xref: /openbmc/linux/arch/ia64/kernel/head.S (revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2)
1/*
2 * Here is where the ball gets rolling as far as the kernel is concerned.
3 * When control is transferred to _start, the bootload has already
4 * loaded us to the correct address.  All that's left to do here is
5 * to set up the kernel's global pointer and jump to the kernel
6 * entry point.
7 *
8 * Copyright (C) 1998-2001, 2003, 2005 Hewlett-Packard Co
9 *	David Mosberger-Tang <davidm@hpl.hp.com>
10 *	Stephane Eranian <eranian@hpl.hp.com>
11 * Copyright (C) 1999 VA Linux Systems
12 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13 * Copyright (C) 1999 Intel Corp.
14 * Copyright (C) 1999 Asit Mallick <Asit.K.Mallick@intel.com>
15 * Copyright (C) 1999 Don Dugger <Don.Dugger@intel.com>
16 * Copyright (C) 2002 Fenghua Yu <fenghua.yu@intel.com>
17 *   -Optimize __ia64_save_fpu() and __ia64_load_fpu() for Itanium 2.
18 */
19
20#include <linux/config.h>
21
22#include <asm/asmmacro.h>
23#include <asm/fpu.h>
24#include <asm/kregs.h>
25#include <asm/mmu_context.h>
26#include <asm/offsets.h>
27#include <asm/pal.h>
28#include <asm/pgtable.h>
29#include <asm/processor.h>
30#include <asm/ptrace.h>
31#include <asm/system.h>
32
33	.section __special_page_section,"ax"
34
35	.global empty_zero_page
36empty_zero_page:
37	.skip PAGE_SIZE
38
39	.global swapper_pg_dir
40swapper_pg_dir:
41	.skip PAGE_SIZE
42
43	.rodata
44halt_msg:
45	stringz "Halting kernel\n"
46
47	.text
48
49	.global start_ap
50
51	/*
52	 * Start the kernel.  When the bootloader passes control to _start(), r28
53	 * points to the address of the boot parameter area.  Execution reaches
54	 * here in physical mode.
55	 */
56GLOBAL_ENTRY(_start)
57start_ap:
58	.prologue
59	.save rp, r0		// terminate unwind chain with a NULL rp
60	.body
61
62	rsm psr.i | psr.ic
63	;;
64	srlz.i
65	;;
66	/*
67	 * Initialize kernel region registers:
68	 *	rr[0]: VHPT enabled, page size = PAGE_SHIFT
69	 *	rr[1]: VHPT enabled, page size = PAGE_SHIFT
70	 *	rr[2]: VHPT enabled, page size = PAGE_SHIFT
71	 *	rr[3]: VHPT enabled, page size = PAGE_SHIFT
72	 *	rr[4]: VHPT enabled, page size = PAGE_SHIFT
73	 *	rr[5]: VHPT enabled, page size = PAGE_SHIFT
74	 *	rr[6]: VHPT disabled, page size = IA64_GRANULE_SHIFT
75	 *	rr[7]: VHPT disabled, page size = IA64_GRANULE_SHIFT
76	 * We initialize all of them to prevent inadvertently assuming
77	 * something about the state of address translation early in boot.
78	 */
79	mov r6=((ia64_rid(IA64_REGION_ID_KERNEL, (0<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
80	movl r7=(0<<61)
81	mov r8=((ia64_rid(IA64_REGION_ID_KERNEL, (1<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
82	movl r9=(1<<61)
83	mov r10=((ia64_rid(IA64_REGION_ID_KERNEL, (2<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
84	movl r11=(2<<61)
85	mov r12=((ia64_rid(IA64_REGION_ID_KERNEL, (3<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
86	movl r13=(3<<61)
87	mov r14=((ia64_rid(IA64_REGION_ID_KERNEL, (4<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
88	movl r15=(4<<61)
89	mov r16=((ia64_rid(IA64_REGION_ID_KERNEL, (5<<61)) << 8) | (PAGE_SHIFT << 2) | 1)
90	movl r17=(5<<61)
91	mov r18=((ia64_rid(IA64_REGION_ID_KERNEL, (6<<61)) << 8) | (IA64_GRANULE_SHIFT << 2))
92	movl r19=(6<<61)
93	mov r20=((ia64_rid(IA64_REGION_ID_KERNEL, (7<<61)) << 8) | (IA64_GRANULE_SHIFT << 2))
94	movl r21=(7<<61)
95	;;
96	mov rr[r7]=r6
97	mov rr[r9]=r8
98	mov rr[r11]=r10
99	mov rr[r13]=r12
100	mov rr[r15]=r14
101	mov rr[r17]=r16
102	mov rr[r19]=r18
103	mov rr[r21]=r20
104	;;
105	/*
106	 * Now pin mappings into the TLB for kernel text and data
107	 */
108	mov r18=KERNEL_TR_PAGE_SHIFT<<2
109	movl r17=KERNEL_START
110	;;
111	mov cr.itir=r18
112	mov cr.ifa=r17
113	mov r16=IA64_TR_KERNEL
114	mov r3=ip
115	movl r18=PAGE_KERNEL
116	;;
117	dep r2=0,r3,0,KERNEL_TR_PAGE_SHIFT
118	;;
119	or r18=r2,r18
120	;;
121	srlz.i
122	;;
123	itr.i itr[r16]=r18
124	;;
125	itr.d dtr[r16]=r18
126	;;
127	srlz.i
128
129	/*
130	 * Switch into virtual mode:
131	 */
132	movl r16=(IA64_PSR_IT|IA64_PSR_IC|IA64_PSR_DT|IA64_PSR_RT|IA64_PSR_DFH|IA64_PSR_BN \
133		  |IA64_PSR_DI)
134	;;
135	mov cr.ipsr=r16
136	movl r17=1f
137	;;
138	mov cr.iip=r17
139	mov cr.ifs=r0
140	;;
141	rfi
142	;;
1431:	// now we are in virtual mode
144
145	// set IVT entry point---can't access I/O ports without it
146	movl r3=ia64_ivt
147	;;
148	mov cr.iva=r3
149	movl r2=FPSR_DEFAULT
150	;;
151	srlz.i
152	movl gp=__gp
153
154	mov ar.fpsr=r2
155	;;
156
157#define isAP	p2	// are we an Application Processor?
158#define isBP	p3	// are we the Bootstrap Processor?
159
160#ifdef CONFIG_SMP
161	/*
162	 * Find the init_task for the currently booting CPU.  At poweron, and in
163	 * UP mode, task_for_booting_cpu is NULL.
164	 */
165	movl r3=task_for_booting_cpu
166 	;;
167	ld8 r3=[r3]
168	movl r2=init_task
169	;;
170	cmp.eq isBP,isAP=r3,r0
171	;;
172(isAP)	mov r2=r3
173#else
174	movl r2=init_task
175	cmp.eq isBP,isAP=r0,r0
176#endif
177	;;
178	tpa r3=r2		// r3 == phys addr of task struct
179	mov r16=-1
180(isBP)	br.cond.dpnt .load_current // BP stack is on region 5 --- no need to map it
181
182	// load mapping for stack (virtaddr in r2, physaddr in r3)
183	rsm psr.ic
184	movl r17=PAGE_KERNEL
185	;;
186	srlz.d
187	dep r18=0,r3,0,12
188	;;
189	or r18=r17,r18
190	dep r2=-1,r3,61,3	// IMVA of task
191	;;
192	mov r17=rr[r2]
193	shr.u r16=r3,IA64_GRANULE_SHIFT
194	;;
195	dep r17=0,r17,8,24
196	;;
197	mov cr.itir=r17
198	mov cr.ifa=r2
199
200	mov r19=IA64_TR_CURRENT_STACK
201	;;
202	itr.d dtr[r19]=r18
203	;;
204	ssm psr.ic
205	srlz.d
206  	;;
207
208.load_current:
209	// load the "current" pointer (r13) and ar.k6 with the current task
210	mov IA64_KR(CURRENT)=r2		// virtual address
211	mov IA64_KR(CURRENT_STACK)=r16
212	mov r13=r2
213	/*
214	 * Reserve space at the top of the stack for "struct pt_regs".  Kernel threads
215	 * don't store interesting values in that structure, but the space still needs
216	 * to be there because time-critical stuff such as the context switching can
217	 * be implemented more efficiently (for example, __switch_to()
218	 * always sets the psr.dfh bit of the task it is switching to).
219	 */
220	addl r12=IA64_STK_OFFSET-IA64_PT_REGS_SIZE-16,r2
221	addl r2=IA64_RBS_OFFSET,r2	// initialize the RSE
222	mov ar.rsc=0		// place RSE in enforced lazy mode
223	;;
224	loadrs			// clear the dirty partition
225	;;
226	mov ar.bspstore=r2	// establish the new RSE stack
227	;;
228	mov ar.rsc=0x3		// place RSE in eager mode
229
230(isBP)	dep r28=-1,r28,61,3	// make address virtual
231(isBP)	movl r2=ia64_boot_param
232	;;
233(isBP)	st8 [r2]=r28		// save the address of the boot param area passed by the bootloader
234
235#ifdef CONFIG_SMP
236(isAP)	br.call.sptk.many rp=start_secondary
237.ret0:
238(isAP)	br.cond.sptk self
239#endif
240
241	// This is executed by the bootstrap processor (bsp) only:
242
243#ifdef CONFIG_IA64_FW_EMU
244	// initialize PAL & SAL emulator:
245	br.call.sptk.many rp=sys_fw_init
246.ret1:
247#endif
248	br.call.sptk.many rp=start_kernel
249.ret2:	addl r3=@ltoff(halt_msg),gp
250	;;
251	alloc r2=ar.pfs,8,0,2,0
252	;;
253	ld8 out0=[r3]
254	br.call.sptk.many b0=console_print
255
256self:	hint @pause
257	br.sptk.many self		// endless loop
258END(_start)
259
260GLOBAL_ENTRY(ia64_save_debug_regs)
261	alloc r16=ar.pfs,1,0,0,0
262	mov r20=ar.lc			// preserve ar.lc
263	mov ar.lc=IA64_NUM_DBG_REGS-1
264	mov r18=0
265	add r19=IA64_NUM_DBG_REGS*8,in0
266	;;
2671:	mov r16=dbr[r18]
268#ifdef CONFIG_ITANIUM
269	;;
270	srlz.d
271#endif
272	mov r17=ibr[r18]
273	add r18=1,r18
274	;;
275	st8.nta [in0]=r16,8
276	st8.nta [r19]=r17,8
277	br.cloop.sptk.many 1b
278	;;
279	mov ar.lc=r20			// restore ar.lc
280	br.ret.sptk.many rp
281END(ia64_save_debug_regs)
282
283GLOBAL_ENTRY(ia64_load_debug_regs)
284	alloc r16=ar.pfs,1,0,0,0
285	lfetch.nta [in0]
286	mov r20=ar.lc			// preserve ar.lc
287	add r19=IA64_NUM_DBG_REGS*8,in0
288	mov ar.lc=IA64_NUM_DBG_REGS-1
289	mov r18=-1
290	;;
2911:	ld8.nta r16=[in0],8
292	ld8.nta r17=[r19],8
293	add r18=1,r18
294	;;
295	mov dbr[r18]=r16
296#ifdef CONFIG_ITANIUM
297	;;
298	srlz.d				// Errata 132 (NoFix status)
299#endif
300	mov ibr[r18]=r17
301	br.cloop.sptk.many 1b
302	;;
303	mov ar.lc=r20			// restore ar.lc
304	br.ret.sptk.many rp
305END(ia64_load_debug_regs)
306
307GLOBAL_ENTRY(__ia64_save_fpu)
308	alloc r2=ar.pfs,1,4,0,0
309	adds loc0=96*16-16,in0
310	adds loc1=96*16-16-128,in0
311	;;
312	stf.spill.nta [loc0]=f127,-256
313	stf.spill.nta [loc1]=f119,-256
314	;;
315	stf.spill.nta [loc0]=f111,-256
316	stf.spill.nta [loc1]=f103,-256
317	;;
318	stf.spill.nta [loc0]=f95,-256
319	stf.spill.nta [loc1]=f87,-256
320	;;
321	stf.spill.nta [loc0]=f79,-256
322	stf.spill.nta [loc1]=f71,-256
323	;;
324	stf.spill.nta [loc0]=f63,-256
325	stf.spill.nta [loc1]=f55,-256
326	adds loc2=96*16-32,in0
327	;;
328	stf.spill.nta [loc0]=f47,-256
329	stf.spill.nta [loc1]=f39,-256
330	adds loc3=96*16-32-128,in0
331	;;
332	stf.spill.nta [loc2]=f126,-256
333	stf.spill.nta [loc3]=f118,-256
334	;;
335	stf.spill.nta [loc2]=f110,-256
336	stf.spill.nta [loc3]=f102,-256
337	;;
338	stf.spill.nta [loc2]=f94,-256
339	stf.spill.nta [loc3]=f86,-256
340	;;
341	stf.spill.nta [loc2]=f78,-256
342	stf.spill.nta [loc3]=f70,-256
343	;;
344	stf.spill.nta [loc2]=f62,-256
345	stf.spill.nta [loc3]=f54,-256
346	adds loc0=96*16-48,in0
347	;;
348	stf.spill.nta [loc2]=f46,-256
349	stf.spill.nta [loc3]=f38,-256
350	adds loc1=96*16-48-128,in0
351	;;
352	stf.spill.nta [loc0]=f125,-256
353	stf.spill.nta [loc1]=f117,-256
354	;;
355	stf.spill.nta [loc0]=f109,-256
356	stf.spill.nta [loc1]=f101,-256
357	;;
358	stf.spill.nta [loc0]=f93,-256
359	stf.spill.nta [loc1]=f85,-256
360	;;
361	stf.spill.nta [loc0]=f77,-256
362	stf.spill.nta [loc1]=f69,-256
363	;;
364	stf.spill.nta [loc0]=f61,-256
365	stf.spill.nta [loc1]=f53,-256
366	adds loc2=96*16-64,in0
367	;;
368	stf.spill.nta [loc0]=f45,-256
369	stf.spill.nta [loc1]=f37,-256
370	adds loc3=96*16-64-128,in0
371	;;
372	stf.spill.nta [loc2]=f124,-256
373	stf.spill.nta [loc3]=f116,-256
374	;;
375	stf.spill.nta [loc2]=f108,-256
376	stf.spill.nta [loc3]=f100,-256
377	;;
378	stf.spill.nta [loc2]=f92,-256
379	stf.spill.nta [loc3]=f84,-256
380	;;
381	stf.spill.nta [loc2]=f76,-256
382	stf.spill.nta [loc3]=f68,-256
383	;;
384	stf.spill.nta [loc2]=f60,-256
385	stf.spill.nta [loc3]=f52,-256
386	adds loc0=96*16-80,in0
387	;;
388	stf.spill.nta [loc2]=f44,-256
389	stf.spill.nta [loc3]=f36,-256
390	adds loc1=96*16-80-128,in0
391	;;
392	stf.spill.nta [loc0]=f123,-256
393	stf.spill.nta [loc1]=f115,-256
394	;;
395	stf.spill.nta [loc0]=f107,-256
396	stf.spill.nta [loc1]=f99,-256
397	;;
398	stf.spill.nta [loc0]=f91,-256
399	stf.spill.nta [loc1]=f83,-256
400	;;
401	stf.spill.nta [loc0]=f75,-256
402	stf.spill.nta [loc1]=f67,-256
403	;;
404	stf.spill.nta [loc0]=f59,-256
405	stf.spill.nta [loc1]=f51,-256
406	adds loc2=96*16-96,in0
407	;;
408	stf.spill.nta [loc0]=f43,-256
409	stf.spill.nta [loc1]=f35,-256
410	adds loc3=96*16-96-128,in0
411	;;
412	stf.spill.nta [loc2]=f122,-256
413	stf.spill.nta [loc3]=f114,-256
414	;;
415	stf.spill.nta [loc2]=f106,-256
416	stf.spill.nta [loc3]=f98,-256
417	;;
418	stf.spill.nta [loc2]=f90,-256
419	stf.spill.nta [loc3]=f82,-256
420	;;
421	stf.spill.nta [loc2]=f74,-256
422	stf.spill.nta [loc3]=f66,-256
423	;;
424	stf.spill.nta [loc2]=f58,-256
425	stf.spill.nta [loc3]=f50,-256
426	adds loc0=96*16-112,in0
427	;;
428	stf.spill.nta [loc2]=f42,-256
429	stf.spill.nta [loc3]=f34,-256
430	adds loc1=96*16-112-128,in0
431	;;
432	stf.spill.nta [loc0]=f121,-256
433	stf.spill.nta [loc1]=f113,-256
434	;;
435	stf.spill.nta [loc0]=f105,-256
436	stf.spill.nta [loc1]=f97,-256
437	;;
438	stf.spill.nta [loc0]=f89,-256
439	stf.spill.nta [loc1]=f81,-256
440	;;
441	stf.spill.nta [loc0]=f73,-256
442	stf.spill.nta [loc1]=f65,-256
443	;;
444	stf.spill.nta [loc0]=f57,-256
445	stf.spill.nta [loc1]=f49,-256
446	adds loc2=96*16-128,in0
447	;;
448	stf.spill.nta [loc0]=f41,-256
449	stf.spill.nta [loc1]=f33,-256
450	adds loc3=96*16-128-128,in0
451	;;
452	stf.spill.nta [loc2]=f120,-256
453	stf.spill.nta [loc3]=f112,-256
454	;;
455	stf.spill.nta [loc2]=f104,-256
456	stf.spill.nta [loc3]=f96,-256
457	;;
458	stf.spill.nta [loc2]=f88,-256
459	stf.spill.nta [loc3]=f80,-256
460	;;
461	stf.spill.nta [loc2]=f72,-256
462	stf.spill.nta [loc3]=f64,-256
463	;;
464	stf.spill.nta [loc2]=f56,-256
465	stf.spill.nta [loc3]=f48,-256
466	;;
467	stf.spill.nta [loc2]=f40
468	stf.spill.nta [loc3]=f32
469	br.ret.sptk.many rp
470END(__ia64_save_fpu)
471
472GLOBAL_ENTRY(__ia64_load_fpu)
473	alloc r2=ar.pfs,1,2,0,0
474	adds r3=128,in0
475	adds r14=256,in0
476	adds r15=384,in0
477	mov loc0=512
478	mov loc1=-1024+16
479	;;
480	ldf.fill.nta f32=[in0],loc0
481	ldf.fill.nta f40=[ r3],loc0
482	ldf.fill.nta f48=[r14],loc0
483	ldf.fill.nta f56=[r15],loc0
484	;;
485	ldf.fill.nta f64=[in0],loc0
486	ldf.fill.nta f72=[ r3],loc0
487	ldf.fill.nta f80=[r14],loc0
488	ldf.fill.nta f88=[r15],loc0
489	;;
490	ldf.fill.nta f96=[in0],loc1
491	ldf.fill.nta f104=[ r3],loc1
492	ldf.fill.nta f112=[r14],loc1
493	ldf.fill.nta f120=[r15],loc1
494	;;
495	ldf.fill.nta f33=[in0],loc0
496	ldf.fill.nta f41=[ r3],loc0
497	ldf.fill.nta f49=[r14],loc0
498	ldf.fill.nta f57=[r15],loc0
499	;;
500	ldf.fill.nta f65=[in0],loc0
501	ldf.fill.nta f73=[ r3],loc0
502	ldf.fill.nta f81=[r14],loc0
503	ldf.fill.nta f89=[r15],loc0
504	;;
505	ldf.fill.nta f97=[in0],loc1
506	ldf.fill.nta f105=[ r3],loc1
507	ldf.fill.nta f113=[r14],loc1
508	ldf.fill.nta f121=[r15],loc1
509	;;
510	ldf.fill.nta f34=[in0],loc0
511	ldf.fill.nta f42=[ r3],loc0
512	ldf.fill.nta f50=[r14],loc0
513	ldf.fill.nta f58=[r15],loc0
514	;;
515	ldf.fill.nta f66=[in0],loc0
516	ldf.fill.nta f74=[ r3],loc0
517	ldf.fill.nta f82=[r14],loc0
518	ldf.fill.nta f90=[r15],loc0
519	;;
520	ldf.fill.nta f98=[in0],loc1
521	ldf.fill.nta f106=[ r3],loc1
522	ldf.fill.nta f114=[r14],loc1
523	ldf.fill.nta f122=[r15],loc1
524	;;
525	ldf.fill.nta f35=[in0],loc0
526	ldf.fill.nta f43=[ r3],loc0
527	ldf.fill.nta f51=[r14],loc0
528	ldf.fill.nta f59=[r15],loc0
529	;;
530	ldf.fill.nta f67=[in0],loc0
531	ldf.fill.nta f75=[ r3],loc0
532	ldf.fill.nta f83=[r14],loc0
533	ldf.fill.nta f91=[r15],loc0
534	;;
535	ldf.fill.nta f99=[in0],loc1
536	ldf.fill.nta f107=[ r3],loc1
537	ldf.fill.nta f115=[r14],loc1
538	ldf.fill.nta f123=[r15],loc1
539	;;
540	ldf.fill.nta f36=[in0],loc0
541	ldf.fill.nta f44=[ r3],loc0
542	ldf.fill.nta f52=[r14],loc0
543	ldf.fill.nta f60=[r15],loc0
544	;;
545	ldf.fill.nta f68=[in0],loc0
546	ldf.fill.nta f76=[ r3],loc0
547	ldf.fill.nta f84=[r14],loc0
548	ldf.fill.nta f92=[r15],loc0
549	;;
550	ldf.fill.nta f100=[in0],loc1
551	ldf.fill.nta f108=[ r3],loc1
552	ldf.fill.nta f116=[r14],loc1
553	ldf.fill.nta f124=[r15],loc1
554	;;
555	ldf.fill.nta f37=[in0],loc0
556	ldf.fill.nta f45=[ r3],loc0
557	ldf.fill.nta f53=[r14],loc0
558	ldf.fill.nta f61=[r15],loc0
559	;;
560	ldf.fill.nta f69=[in0],loc0
561	ldf.fill.nta f77=[ r3],loc0
562	ldf.fill.nta f85=[r14],loc0
563	ldf.fill.nta f93=[r15],loc0
564	;;
565	ldf.fill.nta f101=[in0],loc1
566	ldf.fill.nta f109=[ r3],loc1
567	ldf.fill.nta f117=[r14],loc1
568	ldf.fill.nta f125=[r15],loc1
569	;;
570	ldf.fill.nta f38 =[in0],loc0
571	ldf.fill.nta f46 =[ r3],loc0
572	ldf.fill.nta f54 =[r14],loc0
573	ldf.fill.nta f62 =[r15],loc0
574	;;
575	ldf.fill.nta f70 =[in0],loc0
576	ldf.fill.nta f78 =[ r3],loc0
577	ldf.fill.nta f86 =[r14],loc0
578	ldf.fill.nta f94 =[r15],loc0
579	;;
580	ldf.fill.nta f102=[in0],loc1
581	ldf.fill.nta f110=[ r3],loc1
582	ldf.fill.nta f118=[r14],loc1
583	ldf.fill.nta f126=[r15],loc1
584	;;
585	ldf.fill.nta f39 =[in0],loc0
586	ldf.fill.nta f47 =[ r3],loc0
587	ldf.fill.nta f55 =[r14],loc0
588	ldf.fill.nta f63 =[r15],loc0
589	;;
590	ldf.fill.nta f71 =[in0],loc0
591	ldf.fill.nta f79 =[ r3],loc0
592	ldf.fill.nta f87 =[r14],loc0
593	ldf.fill.nta f95 =[r15],loc0
594	;;
595	ldf.fill.nta f103=[in0]
596	ldf.fill.nta f111=[ r3]
597	ldf.fill.nta f119=[r14]
598	ldf.fill.nta f127=[r15]
599	br.ret.sptk.many rp
600END(__ia64_load_fpu)
601
602GLOBAL_ENTRY(__ia64_init_fpu)
603	stf.spill [sp]=f0		// M3
604	mov	 f32=f0			// F
605	nop.b	 0
606
607	ldfps	 f33,f34=[sp]		// M0
608	ldfps	 f35,f36=[sp]		// M1
609	mov      f37=f0			// F
610	;;
611
612	setf.s	 f38=r0			// M2
613	setf.s	 f39=r0			// M3
614	mov      f40=f0			// F
615
616	ldfps	 f41,f42=[sp]		// M0
617	ldfps	 f43,f44=[sp]		// M1
618	mov      f45=f0			// F
619
620	setf.s	 f46=r0			// M2
621	setf.s	 f47=r0			// M3
622	mov      f48=f0			// F
623
624	ldfps	 f49,f50=[sp]		// M0
625	ldfps	 f51,f52=[sp]		// M1
626	mov      f53=f0			// F
627
628	setf.s	 f54=r0			// M2
629	setf.s	 f55=r0			// M3
630	mov      f56=f0			// F
631
632	ldfps	 f57,f58=[sp]		// M0
633	ldfps	 f59,f60=[sp]		// M1
634	mov      f61=f0			// F
635
636	setf.s	 f62=r0			// M2
637	setf.s	 f63=r0			// M3
638	mov      f64=f0			// F
639
640	ldfps	 f65,f66=[sp]		// M0
641	ldfps	 f67,f68=[sp]		// M1
642	mov      f69=f0			// F
643
644	setf.s	 f70=r0			// M2
645	setf.s	 f71=r0			// M3
646	mov      f72=f0			// F
647
648	ldfps	 f73,f74=[sp]		// M0
649	ldfps	 f75,f76=[sp]		// M1
650	mov      f77=f0			// F
651
652	setf.s	 f78=r0			// M2
653	setf.s	 f79=r0			// M3
654	mov      f80=f0			// F
655
656	ldfps	 f81,f82=[sp]		// M0
657	ldfps	 f83,f84=[sp]		// M1
658	mov      f85=f0			// F
659
660	setf.s	 f86=r0			// M2
661	setf.s	 f87=r0			// M3
662	mov      f88=f0			// F
663
664	/*
665	 * When the instructions are cached, it would be faster to initialize
666	 * the remaining registers with simply mov instructions (F-unit).
667	 * This gets the time down to ~29 cycles.  However, this would use up
668	 * 33 bundles, whereas continuing with the above pattern yields
669	 * 10 bundles and ~30 cycles.
670	 */
671
672	ldfps	 f89,f90=[sp]		// M0
673	ldfps	 f91,f92=[sp]		// M1
674	mov      f93=f0			// F
675
676	setf.s	 f94=r0			// M2
677	setf.s	 f95=r0			// M3
678	mov      f96=f0			// F
679
680	ldfps	 f97,f98=[sp]		// M0
681	ldfps	 f99,f100=[sp]		// M1
682	mov      f101=f0		// F
683
684	setf.s	 f102=r0		// M2
685	setf.s	 f103=r0		// M3
686	mov      f104=f0		// F
687
688	ldfps	 f105,f106=[sp]		// M0
689	ldfps	 f107,f108=[sp]		// M1
690	mov      f109=f0		// F
691
692	setf.s	 f110=r0		// M2
693	setf.s	 f111=r0		// M3
694	mov      f112=f0		// F
695
696	ldfps	 f113,f114=[sp]		// M0
697	ldfps	 f115,f116=[sp]		// M1
698	mov      f117=f0		// F
699
700	setf.s	 f118=r0		// M2
701	setf.s	 f119=r0		// M3
702	mov      f120=f0		// F
703
704	ldfps	 f121,f122=[sp]		// M0
705	ldfps	 f123,f124=[sp]		// M1
706	mov      f125=f0		// F
707
708	setf.s	 f126=r0		// M2
709	setf.s	 f127=r0		// M3
710	br.ret.sptk.many rp		// F
711END(__ia64_init_fpu)
712
713/*
714 * Switch execution mode from virtual to physical
715 *
716 * Inputs:
717 *	r16 = new psr to establish
718 * Output:
719 *	r19 = old virtual address of ar.bsp
720 *	r20 = old virtual address of sp
721 *
722 * Note: RSE must already be in enforced lazy mode
723 */
724GLOBAL_ENTRY(ia64_switch_mode_phys)
725 {
726	alloc r2=ar.pfs,0,0,0,0
727	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
728	mov r15=ip
729 }
730	;;
731 {
732	flushrs				// must be first insn in group
733	srlz.i
734 }
735	;;
736	mov cr.ipsr=r16			// set new PSR
737	add r3=1f-ia64_switch_mode_phys,r15
738
739	mov r19=ar.bsp
740	mov r20=sp
741	mov r14=rp			// get return address into a general register
742	;;
743
744	// going to physical mode, use tpa to translate virt->phys
745	tpa r17=r19
746	tpa r3=r3
747	tpa sp=sp
748	tpa r14=r14
749	;;
750
751	mov r18=ar.rnat			// save ar.rnat
752	mov ar.bspstore=r17		// this steps on ar.rnat
753	mov cr.iip=r3
754	mov cr.ifs=r0
755	;;
756	mov ar.rnat=r18			// restore ar.rnat
757	rfi				// must be last insn in group
758	;;
7591:	mov rp=r14
760	br.ret.sptk.many rp
761END(ia64_switch_mode_phys)
762
763/*
764 * Switch execution mode from physical to virtual
765 *
766 * Inputs:
767 *	r16 = new psr to establish
768 *	r19 = new bspstore to establish
769 *	r20 = new sp to establish
770 *
771 * Note: RSE must already be in enforced lazy mode
772 */
773GLOBAL_ENTRY(ia64_switch_mode_virt)
774 {
775	alloc r2=ar.pfs,0,0,0,0
776	rsm psr.i | psr.ic		// disable interrupts and interrupt collection
777	mov r15=ip
778 }
779	;;
780 {
781	flushrs				// must be first insn in group
782	srlz.i
783 }
784	;;
785	mov cr.ipsr=r16			// set new PSR
786	add r3=1f-ia64_switch_mode_virt,r15
787
788	mov r14=rp			// get return address into a general register
789	;;
790
791	// going to virtual
792	//   - for code addresses, set upper bits of addr to KERNEL_START
793	//   - for stack addresses, copy from input argument
794	movl r18=KERNEL_START
795	dep r3=0,r3,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
796	dep r14=0,r14,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
797	mov sp=r20
798	;;
799	or r3=r3,r18
800	or r14=r14,r18
801	;;
802
803	mov r18=ar.rnat			// save ar.rnat
804	mov ar.bspstore=r19		// this steps on ar.rnat
805	mov cr.iip=r3
806	mov cr.ifs=r0
807	;;
808	mov ar.rnat=r18			// restore ar.rnat
809	rfi				// must be last insn in group
810	;;
8111:	mov rp=r14
812	br.ret.sptk.many rp
813END(ia64_switch_mode_virt)
814
815GLOBAL_ENTRY(ia64_delay_loop)
816	.prologue
817{	nop 0			// work around GAS unwind info generation bug...
818	.save ar.lc,r2
819	mov r2=ar.lc
820	.body
821	;;
822	mov ar.lc=r32
823}
824	;;
825	// force loop to be 32-byte aligned (GAS bug means we cannot use .align
826	// inside function body without corrupting unwind info).
827{	nop 0 }
8281:	br.cloop.sptk.few 1b
829	;;
830	mov ar.lc=r2
831	br.ret.sptk.many rp
832END(ia64_delay_loop)
833
834/*
835 * Return a CPU-local timestamp in nano-seconds.  This timestamp is
836 * NOT synchronized across CPUs its return value must never be
837 * compared against the values returned on another CPU.  The usage in
838 * kernel/sched.c ensures that.
839 *
840 * The return-value of sched_clock() is NOT supposed to wrap-around.
841 * If it did, it would cause some scheduling hiccups (at the worst).
842 * Fortunately, with a 64-bit cycle-counter ticking at 100GHz, even
843 * that would happen only once every 5+ years.
844 *
845 * The code below basically calculates:
846 *
847 *   (ia64_get_itc() * local_cpu_data->nsec_per_cyc) >> IA64_NSEC_PER_CYC_SHIFT
848 *
849 * except that the multiplication and the shift are done with 128-bit
850 * intermediate precision so that we can produce a full 64-bit result.
851 */
852GLOBAL_ENTRY(sched_clock)
853	addl r8=THIS_CPU(cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
854	mov.m r9=ar.itc		// fetch cycle-counter				(35 cyc)
855	;;
856	ldf8 f8=[r8]
857	;;
858	setf.sig f9=r9		// certain to stall, so issue it _after_ ldf8...
859	;;
860	xmpy.lu f10=f9,f8	// calculate low 64 bits of 128-bit product	(4 cyc)
861	xmpy.hu f11=f9,f8	// calculate high 64 bits of 128-bit product
862	;;
863	getf.sig r8=f10		//						(5 cyc)
864	getf.sig r9=f11
865	;;
866	shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
867	br.ret.sptk.many rp
868END(sched_clock)
869
870GLOBAL_ENTRY(start_kernel_thread)
871	.prologue
872	.save rp, r0				// this is the end of the call-chain
873	.body
874	alloc r2 = ar.pfs, 0, 0, 2, 0
875	mov out0 = r9
876	mov out1 = r11;;
877	br.call.sptk.many rp = kernel_thread_helper;;
878	mov out0 = r8
879	br.call.sptk.many rp = sys_exit;;
8801:	br.sptk.few 1b				// not reached
881END(start_kernel_thread)
882
883#ifdef CONFIG_IA64_BRL_EMU
884
885/*
886 *  Assembly routines used by brl_emu.c to set preserved register state.
887 */
888
889#define SET_REG(reg)				\
890 GLOBAL_ENTRY(ia64_set_##reg);			\
891	alloc r16=ar.pfs,1,0,0,0;		\
892	mov reg=r32;				\
893	;;					\
894	br.ret.sptk.many rp;			\
895 END(ia64_set_##reg)
896
897SET_REG(b1);
898SET_REG(b2);
899SET_REG(b3);
900SET_REG(b4);
901SET_REG(b5);
902
903#endif /* CONFIG_IA64_BRL_EMU */
904
905#ifdef CONFIG_SMP
906	/*
907	 * This routine handles spinlock contention.  It uses a non-standard calling
908	 * convention to avoid converting leaf routines into interior routines.  Because
909	 * of this special convention, there are several restrictions:
910	 *
911	 * - do not use gp relative variables, this code is called from the kernel
912	 *   and from modules, r1 is undefined.
913	 * - do not use stacked registers, the caller owns them.
914	 * - do not use the scratch stack space, the caller owns it.
915	 * - do not use any registers other than the ones listed below
916	 *
917	 * Inputs:
918	 *   ar.pfs - saved CFM of caller
919	 *   ar.ccv - 0 (and available for use)
920	 *   r27    - flags from spin_lock_irqsave or 0.  Must be preserved.
921	 *   r28    - available for use.
922	 *   r29    - available for use.
923	 *   r30    - available for use.
924	 *   r31    - address of lock, available for use.
925	 *   b6     - return address
926	 *   p14    - available for use.
927	 *   p15    - used to track flag status.
928	 *
929	 * If you patch this code to use more registers, do not forget to update
930	 * the clobber lists for spin_lock() in include/asm-ia64/spinlock.h.
931	 */
932
933#if __GNUC__ < 3 || (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
934
935GLOBAL_ENTRY(ia64_spinlock_contention_pre3_4)
936	.prologue
937	.save ar.pfs, r0	// this code effectively has a zero frame size
938	.save rp, r28
939	.body
940	nop 0
941	tbit.nz p15,p0=r27,IA64_PSR_I_BIT
942	.restore sp		// pop existing prologue after next insn
943	mov b6 = r28
944	.prologue
945	.save ar.pfs, r0
946	.altrp b6
947	.body
948	;;
949(p15)	ssm psr.i		// reenable interrupts if they were on
950				// DavidM says that srlz.d is slow and is not required in this case
951.wait:
952	// exponential backoff, kdb, lockmeter etc. go in here
953	hint @pause
954	ld4 r30=[r31]		// don't use ld4.bias; if it's contended, we won't write the word
955	nop 0
956	;;
957	cmp4.ne p14,p0=r30,r0
958(p14)	br.cond.sptk.few .wait
959(p15)	rsm psr.i		// disable interrupts if we reenabled them
960	br.cond.sptk.few b6	// lock is now free, try to acquire
961	.global ia64_spinlock_contention_pre3_4_end	// for kernprof
962ia64_spinlock_contention_pre3_4_end:
963END(ia64_spinlock_contention_pre3_4)
964
965#else
966
967GLOBAL_ENTRY(ia64_spinlock_contention)
968	.prologue
969	.altrp b6
970	.body
971	tbit.nz p15,p0=r27,IA64_PSR_I_BIT
972	;;
973.wait:
974(p15)	ssm psr.i		// reenable interrupts if they were on
975				// DavidM says that srlz.d is slow and is not required in this case
976.wait2:
977	// exponential backoff, kdb, lockmeter etc. go in here
978	hint @pause
979	ld4 r30=[r31]		// don't use ld4.bias; if it's contended, we won't write the word
980	;;
981	cmp4.ne p14,p0=r30,r0
982	mov r30 = 1
983(p14)	br.cond.sptk.few .wait2
984(p15)	rsm psr.i		// disable interrupts if we reenabled them
985	;;
986	cmpxchg4.acq r30=[r31], r30, ar.ccv
987	;;
988	cmp4.ne p14,p0=r0,r30
989(p14)	br.cond.sptk.few .wait
990
991	br.ret.sptk.many b6	// lock is now taken
992END(ia64_spinlock_contention)
993
994#endif
995
996#endif /* CONFIG_SMP */
997