1/*
2 * ultra.S: Don't expand these all over the place...
3 *
4 * Copyright (C) 1997, 2000, 2008 David S. Miller (davem@davemloft.net)
5 */
6
7#include <asm/asi.h>
8#include <asm/pgtable.h>
9#include <asm/page.h>
10#include <asm/spitfire.h>
11#include <asm/mmu_context.h>
12#include <asm/mmu.h>
13#include <asm/pil.h>
14#include <asm/head.h>
15#include <asm/thread_info.h>
16#include <asm/cacheflush.h>
17#include <asm/hypervisor.h>
18#include <asm/cpudata.h>
19
20	/* Basically, most of the Spitfire vs. Cheetah madness
21	 * has to do with the fact that Cheetah does not support
22	 * IMMU flushes out of the secondary context.  Someone needs
23	 * to throw a south lake birthday party for the folks
24	 * in Microelectronics who refused to fix this shit.
25	 */
26
27	/* This file is meant to be read efficiently by the CPU, not humans.
28	 * Staraj sie tego nikomu nie pierdolnac...
29	 */
30	.text
31	.align		32
32	.globl		__flush_tlb_mm
33__flush_tlb_mm:		/* 18 insns */
34	/* %o0=(ctx & TAG_CONTEXT_BITS), %o1=SECONDARY_CONTEXT */
35	ldxa		[%o1] ASI_DMMU, %g2
36	cmp		%g2, %o0
37	bne,pn		%icc, __spitfire_flush_tlb_mm_slow
38	 mov		0x50, %g3
39	stxa		%g0, [%g3] ASI_DMMU_DEMAP
40	stxa		%g0, [%g3] ASI_IMMU_DEMAP
41	sethi		%hi(KERNBASE), %g3
42	flush		%g3
43	retl
44	 nop
45	nop
46	nop
47	nop
48	nop
49	nop
50	nop
51	nop
52	nop
53	nop
54
55	.align		32
56	.globl		__flush_tlb_page
57__flush_tlb_page:	/* 22 insns */
58	/* %o0 = context, %o1 = vaddr */
59	rdpr		%pstate, %g7
60	andn		%g7, PSTATE_IE, %g2
61	wrpr		%g2, %pstate
62	mov		SECONDARY_CONTEXT, %o4
63	ldxa		[%o4] ASI_DMMU, %g2
64	stxa		%o0, [%o4] ASI_DMMU
65	andcc		%o1, 1, %g0
66	andn		%o1, 1, %o3
67	be,pn		%icc, 1f
68	 or		%o3, 0x10, %o3
69	stxa		%g0, [%o3] ASI_IMMU_DEMAP
701:	stxa		%g0, [%o3] ASI_DMMU_DEMAP
71	membar		#Sync
72	stxa		%g2, [%o4] ASI_DMMU
73	sethi		%hi(KERNBASE), %o4
74	flush		%o4
75	retl
76	 wrpr		%g7, 0x0, %pstate
77	nop
78	nop
79	nop
80	nop
81
82	.align		32
83	.globl		__flush_tlb_pending
84__flush_tlb_pending:	/* 26 insns */
85	/* %o0 = context, %o1 = nr, %o2 = vaddrs[] */
86	rdpr		%pstate, %g7
87	sllx		%o1, 3, %o1
88	andn		%g7, PSTATE_IE, %g2
89	wrpr		%g2, %pstate
90	mov		SECONDARY_CONTEXT, %o4
91	ldxa		[%o4] ASI_DMMU, %g2
92	stxa		%o0, [%o4] ASI_DMMU
931:	sub		%o1, (1 << 3), %o1
94	ldx		[%o2 + %o1], %o3
95	andcc		%o3, 1, %g0
96	andn		%o3, 1, %o3
97	be,pn		%icc, 2f
98	 or		%o3, 0x10, %o3
99	stxa		%g0, [%o3] ASI_IMMU_DEMAP
1002:	stxa		%g0, [%o3] ASI_DMMU_DEMAP
101	membar		#Sync
102	brnz,pt		%o1, 1b
103	 nop
104	stxa		%g2, [%o4] ASI_DMMU
105	sethi		%hi(KERNBASE), %o4
106	flush		%o4
107	retl
108	 wrpr		%g7, 0x0, %pstate
109	nop
110	nop
111	nop
112	nop
113
114	.align		32
115	.globl		__flush_tlb_kernel_range
116__flush_tlb_kernel_range:	/* 16 insns */
117	/* %o0=start, %o1=end */
118	cmp		%o0, %o1
119	be,pn		%xcc, 2f
120	 sethi		%hi(PAGE_SIZE), %o4
121	sub		%o1, %o0, %o3
122	sub		%o3, %o4, %o3
123	or		%o0, 0x20, %o0		! Nucleus
1241:	stxa		%g0, [%o0 + %o3] ASI_DMMU_DEMAP
125	stxa		%g0, [%o0 + %o3] ASI_IMMU_DEMAP
126	membar		#Sync
127	brnz,pt		%o3, 1b
128	 sub		%o3, %o4, %o3
1292:	sethi		%hi(KERNBASE), %o3
130	flush		%o3
131	retl
132	 nop
133	nop
134
135__spitfire_flush_tlb_mm_slow:
136	rdpr		%pstate, %g1
137	wrpr		%g1, PSTATE_IE, %pstate
138	stxa		%o0, [%o1] ASI_DMMU
139	stxa		%g0, [%g3] ASI_DMMU_DEMAP
140	stxa		%g0, [%g3] ASI_IMMU_DEMAP
141	flush		%g6
142	stxa		%g2, [%o1] ASI_DMMU
143	sethi		%hi(KERNBASE), %o1
144	flush		%o1
145	retl
146	 wrpr		%g1, 0, %pstate
147
148/*
149 * The following code flushes one page_size worth.
150 */
151	.section .kprobes.text, "ax"
152	.align		32
153	.globl		__flush_icache_page
154__flush_icache_page:	/* %o0 = phys_page */
155	srlx		%o0, PAGE_SHIFT, %o0
156	sethi		%hi(PAGE_OFFSET), %g1
157	sllx		%o0, PAGE_SHIFT, %o0
158	sethi		%hi(PAGE_SIZE), %g2
159	ldx		[%g1 + %lo(PAGE_OFFSET)], %g1
160	add		%o0, %g1, %o0
1611:	subcc		%g2, 32, %g2
162	bne,pt		%icc, 1b
163	 flush		%o0 + %g2
164	retl
165	 nop
166
167#ifdef DCACHE_ALIASING_POSSIBLE
168
169#if (PAGE_SHIFT != 13)
170#error only page shift of 13 is supported by dcache flush
171#endif
172
173#define DTAG_MASK 0x3
174
175	/* This routine is Spitfire specific so the hardcoded
176	 * D-cache size and line-size are OK.
177	 */
178	.align		64
179	.globl		__flush_dcache_page
180__flush_dcache_page:	/* %o0=kaddr, %o1=flush_icache */
181	sethi		%hi(PAGE_OFFSET), %g1
182	ldx		[%g1 + %lo(PAGE_OFFSET)], %g1
183	sub		%o0, %g1, %o0			! physical address
184	srlx		%o0, 11, %o0			! make D-cache TAG
185	sethi		%hi(1 << 14), %o2		! D-cache size
186	sub		%o2, (1 << 5), %o2		! D-cache line size
1871:	ldxa		[%o2] ASI_DCACHE_TAG, %o3	! load D-cache TAG
188	andcc		%o3, DTAG_MASK, %g0		! Valid?
189	be,pn		%xcc, 2f			! Nope, branch
190	 andn		%o3, DTAG_MASK, %o3		! Clear valid bits
191	cmp		%o3, %o0			! TAG match?
192	bne,pt		%xcc, 2f			! Nope, branch
193	 nop
194	stxa		%g0, [%o2] ASI_DCACHE_TAG	! Invalidate TAG
195	membar		#Sync
1962:	brnz,pt		%o2, 1b
197	 sub		%o2, (1 << 5), %o2		! D-cache line size
198
199	/* The I-cache does not snoop local stores so we
200	 * better flush that too when necessary.
201	 */
202	brnz,pt		%o1, __flush_icache_page
203	 sllx		%o0, 11, %o0
204	retl
205	 nop
206
207#endif /* DCACHE_ALIASING_POSSIBLE */
208
209	.previous
210
211	/* Cheetah specific versions, patched at boot time. */
212__cheetah_flush_tlb_mm: /* 19 insns */
213	rdpr		%pstate, %g7
214	andn		%g7, PSTATE_IE, %g2
215	wrpr		%g2, 0x0, %pstate
216	wrpr		%g0, 1, %tl
217	mov		PRIMARY_CONTEXT, %o2
218	mov		0x40, %g3
219	ldxa		[%o2] ASI_DMMU, %g2
220	srlx		%g2, CTX_PGSZ1_NUC_SHIFT, %o1
221	sllx		%o1, CTX_PGSZ1_NUC_SHIFT, %o1
222	or		%o0, %o1, %o0	/* Preserve nucleus page size fields */
223	stxa		%o0, [%o2] ASI_DMMU
224	stxa		%g0, [%g3] ASI_DMMU_DEMAP
225	stxa		%g0, [%g3] ASI_IMMU_DEMAP
226	stxa		%g2, [%o2] ASI_DMMU
227	sethi		%hi(KERNBASE), %o2
228	flush		%o2
229	wrpr		%g0, 0, %tl
230	retl
231	 wrpr		%g7, 0x0, %pstate
232
233__cheetah_flush_tlb_page:	/* 22 insns */
234	/* %o0 = context, %o1 = vaddr */
235	rdpr		%pstate, %g7
236	andn		%g7, PSTATE_IE, %g2
237	wrpr		%g2, 0x0, %pstate
238	wrpr		%g0, 1, %tl
239	mov		PRIMARY_CONTEXT, %o4
240	ldxa		[%o4] ASI_DMMU, %g2
241	srlx		%g2, CTX_PGSZ1_NUC_SHIFT, %o3
242	sllx		%o3, CTX_PGSZ1_NUC_SHIFT, %o3
243	or		%o0, %o3, %o0	/* Preserve nucleus page size fields */
244	stxa		%o0, [%o4] ASI_DMMU
245	andcc		%o1, 1, %g0
246	be,pn		%icc, 1f
247	 andn		%o1, 1, %o3
248	stxa		%g0, [%o3] ASI_IMMU_DEMAP
2491:	stxa		%g0, [%o3] ASI_DMMU_DEMAP
250	membar		#Sync
251	stxa		%g2, [%o4] ASI_DMMU
252	sethi		%hi(KERNBASE), %o4
253	flush		%o4
254	wrpr		%g0, 0, %tl
255	retl
256	 wrpr		%g7, 0x0, %pstate
257
258__cheetah_flush_tlb_pending:	/* 27 insns */
259	/* %o0 = context, %o1 = nr, %o2 = vaddrs[] */
260	rdpr		%pstate, %g7
261	sllx		%o1, 3, %o1
262	andn		%g7, PSTATE_IE, %g2
263	wrpr		%g2, 0x0, %pstate
264	wrpr		%g0, 1, %tl
265	mov		PRIMARY_CONTEXT, %o4
266	ldxa		[%o4] ASI_DMMU, %g2
267	srlx		%g2, CTX_PGSZ1_NUC_SHIFT, %o3
268	sllx		%o3, CTX_PGSZ1_NUC_SHIFT, %o3
269	or		%o0, %o3, %o0	/* Preserve nucleus page size fields */
270	stxa		%o0, [%o4] ASI_DMMU
2711:	sub		%o1, (1 << 3), %o1
272	ldx		[%o2 + %o1], %o3
273	andcc		%o3, 1, %g0
274	be,pn		%icc, 2f
275	 andn		%o3, 1, %o3
276	stxa		%g0, [%o3] ASI_IMMU_DEMAP
2772:	stxa		%g0, [%o3] ASI_DMMU_DEMAP
278	membar		#Sync
279	brnz,pt		%o1, 1b
280	 nop
281	stxa		%g2, [%o4] ASI_DMMU
282	sethi		%hi(KERNBASE), %o4
283	flush		%o4
284	wrpr		%g0, 0, %tl
285	retl
286	 wrpr		%g7, 0x0, %pstate
287
288#ifdef DCACHE_ALIASING_POSSIBLE
289__cheetah_flush_dcache_page: /* 11 insns */
290	sethi		%hi(PAGE_OFFSET), %g1
291	ldx		[%g1 + %lo(PAGE_OFFSET)], %g1
292	sub		%o0, %g1, %o0
293	sethi		%hi(PAGE_SIZE), %o4
2941:	subcc		%o4, (1 << 5), %o4
295	stxa		%g0, [%o0 + %o4] ASI_DCACHE_INVALIDATE
296	membar		#Sync
297	bne,pt		%icc, 1b
298	 nop
299	retl		/* I-cache flush never needed on Cheetah, see callers. */
300	 nop
301#endif /* DCACHE_ALIASING_POSSIBLE */
302
303	/* Hypervisor specific versions, patched at boot time.  */
304__hypervisor_tlb_tl0_error:
305	save		%sp, -192, %sp
306	mov		%i0, %o0
307	call		hypervisor_tlbop_error
308	 mov		%i1, %o1
309	ret
310	 restore
311
312__hypervisor_flush_tlb_mm: /* 10 insns */
313	mov		%o0, %o2	/* ARG2: mmu context */
314	mov		0, %o0		/* ARG0: CPU lists unimplemented */
315	mov		0, %o1		/* ARG1: CPU lists unimplemented */
316	mov		HV_MMU_ALL, %o3	/* ARG3: flags */
317	mov		HV_FAST_MMU_DEMAP_CTX, %o5
318	ta		HV_FAST_TRAP
319	brnz,pn		%o0, __hypervisor_tlb_tl0_error
320	 mov		HV_FAST_MMU_DEMAP_CTX, %o1
321	retl
322	 nop
323
324__hypervisor_flush_tlb_page: /* 11 insns */
325	/* %o0 = context, %o1 = vaddr */
326	mov		%o0, %g2
327	mov		%o1, %o0              /* ARG0: vaddr + IMMU-bit */
328	mov		%g2, %o1	      /* ARG1: mmu context */
329	mov		HV_MMU_ALL, %o2	      /* ARG2: flags */
330	srlx		%o0, PAGE_SHIFT, %o0
331	sllx		%o0, PAGE_SHIFT, %o0
332	ta		HV_MMU_UNMAP_ADDR_TRAP
333	brnz,pn		%o0, __hypervisor_tlb_tl0_error
334	 mov		HV_MMU_UNMAP_ADDR_TRAP, %o1
335	retl
336	 nop
337
338__hypervisor_flush_tlb_pending: /* 16 insns */
339	/* %o0 = context, %o1 = nr, %o2 = vaddrs[] */
340	sllx		%o1, 3, %g1
341	mov		%o2, %g2
342	mov		%o0, %g3
3431:	sub		%g1, (1 << 3), %g1
344	ldx		[%g2 + %g1], %o0      /* ARG0: vaddr + IMMU-bit */
345	mov		%g3, %o1	      /* ARG1: mmu context */
346	mov		HV_MMU_ALL, %o2	      /* ARG2: flags */
347	srlx		%o0, PAGE_SHIFT, %o0
348	sllx		%o0, PAGE_SHIFT, %o0
349	ta		HV_MMU_UNMAP_ADDR_TRAP
350	brnz,pn		%o0, __hypervisor_tlb_tl0_error
351	 mov		HV_MMU_UNMAP_ADDR_TRAP, %o1
352	brnz,pt		%g1, 1b
353	 nop
354	retl
355	 nop
356
357__hypervisor_flush_tlb_kernel_range: /* 16 insns */
358	/* %o0=start, %o1=end */
359	cmp		%o0, %o1
360	be,pn		%xcc, 2f
361	 sethi		%hi(PAGE_SIZE), %g3
362	mov		%o0, %g1
363	sub		%o1, %g1, %g2
364	sub		%g2, %g3, %g2
3651:	add		%g1, %g2, %o0	/* ARG0: virtual address */
366	mov		0, %o1		/* ARG1: mmu context */
367	mov		HV_MMU_ALL, %o2	/* ARG2: flags */
368	ta		HV_MMU_UNMAP_ADDR_TRAP
369	brnz,pn		%o0, __hypervisor_tlb_tl0_error
370	 mov		HV_MMU_UNMAP_ADDR_TRAP, %o1
371	brnz,pt		%g2, 1b
372	 sub		%g2, %g3, %g2
3732:	retl
374	 nop
375
376#ifdef DCACHE_ALIASING_POSSIBLE
377	/* XXX Niagara and friends have an 8K cache, so no aliasing is
378	 * XXX possible, but nothing explicit in the Hypervisor API
379	 * XXX guarantees this.
380	 */
381__hypervisor_flush_dcache_page:	/* 2 insns */
382	retl
383	 nop
384#endif
385
386tlb_patch_one:
3871:	lduw		[%o1], %g1
388	stw		%g1, [%o0]
389	flush		%o0
390	subcc		%o2, 1, %o2
391	add		%o1, 4, %o1
392	bne,pt		%icc, 1b
393	 add		%o0, 4, %o0
394	retl
395	 nop
396
397	.globl		cheetah_patch_cachetlbops
398cheetah_patch_cachetlbops:
399	save		%sp, -128, %sp
400
401	sethi		%hi(__flush_tlb_mm), %o0
402	or		%o0, %lo(__flush_tlb_mm), %o0
403	sethi		%hi(__cheetah_flush_tlb_mm), %o1
404	or		%o1, %lo(__cheetah_flush_tlb_mm), %o1
405	call		tlb_patch_one
406	 mov		19, %o2
407
408	sethi		%hi(__flush_tlb_page), %o0
409	or		%o0, %lo(__flush_tlb_page), %o0
410	sethi		%hi(__cheetah_flush_tlb_page), %o1
411	or		%o1, %lo(__cheetah_flush_tlb_page), %o1
412	call		tlb_patch_one
413	 mov		22, %o2
414
415	sethi		%hi(__flush_tlb_pending), %o0
416	or		%o0, %lo(__flush_tlb_pending), %o0
417	sethi		%hi(__cheetah_flush_tlb_pending), %o1
418	or		%o1, %lo(__cheetah_flush_tlb_pending), %o1
419	call		tlb_patch_one
420	 mov		27, %o2
421
422#ifdef DCACHE_ALIASING_POSSIBLE
423	sethi		%hi(__flush_dcache_page), %o0
424	or		%o0, %lo(__flush_dcache_page), %o0
425	sethi		%hi(__cheetah_flush_dcache_page), %o1
426	or		%o1, %lo(__cheetah_flush_dcache_page), %o1
427	call		tlb_patch_one
428	 mov		11, %o2
429#endif /* DCACHE_ALIASING_POSSIBLE */
430
431	ret
432	 restore
433
434#ifdef CONFIG_SMP
435	/* These are all called by the slaves of a cross call, at
436	 * trap level 1, with interrupts fully disabled.
437	 *
438	 * Register usage:
439	 *   %g5	mm->context	(all tlb flushes)
440	 *   %g1	address arg 1	(tlb page and range flushes)
441	 *   %g7	address arg 2	(tlb range flush only)
442	 *
443	 *   %g6	scratch 1
444	 *   %g2	scratch 2
445	 *   %g3	scratch 3
446	 *   %g4	scratch 4
447	 */
448	.align		32
449	.globl		xcall_flush_tlb_mm
450xcall_flush_tlb_mm:	/* 21 insns */
451	mov		PRIMARY_CONTEXT, %g2
452	ldxa		[%g2] ASI_DMMU, %g3
453	srlx		%g3, CTX_PGSZ1_NUC_SHIFT, %g4
454	sllx		%g4, CTX_PGSZ1_NUC_SHIFT, %g4
455	or		%g5, %g4, %g5	/* Preserve nucleus page size fields */
456	stxa		%g5, [%g2] ASI_DMMU
457	mov		0x40, %g4
458	stxa		%g0, [%g4] ASI_DMMU_DEMAP
459	stxa		%g0, [%g4] ASI_IMMU_DEMAP
460	stxa		%g3, [%g2] ASI_DMMU
461	retry
462	nop
463	nop
464	nop
465	nop
466	nop
467	nop
468	nop
469	nop
470	nop
471	nop
472
473	.globl		xcall_flush_tlb_page
474xcall_flush_tlb_page:	/* 17 insns */
475	/* %g5=context, %g1=vaddr */
476	mov		PRIMARY_CONTEXT, %g4
477	ldxa		[%g4] ASI_DMMU, %g2
478	srlx		%g2, CTX_PGSZ1_NUC_SHIFT, %g4
479	sllx		%g4, CTX_PGSZ1_NUC_SHIFT, %g4
480	or		%g5, %g4, %g5
481	mov		PRIMARY_CONTEXT, %g4
482	stxa		%g5, [%g4] ASI_DMMU
483	andcc		%g1, 0x1, %g0
484	be,pn		%icc, 2f
485	 andn		%g1, 0x1, %g5
486	stxa		%g0, [%g5] ASI_IMMU_DEMAP
4872:	stxa		%g0, [%g5] ASI_DMMU_DEMAP
488	membar		#Sync
489	stxa		%g2, [%g4] ASI_DMMU
490	retry
491	nop
492	nop
493
494	.globl		xcall_flush_tlb_kernel_range
495xcall_flush_tlb_kernel_range:	/* 25 insns */
496	sethi		%hi(PAGE_SIZE - 1), %g2
497	or		%g2, %lo(PAGE_SIZE - 1), %g2
498	andn		%g1, %g2, %g1
499	andn		%g7, %g2, %g7
500	sub		%g7, %g1, %g3
501	add		%g2, 1, %g2
502	sub		%g3, %g2, %g3
503	or		%g1, 0x20, %g1		! Nucleus
5041:	stxa		%g0, [%g1 + %g3] ASI_DMMU_DEMAP
505	stxa		%g0, [%g1 + %g3] ASI_IMMU_DEMAP
506	membar		#Sync
507	brnz,pt		%g3, 1b
508	 sub		%g3, %g2, %g3
509	retry
510	nop
511	nop
512	nop
513	nop
514	nop
515	nop
516	nop
517	nop
518	nop
519	nop
520	nop
521
522	/* This runs in a very controlled environment, so we do
523	 * not need to worry about BH races etc.
524	 */
525	.globl		xcall_sync_tick
526xcall_sync_tick:
527
528661:	rdpr		%pstate, %g2
529	wrpr		%g2, PSTATE_IG | PSTATE_AG, %pstate
530	.section	.sun4v_2insn_patch, "ax"
531	.word		661b
532	nop
533	nop
534	.previous
535
536	rdpr		%pil, %g2
537	wrpr		%g0, PIL_NORMAL_MAX, %pil
538	sethi		%hi(109f), %g7
539	b,pt		%xcc, etrap_irq
540109:	 or		%g7, %lo(109b), %g7
541#ifdef CONFIG_TRACE_IRQFLAGS
542	call		trace_hardirqs_off
543	 nop
544#endif
545	call		smp_synchronize_tick_client
546	 nop
547	b		rtrap_xcall
548	 ldx		[%sp + PTREGS_OFF + PT_V9_TSTATE], %l1
549
550	.globl		xcall_fetch_glob_regs
551xcall_fetch_glob_regs:
552	sethi		%hi(global_cpu_snapshot), %g1
553	or		%g1, %lo(global_cpu_snapshot), %g1
554	__GET_CPUID(%g2)
555	sllx		%g2, 6, %g3
556	add		%g1, %g3, %g1
557	rdpr		%tstate, %g7
558	stx		%g7, [%g1 + GR_SNAP_TSTATE]
559	rdpr		%tpc, %g7
560	stx		%g7, [%g1 + GR_SNAP_TPC]
561	rdpr		%tnpc, %g7
562	stx		%g7, [%g1 + GR_SNAP_TNPC]
563	stx		%o7, [%g1 + GR_SNAP_O7]
564	stx		%i7, [%g1 + GR_SNAP_I7]
565	/* Don't try this at home kids... */
566	rdpr		%cwp, %g3
567	sub		%g3, 1, %g7
568	wrpr		%g7, %cwp
569	mov		%i7, %g7
570	wrpr		%g3, %cwp
571	stx		%g7, [%g1 + GR_SNAP_RPC]
572	sethi		%hi(trap_block), %g7
573	or		%g7, %lo(trap_block), %g7
574	sllx		%g2, TRAP_BLOCK_SZ_SHIFT, %g2
575	add		%g7, %g2, %g7
576	ldx		[%g7 + TRAP_PER_CPU_THREAD], %g3
577	stx		%g3, [%g1 + GR_SNAP_THREAD]
578	retry
579
580	.globl		xcall_fetch_glob_pmu
581xcall_fetch_glob_pmu:
582	sethi		%hi(global_cpu_snapshot), %g1
583	or		%g1, %lo(global_cpu_snapshot), %g1
584	__GET_CPUID(%g2)
585	sllx		%g2, 6, %g3
586	add		%g1, %g3, %g1
587	rd		%pic, %g7
588	stx		%g7, [%g1 + (4 * 8)]
589	rd		%pcr, %g7
590	stx		%g7, [%g1 + (0 * 8)]
591	retry
592
593	.globl		xcall_fetch_glob_pmu_n4
594xcall_fetch_glob_pmu_n4:
595	sethi		%hi(global_cpu_snapshot), %g1
596	or		%g1, %lo(global_cpu_snapshot), %g1
597	__GET_CPUID(%g2)
598	sllx		%g2, 6, %g3
599	add		%g1, %g3, %g1
600
601	ldxa		[%g0] ASI_PIC, %g7
602	stx		%g7, [%g1 + (4 * 8)]
603	mov		0x08, %g3
604	ldxa		[%g3] ASI_PIC, %g7
605	stx		%g7, [%g1 + (5 * 8)]
606	mov		0x10, %g3
607	ldxa		[%g3] ASI_PIC, %g7
608	stx		%g7, [%g1 + (6 * 8)]
609	mov		0x18, %g3
610	ldxa		[%g3] ASI_PIC, %g7
611	stx		%g7, [%g1 + (7 * 8)]
612
613	mov		%o0, %g2
614	mov		%o1, %g3
615	mov		%o5, %g7
616
617	mov		HV_FAST_VT_GET_PERFREG, %o5
618	mov		3, %o0
619	ta		HV_FAST_TRAP
620	stx		%o1, [%g1 + (3 * 8)]
621	mov		HV_FAST_VT_GET_PERFREG, %o5
622	mov		2, %o0
623	ta		HV_FAST_TRAP
624	stx		%o1, [%g1 + (2 * 8)]
625	mov		HV_FAST_VT_GET_PERFREG, %o5
626	mov		1, %o0
627	ta		HV_FAST_TRAP
628	stx		%o1, [%g1 + (1 * 8)]
629	mov		HV_FAST_VT_GET_PERFREG, %o5
630	mov		0, %o0
631	ta		HV_FAST_TRAP
632	stx		%o1, [%g1 + (0 * 8)]
633
634	mov		%g2, %o0
635	mov		%g3, %o1
636	mov		%g7, %o5
637
638	retry
639
640#ifdef DCACHE_ALIASING_POSSIBLE
641	.align		32
642	.globl		xcall_flush_dcache_page_cheetah
643xcall_flush_dcache_page_cheetah: /* %g1 == physical page address */
644	sethi		%hi(PAGE_SIZE), %g3
6451:	subcc		%g3, (1 << 5), %g3
646	stxa		%g0, [%g1 + %g3] ASI_DCACHE_INVALIDATE
647	membar		#Sync
648	bne,pt		%icc, 1b
649	 nop
650	retry
651	nop
652#endif /* DCACHE_ALIASING_POSSIBLE */
653
654	.globl		xcall_flush_dcache_page_spitfire
655xcall_flush_dcache_page_spitfire: /* %g1 == physical page address
656				     %g7 == kernel page virtual address
657				     %g5 == (page->mapping != NULL)  */
658#ifdef DCACHE_ALIASING_POSSIBLE
659	srlx		%g1, (13 - 2), %g1	! Form tag comparitor
660	sethi		%hi(L1DCACHE_SIZE), %g3	! D$ size == 16K
661	sub		%g3, (1 << 5), %g3	! D$ linesize == 32
6621:	ldxa		[%g3] ASI_DCACHE_TAG, %g2
663	andcc		%g2, 0x3, %g0
664	be,pn		%xcc, 2f
665	 andn		%g2, 0x3, %g2
666	cmp		%g2, %g1
667
668	bne,pt		%xcc, 2f
669	 nop
670	stxa		%g0, [%g3] ASI_DCACHE_TAG
671	membar		#Sync
6722:	cmp		%g3, 0
673	bne,pt		%xcc, 1b
674	 sub		%g3, (1 << 5), %g3
675
676	brz,pn		%g5, 2f
677#endif /* DCACHE_ALIASING_POSSIBLE */
678	 sethi		%hi(PAGE_SIZE), %g3
679
6801:	flush		%g7
681	subcc		%g3, (1 << 5), %g3
682	bne,pt		%icc, 1b
683	 add		%g7, (1 << 5), %g7
684
6852:	retry
686	nop
687	nop
688
689	/* %g5:	error
690	 * %g6:	tlb op
691	 */
692__hypervisor_tlb_xcall_error:
693	mov	%g5, %g4
694	mov	%g6, %g5
695	ba,pt	%xcc, etrap
696	 rd	%pc, %g7
697	mov	%l4, %o0
698	call	hypervisor_tlbop_error_xcall
699	 mov	%l5, %o1
700	ba,a,pt	%xcc, rtrap
701
702	.globl		__hypervisor_xcall_flush_tlb_mm
703__hypervisor_xcall_flush_tlb_mm: /* 21 insns */
704	/* %g5=ctx, g1,g2,g3,g4,g7=scratch, %g6=unusable */
705	mov		%o0, %g2
706	mov		%o1, %g3
707	mov		%o2, %g4
708	mov		%o3, %g1
709	mov		%o5, %g7
710	clr		%o0		/* ARG0: CPU lists unimplemented */
711	clr		%o1		/* ARG1: CPU lists unimplemented */
712	mov		%g5, %o2	/* ARG2: mmu context */
713	mov		HV_MMU_ALL, %o3	/* ARG3: flags */
714	mov		HV_FAST_MMU_DEMAP_CTX, %o5
715	ta		HV_FAST_TRAP
716	mov		HV_FAST_MMU_DEMAP_CTX, %g6
717	brnz,pn		%o0, __hypervisor_tlb_xcall_error
718	 mov		%o0, %g5
719	mov		%g2, %o0
720	mov		%g3, %o1
721	mov		%g4, %o2
722	mov		%g1, %o3
723	mov		%g7, %o5
724	membar		#Sync
725	retry
726
727	.globl		__hypervisor_xcall_flush_tlb_page
728__hypervisor_xcall_flush_tlb_page: /* 17 insns */
729	/* %g5=ctx, %g1=vaddr */
730	mov		%o0, %g2
731	mov		%o1, %g3
732	mov		%o2, %g4
733	mov		%g1, %o0	        /* ARG0: virtual address */
734	mov		%g5, %o1		/* ARG1: mmu context */
735	mov		HV_MMU_ALL, %o2		/* ARG2: flags */
736	srlx		%o0, PAGE_SHIFT, %o0
737	sllx		%o0, PAGE_SHIFT, %o0
738	ta		HV_MMU_UNMAP_ADDR_TRAP
739	mov		HV_MMU_UNMAP_ADDR_TRAP, %g6
740	brnz,a,pn	%o0, __hypervisor_tlb_xcall_error
741	 mov		%o0, %g5
742	mov		%g2, %o0
743	mov		%g3, %o1
744	mov		%g4, %o2
745	membar		#Sync
746	retry
747
748	.globl		__hypervisor_xcall_flush_tlb_kernel_range
749__hypervisor_xcall_flush_tlb_kernel_range: /* 25 insns */
750	/* %g1=start, %g7=end, g2,g3,g4,g5,g6=scratch */
751	sethi		%hi(PAGE_SIZE - 1), %g2
752	or		%g2, %lo(PAGE_SIZE - 1), %g2
753	andn		%g1, %g2, %g1
754	andn		%g7, %g2, %g7
755	sub		%g7, %g1, %g3
756	add		%g2, 1, %g2
757	sub		%g3, %g2, %g3
758	mov		%o0, %g2
759	mov		%o1, %g4
760	mov		%o2, %g7
7611:	add		%g1, %g3, %o0	/* ARG0: virtual address */
762	mov		0, %o1		/* ARG1: mmu context */
763	mov		HV_MMU_ALL, %o2	/* ARG2: flags */
764	ta		HV_MMU_UNMAP_ADDR_TRAP
765	mov		HV_MMU_UNMAP_ADDR_TRAP, %g6
766	brnz,pn		%o0, __hypervisor_tlb_xcall_error
767	 mov		%o0, %g5
768	sethi		%hi(PAGE_SIZE), %o2
769	brnz,pt		%g3, 1b
770	 sub		%g3, %o2, %g3
771	mov		%g2, %o0
772	mov		%g4, %o1
773	mov		%g7, %o2
774	membar		#Sync
775	retry
776
777	/* These just get rescheduled to PIL vectors. */
778	.globl		xcall_call_function
779xcall_call_function:
780	wr		%g0, (1 << PIL_SMP_CALL_FUNC), %set_softint
781	retry
782
783	.globl		xcall_call_function_single
784xcall_call_function_single:
785	wr		%g0, (1 << PIL_SMP_CALL_FUNC_SNGL), %set_softint
786	retry
787
788	.globl		xcall_receive_signal
789xcall_receive_signal:
790	wr		%g0, (1 << PIL_SMP_RECEIVE_SIGNAL), %set_softint
791	retry
792
793	.globl		xcall_capture
794xcall_capture:
795	wr		%g0, (1 << PIL_SMP_CAPTURE), %set_softint
796	retry
797
798	.globl		xcall_new_mmu_context_version
799xcall_new_mmu_context_version:
800	wr		%g0, (1 << PIL_SMP_CTX_NEW_VERSION), %set_softint
801	retry
802
803#ifdef CONFIG_KGDB
804	.globl		xcall_kgdb_capture
805xcall_kgdb_capture:
806	wr		%g0, (1 << PIL_KGDB_CAPTURE), %set_softint
807	retry
808#endif
809
810#endif /* CONFIG_SMP */
811
812
813	.globl		hypervisor_patch_cachetlbops
814hypervisor_patch_cachetlbops:
815	save		%sp, -128, %sp
816
817	sethi		%hi(__flush_tlb_mm), %o0
818	or		%o0, %lo(__flush_tlb_mm), %o0
819	sethi		%hi(__hypervisor_flush_tlb_mm), %o1
820	or		%o1, %lo(__hypervisor_flush_tlb_mm), %o1
821	call		tlb_patch_one
822	 mov		10, %o2
823
824	sethi		%hi(__flush_tlb_page), %o0
825	or		%o0, %lo(__flush_tlb_page), %o0
826	sethi		%hi(__hypervisor_flush_tlb_page), %o1
827	or		%o1, %lo(__hypervisor_flush_tlb_page), %o1
828	call		tlb_patch_one
829	 mov		11, %o2
830
831	sethi		%hi(__flush_tlb_pending), %o0
832	or		%o0, %lo(__flush_tlb_pending), %o0
833	sethi		%hi(__hypervisor_flush_tlb_pending), %o1
834	or		%o1, %lo(__hypervisor_flush_tlb_pending), %o1
835	call		tlb_patch_one
836	 mov		16, %o2
837
838	sethi		%hi(__flush_tlb_kernel_range), %o0
839	or		%o0, %lo(__flush_tlb_kernel_range), %o0
840	sethi		%hi(__hypervisor_flush_tlb_kernel_range), %o1
841	or		%o1, %lo(__hypervisor_flush_tlb_kernel_range), %o1
842	call		tlb_patch_one
843	 mov		16, %o2
844
845#ifdef DCACHE_ALIASING_POSSIBLE
846	sethi		%hi(__flush_dcache_page), %o0
847	or		%o0, %lo(__flush_dcache_page), %o0
848	sethi		%hi(__hypervisor_flush_dcache_page), %o1
849	or		%o1, %lo(__hypervisor_flush_dcache_page), %o1
850	call		tlb_patch_one
851	 mov		2, %o2
852#endif /* DCACHE_ALIASING_POSSIBLE */
853
854#ifdef CONFIG_SMP
855	sethi		%hi(xcall_flush_tlb_mm), %o0
856	or		%o0, %lo(xcall_flush_tlb_mm), %o0
857	sethi		%hi(__hypervisor_xcall_flush_tlb_mm), %o1
858	or		%o1, %lo(__hypervisor_xcall_flush_tlb_mm), %o1
859	call		tlb_patch_one
860	 mov		21, %o2
861
862	sethi		%hi(xcall_flush_tlb_page), %o0
863	or		%o0, %lo(xcall_flush_tlb_page), %o0
864	sethi		%hi(__hypervisor_xcall_flush_tlb_page), %o1
865	or		%o1, %lo(__hypervisor_xcall_flush_tlb_page), %o1
866	call		tlb_patch_one
867	 mov		17, %o2
868
869	sethi		%hi(xcall_flush_tlb_kernel_range), %o0
870	or		%o0, %lo(xcall_flush_tlb_kernel_range), %o0
871	sethi		%hi(__hypervisor_xcall_flush_tlb_kernel_range), %o1
872	or		%o1, %lo(__hypervisor_xcall_flush_tlb_kernel_range), %o1
873	call		tlb_patch_one
874	 mov		25, %o2
875#endif /* CONFIG_SMP */
876
877	ret
878	 restore
879