf.spill.nta [loc1]=f35,-256
571	adds loc3=96*16-96-128,in0
572	;;
573	stf.spill.nta [loc2]=f122,-256
574	stf.spill.nta [loc3]=f114,-256
575	;;
576	stf.spill.nta [loc2]=f106,-256
577	stf.spill.nta [loc3]=f98,-256
578	;;
579	stf.spill.nta [loc2]=f90,-256
580	stf.spill.nta [loc3]=f82,-256
581	;;
582	stf.spill.nta [loc2]=f74,-256
583	stf.spill.nta [loc3]=f66,-256
584	;;
585	stf.spill.nta [loc2]=f58,-256
586	stf.spill.nta [loc3]=f50,-256
587	adds loc0=96*16-112,in0
588	;;
589	stf.spill.nta [loc2]=f42,-256
590	stf.spill.nta [loc3]=f34,-256
591	adds loc1=96*16-112-128,in0
592	;;
593	stf.spill.nta [loc0]=f121,-256
594	stf.spill.nta [loc1]=f113,-256
595	;;
596	stf.spill.nta [loc0]=f105,-256
597	stf.spill.nta [loc1]=f97,-256
598	;;
599	stf.spill.nta [loc0]=f89,-256
600	stf.spill.nta [loc1]=f81,-256
601	;;
602	stf.spill.nta [loc0]=f73,-256
603	stf.spill.nta [loc1]=f65,-256
604	;;
605	stf.spill.nta [loc0]=f57,-256
606	stf.spill.nta [loc1]=f49,-256
607	adds loc2=96*16-128,in0
608	;;
609	stf.spill.nta [loc0]=f41,-256
610	stf.spill.nta [loc1]=f33,-256
611	adds loc3=96*16-128-128,in0
612	;;
613	stf.spill.nta [loc2]=f120,-256
614	stf.spill.nta [loc3]=f112,-256
615	;;
616	stf.spill.nta [loc2]=f104,-256
617	stf.spill.nta [loc3]=f96,-256
618	;;
619	stf.spill.nta [loc2]=f88,-256
620	stf.spill.nta [loc3]=f80,-256
621	;;
622	stf.spill.nta [loc2]=f72,-256
623	stf.spill.nta [loc3]=f64,-256
624	;;
625	stf.spill.nta [loc2]=f56,-256
626	stf.spill.nta [loc3]=f48,-256
627	;;
628	stf.spill.nta [loc2]=f40
629	stf.spill.nta [loc3]=f32
630	br.ret.sptk.many rp
631END(__ia64_save_fpu)
632
633GLOBAL_ENTRY(__ia64_load_fpu)
634	alloc r2=ar.pfs,1,2,0,0
635	adds r3=128,in0
636	adds r14=256,in0
637	adds r15=384,in0
638	mov loc0=512
639	mov loc1=-1024+16
640	;;
641	ldf.fill.nta f32=[in0],loc0
642	ldf.fill.nta f40=[ r3],loc0
643	ldf.fill.nta f48=[r14],loc0
644	ldf.fill.nta f56=[r15],loc0
645	;;
646	ldf.fill.nta f64=[in0],loc0
647	ldf.fill.nta f72=[ r3],loc0
648	ldf.fill.nta f80=[r14],loc0
649	ldf.fill.nta f88=[r15],loc0
650	;;
651	ldf.fill.nta f96=[in0],loc1
652	ldf.fill.nta f104=[ r3],loc1
653	ldf.fill.nta f112=[r14],loc1
654	ldf.fill.nta f120=[r15],loc1
655	;;
656	ldf.fill.nta f33=[in0],loc0
657	ldf.fill.nta f41=[ r3],loc0
658	ldf.fill.nta f49=[r14],loc0
659	ldf.fill.nta f57=[r15],loc0
660	;;
661	ldf.fill.nta f65=[in0],loc0
662	ldf.fill.nta f73=[ r3],loc0
663	ldf.fill.nta f81=[r14],loc0
664	ldf.fill.nta f89=[r15],loc0
665	;;
666	ldf.fill.nta f97=[in0],loc1
667	ldf.fill.nta f105=[ r3],loc1
668	ldf.fill.nta f113=[r14],loc1
669	ldf.fill.nta f121=[r15],loc1
670	;;
671	ldf.fill.nta f34=[in0],loc0
672	ldf.fill.nta f42=[ r3],loc0
673	ldf.fill.nta f50=[r14],loc0
674	ldf.fill.nta f58=[r15],loc0
675	;;
676	ldf.fill.nta f66=[in0],loc0
677	ldf.fill.nta f74=[ r3],loc0
678	ldf.fill.nta f82=[r14],loc0
679	ldf.fill.nta f90=[r15],loc0
680	;;
681	ldf.fill.nta f98=[in0],loc1
682	ldf.fill.nta f106=[ r3],loc1
683	ldf.fill.nta f114=[r14],loc1
684	ldf.fill.nta f122=[r15],loc1
685	;;
686	ldf.fill.nta f35=[in0],loc0
687	ldf.fill.nta f43=[ r3],loc0
688	ldf.fill.nta f51=[r14],loc0
689	ldf.fill.nta f59=[r15],loc0
690	;;
691	ldf.fill.nta f67=[in0],loc0
692	ldf.fill.nta f75=[ r3],loc0
693	ldf.fill.nta f83=[r14],loc0
694	ldf.fill.nta f91=[r15],loc0
695	;;
696	ldf.fill.nta f99=[in0],loc1
697	ldf.fill.nta f107=[ r3],loc1
698	ldf.fill.nta f115=[r14],loc1
699	ldf.fill.nta f123=[r15],loc1
700	;;
701	ldf.fill.nta f36=[in0],loc0
702	ldf.fill.nta f44=[ r3],loc0
703	ldf.fill.nta f52=[r14],loc0
704	ldf.fill.nta f60=[r15],loc0
705	;;
706	ldf.fill.nta f68=[in0],loc0
707	ldf.fill.nta f76=[ r3],loc0
708	ldf.fill.nta f84=[r14],loc0
709	ldf.fill.nta f92=[r15],loc0
710	;;
711	ldf.fill.nta f100=[in0],loc1
712	ldf.fill.nta f108=[ r3],loc1
713	ldf.fill.nta f116=[r14],loc1
714	ldf.fill.nta f124=[r15],loc1
715	;;
716	ldf.fill.nta f37=[in0],loc0
717	ldf.fill.nta f45=[ r3],loc0
718	ldf.fill.nta f53=[r14],loc0
719	ldf.fill.nta f61=[r15],loc0
720	;;
721	ldf.fill.nta f69=[in0],loc0
722	ldf.fill.nta f77=[ r3],loc0
723	ldf.fill.nta f85=[r14],loc0
724	ldf.fill.nta f93=[r15],loc0
725	;;
726	ldf.fill.nta f101=[in0],loc1
727	ldf.fill.nta f109=[ r3],loc1
728	ldf.fill.nta f117=[r14],loc1
729	ldf.fill.nta f125=[r15],loc1
730	;;
731	ldf.fill.nta f38 =[in0],loc0
732	ldf.fill.nta f46 =[ r3],loc0
733	ldf.fill.nta f54 =[r14],loc0
734	ldf.fill.nta f62 =[r15],loc0
735	;;
736	ldf.fill.nta f70 =[in0],loc0
737	ldf.fill.nta f78 =[ r3],loc0
738	ldf.fill.nta f86 =[r14],loc0
739	ldf.fill.nta f94 =[r15],loc0
740	;;
741	ldf.fill.nta f102=[in0],loc1
742	ldf.fill.nta f110=[ r3],loc1
743	ldf.fill.nta f118=[r14],loc1
744	ldf.fill.nta f126=[r15],loc1
745	;;
746	ldf.fill.nta f39 =[in0],loc0
747	ldf.fill.nta f47 =[ r3],loc0
748	ldf.fill.nta f55 =[r14],loc0
749	ldf.fill.nta f63 =[r15],loc0
750	;;
751	ldf.fill.nta f71 =[in0],loc0
752	ldf.fill.nta f79 =[ r3],loc0
753	ldf.fill.nta f87 =[r14],loc0
754	ldf.fill.nta f95 =[r15],loc0
755	;;
756	ldf.fill.nta f103=[in0]
757	ldf.fill.nta f111=[ r3]
758	ldf.fill.nta f119=[r14]
759	ldf.fill.nta f127=[r15]
760	br.ret.sptk.many rp
761END(__ia64_load_fpu)
762
763GLOBAL_ENTRY(__ia64_init_fpu)
764	stf.spill [sp]=f0		// M3
765	mov	 f32=f0			// F
766	nop.b	 0
767
768	ldfps	 f33,f34=[sp]		// M0
769	ldfps	 f35,f36=[sp]		// M1
770	mov      f37=f0			// F
771	;;
772
773	setf.s	 f38=r0			// M2
774	setf.s	 f39=r0			// M3
775	mov      f40=f0			// F
776
777	ldfps	 f41,f42=[sp]		// M0
778	ldfps	 f43,f44=[sp]		// M1
779	mov      f45=f0			// F
780
781	setf.s	 f46=r0			// M2
782	setf.s	 f47=r0			// M3
783	mov      f48=f0			// F
784
785	ldfps	 f49,f50=[sp]		// M0
786	ldfps	 f51,f52=[sp]		// M1
787	mov      f53=f0			// F
788
789	setf.s	 f54=r0			// M2
790	setf.s	 f55=r0			// M3
791	mov      f56=f0			// F
792
793	ldfps	 f57,f58=[sp]		// M0
794	ldfps	 f59,f60=[sp]		// M1
795	mov      f61=f0			// F
796
797	setf.s	 f62=r0			// M2
798	setf.s	 f63=r0			// M3
799	mov      f64=f0			// F
800
801	ldfps	 f65,f66=[sp]		// M0
802	ldfps	 f67,f68=[sp]		// M1
803	mov      f69=f0			// F
804
805	setf.s	 f70=r0			// M2
806	setf.s	 f71=r0			// M3
807	mov      f72=f0			// F
808
809	ldfps	 f73,f74=[sp]		// M0
810	ldfps	 f75,f76=[sp]		// M1
811	mov      f77=f0			// F
812
813	setf.s	 f78=r0			// M2
814	setf.s	 f79=r0			// M3
815	mov      f80=f0			// F
816
817	ldfps	 f81,f82=[sp]		// M0
818	ldfps	 f83,f84=[sp]		// M1
819	mov      f85=f0			// F
820
821	setf.s	 f86=r0			// M2
822	setf.s	 f87=r0			// M3
823	mov      f88=f0			// F
824
825	/*
826	 * When the instructions are cached, it would be faster to initialize
827	 * the remaining registers with simply mov instructions (F-unit).
828	 * This gets the time down to ~29 cycles.  However, this would use up
829	 * 33 bundles, whereas continuing with the above pattern yields
830	 * 10 bundles and ~30 cycles.
831	 */
832
833	ldfps	 f89,f90=[sp]		// M0
834	ldfps	 f91,f92=[sp]		// M1
835	mov      f93=f0			// F
836
837	setf.s	 f94=r0			// M2
838	setf.s	 f95=r0			// M3
839	mov      f96=f0			// F
840
841	ldfps	 f97,f98=[sp]		// M0
842	ldfps	 f99,f100=[sp]		// M1
843	mov      f101=f0		// F
844
845	setf.s	 f102=r0		// M2
846	setf.s	 f103=r0		// M3
847	mov      f104=f0		// F
848
849	ldfps	 f105,f106=[sp]		// M0
850	ldfps	 f107,f108=[sp]		// M1
851	mov      f109=f0		// F
852
853	setf.s	 f110=r0		// M2
854	setf.s	 f111=r0		// M3
855	mov      f112=f0		// F
856
857	ldfps	 f113,f114=[sp]		// M0
858	ldfps	 f115,f116=[sp]		// M1
859	mov      f117=f0		// F
860
861	setf.s	 f118=r0		// M2
862	setf.s	 f119=r0		// M3
863	mov      f120=f0		// F
864
865	ldfps	 f121,f122=[sp]		// M0
866	ldfps	 f123,f124=[sp]		// M1
867	mov      f125=f0		// F
868
869	setf.s	 f126=r0		// M2
870	setf.s	 f127=r0		// M3
871	br.ret.sptk.many rp		// F
872END(__ia64_init_fpu)
873
874/*
875 * Switch execution mode from virtual to physical
876 *
877 * Inputs:
878 *	r16 = new psr to establish
879 * Output:
880 *	r19 = old virtual address of ar.bsp
881 *	r20 = old virtual address of sp
882 *
883 * Note: RSE must already be in enforced lazy mode
884 */
885GLOBAL_ENTRY(ia64_switch_mode_phys)
886 {
887	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
888	mov r15=ip
889 }
890	;;
891 {
892	flushrs				// must be first insn in group
893	srlz.i
894 }
895	;;
896	mov cr.ipsr=r16			// set new PSR
897	add r3=1f-ia64_switch_mode_phys,r15
898
899	mov r19=ar.bsp
900	mov r20=sp
901	mov r14=rp			// get return address into a general register
902	;;
903
904	// going to physical mode, use tpa to translate virt->phys
905	tpa r17=r19
906	tpa r3=r3
907	tpa sp=sp
908	tpa r14=r14
909	;;
910
911	mov r18=ar.rnat			// save ar.rnat
912	mov ar.bspstore=r17		// this steps on ar.rnat
913	mov cr.iip=r3
914	mov cr.ifs=r0
915	;;
916	mov ar.rnat=r18			// restore ar.rnat
917	rfi				// must be last insn in group
918	;;
9191:	mov rp=r14
920	br.ret.sptk.many rp
921END(ia64_switch_mode_phys)
922
923/*
924 * Switch execution mode from physical to virtual
925 *
926 * Inputs:
927 *	r16 = new psr to establish
928 *	r19 = new bspstore to establish
929 *	r20 = new sp to establish
930 *
931 * Note: RSE must already be in enforced lazy mode
932 */
933GLOBAL_ENTRY(ia64_switch_mode_virt)
934 {
935	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
936	mov r15=ip
937 }
938	;;
939 {
940	flushrs				// must be first insn in group
941	srlz.i
942 }
943	;;
944	mov cr.ipsr=r16			// set new PSR
945	add r3=1f-ia64_switch_mode_virt,r15
946
947	mov r14=rp			// get return address into a general register
948	;;
949
950	// going to virtual
951	//   - for code addresses, set upper bits of addr to KERNEL_START
952	//   - for stack addresses, copy from input argument
953	movl r18=KERNEL_START
954	dep r3=0,r3,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
955	dep r14=0,r14,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
956	mov sp=r20
957	;;
958	or r3=r3,r18
959	or r14=r14,r18
960	;;
961
962	mov r18=ar.rnat			// save ar.rnat
963	mov ar.bspstore=r19		// this steps on ar.rnat
964	mov cr.iip=r3
965	mov cr.ifs=r0
966	;;
967	mov ar.rnat=r18			// restore ar.rnat
968	rfi				// must be last insn in group
969	;;
9701:	mov rp=r14
971	br.ret.sptk.many rp
972END(ia64_switch_mode_virt)
973
974GLOBAL_ENTRY(ia64_delay_loop)
975	.prologue
976{	nop 0			// work around GAS unwind info generation bug...
977	.save ar.lc,r2
978	mov r2=ar.lc
979	.body
980	;;
981	mov ar.lc=r32
982}
983	;;
984	// force loop to be 32-byte aligned (GAS bug means we cannot use .align
985	// inside function body without corrupting unwind info).
986{	nop 0 }
9871:	br.cloop.sptk.few 1b
988	;;
989	mov ar.lc=r2
990	br.ret.sptk.many rp
991END(ia64_delay_loop)
992
993/*
994 * Return a CPU-local timestamp in nano-seconds.  This timestamp is
995 * NOT synchronized across CPUs its return value must never be
996 * compared against the values returned on another CPU.  The usage in
997 * kernel/sched/core.c ensures that.
998 *
999 * The return-value of sched_clock() is NOT supposed to wrap-around.
1000 * If it did, it would cause some scheduling hiccups (at the worst).
1001 * Fortunately, with a 64-bit cycle-counter ticking at 100GHz, even
1002 * that would happen only once every 5+ years.
1003 *
1004 * The code below basically calculates:
1005 *
1006 *   (ia64_get_itc() * local_cpu_data->nsec_per_cyc) >> IA64_NSEC_PER_CYC_SHIFT
1007 *
1008 * except that the multiplication and the shift are done with 128-bit
1009 * intermediate precision so that we can produce a full 64-bit result.
1010 */
1011GLOBAL_ENTRY(ia64_native_sched_clock)
1012	addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1013	mov.m r9=ar.itc		// fetch cycle-counter				(35 cyc)
1014	;;
1015	ldf8 f8=[r8]
1016	;;
1017	setf.sig f9=r9		// certain to stall, so issue it _after_ ldf8...
1018	;;
1019	xmpy.lu f10=f9,f8	// calculate low 64 bits of 128-bit product	(4 cyc)
1020	xmpy.hu f11=f9,f8	// calculate high 64 bits of 128-bit product
1021	;;
1022	getf.sig r8=f10		//						(5 cyc)
1023	getf.sig r9=f11
1024	;;
1025	shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1026	br.ret.sptk.many rp
1027END(ia64_native_sched_clock)
1028
1029#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1030GLOBAL_ENTRY(cycle_to_nsec)
1031	alloc r16=ar.pfs,1,0,0,0
1032	addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1033	;;
1034	ldf8 f8=[r8]
1035	;;
1036	setf.sig f9=r32
1037	;;
1038	xmpy.lu f10=f9,f8	// calculate low 64 bits of 128-bit product	(4 cyc)
1039	xmpy.hu f11=f9,f8	// calculate high 64 bits of 128-bit product
1040	;;
1041	getf.sig r8=f10		//						(5 cyc)
1042	getf.sig r9=f11
1043	;;
1044	shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1045	br.ret.sptk.many rp
1046END(cycle_to_nsec)
1047#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
1048
1049#ifdef CONFIG_IA64_BRL_EMU
1050
1051/*
1052 *  Assembly routines used by brl_emu.c to set preserved register state.
1053 */
1054
1055#define SET_REG(reg)				\
1056 GLOBAL_ENTRY(ia64_set_##reg);			\
1057	alloc r16=ar.pfs,1,0,0,0;		\
1058	mov reg=r32;				\
1059	;;					\
1060	br.ret.sptk.many rp;			\
1061 END(ia64_set_##reg)
1062
1063SET_REG(b1);
1064SET_REG(b2);
1065SET_REG(b3);
1066SET_REG(b4);
1067SET_REG(b5);
1068
1069#endif /* CONFIG_IA64_BRL_EMU */
1070
1071#ifdef CONFIG_SMP
1072
1073#ifdef CONFIG_HOTPLUG_CPU
1074GLOBAL_ENTRY(ia64_jump_to_sal)
1075	alloc r16=ar.pfs,1,0,0,0;;
1076	rsm psr.i  | psr.ic
1077{
1078	flushrs
1079	srlz.i
1080}
1081	tpa r25=in0
1082	movl r18=tlb_purge_done;;
1083	DATA_VA_TO_PA(r18);;
1084	mov b1=r18 	// Return location
1085	movl r18=ia64_do_tlb_purge;;
1086	DATA_VA_TO_PA(r18);;
1087	mov b2=r18 	// doing tlb_flush work
1088	mov ar.rsc=0  // Put RSE  in enforced lazy, LE mode
1089	movl r17=1f;;
1090	DATA_VA_TO_PA(r17);;
1091	mov cr.iip=r17
1092	movl r16=SAL_PSR_BITS_TO_SET;;
1093	mov cr.ipsr=r16
1094	mov cr.ifs=r0;;
1095	rfi;;			// note: this unmask MCA/INIT (psr.mc)
10961:
1097	/*
1098	 * Invalidate all TLB data/inst
1099	 */
1100	br.sptk.many b2;; // jump to tlb purge code
1101
1102tlb_purge_done:
1103	RESTORE_REGION_REGS(r25, r17,r18,r19);;
1104	RESTORE_REG(b0, r25, r17);;
1105	RESTORE_REG(b1, r25, r17);;
1106	RESTORE_REG(b2, r25, r17);;
1107	RESTORE_REG(b3, r25, r17);;
1108	RESTORE_REG(b4, r25, r17);;
1109	RESTORE_REG(b5, r25, r17);;
1110	ld8 r1=[r25],0x08;;
1111	ld8 r12=[r25],0x08;;
1112	ld8 r13=[r25],0x08;;
1113	RESTORE_REG(ar.fpsr, r25, r17);;
1114	RESTORE_REG(ar.pfs, r25, r17);;
1115	RESTORE_REG(ar.rnat, r25, r17);;
1116	RESTORE_REG(ar.unat, r25, r17);;
1117	RESTORE_REG(ar.bspstore, r25, r17);;
1118	RESTORE_REG(cr.dcr, r25, r17);;
1119	RESTORE_REG(cr.iva, r25, r17);;
1120	RESTORE_REG(cr.pta, r25, r17);;
1121	srlz.d;;	// required not to violate RAW dependency
1122	RESTORE_REG(cr.itv, r25, r17);;
1123	RESTORE_REG(cr.pmv, r25, r17);;
1124	RESTORE_REG(cr.cmcv, r25, r17);;
1125	RESTORE_REG(cr.lrr0, r25, r17);;
1126	RESTORE_REG(cr.lrr1, r25, r17);;
1127	ld8 r4=[r25],0x08;;
1128	ld8 r5=[r25],0x08;;
1129	ld8 r6=[r25],0x08;;
1130	ld8 r7=[r25],0x08;;
1131	ld8 r17=[r25],0x08;;
1132	mov pr=r17,-1;;
1133	RESTORE_REG(ar.lc, r25, r17);;
1134	/*
1135	 * Now Restore floating point regs
1136	 */
1137	ldf.fill.nta f2=[r25],16;;
1138	ldf.fill.nta f3=[r25],16;;
1139	ldf.fill.nta f4=[r25],16;;
1140	ldf.fill.nta f5=[r25],16;;
1141	ldf.fill.nta f16=[r25],16;;
1142	ldf.fill.nta f17=[r25],16;;
1143	ldf.fill.nta f18=[r25],16;;
1144	ldf.fill.nta f19=[r25],16;;
1145	ldf.fill.nta f20=[r25],16;;
1146	ldf.fill.nta f21=[r25],16;;
1147	ldf.fill.nta f22=[r25],16;;
1148	ldf.fill.nta f23=[r25],16;;
1149	ldf.fill.nta f24=[r25],16;;
1150	ldf.fill.nta f25=[r25],16;;
1151	ldf.fill.nta f26=[r25],16;;
1152	ldf.fill.nta f27=[r25],16;;
1153	ldf.fill.nta f28=[r25],16;;
1154	ldf.fill.nta f29=[r25],16;;
1155	ldf.fill.nta f30=[r25],16;;
1156	ldf.fill.nta f31=[r25],16;;
1157
1158	/*
1159	 * Now that we have done all the register restores
1160	 * we are now ready for the big DIVE to SAL Land
1161	 */
1162	ssm psr.ic;;
1163	srlz.d;;
1164	br.ret.sptk.many b0;;
1165END(ia64_jump_to_sal)
1166#endif /* CONFIG_HOTPLUG_CPU */
1167
1168#endif /* CONFIG_SMP */
1169