xref: /titanic_52/usr/src/uts/sfmmu/ml/sfmmu_asm.s (revision 6c9596d46e3a733328712fdad3ea5ee362795acc)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28/*
29 * SFMMU primitives.  These primitives should only be used by sfmmu
30 * routines.
31 */
32
33#if defined(lint)
34#include <sys/types.h>
35#else	/* lint */
36#include "assym.h"
37#endif	/* lint */
38
39#include <sys/asm_linkage.h>
40#include <sys/machtrap.h>
41#include <sys/machasi.h>
42#include <sys/sun4asi.h>
43#include <sys/pte.h>
44#include <sys/mmu.h>
45#include <vm/hat_sfmmu.h>
46#include <vm/seg_spt.h>
47#include <sys/machparam.h>
48#include <sys/privregs.h>
49#include <sys/scb.h>
50#include <sys/intreg.h>
51#include <sys/machthread.h>
52#include <sys/intr.h>
53#include <sys/clock.h>
54#include <sys/trapstat.h>
55
56#ifdef TRAPTRACE
57#include <sys/traptrace.h>
58
59/*
60 * Tracing macro. Adds two instructions if TRAPTRACE is defined.
61 */
62#define	TT_TRACE(label)		\
63	ba	label		;\
64	rd	%pc, %g7
65#else
66
67#define	TT_TRACE(label)
68
69#endif /* TRAPTRACE */
70
71#ifndef	lint
72
73#if (TTE_SUSPEND_SHIFT > 0)
74#define	TTE_SUSPEND_INT_SHIFT(reg)				\
75	sllx	reg, TTE_SUSPEND_SHIFT, reg
76#else
77#define	TTE_SUSPEND_INT_SHIFT(reg)
78#endif
79
80#endif /* lint */
81
82#ifndef	lint
83
84/*
85 * Assumes TSBE_TAG is 0
86 * Assumes TSBE_INTHI is 0
87 * Assumes TSBREG.split is 0
88 */
89
90#if TSBE_TAG != 0
91#error "TSB_UPDATE and TSB_INVALIDATE assume TSBE_TAG = 0"
92#endif
93
94#if TSBTAG_INTHI != 0
95#error "TSB_UPDATE and TSB_INVALIDATE assume TSBTAG_INTHI = 0"
96#endif
97
98/*
99 * The following code assumes the tsb is not split.
100 *
101 * With TSBs no longer shared between processes, it's no longer
102 * necessary to hash the context bits into the tsb index to get
103 * tsb coloring; the new implementation treats the TSB as a
104 * direct-mapped, virtually-addressed cache.
105 *
106 * In:
107 *    vpshift = virtual page shift; e.g. 13 for 8K TTEs (constant or ro)
108 *    tsbbase = base address of TSB (clobbered)
109 *    tagacc = tag access register (clobbered)
110 *    szc = size code of TSB (ro)
111 *    tmp = scratch reg
112 * Out:
113 *    tsbbase = pointer to entry in TSB
114 */
115#define	GET_TSBE_POINTER(vpshift, tsbbase, tagacc, szc, tmp)		\
116	mov	TSB_ENTRIES(0), tmp	/* nentries in TSB size 0 */	;\
117	srlx	tagacc, vpshift, tagacc 				;\
118	sllx	tmp, szc, tmp		/* tmp = nentries in TSB */	;\
119	sub	tmp, 1, tmp		/* mask = nentries - 1 */	;\
120	and	tagacc, tmp, tmp	/* tsbent = virtpage & mask */	;\
121	sllx	tmp, TSB_ENTRY_SHIFT, tmp	/* entry num --> ptr */	;\
122	add	tsbbase, tmp, tsbbase	/* add entry offset to TSB base */
123
124/*
125 * When the kpm TSB is used it is assumed that it is direct mapped
126 * using (vaddr>>vpshift)%tsbsz as the index.
127 *
128 * Note that, for now, the kpm TSB and kernel TSB are the same for
129 * each mapping size.  However that need not always be the case.  If
130 * the trap handlers are updated to search a different TSB for kpm
131 * addresses than for kernel addresses then kpm_tsbbase and kpm_tsbsz
132 * (and/or kpmsm_tsbbase/kpmsm_tsbsz) may be entirely independent.
133 *
134 * In:
135 *    vpshift = virtual page shift; e.g. 13 for 8K TTEs (constant or ro)
136 *    vaddr = virtual address (clobbered)
137 *    tsbp, szc, tmp = scratch
138 * Out:
139 *    tsbp = pointer to entry in TSB
140 */
141#define	GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr, szc, tmp)		\
142	cmp	vpshift, MMU_PAGESHIFT					;\
143	bne,pn	%icc, 1f		/* branch if large case */	;\
144	  sethi	%hi(kpmsm_tsbsz), szc					;\
145	sethi	%hi(kpmsm_tsbbase), tsbp				;\
146	ld	[szc + %lo(kpmsm_tsbsz)], szc				;\
147	ldx	[tsbp + %lo(kpmsm_tsbbase)], tsbp			;\
148	ba,pt	%icc, 2f						;\
149	  nop								;\
1501:	sethi	%hi(kpm_tsbsz), szc					;\
151	sethi	%hi(kpm_tsbbase), tsbp					;\
152	ld	[szc + %lo(kpm_tsbsz)], szc				;\
153	ldx	[tsbp + %lo(kpm_tsbbase)], tsbp				;\
1542:	GET_TSBE_POINTER(vpshift, tsbp, vaddr, szc, tmp)
155
156/*
157 * Lock the TSBE at virtual address tsbep.
158 *
159 * tsbep = TSBE va (ro)
160 * tmp1, tmp2 = scratch registers (clobbered)
161 * label = label to use for branches (text)
162 * %asi = ASI to use for TSB access
163 *
164 * NOTE that we flush the TSB using fast VIS instructions that
165 * set all 1's in the TSB tag, so TSBTAG_LOCKED|TSBTAG_INVALID must
166 * not be treated as a locked entry or we'll get stuck spinning on
167 * an entry that isn't locked but really invalid.
168 */
169
170#if defined(UTSB_PHYS)
171
172#define	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			\
173	lda	[tsbep]ASI_MEM, tmp1					;\
174label:									;\
175	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
176	cmp	tmp1, tmp2 						;\
177	be,a,pn	%icc, label/**/b	/* if locked spin */		;\
178	  lda	[tsbep]ASI_MEM, tmp1					;\
179	casa	[tsbep]ASI_MEM, tmp1, tmp2				;\
180	cmp	tmp1, tmp2 						;\
181	bne,a,pn %icc, label/**/b	/* didn't lock so try again */	;\
182	  lda	[tsbep]ASI_MEM, tmp1					;\
183	/* tsbe lock acquired */					;\
184	membar #StoreStore
185
186#else /* UTSB_PHYS */
187
188#define	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			\
189	lda	[tsbep]%asi, tmp1					;\
190label:									;\
191	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
192	cmp	tmp1, tmp2 						;\
193	be,a,pn	%icc, label/**/b	/* if locked spin */		;\
194	  lda	[tsbep]%asi, tmp1					;\
195	casa	[tsbep]%asi, tmp1, tmp2					;\
196	cmp	tmp1, tmp2 						;\
197	bne,a,pn %icc, label/**/b	/* didn't lock so try again */	;\
198	  lda	[tsbep]%asi, tmp1					;\
199	/* tsbe lock acquired */					;\
200	membar #StoreStore
201
202#endif /* UTSB_PHYS */
203
204/*
205 * Atomically write TSBE at virtual address tsbep.
206 *
207 * tsbep = TSBE va (ro)
208 * tte = TSBE TTE (ro)
209 * tagtarget = TSBE tag (ro)
210 * %asi = ASI to use for TSB access
211 */
212
213#if defined(UTSB_PHYS)
214
215#define	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		\
216	add	tsbep, TSBE_TTE, tmp1					;\
217	stxa	tte, [tmp1]ASI_MEM		/* write tte data */	;\
218	membar #StoreStore						;\
219	add	tsbep, TSBE_TAG, tmp1					;\
220	stxa	tagtarget, [tmp1]ASI_MEM	/* write tte tag & unlock */
221
222#else /* UTSB_PHYS */
223
224#define	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget,tmp1)		\
225	stxa	tte, [tsbep + TSBE_TTE]%asi	/* write tte data */	;\
226	membar #StoreStore						;\
227	stxa	tagtarget, [tsbep + TSBE_TAG]%asi /* write tte tag & unlock */
228
229#endif /* UTSB_PHYS */
230
231/*
232 * Load an entry into the TSB at TL > 0.
233 *
234 * tsbep = pointer to the TSBE to load as va (ro)
235 * tte = value of the TTE retrieved and loaded (wo)
236 * tagtarget = tag target register.  To get TSBE tag to load,
237 *   we need to mask off the context and leave only the va (clobbered)
238 * ttepa = pointer to the TTE to retrieve/load as pa (ro)
239 * tmp1, tmp2 = scratch registers
240 * label = label to use for branches (text)
241 * %asi = ASI to use for TSB access
242 */
243
244#if defined(UTSB_PHYS)
245
246#define	TSB_UPDATE_TL(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
247	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
248	/*								;\
249	 * I don't need to update the TSB then check for the valid tte.	;\
250	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
251	 * we always invalidate the hash table before we unload the TSB.;\
252	 */								;\
253	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
254	ldxa	[ttepa]ASI_MEM, tte					;\
255	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
256	sethi	%hi(TSBTAG_INVALID), tmp2				;\
257	add	tsbep, TSBE_TAG, tmp1					;\
258	brgez,a,pn tte, label/**/f					;\
259	 sta	tmp2, [tmp1]ASI_MEM			/* unlock */	;\
260	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
261label:
262
263#else /* UTSB_PHYS */
264
265#define	TSB_UPDATE_TL(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
266	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
267	/*								;\
268	 * I don't need to update the TSB then check for the valid tte.	;\
269	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
270	 * we always invalidate the hash table before we unload the TSB.;\
271	 */								;\
272	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
273	ldxa	[ttepa]ASI_MEM, tte					;\
274	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
275	sethi	%hi(TSBTAG_INVALID), tmp2				;\
276	brgez,a,pn tte, label/**/f					;\
277	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
278	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
279label:
280
281#endif /* UTSB_PHYS */
282
283/*
284 * Load a 32M/256M Panther TSB entry into the TSB at TL > 0,
285 *   for ITLB synthesis.
286 *
287 * tsbep = pointer to the TSBE to load as va (ro)
288 * tte = 4M pfn offset (in), value of the TTE retrieved and loaded (out)
289 *   with exec_perm turned off and exec_synth turned on
290 * tagtarget = tag target register.  To get TSBE tag to load,
291 *   we need to mask off the context and leave only the va (clobbered)
292 * ttepa = pointer to the TTE to retrieve/load as pa (ro)
293 * tmp1, tmp2 = scratch registers
294 * label = label to use for branch (text)
295 * %asi = ASI to use for TSB access
296 */
297
298#define	TSB_UPDATE_TL_PN(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
299	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
300	/*								;\
301	 * I don't need to update the TSB then check for the valid tte.	;\
302	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
303	 * we always invalidate the hash table before we unload the TSB.;\
304	 * Or in 4M pfn offset to TTE and set the exec_perm bit to 0	;\
305	 * and exec_synth bit to 1.					;\
306	 */								;\
307	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
308	mov	tte, tmp1						;\
309	ldxa	[ttepa]ASI_MEM, tte					;\
310	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
311	sethi	%hi(TSBTAG_INVALID), tmp2				;\
312	brgez,a,pn tte, label/**/f					;\
313	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
314	or	tte, tmp1, tte						;\
315	andn	tte, TTE_EXECPRM_INT, tte				;\
316	or	tte, TTE_E_SYNTH_INT, tte				;\
317	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
318label:
319
320/*
321 * Build a 4M pfn offset for a Panther 32M/256M page, for ITLB synthesis.
322 *
323 * tte = value of the TTE, used to get tte_size bits (ro)
324 * tagaccess = tag access register, used to get 4M pfn bits (ro)
325 * pfn = 4M pfn bits shifted to offset for tte (out)
326 * tmp1 = scratch register
327 * label = label to use for branch (text)
328 */
329
330#define	GET_4M_PFN_OFF(tte, tagaccess, pfn, tmp, label)			\
331	/*								;\
332	 * Get 4M bits from tagaccess for 32M, 256M pagesizes.		;\
333	 * Return them, shifted, in pfn.				;\
334	 */								;\
335	srlx	tagaccess, MMU_PAGESHIFT4M, tagaccess			;\
336	srlx	tte, TTE_SZ_SHFT, tmp		/* isolate the */	;\
337	andcc	tmp, TTE_SZ_BITS, %g0		/* tte_size bits */	;\
338	bz,a,pt %icc, label/**/f		/* if 0, is */		;\
339	  and	tagaccess, 0x7, tagaccess	/* 32M page size */	;\
340	and	tagaccess, 0x3f, tagaccess /* else 256M page size */	;\
341label:									;\
342	sllx	tagaccess, MMU_PAGESHIFT4M, pfn
343
344/*
345 * Add 4M TTE size code to a tte for a Panther 32M/256M page,
346 * for ITLB synthesis.
347 *
348 * tte = value of the TTE, used to get tte_size bits (rw)
349 * tmp1 = scratch register
350 */
351
352#define	SET_TTE4M_PN(tte, tmp)						\
353	/*								;\
354	 * Set 4M pagesize tte bits. 					;\
355	 */								;\
356	set	TTE4M, tmp						;\
357	sllx	tmp, TTE_SZ_SHFT, tmp					;\
358	or	tte, tmp, tte
359
360/*
361 * Load an entry into the TSB at TL=0.
362 *
363 * tsbep = pointer to the TSBE to load as va (ro)
364 * tteva = pointer to the TTE to load as va (ro)
365 * tagtarget = TSBE tag to load (which contains no context), synthesized
366 * to match va of MMU tag target register only (ro)
367 * tmp1, tmp2 = scratch registers (clobbered)
368 * label = label to use for branches (text)
369 * %asi = ASI to use for TSB access
370 */
371
372#if defined(UTSB_PHYS)
373
374#define	TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label)		\
375	/* can't rd tteva after locking tsb because it can tlb miss */	;\
376	ldx	[tteva], tteva			/* load tte */		;\
377	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
378	sethi	%hi(TSBTAG_INVALID), tmp2				;\
379	add	tsbep, TSBE_TAG, tmp1					;\
380	brgez,a,pn tteva, label/**/f					;\
381	 sta	tmp2, [tmp1]ASI_MEM			/* unlock */	;\
382	TSB_INSERT_UNLOCK_ENTRY(tsbep, tteva, tagtarget, tmp1)		;\
383label:
384
385#else /* UTSB_PHYS */
386
387#define	TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label)		\
388	/* can't rd tteva after locking tsb because it can tlb miss */	;\
389	ldx	[tteva], tteva			/* load tte */		;\
390	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
391	sethi	%hi(TSBTAG_INVALID), tmp2				;\
392	brgez,a,pn tteva, label/**/f					;\
393	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
394	TSB_INSERT_UNLOCK_ENTRY(tsbep, tteva, tagtarget, tmp1)		;\
395label:
396
397#endif /* UTSB_PHYS */
398
399/*
400 * Invalidate a TSB entry in the TSB.
401 *
402 * NOTE: TSBE_TAG is assumed to be zero.  There is a compile time check
403 *	 about this earlier to ensure this is true.  Thus when we are
404 *	 directly referencing tsbep below, we are referencing the tte_tag
405 *	 field of the TSBE.  If this  offset ever changes, the code below
406 *	 will need to be modified.
407 *
408 * tsbep = pointer to TSBE as va (ro)
409 * tag = invalidation is done if this matches the TSBE tag (ro)
410 * tmp1 - tmp3 = scratch registers (clobbered)
411 * label = label name to use for branches (text)
412 * %asi = ASI to use for TSB access
413 */
414
415#if defined(UTSB_PHYS)
416
417#define	TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label)		\
418	lda	[tsbep]ASI_MEM, tmp1	/* tmp1 = tsbe tag */		;\
419	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
420label/**/1:								;\
421	cmp	tmp1, tmp2		/* see if tsbe is locked, if */	;\
422	be,a,pn	%icc, label/**/1	/* so, loop until unlocked */	;\
423	  lda	[tsbep]ASI_MEM, tmp1	/* reloading value each time */	;\
424	ldxa	[tsbep]ASI_MEM, tmp3	/* tmp3 = tsbe tag */		;\
425	cmp	tag, tmp3		/* compare tags */		;\
426	bne,pt	%xcc, label/**/2	/* if different, do nothing */	;\
427	  sethi	%hi(TSBTAG_INVALID), tmp3				;\
428	casa	[tsbep]ASI_MEM, tmp1, tmp3 /* try to set tag invalid */	;\
429	cmp	tmp1, tmp3		/* if not successful */		;\
430	bne,a,pn %icc, label/**/1	/* start over from the top */	;\
431	  lda	[tsbep]ASI_MEM, tmp1	/* reloading tsbe tag */	;\
432label/**/2:
433
434#else /* UTSB_PHYS */
435
436#define	TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label)		\
437	lda	[tsbep]%asi, tmp1	/* tmp1 = tsbe tag */		;\
438	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
439label/**/1:								;\
440	cmp	tmp1, tmp2		/* see if tsbe is locked, if */	;\
441	be,a,pn	%icc, label/**/1	/* so, loop until unlocked */	;\
442	  lda	[tsbep]%asi, tmp1	/* reloading value each time */	;\
443	ldxa	[tsbep]%asi, tmp3	/* tmp3 = tsbe tag */		;\
444	cmp	tag, tmp3		/* compare tags */		;\
445	bne,pt	%xcc, label/**/2	/* if different, do nothing */	;\
446	  sethi	%hi(TSBTAG_INVALID), tmp3				;\
447	casa	[tsbep]%asi, tmp1, tmp3	/* try to set tag invalid */	;\
448	cmp	tmp1, tmp3		/* if not successful */		;\
449	bne,a,pn %icc, label/**/1	/* start over from the top */	;\
450	  lda	[tsbep]%asi, tmp1	/* reloading tsbe tag */	;\
451label/**/2:
452
453#endif /* UTSB_PHYS */
454
455#if TSB_SOFTSZ_MASK < TSB_SZ_MASK
456#error	- TSB_SOFTSZ_MASK too small
457#endif
458
459
460/*
461 * An implementation of setx which will be hot patched at run time.
462 * since it is being hot patched, there is no value passed in.
463 * Thus, essentially we are implementing
464 *	setx value, tmp, dest
465 * where value is RUNTIME_PATCH (aka 0) in this case.
466 */
467#define	RUNTIME_PATCH_SETX(dest, tmp)					\
468	sethi	%hh(RUNTIME_PATCH), tmp					;\
469	sethi	%lm(RUNTIME_PATCH), dest				;\
470	or	tmp, %hm(RUNTIME_PATCH), tmp				;\
471	or	dest, %lo(RUNTIME_PATCH), dest				;\
472	sllx	tmp, 32, tmp						;\
473	nop				/* for perf reasons */		;\
474	or	tmp, dest, dest		/* contents of patched value */
475
476
477#endif (lint)
478
479
480#if defined (lint)
481
482/*
483 * sfmmu related subroutines
484 */
485uint_t
486sfmmu_disable_intrs()
487{ return(0); }
488
489/* ARGSUSED */
490void
491sfmmu_enable_intrs(uint_t pstate_save)
492{}
493
494/* ARGSUSED */
495int
496sfmmu_alloc_ctx(sfmmu_t *sfmmup, int allocflag, struct cpu *cp, int shflag)
497{ return(0); }
498
499/*
500 * Use cas, if tte has changed underneath us then reread and try again.
501 * In the case of a retry, it will update sttep with the new original.
502 */
503/* ARGSUSED */
504int
505sfmmu_modifytte(tte_t *sttep, tte_t *stmodttep, tte_t *dttep)
506{ return(0); }
507
508/*
509 * Use cas, if tte has changed underneath us then return 1, else return 0
510 */
511/* ARGSUSED */
512int
513sfmmu_modifytte_try(tte_t *sttep, tte_t *stmodttep, tte_t *dttep)
514{ return(0); }
515
516/* ARGSUSED */
517void
518sfmmu_copytte(tte_t *sttep, tte_t *dttep)
519{}
520
521/*ARGSUSED*/
522struct tsbe *
523sfmmu_get_tsbe(uint64_t tsbeptr, caddr_t vaddr, int vpshift, int tsb_szc)
524{ return(0); }
525
526/*ARGSUSED*/
527uint64_t
528sfmmu_make_tsbtag(caddr_t va)
529{ return(0); }
530
531#else	/* lint */
532
533	.seg	".data"
534	.global	sfmmu_panic1
535sfmmu_panic1:
536	.asciz	"sfmmu_asm: interrupts already disabled"
537
538	.global	sfmmu_panic3
539sfmmu_panic3:
540	.asciz	"sfmmu_asm: sfmmu_vatopfn called for user"
541
542	.global	sfmmu_panic4
543sfmmu_panic4:
544	.asciz	"sfmmu_asm: 4M tsb pointer mis-match"
545
546	.global	sfmmu_panic5
547sfmmu_panic5:
548	.asciz	"sfmmu_asm: no unlocked TTEs in TLB 0"
549
550	.global	sfmmu_panic6
551sfmmu_panic6:
552	.asciz	"sfmmu_asm: interrupts not disabled"
553
554	.global	sfmmu_panic7
555sfmmu_panic7:
556	.asciz	"sfmmu_asm: kernel as"
557
558	.global	sfmmu_panic8
559sfmmu_panic8:
560	.asciz	"sfmmu_asm: gnum is zero"
561
562	.global	sfmmu_panic9
563sfmmu_panic9:
564	.asciz	"sfmmu_asm: cnum is greater than MAX_SFMMU_CTX_VAL"
565
566	.global	sfmmu_panic10
567sfmmu_panic10:
568	.asciz	"sfmmu_asm: valid SCD with no 3rd scd TSB"
569
570        ENTRY(sfmmu_disable_intrs)
571        rdpr    %pstate, %o0
572#ifdef DEBUG
573	PANIC_IF_INTR_DISABLED_PSTR(%o0, sfmmu_di_l0, %g1)
574#endif /* DEBUG */
575        retl
576          wrpr   %o0, PSTATE_IE, %pstate
577        SET_SIZE(sfmmu_disable_intrs)
578
579	ENTRY(sfmmu_enable_intrs)
580        retl
581          wrpr    %g0, %o0, %pstate
582        SET_SIZE(sfmmu_enable_intrs)
583
584/*
585 * This routine is called both by resume() and sfmmu_get_ctx() to
586 * allocate a new context for the process on a MMU.
587 * if allocflag == 1, then alloc ctx when HAT mmu cnum == INVALID .
588 * if allocflag == 0, then do not alloc ctx if HAT mmu cnum == INVALID, which
589 * is the case when sfmmu_alloc_ctx is called from resume().
590 *
591 * The caller must disable interrupts before entering this routine.
592 * To reduce ctx switch overhead, the code contains both 'fast path' and
593 * 'slow path' code. The fast path code covers the common case where only
594 * a quick check is needed and the real ctx allocation is not required.
595 * It can be done without holding the per-process (PP) lock.
596 * The 'slow path' code must be protected by the PP Lock and performs ctx
597 * allocation.
598 * Hardware context register and HAT mmu cnum are updated accordingly.
599 *
600 * %o0 - sfmmup
601 * %o1 - allocflag
602 * %o2 - CPU
603 * %o3 - sfmmu private/shared flag
604 *
605 * ret - 0: no ctx is allocated
606 *       1: a ctx is allocated
607 */
608        ENTRY_NP(sfmmu_alloc_ctx)
609
610#ifdef DEBUG
611	sethi   %hi(ksfmmup), %g1
612	ldx     [%g1 + %lo(ksfmmup)], %g1
613	cmp     %g1, %o0
614	bne,pt   %xcc, 0f
615	  nop
616
617	sethi   %hi(panicstr), %g1		! if kernel as, panic
618        ldx     [%g1 + %lo(panicstr)], %g1
619        tst     %g1
620        bnz,pn  %icc, 7f
621          nop
622
623	sethi	%hi(sfmmu_panic7), %o0
624	call	panic
625	  or	%o0, %lo(sfmmu_panic7), %o0
626
6277:
628	retl
629	  mov	%g0, %o0			! %o0 = ret = 0
630
6310:
632	PANIC_IF_INTR_ENABLED_PSTR(sfmmu_ei_l1, %g1)
633#endif /* DEBUG */
634
635	mov	%o3, %g1			! save sfmmu pri/sh flag in %g1
636
637	! load global mmu_ctxp info
638	ldx	[%o2 + CPU_MMU_CTXP], %o3		! %o3 = mmu_ctx_t ptr
639        lduw	[%o2 + CPU_MMU_IDX], %g2		! %g2 = mmu index
640
641	! load global mmu_ctxp gnum
642	ldx	[%o3 + MMU_CTX_GNUM], %o4		! %o4 = mmu_ctxp->gnum
643
644#ifdef DEBUG
645	cmp	%o4, %g0		! mmu_ctxp->gnum should never be 0
646	bne,pt	%xcc, 3f
647	  nop
648
649	sethi   %hi(panicstr), %g1	! test if panicstr is already set
650        ldx     [%g1 + %lo(panicstr)], %g1
651        tst     %g1
652        bnz,pn  %icc, 1f
653          nop
654
655	sethi	%hi(sfmmu_panic8), %o0
656	call	panic
657	  or	%o0, %lo(sfmmu_panic8), %o0
6581:
659	retl
660	  mov	%g0, %o0			! %o0 = ret = 0
6613:
662#endif
663
664	! load HAT sfmmu_ctxs[mmuid] gnum, cnum
665
666	sllx	%g2, SFMMU_MMU_CTX_SHIFT, %g2
667	add	%o0, %g2, %g2		! %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
668
669	/*
670	 * %g5 = sfmmu gnum returned
671	 * %g6 = sfmmu cnum returned
672	 * %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
673	 * %g4 = scratch
674	 *
675	 * Fast path code, do a quick check.
676	 */
677	SFMMU_MMUID_GNUM_CNUM(%g2, %g5, %g6, %g4)
678
679	cmp	%g6, INVALID_CONTEXT		! hat cnum == INVALID ??
680	bne,pt	%icc, 1f			! valid hat cnum, check gnum
681	  nop
682
683	! cnum == INVALID, check allocflag
684	mov	%g0, %g4	! %g4 = ret = 0
685	brz,pt  %o1, 8f		! allocflag == 0, skip ctx allocation, bail
686	  mov	%g6, %o1
687
688	! (invalid HAT cnum) && (allocflag == 1)
689	ba,pt	%icc, 2f
690	  nop
6911:
692	! valid HAT cnum, check gnum
693	cmp	%g5, %o4
694	mov	1, %g4				!%g4 = ret = 1
695	be,a,pt	%icc, 8f			! gnum unchanged, go to done
696	  mov	%g6, %o1
697
6982:
699	/*
700	 * Grab per process (PP) sfmmu_ctx_lock spinlock,
701	 * followed by the 'slow path' code.
702	 */
703	ldstub	[%o0 + SFMMU_CTX_LOCK], %g3	! %g3 = per process (PP) lock
7043:
705	brz	%g3, 5f
706	  nop
7074:
708	brnz,a,pt       %g3, 4b				! spin if lock is 1
709	  ldub	[%o0 + SFMMU_CTX_LOCK], %g3
710	ba	%xcc, 3b				! retry the lock
711	  ldstub	[%o0 + SFMMU_CTX_LOCK], %g3    ! %g3 = PP lock
712
7135:
714	membar  #LoadLoad
715	/*
716	 * %g5 = sfmmu gnum returned
717	 * %g6 = sfmmu cnum returned
718	 * %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
719	 * %g4 = scratch
720	 */
721	SFMMU_MMUID_GNUM_CNUM(%g2, %g5, %g6, %g4)
722
723	cmp	%g6, INVALID_CONTEXT		! hat cnum == INVALID ??
724	bne,pt	%icc, 1f			! valid hat cnum, check gnum
725	  nop
726
727	! cnum == INVALID, check allocflag
728	mov	%g0, %g4	! %g4 = ret = 0
729	brz,pt	%o1, 2f		! allocflag == 0, called from resume, set hw
730	  mov	%g6, %o1
731
732	! (invalid HAT cnum) && (allocflag == 1)
733	ba,pt	%icc, 6f
734	  nop
7351:
736	! valid HAT cnum, check gnum
737	cmp	%g5, %o4
738	mov	1, %g4				! %g4 = ret  = 1
739	be,a,pt	%icc, 2f			! gnum unchanged, go to done
740	  mov	%g6, %o1
741
742	ba,pt	%icc, 6f
743	  nop
7442:
745	membar  #LoadStore|#StoreStore
746	ba,pt %icc, 8f
747	  clrb  [%o0 + SFMMU_CTX_LOCK]
7486:
749	/*
750	 * We get here if we do not have a valid context, or
751	 * the HAT gnum does not match global gnum. We hold
752	 * sfmmu_ctx_lock spinlock. Allocate that context.
753	 *
754	 * %o3 = mmu_ctxp
755	 */
756	add	%o3, MMU_CTX_CNUM, %g3
757	ld	[%o3 + MMU_CTX_NCTXS], %g4
758
759	/*
760         * %g2 = &sfmmu_ctx_t[mmuid] - SFMMU_CTXS;
761         * %g3 = mmu cnum address
762	 * %g4 = mmu nctxs
763	 *
764	 * %o0 = sfmmup
765	 * %o1 = mmu current cnum value (used as new cnum)
766	 * %o4 = mmu gnum
767	 *
768	 * %o5 = scratch
769	 */
770	ld	[%g3], %o1
7710:
772	cmp	%o1, %g4
773	bl,a,pt %icc, 1f
774	  add	%o1, 1, %o5		! %o5 = mmu_ctxp->cnum + 1
775
776	/*
777	 * cnum reachs max, bail, so wrap around can be performed later.
778	 */
779	set	INVALID_CONTEXT, %o1
780	/*
781	 * When the routine is called by shared ctx, we want to set
782	 * both private and shared ctx regs to INVALID. In order to
783	 * do so, we set the sfmmu priv/shared flag to 'private' regardless.
784	 * so that private ctx reg will be set to invalid.
785	 * Note that values written to private context register are
786	 * automatically written to shared context register as well.
787	 */
788	mov	%g0, %g1		! %g1 = sfmmu private/shared flag
789	mov	%g0, %g4		! %g4 = ret = 0
790
791	membar  #LoadStore|#StoreStore
792	ba,pt	%icc, 8f
793	  clrb	[%o0 + SFMMU_CTX_LOCK]
7941:
795	! %g3 = addr of mmu_ctxp->cnum
796	! %o5 = mmu_ctxp->cnum + 1
797	cas	[%g3], %o1, %o5
798	cmp	%o1, %o5
799	bne,a,pn %xcc, 0b	! cas failed
800	  ld	[%g3], %o1
801
802#ifdef DEBUG
803        set	MAX_SFMMU_CTX_VAL, %o5
804	cmp	%o1, %o5
805	ble,pt %icc, 2f
806	  nop
807
808	sethi	%hi(sfmmu_panic9), %o0
809	call	panic
810	  or	%o0, %lo(sfmmu_panic9), %o0
8112:
812#endif
813	! update hat gnum and cnum
814	sllx	%o4, SFMMU_MMU_GNUM_RSHIFT, %o4
815	or	%o4, %o1, %o4
816	stx	%o4, [%g2 + SFMMU_CTXS]
817
818	membar  #LoadStore|#StoreStore
819	clrb	[%o0 + SFMMU_CTX_LOCK]
820
821	mov	1, %g4			! %g4 = ret = 1
8228:
823	/*
824	 * program the secondary context register
825	 *
826	 * %o1 = cnum
827	 * %g1 = sfmmu private/shared flag (0:private,  1:shared)
828	 */
829
830#ifdef	sun4u
831	ldub	[%o0 + SFMMU_CEXT], %o2
832	sll	%o2, CTXREG_EXT_SHIFT, %o2
833	or	%o1, %o2, %o1
834#endif
835	SET_SECCTX(%o1, %g1, %o4, %o5)
836
837	retl
838	  mov	%g4, %o0			! %o0 = ret
839
840	SET_SIZE(sfmmu_alloc_ctx)
841
842	ENTRY_NP(sfmmu_modifytte)
843	ldx	[%o2], %g3			/* current */
844	ldx	[%o0], %g1			/* original */
8452:
846	ldx	[%o1], %g2			/* modified */
847	cmp	%g2, %g3			/* is modified = current? */
848	be,a,pt	%xcc,1f				/* yes, don't write */
849	stx	%g3, [%o0]			/* update new original */
850	casx	[%o2], %g1, %g2
851	cmp	%g1, %g2
852	be,pt	%xcc, 1f			/* cas succeeded - return */
853	  nop
854	ldx	[%o2], %g3			/* new current */
855	stx	%g3, [%o0]			/* save as new original */
856	ba,pt	%xcc, 2b
857	  mov	%g3, %g1
8581:	retl
859	membar	#StoreLoad
860	SET_SIZE(sfmmu_modifytte)
861
862	ENTRY_NP(sfmmu_modifytte_try)
863	ldx	[%o1], %g2			/* modified */
864	ldx	[%o2], %g3			/* current */
865	ldx	[%o0], %g1			/* original */
866	cmp	%g3, %g2			/* is modified = current? */
867	be,a,pn %xcc,1f				/* yes, don't write */
868	mov	0, %o1				/* as if cas failed. */
869
870	casx	[%o2], %g1, %g2
871	membar	#StoreLoad
872	cmp	%g1, %g2
873	movne	%xcc, -1, %o1			/* cas failed. */
874	move	%xcc, 1, %o1			/* cas succeeded. */
8751:
876	stx	%g2, [%o0]			/* report "current" value */
877	retl
878	mov	%o1, %o0
879	SET_SIZE(sfmmu_modifytte_try)
880
881	ENTRY_NP(sfmmu_copytte)
882	ldx	[%o0], %g1
883	retl
884	stx	%g1, [%o1]
885	SET_SIZE(sfmmu_copytte)
886
887
888	/*
889	 * Calculate a TSB entry pointer for the given TSB, va, pagesize.
890	 * %o0 = TSB base address (in), pointer to TSB entry (out)
891	 * %o1 = vaddr (in)
892	 * %o2 = vpshift (in)
893	 * %o3 = tsb size code (in)
894	 * %o4 = scratch register
895	 */
896	ENTRY_NP(sfmmu_get_tsbe)
897	GET_TSBE_POINTER(%o2, %o0, %o1, %o3, %o4)
898	retl
899	nop
900	SET_SIZE(sfmmu_get_tsbe)
901
902	/*
903	 * Return a TSB tag for the given va.
904	 * %o0 = va (in/clobbered)
905	 * %o0 = va shifted to be in tsb tag format (with no context) (out)
906	 */
907	ENTRY_NP(sfmmu_make_tsbtag)
908	retl
909	srln	%o0, TTARGET_VA_SHIFT, %o0
910	SET_SIZE(sfmmu_make_tsbtag)
911
912#endif /* lint */
913
914/*
915 * Other sfmmu primitives
916 */
917
918
919#if defined (lint)
920void
921sfmmu_patch_ktsb(void)
922{
923}
924
925void
926sfmmu_kpm_patch_tlbm(void)
927{
928}
929
930void
931sfmmu_kpm_patch_tsbm(void)
932{
933}
934
935void
936sfmmu_patch_shctx(void)
937{
938}
939
940/* ARGSUSED */
941void
942sfmmu_load_tsbe(struct tsbe *tsbep, uint64_t vaddr, tte_t *ttep, int phys)
943{
944}
945
946/* ARGSUSED */
947void
948sfmmu_unload_tsbe(struct tsbe *tsbep, uint64_t vaddr, int phys)
949{
950}
951
952/* ARGSUSED */
953void
954sfmmu_kpm_load_tsb(caddr_t addr, tte_t *ttep, int vpshift)
955{
956}
957
958/* ARGSUSED */
959void
960sfmmu_kpm_unload_tsb(caddr_t addr, int vpshift)
961{
962}
963
964#else /* lint */
965
966#define	I_SIZE		4
967
968	ENTRY_NP(sfmmu_fix_ktlb_traptable)
969	/*
970	 * %o0 = start of patch area
971	 * %o1 = size code of TSB to patch
972	 * %o3 = scratch
973	 */
974	/* fix sll */
975	ld	[%o0], %o3			/* get sll */
976	sub	%o3, %o1, %o3			/* decrease shift by tsb szc */
977	st	%o3, [%o0]			/* write sll */
978	flush	%o0
979	/* fix srl */
980	add	%o0, I_SIZE, %o0		/* goto next instr. */
981	ld	[%o0], %o3			/* get srl */
982	sub	%o3, %o1, %o3			/* decrease shift by tsb szc */
983	st	%o3, [%o0]			/* write srl */
984	retl
985	flush	%o0
986	SET_SIZE(sfmmu_fix_ktlb_traptable)
987
988	ENTRY_NP(sfmmu_fixup_ktsbbase)
989	/*
990	 * %o0 = start of patch area
991	 * %o5 = kernel virtual or physical tsb base address
992	 * %o2, %o3 are used as scratch registers.
993	 */
994	/* fixup sethi instruction */
995	ld	[%o0], %o3
996	srl	%o5, 10, %o2			! offset is bits 32:10
997	or	%o3, %o2, %o3			! set imm22
998	st	%o3, [%o0]
999	/* fixup offset of lduw/ldx */
1000	add	%o0, I_SIZE, %o0		! next instr
1001	ld	[%o0], %o3
1002	and	%o5, 0x3ff, %o2			! set imm13 to bits 9:0
1003	or	%o3, %o2, %o3
1004	st	%o3, [%o0]
1005	retl
1006	flush	%o0
1007	SET_SIZE(sfmmu_fixup_ktsbbase)
1008
1009	ENTRY_NP(sfmmu_fixup_setx)
1010	/*
1011	 * %o0 = start of patch area
1012	 * %o4 = 64 bit value to patch
1013	 * %o2, %o3 are used as scratch registers.
1014	 *
1015	 * Note: Assuming that all parts of the instructions which need to be
1016	 *	 patched correspond to RUNTIME_PATCH (aka 0)
1017	 *
1018	 * Note the implementation of setx which is being patched is as follows:
1019	 *
1020	 * sethi   %hh(RUNTIME_PATCH), tmp
1021	 * sethi   %lm(RUNTIME_PATCH), dest
1022	 * or      tmp, %hm(RUNTIME_PATCH), tmp
1023	 * or      dest, %lo(RUNTIME_PATCH), dest
1024	 * sllx    tmp, 32, tmp
1025	 * nop
1026	 * or      tmp, dest, dest
1027	 *
1028	 * which differs from the implementation in the
1029	 * "SPARC Architecture Manual"
1030	 */
1031	/* fixup sethi instruction */
1032	ld	[%o0], %o3
1033	srlx	%o4, 42, %o2			! bits [63:42]
1034	or	%o3, %o2, %o3			! set imm22
1035	st	%o3, [%o0]
1036	/* fixup sethi instruction */
1037	add	%o0, I_SIZE, %o0		! next instr
1038	ld	[%o0], %o3
1039	sllx	%o4, 32, %o2			! clear upper bits
1040	srlx	%o2, 42, %o2			! bits [31:10]
1041	or	%o3, %o2, %o3			! set imm22
1042	st	%o3, [%o0]
1043	/* fixup or instruction */
1044	add	%o0, I_SIZE, %o0		! next instr
1045	ld	[%o0], %o3
1046	srlx	%o4, 32, %o2			! bits [63:32]
1047	and	%o2, 0x3ff, %o2			! bits [41:32]
1048	or	%o3, %o2, %o3			! set imm
1049	st	%o3, [%o0]
1050	/* fixup or instruction */
1051	add	%o0, I_SIZE, %o0		! next instr
1052	ld	[%o0], %o3
1053	and	%o4, 0x3ff, %o2			! bits [9:0]
1054	or	%o3, %o2, %o3			! set imm
1055	st	%o3, [%o0]
1056	retl
1057	flush	%o0
1058	SET_SIZE(sfmmu_fixup_setx)
1059
1060	ENTRY_NP(sfmmu_fixup_or)
1061	/*
1062	 * %o0 = start of patch area
1063	 * %o4 = 32 bit value to patch
1064	 * %o2, %o3 are used as scratch registers.
1065	 * Note: Assuming that all parts of the instructions which need to be
1066	 *	 patched correspond to RUNTIME_PATCH (aka 0)
1067	 */
1068	ld	[%o0], %o3
1069	and	%o4, 0x3ff, %o2			! bits [9:0]
1070	or	%o3, %o2, %o3			! set imm
1071	st	%o3, [%o0]
1072	retl
1073	flush	%o0
1074	SET_SIZE(sfmmu_fixup_or)
1075
1076	ENTRY_NP(sfmmu_fixup_shiftx)
1077	/*
1078	 * %o0 = start of patch area
1079	 * %o4 = signed int immediate value to add to sllx/srlx imm field
1080	 * %o2, %o3 are used as scratch registers.
1081	 *
1082	 * sllx/srlx store the 6 bit immediate value in the lowest order bits
1083	 * so we do a simple add.  The caller must be careful to prevent
1084	 * overflow, which could easily occur if the initial value is nonzero!
1085	 */
1086	ld	[%o0], %o3			! %o3 = instruction to patch
1087	and	%o3, 0x3f, %o2			! %o2 = existing imm value
1088	add	%o2, %o4, %o2			! %o2 = new imm value
1089	andn	%o3, 0x3f, %o3			! clear old imm value
1090	and	%o2, 0x3f, %o2			! truncate new imm value
1091	or	%o3, %o2, %o3			! set new imm value
1092	st	%o3, [%o0]			! store updated instruction
1093	retl
1094	flush	%o0
1095	SET_SIZE(sfmmu_fixup_shiftx)
1096
1097	ENTRY_NP(sfmmu_fixup_mmu_asi)
1098	/*
1099	 * Patch imm_asi of all ldda instructions in the MMU
1100	 * trap handlers.  We search MMU_PATCH_INSTR instructions
1101	 * starting from the itlb miss handler (trap 0x64).
1102	 * %o0 = address of tt[0,1]_itlbmiss
1103	 * %o1 = imm_asi to setup, shifted by appropriate offset.
1104	 * %o3 = number of instructions to search
1105	 * %o4 = reserved by caller: called from leaf routine
1106	 */
11071:	ldsw	[%o0], %o2			! load instruction to %o2
1108	brgez,pt %o2, 2f
1109	  srl	%o2, 30, %o5
1110	btst	1, %o5				! test bit 30; skip if not set
1111	bz,pt	%icc, 2f
1112	  sllx	%o2, 39, %o5			! bit 24 -> bit 63
1113	srlx	%o5, 58, %o5			! isolate op3 part of opcode
1114	xor	%o5, 0x13, %o5			! 01 0011 binary == ldda
1115	brnz,pt	%o5, 2f				! skip if not a match
1116	  or	%o2, %o1, %o2			! or in imm_asi
1117	st	%o2, [%o0]			! write patched instruction
11182:	dec	%o3
1119	brnz,a,pt %o3, 1b			! loop until we're done
1120	  add	%o0, I_SIZE, %o0
1121	retl
1122	flush	%o0
1123	SET_SIZE(sfmmu_fixup_mmu_asi)
1124
1125	/*
1126	 * Patch immediate ASI used to access the TSB in the
1127	 * trap table.
1128	 * inputs: %o0 = value of ktsb_phys
1129	 */
1130	ENTRY_NP(sfmmu_patch_mmu_asi)
1131	mov	%o7, %o4			! save return pc in %o4
1132	movrnz	%o0, ASI_QUAD_LDD_PHYS, %o3
1133	movrz	%o0, ASI_NQUAD_LD, %o3
1134	sll	%o3, 5, %o1			! imm_asi offset
1135	mov	6, %o3				! number of instructions
1136	sethi	%hi(dktsb), %o0			! to search
1137	call	sfmmu_fixup_mmu_asi		! patch kdtlb miss
1138	  or	%o0, %lo(dktsb), %o0
1139	mov	6, %o3				! number of instructions
1140	sethi	%hi(dktsb4m), %o0		! to search
1141	call	sfmmu_fixup_mmu_asi		! patch kdtlb4m miss
1142	  or	%o0, %lo(dktsb4m), %o0
1143	mov	6, %o3				! number of instructions
1144	sethi	%hi(iktsb), %o0			! to search
1145	call	sfmmu_fixup_mmu_asi		! patch kitlb miss
1146	  or	%o0, %lo(iktsb), %o0
1147	mov	6, %o3				! number of instructions
1148	sethi	%hi(iktsb4m), %o0		! to search
1149	call	sfmmu_fixup_mmu_asi		! patch kitlb4m miss
1150	  or	%o0, %lo(iktsb4m), %o0
1151	mov	%o4, %o7			! retore return pc -- leaf
1152	retl
1153	nop
1154	SET_SIZE(sfmmu_patch_mmu_asi)
1155
1156	ENTRY_NP(sfmmu_patch_ktsb)
1157	/*
1158	 * We need to fix iktsb, dktsb, et. al.
1159	 */
1160	save	%sp, -SA(MINFRAME), %sp
1161	set	ktsb_phys, %o1
1162	ld	[%o1], %o4
1163	set	ktsb_base, %o5
1164	set	ktsb4m_base, %l1
1165	brz,pt	%o4, 1f
1166	  nop
1167	set	ktsb_pbase, %o5
1168	set	ktsb4m_pbase, %l1
11691:
1170	sethi	%hi(ktsb_szcode), %o1
1171	ld	[%o1 + %lo(ktsb_szcode)], %o1	/* %o1 = ktsb size code */
1172
1173	sethi	%hi(iktsb), %o0
1174	call	sfmmu_fix_ktlb_traptable
1175	  or	%o0, %lo(iktsb), %o0
1176
1177	sethi	%hi(dktsb), %o0
1178	call	sfmmu_fix_ktlb_traptable
1179	  or	%o0, %lo(dktsb), %o0
1180
1181	sethi	%hi(ktsb4m_szcode), %o1
1182	ld	[%o1 + %lo(ktsb4m_szcode)], %o1	/* %o1 = ktsb4m size code */
1183
1184	sethi	%hi(iktsb4m), %o0
1185	call	sfmmu_fix_ktlb_traptable
1186	  or	%o0, %lo(iktsb4m), %o0
1187
1188	sethi	%hi(dktsb4m), %o0
1189	call	sfmmu_fix_ktlb_traptable
1190	  or	%o0, %lo(dktsb4m), %o0
1191
1192#ifndef sun4v
1193	mov	ASI_N, %o2
1194	movrnz	%o4, ASI_MEM, %o2	! setup kernel 32bit ASI to patch
1195	mov	%o2, %o4		! sfmmu_fixup_or needs this in %o4
1196	sethi	%hi(tsb_kernel_patch_asi), %o0
1197	call	sfmmu_fixup_or
1198	  or	%o0, %lo(tsb_kernel_patch_asi), %o0
1199#endif
1200
1201	ldx 	[%o5], %o4		! load ktsb base addr (VA or PA)
1202
1203	sethi	%hi(dktsbbase), %o0
1204	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1205	  or	%o0, %lo(dktsbbase), %o0
1206
1207	sethi	%hi(iktsbbase), %o0
1208	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1209	  or	%o0, %lo(iktsbbase), %o0
1210
1211	sethi	%hi(sfmmu_kprot_patch_ktsb_base), %o0
1212	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1213	  or	%o0, %lo(sfmmu_kprot_patch_ktsb_base), %o0
1214
1215#ifdef sun4v
1216	sethi	%hi(sfmmu_dslow_patch_ktsb_base), %o0
1217	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1218	  or	%o0, %lo(sfmmu_dslow_patch_ktsb_base), %o0
1219#endif /* sun4v */
1220
1221	ldx 	[%l1], %o4		! load ktsb4m base addr (VA or PA)
1222
1223	sethi	%hi(dktsb4mbase), %o0
1224	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1225	  or	%o0, %lo(dktsb4mbase), %o0
1226
1227	sethi	%hi(iktsb4mbase), %o0
1228	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1229	  or	%o0, %lo(iktsb4mbase), %o0
1230
1231	sethi	%hi(sfmmu_kprot_patch_ktsb4m_base), %o0
1232	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1233	  or	%o0, %lo(sfmmu_kprot_patch_ktsb4m_base), %o0
1234
1235#ifdef sun4v
1236	sethi	%hi(sfmmu_dslow_patch_ktsb4m_base), %o0
1237	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1238	  or	%o0, %lo(sfmmu_dslow_patch_ktsb4m_base), %o0
1239#endif /* sun4v */
1240
1241	set	ktsb_szcode, %o4
1242	ld	[%o4], %o4
1243	sethi	%hi(sfmmu_kprot_patch_ktsb_szcode), %o0
1244	call	sfmmu_fixup_or		! patch value of ktsb_szcode
1245	  or	%o0, %lo(sfmmu_kprot_patch_ktsb_szcode), %o0
1246
1247#ifdef sun4v
1248	sethi	%hi(sfmmu_dslow_patch_ktsb_szcode), %o0
1249	call	sfmmu_fixup_or		! patch value of ktsb_szcode
1250	  or	%o0, %lo(sfmmu_dslow_patch_ktsb_szcode), %o0
1251#endif /* sun4v */
1252
1253	set	ktsb4m_szcode, %o4
1254	ld	[%o4], %o4
1255	sethi	%hi(sfmmu_kprot_patch_ktsb4m_szcode), %o0
1256	call	sfmmu_fixup_or		! patch value of ktsb4m_szcode
1257	  or	%o0, %lo(sfmmu_kprot_patch_ktsb4m_szcode), %o0
1258
1259#ifdef sun4v
1260	sethi	%hi(sfmmu_dslow_patch_ktsb4m_szcode), %o0
1261	call	sfmmu_fixup_or		! patch value of ktsb4m_szcode
1262	  or	%o0, %lo(sfmmu_dslow_patch_ktsb4m_szcode), %o0
1263#endif /* sun4v */
1264
1265	ret
1266	restore
1267	SET_SIZE(sfmmu_patch_ktsb)
1268
1269	ENTRY_NP(sfmmu_kpm_patch_tlbm)
1270	/*
1271	 * Fixup trap handlers in common segkpm case.  This is reserved
1272	 * for future use should kpm TSB be changed to be other than the
1273	 * kernel TSB.
1274	 */
1275	retl
1276	nop
1277	SET_SIZE(sfmmu_kpm_patch_tlbm)
1278
1279	ENTRY_NP(sfmmu_kpm_patch_tsbm)
1280	/*
1281	 * nop the branch to sfmmu_kpm_dtsb_miss_small
1282	 * in the case where we are using large pages for
1283	 * seg_kpm (and hence must probe the second TSB for
1284	 * seg_kpm VAs)
1285	 */
1286	set	dktsb4m_kpmcheck_small, %o0
1287	MAKE_NOP_INSTR(%o1)
1288	st	%o1, [%o0]
1289	flush	%o0
1290	retl
1291	nop
1292	SET_SIZE(sfmmu_kpm_patch_tsbm)
1293
1294	ENTRY_NP(sfmmu_patch_utsb)
1295#ifdef UTSB_PHYS
1296	retl
1297	nop
1298#else /* UTSB_PHYS */
1299	/*
1300	 * We need to hot patch utsb_vabase and utsb4m_vabase
1301	 */
1302	save	%sp, -SA(MINFRAME), %sp
1303
1304	/* patch value of utsb_vabase */
1305	set	utsb_vabase, %o1
1306	ldx	[%o1], %o4
1307	sethi	%hi(sfmmu_uprot_get_1st_tsbe_ptr), %o0
1308	call	sfmmu_fixup_setx
1309	  or	%o0, %lo(sfmmu_uprot_get_1st_tsbe_ptr), %o0
1310	sethi	%hi(sfmmu_uitlb_get_1st_tsbe_ptr), %o0
1311	call	sfmmu_fixup_setx
1312	  or	%o0, %lo(sfmmu_uitlb_get_1st_tsbe_ptr), %o0
1313	sethi	%hi(sfmmu_udtlb_get_1st_tsbe_ptr), %o0
1314	call	sfmmu_fixup_setx
1315	  or	%o0, %lo(sfmmu_udtlb_get_1st_tsbe_ptr), %o0
1316
1317	/* patch value of utsb4m_vabase */
1318	set	utsb4m_vabase, %o1
1319	ldx	[%o1], %o4
1320	sethi	%hi(sfmmu_uprot_get_2nd_tsb_base), %o0
1321	call	sfmmu_fixup_setx
1322	  or	%o0, %lo(sfmmu_uprot_get_2nd_tsb_base), %o0
1323	sethi	%hi(sfmmu_uitlb_get_2nd_tsb_base), %o0
1324	call	sfmmu_fixup_setx
1325	  or	%o0, %lo(sfmmu_uitlb_get_2nd_tsb_base), %o0
1326	sethi	%hi(sfmmu_udtlb_get_2nd_tsb_base), %o0
1327	call	sfmmu_fixup_setx
1328	  or	%o0, %lo(sfmmu_udtlb_get_2nd_tsb_base), %o0
1329
1330	/*
1331	 * Patch TSB base register masks and shifts if needed.
1332	 * By default the TSB base register contents are set up for 4M slab.
1333	 * If we're using a smaller slab size and reserved VA range we need
1334	 * to patch up those values here.
1335	 */
1336	set	tsb_slab_shift, %o1
1337	set	MMU_PAGESHIFT4M, %o4
1338	lduw	[%o1], %o3
1339	subcc	%o4, %o3, %o4
1340	bz,pt	%icc, 1f
1341	  /* delay slot safe */
1342
1343	/* patch reserved VA range size if needed. */
1344	sethi	%hi(sfmmu_tsb_1st_resv_offset), %o0
1345	call	sfmmu_fixup_shiftx
1346	  or	%o0, %lo(sfmmu_tsb_1st_resv_offset), %o0
1347	call	sfmmu_fixup_shiftx
1348	  add	%o0, I_SIZE, %o0
1349	sethi	%hi(sfmmu_tsb_2nd_resv_offset), %o0
1350	call	sfmmu_fixup_shiftx
1351	  or	%o0, %lo(sfmmu_tsb_2nd_resv_offset), %o0
1352	call	sfmmu_fixup_shiftx
1353	  add	%o0, I_SIZE, %o0
13541:
1355	/* patch TSBREG_VAMASK used to set up TSB base register */
1356	set	tsb_slab_mask, %o1
1357	ldx	[%o1], %o4
1358	sethi	%hi(sfmmu_tsb_1st_tsbreg_vamask), %o0
1359	call	sfmmu_fixup_or
1360	  or	%o0, %lo(sfmmu_tsb_1st_tsbreg_vamask), %o0
1361	sethi	%hi(sfmmu_tsb_2nd_tsbreg_vamask), %o0
1362	call	sfmmu_fixup_or
1363	  or	%o0, %lo(sfmmu_tsb_2nd_tsbreg_vamask), %o0
1364
1365	ret
1366	restore
1367#endif /* UTSB_PHYS */
1368	SET_SIZE(sfmmu_patch_utsb)
1369
1370	ENTRY_NP(sfmmu_patch_shctx)
1371#ifdef sun4u
1372	retl
1373	  nop
1374#else /* sun4u */
1375	set	sfmmu_shctx_cpu_mondo_patch, %o0
1376	MAKE_JMP_INSTR(5, %o1, %o2)	! jmp       %g5
1377	st	%o1, [%o0]
1378	flush	%o0
1379	MAKE_NOP_INSTR(%o1)
1380	add	%o0, I_SIZE, %o0	! next instr
1381	st	%o1, [%o0]
1382	flush	%o0
1383
1384	set	sfmmu_shctx_user_rtt_patch, %o0
1385	st      %o1, [%o0]		! nop 1st instruction
1386	flush	%o0
1387	add     %o0, I_SIZE, %o0
1388	st      %o1, [%o0]		! nop 2nd instruction
1389	flush	%o0
1390	add     %o0, I_SIZE, %o0
1391	st      %o1, [%o0]		! nop 3rd instruction
1392	flush	%o0
1393	add     %o0, I_SIZE, %o0
1394	st      %o1, [%o0]		! nop 4th instruction
1395	flush	%o0
1396	add     %o0, I_SIZE, %o0
1397	st      %o1, [%o0]		! nop 5th instruction
1398	retl
1399	  flush	%o0
1400#endif /* sun4u */
1401	SET_SIZE(sfmmu_patch_shctx)
1402
1403	/*
1404	 * Routine that loads an entry into a tsb using virtual addresses.
1405	 * Locking is required since all cpus can use the same TSB.
1406	 * Note that it is no longer required to have a valid context
1407	 * when calling this function.
1408	 */
1409	ENTRY_NP(sfmmu_load_tsbe)
1410	/*
1411	 * %o0 = pointer to tsbe to load
1412	 * %o1 = tsb tag
1413	 * %o2 = virtual pointer to TTE
1414	 * %o3 = 1 if physical address in %o0 else 0
1415	 */
1416	rdpr	%pstate, %o5
1417#ifdef DEBUG
1418	PANIC_IF_INTR_DISABLED_PSTR(%o5, sfmmu_di_l2, %g1)
1419#endif /* DEBUG */
1420
1421	wrpr	%o5, PSTATE_IE, %pstate		/* disable interrupts */
1422
1423	SETUP_TSB_ASI(%o3, %g3)
1424	TSB_UPDATE(%o0, %o2, %o1, %g1, %g2, 1)
1425
1426	wrpr	%g0, %o5, %pstate		/* enable interrupts */
1427
1428	retl
1429	membar	#StoreStore|#StoreLoad
1430	SET_SIZE(sfmmu_load_tsbe)
1431
1432	/*
1433	 * Flush TSB of a given entry if the tag matches.
1434	 */
1435	ENTRY(sfmmu_unload_tsbe)
1436	/*
1437	 * %o0 = pointer to tsbe to be flushed
1438	 * %o1 = tag to match
1439	 * %o2 = 1 if physical address in %o0 else 0
1440	 */
1441	SETUP_TSB_ASI(%o2, %g1)
1442	TSB_INVALIDATE(%o0, %o1, %g1, %o2, %o3, unload_tsbe)
1443	retl
1444	membar	#StoreStore|#StoreLoad
1445	SET_SIZE(sfmmu_unload_tsbe)
1446
1447	/*
1448	 * Routine that loads a TTE into the kpm TSB from C code.
1449	 * Locking is required since kpm TSB is shared among all CPUs.
1450	 */
1451	ENTRY_NP(sfmmu_kpm_load_tsb)
1452	/*
1453	 * %o0 = vaddr
1454	 * %o1 = ttep
1455	 * %o2 = virtpg to TSB index shift (e.g. TTE pagesize shift)
1456	 */
1457	rdpr	%pstate, %o5			! %o5 = saved pstate
1458#ifdef DEBUG
1459	PANIC_IF_INTR_DISABLED_PSTR(%o5, sfmmu_di_l3, %g1)
1460#endif /* DEBUG */
1461	wrpr	%o5, PSTATE_IE, %pstate		! disable interrupts
1462
1463#ifndef sun4v
1464	sethi	%hi(ktsb_phys), %o4
1465	mov	ASI_N, %o3
1466	ld	[%o4 + %lo(ktsb_phys)], %o4
1467	movrnz	%o4, ASI_MEM, %o3
1468	mov	%o3, %asi
1469#endif
1470	mov	%o0, %g1			! %g1 = vaddr
1471
1472	/* GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr (clobbers), tmp1, tmp2) */
1473	GET_KPM_TSBE_POINTER(%o2, %g2, %g1, %o3, %o4)
1474	/* %g2 = tsbep, %g1 clobbered */
1475
1476	srlx	%o0, TTARGET_VA_SHIFT, %g1;	! %g1 = tag target
1477	/* TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label) */
1478	TSB_UPDATE(%g2, %o1, %g1, %o3, %o4, 1)
1479
1480	wrpr	%g0, %o5, %pstate		! enable interrupts
1481	retl
1482	  membar #StoreStore|#StoreLoad
1483	SET_SIZE(sfmmu_kpm_load_tsb)
1484
1485	/*
1486	 * Routine that shoots down a TTE in the kpm TSB or in the
1487	 * kernel TSB depending on virtpg. Locking is required since
1488	 * kpm/kernel TSB is shared among all CPUs.
1489	 */
1490	ENTRY_NP(sfmmu_kpm_unload_tsb)
1491	/*
1492	 * %o0 = vaddr
1493	 * %o1 = virtpg to TSB index shift (e.g. TTE page shift)
1494	 */
1495#ifndef sun4v
1496	sethi	%hi(ktsb_phys), %o4
1497	mov	ASI_N, %o3
1498	ld	[%o4 + %lo(ktsb_phys)], %o4
1499	movrnz	%o4, ASI_MEM, %o3
1500	mov	%o3, %asi
1501#endif
1502	mov	%o0, %g1			! %g1 = vaddr
1503
1504	/* GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr (clobbers), tmp1, tmp2) */
1505	GET_KPM_TSBE_POINTER(%o1, %g2, %g1, %o3, %o4)
1506	/* %g2 = tsbep, %g1 clobbered */
1507
1508	srlx	%o0, TTARGET_VA_SHIFT, %g1;	! %g1 = tag target
1509	/* TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label) */
1510	TSB_INVALIDATE(%g2, %g1, %o3, %o4, %o1, kpm_tsbinval)
1511
1512	retl
1513	  membar	#StoreStore|#StoreLoad
1514	SET_SIZE(sfmmu_kpm_unload_tsb)
1515
1516#endif /* lint */
1517
1518
1519#if defined (lint)
1520
1521/*ARGSUSED*/
1522pfn_t
1523sfmmu_ttetopfn(tte_t *tte, caddr_t vaddr)
1524{ return(0); }
1525
1526#else /* lint */
1527
1528	ENTRY_NP(sfmmu_ttetopfn)
1529	ldx	[%o0], %g1			/* read tte */
1530	TTETOPFN(%g1, %o1, sfmmu_ttetopfn_l1, %g2, %g3, %g4)
1531	/*
1532	 * g1 = pfn
1533	 */
1534	retl
1535	mov	%g1, %o0
1536	SET_SIZE(sfmmu_ttetopfn)
1537
1538#endif /* !lint */
1539
1540
1541#if defined (lint)
1542/*
1543 * The sfmmu_hblk_hash_add is the assembly primitive for adding hmeblks to the
1544 * the hash list.
1545 */
1546/* ARGSUSED */
1547void
1548sfmmu_hblk_hash_add(struct hmehash_bucket *hmebp, struct hme_blk *hmeblkp,
1549	uint64_t hblkpa)
1550{
1551}
1552
1553/*
1554 * The sfmmu_hblk_hash_rm is the assembly primitive to remove hmeblks from the
1555 * hash list.
1556 */
1557/* ARGSUSED */
1558void
1559sfmmu_hblk_hash_rm(struct hmehash_bucket *hmebp, struct hme_blk *hmeblkp,
1560	uint64_t hblkpa, struct hme_blk *prev_hblkp)
1561{
1562}
1563#else /* lint */
1564
1565/*
1566 * Functions to grab/release hme bucket list lock.  I only use a byte
1567 * instead of the whole int because eventually we might want to
1568 * put some counters on the other bytes (of course, these routines would
1569 * have to change).  The code that grab this lock should execute
1570 * with interrupts disabled and hold the lock for the least amount of time
1571 * possible.
1572 */
1573
1574/*
1575 * Even though hmeh_listlock is updated using pa there's no need to flush
1576 * dcache since hmeh_listlock will be restored to the original value (0)
1577 * before interrupts are reenabled.
1578 */
1579
1580/*
1581 * For sparcv9 hme hash buckets may not be in the nucleus.  hme hash update
1582 * routines still use virtual addresses to update the bucket fields. But they
1583 * must not cause a TLB miss after grabbing the low level bucket lock. To
1584 * achieve this we must make sure the bucket structure is completely within an
1585 * 8K page.
1586 */
1587
1588#if (HMEBUCK_SIZE & (HMEBUCK_SIZE - 1))
1589#error - the size of hmehash_bucket structure is not power of 2
1590#endif
1591
1592#define HMELOCK_ENTER(hmebp, tmp1, tmp2, label1, asi)           \
1593	mov     0xff, tmp2                                      ;\
1594	add     hmebp, HMEBUCK_LOCK, tmp1                       ;\
1595label1:                                                         ;\
1596	casa    [tmp1]asi, %g0, tmp2                            ;\
1597	brnz,pn tmp2, label1                                    ;\
1598	mov     0xff, tmp2                                      ;\
1599	membar  #LoadLoad
1600
1601#define HMELOCK_EXIT(hmebp, tmp1, asi)                          \
1602	membar  #LoadStore|#StoreStore                          ;\
1603	add     hmebp, HMEBUCK_LOCK, tmp1                       ;\
1604	sta     %g0, [tmp1]asi
1605
1606	.seg	".data"
1607hblk_add_panic1:
1608	.ascii	"sfmmu_hblk_hash_add: interrupts disabled"
1609	.byte	0
1610hblk_add_panic2:
1611	.ascii	"sfmmu_hblk_hash_add: va hmeblkp is NULL but pa is not"
1612	.byte	0
1613	.align	4
1614	.seg	".text"
1615
1616	ENTRY_NP(sfmmu_hblk_hash_add)
1617	/*
1618	 * %o0 = hmebp
1619	 * %o1 = hmeblkp
1620	 * %o2 = hblkpa
1621	 */
1622	rdpr	%pstate, %o5
1623#ifdef DEBUG
1624	andcc	%o5, PSTATE_IE, %g0		/* if interrupts already */
1625	bnz,pt %icc, 3f				/* disabled, panic	 */
1626	  nop
1627	save	%sp, -SA(MINFRAME), %sp
1628	sethi	%hi(hblk_add_panic1), %o0
1629	call	panic
1630	 or	%o0, %lo(hblk_add_panic1), %o0
1631	ret
1632	restore
1633
16343:
1635#endif /* DEBUG */
1636	wrpr	%o5, PSTATE_IE, %pstate		/* disable interrupts */
1637	mov	%o2, %g1
1638
1639	/*
1640	 * g1 = hblkpa
1641	 */
1642	ldn	[%o0 + HMEBUCK_HBLK], %o4	/* next hmeblk */
1643	ldx	[%o0 + HMEBUCK_NEXTPA], %g2	/* g2 = next hblkpa */
1644#ifdef	DEBUG
1645	cmp	%o4, %g0
1646	bne,pt %xcc, 1f
1647	 nop
1648	brz,pt %g2, 1f
1649	 nop
1650	wrpr	%g0, %o5, %pstate		/* enable interrupts */
1651	save	%sp, -SA(MINFRAME), %sp
1652	sethi	%hi(hblk_add_panic2), %o0
1653	call	panic
1654	  or	%o0, %lo(hblk_add_panic2), %o0
1655	ret
1656	restore
16571:
1658#endif /* DEBUG */
1659	/*
1660	 * We update hmeblks entries before grabbing lock because the stores
1661	 * could take a tlb miss and require the hash lock.  The buckets
1662	 * are part of the nucleus so we are cool with those stores.
1663	 *
1664	 * if buckets are not part of the nucleus our game is to
1665	 * not touch any other page via va until we drop the lock.
1666	 * This guarantees we won't get a tlb miss before the lock release
1667	 * since interrupts are disabled.
1668	 */
1669	stn	%o4, [%o1 + HMEBLK_NEXT]	/* update hmeblk's next */
1670	stx	%g2, [%o1 + HMEBLK_NEXTPA]	/* update hmeblk's next pa */
1671	HMELOCK_ENTER(%o0, %o2, %o3, hashadd1, ASI_N)
1672	stn	%o1, [%o0 + HMEBUCK_HBLK]	/* update bucket hblk next */
1673	stx	%g1, [%o0 + HMEBUCK_NEXTPA]	/* add hmeblk to list */
1674	HMELOCK_EXIT(%o0, %g2, ASI_N)
1675	retl
1676	  wrpr	%g0, %o5, %pstate		/* enable interrupts */
1677	SET_SIZE(sfmmu_hblk_hash_add)
1678
1679	ENTRY_NP(sfmmu_hblk_hash_rm)
1680	/*
1681	 * This function removes an hmeblk from the hash chain.
1682	 * It is written to guarantee we don't take a tlb miss
1683	 * by using physical addresses to update the list.
1684	 *
1685	 * %o0 = hmebp
1686	 * %o1 = hmeblkp
1687	 * %o2 = hmeblkp previous pa
1688	 * %o3 = hmeblkp previous
1689	 */
1690
1691	mov	%o3, %o4			/* o4 = hmeblkp previous */
1692
1693	rdpr	%pstate, %o5
1694#ifdef DEBUG
1695	PANIC_IF_INTR_DISABLED_PSTR(%o5, sfmmu_di_l4, %g1)
1696#endif /* DEBUG */
1697	/*
1698	 * disable interrupts, clear Address Mask to access 64 bit physaddr
1699	 */
1700	andn    %o5, PSTATE_IE, %g1
1701	wrpr    %g1, 0, %pstate
1702
1703#ifndef sun4v
1704	sethi   %hi(dcache_line_mask), %g4
1705	ld      [%g4 + %lo(dcache_line_mask)], %g4
1706#endif /* sun4v */
1707
1708	/*
1709	 * if buckets are not part of the nucleus our game is to
1710	 * not touch any other page via va until we drop the lock.
1711	 * This guarantees we won't get a tlb miss before the lock release
1712	 * since interrupts are disabled.
1713	 */
1714	HMELOCK_ENTER(%o0, %g1, %g3, hashrm1, ASI_N)
1715	ldn	[%o0 + HMEBUCK_HBLK], %g2	/* first hmeblk in list */
1716	cmp	%g2, %o1
1717	bne,pt	%ncc,1f
1718	 mov	ASI_MEM, %asi
1719	/*
1720	 * hmeblk is first on list
1721	 */
1722	ldx	[%o0 + HMEBUCK_NEXTPA], %g2	/* g2 = hmeblk pa */
1723	ldna	[%g2 + HMEBLK_NEXT] %asi, %o3	/* read next hmeblk va */
1724	ldxa	[%g2 + HMEBLK_NEXTPA] %asi, %g1	/* read next hmeblk pa */
1725	stn	%o3, [%o0 + HMEBUCK_HBLK]	/* write va */
1726	ba,pt	%xcc, 2f
1727	  stx	%g1, [%o0 + HMEBUCK_NEXTPA]	/* write pa */
17281:
1729	/* hmeblk is not first on list */
1730
1731	mov	%o2, %g3
1732#ifndef sun4v
1733	GET_CPU_IMPL(%g2)
1734	cmp 	%g2, CHEETAH_IMPL
1735	bge,a,pt %icc, hblk_hash_rm_1
1736	  and	%o4, %g4, %g2
1737	cmp	%g2, SPITFIRE_IMPL
1738	blt	%icc, hblk_hash_rm_2		/* no flushing needed for OPL */
1739	  and	%o4, %g4, %g2
1740	stxa	%g0, [%g2]ASI_DC_TAG		/* flush prev pa from dcache */
1741	add	%o4, HMEBLK_NEXT, %o4
1742	and	%o4, %g4, %g2
1743	ba	hblk_hash_rm_2
1744	stxa	%g0, [%g2]ASI_DC_TAG		/* flush prev va from dcache */
1745hblk_hash_rm_1:
1746
1747	stxa	%g0, [%g3]ASI_DC_INVAL		/* flush prev pa from dcache */
1748	membar	#Sync
1749	add     %g3, HMEBLK_NEXT, %g2
1750	stxa	%g0, [%g2]ASI_DC_INVAL		/* flush prev va from dcache */
1751hblk_hash_rm_2:
1752	membar	#Sync
1753#endif /* sun4v */
1754	ldxa	[%g3 + HMEBLK_NEXTPA] %asi, %g2	/* g2 = hmeblk pa */
1755	ldna	[%g2 + HMEBLK_NEXT] %asi, %o3	/* read next hmeblk va */
1756	ldxa	[%g2 + HMEBLK_NEXTPA] %asi, %g1	/* read next hmeblk pa */
1757	stna	%o3, [%g3 + HMEBLK_NEXT] %asi	/* write va */
1758	stxa	%g1, [%g3 + HMEBLK_NEXTPA] %asi	/* write pa */
17592:
1760	HMELOCK_EXIT(%o0, %g2, ASI_N)
1761	retl
1762	  wrpr	%g0, %o5, %pstate		/* enable interrupts */
1763	SET_SIZE(sfmmu_hblk_hash_rm)
1764
1765#endif /* lint */
1766
1767/*
1768 * These macros are used to update global sfmmu hme hash statistics
1769 * in perf critical paths. It is only enabled in debug kernels or
1770 * if SFMMU_STAT_GATHER is defined
1771 */
1772#if defined(DEBUG) || defined(SFMMU_STAT_GATHER)
1773#define	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tmp1, tmp2)			\
1774	ldn	[tsbarea + TSBMISS_KHATID], tmp1			;\
1775	mov	HATSTAT_KHASH_SEARCH, tmp2				;\
1776	cmp	tmp1, hatid						;\
1777	movne	%ncc, HATSTAT_UHASH_SEARCH, tmp2			;\
1778	set	sfmmu_global_stat, tmp1					;\
1779	add	tmp1, tmp2, tmp1					;\
1780	ld	[tmp1], tmp2						;\
1781	inc	tmp2							;\
1782	st	tmp2, [tmp1]
1783
1784#define	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)			\
1785	ldn	[tsbarea + TSBMISS_KHATID], tmp1			;\
1786	mov	HATSTAT_KHASH_LINKS, tmp2				;\
1787	cmp	tmp1, hatid						;\
1788	movne	%ncc, HATSTAT_UHASH_LINKS, tmp2				;\
1789	set	sfmmu_global_stat, tmp1					;\
1790	add	tmp1, tmp2, tmp1					;\
1791	ld	[tmp1], tmp2						;\
1792	inc	tmp2							;\
1793	st	tmp2, [tmp1]
1794
1795
1796#else /* DEBUG || SFMMU_STAT_GATHER */
1797
1798#define	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tmp1, tmp2)
1799
1800#define	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)
1801
1802#endif  /* DEBUG || SFMMU_STAT_GATHER */
1803
1804/*
1805 * This macro is used to update global sfmmu kstas in non
1806 * perf critical areas so they are enabled all the time
1807 */
1808#define	HAT_GLOBAL_STAT(statname, tmp1, tmp2)				\
1809	sethi	%hi(sfmmu_global_stat), tmp1				;\
1810	add	tmp1, statname, tmp1					;\
1811	ld	[tmp1 + %lo(sfmmu_global_stat)], tmp2			;\
1812	inc	tmp2							;\
1813	st	tmp2, [tmp1 + %lo(sfmmu_global_stat)]
1814
1815/*
1816 * These macros are used to update per cpu stats in non perf
1817 * critical areas so they are enabled all the time
1818 */
1819#define	HAT_PERCPU_STAT32(tsbarea, stat, tmp1)				\
1820	ld	[tsbarea + stat], tmp1					;\
1821	inc	tmp1							;\
1822	st	tmp1, [tsbarea + stat]
1823
1824/*
1825 * These macros are used to update per cpu stats in non perf
1826 * critical areas so they are enabled all the time
1827 */
1828#define	HAT_PERCPU_STAT16(tsbarea, stat, tmp1)				\
1829	lduh	[tsbarea + stat], tmp1					;\
1830	inc	tmp1							;\
1831	stuh	tmp1, [tsbarea + stat]
1832
1833#if defined(KPM_TLBMISS_STATS_GATHER)
1834	/*
1835	 * Count kpm dtlb misses separately to allow a different
1836	 * evaluation of hme and kpm tlbmisses. kpm tsb hits can
1837	 * be calculated by (kpm_dtlb_misses - kpm_tsb_misses).
1838	 */
1839#define	KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label)		\
1840	brgez	tagacc, label	/* KPM VA? */				;\
1841	nop								;\
1842	CPU_INDEX(tmp1, tsbma)						;\
1843	sethi	%hi(kpmtsbm_area), tsbma				;\
1844	sllx	tmp1, KPMTSBM_SHIFT, tmp1				;\
1845	or	tsbma, %lo(kpmtsbm_area), tsbma				;\
1846	add	tsbma, tmp1, tsbma		/* kpmtsbm area */	;\
1847	/* VA range check */						;\
1848	ldx	[tsbma + KPMTSBM_VBASE], val				;\
1849	cmp	tagacc, val						;\
1850	blu,pn	%xcc, label						;\
1851	  ldx	[tsbma + KPMTSBM_VEND], tmp1				;\
1852	cmp	tagacc, tmp1						;\
1853	bgeu,pn	%xcc, label						;\
1854	  lduw	[tsbma + KPMTSBM_DTLBMISS], val				;\
1855	inc	val							;\
1856	st	val, [tsbma + KPMTSBM_DTLBMISS]				;\
1857label:
1858#else
1859#define	KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label)
1860#endif	/* KPM_TLBMISS_STATS_GATHER */
1861
1862#if defined (lint)
1863/*
1864 * The following routines are jumped to from the mmu trap handlers to do
1865 * the setting up to call systrap.  They are separate routines instead of
1866 * being part of the handlers because the handlers would exceed 32
1867 * instructions and since this is part of the slow path the jump
1868 * cost is irrelevant.
1869 */
1870void
1871sfmmu_pagefault(void)
1872{
1873}
1874
1875void
1876sfmmu_mmu_trap(void)
1877{
1878}
1879
1880void
1881sfmmu_window_trap(void)
1882{
1883}
1884
1885void
1886sfmmu_kpm_exception(void)
1887{
1888}
1889
1890#else /* lint */
1891
1892#ifdef	PTL1_PANIC_DEBUG
1893	.seg	".data"
1894	.global	test_ptl1_panic
1895test_ptl1_panic:
1896	.word	0
1897	.align	8
1898
1899	.seg	".text"
1900	.align	4
1901#endif	/* PTL1_PANIC_DEBUG */
1902
1903
1904	ENTRY_NP(sfmmu_pagefault)
1905	SET_GL_REG(1)
1906	USE_ALTERNATE_GLOBALS(%g5)
1907	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g6, %g4)
1908	rdpr	%tt, %g6
1909	cmp	%g6, FAST_IMMU_MISS_TT
1910	be,a,pn	%icc, 1f
1911	  mov	T_INSTR_MMU_MISS, %g3
1912	cmp	%g6, T_INSTR_MMU_MISS
1913	be,a,pn	%icc, 1f
1914	  mov	T_INSTR_MMU_MISS, %g3
1915	mov	%g5, %g2
1916	mov	T_DATA_PROT, %g3		/* arg2 = traptype */
1917	cmp	%g6, FAST_DMMU_MISS_TT
1918	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1919	cmp	%g6, T_DATA_MMU_MISS
1920	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1921
1922#ifdef  PTL1_PANIC_DEBUG
1923	/* check if we want to test the tl1 panic */
1924	sethi	%hi(test_ptl1_panic), %g4
1925	ld	[%g4 + %lo(test_ptl1_panic)], %g1
1926	st	%g0, [%g4 + %lo(test_ptl1_panic)]
1927	cmp	%g1, %g0
1928	bne,a,pn %icc, ptl1_panic
1929	  or	%g0, PTL1_BAD_DEBUG, %g1
1930#endif	/* PTL1_PANIC_DEBUG */
19311:
1932	HAT_GLOBAL_STAT(HATSTAT_PAGEFAULT, %g6, %g4)
1933	/*
1934	 * g2 = tag access reg
1935	 * g3.l = type
1936	 * g3.h = 0
1937	 */
1938	sethi	%hi(trap), %g1
1939	or	%g1, %lo(trap), %g1
19402:
1941	ba,pt	%xcc, sys_trap
1942	  mov	-1, %g4
1943	SET_SIZE(sfmmu_pagefault)
1944
1945	ENTRY_NP(sfmmu_mmu_trap)
1946	SET_GL_REG(1)
1947	USE_ALTERNATE_GLOBALS(%g5)
1948	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g4, %g6)
1949	rdpr	%tt, %g6
1950	cmp	%g6, FAST_IMMU_MISS_TT
1951	be,a,pn	%icc, 1f
1952	  mov	T_INSTR_MMU_MISS, %g3
1953	cmp	%g6, T_INSTR_MMU_MISS
1954	be,a,pn	%icc, 1f
1955	  mov	T_INSTR_MMU_MISS, %g3
1956	mov	%g5, %g2
1957	mov	T_DATA_PROT, %g3		/* arg2 = traptype */
1958	cmp	%g6, FAST_DMMU_MISS_TT
1959	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1960	cmp	%g6, T_DATA_MMU_MISS
1961	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
19621:
1963	/*
1964	 * g2 = tag access reg
1965	 * g3 = type
1966	 */
1967	sethi	%hi(sfmmu_tsbmiss_exception), %g1
1968	or	%g1, %lo(sfmmu_tsbmiss_exception), %g1
1969	ba,pt	%xcc, sys_trap
1970	  mov	-1, %g4
1971	/*NOTREACHED*/
1972	SET_SIZE(sfmmu_mmu_trap)
1973
1974	ENTRY_NP(sfmmu_suspend_tl)
1975	SET_GL_REG(1)
1976	USE_ALTERNATE_GLOBALS(%g5)
1977	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g4, %g3)
1978	rdpr	%tt, %g6
1979	cmp	%g6, FAST_IMMU_MISS_TT
1980	be,a,pn	%icc, 1f
1981	  mov	T_INSTR_MMU_MISS, %g3
1982	mov	%g5, %g2
1983	cmp	%g6, FAST_DMMU_MISS_TT
1984	move	%icc, T_DATA_MMU_MISS, %g3
1985	movne	%icc, T_DATA_PROT, %g3
19861:
1987	sethi	%hi(sfmmu_tsbmiss_suspended), %g1
1988	or	%g1, %lo(sfmmu_tsbmiss_suspended), %g1
1989	/* g1 = TL0 handler, g2 = tagacc, g3 = trap type */
1990	ba,pt	%xcc, sys_trap
1991	  mov	PIL_15, %g4
1992	/*NOTREACHED*/
1993	SET_SIZE(sfmmu_suspend_tl)
1994
1995	/*
1996	 * No %g registers in use at this point.
1997	 */
1998	ENTRY_NP(sfmmu_window_trap)
1999	rdpr	%tpc, %g1
2000#ifdef sun4v
2001#ifdef DEBUG
2002	/* We assume previous %gl was 1 */
2003	rdpr	%tstate, %g4
2004	srlx	%g4, TSTATE_GL_SHIFT, %g4
2005	and	%g4, TSTATE_GL_MASK, %g4
2006	cmp	%g4, 1
2007	bne,a,pn %icc, ptl1_panic
2008	  mov	PTL1_BAD_WTRAP, %g1
2009#endif /* DEBUG */
2010	/* user miss at tl>1. better be the window handler or user_rtt */
2011	/* in user_rtt? */
2012	set	rtt_fill_start, %g4
2013	cmp	%g1, %g4
2014	blu,pn %xcc, 6f
2015	 .empty
2016	set	rtt_fill_end, %g4
2017	cmp	%g1, %g4
2018	bgeu,pn %xcc, 6f
2019	 nop
2020	set	fault_rtt_fn1, %g1
2021	wrpr	%g0, %g1, %tnpc
2022	ba,a	7f
20236:
2024	! must save this trap level before descending trap stack
2025	! no need to save %tnpc, either overwritten or discarded
2026	! already got it: rdpr	%tpc, %g1
2027	rdpr	%tstate, %g6
2028	rdpr	%tt, %g7
2029	! trap level saved, go get underlying trap type
2030	rdpr	%tl, %g5
2031	sub	%g5, 1, %g3
2032	wrpr	%g3, %tl
2033	rdpr	%tt, %g2
2034	wrpr	%g5, %tl
2035	! restore saved trap level
2036	wrpr	%g1, %tpc
2037	wrpr	%g6, %tstate
2038	wrpr	%g7, %tt
2039#else /* sun4v */
2040	/* user miss at tl>1. better be the window handler */
2041	rdpr	%tl, %g5
2042	sub	%g5, 1, %g3
2043	wrpr	%g3, %tl
2044	rdpr	%tt, %g2
2045	wrpr	%g5, %tl
2046#endif /* sun4v */
2047	and	%g2, WTRAP_TTMASK, %g4
2048	cmp	%g4, WTRAP_TYPE
2049	bne,pn	%xcc, 1f
2050	 nop
2051	/* tpc should be in the trap table */
2052	set	trap_table, %g4
2053	cmp	%g1, %g4
2054	blt,pn %xcc, 1f
2055	 .empty
2056	set	etrap_table, %g4
2057	cmp	%g1, %g4
2058	bge,pn %xcc, 1f
2059	 .empty
2060	andn	%g1, WTRAP_ALIGN, %g1	/* 128 byte aligned */
2061	add	%g1, WTRAP_FAULTOFF, %g1
2062	wrpr	%g0, %g1, %tnpc
20637:
2064	/*
2065	 * some wbuf handlers will call systrap to resolve the fault
2066	 * we pass the trap type so they figure out the correct parameters.
2067	 * g5 = trap type, g6 = tag access reg
2068	 */
2069
2070	/*
2071	 * only use g5, g6, g7 registers after we have switched to alternate
2072	 * globals.
2073	 */
2074	SET_GL_REG(1)
2075	USE_ALTERNATE_GLOBALS(%g5)
2076	GET_MMU_D_TAGACC(%g6 /*dtag*/, %g5 /*scratch*/)
2077	rdpr	%tt, %g7
2078	cmp	%g7, FAST_IMMU_MISS_TT
2079	be,a,pn	%icc, ptl1_panic
2080	  mov	PTL1_BAD_WTRAP, %g1
2081	cmp	%g7, T_INSTR_MMU_MISS
2082	be,a,pn	%icc, ptl1_panic
2083	  mov	PTL1_BAD_WTRAP, %g1
2084	mov	T_DATA_PROT, %g5
2085	cmp	%g7, FAST_DMMU_MISS_TT
2086	move	%icc, T_DATA_MMU_MISS, %g5
2087	cmp	%g7, T_DATA_MMU_MISS
2088	move	%icc, T_DATA_MMU_MISS, %g5
2089	! XXXQ AGS re-check out this one
2090	done
20911:
2092	CPU_PADDR(%g1, %g4)
2093	add	%g1, CPU_TL1_HDLR, %g1
2094	lda	[%g1]ASI_MEM, %g4
2095	brnz,a,pt %g4, sfmmu_mmu_trap
2096	  sta	%g0, [%g1]ASI_MEM
2097	ba,pt	%icc, ptl1_panic
2098	  mov	PTL1_BAD_TRAP, %g1
2099	SET_SIZE(sfmmu_window_trap)
2100
2101	ENTRY_NP(sfmmu_kpm_exception)
2102	/*
2103	 * We have accessed an unmapped segkpm address or a legal segkpm
2104	 * address which is involved in a VAC alias conflict prevention.
2105	 * Before we go to trap(), check to see if CPU_DTRACE_NOFAULT is
2106	 * set. If it is, we will instead note that a fault has occurred
2107	 * by setting CPU_DTRACE_BADADDR and issue a "done" (instead of
2108	 * a "retry"). This will step over the faulting instruction.
2109	 * Note that this means that a legal segkpm address involved in
2110	 * a VAC alias conflict prevention (a rare case to begin with)
2111	 * cannot be used in DTrace.
2112	 */
2113	CPU_INDEX(%g1, %g2)
2114	set	cpu_core, %g2
2115	sllx	%g1, CPU_CORE_SHIFT, %g1
2116	add	%g1, %g2, %g1
2117	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
2118	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
2119	bz	0f
2120	or	%g2, CPU_DTRACE_BADADDR, %g2
2121	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
2122	GET_MMU_D_ADDR(%g3, /*scratch*/ %g4)
2123	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
2124	done
21250:
2126	TSTAT_CHECK_TL1(1f, %g1, %g2)
21271:
2128	SET_GL_REG(1)
2129	USE_ALTERNATE_GLOBALS(%g5)
2130	GET_MMU_D_TAGACC(%g2 /* tagacc */, %g4 /*scratch*/)
2131	mov	T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
2132	/*
2133	 * g2=tagacc g3.l=type g3.h=0
2134	 */
2135	sethi	%hi(trap), %g1
2136	or	%g1, %lo(trap), %g1
2137	ba,pt	%xcc, sys_trap
2138	mov	-1, %g4
2139	SET_SIZE(sfmmu_kpm_exception)
2140
2141#endif /* lint */
2142
2143#if defined (lint)
2144
2145void
2146sfmmu_tsb_miss(void)
2147{
2148}
2149
2150void
2151sfmmu_kpm_dtsb_miss(void)
2152{
2153}
2154
2155void
2156sfmmu_kpm_dtsb_miss_small(void)
2157{
2158}
2159
2160#else /* lint */
2161
2162#if (IMAP_SEG != 0)
2163#error - ism_map->ism_seg offset is not zero
2164#endif
2165
2166/*
2167 * Copies ism mapping for this ctx in param "ism" if this is a ISM
2168 * tlb miss and branches to label "ismhit". If this is not an ISM
2169 * process or an ISM tlb miss it falls thru.
2170 *
2171 * Checks to see if the vaddr passed in via tagacc is in an ISM segment for
2172 * this process.
2173 * If so, it will branch to label "ismhit".  If not, it will fall through.
2174 *
2175 * Also hat_unshare() will set the context for this process to INVALID_CONTEXT
2176 * so that any other threads of this process will not try and walk the ism
2177 * maps while they are being changed.
2178 *
2179 * NOTE: We will never have any holes in our ISM maps. sfmmu_share/unshare
2180 *       will make sure of that. This means we can terminate our search on
2181 *       the first zero mapping we find.
2182 *
2183 * Parameters:
2184 * tagacc	= (pseudo-)tag access register (vaddr + ctx) (in)
2185 * tsbmiss	= address of tsb miss area (in)
2186 * ismseg	= contents of ism_seg for this ism map (out)
2187 * ismhat	= physical address of imap_ismhat for this ism map (out)
2188 * tmp1		= scratch reg (CLOBBERED)
2189 * tmp2		= scratch reg (CLOBBERED)
2190 * tmp3		= scratch reg (CLOBBERED)
2191 * label:    temporary labels
2192 * ismhit:   label where to jump to if an ism dtlb miss
2193 * exitlabel:label where to jump if hat is busy due to hat_unshare.
2194 */
2195#define ISM_CHECK(tagacc, tsbmiss, ismseg, ismhat, tmp1, tmp2, tmp3 \
2196	label, ismhit)							\
2197	ldx	[tsbmiss + TSBMISS_ISMBLKPA], tmp1 /* tmp1 = &ismblk */	;\
2198	brlz,pt  tmp1, label/**/3		/* exit if -1 */	;\
2199	  add	tmp1, IBLK_MAPS, ismhat	/* ismhat = &ismblk.map[0] */	;\
2200label/**/1:								;\
2201	ldxa	[ismhat]ASI_MEM, ismseg	/* ismblk.map[0].ism_seg */	;\
2202	mov	tmp1, tmp3	/* update current ismblkpa head */	;\
2203label/**/2:								;\
2204	brz,pt  ismseg, label/**/3		/* no mapping */	;\
2205	  add	ismhat, IMAP_VB_SHIFT, tmp1 /* tmp1 = vb_shift addr */	;\
2206	lduba	[tmp1]ASI_MEM, tmp1 		/* tmp1 = vb shift*/	;\
2207	srlx	ismseg, tmp1, tmp2		/* tmp2 = vbase */	;\
2208	srlx	tagacc, tmp1, tmp1		/* tmp1 =  va seg*/	;\
2209	sub	tmp1, tmp2, tmp2		/* tmp2 = va - vbase */	;\
2210	add	ismhat, IMAP_SZ_MASK, tmp1 /* tmp1 = sz_mask addr */	;\
2211	lda	[tmp1]ASI_MEM, tmp1		/* tmp1 = sz_mask */	;\
2212	and	ismseg, tmp1, tmp1		/* tmp1 = size */	;\
2213	cmp	tmp2, tmp1		 	/* check va <= offset*/	;\
2214	blu,a,pt  %xcc, ismhit			/* ism hit */		;\
2215	  add	ismhat, IMAP_ISMHAT, ismhat 	/* ismhat = &ism_sfmmu*/ ;\
2216									;\
2217	add	ismhat, ISM_MAP_SZ, ismhat /* ismhat += sizeof(map) */ 	;\
2218	add	tmp3, (IBLK_MAPS + ISM_MAP_SLOTS * ISM_MAP_SZ), tmp1	;\
2219	cmp	ismhat, tmp1						;\
2220	bl,pt	%xcc, label/**/2		/* keep looking  */	;\
2221	  ldxa	[ismhat]ASI_MEM, ismseg	/* ismseg = map[ismhat] */	;\
2222									;\
2223	add	tmp3, IBLK_NEXTPA, tmp1					;\
2224	ldxa	[tmp1]ASI_MEM, tmp1		/* check blk->nextpa */	;\
2225	brgez,pt tmp1, label/**/1		/* continue if not -1*/	;\
2226	  add	tmp1, IBLK_MAPS, ismhat	/* ismhat = &ismblk.map[0]*/	;\
2227label/**/3:
2228
2229/*
2230 * Returns the hme hash bucket (hmebp) given the vaddr, and the hatid
2231 * It also returns the virtual pg for vaddr (ie. vaddr << hmeshift)
2232 * Parameters:
2233 * tagacc = reg containing virtual address
2234 * hatid = reg containing sfmmu pointer
2235 * hmeshift = constant/register to shift vaddr to obtain vapg
2236 * hmebp = register where bucket pointer will be stored
2237 * vapg = register where virtual page will be stored
2238 * tmp1, tmp2 = tmp registers
2239 */
2240
2241
2242#define	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, hmebp,	\
2243	vapg, label, tmp1, tmp2)					\
2244	sllx	tagacc, TAGACC_CTX_LSHIFT, tmp1				;\
2245	brnz,a,pt tmp1, label/**/1					;\
2246	  ld    [tsbarea + TSBMISS_UHASHSZ], hmebp			;\
2247	ld	[tsbarea + TSBMISS_KHASHSZ], hmebp			;\
2248	ba,pt	%xcc, label/**/2					;\
2249	  ldx	[tsbarea + TSBMISS_KHASHSTART], tmp1			;\
2250label/**/1:								;\
2251	ldx	[tsbarea + TSBMISS_UHASHSTART], tmp1			;\
2252label/**/2:								;\
2253	srlx	tagacc, hmeshift, vapg					;\
2254	xor	vapg, hatid, tmp2	/* hatid ^ (vaddr >> shift) */	;\
2255	and	tmp2, hmebp, hmebp	/* index into hme_hash */	;\
2256	mulx	hmebp, HMEBUCK_SIZE, hmebp				;\
2257	add	hmebp, tmp1, hmebp
2258
2259/*
2260 * hashtag includes bspage + hashno (64 bits).
2261 */
2262
2263#define	MAKE_HASHTAG(vapg, hatid, hmeshift, hashno, hblktag)		\
2264	sllx	vapg, hmeshift, vapg					;\
2265	mov	hashno, hblktag						;\
2266	sllx	hblktag, HTAG_REHASH_SHIFT, hblktag			;\
2267	or	vapg, hblktag, hblktag
2268
2269/*
2270 * Function to traverse hmeblk hash link list and find corresponding match.
2271 * The search is done using physical pointers. It returns the physical address
2272 * and virtual address pointers to the hmeblk that matches with the tag
2273 * provided.
2274 * Parameters:
2275 * hmebp	= register that points to hme hash bucket, also used as
2276 *		  tmp reg (clobbered)
2277 * hmeblktag	= register with hmeblk tag match
2278 * hatid	= register with hatid
2279 * hmeblkpa	= register where physical ptr will be stored
2280 * hmeblkva	= register where virtual ptr will be stored
2281 * tmp1		= tmp reg
2282 * label: temporary label
2283 */
2284
2285#define	HMEHASH_SEARCH(hmebp, hmeblktag, hatid, hmeblkpa, hmeblkva,	\
2286	tsbarea, tmp1, label)					 	\
2287	add     hmebp, HMEBUCK_NEXTPA, hmeblkpa				;\
2288	ldxa    [hmeblkpa]ASI_MEM, hmeblkpa				;\
2289	add     hmebp, HMEBUCK_HBLK, hmeblkva				;\
2290	ldxa    [hmeblkva]ASI_MEM, hmeblkva				;\
2291	HAT_HSEARCH_DBSTAT(hatid, tsbarea, hmebp, tmp1)			;\
2292label/**/1:								;\
2293	brz,pn	hmeblkva, label/**/2					;\
2294	HAT_HLINK_DBSTAT(hatid, tsbarea, hmebp, tmp1)			;\
2295	add	hmeblkpa, HMEBLK_TAG, hmebp				;\
2296	ldxa	[hmebp]ASI_MEM, tmp1	 /* read 1st part of tag */	;\
2297	add	hmebp, CLONGSIZE, hmebp					;\
2298	ldxa	[hmebp]ASI_MEM, hmebp 	/* read 2nd part of tag */	;\
2299	xor	tmp1, hmeblktag, tmp1					;\
2300	xor	hmebp, hatid, hmebp					;\
2301	or	hmebp, tmp1, hmebp					;\
2302	brz,pn	hmebp, label/**/2	/* branch on hit */		;\
2303	  add	hmeblkpa, HMEBLK_NEXT, hmebp				;\
2304	ldna	[hmebp]ASI_MEM, hmeblkva	/* hmeblk ptr va */	;\
2305	add	hmeblkpa, HMEBLK_NEXTPA, hmebp				;\
2306	ba,pt	%xcc, label/**/1					;\
2307	  ldxa	[hmebp]ASI_MEM, hmeblkpa	/* hmeblk ptr pa */	;\
2308label/**/2:
2309
2310/*
2311 * Function to traverse hmeblk hash link list and find corresponding match.
2312 * The search is done using physical pointers. It returns the physical address
2313 * and virtual address pointers to the hmeblk that matches with the tag
2314 * provided.
2315 * Parameters:
2316 * hmeblktag	= register with hmeblk tag match (rid field is 0)
2317 * hatid	= register with hatid (pointer to SRD)
2318 * hmeblkpa	= register where physical ptr will be stored
2319 * hmeblkva	= register where virtual ptr will be stored
2320 * tmp1		= tmp reg
2321 * tmp2		= tmp reg
2322 * label: temporary label
2323 */
2324
2325#define	HMEHASH_SEARCH_SHME(hmeblktag, hatid, hmeblkpa, hmeblkva,	\
2326	tsbarea, tmp1, tmp2, label)			 		\
2327label/**/1:								;\
2328	brz,pn	hmeblkva, label/**/4					;\
2329	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)			;\
2330	add	hmeblkpa, HMEBLK_TAG, tmp2				;\
2331	ldxa	[tmp2]ASI_MEM, tmp1	 /* read 1st part of tag */	;\
2332	add	tmp2, CLONGSIZE, tmp2					;\
2333	ldxa	[tmp2]ASI_MEM, tmp2 	/* read 2nd part of tag */	;\
2334	xor	tmp1, hmeblktag, tmp1					;\
2335	xor	tmp2, hatid, tmp2					;\
2336	brz,pn	tmp2, label/**/3	/* branch on hit */		;\
2337	  add	hmeblkpa, HMEBLK_NEXT, tmp2				;\
2338label/**/2:								;\
2339	ldna	[tmp2]ASI_MEM, hmeblkva	/* hmeblk ptr va */		;\
2340	add	hmeblkpa, HMEBLK_NEXTPA, tmp2				;\
2341	ba,pt	%xcc, label/**/1					;\
2342	  ldxa	[tmp2]ASI_MEM, hmeblkpa	/* hmeblk ptr pa */		;\
2343label/**/3:								;\
2344	cmp	tmp1, SFMMU_MAX_HME_REGIONS				;\
2345	bgeu,pt	%xcc, label/**/2					;\
2346	  add	hmeblkpa, HMEBLK_NEXT, tmp2				;\
2347	and	tmp1, BT_ULMASK, tmp2					;\
2348	srlx	tmp1, BT_ULSHIFT, tmp1					;\
2349	sllx	tmp1, CLONGSHIFT, tmp1					;\
2350	add	tsbarea, tmp1, tmp1					;\
2351	ldx	[tmp1 + TSBMISS_SHMERMAP], tmp1				;\
2352	srlx	tmp1, tmp2, tmp1					;\
2353	btst	0x1, tmp1						;\
2354	bz,pn	%xcc, label/**/2					;\
2355	  add	hmeblkpa, HMEBLK_NEXT, tmp2				;\
2356label/**/4:
2357
2358#if ((1 << SFHME_SHIFT) != SFHME_SIZE)
2359#error HMEBLK_TO_HMENT assumes sf_hment is power of 2 in size
2360#endif
2361
2362/*
2363 * HMEBLK_TO_HMENT is a macro that given an hmeblk and a vaddr returns
2364 * he offset for the corresponding hment.
2365 * Parameters:
2366 * In:
2367 *	vaddr = register with virtual address
2368 *	hmeblkpa = physical pointer to hme_blk
2369 * Out:
2370 *	hmentoff = register where hment offset will be stored
2371 *	hmemisc = hblk_misc
2372 * Scratch:
2373 *	tmp1
2374 */
2375#define	HMEBLK_TO_HMENT(vaddr, hmeblkpa, hmentoff, hmemisc, tmp1, label1)\
2376	add	hmeblkpa, HMEBLK_MISC, hmentoff				;\
2377	lda	[hmentoff]ASI_MEM, hmemisc 				;\
2378	andcc	hmemisc, HBLK_SZMASK, %g0				;\
2379	bnz,a,pn  %icc, label1		/* if sz != TTE8K branch */	;\
2380	  or	%g0, HMEBLK_HME1, hmentoff				;\
2381	srl	vaddr, MMU_PAGESHIFT, tmp1				;\
2382	and	tmp1, NHMENTS - 1, tmp1		/* tmp1 = index */	;\
2383	sllx	tmp1, SFHME_SHIFT, tmp1					;\
2384	add	tmp1, HMEBLK_HME1, hmentoff				;\
2385label1:
2386
2387/*
2388 * GET_TTE is a macro that returns a TTE given a tag and hatid.
2389 *
2390 * tagacc	= (pseudo-)tag access register (in)
2391 * hatid	= sfmmu pointer for TSB miss (in)
2392 * tte		= tte for TLB miss if found, otherwise clobbered (out)
2393 * hmeblkpa	= PA of hment if found, otherwise clobbered (out)
2394 * hmeblkva	= VA of hment if found, otherwise clobbered (out)
2395 * tsbarea	= pointer to the tsbmiss area for this cpu. (in)
2396 * hmemisc	= hblk_misc if TTE is found (out), otherwise clobbered
2397 * hmeshift	= constant/register to shift VA to obtain the virtual pfn
2398 *		  for this page size.
2399 * hashno	= constant/register hash number
2400 * label	= temporary label for branching within macro.
2401 * foundlabel	= label to jump to when tte is found.
2402 * suspendlabel= label to jump to when tte is suspended.
2403 * exitlabel	= label to jump to when tte is not found.
2404 *
2405 */
2406#define GET_TTE(tagacc, hatid, tte, hmeblkpa, hmeblkva, tsbarea, hmemisc, \
2407		hmeshift, hashno, label, foundlabel, suspendlabel, exitlabel) \
2408									;\
2409	stn	tagacc, [tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)]	;\
2410	stn	hatid, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)]	;\
2411	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, tte,		\
2412		hmeblkpa, label/**/5, hmemisc, hmeblkva)		;\
2413									;\
2414	/*								;\
2415	 * tagacc = tagacc						;\
2416	 * hatid = hatid						;\
2417	 * tsbarea = tsbarea						;\
2418	 * tte   = hmebp (hme bucket pointer)				;\
2419	 * hmeblkpa  = vapg  (virtual page)				;\
2420	 * hmemisc, hmeblkva = scratch					;\
2421	 */								;\
2422	MAKE_HASHTAG(hmeblkpa, hatid, hmeshift, hashno, hmemisc)	;\
2423	or	hmemisc, SFMMU_INVALID_SHMERID, hmemisc			;\
2424									;\
2425	/*								;\
2426	 * tagacc = tagacc						;\
2427	 * hatid = hatid						;\
2428	 * tte   = hmebp						;\
2429	 * hmeblkpa  = CLOBBERED					;\
2430	 * hmemisc  = htag_bspage+hashno+invalid_rid			;\
2431	 * hmeblkva  = scratch						;\
2432	 */								;\
2433	stn	tte, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)]	;\
2434	HMELOCK_ENTER(tte, hmeblkpa, hmeblkva, label/**/3, ASI_MEM)	;\
2435	HMEHASH_SEARCH(tte, hmemisc, hatid, hmeblkpa, hmeblkva, 	\
2436		tsbarea, tagacc, label/**/1)				;\
2437	/*								;\
2438	 * tagacc = CLOBBERED						;\
2439	 * tte = CLOBBERED						;\
2440	 * hmeblkpa = hmeblkpa						;\
2441	 * hmeblkva = hmeblkva						;\
2442	 */								;\
2443	brnz,pt	hmeblkva, label/**/4	/* branch if hmeblk found */	;\
2444	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)], tagacc	;\
2445	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hmeblkva	;\
2446	HMELOCK_EXIT(hmeblkva, hmeblkva, ASI_MEM)  /* drop lock */	;\
2447	ba,pt	%xcc, exitlabel		/* exit if hblk not found */	;\
2448	  nop								;\
2449label/**/4:								;\
2450	/*								;\
2451	 * We have found the hmeblk containing the hment.		;\
2452	 * Now we calculate the corresponding tte.			;\
2453	 *								;\
2454	 * tagacc = tagacc						;\
2455	 * hatid = hatid						;\
2456	 * tte   = clobbered						;\
2457	 * hmeblkpa  = hmeblkpa						;\
2458	 * hmemisc  = hblktag						;\
2459	 * hmeblkva  = hmeblkva 					;\
2460	 */								;\
2461	HMEBLK_TO_HMENT(tagacc, hmeblkpa, hatid, hmemisc, tte,		\
2462		label/**/2)						;\
2463									;\
2464	/*								;\
2465	 * tagacc = tagacc						;\
2466	 * hatid = hmentoff						;\
2467	 * tte   = clobbered						;\
2468	 * hmeblkpa  = hmeblkpa						;\
2469	 * hmemisc  = hblk_misc						;\
2470	 * hmeblkva  = hmeblkva 					;\
2471	 */								;\
2472									;\
2473	add	hatid, SFHME_TTE, hatid					;\
2474	add	hmeblkpa, hatid, hmeblkpa				;\
2475	ldxa	[hmeblkpa]ASI_MEM, tte	/* MMU_READTTE through pa */	;\
2476	add	hmeblkva, hatid, hmeblkva				;\
2477	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2478	HMELOCK_EXIT(hatid, hatid, ASI_MEM)	/* drop lock */		;\
2479	set	TTE_SUSPEND, hatid					;\
2480	TTE_SUSPEND_INT_SHIFT(hatid)					;\
2481	btst	tte, hatid						;\
2482	bz,pt	%xcc, foundlabel					;\
2483	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2484									;\
2485	/*								;\
2486	 * Mapping is suspended, so goto suspend label.			;\
2487	 */								;\
2488	ba,pt	%xcc, suspendlabel					;\
2489	  nop
2490
2491/*
2492 * GET_SHME_TTE is similar to GET_TTE() except it searches
2493 * shared hmeblks via HMEHASH_SEARCH_SHME() macro.
2494 * If valid tte is found, hmemisc = shctx flag, i.e., shme is
2495 * either 0 (not part of scd) or 1 (part of scd).
2496 */
2497#define GET_SHME_TTE(tagacc, hatid, tte, hmeblkpa, hmeblkva, tsbarea,	\
2498		hmemisc, hmeshift, hashno, label, foundlabel,		\
2499		suspendlabel, exitlabel)				\
2500									;\
2501	stn	tagacc, [tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)]	;\
2502	stn	hatid, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)]	;\
2503	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, tte,		\
2504		hmeblkpa, label/**/5, hmemisc, hmeblkva)		;\
2505									;\
2506	/*								;\
2507	 * tagacc = tagacc						;\
2508	 * hatid = hatid						;\
2509	 * tsbarea = tsbarea						;\
2510	 * tte   = hmebp (hme bucket pointer)				;\
2511	 * hmeblkpa  = vapg  (virtual page)				;\
2512	 * hmemisc, hmeblkva = scratch					;\
2513	 */								;\
2514	MAKE_HASHTAG(hmeblkpa, hatid, hmeshift, hashno, hmemisc)	;\
2515									;\
2516	/*								;\
2517	 * tagacc = tagacc						;\
2518	 * hatid = hatid						;\
2519	 * tsbarea = tsbarea						;\
2520	 * tte   = hmebp						;\
2521	 * hmemisc  = htag_bspage + hashno + 0 (for rid)		;\
2522	 * hmeblkpa  = CLOBBERED					;\
2523	 * hmeblkva  = scratch						;\
2524	 */								;\
2525	stn	tte, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)]	;\
2526	HMELOCK_ENTER(tte, hmeblkpa, hmeblkva, label/**/3, ASI_MEM)	;\
2527									;\
2528	add     tte, HMEBUCK_NEXTPA, hmeblkpa				;\
2529	ldxa    [hmeblkpa]ASI_MEM, hmeblkpa				;\
2530	add     tte, HMEBUCK_HBLK, hmeblkva				;\
2531	ldxa    [hmeblkva]ASI_MEM, hmeblkva				;\
2532	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tagacc, tte)			;\
2533									;\
2534label/**/8:								;\
2535	HMEHASH_SEARCH_SHME(hmemisc, hatid, hmeblkpa, hmeblkva, 	\
2536		tsbarea, tagacc, tte, label/**/1)			;\
2537	/*								;\
2538	 * tagacc = CLOBBERED						;\
2539	 * tte = CLOBBERED						;\
2540	 * hmeblkpa = hmeblkpa						;\
2541	 * hmeblkva = hmeblkva						;\
2542	 */								;\
2543	brnz,pt	hmeblkva, label/**/4	/* branch if hmeblk found */	;\
2544	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)], tagacc	;\
2545	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hmeblkva	;\
2546	HMELOCK_EXIT(hmeblkva, hmeblkva, ASI_MEM)  /* drop lock */	;\
2547	ba,pt	%xcc, exitlabel		/* exit if hblk not found */	;\
2548	  nop								;\
2549label/**/4:								;\
2550	/*								;\
2551	 * We have found the hmeblk containing the hment.		;\
2552	 * Now we calculate the corresponding tte.			;\
2553	 *								;\
2554	 * tagacc = tagacc						;\
2555	 * hatid = hatid						;\
2556	 * tte   = clobbered						;\
2557	 * hmeblkpa  = hmeblkpa						;\
2558	 * hmemisc  = hblktag						;\
2559	 * hmeblkva  = hmeblkva 					;\
2560	 * tsbarea = tsbmiss area					;\
2561	 */								;\
2562	HMEBLK_TO_HMENT(tagacc, hmeblkpa, hatid, hmemisc, tte,		\
2563		label/**/2)						;\
2564									;\
2565	/*								;\
2566	 * tagacc = tagacc						;\
2567	 * hatid = hmentoff						;\
2568	 * tte = clobbered						;\
2569	 * hmeblkpa  = hmeblkpa						;\
2570	 * hmemisc  = hblk_misc						;\
2571	 * hmeblkva  = hmeblkva						;\
2572	 * tsbarea = tsbmiss area					;\
2573	 */								;\
2574									;\
2575	add	hatid, SFHME_TTE, hatid					;\
2576	add	hmeblkpa, hatid, hmeblkpa				;\
2577	ldxa	[hmeblkpa]ASI_MEM, tte	/* MMU_READTTE through pa */	;\
2578	brlz,pt tte, label/**/6						;\
2579	  add	hmeblkva, hatid, hmeblkva				;\
2580	btst	HBLK_SZMASK, hmemisc					;\
2581	bnz,a,pt %icc, label/**/7					;\
2582	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2583									;\
2584	/*								;\
2585 	 * We found an invalid 8K tte in shme.				;\
2586	 * it may not belong to shme's region since			;\
2587	 * region size/alignment granularity is 8K but different	;\
2588	 * regions don't share hmeblks. Continue the search.		;\
2589	 */								;\
2590	sub	hmeblkpa, hatid, hmeblkpa				;\
2591	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2592	srlx	tagacc, hmeshift, tte					;\
2593	add	hmeblkpa, HMEBLK_NEXT, hmeblkva				;\
2594	ldxa	[hmeblkva]ASI_MEM, hmeblkva				;\
2595	add	hmeblkpa, HMEBLK_NEXTPA, hmeblkpa			;\
2596	ldxa	[hmeblkpa]ASI_MEM, hmeblkpa				;\
2597	MAKE_HASHTAG(tte, hatid, hmeshift, hashno, hmemisc)		;\
2598	ba,a,pt	%xcc, label/**/8					;\
2599label/**/6:								;\
2600	GET_SCDSHMERMAP(tsbarea, hmeblkpa, hatid, hmemisc)		;\
2601	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2602label/**/7:								;\
2603	HMELOCK_EXIT(hatid, hatid, ASI_MEM)	/* drop lock */		;\
2604	set	TTE_SUSPEND, hatid					;\
2605	TTE_SUSPEND_INT_SHIFT(hatid)					;\
2606	btst	tte, hatid						;\
2607	bz,pt	%xcc, foundlabel					;\
2608	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2609									;\
2610	/*								;\
2611	 * Mapping is suspended, so goto suspend label.			;\
2612	 */								;\
2613	ba,pt	%xcc, suspendlabel					;\
2614	  nop
2615
2616	/*
2617	 * KERNEL PROTECTION HANDLER
2618	 *
2619	 * g1 = tsb8k pointer register (clobbered)
2620	 * g2 = tag access register (ro)
2621	 * g3 - g7 = scratch registers
2622	 *
2623	 * Note: This function is patched at runtime for performance reasons.
2624	 * 	 Any changes here require sfmmu_patch_ktsb fixed.
2625	 */
2626	ENTRY_NP(sfmmu_kprot_trap)
2627	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
2628sfmmu_kprot_patch_ktsb_base:
2629	RUNTIME_PATCH_SETX(%g1, %g6)
2630	/* %g1 = contents of ktsb_base or ktsb_pbase */
2631sfmmu_kprot_patch_ktsb_szcode:
2632	or	%g0, RUNTIME_PATCH, %g3	! ktsb_szcode (hot patched)
2633
2634	GET_TSBE_POINTER(MMU_PAGESHIFT, %g1, %g7, %g3, %g5)
2635	! %g1 = First TSB entry pointer, as TSB miss handler expects
2636
2637	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
2638sfmmu_kprot_patch_ktsb4m_base:
2639	RUNTIME_PATCH_SETX(%g3, %g6)
2640	/* %g3 = contents of ktsb4m_base or ktsb4m_pbase */
2641sfmmu_kprot_patch_ktsb4m_szcode:
2642	or	%g0, RUNTIME_PATCH, %g6	! ktsb4m_szcode (hot patched)
2643
2644	GET_TSBE_POINTER(MMU_PAGESHIFT4M, %g3, %g7, %g6, %g5)
2645	! %g3 = 4M tsb entry pointer, as TSB miss handler expects
2646
2647        CPU_TSBMISS_AREA(%g6, %g7)
2648        HAT_PERCPU_STAT16(%g6, TSBMISS_KPROTS, %g7)
2649	ba,pt	%xcc, sfmmu_tsb_miss_tt
2650	  nop
2651
2652	/*
2653	 * USER PROTECTION HANDLER
2654	 *
2655	 * g1 = tsb8k pointer register (ro)
2656	 * g2 = tag access register (ro)
2657	 * g3 = faulting context (clobbered, currently not used)
2658	 * g4 - g7 = scratch registers
2659	 */
2660	ALTENTRY(sfmmu_uprot_trap)
2661#ifdef sun4v
2662	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
2663	/* %g1 = first TSB entry ptr now, %g2 preserved */
2664
2665	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)	/* get 2nd utsbreg */
2666	brlz,pt %g3, 9f				/* check for 2nd TSB */
2667	  nop
2668
2669	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2670	/* %g3 = second TSB entry ptr now, %g2 preserved */
2671
2672#else /* sun4v */
2673#ifdef UTSB_PHYS
2674	/* g1 = first TSB entry ptr */
2675	GET_2ND_TSBREG(%g3)
2676	brlz,pt %g3, 9f			/* check for 2nd TSB */
2677	  nop
2678
2679	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2680	/* %g3 = second TSB entry ptr now, %g2 preserved */
2681#else /* UTSB_PHYS */
2682	brgez,pt %g1, 9f		/* check for 2nd TSB */
2683	  mov	-1, %g3			/* set second tsbe ptr to -1 */
2684
2685	mov	%g2, %g7
2686	GET_2ND_TSBE_PTR(%g7, %g1, %g3, %g4, %g5, sfmmu_uprot)
2687	/* %g3 = second TSB entry ptr now, %g7 clobbered */
2688	mov	%g1, %g7
2689	GET_1ST_TSBE_PTR(%g7, %g1, %g5, sfmmu_uprot)
2690#endif /* UTSB_PHYS */
2691#endif /* sun4v */
26929:
2693	CPU_TSBMISS_AREA(%g6, %g7)
2694	HAT_PERCPU_STAT16(%g6, TSBMISS_UPROTS, %g7)
2695	ba,pt	%xcc, sfmmu_tsb_miss_tt		/* branch TSB miss handler */
2696	  nop
2697
2698	/*
2699	 * Kernel 8K page iTLB miss.  We also get here if we took a
2700	 * fast instruction access mmu miss trap while running in
2701	 * invalid context.
2702	 *
2703	 * %g1 = 8K TSB pointer register (not used, clobbered)
2704	 * %g2 = tag access register (used)
2705	 * %g3 = faulting context id (used)
2706	 * %g7 = TSB tag to match (used)
2707	 */
2708	.align	64
2709	ALTENTRY(sfmmu_kitlb_miss)
2710	brnz,pn %g3, tsb_tl0_noctxt
2711	  nop
2712
2713	/* kernel miss */
2714	/* get kernel tsb pointer */
2715	/* we patch the next set of instructions at run time */
2716	/* NOTE: any changes here require sfmmu_patch_ktsb fixed */
2717iktsbbase:
2718	RUNTIME_PATCH_SETX(%g4, %g5)
2719	/* %g4 = contents of ktsb_base or ktsb_pbase */
2720
2721iktsb:	sllx	%g2, 64-(TAGACC_SHIFT + TSB_START_SIZE + RUNTIME_PATCH), %g1
2722	srlx	%g1, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g1
2723	or	%g4, %g1, %g1			! form tsb ptr
2724	ldda	[%g1]RUNTIME_PATCH, %g4		! %g4 = tag, %g5 = data
2725	cmp	%g4, %g7
2726	bne,pn	%xcc, iktsb4mbase		! check 4m ktsb
2727	  srlx    %g2, MMU_PAGESHIFT4M, %g3	! use 4m virt-page as TSB index
2728
2729	andcc %g5, TTE_EXECPRM_INT, %g0		! check exec bit
2730	bz,pn	%icc, exec_fault
2731	  nop
2732	TT_TRACE(trace_tsbhit)			! 2 instr traptrace
2733	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2734	retry
2735
2736iktsb4mbase:
2737        RUNTIME_PATCH_SETX(%g4, %g6)
2738        /* %g4 = contents of ktsb4m_base or ktsb4m_pbase */
2739iktsb4m:
2740	sllx    %g3, 64-(TSB_START_SIZE + RUNTIME_PATCH), %g3
2741        srlx    %g3, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g3
2742	add	%g4, %g3, %g3			! %g3 = 4m tsbe ptr
2743	ldda	[%g3]RUNTIME_PATCH, %g4		! %g4 = tag, %g5 = data
2744	cmp	%g4, %g7
2745	bne,pn	%xcc, sfmmu_tsb_miss_tt		! branch on miss
2746	  andcc %g5, TTE_EXECPRM_INT, %g0		! check exec bit
2747	bz,pn	%icc, exec_fault
2748	  nop
2749	TT_TRACE(trace_tsbhit)			! 2 instr traptrace
2750	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2751	retry
2752
2753	/*
2754	 * Kernel dTLB miss.  We also get here if we took a fast data
2755	 * access mmu miss trap while running in invalid context.
2756	 *
2757	 * Note: for now we store kpm TTEs in the kernel TSB as usual.
2758	 *	We select the TSB miss handler to branch to depending on
2759	 *	the virtual address of the access.  In the future it may
2760	 *	be desirable to separate kpm TTEs into their own TSB,
2761	 *	in which case all that needs to be done is to set
2762	 *	kpm_tsbbase/kpm_tsbsz to point to the new TSB and branch
2763	 *	early in the miss if we detect a kpm VA to a new handler.
2764	 *
2765	 * %g1 = 8K TSB pointer register (not used, clobbered)
2766	 * %g2 = tag access register (used)
2767	 * %g3 = faulting context id (used)
2768	 */
2769	.align	64
2770	ALTENTRY(sfmmu_kdtlb_miss)
2771	brnz,pn	%g3, tsb_tl0_noctxt		/* invalid context? */
2772	  nop
2773
2774	/* Gather some stats for kpm misses in the TLB. */
2775	/* KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label) */
2776	KPM_TLBMISS_STAT_INCR(%g2, %g4, %g5, %g6, kpmtlbm_stat_out)
2777
2778	/*
2779	 * Get first TSB offset and look for 8K/64K/512K mapping
2780	 * using the 8K virtual page as the index.
2781	 *
2782	 * We patch the next set of instructions at run time;
2783	 * any changes here require sfmmu_patch_ktsb changes too.
2784	 */
2785dktsbbase:
2786	RUNTIME_PATCH_SETX(%g7, %g6)
2787	/* %g7 = contents of ktsb_base or ktsb_pbase */
2788
2789dktsb:	sllx	%g2, 64-(TAGACC_SHIFT + TSB_START_SIZE + RUNTIME_PATCH), %g1
2790	srlx	%g1, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g1
2791
2792	/*
2793	 * At this point %g1 is our index into the TSB.
2794	 * We just masked off enough bits of the VA depending
2795	 * on our TSB size code.
2796	 */
2797	ldda	[%g7 + %g1]RUNTIME_PATCH, %g4	! %g4 = tag, %g5 = data
2798	srlx	%g2, TAG_VALO_SHIFT, %g6	! make tag to compare
2799	cmp	%g6, %g4			! compare tag
2800	bne,pn	%xcc, dktsb4m_kpmcheck_small
2801	  add	%g7, %g1, %g1			/* form tsb ptr */
2802	TT_TRACE(trace_tsbhit)
2803	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2804	/* trapstat expects tte in %g5 */
2805	retry
2806
2807	/*
2808	 * If kpm is using large pages, the following instruction needs
2809	 * to be patched to a nop at boot time (by sfmmu_kpm_patch_tsbm)
2810	 * so that we will probe the 4M TSB regardless of the VA.  In
2811	 * the case kpm is using small pages, we know no large kernel
2812	 * mappings are located above 0x80000000.00000000 so we skip the
2813	 * probe as an optimization.
2814	 */
2815dktsb4m_kpmcheck_small:
2816	brlz,pn %g2, sfmmu_kpm_dtsb_miss_small
2817	  /* delay slot safe, below */
2818
2819	/*
2820	 * Get second TSB offset and look for 4M mapping
2821	 * using 4M virtual page as the TSB index.
2822	 *
2823	 * Here:
2824	 * %g1 = 8K TSB pointer.  Don't squash it.
2825	 * %g2 = tag access register (we still need it)
2826	 */
2827	srlx	%g2, MMU_PAGESHIFT4M, %g3
2828
2829	/*
2830	 * We patch the next set of instructions at run time;
2831	 * any changes here require sfmmu_patch_ktsb changes too.
2832	 */
2833dktsb4mbase:
2834	RUNTIME_PATCH_SETX(%g7, %g6)
2835	/* %g7 = contents of ktsb4m_base or ktsb4m_pbase */
2836dktsb4m:
2837	sllx	%g3, 64-(TSB_START_SIZE + RUNTIME_PATCH), %g3
2838	srlx	%g3, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g3
2839
2840	/*
2841	 * At this point %g3 is our index into the TSB.
2842	 * We just masked off enough bits of the VA depending
2843	 * on our TSB size code.
2844	 */
2845	ldda	[%g7 + %g3]RUNTIME_PATCH, %g4	! %g4 = tag, %g5 = data
2846	srlx	%g2, TAG_VALO_SHIFT, %g6	! make tag to compare
2847	cmp	%g6, %g4			! compare tag
2848
2849dktsb4m_tsbmiss:
2850	bne,pn	%xcc, dktsb4m_kpmcheck
2851	  add	%g7, %g3, %g3			! %g3 = kernel second TSB ptr
2852	TT_TRACE(trace_tsbhit)
2853	/* we don't check TTE size here since we assume 4M TSB is separate */
2854	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2855	/* trapstat expects tte in %g5 */
2856	retry
2857
2858	/*
2859	 * So, we failed to find a valid TTE to match the faulting
2860	 * address in either TSB.  There are a few cases that could land
2861	 * us here:
2862	 *
2863	 * 1) This is a kernel VA below 0x80000000.00000000.  We branch
2864	 *    to sfmmu_tsb_miss_tt to handle the miss.
2865	 * 2) We missed on a kpm VA, and we didn't find the mapping in the
2866	 *    4M TSB.  Let segkpm handle it.
2867	 *
2868	 * Note that we shouldn't land here in the case of a kpm VA when
2869	 * kpm_smallpages is active -- we handled that case earlier at
2870	 * dktsb4m_kpmcheck_small.
2871	 *
2872	 * At this point:
2873	 *  g1 = 8K-indexed primary TSB pointer
2874	 *  g2 = tag access register
2875	 *  g3 = 4M-indexed secondary TSB pointer
2876	 */
2877dktsb4m_kpmcheck:
2878	cmp	%g2, %g0
2879	bl,pn	%xcc, sfmmu_kpm_dtsb_miss
2880	  nop
2881	ba,a,pt	%icc, sfmmu_tsb_miss_tt
2882	  nop
2883
2884#ifdef sun4v
2885	/*
2886	 * User instruction miss w/ single TSB.
2887	 * The first probe covers 8K, 64K, and 512K page sizes,
2888	 * because 64K and 512K mappings are replicated off 8K
2889	 * pointer.
2890	 *
2891	 * g1 = tsb8k pointer register
2892	 * g2 = tag access register
2893	 * g3 - g6 = scratch registers
2894	 * g7 = TSB tag to match
2895	 */
2896	.align	64
2897	ALTENTRY(sfmmu_uitlb_fastpath)
2898
2899	PROBE_1ST_ITSB(%g1, %g7, uitlb_fast_8k_probefail)
2900	/* g4 - g5 = clobbered by PROBE_1ST_ITSB */
2901	ba,pn	%xcc, sfmmu_tsb_miss_tt
2902	  mov	-1, %g3
2903
2904	/*
2905	 * User data miss w/ single TSB.
2906	 * The first probe covers 8K, 64K, and 512K page sizes,
2907	 * because 64K and 512K mappings are replicated off 8K
2908	 * pointer.
2909	 *
2910	 * g1 = tsb8k pointer register
2911	 * g2 = tag access register
2912	 * g3 - g6 = scratch registers
2913	 * g7 = TSB tag to match
2914	 */
2915	.align 64
2916	ALTENTRY(sfmmu_udtlb_fastpath)
2917
2918	PROBE_1ST_DTSB(%g1, %g7, udtlb_fast_8k_probefail)
2919	/* g4 - g5 = clobbered by PROBE_1ST_DTSB */
2920	ba,pn	%xcc, sfmmu_tsb_miss_tt
2921	  mov	-1, %g3
2922
2923	/*
2924	 * User instruction miss w/ multiple TSBs (sun4v).
2925	 * The first probe covers 8K, 64K, and 512K page sizes,
2926	 * because 64K and 512K mappings are replicated off 8K
2927	 * pointer.  Second probe covers 4M page size only.
2928	 *
2929	 * Just like sfmmu_udtlb_slowpath, except:
2930	 *   o Uses ASI_ITLB_IN
2931	 *   o checks for execute permission
2932	 *   o No ISM prediction.
2933	 *
2934	 * g1 = tsb8k pointer register
2935	 * g2 = tag access register
2936	 * g3 - g6 = scratch registers
2937	 * g7 = TSB tag to match
2938	 */
2939	.align	64
2940	ALTENTRY(sfmmu_uitlb_slowpath)
2941
2942	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
2943	PROBE_1ST_ITSB(%g1, %g7, uitlb_8k_probefail)
2944	/* g4 - g5 = clobbered here */
2945
2946	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2947	/* g1 = first TSB pointer, g3 = second TSB pointer */
2948	srlx	%g2, TAG_VALO_SHIFT, %g7
2949	PROBE_2ND_ITSB(%g3, %g7)
2950	/* NOT REACHED */
2951
2952#else /* sun4v */
2953
2954	/*
2955	 * User instruction miss w/ multiple TSBs (sun4u).
2956	 * The first probe covers 8K, 64K, and 512K page sizes,
2957	 * because 64K and 512K mappings are replicated off 8K
2958	 * pointer.  Second probe covers 4M page size only.
2959	 *
2960	 * Just like sfmmu_udtlb_slowpath, except:
2961	 *   o Uses ASI_ITLB_IN
2962	 *   o checks for execute permission
2963	 *   o No ISM prediction.
2964	 *
2965	 * g1 = tsb8k pointer register
2966	 * g2 = tag access register
2967	 * g3 = 2nd tsbreg if defined UTSB_PHYS, else scratch
2968	 * g4 - g6 = scratch registers
2969	 * g7 = TSB tag to match
2970	 */
2971	.align	64
2972	ALTENTRY(sfmmu_uitlb_slowpath)
2973
2974#ifdef UTSB_PHYS
2975	/*
2976	 * g1 = 1st TSB entry pointer
2977	 * g3 = 2nd TSB base register
2978	 * Need 2nd TSB entry pointer for 2nd probe.
2979	 */
2980	PROBE_1ST_ITSB(%g1, %g7, uitlb_8k_probefail)
2981
2982	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2983#else /* UTSB_PHYS */
2984	mov	%g1, %g3	/* save tsb8k reg in %g3 */
2985	GET_1ST_TSBE_PTR(%g3, %g1, %g5, sfmmu_uitlb)
2986	PROBE_1ST_ITSB(%g1, %g7, uitlb_8k_probefail)
2987
2988	mov	%g2, %g6	/* GET_2ND_TSBE_PTR clobbers tagacc */
2989	mov	%g3, %g7	/* copy tsb8k reg in %g7 */
2990	GET_2ND_TSBE_PTR(%g6, %g7, %g3, %g4, %g5, sfmmu_uitlb)
2991#endif /* UTSB_PHYS */
2992	/* g1 = first TSB pointer, g3 = second TSB pointer */
2993	srlx	%g2, TAG_VALO_SHIFT, %g7
2994	PROBE_2ND_ITSB(%g3, %g7, isynth)
2995	/* NOT REACHED */
2996#endif /* sun4v */
2997
2998	/*
2999	 * User data miss w/ multiple TSBs.
3000	 * The first probe covers 8K, 64K, and 512K page sizes,
3001	 * because 64K and 512K mappings are replicated off 8K
3002	 * pointer.  Second probe covers 4M page size only.
3003	 *
3004	 * We consider probing for 4M pages first if the VA falls
3005	 * in a range that's likely to be ISM.
3006	 *
3007	 * g1 = tsb8k pointer register
3008	 * g2 = tag access register
3009	 * g3 = 2nd tsbreg if defined UTSB_PHYS, else scratch
3010	 * g4 - g6 = scratch registers
3011	 * g7 = TSB tag to match
3012	 */
3013	.align 64
3014	ALTENTRY(sfmmu_udtlb_slowpath)
3015
3016	/*
3017	 * Check for ISM.  If it exists, look for 4M mappings in the second TSB
3018	 * first, then probe for other mappings in the first TSB if that fails.
3019	 */
3020	srax	%g2, PREDISM_BASESHIFT, %g6	/* g6 > 0 : ISM predicted */
3021	brgz,pn %g6, udtlb_miss_probesecond	/* check for ISM */
3022	  mov	%g1, %g3
3023
3024udtlb_miss_probefirst:
3025	/*
3026	 * g1 = 8K TSB pointer register
3027	 * g2 = tag access register
3028	 * g3 = (potentially) second TSB entry ptr
3029	 * g6 = ism pred.
3030	 * g7 = vpg_4m
3031	 */
3032#ifdef sun4v
3033	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
3034	PROBE_1ST_DTSB(%g1, %g7, udtlb_first_probefail)
3035
3036	/*
3037	 * Here:
3038	 *   g1 = first TSB pointer
3039	 *   g2 = tag access reg
3040	 *   g3 = second TSB ptr IFF ISM pred. (else don't care)
3041	 */
3042	brgz,pn	%g6, sfmmu_tsb_miss_tt
3043	  nop
3044#else /* sun4v */
3045#ifndef UTSB_PHYS
3046	mov	%g1, %g4
3047	GET_1ST_TSBE_PTR(%g4, %g1, %g5, sfmmu_udtlb)
3048#endif UTSB_PHYS
3049	PROBE_1ST_DTSB(%g1, %g7, udtlb_first_probefail)
3050
3051	/*
3052	 * Here:
3053	 *   g1 = first TSB pointer
3054	 *   g2 = tag access reg
3055	 *   g3 = second TSB ptr IFF ISM pred. (else don't care)
3056	 */
3057	brgz,pn	%g6, sfmmu_tsb_miss_tt
3058	  nop
3059#ifndef UTSB_PHYS
3060	ldxa	[%g0]ASI_DMMU_TSB_8K, %g3
3061#endif UTSB_PHYS
3062	/* fall through in 8K->4M probe order */
3063#endif /* sun4v */
3064
3065udtlb_miss_probesecond:
3066	/*
3067	 * Look in the second TSB for the TTE
3068	 * g1 = First TSB entry ptr if !ISM pred, TSB8K ptr reg if ISM pred.
3069	 * g2 = tag access reg
3070	 * g3 = 8K TSB pointer register
3071	 * g6 = ism pred.
3072	 * g7 = vpg_4m
3073	 */
3074#ifdef sun4v
3075	/* GET_2ND_TSBE_PTR(tagacc, tsbe_ptr, tmp1, tmp2) */
3076	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
3077	/* %g2 is okay, no need to reload, %g3 = second tsbe ptr */
3078#else /* sun4v */
3079#ifdef UTSB_PHYS
3080	GET_2ND_TSBREG(%g3)
3081	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
3082	/* tagacc (%g2) is okay, no need to reload, %g3 = second tsbe ptr */
3083#else /* UTSB_PHYS */
3084	mov	%g3, %g7
3085	GET_2ND_TSBE_PTR(%g2, %g7, %g3, %g4, %g5, sfmmu_udtlb)
3086	/* %g2 clobbered, %g3 =second tsbe ptr */
3087	mov	MMU_TAG_ACCESS, %g2
3088	ldxa	[%g2]ASI_DMMU, %g2
3089#endif /* UTSB_PHYS */
3090#endif /* sun4v */
3091
3092	srlx	%g2, TAG_VALO_SHIFT, %g7
3093	PROBE_2ND_DTSB(%g3, %g7, udtlb_4m_probefail)
3094	/* g4 - g5 = clobbered here; %g7 still vpg_4m at this point */
3095	brgz,pn	%g6, udtlb_miss_probefirst
3096	  nop
3097
3098	/* fall through to sfmmu_tsb_miss_tt */
3099
3100	ALTENTRY(sfmmu_tsb_miss_tt)
3101	TT_TRACE(trace_tsbmiss)
3102	/*
3103	 * We get here if there is a TSB miss OR a write protect trap.
3104	 *
3105	 * g1 = First TSB entry pointer
3106	 * g2 = tag access register
3107	 * g3 = 4M TSB entry pointer; -1 if no 2nd TSB
3108	 * g4 - g7 = scratch registers
3109	 */
3110
3111	ALTENTRY(sfmmu_tsb_miss)
3112
3113	/*
3114	 * If trapstat is running, we need to shift the %tpc and %tnpc to
3115	 * point to trapstat's TSB miss return code (note that trapstat
3116	 * itself will patch the correct offset to add).
3117	 */
3118	rdpr	%tl, %g7
3119	cmp	%g7, 1
3120	ble,pt	%xcc, 0f
3121	  sethi	%hi(KERNELBASE), %g6
3122	rdpr	%tpc, %g7
3123	or	%g6, %lo(KERNELBASE), %g6
3124	cmp	%g7, %g6
3125	bgeu,pt	%xcc, 0f
3126	/* delay slot safe */
3127
3128	ALTENTRY(tsbmiss_trapstat_patch_point)
3129	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
3130	wrpr	%g7, %tpc
3131	add	%g7, 4, %g7
3132	wrpr	%g7, %tnpc
31330:
3134	CPU_TSBMISS_AREA(%g6, %g7)
3135
3136	stn	%g1, [%g6 + TSBMISS_TSBPTR]	/* save first tsb pointer */
3137	stn	%g3, [%g6 + TSBMISS_TSBPTR4M]	/* save second tsb pointer */
3138
3139	sllx	%g2, TAGACC_CTX_LSHIFT, %g3
3140	brz,a,pn %g3, 1f			/* skip ahead if kernel */
3141	  ldn	[%g6 + TSBMISS_KHATID], %g7
3142	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctxnum */
3143	ldn	[%g6 + TSBMISS_UHATID], %g7     /* g7 = hatid */
3144
3145	HAT_PERCPU_STAT32(%g6, TSBMISS_UTSBMISS, %g5)
3146
3147	cmp	%g3, INVALID_CONTEXT
3148	be,pn	%icc, tsb_tl0_noctxt		/* no ctx miss exception */
3149	  stn	%g7, [%g6 + (TSBMISS_SCRATCH + TSBMISS_HATID)]
3150
3151#ifdef sun4v
3152        ldub    [%g6 + TSBMISS_URTTEFLAGS], %g7	/* clear ctx1 flag set from */
3153        andn    %g7, HAT_CHKCTX1_FLAG, %g7	/* the previous tsb miss    */
3154        stub    %g7, [%g6 + TSBMISS_URTTEFLAGS]
3155#endif
3156
3157	ISM_CHECK(%g2, %g6, %g3, %g4, %g5, %g7, %g1, tsb_l1, tsb_ism)
3158	/*
3159	 * The miss wasn't in an ISM segment.
3160	 *
3161	 * %g1 %g3, %g4, %g5, %g7 all clobbered
3162	 * %g2 = (pseudo) tag access
3163	 */
3164
3165	ba,pt	%icc, 2f
3166	  ldn	[%g6 + (TSBMISS_SCRATCH + TSBMISS_HATID)], %g7
3167
31681:
3169	HAT_PERCPU_STAT32(%g6, TSBMISS_KTSBMISS, %g5)
3170	/*
3171	 * 8K and 64K hash.
3172	 */
31732:
3174
3175	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3176		MMU_PAGESHIFT64K, TTE64K, tsb_l8K, tsb_checktte,
3177		sfmmu_suspend_tl, tsb_512K)
3178	/* NOT REACHED */
3179
3180tsb_512K:
3181	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3182	brz,pn	%g5, 3f
3183	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3184	and	%g4, HAT_512K_FLAG, %g5
3185
3186	/*
3187	 * Note that there is a small window here where we may have
3188	 * a 512k page in the hash list but have not set the HAT_512K_FLAG
3189	 * flag yet, so we will skip searching the 512k hash list.
3190	 * In this case we will end up in pagefault which will find
3191	 * the mapping and return.  So, in this instance we will end up
3192	 * spending a bit more time resolving this TSB miss, but it can
3193	 * only happen once per process and even then, the chances of that
3194	 * are very small, so it's not worth the extra overhead it would
3195	 * take to close this window.
3196	 */
3197	brz,pn	%g5, tsb_4M
3198	  nop
31993:
3200	/*
3201	 * 512K hash
3202	 */
3203
3204	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3205		MMU_PAGESHIFT512K, TTE512K, tsb_l512K, tsb_checktte,
3206		sfmmu_suspend_tl, tsb_4M)
3207	/* NOT REACHED */
3208
3209tsb_4M:
3210	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3211	brz,pn	%g5, 4f
3212	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3213	and	%g4, HAT_4M_FLAG, %g5
3214	brz,pn	%g5, tsb_32M
3215	  nop
32164:
3217	/*
3218	 * 4M hash
3219	 */
3220
3221	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3222		MMU_PAGESHIFT4M, TTE4M, tsb_l4M, tsb_checktte,
3223		sfmmu_suspend_tl, tsb_32M)
3224	/* NOT REACHED */
3225
3226tsb_32M:
3227	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3228#ifdef sun4v
3229        brz,pn	%g5, 6f
3230#else
3231	brz,pn	%g5, tsb_pagefault
3232#endif
3233	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3234	and	%g4, HAT_32M_FLAG, %g5
3235	brz,pn	%g5, tsb_256M
3236	  nop
32375:
3238	/*
3239	 * 32M hash
3240	 */
3241
3242	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3243		MMU_PAGESHIFT32M, TTE32M, tsb_l32M, tsb_checktte,
3244		sfmmu_suspend_tl, tsb_256M)
3245	/* NOT REACHED */
3246
3247#ifdef sun4u
3248#define tsb_shme        tsb_pagefault
3249#endif
3250tsb_256M:
3251	ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3252	and	%g4, HAT_256M_FLAG, %g5
3253	brz,pn	%g5, tsb_shme
3254	  nop
32556:
3256	/*
3257	 * 256M hash
3258	 */
3259
3260	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3261	    MMU_PAGESHIFT256M, TTE256M, tsb_l256M, tsb_checktte,
3262	    sfmmu_suspend_tl, tsb_shme)
3263	/* NOT REACHED */
3264
3265tsb_checktte:
3266	/*
3267	 * g1 = hblk_misc
3268	 * g2 = tagacc
3269	 * g3 = tte
3270	 * g4 = tte pa
3271	 * g5 = tte va
3272	 * g6 = tsbmiss area
3273	 * g7 = hatid
3274	 */
3275	brlz,a,pt %g3, tsb_validtte
3276	  rdpr	%tt, %g7
3277
3278#ifdef sun4u
3279#undef tsb_shme
3280	ba	tsb_pagefault
3281	  nop
3282#else
3283
3284tsb_shme:
3285	/*
3286	 * g2 = tagacc
3287	 * g6 = tsbmiss area
3288	 */
3289	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3290	brz,pn	%g5, tsb_pagefault
3291	  nop
3292	ldx	[%g6 + TSBMISS_SHARED_UHATID], %g7	/* g7 = srdp */
3293	brz,pn	%g7, tsb_pagefault
3294	  nop
3295
3296	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3297		MMU_PAGESHIFT64K, TTE64K, tsb_shme_l8K, tsb_shme_checktte,
3298		sfmmu_suspend_tl, tsb_shme_512K)
3299	/* NOT REACHED */
3300
3301tsb_shme_512K:
3302	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3303	and	%g4, HAT_512K_FLAG, %g5
3304	brz,pn	%g5, tsb_shme_4M
3305	  nop
3306
3307	/*
3308	 * 512K hash
3309	 */
3310
3311	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3312		MMU_PAGESHIFT512K, TTE512K, tsb_shme_l512K, tsb_shme_checktte,
3313		sfmmu_suspend_tl, tsb_shme_4M)
3314	/* NOT REACHED */
3315
3316tsb_shme_4M:
3317	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3318	and	%g4, HAT_4M_FLAG, %g5
3319	brz,pn	%g5, tsb_shme_32M
3320	  nop
33214:
3322	/*
3323	 * 4M hash
3324	 */
3325	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3326		MMU_PAGESHIFT4M, TTE4M, tsb_shme_l4M, tsb_shme_checktte,
3327		sfmmu_suspend_tl, tsb_shme_32M)
3328	/* NOT REACHED */
3329
3330tsb_shme_32M:
3331	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3332	and	%g4, HAT_32M_FLAG, %g5
3333	brz,pn	%g5, tsb_shme_256M
3334	  nop
3335
3336	/*
3337	 * 32M hash
3338	 */
3339
3340	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3341		MMU_PAGESHIFT32M, TTE32M, tsb_shme_l32M, tsb_shme_checktte,
3342		sfmmu_suspend_tl, tsb_shme_256M)
3343	/* NOT REACHED */
3344
3345tsb_shme_256M:
3346	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3347	and	%g4, HAT_256M_FLAG, %g5
3348	brz,pn	%g5, tsb_pagefault
3349	  nop
3350
3351	/*
3352	 * 256M hash
3353	 */
3354
3355	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1,
3356	    MMU_PAGESHIFT256M, TTE256M, tsb_shme_l256M, tsb_shme_checktte,
3357	    sfmmu_suspend_tl, tsb_pagefault)
3358	/* NOT REACHED */
3359
3360tsb_shme_checktte:
3361
3362	brgez,pn %g3, tsb_pagefault
3363	  rdpr	%tt, %g7
3364	/*
3365	 * g1 = ctx1 flag
3366	 * g3 = tte
3367	 * g4 = tte pa
3368	 * g5 = tte va
3369	 * g6 = tsbmiss area
3370	 * g7 = tt
3371	 */
3372
3373	brz,pt  %g1, tsb_validtte
3374	  nop
3375	ldub    [%g6 + TSBMISS_URTTEFLAGS], %g1
3376	  or	%g1, HAT_CHKCTX1_FLAG, %g1
3377	stub    %g1, [%g6 + TSBMISS_URTTEFLAGS]
3378
3379	SAVE_CTX1(%g7, %g2, %g1, tsb_shmel)
3380#endif /* sun4u */
3381
3382tsb_validtte:
3383	/*
3384	 * g3 = tte
3385	 * g4 = tte pa
3386	 * g5 = tte va
3387	 * g6 = tsbmiss area
3388	 * g7 = tt
3389	 */
3390
3391	/*
3392	 * Set ref/mod bits if this is a prot trap.  Usually, it isn't.
3393	 */
3394	cmp	%g7, FAST_PROT_TT
3395	bne,pt	%icc, 4f
3396	  nop
3397
3398	TTE_SET_REFMOD_ML(%g3, %g4, %g5, %g6, %g7, tsb_lset_refmod,
3399	    tsb_protfault)
3400
3401	rdpr	%tt, %g5
3402	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
3403#ifdef sun4v
3404	MMU_FAULT_STATUS_AREA(%g7)
3405	ldx	[%g7 + MMFSA_D_ADDR], %g5	/* save fault addr for later */
3406#endif
3407	ba,pt	%xcc, tsb_update_tl1
3408	  nop
3409
34104:
3411	/*
3412	 * If ITLB miss check exec bit.
3413	 * If not set treat as invalid TTE.
3414	 */
3415	cmp     %g7, T_INSTR_MMU_MISS
3416	be,pn	%icc, 5f
3417	  andcc   %g3, TTE_EXECPRM_INT, %g0	/* check execute bit is set */
3418	cmp     %g7, FAST_IMMU_MISS_TT
3419	bne,pt %icc, 3f
3420	  andcc   %g3, TTE_EXECPRM_INT, %g0	/* check execute bit is set */
34215:
3422	bz,pn %icc, tsb_protfault
3423	  nop
3424
34253:
3426	/*
3427	 * Set reference bit if not already set
3428	 */
3429	TTE_SET_REF_ML(%g3, %g4, %g5, %g6, %g7, tsb_lset_ref)
3430
3431	/*
3432	 * Now, load into TSB/TLB.  At this point:
3433	 * g3 = tte
3434	 * g4 = patte
3435	 * g6 = tsbmiss area
3436	 */
3437	rdpr	%tt, %g5
3438#ifdef sun4v
3439	MMU_FAULT_STATUS_AREA(%g2)
3440	cmp	%g5, T_INSTR_MMU_MISS
3441	be,a,pt	%icc, 9f
3442	  nop
3443	cmp	%g5, FAST_IMMU_MISS_TT
3444	be,a,pt	%icc, 9f
3445	  nop
3446	add	%g2, MMFSA_D_, %g2
34479:
3448	ldx	[%g2 + MMFSA_CTX_], %g7
3449	sllx	%g7, TTARGET_CTX_SHIFT, %g7
3450	ldx	[%g2 + MMFSA_ADDR_], %g2
3451	mov	%g2, %g5		! save the fault addr for later use
3452	srlx	%g2, TTARGET_VA_SHIFT, %g2
3453	or	%g2, %g7, %g2
3454#else
3455	cmp	%g5, FAST_IMMU_MISS_TT
3456	be,a,pt	%icc, tsb_update_tl1
3457	  ldxa	[%g0]ASI_IMMU, %g2
3458	ldxa	[%g0]ASI_DMMU, %g2
3459#endif
3460tsb_update_tl1:
3461	srlx	%g2, TTARGET_CTX_SHIFT, %g7
3462	brz,pn	%g7, tsb_kernel
3463#ifdef sun4v
3464	  and	%g3, TTE_SZ_BITS, %g7	! assumes TTE_SZ_SHFT is 0
3465#else
3466	  srlx	%g3, TTE_SZ_SHFT, %g7
3467#endif
3468
3469tsb_user:
3470#ifdef sun4v
3471	cmp	%g7, TTE4M
3472	bge,pn	%icc, tsb_user4m
3473	  nop
3474#else /* sun4v */
3475	cmp	%g7, TTESZ_VALID | TTE4M
3476	be,pn	%icc, tsb_user4m
3477	  srlx	%g3, TTE_SZ2_SHFT, %g7
3478	andcc	%g7, TTE_SZ2_BITS, %g7		! check 32/256MB
3479#ifdef ITLB_32M_256M_SUPPORT
3480	bnz,pn	%icc, tsb_user4m
3481	  nop
3482#else /* ITLB_32M_256M_SUPPORT */
3483	bnz,a,pn %icc, tsb_user_pn_synth
3484	 cmp	%g5, FAST_IMMU_MISS_TT
3485#endif /* ITLB_32M_256M_SUPPORT */
3486#endif /* sun4v */
3487
3488tsb_user8k:
3489#ifdef sun4v
3490	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g7
3491	and	%g7, HAT_CHKCTX1_FLAG, %g1
3492	brz,a,pn %g1, 1f
3493	  ldn	[%g6 + TSBMISS_TSBPTR], %g1	! g1 = first TSB ptr
3494	GET_UTSBREG_SHCTX(%g6, TSBMISS_TSBSCDPTR, %g1)
3495	brlz,a,pn %g1, ptl1_panic			! if no shared tsb
3496	  mov PTL1_NO_SCDTSB8K, %g1			! panic
3497	GET_3RD_TSBE_PTR(%g5, %g1, %g6, %g7)
34981:
3499#else
3500	ldn	[%g6 + TSBMISS_TSBPTR], %g1	! g1 = first TSB ptr
3501
3502#ifndef UTSB_PHYS
3503	mov	ASI_N, %g7	! user TSBs accessed by VA
3504	mov	%g7, %asi
3505#endif /* UTSB_PHYS */
3506
3507#endif /* sun4v */
3508
3509	TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, 5)
3510
3511#ifdef sun4v
3512	rdpr    %tt, %g5
3513	cmp	%g5, T_INSTR_MMU_MISS
3514	be,a,pn	%xcc, 9f
3515	  mov	%g3, %g5
3516#endif /* sun4v */
3517	cmp	%g5, FAST_IMMU_MISS_TT
3518	be,pn	%xcc, 9f
3519	  mov	%g3, %g5
3520
3521	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3522	! trapstat wants TTE in %g5
3523	retry
35249:
3525	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3526	! trapstat wants TTE in %g5
3527	retry
3528
3529tsb_user4m:
3530#ifdef sun4v
3531	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g7
3532	and	%g7, HAT_CHKCTX1_FLAG, %g1
3533	brz,a,pn %g1, 4f
3534	  ldn	[%g6 + TSBMISS_TSBPTR4M], %g1		! g1 = TSB ptr
3535	GET_UTSBREG_SHCTX(%g6, TSBMISS_TSBSCDPTR4M, %g1)
3536	brlz,a,pn %g1, 5f				! if no shared 2nd tsb
3537	  nop
3538	GET_4TH_TSBE_PTR(%g5, %g1, %g6, %g7)
3539#else
3540	ldn	[%g6 + TSBMISS_TSBPTR4M], %g1		! g1 = TSB ptr
3541#endif
35424:
3543	brlz,pn %g1, 5f	/* Check to see if we have 2nd TSB programmed */
3544	  nop
3545
3546#ifndef UTSB_PHYS
3547	mov	ASI_N, %g7	! user TSBs accessed by VA
3548	mov	%g7, %asi
3549#endif /* UTSB_PHYS */
3550
3551        TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, 6)
3552
35535:
3554#ifdef sun4v
3555	rdpr    %tt, %g5
3556        cmp     %g5, T_INSTR_MMU_MISS
3557        be,a,pn %xcc, 9f
3558          mov   %g3, %g5
3559#endif /* sun4v */
3560        cmp     %g5, FAST_IMMU_MISS_TT
3561        be,pn   %xcc, 9f
3562        mov     %g3, %g5
3563
3564        DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3565        ! trapstat wants TTE in %g5
3566        retry
35679:
3568        ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3569        ! trapstat wants TTE in %g5
3570        retry
3571
3572#if !defined(sun4v) && !defined(ITLB_32M_256M_SUPPORT)
3573	/*
3574	 * Panther ITLB synthesis.
3575	 * The Panther 32M and 256M ITLB code simulates these two large page
3576	 * sizes with 4M pages, to provide support for programs, for example
3577	 * Java, that may copy instructions into a 32M or 256M data page and
3578	 * then execute them. The code below generates the 4M pfn bits and
3579	 * saves them in the modified 32M/256M ttes in the TSB. If the tte is
3580	 * stored in the DTLB to map a 32M/256M page, the 4M pfn offset bits
3581	 * are ignored by the hardware.
3582	 *
3583	 * Now, load into TSB/TLB.  At this point:
3584	 * g2 = tagtarget
3585	 * g3 = tte
3586	 * g4 = patte
3587	 * g5 = tt
3588	 * g6 = tsbmiss area
3589	 */
3590tsb_user_pn_synth:
3591	be,pt	%xcc, tsb_user_itlb_synth	/* ITLB miss */
3592	  andcc %g3, TTE_EXECPRM_INT, %g0	/* is execprm bit set */
3593	bz,pn %icc, 4b				/* if not, been here before */
3594	  ldn	[%g6 + TSBMISS_TSBPTR4M], %g1	/* g1 = tsbp */
3595	brlz,a,pn %g1, 5f			/* no 2nd tsb */
3596	  mov	%g3, %g5
3597
3598	mov	MMU_TAG_ACCESS, %g7
3599	ldxa	[%g7]ASI_DMMU, %g6		/* get tag access va */
3600	GET_4M_PFN_OFF(%g3, %g6, %g5, %g7, 1)	/* make 4M pfn offset */
3601
3602	mov	ASI_N, %g7	/* user TSBs always accessed by VA */
3603	mov	%g7, %asi
3604	TSB_UPDATE_TL_PN(%g1, %g5, %g2, %g4, %g7, %g3, 4) /* update TSB */
36055:
3606        DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3607        retry
3608
3609tsb_user_itlb_synth:
3610	ldn	[%g6 + TSBMISS_TSBPTR4M], %g1		/* g1 = tsbp */
3611
3612	mov	MMU_TAG_ACCESS, %g7
3613	ldxa	[%g7]ASI_IMMU, %g6		/* get tag access va */
3614	GET_4M_PFN_OFF(%g3, %g6, %g5, %g7, 2)	/* make 4M pfn offset */
3615	brlz,a,pn %g1, 7f	/* Check to see if we have 2nd TSB programmed */
3616	  or	%g5, %g3, %g5			/* add 4M bits to TTE */
3617
3618	mov	ASI_N, %g7	/* user TSBs always accessed by VA */
3619	mov	%g7, %asi
3620	TSB_UPDATE_TL_PN(%g1, %g5, %g2, %g4, %g7, %g3, 6) /* update TSB */
36217:
3622	SET_TTE4M_PN(%g5, %g7)			/* add TTE4M pagesize to TTE */
3623        ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3624        retry
3625#endif /* sun4v && ITLB_32M_256M_SUPPORT */
3626
3627tsb_kernel:
3628#ifdef sun4v
3629	rdpr	%tt, %g5
3630	cmp	%g7, TTE4M
3631	bge,pn	%icc, 5f
3632#else
3633	cmp	%g7, TTESZ_VALID | TTE4M	! no 32M or 256M support
3634	be,pn	%icc, 5f
3635#endif
3636	  nop
3637	ldn	[%g6 + TSBMISS_TSBPTR], %g1	! g1 = 8k tsbptr
3638	ba,pt	%xcc, 6f
3639	  nop
36405:
3641	ldn	[%g6 + TSBMISS_TSBPTR4M], %g1	! g1 = 4m tsbptr
3642	brlz,pn	%g1, 3f		/* skip programming if 4m TSB ptr is -1 */
3643	  nop
36446:
3645#ifndef sun4v
3646tsb_kernel_patch_asi:
3647	or	%g0, RUNTIME_PATCH, %g6
3648	mov	%g6, %asi	! XXX avoid writing to %asi !!
3649#endif
3650	TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, 7)
36513:
3652#ifdef sun4v
3653	cmp	%g5, T_INSTR_MMU_MISS
3654	be,a,pn	%icc, 1f
3655	  mov	%g3, %g5			! trapstat wants TTE in %g5
3656#endif /* sun4v */
3657	cmp	%g5, FAST_IMMU_MISS_TT
3658	be,pn	%icc, 1f
3659	  mov	%g3, %g5			! trapstat wants TTE in %g5
3660	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3661	! trapstat wants TTE in %g5
3662	retry
36631:
3664	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3665	! trapstat wants TTE in %g5
3666	retry
3667
3668tsb_ism:
3669	/*
3670	 * This is an ISM [i|d]tlb miss.  We optimize for largest
3671	 * page size down to smallest.
3672	 *
3673	 * g2 = vaddr + ctx(or ctxtype (sun4v)) aka (pseudo-)tag access
3674	 *	register
3675	 * g3 = ismmap->ism_seg
3676	 * g4 = physical address of ismmap->ism_sfmmu
3677	 * g6 = tsbmiss area
3678	 */
3679	ldna	[%g4]ASI_MEM, %g7		/* g7 = ism hatid */
3680	brz,a,pn %g7, ptl1_panic		/* if zero jmp ahead */
3681	  mov	PTL1_BAD_ISM, %g1
3682						/* g5 = pa of imap_vb_shift */
3683	sub	%g4, (IMAP_ISMHAT - IMAP_VB_SHIFT), %g5
3684	lduba	[%g5]ASI_MEM, %g4		/* g4 = imap_vb_shift */
3685	srlx	%g3, %g4, %g3			/* clr size field */
3686	set	TAGACC_CTX_MASK, %g1		/* mask off ctx number */
3687	sllx    %g3, %g4, %g3                   /* g3 = ism vbase */
3688	and     %g2, %g1, %g4                   /* g4 = ctx number */
3689	andn    %g2, %g1, %g1                   /* g1 = tlb miss vaddr */
3690	sub     %g1, %g3, %g2                   /* g2 = offset in ISM seg */
3691	or      %g2, %g4, %g2                   /* g2 = (pseudo-)tagacc */
3692	sub     %g5, (IMAP_VB_SHIFT - IMAP_HATFLAGS), %g5
3693	lduha   [%g5]ASI_MEM, %g4               /* g5 = pa of imap_hatflags */
3694#ifdef sun4v
3695	and     %g4, HAT_CTX1_FLAG, %g5         /* g5 = imap_hatflags */
3696	brz,pt %g5, tsb_chk4M_ism
3697	  nop
3698	ldub    [%g6 + TSBMISS_URTTEFLAGS], %g5
3699	or      %g5, HAT_CHKCTX1_FLAG, %g5
3700	stub    %g5, [%g6 + TSBMISS_URTTEFLAGS]
3701	rdpr    %tt, %g5
3702	SAVE_CTX1(%g5, %g3, %g1, tsb_shctxl)
3703#endif
3704	/*
3705	 * ISM pages are always locked down.
3706	 * If we can't find the tte then pagefault
3707	 * and let the spt segment driver resolve it.
3708	 *
3709	 * g2 = tagacc w/ISM vaddr (offset in ISM seg)
3710	 * g4 = imap_hatflags
3711	 * g6 = tsb miss area
3712	 * g7 = ISM hatid
3713	 */
3714
3715tsb_chk4M_ism:
3716	and	%g4, HAT_4M_FLAG, %g5		/* g4 = imap_hatflags */
3717	brnz,pt	%g5, tsb_ism_4M			/* branch if 4M pages */
3718	  nop
3719
3720tsb_ism_32M:
3721	and	%g4, HAT_32M_FLAG, %g5		/* check default 32M next */
3722	brz,pn	%g5, tsb_ism_256M
3723	  nop
3724
3725	/*
3726	 * 32M hash.
3727	 */
3728
3729	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1, MMU_PAGESHIFT32M,
3730	    TTE32M, tsb_ism_l32M, tsb_ism_32M_found, sfmmu_suspend_tl,
3731	    tsb_ism_4M)
3732	/* NOT REACHED */
3733
3734tsb_ism_32M_found:
3735	brlz,a,pt %g3, tsb_validtte
3736	  rdpr	%tt, %g7
3737	ba,pt	%xcc, tsb_ism_4M
3738	  nop
3739
3740tsb_ism_256M:
3741	and	%g4, HAT_256M_FLAG, %g5		/* 256M is last resort */
3742	brz,a,pn %g5, ptl1_panic
3743	  mov	PTL1_BAD_ISM, %g1
3744
3745	/*
3746	 * 256M hash.
3747	 */
3748	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1, MMU_PAGESHIFT256M,
3749	    TTE256M, tsb_ism_l256M, tsb_ism_256M_found, sfmmu_suspend_tl,
3750	    tsb_ism_4M)
3751
3752tsb_ism_256M_found:
3753	brlz,a,pt %g3, tsb_validtte
3754	  rdpr	%tt, %g7
3755
3756tsb_ism_4M:
3757	/*
3758	 * 4M hash.
3759	 */
3760	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1, MMU_PAGESHIFT4M,
3761	    TTE4M, tsb_ism_l4M, tsb_ism_4M_found, sfmmu_suspend_tl,
3762	    tsb_ism_8K)
3763	/* NOT REACHED */
3764
3765tsb_ism_4M_found:
3766	brlz,a,pt %g3, tsb_validtte
3767	  rdpr	%tt, %g7
3768
3769tsb_ism_8K:
3770	/*
3771	 * 8K and 64K hash.
3772	 */
3773
3774	GET_TTE(%g2, %g7, %g3, %g4, %g5, %g6, %g1, MMU_PAGESHIFT64K,
3775	    TTE64K, tsb_ism_l8K, tsb_ism_8K_found, sfmmu_suspend_tl,
3776	    tsb_pagefault)
3777	/* NOT REACHED */
3778
3779tsb_ism_8K_found:
3780	brlz,a,pt %g3, tsb_validtte
3781	  rdpr	%tt, %g7
3782
3783tsb_pagefault:
3784	rdpr	%tt, %g7
3785	cmp	%g7, FAST_PROT_TT
3786	be,a,pn	%icc, tsb_protfault
3787	  wrpr	%g0, FAST_DMMU_MISS_TT, %tt
3788
3789tsb_protfault:
3790	/*
3791	 * we get here if we couldn't find a valid tte in the hash.
3792	 *
3793	 * If user and we are at tl>1 we go to window handling code.
3794	 *
3795	 * If kernel and the fault is on the same page as our stack
3796	 * pointer, then we know the stack is bad and the trap handler
3797	 * will fail, so we call ptl1_panic with PTL1_BAD_STACK.
3798	 *
3799	 * If this is a kernel trap and tl>1, panic.
3800	 *
3801	 * Otherwise we call pagefault.
3802	 */
3803	cmp	%g7, FAST_IMMU_MISS_TT
3804#ifdef sun4v
3805	MMU_FAULT_STATUS_AREA(%g4)
3806	ldx	[%g4 + MMFSA_I_CTX], %g5
3807	ldx	[%g4 + MMFSA_D_CTX], %g4
3808	move	%icc, %g5, %g4
3809	cmp	%g7, T_INSTR_MMU_MISS
3810	move	%icc, %g5, %g4
3811#else
3812	mov	MMU_TAG_ACCESS, %g4
3813	ldxa	[%g4]ASI_DMMU, %g2
3814	ldxa	[%g4]ASI_IMMU, %g5
3815	move	%icc, %g5, %g2
3816	cmp	%g7, T_INSTR_MMU_MISS
3817	move	%icc, %g5, %g2
3818	sllx	%g2, TAGACC_CTX_LSHIFT, %g4
3819#endif
3820	brnz,pn	%g4, 3f				/* skip if not kernel */
3821	  rdpr	%tl, %g5
3822
3823	add	%sp, STACK_BIAS, %g3
3824	srlx	%g3, MMU_PAGESHIFT, %g3
3825	srlx	%g2, MMU_PAGESHIFT, %g4
3826	cmp	%g3, %g4
3827	be,a,pn	%icc, ptl1_panic		/* panic if bad %sp */
3828	  mov	PTL1_BAD_STACK, %g1
3829
3830	cmp	%g5, 1
3831	ble,pt	%icc, 2f
3832	  nop
3833	TSTAT_CHECK_TL1(2f, %g1, %g2)
3834	rdpr	%tt, %g2
3835	cmp	%g2, FAST_PROT_TT
3836	mov	PTL1_BAD_KPROT_FAULT, %g1
3837	movne	%icc, PTL1_BAD_KMISS, %g1
3838	ba,pt	%icc, ptl1_panic
3839	  nop
3840
38412:
3842	/*
3843	 * We are taking a pagefault in the kernel on a kernel address.  If
3844	 * CPU_DTRACE_NOFAULT is set in the cpuc_dtrace_flags, we don't actually
3845	 * want to call sfmmu_pagefault -- we will instead note that a fault
3846	 * has occurred by setting CPU_DTRACE_BADADDR and issue a "done"
3847	 * (instead of a "retry").  This will step over the faulting
3848	 * instruction.
3849	 */
3850	CPU_INDEX(%g1, %g2)
3851	set	cpu_core, %g2
3852	sllx	%g1, CPU_CORE_SHIFT, %g1
3853	add	%g1, %g2, %g1
3854	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
3855	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
3856	bz	sfmmu_pagefault
3857	or	%g2, CPU_DTRACE_BADADDR, %g2
3858	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
3859	GET_MMU_D_ADDR(%g3, %g4)
3860	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
3861	done
3862
38633:
3864	cmp	%g5, 1
3865	ble,pt	%icc, 4f
3866	  nop
3867	TSTAT_CHECK_TL1(4f, %g1, %g2)
3868	ba,pt	%icc, sfmmu_window_trap
3869	  nop
3870
38714:
3872	/*
3873	 * We are taking a pagefault on a non-kernel address.  If we are in
3874	 * the kernel (e.g., due to a copyin()), we will check cpuc_dtrace_flags
3875	 * and (if CPU_DTRACE_NOFAULT is set) will proceed as outlined above.
3876	 */
3877	CPU_INDEX(%g1, %g2)
3878	set	cpu_core, %g2
3879	sllx	%g1, CPU_CORE_SHIFT, %g1
3880	add	%g1, %g2, %g1
3881	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
3882	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
3883	bz	sfmmu_mmu_trap
3884	or	%g2, CPU_DTRACE_BADADDR, %g2
3885	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
3886	GET_MMU_D_ADDR(%g3, %g4)
3887	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
3888
3889	/*
3890	 * Be sure that we're actually taking this miss from the kernel --
3891	 * otherwise we have managed to return to user-level with
3892	 * CPU_DTRACE_NOFAULT set in cpuc_dtrace_flags.
3893	 */
3894	rdpr	%tstate, %g2
3895	btst	TSTATE_PRIV, %g2
3896	bz,a	ptl1_panic
3897	  mov	PTL1_BAD_DTRACE_FLAGS, %g1
3898	done
3899
3900	ALTENTRY(tsb_tl0_noctxt)
3901	/*
3902	 * If we have no context, check to see if CPU_DTRACE_NOFAULT is set;
3903	 * if it is, indicated that we have faulted and issue a done.
3904	 */
3905	CPU_INDEX(%g5, %g6)
3906	set	cpu_core, %g6
3907	sllx	%g5, CPU_CORE_SHIFT, %g5
3908	add	%g5, %g6, %g5
3909	lduh	[%g5 + CPUC_DTRACE_FLAGS], %g6
3910	andcc	%g6, CPU_DTRACE_NOFAULT, %g0
3911	bz	1f
3912	or	%g6, CPU_DTRACE_BADADDR, %g6
3913	stuh	%g6, [%g5 + CPUC_DTRACE_FLAGS]
3914	GET_MMU_D_ADDR(%g3, %g4)
3915	stx	%g3, [%g5 + CPUC_DTRACE_ILLVAL]
3916
3917	/*
3918	 * Be sure that we're actually taking this miss from the kernel --
3919	 * otherwise we have managed to return to user-level with
3920	 * CPU_DTRACE_NOFAULT set in cpuc_dtrace_flags.
3921	 */
3922	rdpr	%tstate, %g5
3923	btst	TSTATE_PRIV, %g5
3924	bz,a	ptl1_panic
3925	  mov	PTL1_BAD_DTRACE_FLAGS, %g1
3926	TSTAT_CHECK_TL1(2f, %g1, %g2);
39272:
3928	done
3929
39301:
3931	rdpr	%tt, %g5
3932	cmp	%g5, FAST_IMMU_MISS_TT
3933#ifdef sun4v
3934	MMU_FAULT_STATUS_AREA(%g2)
3935	be,a,pt	%icc, 2f
3936	  ldx	[%g2 + MMFSA_I_CTX], %g3
3937	cmp	%g5, T_INSTR_MMU_MISS
3938	be,a,pt	%icc, 2f
3939	  ldx	[%g2 + MMFSA_I_CTX], %g3
3940	ldx	[%g2 + MMFSA_D_CTX], %g3
39412:
3942#else
3943	mov	MMU_TAG_ACCESS, %g2
3944	be,a,pt	%icc, 2f
3945	  ldxa	[%g2]ASI_IMMU, %g3
3946	ldxa	[%g2]ASI_DMMU, %g3
39472:	sllx	%g3, TAGACC_CTX_LSHIFT, %g3
3948#endif
3949	brz,a,pn %g3, ptl1_panic		! panic if called for kernel
3950	  mov	PTL1_BAD_CTX_STEAL, %g1		! since kernel ctx was stolen
3951	rdpr	%tl, %g5
3952	cmp	%g5, 1
3953	ble,pt	%icc, sfmmu_mmu_trap
3954	  nop
3955	TSTAT_CHECK_TL1(sfmmu_mmu_trap, %g1, %g2)
3956	ba,pt	%icc, sfmmu_window_trap
3957	  nop
3958	SET_SIZE(sfmmu_tsb_miss)
3959#endif /* lint */
3960
3961#if defined (lint)
3962/*
3963 * This routine will look for a user or kernel vaddr in the hash
3964 * structure.  It returns a valid pfn or PFN_INVALID.  It doesn't
3965 * grab any locks.  It should only be used by other sfmmu routines.
3966 */
3967/* ARGSUSED */
3968pfn_t
3969sfmmu_vatopfn(caddr_t vaddr, sfmmu_t *sfmmup, tte_t *ttep)
3970{
3971	return(0);
3972}
3973
3974/* ARGSUSED */
3975pfn_t
3976sfmmu_kvaszc2pfn(caddr_t vaddr, int hashno)
3977{
3978	return(0);
3979}
3980
3981#else /* lint */
3982
3983	ENTRY_NP(sfmmu_vatopfn)
3984 	/*
3985 	 * disable interrupts
3986 	 */
3987 	rdpr	%pstate, %o3
3988#ifdef DEBUG
3989	PANIC_IF_INTR_DISABLED_PSTR(%o3, sfmmu_di_l5, %g1)
3990#endif
3991	/*
3992	 * disable interrupts to protect the TSBMISS area
3993	 */
3994	andn    %o3, PSTATE_IE, %o5
3995	wrpr    %o5, 0, %pstate
3996
3997	/*
3998	 * o0 = vaddr
3999	 * o1 = sfmmup
4000	 * o2 = ttep
4001	 */
4002	CPU_TSBMISS_AREA(%g1, %o5)
4003	ldn	[%g1 + TSBMISS_KHATID], %o4
4004	cmp	%o4, %o1
4005	bne,pn	%ncc, vatopfn_nokernel
4006	  mov	TTE64K, %g5			/* g5 = rehash # */
4007	mov %g1,%o5				/* o5 = tsbmiss_area */
4008	/*
4009	 * o0 = vaddr
4010	 * o1 & o4 = hatid
4011	 * o2 = ttep
4012	 * o5 = tsbmiss area
4013	 */
4014	mov	HBLK_RANGE_SHIFT, %g6
40151:
4016
4017	/*
4018	 * o0 = vaddr
4019	 * o1 = sfmmup
4020	 * o2 = ttep
4021	 * o3 = old %pstate
4022	 * o4 = hatid
4023	 * o5 = tsbmiss
4024	 * g5 = rehash #
4025	 * g6 = hmeshift
4026	 *
4027	 * The first arg to GET_TTE is actually tagaccess register
4028	 * not just vaddr. Since this call is for kernel we need to clear
4029	 * any lower vaddr bits that would be interpreted as ctx bits.
4030	 */
4031	set     TAGACC_CTX_MASK, %g1
4032	andn    %o0, %g1, %o0
4033	GET_TTE(%o0, %o4, %g1, %g2, %g3, %o5, %g4, %g6, %g5,
4034		vatopfn_l1, kvtop_hblk_found, tsb_suspend, kvtop_nohblk)
4035
4036kvtop_hblk_found:
4037	/*
4038	 * o0 = vaddr
4039	 * o1 = sfmmup
4040	 * o2 = ttep
4041	 * g1 = tte
4042	 * g2 = tte pa
4043	 * g3 = tte va
4044	 * o2 = tsbmiss area
4045	 * o1 = hat id
4046	 */
4047	brgez,a,pn %g1, 6f			/* if tte invalid goto tl0 */
4048	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
4049	stx %g1,[%o2]				/* put tte into *ttep */
4050	TTETOPFN(%g1, %o0, vatopfn_l2, %g2, %g3, %g4)
4051	/*
4052	 * o0 = vaddr
4053	 * o1 = sfmmup
4054	 * o2 = ttep
4055	 * g1 = pfn
4056	 */
4057	ba,pt	%xcc, 6f
4058	  mov	%g1, %o0
4059
4060kvtop_nohblk:
4061	/*
4062	 * we get here if we couldn't find valid hblk in hash.  We rehash
4063	 * if neccesary.
4064	 */
4065	ldn	[%o5 + (TSBMISS_SCRATCH + TSB_TAGACC)], %o0
4066#ifdef sun4v
4067	cmp	%g5, MAX_HASHCNT
4068#else
4069	cmp	%g5, DEFAULT_MAX_HASHCNT	/* no 32/256M kernel pages */
4070#endif
4071	be,a,pn	%icc, 6f
4072	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
4073	mov	%o1, %o4			/* restore hatid */
4074#ifdef sun4v
4075        add	%g5, 2, %g5
4076	cmp	%g5, 3
4077	move	%icc, MMU_PAGESHIFT4M, %g6
4078	ba,pt	%icc, 1b
4079	movne	%icc, MMU_PAGESHIFT256M, %g6
4080#else
4081        inc	%g5
4082	cmp	%g5, 2
4083	move	%icc, MMU_PAGESHIFT512K, %g6
4084	ba,pt	%icc, 1b
4085	movne	%icc, MMU_PAGESHIFT4M, %g6
4086#endif
40876:
4088	retl
4089 	  wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
4090
4091tsb_suspend:
4092	/*
4093	 * o0 = vaddr
4094	 * o1 = sfmmup
4095	 * o2 = ttep
4096	 * g1 = tte
4097	 * g2 = tte pa
4098	 * g3 = tte va
4099	 * o2 = tsbmiss area  use o5 instead of o2 for tsbmiss
4100	 */
4101	stx %g1,[%o2]				/* put tte into *ttep */
4102	brgez,a,pn %g1, 8f			/* if tte invalid goto 8: */
4103	  sub	%g0, 1, %o0			/* output = PFN_INVALID */
4104	sub	%g0, 2, %o0			/* output = PFN_SUSPENDED */
41058:
4106	retl
4107	 wrpr	%g0, %o3, %pstate		/* enable interrupts */
4108
4109vatopfn_nokernel:
4110	/*
4111	 * This routine does NOT support user addresses
4112	 * There is a routine in C that supports this.
4113	 * The only reason why we don't have the C routine
4114	 * support kernel addresses as well is because
4115	 * we do va_to_pa while holding the hashlock.
4116	 */
4117 	wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
4118	save	%sp, -SA(MINFRAME), %sp
4119	sethi	%hi(sfmmu_panic3), %o0
4120	call	panic
4121	 or	%o0, %lo(sfmmu_panic3), %o0
4122
4123	SET_SIZE(sfmmu_vatopfn)
4124
4125	/*
4126	 * %o0 = vaddr
4127	 * %o1 = hashno (aka szc)
4128	 *
4129	 *
4130	 * This routine is similar to sfmmu_vatopfn() but will only look for
4131	 * a kernel vaddr in the hash structure for the specified rehash value.
4132	 * It's just an optimization for the case when pagesize for a given
4133	 * va range is already known (e.g. large page heap) and we don't want
4134	 * to start the search with rehash value 1 as sfmmu_vatopfn() does.
4135	 *
4136	 * Returns valid pfn or PFN_INVALID if
4137	 * tte for specified rehash # is not found, invalid or suspended.
4138	 */
4139	ENTRY_NP(sfmmu_kvaszc2pfn)
4140 	/*
4141 	 * disable interrupts
4142 	 */
4143 	rdpr	%pstate, %o3
4144#ifdef DEBUG
4145	PANIC_IF_INTR_DISABLED_PSTR(%o3, sfmmu_di_l6, %g1)
4146#endif
4147	/*
4148	 * disable interrupts to protect the TSBMISS area
4149	 */
4150	andn    %o3, PSTATE_IE, %o5
4151	wrpr    %o5, 0, %pstate
4152
4153	CPU_TSBMISS_AREA(%g1, %o5)
4154	ldn	[%g1 + TSBMISS_KHATID], %o4
4155	sll	%o1, 1, %g6
4156	add	%g6, %o1, %g6
4157	add	%g6, MMU_PAGESHIFT, %g6
4158	/*
4159	 * %o0 = vaddr
4160	 * %o1 = hashno
4161	 * %o3 = old %pstate
4162	 * %o4 = ksfmmup
4163	 * %g1 = tsbmiss area
4164	 * %g6 = hmeshift
4165	 */
4166
4167	/*
4168	 * The first arg to GET_TTE is actually tagaccess register
4169	 * not just vaddr. Since this call is for kernel we need to clear
4170	 * any lower vaddr bits that would be interpreted as ctx bits.
4171	 */
4172	srlx	%o0, MMU_PAGESHIFT, %o0
4173	sllx	%o0, MMU_PAGESHIFT, %o0
4174	GET_TTE(%o0, %o4, %g3, %g4, %g5, %g1, %o5, %g6, %o1,
4175		kvaszc2pfn_l1, kvaszc2pfn_hblk_found, kvaszc2pfn_nohblk,
4176		kvaszc2pfn_nohblk)
4177
4178kvaszc2pfn_hblk_found:
4179	/*
4180	 * %g3 = tte
4181	 * %o0 = vaddr
4182	 */
4183	brgez,a,pn %g3, 1f			/* check if tte is invalid */
4184	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
4185	TTETOPFN(%g3, %o0, kvaszc2pfn_l2, %g2, %g4, %g5)
4186	/*
4187	 * g3 = pfn
4188	 */
4189	ba,pt	%xcc, 1f
4190	  mov	%g3, %o0
4191
4192kvaszc2pfn_nohblk:
4193	mov	-1, %o0
4194
41951:
4196	retl
4197 	  wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
4198
4199	SET_SIZE(sfmmu_kvaszc2pfn)
4200
4201#endif /* lint */
4202
4203
4204
4205#if !defined(lint)
4206
4207/*
4208 * kpm lock used between trap level tsbmiss handler and kpm C level.
4209 */
4210#define KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi)			\
4211	mov     0xff, tmp1						;\
4212label1:									;\
4213	casa    [kpmlckp]asi, %g0, tmp1					;\
4214	brnz,pn tmp1, label1						;\
4215	mov     0xff, tmp1						;\
4216	membar  #LoadLoad
4217
4218#define KPMLOCK_EXIT(kpmlckp, asi)					\
4219	membar  #LoadStore|#StoreStore					;\
4220	sta     %g0, [kpmlckp]asi
4221
4222/*
4223 * Lookup a memseg for a given pfn and if found, return the physical
4224 * address of the corresponding struct memseg in mseg, otherwise
4225 * return MSEG_NULLPTR_PA. The kpmtsbm pointer must be provided in
4226 * tsbmp, %asi is assumed to be ASI_MEM.
4227 * This lookup is done by strictly traversing only the physical memseg
4228 * linkage. The more generic approach, to check the virtual linkage
4229 * before using the physical (used e.g. with hmehash buckets), cannot
4230 * be used here. Memory DR operations can run in parallel to this
4231 * lookup w/o any locks and updates of the physical and virtual linkage
4232 * cannot be done atomically wrt. to each other. Because physical
4233 * address zero can be valid physical address, MSEG_NULLPTR_PA acts
4234 * as "physical NULL" pointer.
4235 */
4236#define	PAGE_NUM2MEMSEG_NOLOCK_PA(pfn, mseg, tsbmp, tmp1, tmp2, tmp3, label) \
4237	sethi	%hi(mhash_per_slot), tmp3 /* no tsbmp use due to DR */	;\
4238	ldx	[tmp3 + %lo(mhash_per_slot)], mseg			;\
4239	udivx	pfn, mseg, mseg						;\
4240	ldx	[tsbmp + KPMTSBM_MSEGPHASHPA], tmp1			;\
4241	and	mseg, SFMMU_N_MEM_SLOTS - 1, mseg			;\
4242	sllx	mseg, SFMMU_MEM_HASH_ENTRY_SHIFT, mseg			;\
4243	add	tmp1, mseg, tmp1					;\
4244	ldxa	[tmp1]%asi, mseg					;\
4245	cmp	mseg, MSEG_NULLPTR_PA					;\
4246	be,pn	%xcc, label/**/1		/* if not found */	;\
4247	  nop								;\
4248	ldxa	[mseg + MEMSEG_PAGES_BASE]%asi, tmp1			;\
4249	cmp	pfn, tmp1			/* pfn - pages_base */	;\
4250	blu,pn	%xcc, label/**/1					;\
4251	  ldxa	[mseg + MEMSEG_PAGES_END]%asi, tmp2			;\
4252	cmp	pfn, tmp2			/* pfn - pages_end */	;\
4253	bgeu,pn	%xcc, label/**/1					;\
4254	  sub	pfn, tmp1, tmp1			/* pfn - pages_base */	;\
4255	mulx	tmp1, PAGE_SIZE, tmp1					;\
4256	ldxa	[mseg + MEMSEG_PAGESPA]%asi, tmp2	/* pages */	;\
4257	add	tmp2, tmp1, tmp1			/* pp */	;\
4258	lduwa	[tmp1 + PAGE_PAGENUM]%asi, tmp2				;\
4259	cmp	tmp2, pfn						;\
4260	be,pt	%xcc, label/**/_ok			/* found */	;\
4261label/**/1:								;\
4262	/* brute force lookup */					;\
4263	sethi	%hi(memsegspa), tmp3 /* no tsbmp use due to DR */	;\
4264	ldx	[tmp3 + %lo(memsegspa)], mseg				;\
4265label/**/2:								;\
4266	cmp	mseg, MSEG_NULLPTR_PA					;\
4267	be,pn	%xcc, label/**/_ok		/* if not found */	;\
4268	  nop								;\
4269	ldxa	[mseg + MEMSEG_PAGES_BASE]%asi, tmp1			;\
4270	cmp	pfn, tmp1			/* pfn - pages_base */	;\
4271	blu,a,pt %xcc, label/**/2					;\
4272	  ldxa	[mseg + MEMSEG_NEXTPA]%asi, mseg			;\
4273	ldxa	[mseg + MEMSEG_PAGES_END]%asi, tmp2			;\
4274	cmp	pfn, tmp2			/* pfn - pages_end */	;\
4275	bgeu,a,pt %xcc, label/**/2					;\
4276	  ldxa	[mseg + MEMSEG_NEXTPA]%asi, mseg			;\
4277label/**/_ok:
4278
4279	/*
4280	 * kpm tsb miss handler large pages
4281	 * g1 = 8K kpm TSB entry pointer
4282	 * g2 = tag access register
4283	 * g3 = 4M kpm TSB entry pointer
4284	 */
4285	ALTENTRY(sfmmu_kpm_dtsb_miss)
4286	TT_TRACE(trace_tsbmiss)
4287
4288	CPU_INDEX(%g7, %g6)
4289	sethi	%hi(kpmtsbm_area), %g6
4290	sllx	%g7, KPMTSBM_SHIFT, %g7
4291	or	%g6, %lo(kpmtsbm_area), %g6
4292	add	%g6, %g7, %g6			/* g6 = kpmtsbm ptr */
4293
4294	/* check enable flag */
4295	ldub	[%g6 + KPMTSBM_FLAGS], %g4
4296	and	%g4, KPMTSBM_ENABLE_FLAG, %g5
4297	brz,pn	%g5, sfmmu_tsb_miss		/* if kpm not enabled */
4298	  nop
4299
4300	/* VA range check */
4301	ldx	[%g6 + KPMTSBM_VBASE], %g7
4302	cmp	%g2, %g7
4303	blu,pn	%xcc, sfmmu_tsb_miss
4304	  ldx	[%g6 + KPMTSBM_VEND], %g5
4305	cmp	%g2, %g5
4306	bgeu,pn	%xcc, sfmmu_tsb_miss
4307	  stx	%g3, [%g6 + KPMTSBM_TSBPTR]
4308
4309	/*
4310	 * check TL tsbmiss handling flag
4311	 * bump tsbmiss counter
4312	 */
4313	lduw	[%g6 + KPMTSBM_TSBMISS], %g5
4314#ifdef	DEBUG
4315	and	%g4, KPMTSBM_TLTSBM_FLAG, %g3
4316	inc	%g5
4317	brz,pn	%g3, sfmmu_kpm_exception
4318	  st	%g5, [%g6 + KPMTSBM_TSBMISS]
4319#else
4320	inc	%g5
4321	st	%g5, [%g6 + KPMTSBM_TSBMISS]
4322#endif
4323	/*
4324	 * At this point:
4325	 *  g1 = 8K kpm TSB pointer (not used)
4326	 *  g2 = tag access register
4327	 *  g3 = clobbered
4328	 *  g6 = per-CPU kpm tsbmiss area
4329	 *  g7 = kpm_vbase
4330	 */
4331
4332	/* vaddr2pfn */
4333	ldub	[%g6 + KPMTSBM_SZSHIFT], %g3
4334	sub	%g2, %g7, %g4			/* paddr = vaddr-kpm_vbase */
4335	srax    %g4, %g3, %g2			/* which alias range (r) */
4336	brnz,pn	%g2, sfmmu_kpm_exception	/* if (r != 0) goto C handler */
4337	  srlx	%g4, MMU_PAGESHIFT, %g2		/* %g2 = pfn */
4338
4339	/*
4340	 * Setup %asi
4341	 * mseg_pa = page_numtomemseg_nolock(pfn)
4342	 * if (mseg_pa == NULL) sfmmu_kpm_exception
4343	 * g2=pfn
4344	 */
4345	mov	ASI_MEM, %asi
4346	PAGE_NUM2MEMSEG_NOLOCK_PA(%g2, %g3, %g6, %g4, %g5, %g7, kpmtsbmp2m)
4347	cmp	%g3, MSEG_NULLPTR_PA
4348	be,pn	%xcc, sfmmu_kpm_exception	/* if mseg not found */
4349	  nop
4350
4351	/*
4352	 * inx = ptokpmp((kpmptop((ptopkpmp(pfn))) - mseg_pa->kpm_pbase));
4353	 * g2=pfn g3=mseg_pa
4354	 */
4355	ldub	[%g6 + KPMTSBM_KPMP2PSHFT], %g5
4356	ldxa	[%g3 + MEMSEG_KPM_PBASE]%asi, %g7
4357	srlx	%g2, %g5, %g4
4358	sllx	%g4, %g5, %g4
4359	sub	%g4, %g7, %g4
4360	srlx	%g4, %g5, %g4
4361
4362	/*
4363	 * Validate inx value
4364	 * g2=pfn g3=mseg_pa g4=inx
4365	 */
4366#ifdef	DEBUG
4367	ldxa	[%g3 + MEMSEG_KPM_NKPMPGS]%asi, %g5
4368	cmp	%g4, %g5			/* inx - nkpmpgs */
4369	bgeu,pn	%xcc, sfmmu_kpm_exception	/* if out of range */
4370	  ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4371#else
4372	ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4373#endif
4374	/*
4375	 * kp = &mseg_pa->kpm_pages[inx]
4376	 */
4377	sllx	%g4, KPMPAGE_SHIFT, %g4		/* kpm_pages offset */
4378	ldxa	[%g3 + MEMSEG_KPM_PAGES]%asi, %g5 /* kpm_pages */
4379	add	%g5, %g4, %g5			/* kp */
4380
4381	/*
4382	 * KPMP_HASH(kp)
4383	 * g2=pfn g3=mseg_pa g4=offset g5=kp g7=kpmp_table_sz
4384	 */
4385	ldub	[%g6 + KPMTSBM_KPMPSHIFT], %g1	/* kpmp_shift */
4386	sub	%g7, 1, %g7			/* mask */
4387	srlx	%g5, %g1, %g1			/* x = ksp >> kpmp_shift */
4388	add	%g5, %g1, %g5			/* y = ksp + x */
4389	and 	%g5, %g7, %g5			/* hashinx = y & mask */
4390
4391	/*
4392	 * Calculate physical kpm_page pointer
4393	 * g2=pfn g3=mseg_pa g4=offset g5=hashinx
4394	 */
4395	ldxa	[%g3 + MEMSEG_KPM_PAGESPA]%asi, %g1 /* kpm_pagespa */
4396	add	%g1, %g4, %g1			/* kp_pa */
4397
4398	/*
4399	 * Calculate physical hash lock address
4400	 * g1=kp_refcntc_pa g2=pfn g5=hashinx
4401	 */
4402	ldx	[%g6 + KPMTSBM_KPMPTABLEPA], %g4 /* kpmp_tablepa */
4403	sllx	%g5, KPMHLK_SHIFT, %g5
4404	add	%g4, %g5, %g3
4405	add	%g3, KPMHLK_LOCK, %g3		/* hlck_pa */
4406
4407	/*
4408	 * Assemble tte
4409	 * g1=kp_pa g2=pfn g3=hlck_pa
4410	 */
4411#ifdef sun4v
4412	sethi	%hi(TTE_VALID_INT), %g5		/* upper part */
4413	sllx	%g5, 32, %g5
4414	mov	(TTE_CP_INT|TTE_CV_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4415	or	%g4, TTE4M, %g4
4416	or	%g5, %g4, %g5
4417#else
4418	sethi	%hi(TTE_VALID_INT), %g4
4419	mov	TTE4M, %g5
4420	sllx	%g5, TTE_SZ_SHFT_INT, %g5
4421	or	%g5, %g4, %g5			/* upper part */
4422	sllx	%g5, 32, %g5
4423	mov	(TTE_CP_INT|TTE_CV_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4424	or	%g5, %g4, %g5
4425#endif
4426	sllx	%g2, MMU_PAGESHIFT, %g4
4427	or	%g5, %g4, %g5			/* tte */
4428	ldx	[%g6 + KPMTSBM_TSBPTR], %g4
4429	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
4430
4431	/*
4432	 * tsb dropin
4433	 * g1=kp_pa g2=ttarget g3=hlck_pa g4=kpmtsbp4m g5=tte g6=kpmtsbm_area
4434	 */
4435
4436	/* KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi) */
4437	KPMLOCK_ENTER(%g3, %g7, kpmtsbmhdlr1, ASI_MEM)
4438
4439	/* use C-handler if there's no go for dropin */
4440	ldsha	[%g1 + KPMPAGE_REFCNTC]%asi, %g7 /* kp_refcntc */
4441	cmp	%g7, -1
4442	bne,pn	%xcc, 5f	/* use C-handler if there's no go for dropin */
4443	  nop
4444
4445#ifdef	DEBUG
4446	/* double check refcnt */
4447	ldsha	[%g1 + KPMPAGE_REFCNT]%asi, %g7
4448	brz,pn	%g7, 5f			/* let C-handler deal with this */
4449	  nop
4450#endif
4451
4452#ifndef sun4v
4453	ldub	[%g6 + KPMTSBM_FLAGS], %g7
4454	mov	ASI_N, %g1
4455	andcc	%g7, KPMTSBM_TSBPHYS_FLAG, %g0
4456	movnz	%icc, ASI_MEM, %g1
4457	mov	%g1, %asi
4458#endif
4459
4460	/* TSB_LOCK_ENTRY(tsbp, tmp1, tmp2, label) (needs %asi set) */
4461	TSB_LOCK_ENTRY(%g4, %g1, %g7, 6)
4462
4463	/* TSB_INSERT_UNLOCK_ENTRY(tsbp, tte, tagtarget, tmp) */
4464	TSB_INSERT_UNLOCK_ENTRY(%g4, %g5, %g2, %g7)
4465
4466	DTLB_STUFF(%g5, %g1, %g2, %g4, %g6)
4467
4468	/* KPMLOCK_EXIT(kpmlckp, asi) */
4469	KPMLOCK_EXIT(%g3, ASI_MEM)
4470
4471	/*
4472	 * If trapstat is running, we need to shift the %tpc and %tnpc to
4473	 * point to trapstat's TSB miss return code (note that trapstat
4474	 * itself will patch the correct offset to add).
4475	 * Note: TTE is expected in %g5 (allows per pagesize reporting).
4476	 */
4477	rdpr	%tl, %g7
4478	cmp	%g7, 1
4479	ble	%icc, 0f
4480	sethi	%hi(KERNELBASE), %g6
4481	rdpr	%tpc, %g7
4482	or	%g6, %lo(KERNELBASE), %g6
4483	cmp	%g7, %g6
4484	bgeu	%xcc, 0f
4485	ALTENTRY(tsbmiss_trapstat_patch_point_kpm)
4486	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
4487	wrpr	%g7, %tpc
4488	add	%g7, 4, %g7
4489	wrpr	%g7, %tnpc
44900:
4491	retry
44925:
4493	/* g3=hlck_pa */
4494	KPMLOCK_EXIT(%g3, ASI_MEM)
4495	ba,pt	%icc, sfmmu_kpm_exception
4496	  nop
4497	SET_SIZE(sfmmu_kpm_dtsb_miss)
4498
4499	/*
4500	 * kpm tsbmiss handler for smallpages
4501	 * g1 = 8K kpm TSB pointer
4502	 * g2 = tag access register
4503	 * g3 = 4M kpm TSB pointer
4504	 */
4505	ALTENTRY(sfmmu_kpm_dtsb_miss_small)
4506	TT_TRACE(trace_tsbmiss)
4507	CPU_INDEX(%g7, %g6)
4508	sethi	%hi(kpmtsbm_area), %g6
4509	sllx	%g7, KPMTSBM_SHIFT, %g7
4510	or	%g6, %lo(kpmtsbm_area), %g6
4511	add	%g6, %g7, %g6			/* g6 = kpmtsbm ptr */
4512
4513	/* check enable flag */
4514	ldub	[%g6 + KPMTSBM_FLAGS], %g4
4515	and	%g4, KPMTSBM_ENABLE_FLAG, %g5
4516	brz,pn	%g5, sfmmu_tsb_miss		/* if kpm not enabled */
4517	  nop
4518
4519	/*
4520	 * VA range check
4521	 * On fail: goto sfmmu_tsb_miss
4522	 */
4523	ldx	[%g6 + KPMTSBM_VBASE], %g7
4524	cmp	%g2, %g7
4525	blu,pn	%xcc, sfmmu_tsb_miss
4526	  ldx	[%g6 + KPMTSBM_VEND], %g5
4527	cmp	%g2, %g5
4528	bgeu,pn	%xcc, sfmmu_tsb_miss
4529	  stx	%g1, [%g6 + KPMTSBM_TSBPTR]	/* save 8K kpm TSB pointer */
4530
4531	/*
4532	 * check TL tsbmiss handling flag
4533	 * bump tsbmiss counter
4534	 */
4535	lduw	[%g6 + KPMTSBM_TSBMISS], %g5
4536#ifdef	DEBUG
4537	and	%g4, KPMTSBM_TLTSBM_FLAG, %g1
4538	inc	%g5
4539	brz,pn	%g1, sfmmu_kpm_exception
4540	  st	%g5, [%g6 + KPMTSBM_TSBMISS]
4541#else
4542	inc	%g5
4543	st	%g5, [%g6 + KPMTSBM_TSBMISS]
4544#endif
4545	/*
4546	 * At this point:
4547	 *  g1 = clobbered
4548	 *  g2 = tag access register
4549	 *  g3 = 4M kpm TSB pointer (not used)
4550	 *  g6 = per-CPU kpm tsbmiss area
4551	 *  g7 = kpm_vbase
4552	 */
4553
4554	/* vaddr2pfn */
4555	ldub	[%g6 + KPMTSBM_SZSHIFT], %g3
4556	sub	%g2, %g7, %g4			/* paddr = vaddr-kpm_vbase */
4557	srax    %g4, %g3, %g2			/* which alias range (r) */
4558	brnz,pn	%g2, sfmmu_kpm_exception	/* if (r != 0) goto C handler */
4559	  srlx	%g4, MMU_PAGESHIFT, %g2		/* %g2 = pfn */
4560
4561	/*
4562	 * Setup %asi
4563	 * mseg_pa = page_numtomemseg_nolock_pa(pfn)
4564	 * if (mseg not found) sfmmu_kpm_exception
4565	 * g2=pfn
4566	 */
4567	mov	ASI_MEM, %asi
4568	PAGE_NUM2MEMSEG_NOLOCK_PA(%g2, %g3, %g6, %g4, %g5, %g7, kpmtsbmsp2m)
4569	cmp	%g3, MSEG_NULLPTR_PA
4570	be,pn	%xcc, sfmmu_kpm_exception	/* if mseg not found */
4571	  nop
4572
4573	/*
4574	 * inx = pfn - mseg_pa->kpm_pbase
4575	 * g2=pfn g3=mseg_pa
4576	 */
4577	ldxa	[%g3 + MEMSEG_KPM_PBASE]%asi, %g7
4578	sub	%g2, %g7, %g4
4579
4580#ifdef	DEBUG
4581	/*
4582	 * Validate inx value
4583	 * g2=pfn g3=mseg_pa g4=inx
4584	 */
4585	ldxa	[%g3 + MEMSEG_KPM_NKPMPGS]%asi, %g5
4586	cmp	%g4, %g5			/* inx - nkpmpgs */
4587	bgeu,pn	%xcc, sfmmu_kpm_exception	/* if out of range */
4588	  ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4589#else
4590	ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4591#endif
4592	/* ksp = &mseg_pa->kpm_spages[inx] */
4593	ldxa	[%g3 + MEMSEG_KPM_SPAGES]%asi, %g5
4594	add	%g5, %g4, %g5			/* ksp */
4595
4596	/*
4597	 * KPMP_SHASH(kp)
4598	 * g2=pfn g3=mseg_pa g4=inx g5=ksp g7=kpmp_stable_sz
4599	 */
4600	ldub	[%g6 + KPMTSBM_KPMPSHIFT], %g1	/* kpmp_shift */
4601	sub	%g7, 1, %g7			/* mask */
4602	sllx	%g5, %g1, %g1			/* x = ksp << kpmp_shift */
4603	add	%g5, %g1, %g5			/* y = ksp + x */
4604	and 	%g5, %g7, %g5			/* hashinx = y & mask */
4605
4606	/*
4607	 * Calculate physical kpm_spage pointer
4608	 * g2=pfn g3=mseg_pa g4=offset g5=hashinx
4609	 */
4610	ldxa	[%g3 + MEMSEG_KPM_PAGESPA]%asi, %g1 /* kpm_spagespa */
4611	add	%g1, %g4, %g1			/* ksp_pa */
4612
4613	/*
4614	 * Calculate physical hash lock address.
4615	 * Note: Changes in kpm_shlk_t must be reflected here.
4616	 * g1=ksp_pa g2=pfn g5=hashinx
4617	 */
4618	ldx	[%g6 + KPMTSBM_KPMPTABLEPA], %g4 /* kpmp_stablepa */
4619	sllx	%g5, KPMSHLK_SHIFT, %g5
4620	add	%g4, %g5, %g3			/* hlck_pa */
4621
4622	/*
4623	 * Assemble tte
4624	 * g1=ksp_pa g2=pfn g3=hlck_pa
4625	 */
4626	sethi	%hi(TTE_VALID_INT), %g5		/* upper part */
4627	sllx	%g5, 32, %g5
4628	mov	(TTE_CP_INT|TTE_CV_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4629	or	%g5, %g4, %g5
4630	sllx	%g2, MMU_PAGESHIFT, %g4
4631	or	%g5, %g4, %g5			/* tte */
4632	ldx	[%g6 + KPMTSBM_TSBPTR], %g4
4633	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
4634
4635	/*
4636	 * tsb dropin
4637	 * g1=ksp_pa g2=ttarget g3=hlck_pa g4=ktsbp g5=tte
4638	 */
4639
4640	/* KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi) */
4641	KPMLOCK_ENTER(%g3, %g7, kpmtsbsmlock, ASI_MEM)
4642
4643	/* use C-handler if there's no go for dropin */
4644	ldsba	[%g1 + KPMSPAGE_MAPPED]%asi, %g7 /* kp_mapped */
4645	cmp	%g7, -1
4646	bne,pn	%xcc, 5f
4647	  nop
4648
4649#ifndef sun4v
4650	ldub	[%g6 + KPMTSBM_FLAGS], %g7
4651	mov	ASI_N, %g1
4652	andcc	%g7, KPMTSBM_TSBPHYS_FLAG, %g0
4653	movnz	%icc, ASI_MEM, %g1
4654	mov	%g1, %asi
4655#endif
4656
4657	/* TSB_LOCK_ENTRY(tsbp, tmp1, tmp2, label) (needs %asi set) */
4658	TSB_LOCK_ENTRY(%g4, %g1, %g7, 6)
4659
4660	/* TSB_INSERT_UNLOCK_ENTRY(tsbp, tte, tagtarget, tmp) */
4661	TSB_INSERT_UNLOCK_ENTRY(%g4, %g5, %g2, %g7)
4662
4663	DTLB_STUFF(%g5, %g2, %g4, %g5, %g6)
4664
4665	/* KPMLOCK_EXIT(kpmlckp, asi) */
4666	KPMLOCK_EXIT(%g3, ASI_MEM)
4667
4668	/*
4669	 * If trapstat is running, we need to shift the %tpc and %tnpc to
4670	 * point to trapstat's TSB miss return code (note that trapstat
4671	 * itself will patch the correct offset to add).
4672	 * Note: TTE is expected in %g5 (allows per pagesize reporting).
4673	 */
4674	rdpr	%tl, %g7
4675	cmp	%g7, 1
4676	ble	%icc, 0f
4677	sethi	%hi(KERNELBASE), %g6
4678	rdpr	%tpc, %g7
4679	or	%g6, %lo(KERNELBASE), %g6
4680	cmp	%g7, %g6
4681	bgeu	%xcc, 0f
4682	ALTENTRY(tsbmiss_trapstat_patch_point_kpm_small)
4683	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
4684	wrpr	%g7, %tpc
4685	add	%g7, 4, %g7
4686	wrpr	%g7, %tnpc
46870:
4688	retry
46895:
4690	/* g3=hlck_pa */
4691	KPMLOCK_EXIT(%g3, ASI_MEM)
4692	ba,pt	%icc, sfmmu_kpm_exception
4693	  nop
4694	SET_SIZE(sfmmu_kpm_dtsb_miss_small)
4695
4696#if (1<< KPMTSBM_SHIFT) != KPMTSBM_SIZE
4697#error - KPMTSBM_SHIFT does not correspond to size of kpmtsbm struct
4698#endif
4699
4700#endif /* lint */
4701
4702#ifdef	lint
4703/*
4704 * Enable/disable tsbmiss handling at trap level for a kpm (large) page.
4705 * Called from C-level, sets/clears "go" indication for trap level handler.
4706 * khl_lock is a low level spin lock to protect the kp_tsbmtl field.
4707 * Assumed that &kp->kp_refcntc is checked for zero or -1 at C-level.
4708 * Assumes khl_mutex is held when called from C-level.
4709 */
4710/* ARGSUSED */
4711void
4712sfmmu_kpm_tsbmtl(short *kp_refcntc, uint_t *khl_lock, int cmd)
4713{
4714}
4715
4716/*
4717 * kpm_smallpages: stores val to byte at address mapped within
4718 * low level lock brackets. The old value is returned.
4719 * Called from C-level.
4720 */
4721/* ARGSUSED */
4722int
4723sfmmu_kpm_stsbmtl(char *mapped, uint_t *kshl_lock, int val)
4724{
4725	return (0);
4726}
4727
4728#else /* lint */
4729
4730	.seg	".data"
4731sfmmu_kpm_tsbmtl_panic:
4732	.ascii	"sfmmu_kpm_tsbmtl: interrupts disabled"
4733	.byte	0
4734sfmmu_kpm_stsbmtl_panic:
4735	.ascii	"sfmmu_kpm_stsbmtl: interrupts disabled"
4736	.byte	0
4737	.align	4
4738	.seg	".text"
4739
4740	ENTRY_NP(sfmmu_kpm_tsbmtl)
4741	rdpr	%pstate, %o3
4742	/*
4743	 * %o0 = &kp_refcntc
4744	 * %o1 = &khl_lock
4745	 * %o2 = 0/1 (off/on)
4746	 * %o3 = pstate save
4747	 */
4748#ifdef DEBUG
4749	andcc	%o3, PSTATE_IE, %g0		/* if interrupts already */
4750	bnz,pt %icc, 1f				/* disabled, panic	 */
4751	  nop
4752	save	%sp, -SA(MINFRAME), %sp
4753	sethi	%hi(sfmmu_kpm_tsbmtl_panic), %o0
4754	call	panic
4755	 or	%o0, %lo(sfmmu_kpm_tsbmtl_panic), %o0
4756	ret
4757	restore
47581:
4759#endif /* DEBUG */
4760	wrpr	%o3, PSTATE_IE, %pstate		/* disable interrupts */
4761
4762	KPMLOCK_ENTER(%o1, %o4, kpmtsbmtl1, ASI_N)
4763	mov	-1, %o5
4764	brz,a	%o2, 2f
4765	  mov	0, %o5
47662:
4767	sth	%o5, [%o0]
4768	KPMLOCK_EXIT(%o1, ASI_N)
4769
4770	retl
4771	  wrpr	%g0, %o3, %pstate		/* enable interrupts */
4772	SET_SIZE(sfmmu_kpm_tsbmtl)
4773
4774	ENTRY_NP(sfmmu_kpm_stsbmtl)
4775	rdpr	%pstate, %o3
4776	/*
4777	 * %o0 = &mapped
4778	 * %o1 = &kshl_lock
4779	 * %o2 = val
4780	 * %o3 = pstate save
4781	 */
4782#ifdef DEBUG
4783	andcc	%o3, PSTATE_IE, %g0		/* if interrupts already */
4784	bnz,pt %icc, 1f				/* disabled, panic	 */
4785	  nop
4786	save	%sp, -SA(MINFRAME), %sp
4787	sethi	%hi(sfmmu_kpm_stsbmtl_panic), %o0
4788	call	panic
4789	  or	%o0, %lo(sfmmu_kpm_stsbmtl_panic), %o0
4790	ret
4791	restore
47921:
4793#endif /* DEBUG */
4794	wrpr	%o3, PSTATE_IE, %pstate		/* disable interrupts */
4795
4796	KPMLOCK_ENTER(%o1, %o4, kpmstsbmtl1, ASI_N)
4797	ldsb	[%o0], %o5
4798	stb	%o2, [%o0]
4799	KPMLOCK_EXIT(%o1, ASI_N)
4800
4801	mov	%o5, %o0			/* return old val */
4802	retl
4803	  wrpr	%g0, %o3, %pstate		/* enable interrupts */
4804	SET_SIZE(sfmmu_kpm_stsbmtl)
4805
4806#endif /* lint */
4807
4808#ifndef lint
4809#ifdef sun4v
4810	/*
4811	 * User/kernel data miss w// multiple TSBs
4812	 * The first probe covers 8K, 64K, and 512K page sizes,
4813	 * because 64K and 512K mappings are replicated off 8K
4814	 * pointer.  Second probe covers 4M page size only.
4815	 *
4816	 * MMU fault area contains miss address and context.
4817	 */
4818	ALTENTRY(sfmmu_slow_dmmu_miss)
4819	GET_MMU_D_PTAGACC_CTXTYPE(%g2, %g3)	! %g2 = ptagacc, %g3 = ctx type
4820
4821slow_miss_common:
4822	/*
4823	 *  %g2 = tagacc register (needed for sfmmu_tsb_miss_tt)
4824	 *  %g3 = ctx (cannot be INVALID_CONTEXT)
4825	 */
4826	brnz,pt	%g3, 8f			! check for user context
4827	  nop
4828
4829	/*
4830	 * Kernel miss
4831	 * Get 8K and 4M TSB pointers in %g1 and %g3 and
4832	 * branch to sfmmu_tsb_miss_tt to handle it.
4833	 */
4834	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
4835sfmmu_dslow_patch_ktsb_base:
4836	RUNTIME_PATCH_SETX(%g1, %g6)	! %g1 = contents of ktsb_pbase
4837sfmmu_dslow_patch_ktsb_szcode:
4838	or	%g0, RUNTIME_PATCH, %g3	! ktsb_szcode (hot patched)
4839
4840	GET_TSBE_POINTER(MMU_PAGESHIFT, %g1, %g7, %g3, %g5)
4841	! %g1 = First TSB entry pointer, as TSB miss handler expects
4842
4843	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
4844sfmmu_dslow_patch_ktsb4m_base:
4845	RUNTIME_PATCH_SETX(%g3, %g6)	! %g3 = contents of ktsb4m_pbase
4846sfmmu_dslow_patch_ktsb4m_szcode:
4847	or	%g0, RUNTIME_PATCH, %g6	! ktsb4m_szcode (hot patched)
4848
4849	GET_TSBE_POINTER(MMU_PAGESHIFT4M, %g3, %g7, %g6, %g5)
4850	! %g3 = 4M tsb entry pointer, as TSB miss handler expects
4851	ba,a,pt	%xcc, sfmmu_tsb_miss_tt
4852	.empty
4853
48548:
4855	/*
4856	 * User miss
4857	 * Get first TSB pointer in %g1
4858	 * Get second TSB pointer (or NULL if no second TSB) in %g3
4859	 * Branch to sfmmu_tsb_miss_tt to handle it
4860	 */
4861	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
4862	/* %g1 = first TSB entry ptr now, %g2 preserved */
4863
4864	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)	/* get 2nd utsbreg */
4865	brlz,pt %g3, sfmmu_tsb_miss_tt		/* done if no 2nd TSB */
4866	  nop
4867
4868	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
4869	/* %g3 = second TSB entry ptr now, %g2 preserved */
48709:
4871	ba,a,pt	%xcc, sfmmu_tsb_miss_tt
4872	.empty
4873	SET_SIZE(sfmmu_slow_dmmu_miss)
4874
4875
4876	/*
4877	 * User/kernel instruction miss w/ multiple TSBs
4878	 * The first probe covers 8K, 64K, and 512K page sizes,
4879	 * because 64K and 512K mappings are replicated off 8K
4880	 * pointer.  Second probe covers 4M page size only.
4881	 *
4882	 * MMU fault area contains miss address and context.
4883	 */
4884	ALTENTRY(sfmmu_slow_immu_miss)
4885	GET_MMU_I_PTAGACC_CTXTYPE(%g2, %g3)
4886	ba,a,pt	%xcc, slow_miss_common
4887	SET_SIZE(sfmmu_slow_immu_miss)
4888
4889#endif /* sun4v */
4890#endif	/* lint */
4891
4892#ifndef lint
4893
4894/*
4895 * Per-CPU tsbmiss areas to avoid cache misses in TSB miss handlers.
4896 */
4897	.seg	".data"
4898	.align	64
4899	.global tsbmiss_area
4900tsbmiss_area:
4901	.skip	(TSBMISS_SIZE * NCPU)
4902
4903	.align	64
4904	.global kpmtsbm_area
4905kpmtsbm_area:
4906	.skip	(KPMTSBM_SIZE * NCPU)
4907#endif	/* lint */
4908