xref: /openbmc/linux/arch/ia64/kernel/head.S (revision 9b93eb47)
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 <asm/asmmacro.h>
25#include <asm/fpu.h>
26#include <asm/kregs.h>
27#include <asm/mmu_context.h>
28#include <asm/asm-offsets.h>
29#include <asm/pal.h>
30#include <asm/pgtable.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#ifdef CONFIG_IA64_FW_EMU
408	// initialize PAL & SAL emulator:
409	br.call.sptk.many rp=sys_fw_init
410.ret1:
411#endif
412	br.call.sptk.many rp=start_kernel
413.ret2:	addl r3=@ltoff(halt_msg),gp
414	;;
415	alloc r2=ar.pfs,8,0,2,0
416	;;
417	ld8 out0=[r3]
418	br.call.sptk.many b0=console_print
419
420self:	hint @pause
421	br.sptk.many self		// endless loop
422END(_start)
423
424	.text
425
426GLOBAL_ENTRY(ia64_save_debug_regs)
427	alloc r16=ar.pfs,1,0,0,0
428	mov r20=ar.lc			// preserve ar.lc
429	mov ar.lc=IA64_NUM_DBG_REGS-1
430	mov r18=0
431	add r19=IA64_NUM_DBG_REGS*8,in0
432	;;
4331:	mov r16=dbr[r18]
434#ifdef CONFIG_ITANIUM
435	;;
436	srlz.d
437#endif
438	mov r17=ibr[r18]
439	add r18=1,r18
440	;;
441	st8.nta [in0]=r16,8
442	st8.nta [r19]=r17,8
443	br.cloop.sptk.many 1b
444	;;
445	mov ar.lc=r20			// restore ar.lc
446	br.ret.sptk.many rp
447END(ia64_save_debug_regs)
448
449GLOBAL_ENTRY(ia64_load_debug_regs)
450	alloc r16=ar.pfs,1,0,0,0
451	lfetch.nta [in0]
452	mov r20=ar.lc			// preserve ar.lc
453	add r19=IA64_NUM_DBG_REGS*8,in0
454	mov ar.lc=IA64_NUM_DBG_REGS-1
455	mov r18=-1
456	;;
4571:	ld8.nta r16=[in0],8
458	ld8.nta r17=[r19],8
459	add r18=1,r18
460	;;
461	mov dbr[r18]=r16
462#ifdef CONFIG_ITANIUM
463	;;
464	srlz.d				// Errata 132 (NoFix status)
465#endif
466	mov ibr[r18]=r17
467	br.cloop.sptk.many 1b
468	;;
469	mov ar.lc=r20			// restore ar.lc
470	br.ret.sptk.many rp
471END(ia64_load_debug_regs)
472
473GLOBAL_ENTRY(__ia64_save_fpu)
474	alloc r2=ar.pfs,1,4,0,0
475	adds loc0=96*16-16,in0
476	adds loc1=96*16-16-128,in0
477	;;
478	stf.spill.nta [loc0]=f127,-256
479	stf.spill.nta [loc1]=f119,-256
480	;;
481	stf.spill.nta [loc0]=f111,-256
482	stf.spill.nta [loc1]=f103,-256
483	;;
484	stf.spill.nta [loc0]=f95,-256
485	stf.spill.nta [loc1]=f87,-256
486	;;
487	stf.spill.nta [loc0]=f79,-256
488	stf.spill.nta [loc1]=f71,-256
489	;;
490	stf.spill.nta [loc0]=f63,-256
491	stf.spill.nta [loc1]=f55,-256
492	adds loc2=96*16-32,in0
493	;;
494	stf.spill.nta [loc0]=f47,-256
495	stf.spill.nta [loc1]=f39,-256
496	adds loc3=96*16-32-128,in0
497	;;
498	stf.spill.nta [loc2]=f126,-256
499	stf.spill.nta [loc3]=f118,-256
500	;;
501	stf.spill.nta [loc2]=f110,-256
502	stf.spill.nta [loc3]=f102,-256
503	;;
504	stf.spill.nta [loc2]=f94,-256
505	stf.spill.nta [loc3]=f86,-256
506	;;
507	stf.spill.nta [loc2]=f78,-256
508	stf.spill.nta [loc3]=f70,-256
509	;;
510	stf.spill.nta [loc2]=f62,-256
511	stf.spill.nta [loc3]=f54,-256
512	adds loc0=96*16-48,in0
513	;;
514	stf.spill.nta [loc2]=f46,-256
515	stf.spill.nta [loc3]=f38,-256
516	adds loc1=96*16-48-128,in0
517	;;
518	stf.spill.nta [loc0]=f125,-256
519	stf.spill.nta [loc1]=f117,-256
520	;;
521	stf.spill.nta [loc0]=f109,-256
522	stf.spill.nta [loc1]=f101,-256
523	;;
524	stf.spill.nta [loc0]=f93,-256
525	stf.spill.nta [loc1]=f85,-256
526	;;
527	stf.spill.nta [loc0]=f77,-256
528	stf.spill.nta [loc1]=f69,-256
529	;;
530	stf.spill.nta [loc0]=f61,-256
531	stf.spill.nta [loc1]=f53,-256
532	adds loc2=96*16-64,in0
533	;;
534	stf.spill.nta [loc0]=f45,-256
535	stf.spill.nta [loc1]=f37,-256
536	adds loc3=96*16-64-128,in0
537	;;
538	stf.spill.nta [loc2]=f124,-256
539	stf.spill.nta [loc3]=f116,-256
540	;;
541	stf.spill.nta [loc2]=f108,-256
542	stf.spill.nta [loc3]=f100,-256
543	;;
544	stf.spill.nta [loc2]=f92,-256
545	stf.spill.nta [loc3]=f84,-256
546	;;
547	stf.spill.nta [loc2]=f76,-256
548	stf.spill.nta [loc3]=f68,-256
549	;;
550	stf.spill.nta [loc2]=f60,-256
551	stf.spill.nta [loc3]=f52,-256
552	adds loc0=96*16-80,in0
553	;;
554	stf.spill.nta [loc2]=f44,-256
555	stf.spill.nta [loc3]=f36,-256
556	adds loc1=96*16-80-128,in0
557	;;
558	stf.spill.nta [loc0]=f123,-256
559	stf.spill.nta [loc1]=f115,-256
560	;;
561	stf.spill.nta [loc0]=f107,-256
562	stf.spill.nta [loc1]=f99,-256
563	;;
564	stf.spill.nta [loc0]=f91,-256
565	stf.spill.nta [loc1]=f83,-256
566	;;
567	stf.spill.nta [loc0]=f75,-256
568	stf.spill.nta [loc1]=f67,-256
569	;;
570	stf.spill.nta [loc0]=f59,-256
571	stf.spill.nta [loc1]=f51,-256
572	adds loc2=96*16-96,in0
573	;;
574	stf.spill.nta [loc0]=f43,-256
575	stf.spill.nta [loc1]=f35,-256
576	adds loc3=96*16-96-128,in0
577	;;
578	stf.spill.nta [loc2]=f122,-256
579	stf.spill.nta [loc3]=f114,-256
580	;;
581	stf.spill.nta [loc2]=f106,-256
582	stf.spill.nta [loc3]=f98,-256
583	;;
584	stf.spill.nta [loc2]=f90,-256
585	stf.spill.nta [loc3]=f82,-256
586	;;
587	stf.spill.nta [loc2]=f74,-256
588	stf.spill.nta [loc3]=f66,-256
589	;;
590	stf.spill.nta [loc2]=f58,-256
591	stf.spill.nta [loc3]=f50,-256
592	adds loc0=96*16-112,in0
593	;;
594	stf.spill.nta [loc2]=f42,-256
595	stf.spill.nta [loc3]=f34,-256
596	adds loc1=96*16-112-128,in0
597	;;
598	stf.spill.nta [loc0]=f121,-256
599	stf.spill.nta [loc1]=f113,-256
600	;;
601	stf.spill.nta [loc0]=f105,-256
602	stf.spill.nta [loc1]=f97,-256
603	;;
604	stf.spill.nta [loc0]=f89,-256
605	stf.spill.nta [loc1]=f81,-256
606	;;
607	stf.spill.nta [loc0]=f73,-256
608	stf.spill.nta [loc1]=f65,-256
609	;;
610	stf.spill.nta [loc0]=f57,-256
611	stf.spill.nta [loc1]=f49,-256
612	adds loc2=96*16-128,in0
613	;;
614	stf.spill.nta [loc0]=f41,-256
615	stf.spill.nta [loc1]=f33,-256
616	adds loc3=96*16-128-128,in0
617	;;
618	stf.spill.nta [loc2]=f120,-256
619	stf.spill.nta [loc3]=f112,-256
620	;;
621	stf.spill.nta [loc2]=f104,-256
622	stf.spill.nta [loc3]=f96,-256
623	;;
624	stf.spill.nta [loc2]=f88,-256
625	stf.spill.nta [loc3]=f80,-256
626	;;
627	stf.spill.nta [loc2]=f72,-256
628	stf.spill.nta [loc3]=f64,-256
629	;;
630	stf.spill.nta [loc2]=f56,-256
631	stf.spill.nta [loc3]=f48,-256
632	;;
633	stf.spill.nta [loc2]=f40
634	stf.spill.nta [loc3]=f32
635	br.ret.sptk.many rp
636END(__ia64_save_fpu)
637
638GLOBAL_ENTRY(__ia64_load_fpu)
639	alloc r2=ar.pfs,1,2,0,0
640	adds r3=128,in0
641	adds r14=256,in0
642	adds r15=384,in0
643	mov loc0=512
644	mov loc1=-1024+16
645	;;
646	ldf.fill.nta f32=[in0],loc0
647	ldf.fill.nta f40=[ r3],loc0
648	ldf.fill.nta f48=[r14],loc0
649	ldf.fill.nta f56=[r15],loc0
650	;;
651	ldf.fill.nta f64=[in0],loc0
652	ldf.fill.nta f72=[ r3],loc0
653	ldf.fill.nta f80=[r14],loc0
654	ldf.fill.nta f88=[r15],loc0
655	;;
656	ldf.fill.nta f96=[in0],loc1
657	ldf.fill.nta f104=[ r3],loc1
658	ldf.fill.nta f112=[r14],loc1
659	ldf.fill.nta f120=[r15],loc1
660	;;
661	ldf.fill.nta f33=[in0],loc0
662	ldf.fill.nta f41=[ r3],loc0
663	ldf.fill.nta f49=[r14],loc0
664	ldf.fill.nta f57=[r15],loc0
665	;;
666	ldf.fill.nta f65=[in0],loc0
667	ldf.fill.nta f73=[ r3],loc0
668	ldf.fill.nta f81=[r14],loc0
669	ldf.fill.nta f89=[r15],loc0
670	;;
671	ldf.fill.nta f97=[in0],loc1
672	ldf.fill.nta f105=[ r3],loc1
673	ldf.fill.nta f113=[r14],loc1
674	ldf.fill.nta f121=[r15],loc1
675	;;
676	ldf.fill.nta f34=[in0],loc0
677	ldf.fill.nta f42=[ r3],loc0
678	ldf.fill.nta f50=[r14],loc0
679	ldf.fill.nta f58=[r15],loc0
680	;;
681	ldf.fill.nta f66=[in0],loc0
682	ldf.fill.nta f74=[ r3],loc0
683	ldf.fill.nta f82=[r14],loc0
684	ldf.fill.nta f90=[r15],loc0
685	;;
686	ldf.fill.nta f98=[in0],loc1
687	ldf.fill.nta f106=[ r3],loc1
688	ldf.fill.nta f114=[r14],loc1
689	ldf.fill.nta f122=[r15],loc1
690	;;
691	ldf.fill.nta f35=[in0],loc0
692	ldf.fill.nta f43=[ r3],loc0
693	ldf.fill.nta f51=[r14],loc0
694	ldf.fill.nta f59=[r15],loc0
695	;;
696	ldf.fill.nta f67=[in0],loc0
697	ldf.fill.nta f75=[ r3],loc0
698	ldf.fill.nta f83=[r14],loc0
699	ldf.fill.nta f91=[r15],loc0
700	;;
701	ldf.fill.nta f99=[in0],loc1
702	ldf.fill.nta f107=[ r3],loc1
703	ldf.fill.nta f115=[r14],loc1
704	ldf.fill.nta f123=[r15],loc1
705	;;
706	ldf.fill.nta f36=[in0],loc0
707	ldf.fill.nta f44=[ r3],loc0
708	ldf.fill.nta f52=[r14],loc0
709	ldf.fill.nta f60=[r15],loc0
710	;;
711	ldf.fill.nta f68=[in0],loc0
712	ldf.fill.nta f76=[ r3],loc0
713	ldf.fill.nta f84=[r14],loc0
714	ldf.fill.nta f92=[r15],loc0
715	;;
716	ldf.fill.nta f100=[in0],loc1
717	ldf.fill.nta f108=[ r3],loc1
718	ldf.fill.nta f116=[r14],loc1
719	ldf.fill.nta f124=[r15],loc1
720	;;
721	ldf.fill.nta f37=[in0],loc0
722	ldf.fill.nta f45=[ r3],loc0
723	ldf.fill.nta f53=[r14],loc0
724	ldf.fill.nta f61=[r15],loc0
725	;;
726	ldf.fill.nta f69=[in0],loc0
727	ldf.fill.nta f77=[ r3],loc0
728	ldf.fill.nta f85=[r14],loc0
729	ldf.fill.nta f93=[r15],loc0
730	;;
731	ldf.fill.nta f101=[in0],loc1
732	ldf.fill.nta f109=[ r3],loc1
733	ldf.fill.nta f117=[r14],loc1
734	ldf.fill.nta f125=[r15],loc1
735	;;
736	ldf.fill.nta f38 =[in0],loc0
737	ldf.fill.nta f46 =[ r3],loc0
738	ldf.fill.nta f54 =[r14],loc0
739	ldf.fill.nta f62 =[r15],loc0
740	;;
741	ldf.fill.nta f70 =[in0],loc0
742	ldf.fill.nta f78 =[ r3],loc0
743	ldf.fill.nta f86 =[r14],loc0
744	ldf.fill.nta f94 =[r15],loc0
745	;;
746	ldf.fill.nta f102=[in0],loc1
747	ldf.fill.nta f110=[ r3],loc1
748	ldf.fill.nta f118=[r14],loc1
749	ldf.fill.nta f126=[r15],loc1
750	;;
751	ldf.fill.nta f39 =[in0],loc0
752	ldf.fill.nta f47 =[ r3],loc0
753	ldf.fill.nta f55 =[r14],loc0
754	ldf.fill.nta f63 =[r15],loc0
755	;;
756	ldf.fill.nta f71 =[in0],loc0
757	ldf.fill.nta f79 =[ r3],loc0
758	ldf.fill.nta f87 =[r14],loc0
759	ldf.fill.nta f95 =[r15],loc0
760	;;
761	ldf.fill.nta f103=[in0]
762	ldf.fill.nta f111=[ r3]
763	ldf.fill.nta f119=[r14]
764	ldf.fill.nta f127=[r15]
765	br.ret.sptk.many rp
766END(__ia64_load_fpu)
767
768GLOBAL_ENTRY(__ia64_init_fpu)
769	stf.spill [sp]=f0		// M3
770	mov	 f32=f0			// F
771	nop.b	 0
772
773	ldfps	 f33,f34=[sp]		// M0
774	ldfps	 f35,f36=[sp]		// M1
775	mov      f37=f0			// F
776	;;
777
778	setf.s	 f38=r0			// M2
779	setf.s	 f39=r0			// M3
780	mov      f40=f0			// F
781
782	ldfps	 f41,f42=[sp]		// M0
783	ldfps	 f43,f44=[sp]		// M1
784	mov      f45=f0			// F
785
786	setf.s	 f46=r0			// M2
787	setf.s	 f47=r0			// M3
788	mov      f48=f0			// F
789
790	ldfps	 f49,f50=[sp]		// M0
791	ldfps	 f51,f52=[sp]		// M1
792	mov      f53=f0			// F
793
794	setf.s	 f54=r0			// M2
795	setf.s	 f55=r0			// M3
796	mov      f56=f0			// F
797
798	ldfps	 f57,f58=[sp]		// M0
799	ldfps	 f59,f60=[sp]		// M1
800	mov      f61=f0			// F
801
802	setf.s	 f62=r0			// M2
803	setf.s	 f63=r0			// M3
804	mov      f64=f0			// F
805
806	ldfps	 f65,f66=[sp]		// M0
807	ldfps	 f67,f68=[sp]		// M1
808	mov      f69=f0			// F
809
810	setf.s	 f70=r0			// M2
811	setf.s	 f71=r0			// M3
812	mov      f72=f0			// F
813
814	ldfps	 f73,f74=[sp]		// M0
815	ldfps	 f75,f76=[sp]		// M1
816	mov      f77=f0			// F
817
818	setf.s	 f78=r0			// M2
819	setf.s	 f79=r0			// M3
820	mov      f80=f0			// F
821
822	ldfps	 f81,f82=[sp]		// M0
823	ldfps	 f83,f84=[sp]		// M1
824	mov      f85=f0			// F
825
826	setf.s	 f86=r0			// M2
827	setf.s	 f87=r0			// M3
828	mov      f88=f0			// F
829
830	/*
831	 * When the instructions are cached, it would be faster to initialize
832	 * the remaining registers with simply mov instructions (F-unit).
833	 * This gets the time down to ~29 cycles.  However, this would use up
834	 * 33 bundles, whereas continuing with the above pattern yields
835	 * 10 bundles and ~30 cycles.
836	 */
837
838	ldfps	 f89,f90=[sp]		// M0
839	ldfps	 f91,f92=[sp]		// M1
840	mov      f93=f0			// F
841
842	setf.s	 f94=r0			// M2
843	setf.s	 f95=r0			// M3
844	mov      f96=f0			// F
845
846	ldfps	 f97,f98=[sp]		// M0
847	ldfps	 f99,f100=[sp]		// M1
848	mov      f101=f0		// F
849
850	setf.s	 f102=r0		// M2
851	setf.s	 f103=r0		// M3
852	mov      f104=f0		// F
853
854	ldfps	 f105,f106=[sp]		// M0
855	ldfps	 f107,f108=[sp]		// M1
856	mov      f109=f0		// F
857
858	setf.s	 f110=r0		// M2
859	setf.s	 f111=r0		// M3
860	mov      f112=f0		// F
861
862	ldfps	 f113,f114=[sp]		// M0
863	ldfps	 f115,f116=[sp]		// M1
864	mov      f117=f0		// F
865
866	setf.s	 f118=r0		// M2
867	setf.s	 f119=r0		// M3
868	mov      f120=f0		// F
869
870	ldfps	 f121,f122=[sp]		// M0
871	ldfps	 f123,f124=[sp]		// M1
872	mov      f125=f0		// F
873
874	setf.s	 f126=r0		// M2
875	setf.s	 f127=r0		// M3
876	br.ret.sptk.many rp		// F
877END(__ia64_init_fpu)
878
879/*
880 * Switch execution mode from virtual to physical
881 *
882 * Inputs:
883 *	r16 = new psr to establish
884 * Output:
885 *	r19 = old virtual address of ar.bsp
886 *	r20 = old virtual address of sp
887 *
888 * Note: RSE must already be in enforced lazy mode
889 */
890GLOBAL_ENTRY(ia64_switch_mode_phys)
891 {
892	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
893	mov r15=ip
894 }
895	;;
896 {
897	flushrs				// must be first insn in group
898	srlz.i
899 }
900	;;
901	mov cr.ipsr=r16			// set new PSR
902	add r3=1f-ia64_switch_mode_phys,r15
903
904	mov r19=ar.bsp
905	mov r20=sp
906	mov r14=rp			// get return address into a general register
907	;;
908
909	// going to physical mode, use tpa to translate virt->phys
910	tpa r17=r19
911	tpa r3=r3
912	tpa sp=sp
913	tpa r14=r14
914	;;
915
916	mov r18=ar.rnat			// save ar.rnat
917	mov ar.bspstore=r17		// this steps on ar.rnat
918	mov cr.iip=r3
919	mov cr.ifs=r0
920	;;
921	mov ar.rnat=r18			// restore ar.rnat
922	rfi				// must be last insn in group
923	;;
9241:	mov rp=r14
925	br.ret.sptk.many rp
926END(ia64_switch_mode_phys)
927
928/*
929 * Switch execution mode from physical to virtual
930 *
931 * Inputs:
932 *	r16 = new psr to establish
933 *	r19 = new bspstore to establish
934 *	r20 = new sp to establish
935 *
936 * Note: RSE must already be in enforced lazy mode
937 */
938GLOBAL_ENTRY(ia64_switch_mode_virt)
939 {
940	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
941	mov r15=ip
942 }
943	;;
944 {
945	flushrs				// must be first insn in group
946	srlz.i
947 }
948	;;
949	mov cr.ipsr=r16			// set new PSR
950	add r3=1f-ia64_switch_mode_virt,r15
951
952	mov r14=rp			// get return address into a general register
953	;;
954
955	// going to virtual
956	//   - for code addresses, set upper bits of addr to KERNEL_START
957	//   - for stack addresses, copy from input argument
958	movl r18=KERNEL_START
959	dep r3=0,r3,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
960	dep r14=0,r14,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
961	mov sp=r20
962	;;
963	or r3=r3,r18
964	or r14=r14,r18
965	;;
966
967	mov r18=ar.rnat			// save ar.rnat
968	mov ar.bspstore=r19		// this steps on ar.rnat
969	mov cr.iip=r3
970	mov cr.ifs=r0
971	;;
972	mov ar.rnat=r18			// restore ar.rnat
973	rfi				// must be last insn in group
974	;;
9751:	mov rp=r14
976	br.ret.sptk.many rp
977END(ia64_switch_mode_virt)
978
979GLOBAL_ENTRY(ia64_delay_loop)
980	.prologue
981{	nop 0			// work around GAS unwind info generation bug...
982	.save ar.lc,r2
983	mov r2=ar.lc
984	.body
985	;;
986	mov ar.lc=r32
987}
988	;;
989	// force loop to be 32-byte aligned (GAS bug means we cannot use .align
990	// inside function body without corrupting unwind info).
991{	nop 0 }
9921:	br.cloop.sptk.few 1b
993	;;
994	mov ar.lc=r2
995	br.ret.sptk.many rp
996END(ia64_delay_loop)
997
998/*
999 * Return a CPU-local timestamp in nano-seconds.  This timestamp is
1000 * NOT synchronized across CPUs its return value must never be
1001 * compared against the values returned on another CPU.  The usage in
1002 * kernel/sched/core.c ensures that.
1003 *
1004 * The return-value of sched_clock() is NOT supposed to wrap-around.
1005 * If it did, it would cause some scheduling hiccups (at the worst).
1006 * Fortunately, with a 64-bit cycle-counter ticking at 100GHz, even
1007 * that would happen only once every 5+ years.
1008 *
1009 * The code below basically calculates:
1010 *
1011 *   (ia64_get_itc() * local_cpu_data->nsec_per_cyc) >> IA64_NSEC_PER_CYC_SHIFT
1012 *
1013 * except that the multiplication and the shift are done with 128-bit
1014 * intermediate precision so that we can produce a full 64-bit result.
1015 */
1016GLOBAL_ENTRY(ia64_native_sched_clock)
1017	addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1018	mov.m r9=ar.itc		// fetch cycle-counter				(35 cyc)
1019	;;
1020	ldf8 f8=[r8]
1021	;;
1022	setf.sig f9=r9		// certain to stall, so issue it _after_ ldf8...
1023	;;
1024	xmpy.lu f10=f9,f8	// calculate low 64 bits of 128-bit product	(4 cyc)
1025	xmpy.hu f11=f9,f8	// calculate high 64 bits of 128-bit product
1026	;;
1027	getf.sig r8=f10		//						(5 cyc)
1028	getf.sig r9=f11
1029	;;
1030	shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1031	br.ret.sptk.many rp
1032END(ia64_native_sched_clock)
1033
1034#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1035GLOBAL_ENTRY(cycle_to_nsec)
1036	alloc r16=ar.pfs,1,0,0,0
1037	addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1038	;;
1039	ldf8 f8=[r8]
1040	;;
1041	setf.sig f9=r32
1042	;;
1043	xmpy.lu f10=f9,f8	// calculate low 64 bits of 128-bit product	(4 cyc)
1044	xmpy.hu f11=f9,f8	// calculate high 64 bits of 128-bit product
1045	;;
1046	getf.sig r8=f10		//						(5 cyc)
1047	getf.sig r9=f11
1048	;;
1049	shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1050	br.ret.sptk.many rp
1051END(cycle_to_nsec)
1052#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
1053
1054#ifdef CONFIG_IA64_BRL_EMU
1055
1056/*
1057 *  Assembly routines used by brl_emu.c to set preserved register state.
1058 */
1059
1060#define SET_REG(reg)				\
1061 GLOBAL_ENTRY(ia64_set_##reg);			\
1062	alloc r16=ar.pfs,1,0,0,0;		\
1063	mov reg=r32;				\
1064	;;					\
1065	br.ret.sptk.many rp;			\
1066 END(ia64_set_##reg)
1067
1068SET_REG(b1);
1069SET_REG(b2);
1070SET_REG(b3);
1071SET_REG(b4);
1072SET_REG(b5);
1073
1074#endif /* CONFIG_IA64_BRL_EMU */
1075
1076#ifdef CONFIG_SMP
1077
1078#ifdef CONFIG_HOTPLUG_CPU
1079GLOBAL_ENTRY(ia64_jump_to_sal)
1080	alloc r16=ar.pfs,1,0,0,0;;
1081	rsm psr.i  | psr.ic
1082{
1083	flushrs
1084	srlz.i
1085}
1086	tpa r25=in0
1087	movl r18=tlb_purge_done;;
1088	DATA_VA_TO_PA(r18);;
1089	mov b1=r18 	// Return location
1090	movl r18=ia64_do_tlb_purge;;
1091	DATA_VA_TO_PA(r18);;
1092	mov b2=r18 	// doing tlb_flush work
1093	mov ar.rsc=0  // Put RSE  in enforced lazy, LE mode
1094	movl r17=1f;;
1095	DATA_VA_TO_PA(r17);;
1096	mov cr.iip=r17
1097	movl r16=SAL_PSR_BITS_TO_SET;;
1098	mov cr.ipsr=r16
1099	mov cr.ifs=r0;;
1100	rfi;;			// note: this unmask MCA/INIT (psr.mc)
11011:
1102	/*
1103	 * Invalidate all TLB data/inst
1104	 */
1105	br.sptk.many b2;; // jump to tlb purge code
1106
1107tlb_purge_done:
1108	RESTORE_REGION_REGS(r25, r17,r18,r19);;
1109	RESTORE_REG(b0, r25, r17);;
1110	RESTORE_REG(b1, r25, r17);;
1111	RESTORE_REG(b2, r25, r17);;
1112	RESTORE_REG(b3, r25, r17);;
1113	RESTORE_REG(b4, r25, r17);;
1114	RESTORE_REG(b5, r25, r17);;
1115	ld8 r1=[r25],0x08;;
1116	ld8 r12=[r25],0x08;;
1117	ld8 r13=[r25],0x08;;
1118	RESTORE_REG(ar.fpsr, r25, r17);;
1119	RESTORE_REG(ar.pfs, r25, r17);;
1120	RESTORE_REG(ar.rnat, r25, r17);;
1121	RESTORE_REG(ar.unat, r25, r17);;
1122	RESTORE_REG(ar.bspstore, r25, r17);;
1123	RESTORE_REG(cr.dcr, r25, r17);;
1124	RESTORE_REG(cr.iva, r25, r17);;
1125	RESTORE_REG(cr.pta, r25, r17);;
1126	srlz.d;;	// required not to violate RAW dependency
1127	RESTORE_REG(cr.itv, r25, r17);;
1128	RESTORE_REG(cr.pmv, r25, r17);;
1129	RESTORE_REG(cr.cmcv, r25, r17);;
1130	RESTORE_REG(cr.lrr0, r25, r17);;
1131	RESTORE_REG(cr.lrr1, r25, r17);;
1132	ld8 r4=[r25],0x08;;
1133	ld8 r5=[r25],0x08;;
1134	ld8 r6=[r25],0x08;;
1135	ld8 r7=[r25],0x08;;
1136	ld8 r17=[r25],0x08;;
1137	mov pr=r17,-1;;
1138	RESTORE_REG(ar.lc, r25, r17);;
1139	/*
1140	 * Now Restore floating point regs
1141	 */
1142	ldf.fill.nta f2=[r25],16;;
1143	ldf.fill.nta f3=[r25],16;;
1144	ldf.fill.nta f4=[r25],16;;
1145	ldf.fill.nta f5=[r25],16;;
1146	ldf.fill.nta f16=[r25],16;;
1147	ldf.fill.nta f17=[r25],16;;
1148	ldf.fill.nta f18=[r25],16;;
1149	ldf.fill.nta f19=[r25],16;;
1150	ldf.fill.nta f20=[r25],16;;
1151	ldf.fill.nta f21=[r25],16;;
1152	ldf.fill.nta f22=[r25],16;;
1153	ldf.fill.nta f23=[r25],16;;
1154	ldf.fill.nta f24=[r25],16;;
1155	ldf.fill.nta f25=[r25],16;;
1156	ldf.fill.nta f26=[r25],16;;
1157	ldf.fill.nta f27=[r25],16;;
1158	ldf.fill.nta f28=[r25],16;;
1159	ldf.fill.nta f29=[r25],16;;
1160	ldf.fill.nta f30=[r25],16;;
1161	ldf.fill.nta f31=[r25],16;;
1162
1163	/*
1164	 * Now that we have done all the register restores
1165	 * we are now ready for the big DIVE to SAL Land
1166	 */
1167	ssm psr.ic;;
1168	srlz.d;;
1169	br.ret.sptk.many b0;;
1170END(ia64_jump_to_sal)
1171#endif /* CONFIG_HOTPLUG_CPU */
1172
1173#endif /* CONFIG_SMP */
1174