xref: /titanic_50/usr/src/uts/sfmmu/ml/sfmmu_asm.s (revision ac7f5757903d7806e03e59f71c10eec36e0deade)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26/*
27 * SFMMU primitives.  These primitives should only be used by sfmmu
28 * routines.
29 */
30
31#if defined(lint)
32#include <sys/types.h>
33#else	/* lint */
34#include "assym.h"
35#endif	/* lint */
36
37#include <sys/asm_linkage.h>
38#include <sys/machtrap.h>
39#include <sys/machasi.h>
40#include <sys/sun4asi.h>
41#include <sys/pte.h>
42#include <sys/mmu.h>
43#include <vm/hat_sfmmu.h>
44#include <vm/seg_spt.h>
45#include <sys/machparam.h>
46#include <sys/privregs.h>
47#include <sys/scb.h>
48#include <sys/intreg.h>
49#include <sys/machthread.h>
50#include <sys/intr.h>
51#include <sys/clock.h>
52#include <sys/trapstat.h>
53
54#ifdef TRAPTRACE
55#include <sys/traptrace.h>
56
57/*
58 * Tracing macro. Adds two instructions if TRAPTRACE is defined.
59 */
60#define	TT_TRACE(label)		\
61	ba	label		;\
62	rd	%pc, %g7
63#else
64
65#define	TT_TRACE(label)
66
67#endif /* TRAPTRACE */
68
69#ifndef	lint
70
71#if (TTE_SUSPEND_SHIFT > 0)
72#define	TTE_SUSPEND_INT_SHIFT(reg)				\
73	sllx	reg, TTE_SUSPEND_SHIFT, reg
74#else
75#define	TTE_SUSPEND_INT_SHIFT(reg)
76#endif
77
78#endif /* lint */
79
80#ifndef	lint
81
82/*
83 * Assumes TSBE_TAG is 0
84 * Assumes TSBE_INTHI is 0
85 * Assumes TSBREG.split is 0
86 */
87
88#if TSBE_TAG != 0
89#error "TSB_UPDATE and TSB_INVALIDATE assume TSBE_TAG = 0"
90#endif
91
92#if TSBTAG_INTHI != 0
93#error "TSB_UPDATE and TSB_INVALIDATE assume TSBTAG_INTHI = 0"
94#endif
95
96/*
97 * The following code assumes the tsb is not split.
98 *
99 * With TSBs no longer shared between processes, it's no longer
100 * necessary to hash the context bits into the tsb index to get
101 * tsb coloring; the new implementation treats the TSB as a
102 * direct-mapped, virtually-addressed cache.
103 *
104 * In:
105 *    vpshift = virtual page shift; e.g. 13 for 8K TTEs (constant or ro)
106 *    tsbbase = base address of TSB (clobbered)
107 *    tagacc = tag access register (clobbered)
108 *    szc = size code of TSB (ro)
109 *    tmp = scratch reg
110 * Out:
111 *    tsbbase = pointer to entry in TSB
112 */
113#define	GET_TSBE_POINTER(vpshift, tsbbase, tagacc, szc, tmp)		\
114	mov	TSB_ENTRIES(0), tmp	/* nentries in TSB size 0 */	;\
115	srlx	tagacc, vpshift, tagacc 				;\
116	sllx	tmp, szc, tmp		/* tmp = nentries in TSB */	;\
117	sub	tmp, 1, tmp		/* mask = nentries - 1 */	;\
118	and	tagacc, tmp, tmp	/* tsbent = virtpage & mask */	;\
119	sllx	tmp, TSB_ENTRY_SHIFT, tmp	/* entry num --> ptr */	;\
120	add	tsbbase, tmp, tsbbase	/* add entry offset to TSB base */
121
122/*
123 * When the kpm TSB is used it is assumed that it is direct mapped
124 * using (vaddr>>vpshift)%tsbsz as the index.
125 *
126 * Note that, for now, the kpm TSB and kernel TSB are the same for
127 * each mapping size.  However that need not always be the case.  If
128 * the trap handlers are updated to search a different TSB for kpm
129 * addresses than for kernel addresses then kpm_tsbbase and kpm_tsbsz
130 * (and/or kpmsm_tsbbase/kpmsm_tsbsz) may be entirely independent.
131 *
132 * In:
133 *    vpshift = virtual page shift; e.g. 13 for 8K TTEs (constant or ro)
134 *    vaddr = virtual address (clobbered)
135 *    tsbp, szc, tmp = scratch
136 * Out:
137 *    tsbp = pointer to entry in TSB
138 */
139#define	GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr, szc, tmp)		\
140	cmp	vpshift, MMU_PAGESHIFT					;\
141	bne,pn	%icc, 1f		/* branch if large case */	;\
142	  sethi	%hi(kpmsm_tsbsz), szc					;\
143	sethi	%hi(kpmsm_tsbbase), tsbp				;\
144	ld	[szc + %lo(kpmsm_tsbsz)], szc				;\
145	ldx	[tsbp + %lo(kpmsm_tsbbase)], tsbp			;\
146	ba,pt	%icc, 2f						;\
147	  nop								;\
1481:	sethi	%hi(kpm_tsbsz), szc					;\
149	sethi	%hi(kpm_tsbbase), tsbp					;\
150	ld	[szc + %lo(kpm_tsbsz)], szc				;\
151	ldx	[tsbp + %lo(kpm_tsbbase)], tsbp				;\
1522:	GET_TSBE_POINTER(vpshift, tsbp, vaddr, szc, tmp)
153
154/*
155 * Lock the TSBE at virtual address tsbep.
156 *
157 * tsbep = TSBE va (ro)
158 * tmp1, tmp2 = scratch registers (clobbered)
159 * label = label to jump to if we fail to lock the tsb entry
160 * %asi = ASI to use for TSB access
161 *
162 * NOTE that we flush the TSB using fast VIS instructions that
163 * set all 1's in the TSB tag, so TSBTAG_LOCKED|TSBTAG_INVALID must
164 * not be treated as a locked entry or we'll get stuck spinning on
165 * an entry that isn't locked but really invalid.
166 */
167
168#if defined(UTSB_PHYS)
169
170#define	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			\
171	lda	[tsbep]ASI_MEM, tmp1					;\
172	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
173	cmp	tmp1, tmp2 						;\
174	be,a,pn	%icc, label		/* if locked ignore */		;\
175	  nop								;\
176	casa	[tsbep]ASI_MEM, tmp1, tmp2				;\
177	cmp	tmp1, tmp2 						;\
178	bne,a,pn %icc, label		/* didn't lock so ignore */	;\
179	  nop								;\
180	/* tsbe lock acquired */					;\
181	membar #StoreStore
182
183#else /* UTSB_PHYS */
184
185#define	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			\
186	lda	[tsbep]%asi, tmp1					;\
187	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
188	cmp	tmp1, tmp2 						;\
189	be,a,pn	%icc, label		/* if locked ignore */		;\
190	  nop								;\
191	casa	[tsbep]%asi, tmp1, tmp2					;\
192	cmp	tmp1, tmp2 						;\
193	bne,a,pn %icc, label		/* didn't lock so ignore */	;\
194	  nop								;\
195	/* tsbe lock acquired */					;\
196	membar #StoreStore
197
198#endif /* UTSB_PHYS */
199
200/*
201 * Atomically write TSBE at virtual address tsbep.
202 *
203 * tsbep = TSBE va (ro)
204 * tte = TSBE TTE (ro)
205 * tagtarget = TSBE tag (ro)
206 * %asi = ASI to use for TSB access
207 */
208
209#if defined(UTSB_PHYS)
210
211#define	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		\
212	add	tsbep, TSBE_TTE, tmp1					;\
213	stxa	tte, [tmp1]ASI_MEM		/* write tte data */	;\
214	membar #StoreStore						;\
215	add	tsbep, TSBE_TAG, tmp1					;\
216	stxa	tagtarget, [tmp1]ASI_MEM	/* write tte tag & unlock */
217
218#else /* UTSB_PHYS */
219
220#define	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget,tmp1)		\
221	stxa	tte, [tsbep + TSBE_TTE]%asi	/* write tte data */	;\
222	membar #StoreStore						;\
223	stxa	tagtarget, [tsbep + TSBE_TAG]%asi /* write tte tag & unlock */
224
225#endif /* UTSB_PHYS */
226
227/*
228 * Load an entry into the TSB at TL > 0.
229 *
230 * tsbep = pointer to the TSBE to load as va (ro)
231 * tte = value of the TTE retrieved and loaded (wo)
232 * tagtarget = tag target register.  To get TSBE tag to load,
233 *   we need to mask off the context and leave only the va (clobbered)
234 * ttepa = pointer to the TTE to retrieve/load as pa (ro)
235 * tmp1, tmp2 = scratch registers
236 * label = label to jump to if we fail to lock the tsb entry
237 * %asi = ASI to use for TSB access
238 */
239
240#if defined(UTSB_PHYS)
241
242#define	TSB_UPDATE_TL(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
243	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
244	/*								;\
245	 * I don't need to update the TSB then check for the valid tte.	;\
246	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
247	 * we always invalidate the hash table before we unload the TSB.;\
248	 */								;\
249	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
250	ldxa	[ttepa]ASI_MEM, tte					;\
251	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
252	sethi	%hi(TSBTAG_INVALID), tmp2				;\
253	add	tsbep, TSBE_TAG, tmp1					;\
254	brgez,a,pn tte, label						;\
255	 sta	tmp2, [tmp1]ASI_MEM			/* unlock */	;\
256	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
257label:
258
259#else /* UTSB_PHYS */
260
261#define	TSB_UPDATE_TL(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
262	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
263	/*								;\
264	 * I don't need to update the TSB then check for the valid tte.	;\
265	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
266	 * we always invalidate the hash table before we unload the TSB.;\
267	 */								;\
268	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
269	ldxa	[ttepa]ASI_MEM, tte					;\
270	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
271	sethi	%hi(TSBTAG_INVALID), tmp2				;\
272	brgez,a,pn tte, label						;\
273	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
274	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
275label:
276
277#endif /* UTSB_PHYS */
278
279/*
280 * Load a 32M/256M Panther TSB entry into the TSB at TL > 0,
281 *   for ITLB synthesis.
282 *
283 * tsbep = pointer to the TSBE to load as va (ro)
284 * tte = 4M pfn offset (in), value of the TTE retrieved and loaded (out)
285 *   with exec_perm turned off and exec_synth turned on
286 * tagtarget = tag target register.  To get TSBE tag to load,
287 *   we need to mask off the context and leave only the va (clobbered)
288 * ttepa = pointer to the TTE to retrieve/load as pa (ro)
289 * tmp1, tmp2 = scratch registers
290 * label = label to use for branch (text)
291 * %asi = ASI to use for TSB access
292 */
293
294#define	TSB_UPDATE_TL_PN(tsbep, tte, tagtarget, ttepa, tmp1, tmp2, label) \
295	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
296	/*								;\
297	 * I don't need to update the TSB then check for the valid tte.	;\
298	 * TSB invalidate will spin till the entry is unlocked.	Note,	;\
299	 * we always invalidate the hash table before we unload the TSB.;\
300	 * Or in 4M pfn offset to TTE and set the exec_perm bit to 0	;\
301	 * and exec_synth bit to 1.					;\
302	 */								;\
303	sllx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
304	mov	tte, tmp1						;\
305	ldxa	[ttepa]ASI_MEM, tte					;\
306	srlx	tagtarget, TTARGET_VA_SHIFT, tagtarget			;\
307	sethi	%hi(TSBTAG_INVALID), tmp2				;\
308	brgez,a,pn tte, label						;\
309	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
310	or	tte, tmp1, tte						;\
311	andn	tte, TTE_EXECPRM_INT, tte				;\
312	or	tte, TTE_E_SYNTH_INT, tte				;\
313	TSB_INSERT_UNLOCK_ENTRY(tsbep, tte, tagtarget, tmp1)		;\
314label:
315
316/*
317 * Build a 4M pfn offset for a Panther 32M/256M page, for ITLB synthesis.
318 *
319 * tte = value of the TTE, used to get tte_size bits (ro)
320 * tagaccess = tag access register, used to get 4M pfn bits (ro)
321 * pfn = 4M pfn bits shifted to offset for tte (out)
322 * tmp1 = scratch register
323 * label = label to use for branch (text)
324 */
325
326#define	GET_4M_PFN_OFF(tte, tagaccess, pfn, tmp, label)			\
327	/*								;\
328	 * Get 4M bits from tagaccess for 32M, 256M pagesizes.		;\
329	 * Return them, shifted, in pfn.				;\
330	 */								;\
331	srlx	tagaccess, MMU_PAGESHIFT4M, tagaccess			;\
332	srlx	tte, TTE_SZ_SHFT, tmp		/* isolate the */	;\
333	andcc	tmp, TTE_SZ_BITS, %g0		/* tte_size bits */	;\
334	bz,a,pt %icc, label/**/f		/* if 0, is */		;\
335	  and	tagaccess, 0x7, tagaccess	/* 32M page size */	;\
336	and	tagaccess, 0x3f, tagaccess /* else 256M page size */	;\
337label:									;\
338	sllx	tagaccess, MMU_PAGESHIFT4M, pfn
339
340/*
341 * Add 4M TTE size code to a tte for a Panther 32M/256M page,
342 * for ITLB synthesis.
343 *
344 * tte = value of the TTE, used to get tte_size bits (rw)
345 * tmp1 = scratch register
346 */
347
348#define	SET_TTE4M_PN(tte, tmp)						\
349	/*								;\
350	 * Set 4M pagesize tte bits. 					;\
351	 */								;\
352	set	TTE4M, tmp						;\
353	sllx	tmp, TTE_SZ_SHFT, tmp					;\
354	or	tte, tmp, tte
355
356/*
357 * Load an entry into the TSB at TL=0.
358 *
359 * tsbep = pointer to the TSBE to load as va (ro)
360 * tteva = pointer to the TTE to load as va (ro)
361 * tagtarget = TSBE tag to load (which contains no context), synthesized
362 * to match va of MMU tag target register only (ro)
363 * tmp1, tmp2 = scratch registers (clobbered)
364 * label = label to use for branches (text)
365 * %asi = ASI to use for TSB access
366 */
367
368#if defined(UTSB_PHYS)
369
370#define	TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label)		\
371	/* can't rd tteva after locking tsb because it can tlb miss */	;\
372	ldx	[tteva], tteva			/* load tte */		;\
373	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
374	sethi	%hi(TSBTAG_INVALID), tmp2				;\
375	add	tsbep, TSBE_TAG, tmp1					;\
376	brgez,a,pn tteva, label						;\
377	 sta	tmp2, [tmp1]ASI_MEM			/* unlock */	;\
378	TSB_INSERT_UNLOCK_ENTRY(tsbep, tteva, tagtarget, tmp1)		;\
379label:
380
381#else /* UTSB_PHYS */
382
383#define	TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label)		\
384	/* can't rd tteva after locking tsb because it can tlb miss */	;\
385	ldx	[tteva], tteva			/* load tte */		;\
386	TSB_LOCK_ENTRY(tsbep, tmp1, tmp2, label)			;\
387	sethi	%hi(TSBTAG_INVALID), tmp2				;\
388	brgez,a,pn tteva, label						;\
389	 sta	tmp2, [tsbep + TSBE_TAG]%asi		/* unlock */	;\
390	TSB_INSERT_UNLOCK_ENTRY(tsbep, tteva, tagtarget, tmp1)		;\
391label:
392
393#endif /* UTSB_PHYS */
394
395/*
396 * Invalidate a TSB entry in the TSB.
397 *
398 * NOTE: TSBE_TAG is assumed to be zero.  There is a compile time check
399 *	 about this earlier to ensure this is true.  Thus when we are
400 *	 directly referencing tsbep below, we are referencing the tte_tag
401 *	 field of the TSBE.  If this  offset ever changes, the code below
402 *	 will need to be modified.
403 *
404 * tsbep = pointer to TSBE as va (ro)
405 * tag = invalidation is done if this matches the TSBE tag (ro)
406 * tmp1 - tmp3 = scratch registers (clobbered)
407 * label = label name to use for branches (text)
408 * %asi = ASI to use for TSB access
409 */
410
411#if defined(UTSB_PHYS)
412
413#define	TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label)		\
414	lda	[tsbep]ASI_MEM, tmp1	/* tmp1 = tsbe tag */		;\
415	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
416label/**/1:								;\
417	cmp	tmp1, tmp2		/* see if tsbe is locked, if */	;\
418	be,a,pn	%icc, label/**/1	/* so, loop until unlocked */	;\
419	  lda	[tsbep]ASI_MEM, tmp1	/* reloading value each time */	;\
420	ldxa	[tsbep]ASI_MEM, tmp3	/* tmp3 = tsbe tag */		;\
421	cmp	tag, tmp3		/* compare tags */		;\
422	bne,pt	%xcc, label/**/2	/* if different, do nothing */	;\
423	  sethi	%hi(TSBTAG_INVALID), tmp3				;\
424	casa	[tsbep]ASI_MEM, tmp1, tmp3 /* try to set tag invalid */	;\
425	cmp	tmp1, tmp3		/* if not successful */		;\
426	bne,a,pn %icc, label/**/1	/* start over from the top */	;\
427	  lda	[tsbep]ASI_MEM, tmp1	/* reloading tsbe tag */	;\
428label/**/2:
429
430#else /* UTSB_PHYS */
431
432#define	TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label)		\
433	lda	[tsbep]%asi, tmp1	/* tmp1 = tsbe tag */		;\
434	sethi	%hi(TSBTAG_LOCKED), tmp2				;\
435label/**/1:								;\
436	cmp	tmp1, tmp2		/* see if tsbe is locked, if */	;\
437	be,a,pn	%icc, label/**/1	/* so, loop until unlocked */	;\
438	  lda	[tsbep]%asi, tmp1	/* reloading value each time */	;\
439	ldxa	[tsbep]%asi, tmp3	/* tmp3 = tsbe tag */		;\
440	cmp	tag, tmp3		/* compare tags */		;\
441	bne,pt	%xcc, label/**/2	/* if different, do nothing */	;\
442	  sethi	%hi(TSBTAG_INVALID), tmp3				;\
443	casa	[tsbep]%asi, tmp1, tmp3	/* try to set tag invalid */	;\
444	cmp	tmp1, tmp3		/* if not successful */		;\
445	bne,a,pn %icc, label/**/1	/* start over from the top */	;\
446	  lda	[tsbep]%asi, tmp1	/* reloading tsbe tag */	;\
447label/**/2:
448
449#endif /* UTSB_PHYS */
450
451#if TSB_SOFTSZ_MASK < TSB_SZ_MASK
452#error	- TSB_SOFTSZ_MASK too small
453#endif
454
455
456/*
457 * An implementation of setx which will be hot patched at run time.
458 * since it is being hot patched, there is no value passed in.
459 * Thus, essentially we are implementing
460 *	setx value, tmp, dest
461 * where value is RUNTIME_PATCH (aka 0) in this case.
462 */
463#define	RUNTIME_PATCH_SETX(dest, tmp)					\
464	sethi	%hh(RUNTIME_PATCH), tmp					;\
465	sethi	%lm(RUNTIME_PATCH), dest				;\
466	or	tmp, %hm(RUNTIME_PATCH), tmp				;\
467	or	dest, %lo(RUNTIME_PATCH), dest				;\
468	sllx	tmp, 32, tmp						;\
469	nop				/* for perf reasons */		;\
470	or	tmp, dest, dest		/* contents of patched value */
471
472#endif /* lint */
473
474
475#if defined (lint)
476
477/*
478 * sfmmu related subroutines
479 */
480uint_t
481sfmmu_disable_intrs()
482{ return(0); }
483
484/* ARGSUSED */
485void
486sfmmu_enable_intrs(uint_t pstate_save)
487{}
488
489/* ARGSUSED */
490int
491sfmmu_alloc_ctx(sfmmu_t *sfmmup, int allocflag, struct cpu *cp, int shflag)
492{ return(0); }
493
494/*
495 * Use cas, if tte has changed underneath us then reread and try again.
496 * In the case of a retry, it will update sttep with the new original.
497 */
498/* ARGSUSED */
499int
500sfmmu_modifytte(tte_t *sttep, tte_t *stmodttep, tte_t *dttep)
501{ return(0); }
502
503/*
504 * Use cas, if tte has changed underneath us then return 1, else return 0
505 */
506/* ARGSUSED */
507int
508sfmmu_modifytte_try(tte_t *sttep, tte_t *stmodttep, tte_t *dttep)
509{ return(0); }
510
511/* ARGSUSED */
512void
513sfmmu_copytte(tte_t *sttep, tte_t *dttep)
514{}
515
516/*ARGSUSED*/
517struct tsbe *
518sfmmu_get_tsbe(uint64_t tsbeptr, caddr_t vaddr, int vpshift, int tsb_szc)
519{ return(0); }
520
521/*ARGSUSED*/
522uint64_t
523sfmmu_make_tsbtag(caddr_t va)
524{ return(0); }
525
526#else	/* lint */
527
528	.seg	".data"
529	.global	sfmmu_panic1
530sfmmu_panic1:
531	.asciz	"sfmmu_asm: interrupts already disabled"
532
533	.global	sfmmu_panic3
534sfmmu_panic3:
535	.asciz	"sfmmu_asm: sfmmu_vatopfn called for user"
536
537	.global	sfmmu_panic4
538sfmmu_panic4:
539	.asciz	"sfmmu_asm: 4M tsb pointer mis-match"
540
541	.global	sfmmu_panic5
542sfmmu_panic5:
543	.asciz	"sfmmu_asm: no unlocked TTEs in TLB 0"
544
545	.global	sfmmu_panic6
546sfmmu_panic6:
547	.asciz	"sfmmu_asm: interrupts not disabled"
548
549	.global	sfmmu_panic7
550sfmmu_panic7:
551	.asciz	"sfmmu_asm: kernel as"
552
553	.global	sfmmu_panic8
554sfmmu_panic8:
555	.asciz	"sfmmu_asm: gnum is zero"
556
557	.global	sfmmu_panic9
558sfmmu_panic9:
559	.asciz	"sfmmu_asm: cnum is greater than MAX_SFMMU_CTX_VAL"
560
561	.global	sfmmu_panic10
562sfmmu_panic10:
563	.asciz	"sfmmu_asm: valid SCD with no 3rd scd TSB"
564
565	.global	sfmmu_panic11
566sfmmu_panic11:
567	.asciz	"sfmmu_asm: ktsb_phys must not be 0 on a sun4v platform"
568
569        ENTRY(sfmmu_disable_intrs)
570        rdpr    %pstate, %o0
571#ifdef DEBUG
572	PANIC_IF_INTR_DISABLED_PSTR(%o0, sfmmu_di_l0, %g1)
573#endif /* DEBUG */
574        retl
575          wrpr   %o0, PSTATE_IE, %pstate
576        SET_SIZE(sfmmu_disable_intrs)
577
578	ENTRY(sfmmu_enable_intrs)
579        retl
580          wrpr    %g0, %o0, %pstate
581        SET_SIZE(sfmmu_enable_intrs)
582
583/*
584 * This routine is called both by resume() and sfmmu_get_ctx() to
585 * allocate a new context for the process on a MMU.
586 * if allocflag == 1, then alloc ctx when HAT mmu cnum == INVALID .
587 * if allocflag == 0, then do not alloc ctx if HAT mmu cnum == INVALID, which
588 * is the case when sfmmu_alloc_ctx is called from resume().
589 *
590 * The caller must disable interrupts before entering this routine.
591 * To reduce ctx switch overhead, the code contains both 'fast path' and
592 * 'slow path' code. The fast path code covers the common case where only
593 * a quick check is needed and the real ctx allocation is not required.
594 * It can be done without holding the per-process (PP) lock.
595 * The 'slow path' code must be protected by the PP Lock and performs ctx
596 * allocation.
597 * Hardware context register and HAT mmu cnum are updated accordingly.
598 *
599 * %o0 - sfmmup
600 * %o1 - allocflag
601 * %o2 - CPU
602 * %o3 - sfmmu private/shared flag
603 *
604 * ret - 0: no ctx is allocated
605 *       1: a ctx is allocated
606 */
607        ENTRY_NP(sfmmu_alloc_ctx)
608
609#ifdef DEBUG
610	sethi   %hi(ksfmmup), %g1
611	ldx     [%g1 + %lo(ksfmmup)], %g1
612	cmp     %g1, %o0
613	bne,pt   %xcc, 0f
614	  nop
615
616	sethi   %hi(panicstr), %g1		! if kernel as, panic
617        ldx     [%g1 + %lo(panicstr)], %g1
618        tst     %g1
619        bnz,pn  %icc, 7f
620          nop
621
622	sethi	%hi(sfmmu_panic7), %o0
623	call	panic
624	  or	%o0, %lo(sfmmu_panic7), %o0
625
6267:
627	retl
628	  mov	%g0, %o0			! %o0 = ret = 0
629
6300:
631	PANIC_IF_INTR_ENABLED_PSTR(sfmmu_ei_l1, %g1)
632#endif /* DEBUG */
633
634	mov	%o3, %g1			! save sfmmu pri/sh flag in %g1
635
636	! load global mmu_ctxp info
637	ldx	[%o2 + CPU_MMU_CTXP], %o3		! %o3 = mmu_ctx_t ptr
638        lduw	[%o2 + CPU_MMU_IDX], %g2		! %g2 = mmu index
639
640	! load global mmu_ctxp gnum
641	ldx	[%o3 + MMU_CTX_GNUM], %o4		! %o4 = mmu_ctxp->gnum
642
643#ifdef DEBUG
644	cmp	%o4, %g0		! mmu_ctxp->gnum should never be 0
645	bne,pt	%xcc, 3f
646	  nop
647
648	sethi   %hi(panicstr), %g1	! test if panicstr is already set
649        ldx     [%g1 + %lo(panicstr)], %g1
650        tst     %g1
651        bnz,pn  %icc, 1f
652          nop
653
654	sethi	%hi(sfmmu_panic8), %o0
655	call	panic
656	  or	%o0, %lo(sfmmu_panic8), %o0
6571:
658	retl
659	  mov	%g0, %o0			! %o0 = ret = 0
6603:
661#endif
662
663	! load HAT sfmmu_ctxs[mmuid] gnum, cnum
664
665	sllx	%g2, SFMMU_MMU_CTX_SHIFT, %g2
666	add	%o0, %g2, %g2		! %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
667
668	/*
669	 * %g5 = sfmmu gnum returned
670	 * %g6 = sfmmu cnum returned
671	 * %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
672	 * %g4 = scratch
673	 *
674	 * Fast path code, do a quick check.
675	 */
676	SFMMU_MMUID_GNUM_CNUM(%g2, %g5, %g6, %g4)
677
678	cmp	%g6, INVALID_CONTEXT		! hat cnum == INVALID ??
679	bne,pt	%icc, 1f			! valid hat cnum, check gnum
680	  nop
681
682	! cnum == INVALID, check allocflag
683	mov	%g0, %g4	! %g4 = ret = 0
684	brz,pt  %o1, 8f		! allocflag == 0, skip ctx allocation, bail
685	  mov	%g6, %o1
686
687	! (invalid HAT cnum) && (allocflag == 1)
688	ba,pt	%icc, 2f
689	  nop
6901:
691	! valid HAT cnum, check gnum
692	cmp	%g5, %o4
693	mov	1, %g4				!%g4 = ret = 1
694	be,a,pt	%icc, 8f			! gnum unchanged, go to done
695	  mov	%g6, %o1
696
6972:
698	/*
699	 * Grab per process (PP) sfmmu_ctx_lock spinlock,
700	 * followed by the 'slow path' code.
701	 */
702	ldstub	[%o0 + SFMMU_CTX_LOCK], %g3	! %g3 = per process (PP) lock
7033:
704	brz	%g3, 5f
705	  nop
7064:
707	brnz,a,pt       %g3, 4b				! spin if lock is 1
708	  ldub	[%o0 + SFMMU_CTX_LOCK], %g3
709	ba	%xcc, 3b				! retry the lock
710	  ldstub	[%o0 + SFMMU_CTX_LOCK], %g3    ! %g3 = PP lock
711
7125:
713	membar  #LoadLoad
714	/*
715	 * %g5 = sfmmu gnum returned
716	 * %g6 = sfmmu cnum returned
717	 * %g2 = &sfmmu_ctxs[mmuid] - SFMMU_CTXS
718	 * %g4 = scratch
719	 */
720	SFMMU_MMUID_GNUM_CNUM(%g2, %g5, %g6, %g4)
721
722	cmp	%g6, INVALID_CONTEXT		! hat cnum == INVALID ??
723	bne,pt	%icc, 1f			! valid hat cnum, check gnum
724	  nop
725
726	! cnum == INVALID, check allocflag
727	mov	%g0, %g4	! %g4 = ret = 0
728	brz,pt	%o1, 2f		! allocflag == 0, called from resume, set hw
729	  mov	%g6, %o1
730
731	! (invalid HAT cnum) && (allocflag == 1)
732	ba,pt	%icc, 6f
733	  nop
7341:
735	! valid HAT cnum, check gnum
736	cmp	%g5, %o4
737	mov	1, %g4				! %g4 = ret  = 1
738	be,a,pt	%icc, 2f			! gnum unchanged, go to done
739	  mov	%g6, %o1
740
741	ba,pt	%icc, 6f
742	  nop
7432:
744	membar  #LoadStore|#StoreStore
745	ba,pt %icc, 8f
746	  clrb  [%o0 + SFMMU_CTX_LOCK]
7476:
748	/*
749	 * We get here if we do not have a valid context, or
750	 * the HAT gnum does not match global gnum. We hold
751	 * sfmmu_ctx_lock spinlock. Allocate that context.
752	 *
753	 * %o3 = mmu_ctxp
754	 */
755	add	%o3, MMU_CTX_CNUM, %g3
756	ld	[%o3 + MMU_CTX_NCTXS], %g4
757
758	/*
759         * %g2 = &sfmmu_ctx_t[mmuid] - SFMMU_CTXS;
760         * %g3 = mmu cnum address
761	 * %g4 = mmu nctxs
762	 *
763	 * %o0 = sfmmup
764	 * %o1 = mmu current cnum value (used as new cnum)
765	 * %o4 = mmu gnum
766	 *
767	 * %o5 = scratch
768	 */
769	ld	[%g3], %o1
7700:
771	cmp	%o1, %g4
772	bl,a,pt %icc, 1f
773	  add	%o1, 1, %o5		! %o5 = mmu_ctxp->cnum + 1
774
775	/*
776	 * cnum reachs max, bail, so wrap around can be performed later.
777	 */
778	set	INVALID_CONTEXT, %o1
779	mov	%g0, %g4		! %g4 = ret = 0
780
781	membar  #LoadStore|#StoreStore
782	ba,pt	%icc, 8f
783	  clrb	[%o0 + SFMMU_CTX_LOCK]
7841:
785	! %g3 = addr of mmu_ctxp->cnum
786	! %o5 = mmu_ctxp->cnum + 1
787	cas	[%g3], %o1, %o5
788	cmp	%o1, %o5
789	bne,a,pn %xcc, 0b	! cas failed
790	  ld	[%g3], %o1
791
792#ifdef DEBUG
793        set	MAX_SFMMU_CTX_VAL, %o5
794	cmp	%o1, %o5
795	ble,pt %icc, 2f
796	  nop
797
798	sethi	%hi(sfmmu_panic9), %o0
799	call	panic
800	  or	%o0, %lo(sfmmu_panic9), %o0
8012:
802#endif
803	! update hat gnum and cnum
804	sllx	%o4, SFMMU_MMU_GNUM_RSHIFT, %o4
805	or	%o4, %o1, %o4
806	stx	%o4, [%g2 + SFMMU_CTXS]
807
808	membar  #LoadStore|#StoreStore
809	clrb	[%o0 + SFMMU_CTX_LOCK]
810
811	mov	1, %g4			! %g4 = ret = 1
8128:
813	/*
814	 * program the secondary context register
815	 *
816	 * %o1 = cnum
817	 * %g1 = sfmmu private/shared flag (0:private,  1:shared)
818	 */
819
820	/*
821	 * When we come here and context is invalid, we want to set both
822	 * private and shared ctx regs to INVALID. In order to
823	 * do so, we set the sfmmu priv/shared flag to 'private' regardless
824	 * so that private ctx reg will be set to invalid.
825	 * Note that on sun4v values written to private context register are
826	 * automatically written to corresponding shared context register as
827	 * well. On sun4u SET_SECCTX() will invalidate shared context register
828	 * when it sets a private secondary context register.
829	 */
830
831	cmp	%o1, INVALID_CONTEXT
832	be,a,pn	%icc, 9f
833	  clr	%g1
8349:
835
836#ifdef	sun4u
837	ldub	[%o0 + SFMMU_CEXT], %o2
838	sll	%o2, CTXREG_EXT_SHIFT, %o2
839	or	%o1, %o2, %o1
840#endif /* sun4u */
841
842	SET_SECCTX(%o1, %g1, %o4, %o5, alloc_ctx_lbl1)
843
844        retl
845          mov   %g4, %o0                        ! %o0 = ret
846
847	SET_SIZE(sfmmu_alloc_ctx)
848
849	ENTRY_NP(sfmmu_modifytte)
850	ldx	[%o2], %g3			/* current */
851	ldx	[%o0], %g1			/* original */
8522:
853	ldx	[%o1], %g2			/* modified */
854	cmp	%g2, %g3			/* is modified = current? */
855	be,a,pt	%xcc,1f				/* yes, don't write */
856	stx	%g3, [%o0]			/* update new original */
857	casx	[%o2], %g1, %g2
858	cmp	%g1, %g2
859	be,pt	%xcc, 1f			/* cas succeeded - return */
860	  nop
861	ldx	[%o2], %g3			/* new current */
862	stx	%g3, [%o0]			/* save as new original */
863	ba,pt	%xcc, 2b
864	  mov	%g3, %g1
8651:	retl
866	membar	#StoreLoad
867	SET_SIZE(sfmmu_modifytte)
868
869	ENTRY_NP(sfmmu_modifytte_try)
870	ldx	[%o1], %g2			/* modified */
871	ldx	[%o2], %g3			/* current */
872	ldx	[%o0], %g1			/* original */
873	cmp	%g3, %g2			/* is modified = current? */
874	be,a,pn %xcc,1f				/* yes, don't write */
875	mov	0, %o1				/* as if cas failed. */
876
877	casx	[%o2], %g1, %g2
878	membar	#StoreLoad
879	cmp	%g1, %g2
880	movne	%xcc, -1, %o1			/* cas failed. */
881	move	%xcc, 1, %o1			/* cas succeeded. */
8821:
883	stx	%g2, [%o0]			/* report "current" value */
884	retl
885	mov	%o1, %o0
886	SET_SIZE(sfmmu_modifytte_try)
887
888	ENTRY_NP(sfmmu_copytte)
889	ldx	[%o0], %g1
890	retl
891	stx	%g1, [%o1]
892	SET_SIZE(sfmmu_copytte)
893
894
895	/*
896	 * Calculate a TSB entry pointer for the given TSB, va, pagesize.
897	 * %o0 = TSB base address (in), pointer to TSB entry (out)
898	 * %o1 = vaddr (in)
899	 * %o2 = vpshift (in)
900	 * %o3 = tsb size code (in)
901	 * %o4 = scratch register
902	 */
903	ENTRY_NP(sfmmu_get_tsbe)
904	GET_TSBE_POINTER(%o2, %o0, %o1, %o3, %o4)
905	retl
906	nop
907	SET_SIZE(sfmmu_get_tsbe)
908
909	/*
910	 * Return a TSB tag for the given va.
911	 * %o0 = va (in/clobbered)
912	 * %o0 = va shifted to be in tsb tag format (with no context) (out)
913	 */
914	ENTRY_NP(sfmmu_make_tsbtag)
915	retl
916	srln	%o0, TTARGET_VA_SHIFT, %o0
917	SET_SIZE(sfmmu_make_tsbtag)
918
919#endif /* lint */
920
921/*
922 * Other sfmmu primitives
923 */
924
925
926#if defined (lint)
927void
928sfmmu_patch_ktsb(void)
929{
930}
931
932void
933sfmmu_kpm_patch_tlbm(void)
934{
935}
936
937void
938sfmmu_kpm_patch_tsbm(void)
939{
940}
941
942void
943sfmmu_patch_shctx(void)
944{
945}
946
947/* ARGSUSED */
948void
949sfmmu_load_tsbe(struct tsbe *tsbep, uint64_t vaddr, tte_t *ttep, int phys)
950{
951}
952
953/* ARGSUSED */
954void
955sfmmu_unload_tsbe(struct tsbe *tsbep, uint64_t vaddr, int phys)
956{
957}
958
959/* ARGSUSED */
960void
961sfmmu_kpm_load_tsb(caddr_t addr, tte_t *ttep, int vpshift)
962{
963}
964
965/* ARGSUSED */
966void
967sfmmu_kpm_unload_tsb(caddr_t addr, int vpshift)
968{
969}
970
971#else /* lint */
972
973#define	I_SIZE		4
974
975	ENTRY_NP(sfmmu_fix_ktlb_traptable)
976	/*
977	 * %o0 = start of patch area
978	 * %o1 = size code of TSB to patch
979	 * %o3 = scratch
980	 */
981	/* fix sll */
982	ld	[%o0], %o3			/* get sll */
983	sub	%o3, %o1, %o3			/* decrease shift by tsb szc */
984	st	%o3, [%o0]			/* write sll */
985	flush	%o0
986	/* fix srl */
987	add	%o0, I_SIZE, %o0		/* goto next instr. */
988	ld	[%o0], %o3			/* get srl */
989	sub	%o3, %o1, %o3			/* decrease shift by tsb szc */
990	st	%o3, [%o0]			/* write srl */
991	retl
992	flush	%o0
993	SET_SIZE(sfmmu_fix_ktlb_traptable)
994
995	ENTRY_NP(sfmmu_fixup_ktsbbase)
996	/*
997	 * %o0 = start of patch area
998	 * %o5 = kernel virtual or physical tsb base address
999	 * %o2, %o3 are used as scratch registers.
1000	 */
1001	/* fixup sethi instruction */
1002	ld	[%o0], %o3
1003	srl	%o5, 10, %o2			! offset is bits 32:10
1004	or	%o3, %o2, %o3			! set imm22
1005	st	%o3, [%o0]
1006	/* fixup offset of lduw/ldx */
1007	add	%o0, I_SIZE, %o0		! next instr
1008	ld	[%o0], %o3
1009	and	%o5, 0x3ff, %o2			! set imm13 to bits 9:0
1010	or	%o3, %o2, %o3
1011	st	%o3, [%o0]
1012	retl
1013	flush	%o0
1014	SET_SIZE(sfmmu_fixup_ktsbbase)
1015
1016	ENTRY_NP(sfmmu_fixup_setx)
1017	/*
1018	 * %o0 = start of patch area
1019	 * %o4 = 64 bit value to patch
1020	 * %o2, %o3 are used as scratch registers.
1021	 *
1022	 * Note: Assuming that all parts of the instructions which need to be
1023	 *	 patched correspond to RUNTIME_PATCH (aka 0)
1024	 *
1025	 * Note the implementation of setx which is being patched is as follows:
1026	 *
1027	 * sethi   %hh(RUNTIME_PATCH), tmp
1028	 * sethi   %lm(RUNTIME_PATCH), dest
1029	 * or      tmp, %hm(RUNTIME_PATCH), tmp
1030	 * or      dest, %lo(RUNTIME_PATCH), dest
1031	 * sllx    tmp, 32, tmp
1032	 * nop
1033	 * or      tmp, dest, dest
1034	 *
1035	 * which differs from the implementation in the
1036	 * "SPARC Architecture Manual"
1037	 */
1038	/* fixup sethi instruction */
1039	ld	[%o0], %o3
1040	srlx	%o4, 42, %o2			! bits [63:42]
1041	or	%o3, %o2, %o3			! set imm22
1042	st	%o3, [%o0]
1043	/* fixup sethi instruction */
1044	add	%o0, I_SIZE, %o0		! next instr
1045	ld	[%o0], %o3
1046	sllx	%o4, 32, %o2			! clear upper bits
1047	srlx	%o2, 42, %o2			! bits [31:10]
1048	or	%o3, %o2, %o3			! set imm22
1049	st	%o3, [%o0]
1050	/* fixup or instruction */
1051	add	%o0, I_SIZE, %o0		! next instr
1052	ld	[%o0], %o3
1053	srlx	%o4, 32, %o2			! bits [63:32]
1054	and	%o2, 0x3ff, %o2			! bits [41:32]
1055	or	%o3, %o2, %o3			! set imm
1056	st	%o3, [%o0]
1057	/* fixup or instruction */
1058	add	%o0, I_SIZE, %o0		! next instr
1059	ld	[%o0], %o3
1060	and	%o4, 0x3ff, %o2			! bits [9:0]
1061	or	%o3, %o2, %o3			! set imm
1062	st	%o3, [%o0]
1063	retl
1064	flush	%o0
1065	SET_SIZE(sfmmu_fixup_setx)
1066
1067	ENTRY_NP(sfmmu_fixup_or)
1068	/*
1069	 * %o0 = start of patch area
1070	 * %o4 = 32 bit value to patch
1071	 * %o2, %o3 are used as scratch registers.
1072	 * Note: Assuming that all parts of the instructions which need to be
1073	 *	 patched correspond to RUNTIME_PATCH (aka 0)
1074	 */
1075	ld	[%o0], %o3
1076	and	%o4, 0x3ff, %o2			! bits [9:0]
1077	or	%o3, %o2, %o3			! set imm
1078	st	%o3, [%o0]
1079	retl
1080	flush	%o0
1081	SET_SIZE(sfmmu_fixup_or)
1082
1083	ENTRY_NP(sfmmu_fixup_shiftx)
1084	/*
1085	 * %o0 = start of patch area
1086	 * %o4 = signed int immediate value to add to sllx/srlx imm field
1087	 * %o2, %o3 are used as scratch registers.
1088	 *
1089	 * sllx/srlx store the 6 bit immediate value in the lowest order bits
1090	 * so we do a simple add.  The caller must be careful to prevent
1091	 * overflow, which could easily occur if the initial value is nonzero!
1092	 */
1093	ld	[%o0], %o3			! %o3 = instruction to patch
1094	and	%o3, 0x3f, %o2			! %o2 = existing imm value
1095	add	%o2, %o4, %o2			! %o2 = new imm value
1096	andn	%o3, 0x3f, %o3			! clear old imm value
1097	and	%o2, 0x3f, %o2			! truncate new imm value
1098	or	%o3, %o2, %o3			! set new imm value
1099	st	%o3, [%o0]			! store updated instruction
1100	retl
1101	flush	%o0
1102	SET_SIZE(sfmmu_fixup_shiftx)
1103
1104	ENTRY_NP(sfmmu_fixup_mmu_asi)
1105	/*
1106	 * Patch imm_asi of all ldda instructions in the MMU
1107	 * trap handlers.  We search MMU_PATCH_INSTR instructions
1108	 * starting from the itlb miss handler (trap 0x64).
1109	 * %o0 = address of tt[0,1]_itlbmiss
1110	 * %o1 = imm_asi to setup, shifted by appropriate offset.
1111	 * %o3 = number of instructions to search
1112	 * %o4 = reserved by caller: called from leaf routine
1113	 */
11141:	ldsw	[%o0], %o2			! load instruction to %o2
1115	brgez,pt %o2, 2f
1116	  srl	%o2, 30, %o5
1117	btst	1, %o5				! test bit 30; skip if not set
1118	bz,pt	%icc, 2f
1119	  sllx	%o2, 39, %o5			! bit 24 -> bit 63
1120	srlx	%o5, 58, %o5			! isolate op3 part of opcode
1121	xor	%o5, 0x13, %o5			! 01 0011 binary == ldda
1122	brnz,pt	%o5, 2f				! skip if not a match
1123	  or	%o2, %o1, %o2			! or in imm_asi
1124	st	%o2, [%o0]			! write patched instruction
11252:	dec	%o3
1126	brnz,a,pt %o3, 1b			! loop until we're done
1127	  add	%o0, I_SIZE, %o0
1128	retl
1129	flush	%o0
1130	SET_SIZE(sfmmu_fixup_mmu_asi)
1131
1132	/*
1133	 * Patch immediate ASI used to access the TSB in the
1134	 * trap table.
1135	 * inputs: %o0 = value of ktsb_phys
1136	 */
1137	ENTRY_NP(sfmmu_patch_mmu_asi)
1138	mov	%o7, %o4			! save return pc in %o4
1139	mov	ASI_QUAD_LDD_PHYS, %o3		! set QUAD_LDD_PHYS by default
1140
1141#ifdef sun4v
1142
1143	/*
1144	 * Check ktsb_phys. It must be non-zero for sun4v, panic if not.
1145	 */
1146
1147	brnz,pt %o0, do_patch
1148	nop
1149
1150	sethi	%hi(sfmmu_panic11), %o0
1151	call	panic
1152	  or	%o0, %lo(sfmmu_panic11), %o0
1153do_patch:
1154
1155#else /* sun4v */
1156	/*
1157	 * Some non-sun4v platforms deploy virtual ktsb (ktsb_phys==0).
1158	 * Note that ASI_NQUAD_LD is not defined/used for sun4v
1159	 */
1160	movrz	%o0, ASI_NQUAD_LD, %o3
1161
1162#endif /* sun4v */
1163
1164	sll	%o3, 5, %o1			! imm_asi offset
1165	mov	6, %o3				! number of instructions
1166	sethi	%hi(dktsb), %o0			! to search
1167	call	sfmmu_fixup_mmu_asi		! patch kdtlb miss
1168	  or	%o0, %lo(dktsb), %o0
1169	mov	6, %o3				! number of instructions
1170	sethi	%hi(dktsb4m), %o0		! to search
1171	call	sfmmu_fixup_mmu_asi		! patch kdtlb4m miss
1172	  or	%o0, %lo(dktsb4m), %o0
1173	mov	6, %o3				! number of instructions
1174	sethi	%hi(iktsb), %o0			! to search
1175	call	sfmmu_fixup_mmu_asi		! patch kitlb miss
1176	  or	%o0, %lo(iktsb), %o0
1177	mov	6, %o3				! number of instructions
1178	sethi	%hi(iktsb4m), %o0		! to search
1179	call	sfmmu_fixup_mmu_asi		! patch kitlb4m miss
1180	  or	%o0, %lo(iktsb4m), %o0
1181	mov	%o4, %o7			! retore return pc -- leaf
1182	retl
1183	nop
1184	SET_SIZE(sfmmu_patch_mmu_asi)
1185
1186
1187	ENTRY_NP(sfmmu_patch_ktsb)
1188	/*
1189	 * We need to fix iktsb, dktsb, et. al.
1190	 */
1191	save	%sp, -SA(MINFRAME), %sp
1192	set	ktsb_phys, %o1
1193	ld	[%o1], %o4
1194	set	ktsb_base, %o5
1195	set	ktsb4m_base, %l1
1196	brz,pt	%o4, 1f
1197	  nop
1198	set	ktsb_pbase, %o5
1199	set	ktsb4m_pbase, %l1
12001:
1201	sethi	%hi(ktsb_szcode), %o1
1202	ld	[%o1 + %lo(ktsb_szcode)], %o1	/* %o1 = ktsb size code */
1203
1204	sethi	%hi(iktsb), %o0
1205	call	sfmmu_fix_ktlb_traptable
1206	  or	%o0, %lo(iktsb), %o0
1207
1208	sethi	%hi(dktsb), %o0
1209	call	sfmmu_fix_ktlb_traptable
1210	  or	%o0, %lo(dktsb), %o0
1211
1212	sethi	%hi(ktsb4m_szcode), %o1
1213	ld	[%o1 + %lo(ktsb4m_szcode)], %o1	/* %o1 = ktsb4m size code */
1214
1215	sethi	%hi(iktsb4m), %o0
1216	call	sfmmu_fix_ktlb_traptable
1217	  or	%o0, %lo(iktsb4m), %o0
1218
1219	sethi	%hi(dktsb4m), %o0
1220	call	sfmmu_fix_ktlb_traptable
1221	  or	%o0, %lo(dktsb4m), %o0
1222
1223#ifndef sun4v
1224	mov	ASI_N, %o2
1225	movrnz	%o4, ASI_MEM, %o2	! setup kernel 32bit ASI to patch
1226	mov	%o2, %o4		! sfmmu_fixup_or needs this in %o4
1227	sethi	%hi(tsb_kernel_patch_asi), %o0
1228	call	sfmmu_fixup_or
1229	  or	%o0, %lo(tsb_kernel_patch_asi), %o0
1230#endif /* !sun4v */
1231
1232	ldx 	[%o5], %o4		! load ktsb base addr (VA or PA)
1233
1234	sethi	%hi(dktsbbase), %o0
1235	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1236	  or	%o0, %lo(dktsbbase), %o0
1237
1238	sethi	%hi(iktsbbase), %o0
1239	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1240	  or	%o0, %lo(iktsbbase), %o0
1241
1242	sethi	%hi(sfmmu_kprot_patch_ktsb_base), %o0
1243	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1244	  or	%o0, %lo(sfmmu_kprot_patch_ktsb_base), %o0
1245
1246#ifdef sun4v
1247	sethi	%hi(sfmmu_dslow_patch_ktsb_base), %o0
1248	call	sfmmu_fixup_setx	! patch value of ktsb base addr
1249	  or	%o0, %lo(sfmmu_dslow_patch_ktsb_base), %o0
1250#endif /* sun4v */
1251
1252	ldx 	[%l1], %o4		! load ktsb4m base addr (VA or PA)
1253
1254	sethi	%hi(dktsb4mbase), %o0
1255	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1256	  or	%o0, %lo(dktsb4mbase), %o0
1257
1258	sethi	%hi(iktsb4mbase), %o0
1259	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1260	  or	%o0, %lo(iktsb4mbase), %o0
1261
1262	sethi	%hi(sfmmu_kprot_patch_ktsb4m_base), %o0
1263	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1264	  or	%o0, %lo(sfmmu_kprot_patch_ktsb4m_base), %o0
1265
1266#ifdef sun4v
1267	sethi	%hi(sfmmu_dslow_patch_ktsb4m_base), %o0
1268	call	sfmmu_fixup_setx	! patch value of ktsb4m base addr
1269	  or	%o0, %lo(sfmmu_dslow_patch_ktsb4m_base), %o0
1270#endif /* sun4v */
1271
1272	set	ktsb_szcode, %o4
1273	ld	[%o4], %o4
1274	sethi	%hi(sfmmu_kprot_patch_ktsb_szcode), %o0
1275	call	sfmmu_fixup_or		! patch value of ktsb_szcode
1276	  or	%o0, %lo(sfmmu_kprot_patch_ktsb_szcode), %o0
1277
1278#ifdef sun4v
1279	sethi	%hi(sfmmu_dslow_patch_ktsb_szcode), %o0
1280	call	sfmmu_fixup_or		! patch value of ktsb_szcode
1281	  or	%o0, %lo(sfmmu_dslow_patch_ktsb_szcode), %o0
1282#endif /* sun4v */
1283
1284	set	ktsb4m_szcode, %o4
1285	ld	[%o4], %o4
1286	sethi	%hi(sfmmu_kprot_patch_ktsb4m_szcode), %o0
1287	call	sfmmu_fixup_or		! patch value of ktsb4m_szcode
1288	  or	%o0, %lo(sfmmu_kprot_patch_ktsb4m_szcode), %o0
1289
1290#ifdef sun4v
1291	sethi	%hi(sfmmu_dslow_patch_ktsb4m_szcode), %o0
1292	call	sfmmu_fixup_or		! patch value of ktsb4m_szcode
1293	  or	%o0, %lo(sfmmu_dslow_patch_ktsb4m_szcode), %o0
1294#endif /* sun4v */
1295
1296	ret
1297	restore
1298	SET_SIZE(sfmmu_patch_ktsb)
1299
1300	ENTRY_NP(sfmmu_kpm_patch_tlbm)
1301	/*
1302	 * Fixup trap handlers in common segkpm case.  This is reserved
1303	 * for future use should kpm TSB be changed to be other than the
1304	 * kernel TSB.
1305	 */
1306	retl
1307	nop
1308	SET_SIZE(sfmmu_kpm_patch_tlbm)
1309
1310	ENTRY_NP(sfmmu_kpm_patch_tsbm)
1311	/*
1312	 * nop the branch to sfmmu_kpm_dtsb_miss_small
1313	 * in the case where we are using large pages for
1314	 * seg_kpm (and hence must probe the second TSB for
1315	 * seg_kpm VAs)
1316	 */
1317	set	dktsb4m_kpmcheck_small, %o0
1318	MAKE_NOP_INSTR(%o1)
1319	st	%o1, [%o0]
1320	flush	%o0
1321	retl
1322	nop
1323	SET_SIZE(sfmmu_kpm_patch_tsbm)
1324
1325	ENTRY_NP(sfmmu_patch_utsb)
1326#ifdef UTSB_PHYS
1327	retl
1328	nop
1329#else /* UTSB_PHYS */
1330	/*
1331	 * We need to hot patch utsb_vabase and utsb4m_vabase
1332	 */
1333	save	%sp, -SA(MINFRAME), %sp
1334
1335	/* patch value of utsb_vabase */
1336	set	utsb_vabase, %o1
1337	ldx	[%o1], %o4
1338	sethi	%hi(sfmmu_uprot_get_1st_tsbe_ptr), %o0
1339	call	sfmmu_fixup_setx
1340	  or	%o0, %lo(sfmmu_uprot_get_1st_tsbe_ptr), %o0
1341	sethi	%hi(sfmmu_uitlb_get_1st_tsbe_ptr), %o0
1342	call	sfmmu_fixup_setx
1343	  or	%o0, %lo(sfmmu_uitlb_get_1st_tsbe_ptr), %o0
1344	sethi	%hi(sfmmu_udtlb_get_1st_tsbe_ptr), %o0
1345	call	sfmmu_fixup_setx
1346	  or	%o0, %lo(sfmmu_udtlb_get_1st_tsbe_ptr), %o0
1347
1348	/* patch value of utsb4m_vabase */
1349	set	utsb4m_vabase, %o1
1350	ldx	[%o1], %o4
1351	sethi	%hi(sfmmu_uprot_get_2nd_tsb_base), %o0
1352	call	sfmmu_fixup_setx
1353	  or	%o0, %lo(sfmmu_uprot_get_2nd_tsb_base), %o0
1354	sethi	%hi(sfmmu_uitlb_get_2nd_tsb_base), %o0
1355	call	sfmmu_fixup_setx
1356	  or	%o0, %lo(sfmmu_uitlb_get_2nd_tsb_base), %o0
1357	sethi	%hi(sfmmu_udtlb_get_2nd_tsb_base), %o0
1358	call	sfmmu_fixup_setx
1359	  or	%o0, %lo(sfmmu_udtlb_get_2nd_tsb_base), %o0
1360
1361	/*
1362	 * Patch TSB base register masks and shifts if needed.
1363	 * By default the TSB base register contents are set up for 4M slab.
1364	 * If we're using a smaller slab size and reserved VA range we need
1365	 * to patch up those values here.
1366	 */
1367	set	tsb_slab_shift, %o1
1368	set	MMU_PAGESHIFT4M, %o4
1369	lduw	[%o1], %o3
1370	subcc	%o4, %o3, %o4
1371	bz,pt	%icc, 1f
1372	  /* delay slot safe */
1373
1374	/* patch reserved VA range size if needed. */
1375	sethi	%hi(sfmmu_tsb_1st_resv_offset), %o0
1376	call	sfmmu_fixup_shiftx
1377	  or	%o0, %lo(sfmmu_tsb_1st_resv_offset), %o0
1378	call	sfmmu_fixup_shiftx
1379	  add	%o0, I_SIZE, %o0
1380	sethi	%hi(sfmmu_tsb_2nd_resv_offset), %o0
1381	call	sfmmu_fixup_shiftx
1382	  or	%o0, %lo(sfmmu_tsb_2nd_resv_offset), %o0
1383	call	sfmmu_fixup_shiftx
1384	  add	%o0, I_SIZE, %o0
13851:
1386	/* patch TSBREG_VAMASK used to set up TSB base register */
1387	set	tsb_slab_mask, %o1
1388	ldx	[%o1], %o4
1389	sethi	%hi(sfmmu_tsb_1st_tsbreg_vamask), %o0
1390	call	sfmmu_fixup_or
1391	  or	%o0, %lo(sfmmu_tsb_1st_tsbreg_vamask), %o0
1392	sethi	%hi(sfmmu_tsb_2nd_tsbreg_vamask), %o0
1393	call	sfmmu_fixup_or
1394	  or	%o0, %lo(sfmmu_tsb_2nd_tsbreg_vamask), %o0
1395
1396	ret
1397	restore
1398#endif /* UTSB_PHYS */
1399	SET_SIZE(sfmmu_patch_utsb)
1400
1401	ENTRY_NP(sfmmu_patch_shctx)
1402#ifdef sun4u
1403	retl
1404	  nop
1405#else /* sun4u */
1406	set	sfmmu_shctx_cpu_mondo_patch, %o0
1407	MAKE_JMP_INSTR(5, %o1, %o2)	! jmp       %g5
1408	st	%o1, [%o0]
1409	flush	%o0
1410	MAKE_NOP_INSTR(%o1)
1411	add	%o0, I_SIZE, %o0	! next instr
1412	st	%o1, [%o0]
1413	flush	%o0
1414
1415	set	sfmmu_shctx_user_rtt_patch, %o0
1416	st      %o1, [%o0]		! nop 1st instruction
1417	flush	%o0
1418	add     %o0, I_SIZE, %o0
1419	st      %o1, [%o0]		! nop 2nd instruction
1420	flush	%o0
1421	add     %o0, I_SIZE, %o0
1422	st      %o1, [%o0]		! nop 3rd instruction
1423	flush	%o0
1424	add     %o0, I_SIZE, %o0
1425	st      %o1, [%o0]		! nop 4th instruction
1426	flush	%o0
1427	add     %o0, I_SIZE, %o0
1428	st      %o1, [%o0]		! nop 5th instruction
1429	flush	%o0
1430	add     %o0, I_SIZE, %o0
1431	st      %o1, [%o0]		! nop 6th instruction
1432	retl
1433	flush	%o0
1434#endif /* sun4u */
1435	SET_SIZE(sfmmu_patch_shctx)
1436
1437	/*
1438	 * Routine that loads an entry into a tsb using virtual addresses.
1439	 * Locking is required since all cpus can use the same TSB.
1440	 * Note that it is no longer required to have a valid context
1441	 * when calling this function.
1442	 */
1443	ENTRY_NP(sfmmu_load_tsbe)
1444	/*
1445	 * %o0 = pointer to tsbe to load
1446	 * %o1 = tsb tag
1447	 * %o2 = virtual pointer to TTE
1448	 * %o3 = 1 if physical address in %o0 else 0
1449	 */
1450	rdpr	%pstate, %o5
1451#ifdef DEBUG
1452	PANIC_IF_INTR_DISABLED_PSTR(%o5, sfmmu_di_l2, %g1)
1453#endif /* DEBUG */
1454
1455	wrpr	%o5, PSTATE_IE, %pstate		/* disable interrupts */
1456
1457	SETUP_TSB_ASI(%o3, %g3)
1458	TSB_UPDATE(%o0, %o2, %o1, %g1, %g2, locked_tsb_l8)
1459
1460	wrpr	%g0, %o5, %pstate		/* enable interrupts */
1461
1462	retl
1463	membar	#StoreStore|#StoreLoad
1464	SET_SIZE(sfmmu_load_tsbe)
1465
1466	/*
1467	 * Flush TSB of a given entry if the tag matches.
1468	 */
1469	ENTRY(sfmmu_unload_tsbe)
1470	/*
1471	 * %o0 = pointer to tsbe to be flushed
1472	 * %o1 = tag to match
1473	 * %o2 = 1 if physical address in %o0 else 0
1474	 */
1475	SETUP_TSB_ASI(%o2, %g1)
1476	TSB_INVALIDATE(%o0, %o1, %g1, %o2, %o3, unload_tsbe)
1477	retl
1478	membar	#StoreStore|#StoreLoad
1479	SET_SIZE(sfmmu_unload_tsbe)
1480
1481	/*
1482	 * Routine that loads a TTE into the kpm TSB from C code.
1483	 * Locking is required since kpm TSB is shared among all CPUs.
1484	 */
1485	ENTRY_NP(sfmmu_kpm_load_tsb)
1486	/*
1487	 * %o0 = vaddr
1488	 * %o1 = ttep
1489	 * %o2 = virtpg to TSB index shift (e.g. TTE pagesize shift)
1490	 */
1491	rdpr	%pstate, %o5			! %o5 = saved pstate
1492#ifdef DEBUG
1493	PANIC_IF_INTR_DISABLED_PSTR(%o5, sfmmu_di_l3, %g1)
1494#endif /* DEBUG */
1495	wrpr	%o5, PSTATE_IE, %pstate		! disable interrupts
1496
1497#ifndef sun4v
1498	sethi	%hi(ktsb_phys), %o4
1499	mov	ASI_N, %o3
1500	ld	[%o4 + %lo(ktsb_phys)], %o4
1501	movrnz	%o4, ASI_MEM, %o3
1502	mov	%o3, %asi
1503#endif /* !sun4v */
1504	mov	%o0, %g1			! %g1 = vaddr
1505
1506	/* GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr (clobbers), tmp1, tmp2) */
1507	GET_KPM_TSBE_POINTER(%o2, %g2, %g1, %o3, %o4)
1508	/* %g2 = tsbep, %g1 clobbered */
1509
1510	srlx	%o0, TTARGET_VA_SHIFT, %g1;	! %g1 = tag target
1511	/* TSB_UPDATE(tsbep, tteva, tagtarget, tmp1, tmp2, label) */
1512	TSB_UPDATE(%g2, %o1, %g1, %o3, %o4, locked_tsb_l9)
1513
1514	wrpr	%g0, %o5, %pstate		! enable interrupts
1515	retl
1516	  membar #StoreStore|#StoreLoad
1517	SET_SIZE(sfmmu_kpm_load_tsb)
1518
1519	/*
1520	 * Routine that shoots down a TTE in the kpm TSB or in the
1521	 * kernel TSB depending on virtpg. Locking is required since
1522	 * kpm/kernel TSB is shared among all CPUs.
1523	 */
1524	ENTRY_NP(sfmmu_kpm_unload_tsb)
1525	/*
1526	 * %o0 = vaddr
1527	 * %o1 = virtpg to TSB index shift (e.g. TTE page shift)
1528	 */
1529#ifndef sun4v
1530	sethi	%hi(ktsb_phys), %o4
1531	mov	ASI_N, %o3
1532	ld	[%o4 + %lo(ktsb_phys)], %o4
1533	movrnz	%o4, ASI_MEM, %o3
1534	mov	%o3, %asi
1535#endif /* !sun4v */
1536	mov	%o0, %g1			! %g1 = vaddr
1537
1538	/* GET_KPM_TSBE_POINTER(vpshift, tsbp, vaddr (clobbers), tmp1, tmp2) */
1539	GET_KPM_TSBE_POINTER(%o1, %g2, %g1, %o3, %o4)
1540	/* %g2 = tsbep, %g1 clobbered */
1541
1542	srlx	%o0, TTARGET_VA_SHIFT, %g1;	! %g1 = tag target
1543	/* TSB_INVALIDATE(tsbep, tag, tmp1, tmp2, tmp3, label) */
1544	TSB_INVALIDATE(%g2, %g1, %o3, %o4, %o1, kpm_tsbinval)
1545
1546	retl
1547	  membar	#StoreStore|#StoreLoad
1548	SET_SIZE(sfmmu_kpm_unload_tsb)
1549
1550#endif /* lint */
1551
1552
1553#if defined (lint)
1554
1555/*ARGSUSED*/
1556pfn_t
1557sfmmu_ttetopfn(tte_t *tte, caddr_t vaddr)
1558{ return(0); }
1559
1560#else /* lint */
1561
1562	ENTRY_NP(sfmmu_ttetopfn)
1563	ldx	[%o0], %g1			/* read tte */
1564	TTETOPFN(%g1, %o1, sfmmu_ttetopfn_l1, %g2, %g3, %g4)
1565	/*
1566	 * g1 = pfn
1567	 */
1568	retl
1569	mov	%g1, %o0
1570	SET_SIZE(sfmmu_ttetopfn)
1571
1572#endif /* !lint */
1573
1574/*
1575 * These macros are used to update global sfmmu hme hash statistics
1576 * in perf critical paths. It is only enabled in debug kernels or
1577 * if SFMMU_STAT_GATHER is defined
1578 */
1579#if defined(DEBUG) || defined(SFMMU_STAT_GATHER)
1580#define	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tmp1, tmp2)			\
1581	ldn	[tsbarea + TSBMISS_KHATID], tmp1			;\
1582	mov	HATSTAT_KHASH_SEARCH, tmp2				;\
1583	cmp	tmp1, hatid						;\
1584	movne	%ncc, HATSTAT_UHASH_SEARCH, tmp2			;\
1585	set	sfmmu_global_stat, tmp1					;\
1586	add	tmp1, tmp2, tmp1					;\
1587	ld	[tmp1], tmp2						;\
1588	inc	tmp2							;\
1589	st	tmp2, [tmp1]
1590
1591#define	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)			\
1592	ldn	[tsbarea + TSBMISS_KHATID], tmp1			;\
1593	mov	HATSTAT_KHASH_LINKS, tmp2				;\
1594	cmp	tmp1, hatid						;\
1595	movne	%ncc, HATSTAT_UHASH_LINKS, tmp2				;\
1596	set	sfmmu_global_stat, tmp1					;\
1597	add	tmp1, tmp2, tmp1					;\
1598	ld	[tmp1], tmp2						;\
1599	inc	tmp2							;\
1600	st	tmp2, [tmp1]
1601
1602
1603#else /* DEBUG || SFMMU_STAT_GATHER */
1604
1605#define	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tmp1, tmp2)
1606
1607#define	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)
1608
1609#endif  /* DEBUG || SFMMU_STAT_GATHER */
1610
1611/*
1612 * This macro is used to update global sfmmu kstas in non
1613 * perf critical areas so they are enabled all the time
1614 */
1615#define	HAT_GLOBAL_STAT(statname, tmp1, tmp2)				\
1616	sethi	%hi(sfmmu_global_stat), tmp1				;\
1617	add	tmp1, statname, tmp1					;\
1618	ld	[tmp1 + %lo(sfmmu_global_stat)], tmp2			;\
1619	inc	tmp2							;\
1620	st	tmp2, [tmp1 + %lo(sfmmu_global_stat)]
1621
1622/*
1623 * These macros are used to update per cpu stats in non perf
1624 * critical areas so they are enabled all the time
1625 */
1626#define	HAT_PERCPU_STAT32(tsbarea, stat, tmp1)				\
1627	ld	[tsbarea + stat], tmp1					;\
1628	inc	tmp1							;\
1629	st	tmp1, [tsbarea + stat]
1630
1631/*
1632 * These macros are used to update per cpu stats in non perf
1633 * critical areas so they are enabled all the time
1634 */
1635#define	HAT_PERCPU_STAT16(tsbarea, stat, tmp1)				\
1636	lduh	[tsbarea + stat], tmp1					;\
1637	inc	tmp1							;\
1638	stuh	tmp1, [tsbarea + stat]
1639
1640#if defined(KPM_TLBMISS_STATS_GATHER)
1641	/*
1642	 * Count kpm dtlb misses separately to allow a different
1643	 * evaluation of hme and kpm tlbmisses. kpm tsb hits can
1644	 * be calculated by (kpm_dtlb_misses - kpm_tsb_misses).
1645	 */
1646#define	KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label)		\
1647	brgez	tagacc, label	/* KPM VA? */				;\
1648	nop								;\
1649	CPU_INDEX(tmp1, tsbma)						;\
1650	sethi	%hi(kpmtsbm_area), tsbma				;\
1651	sllx	tmp1, KPMTSBM_SHIFT, tmp1				;\
1652	or	tsbma, %lo(kpmtsbm_area), tsbma				;\
1653	add	tsbma, tmp1, tsbma		/* kpmtsbm area */	;\
1654	/* VA range check */						;\
1655	ldx	[tsbma + KPMTSBM_VBASE], val				;\
1656	cmp	tagacc, val						;\
1657	blu,pn	%xcc, label						;\
1658	  ldx	[tsbma + KPMTSBM_VEND], tmp1				;\
1659	cmp	tagacc, tmp1						;\
1660	bgeu,pn	%xcc, label						;\
1661	  lduw	[tsbma + KPMTSBM_DTLBMISS], val				;\
1662	inc	val							;\
1663	st	val, [tsbma + KPMTSBM_DTLBMISS]				;\
1664label:
1665#else
1666#define	KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label)
1667#endif	/* KPM_TLBMISS_STATS_GATHER */
1668
1669#if defined (lint)
1670/*
1671 * The following routines are jumped to from the mmu trap handlers to do
1672 * the setting up to call systrap.  They are separate routines instead of
1673 * being part of the handlers because the handlers would exceed 32
1674 * instructions and since this is part of the slow path the jump
1675 * cost is irrelevant.
1676 */
1677void
1678sfmmu_pagefault(void)
1679{
1680}
1681
1682void
1683sfmmu_mmu_trap(void)
1684{
1685}
1686
1687void
1688sfmmu_window_trap(void)
1689{
1690}
1691
1692void
1693sfmmu_kpm_exception(void)
1694{
1695}
1696
1697#else /* lint */
1698
1699#ifdef	PTL1_PANIC_DEBUG
1700	.seg	".data"
1701	.global	test_ptl1_panic
1702test_ptl1_panic:
1703	.word	0
1704	.align	8
1705
1706	.seg	".text"
1707	.align	4
1708#endif	/* PTL1_PANIC_DEBUG */
1709
1710
1711	ENTRY_NP(sfmmu_pagefault)
1712	SET_GL_REG(1)
1713	USE_ALTERNATE_GLOBALS(%g5)
1714	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g6, %g4)
1715	rdpr	%tt, %g6
1716	cmp	%g6, FAST_IMMU_MISS_TT
1717	be,a,pn	%icc, 1f
1718	  mov	T_INSTR_MMU_MISS, %g3
1719	cmp	%g6, T_INSTR_MMU_MISS
1720	be,a,pn	%icc, 1f
1721	  mov	T_INSTR_MMU_MISS, %g3
1722	mov	%g5, %g2
1723	mov	T_DATA_PROT, %g3		/* arg2 = traptype */
1724	cmp	%g6, FAST_DMMU_MISS_TT
1725	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1726	cmp	%g6, T_DATA_MMU_MISS
1727	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1728
1729#ifdef  PTL1_PANIC_DEBUG
1730	/* check if we want to test the tl1 panic */
1731	sethi	%hi(test_ptl1_panic), %g4
1732	ld	[%g4 + %lo(test_ptl1_panic)], %g1
1733	st	%g0, [%g4 + %lo(test_ptl1_panic)]
1734	cmp	%g1, %g0
1735	bne,a,pn %icc, ptl1_panic
1736	  or	%g0, PTL1_BAD_DEBUG, %g1
1737#endif	/* PTL1_PANIC_DEBUG */
17381:
1739	HAT_GLOBAL_STAT(HATSTAT_PAGEFAULT, %g6, %g4)
1740	/*
1741	 * g2 = tag access reg
1742	 * g3.l = type
1743	 * g3.h = 0
1744	 */
1745	sethi	%hi(trap), %g1
1746	or	%g1, %lo(trap), %g1
17472:
1748	ba,pt	%xcc, sys_trap
1749	  mov	-1, %g4
1750	SET_SIZE(sfmmu_pagefault)
1751
1752	ENTRY_NP(sfmmu_mmu_trap)
1753	SET_GL_REG(1)
1754	USE_ALTERNATE_GLOBALS(%g5)
1755	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g4, %g6)
1756	rdpr	%tt, %g6
1757	cmp	%g6, FAST_IMMU_MISS_TT
1758	be,a,pn	%icc, 1f
1759	  mov	T_INSTR_MMU_MISS, %g3
1760	cmp	%g6, T_INSTR_MMU_MISS
1761	be,a,pn	%icc, 1f
1762	  mov	T_INSTR_MMU_MISS, %g3
1763	mov	%g5, %g2
1764	mov	T_DATA_PROT, %g3		/* arg2 = traptype */
1765	cmp	%g6, FAST_DMMU_MISS_TT
1766	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1767	cmp	%g6, T_DATA_MMU_MISS
1768	move	%icc, T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
17691:
1770	/*
1771	 * g2 = tag access reg
1772	 * g3 = type
1773	 */
1774	sethi	%hi(sfmmu_tsbmiss_exception), %g1
1775	or	%g1, %lo(sfmmu_tsbmiss_exception), %g1
1776	ba,pt	%xcc, sys_trap
1777	  mov	-1, %g4
1778	/*NOTREACHED*/
1779	SET_SIZE(sfmmu_mmu_trap)
1780
1781	ENTRY_NP(sfmmu_suspend_tl)
1782	SET_GL_REG(1)
1783	USE_ALTERNATE_GLOBALS(%g5)
1784	GET_MMU_BOTH_TAGACC(%g5 /*dtag*/, %g2 /*itag*/, %g4, %g3)
1785	rdpr	%tt, %g6
1786	cmp	%g6, FAST_IMMU_MISS_TT
1787	be,a,pn	%icc, 1f
1788	  mov	T_INSTR_MMU_MISS, %g3
1789	mov	%g5, %g2
1790	cmp	%g6, FAST_DMMU_MISS_TT
1791	move	%icc, T_DATA_MMU_MISS, %g3
1792	movne	%icc, T_DATA_PROT, %g3
17931:
1794	sethi	%hi(sfmmu_tsbmiss_suspended), %g1
1795	or	%g1, %lo(sfmmu_tsbmiss_suspended), %g1
1796	/* g1 = TL0 handler, g2 = tagacc, g3 = trap type */
1797	ba,pt	%xcc, sys_trap
1798	  mov	PIL_15, %g4
1799	/*NOTREACHED*/
1800	SET_SIZE(sfmmu_suspend_tl)
1801
1802	/*
1803	 * No %g registers in use at this point.
1804	 */
1805	ENTRY_NP(sfmmu_window_trap)
1806	rdpr	%tpc, %g1
1807#ifdef sun4v
1808#ifdef DEBUG
1809	/* We assume previous %gl was 1 */
1810	rdpr	%tstate, %g4
1811	srlx	%g4, TSTATE_GL_SHIFT, %g4
1812	and	%g4, TSTATE_GL_MASK, %g4
1813	cmp	%g4, 1
1814	bne,a,pn %icc, ptl1_panic
1815	  mov	PTL1_BAD_WTRAP, %g1
1816#endif /* DEBUG */
1817	/* user miss at tl>1. better be the window handler or user_rtt */
1818	/* in user_rtt? */
1819	set	rtt_fill_start, %g4
1820	cmp	%g1, %g4
1821	blu,pn %xcc, 6f
1822	 .empty
1823	set	rtt_fill_end, %g4
1824	cmp	%g1, %g4
1825	bgeu,pn %xcc, 6f
1826	 nop
1827	set	fault_rtt_fn1, %g1
1828	wrpr	%g0, %g1, %tnpc
1829	ba,a	7f
18306:
1831	! must save this trap level before descending trap stack
1832	! no need to save %tnpc, either overwritten or discarded
1833	! already got it: rdpr	%tpc, %g1
1834	rdpr	%tstate, %g6
1835	rdpr	%tt, %g7
1836	! trap level saved, go get underlying trap type
1837	rdpr	%tl, %g5
1838	sub	%g5, 1, %g3
1839	wrpr	%g3, %tl
1840	rdpr	%tt, %g2
1841	wrpr	%g5, %tl
1842	! restore saved trap level
1843	wrpr	%g1, %tpc
1844	wrpr	%g6, %tstate
1845	wrpr	%g7, %tt
1846#else /* sun4v */
1847	/* user miss at tl>1. better be the window handler */
1848	rdpr	%tl, %g5
1849	sub	%g5, 1, %g3
1850	wrpr	%g3, %tl
1851	rdpr	%tt, %g2
1852	wrpr	%g5, %tl
1853#endif /* sun4v */
1854	and	%g2, WTRAP_TTMASK, %g4
1855	cmp	%g4, WTRAP_TYPE
1856	bne,pn	%xcc, 1f
1857	 nop
1858	/* tpc should be in the trap table */
1859	set	trap_table, %g4
1860	cmp	%g1, %g4
1861	blt,pn %xcc, 1f
1862	 .empty
1863	set	etrap_table, %g4
1864	cmp	%g1, %g4
1865	bge,pn %xcc, 1f
1866	 .empty
1867	andn	%g1, WTRAP_ALIGN, %g1	/* 128 byte aligned */
1868	add	%g1, WTRAP_FAULTOFF, %g1
1869	wrpr	%g0, %g1, %tnpc
18707:
1871	/*
1872	 * some wbuf handlers will call systrap to resolve the fault
1873	 * we pass the trap type so they figure out the correct parameters.
1874	 * g5 = trap type, g6 = tag access reg
1875	 */
1876
1877	/*
1878	 * only use g5, g6, g7 registers after we have switched to alternate
1879	 * globals.
1880	 */
1881	SET_GL_REG(1)
1882	USE_ALTERNATE_GLOBALS(%g5)
1883	GET_MMU_D_TAGACC(%g6 /*dtag*/, %g5 /*scratch*/)
1884	rdpr	%tt, %g7
1885	cmp	%g7, FAST_IMMU_MISS_TT
1886	be,a,pn	%icc, ptl1_panic
1887	  mov	PTL1_BAD_WTRAP, %g1
1888	cmp	%g7, T_INSTR_MMU_MISS
1889	be,a,pn	%icc, ptl1_panic
1890	  mov	PTL1_BAD_WTRAP, %g1
1891	mov	T_DATA_PROT, %g5
1892	cmp	%g7, FAST_DMMU_MISS_TT
1893	move	%icc, T_DATA_MMU_MISS, %g5
1894	cmp	%g7, T_DATA_MMU_MISS
1895	move	%icc, T_DATA_MMU_MISS, %g5
1896	! XXXQ AGS re-check out this one
1897	done
18981:
1899	CPU_PADDR(%g1, %g4)
1900	add	%g1, CPU_TL1_HDLR, %g1
1901	lda	[%g1]ASI_MEM, %g4
1902	brnz,a,pt %g4, sfmmu_mmu_trap
1903	  sta	%g0, [%g1]ASI_MEM
1904	ba,pt	%icc, ptl1_panic
1905	  mov	PTL1_BAD_TRAP, %g1
1906	SET_SIZE(sfmmu_window_trap)
1907
1908	ENTRY_NP(sfmmu_kpm_exception)
1909	/*
1910	 * We have accessed an unmapped segkpm address or a legal segkpm
1911	 * address which is involved in a VAC alias conflict prevention.
1912	 * Before we go to trap(), check to see if CPU_DTRACE_NOFAULT is
1913	 * set. If it is, we will instead note that a fault has occurred
1914	 * by setting CPU_DTRACE_BADADDR and issue a "done" (instead of
1915	 * a "retry"). This will step over the faulting instruction.
1916	 * Note that this means that a legal segkpm address involved in
1917	 * a VAC alias conflict prevention (a rare case to begin with)
1918	 * cannot be used in DTrace.
1919	 */
1920	CPU_INDEX(%g1, %g2)
1921	set	cpu_core, %g2
1922	sllx	%g1, CPU_CORE_SHIFT, %g1
1923	add	%g1, %g2, %g1
1924	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
1925	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
1926	bz	0f
1927	or	%g2, CPU_DTRACE_BADADDR, %g2
1928	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
1929	GET_MMU_D_ADDR(%g3, /*scratch*/ %g4)
1930	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
1931	done
19320:
1933	TSTAT_CHECK_TL1(1f, %g1, %g2)
19341:
1935	SET_GL_REG(1)
1936	USE_ALTERNATE_GLOBALS(%g5)
1937	GET_MMU_D_TAGACC(%g2 /* tagacc */, %g4 /*scratch*/)
1938	mov	T_DATA_MMU_MISS, %g3	/* arg2 = traptype */
1939	/*
1940	 * g2=tagacc g3.l=type g3.h=0
1941	 */
1942	sethi	%hi(trap), %g1
1943	or	%g1, %lo(trap), %g1
1944	ba,pt	%xcc, sys_trap
1945	mov	-1, %g4
1946	SET_SIZE(sfmmu_kpm_exception)
1947
1948#endif /* lint */
1949
1950#if defined (lint)
1951
1952void
1953sfmmu_tsb_miss(void)
1954{
1955}
1956
1957void
1958sfmmu_kpm_dtsb_miss(void)
1959{
1960}
1961
1962void
1963sfmmu_kpm_dtsb_miss_small(void)
1964{
1965}
1966
1967#else /* lint */
1968
1969#if (IMAP_SEG != 0)
1970#error - ism_map->ism_seg offset is not zero
1971#endif
1972
1973/*
1974 * Copies ism mapping for this ctx in param "ism" if this is a ISM
1975 * tlb miss and branches to label "ismhit". If this is not an ISM
1976 * process or an ISM tlb miss it falls thru.
1977 *
1978 * Checks to see if the vaddr passed in via tagacc is in an ISM segment for
1979 * this process.
1980 * If so, it will branch to label "ismhit".  If not, it will fall through.
1981 *
1982 * Also hat_unshare() will set the context for this process to INVALID_CONTEXT
1983 * so that any other threads of this process will not try and walk the ism
1984 * maps while they are being changed.
1985 *
1986 * NOTE: We will never have any holes in our ISM maps. sfmmu_share/unshare
1987 *       will make sure of that. This means we can terminate our search on
1988 *       the first zero mapping we find.
1989 *
1990 * Parameters:
1991 * tagacc	= (pseudo-)tag access register (vaddr + ctx) (in)
1992 * tsbmiss	= address of tsb miss area (in)
1993 * ismseg	= contents of ism_seg for this ism map (out)
1994 * ismhat	= physical address of imap_ismhat for this ism map (out)
1995 * tmp1		= scratch reg (CLOBBERED)
1996 * tmp2		= scratch reg (CLOBBERED)
1997 * tmp3		= scratch reg (CLOBBERED)
1998 * label:    temporary labels
1999 * ismhit:   label where to jump to if an ism dtlb miss
2000 * exitlabel:label where to jump if hat is busy due to hat_unshare.
2001 */
2002#define ISM_CHECK(tagacc, tsbmiss, ismseg, ismhat, tmp1, tmp2, tmp3 \
2003	label, ismhit)							\
2004	ldx	[tsbmiss + TSBMISS_ISMBLKPA], tmp1 /* tmp1 = &ismblk */	;\
2005	brlz,pt  tmp1, label/**/3		/* exit if -1 */	;\
2006	  add	tmp1, IBLK_MAPS, ismhat	/* ismhat = &ismblk.map[0] */	;\
2007label/**/1:								;\
2008	ldxa	[ismhat]ASI_MEM, ismseg	/* ismblk.map[0].ism_seg */	;\
2009	mov	tmp1, tmp3	/* update current ismblkpa head */	;\
2010label/**/2:								;\
2011	brz,pt  ismseg, label/**/3		/* no mapping */	;\
2012	  add	ismhat, IMAP_VB_SHIFT, tmp1 /* tmp1 = vb_shift addr */	;\
2013	lduba	[tmp1]ASI_MEM, tmp1 		/* tmp1 = vb shift*/	;\
2014	srlx	ismseg, tmp1, tmp2		/* tmp2 = vbase */	;\
2015	srlx	tagacc, tmp1, tmp1		/* tmp1 =  va seg*/	;\
2016	sub	tmp1, tmp2, tmp2		/* tmp2 = va - vbase */	;\
2017	add	ismhat, IMAP_SZ_MASK, tmp1 /* tmp1 = sz_mask addr */	;\
2018	lda	[tmp1]ASI_MEM, tmp1		/* tmp1 = sz_mask */	;\
2019	and	ismseg, tmp1, tmp1		/* tmp1 = size */	;\
2020	cmp	tmp2, tmp1		 	/* check va <= offset*/	;\
2021	blu,a,pt  %xcc, ismhit			/* ism hit */		;\
2022	  add	ismhat, IMAP_ISMHAT, ismhat 	/* ismhat = &ism_sfmmu*/ ;\
2023									;\
2024	add	ismhat, ISM_MAP_SZ, ismhat /* ismhat += sizeof(map) */ 	;\
2025	add	tmp3, (IBLK_MAPS + ISM_MAP_SLOTS * ISM_MAP_SZ), tmp1	;\
2026	cmp	ismhat, tmp1						;\
2027	bl,pt	%xcc, label/**/2		/* keep looking  */	;\
2028	  ldxa	[ismhat]ASI_MEM, ismseg	/* ismseg = map[ismhat] */	;\
2029									;\
2030	add	tmp3, IBLK_NEXTPA, tmp1					;\
2031	ldxa	[tmp1]ASI_MEM, tmp1		/* check blk->nextpa */	;\
2032	brgez,pt tmp1, label/**/1		/* continue if not -1*/	;\
2033	  add	tmp1, IBLK_MAPS, ismhat	/* ismhat = &ismblk.map[0]*/	;\
2034label/**/3:
2035
2036/*
2037 * Returns the hme hash bucket (hmebp) given the vaddr, and the hatid
2038 * It also returns the virtual pg for vaddr (ie. vaddr << hmeshift)
2039 * Parameters:
2040 * tagacc = reg containing virtual address
2041 * hatid = reg containing sfmmu pointer
2042 * hmeshift = constant/register to shift vaddr to obtain vapg
2043 * hmebp = register where bucket pointer will be stored
2044 * vapg = register where virtual page will be stored
2045 * tmp1, tmp2 = tmp registers
2046 */
2047
2048
2049#define	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, hmebp,	\
2050	vapg, label, tmp1, tmp2)					\
2051	sllx	tagacc, TAGACC_CTX_LSHIFT, tmp1				;\
2052	brnz,a,pt tmp1, label/**/1					;\
2053	  ld    [tsbarea + TSBMISS_UHASHSZ], hmebp			;\
2054	ld	[tsbarea + TSBMISS_KHASHSZ], hmebp			;\
2055	ba,pt	%xcc, label/**/2					;\
2056	  ldx	[tsbarea + TSBMISS_KHASHSTART], tmp1			;\
2057label/**/1:								;\
2058	ldx	[tsbarea + TSBMISS_UHASHSTART], tmp1			;\
2059label/**/2:								;\
2060	srlx	tagacc, hmeshift, vapg					;\
2061	xor	vapg, hatid, tmp2	/* hatid ^ (vaddr >> shift) */	;\
2062	and	tmp2, hmebp, hmebp	/* index into hme_hash */	;\
2063	mulx	hmebp, HMEBUCK_SIZE, hmebp				;\
2064	add	hmebp, tmp1, hmebp
2065
2066/*
2067 * hashtag includes bspage + hashno (64 bits).
2068 */
2069
2070#define	MAKE_HASHTAG(vapg, hatid, hmeshift, hashno, hblktag)		\
2071	sllx	vapg, hmeshift, vapg					;\
2072	mov	hashno, hblktag						;\
2073	sllx	hblktag, HTAG_REHASH_SHIFT, hblktag			;\
2074	or	vapg, hblktag, hblktag
2075
2076/*
2077 * Function to traverse hmeblk hash link list and find corresponding match.
2078 * The search is done using physical pointers. It returns the physical address
2079 * pointer to the hmeblk that matches with the tag provided.
2080 * Parameters:
2081 * hmebp	= register that points to hme hash bucket, also used as
2082 *		  tmp reg (clobbered)
2083 * hmeblktag	= register with hmeblk tag match
2084 * hatid	= register with hatid
2085 * hmeblkpa	= register where physical ptr will be stored
2086 * tmp1		= tmp reg
2087 * label: temporary label
2088 */
2089
2090#define	HMEHASH_SEARCH(hmebp, hmeblktag, hatid, hmeblkpa, tsbarea, 	\
2091	tmp1, label)							\
2092	add     hmebp, HMEBUCK_NEXTPA, hmeblkpa				;\
2093	ldxa    [hmeblkpa]ASI_MEM, hmeblkpa				;\
2094	HAT_HSEARCH_DBSTAT(hatid, tsbarea, hmebp, tmp1)			;\
2095label/**/1:								;\
2096	cmp	hmeblkpa, HMEBLK_ENDPA					;\
2097	be,pn   %xcc, label/**/2					;\
2098	HAT_HLINK_DBSTAT(hatid, tsbarea, hmebp, tmp1)			;\
2099	add	hmeblkpa, HMEBLK_TAG, hmebp				;\
2100	ldxa	[hmebp]ASI_MEM, tmp1	 /* read 1st part of tag */	;\
2101	add	hmebp, CLONGSIZE, hmebp					;\
2102	ldxa	[hmebp]ASI_MEM, hmebp 	/* read 2nd part of tag */	;\
2103	xor	tmp1, hmeblktag, tmp1					;\
2104	xor	hmebp, hatid, hmebp					;\
2105	or	hmebp, tmp1, hmebp					;\
2106	brz,pn	hmebp, label/**/2	/* branch on hit */		;\
2107	  add	hmeblkpa, HMEBLK_NEXTPA, hmebp				;\
2108	ba,pt	%xcc, label/**/1					;\
2109	  ldxa	[hmebp]ASI_MEM, hmeblkpa	/* hmeblk ptr pa */	;\
2110label/**/2:
2111
2112/*
2113 * Function to traverse hmeblk hash link list and find corresponding match.
2114 * The search is done using physical pointers. It returns the physical address
2115 * pointer to the hmeblk that matches with the tag
2116 * provided.
2117 * Parameters:
2118 * hmeblktag	= register with hmeblk tag match (rid field is 0)
2119 * hatid	= register with hatid (pointer to SRD)
2120 * hmeblkpa	= register where physical ptr will be stored
2121 * tmp1		= tmp reg
2122 * tmp2		= tmp reg
2123 * label: temporary label
2124 */
2125
2126#define	HMEHASH_SEARCH_SHME(hmeblktag, hatid, hmeblkpa, tsbarea,	\
2127	tmp1, tmp2, label)			 			\
2128label/**/1:								;\
2129	cmp	hmeblkpa, HMEBLK_ENDPA					;\
2130	be,pn   %xcc, label/**/4					;\
2131	HAT_HLINK_DBSTAT(hatid, tsbarea, tmp1, tmp2)			;\
2132	add	hmeblkpa, HMEBLK_TAG, tmp2				;\
2133	ldxa	[tmp2]ASI_MEM, tmp1	 /* read 1st part of tag */	;\
2134	add	tmp2, CLONGSIZE, tmp2					;\
2135	ldxa	[tmp2]ASI_MEM, tmp2 	/* read 2nd part of tag */	;\
2136	xor	tmp1, hmeblktag, tmp1					;\
2137	xor	tmp2, hatid, tmp2					;\
2138	brz,pn	tmp2, label/**/3	/* branch on hit */		;\
2139	  add	hmeblkpa, HMEBLK_NEXTPA, tmp2				;\
2140label/**/2:								;\
2141	ba,pt	%xcc, label/**/1					;\
2142	  ldxa	[tmp2]ASI_MEM, hmeblkpa	/* hmeblk ptr pa */		;\
2143label/**/3:								;\
2144	cmp	tmp1, SFMMU_MAX_HME_REGIONS				;\
2145	bgeu,pt	%xcc, label/**/2					;\
2146	  add	hmeblkpa, HMEBLK_NEXTPA, tmp2				;\
2147	and	tmp1, BT_ULMASK, tmp2					;\
2148	srlx	tmp1, BT_ULSHIFT, tmp1					;\
2149	sllx	tmp1, CLONGSHIFT, tmp1					;\
2150	add	tsbarea, tmp1, tmp1					;\
2151	ldx	[tmp1 + TSBMISS_SHMERMAP], tmp1				;\
2152	srlx	tmp1, tmp2, tmp1					;\
2153	btst	0x1, tmp1						;\
2154	bz,pn	%xcc, label/**/2					;\
2155	  add	hmeblkpa, HMEBLK_NEXTPA, tmp2				;\
2156label/**/4:
2157
2158#if ((1 << SFHME_SHIFT) != SFHME_SIZE)
2159#error HMEBLK_TO_HMENT assumes sf_hment is power of 2 in size
2160#endif
2161
2162/*
2163 * HMEBLK_TO_HMENT is a macro that given an hmeblk and a vaddr returns
2164 * he offset for the corresponding hment.
2165 * Parameters:
2166 * In:
2167 *	vaddr = register with virtual address
2168 *	hmeblkpa = physical pointer to hme_blk
2169 * Out:
2170 *	hmentoff = register where hment offset will be stored
2171 *	hmemisc = hblk_misc
2172 * Scratch:
2173 *	tmp1
2174 */
2175#define	HMEBLK_TO_HMENT(vaddr, hmeblkpa, hmentoff, hmemisc, tmp1, label1)\
2176	add	hmeblkpa, HMEBLK_MISC, hmentoff				;\
2177	lda	[hmentoff]ASI_MEM, hmemisc 				;\
2178	andcc	hmemisc, HBLK_SZMASK, %g0				;\
2179	bnz,a,pn  %icc, label1		/* if sz != TTE8K branch */	;\
2180	  or	%g0, HMEBLK_HME1, hmentoff				;\
2181	srl	vaddr, MMU_PAGESHIFT, tmp1				;\
2182	and	tmp1, NHMENTS - 1, tmp1		/* tmp1 = index */	;\
2183	sllx	tmp1, SFHME_SHIFT, tmp1					;\
2184	add	tmp1, HMEBLK_HME1, hmentoff				;\
2185label1:
2186
2187/*
2188 * GET_TTE is a macro that returns a TTE given a tag and hatid.
2189 *
2190 * tagacc	= (pseudo-)tag access register (in)
2191 * hatid	= sfmmu pointer for TSB miss (in)
2192 * tte		= tte for TLB miss if found, otherwise clobbered (out)
2193 * hmeblkpa	= PA of hment if found, otherwise clobbered (out)
2194 * tsbarea	= pointer to the tsbmiss area for this cpu. (in)
2195 * hmemisc	= hblk_misc if TTE is found (out), otherwise clobbered
2196 * hmeshift	= constant/register to shift VA to obtain the virtual pfn
2197 *		  for this page size.
2198 * hashno	= constant/register hash number
2199 * tmp		= temp value - clobbered
2200 * label	= temporary label for branching within macro.
2201 * foundlabel	= label to jump to when tte is found.
2202 * suspendlabel= label to jump to when tte is suspended.
2203 * exitlabel	= label to jump to when tte is not found.
2204 *
2205 */
2206#define GET_TTE(tagacc, hatid, tte, hmeblkpa, tsbarea, hmemisc, hmeshift, \
2207		 hashno, tmp, label, foundlabel, suspendlabel, exitlabel) \
2208									;\
2209	stn	tagacc, [tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)]	;\
2210	stn	hatid, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)]	;\
2211	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, tte,		\
2212		hmeblkpa, label/**/5, hmemisc, tmp)			;\
2213									;\
2214	/*								;\
2215	 * tagacc = tagacc						;\
2216	 * hatid = hatid						;\
2217	 * tsbarea = tsbarea						;\
2218	 * tte   = hmebp (hme bucket pointer)				;\
2219	 * hmeblkpa  = vapg  (virtual page)				;\
2220	 * hmemisc, tmp = scratch					;\
2221	 */								;\
2222	MAKE_HASHTAG(hmeblkpa, hatid, hmeshift, hashno, hmemisc)	;\
2223	or	hmemisc, SFMMU_INVALID_SHMERID, hmemisc			;\
2224									;\
2225	/*								;\
2226	 * tagacc = tagacc						;\
2227	 * hatid = hatid						;\
2228	 * tte   = hmebp						;\
2229	 * hmeblkpa  = CLOBBERED					;\
2230	 * hmemisc  = htag_bspage+hashno+invalid_rid			;\
2231	 * tmp  = scratch						;\
2232	 */								;\
2233	stn	tte, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)]	;\
2234	HMEHASH_SEARCH(tte, hmemisc, hatid, hmeblkpa, 	 		\
2235		tsbarea, tagacc, label/**/1)				;\
2236	/*								;\
2237	 * tagacc = CLOBBERED						;\
2238	 * tte = CLOBBERED						;\
2239	 * hmeblkpa = hmeblkpa						;\
2240	 * tmp = scratch						;\
2241	 */								;\
2242	cmp	hmeblkpa, HMEBLK_ENDPA					;\
2243	bne,pn   %xcc, label/**/4       /* branch if hmeblk found */    ;\
2244	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)], tagacc	;\
2245	ba,pt	%xcc, exitlabel		/* exit if hblk not found */	;\
2246	  nop								;\
2247label/**/4:								;\
2248	/*								;\
2249	 * We have found the hmeblk containing the hment.		;\
2250	 * Now we calculate the corresponding tte.			;\
2251	 *								;\
2252	 * tagacc = tagacc						;\
2253	 * hatid = hatid						;\
2254	 * tte   = clobbered						;\
2255	 * hmeblkpa  = hmeblkpa						;\
2256	 * hmemisc  = hblktag						;\
2257	 * tmp = scratch						;\
2258	 */								;\
2259	HMEBLK_TO_HMENT(tagacc, hmeblkpa, hatid, hmemisc, tte,		\
2260		label/**/2)						;\
2261									;\
2262	/*								;\
2263	 * tagacc = tagacc						;\
2264	 * hatid = hmentoff						;\
2265	 * tte   = clobbered						;\
2266	 * hmeblkpa  = hmeblkpa						;\
2267	 * hmemisc  = hblk_misc						;\
2268	 * tmp = scratch						;\
2269	 */								;\
2270									;\
2271	add	hatid, SFHME_TTE, hatid					;\
2272	add	hmeblkpa, hatid, hmeblkpa				;\
2273	ldxa	[hmeblkpa]ASI_MEM, tte	/* MMU_READTTE through pa */	;\
2274	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2275	set	TTE_SUSPEND, hatid					;\
2276	TTE_SUSPEND_INT_SHIFT(hatid)					;\
2277	btst	tte, hatid						;\
2278	bz,pt	%xcc, foundlabel					;\
2279	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2280									;\
2281	/*								;\
2282	 * Mapping is suspended, so goto suspend label.			;\
2283	 */								;\
2284	ba,pt	%xcc, suspendlabel					;\
2285	  nop
2286
2287/*
2288 * GET_SHME_TTE is similar to GET_TTE() except it searches
2289 * shared hmeblks via HMEHASH_SEARCH_SHME() macro.
2290 * If valid tte is found, hmemisc = shctx flag, i.e., shme is
2291 * either 0 (not part of scd) or 1 (part of scd).
2292 */
2293#define GET_SHME_TTE(tagacc, hatid, tte, hmeblkpa, tsbarea, hmemisc, 	\
2294		hmeshift, hashno, tmp, label, foundlabel,		\
2295		suspendlabel, exitlabel)				\
2296									;\
2297	stn	tagacc, [tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)]	;\
2298	stn	hatid, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)]	;\
2299	HMEHASH_FUNC_ASM(tagacc, hatid, tsbarea, hmeshift, tte,		\
2300		hmeblkpa, label/**/5, hmemisc, tmp)			;\
2301									;\
2302	/*								;\
2303	 * tagacc = tagacc						;\
2304	 * hatid = hatid						;\
2305	 * tsbarea = tsbarea						;\
2306	 * tte   = hmebp (hme bucket pointer)				;\
2307	 * hmeblkpa  = vapg  (virtual page)				;\
2308	 * hmemisc, tmp = scratch					;\
2309	 */								;\
2310	MAKE_HASHTAG(hmeblkpa, hatid, hmeshift, hashno, hmemisc)	;\
2311									;\
2312	/*								;\
2313	 * tagacc = tagacc						;\
2314	 * hatid = hatid						;\
2315	 * tsbarea = tsbarea						;\
2316	 * tte   = hmebp						;\
2317	 * hmemisc  = htag_bspage + hashno + 0 (for rid)		;\
2318	 * hmeblkpa  = CLOBBERED					;\
2319	 * tmp = scratch						;\
2320	 */								;\
2321	stn	tte, [tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)]	;\
2322									;\
2323	add     tte, HMEBUCK_NEXTPA, hmeblkpa				;\
2324	ldxa    [hmeblkpa]ASI_MEM, hmeblkpa				;\
2325	HAT_HSEARCH_DBSTAT(hatid, tsbarea, tagacc, tte)			;\
2326									;\
2327label/**/8:								;\
2328	HMEHASH_SEARCH_SHME(hmemisc, hatid, hmeblkpa,			\
2329		tsbarea, tagacc, tte, label/**/1)			;\
2330	/*								;\
2331	 * tagacc = CLOBBERED						;\
2332	 * tte = CLOBBERED						;\
2333	 * hmeblkpa = hmeblkpa						;\
2334	 * tmp = scratch						;\
2335	 */								;\
2336	cmp	hmeblkpa, HMEBLK_ENDPA					;\
2337	bne,pn   %xcc, label/**/4       /* branch if hmeblk found */    ;\
2338	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSB_TAGACC)], tagacc	;\
2339	ba,pt	%xcc, exitlabel		/* exit if hblk not found */	;\
2340	  nop								;\
2341label/**/4:								;\
2342	/*								;\
2343	 * We have found the hmeblk containing the hment.		;\
2344	 * Now we calculate the corresponding tte.			;\
2345	 *								;\
2346	 * tagacc = tagacc						;\
2347	 * hatid = hatid						;\
2348	 * tte   = clobbered						;\
2349	 * hmeblkpa  = hmeblkpa						;\
2350	 * hmemisc  = hblktag						;\
2351	 * tsbarea = tsbmiss area					;\
2352	 * tmp = scratch						;\
2353	 */								;\
2354	HMEBLK_TO_HMENT(tagacc, hmeblkpa, hatid, hmemisc, tte,		\
2355		label/**/2)						;\
2356									;\
2357	/*								;\
2358	 * tagacc = tagacc						;\
2359	 * hatid = hmentoff						;\
2360	 * tte = clobbered						;\
2361	 * hmeblkpa  = hmeblkpa						;\
2362	 * hmemisc  = hblk_misc						;\
2363	 * tsbarea = tsbmiss area					;\
2364	 * tmp = scratch						;\
2365	 */								;\
2366									;\
2367	add	hatid, SFHME_TTE, hatid					;\
2368	add	hmeblkpa, hatid, hmeblkpa				;\
2369	ldxa	[hmeblkpa]ASI_MEM, tte	/* MMU_READTTE through pa */	;\
2370	brlz,pt tte, label/**/6						;\
2371	  nop								;\
2372	btst	HBLK_SZMASK, hmemisc					;\
2373	bnz,a,pt %icc, label/**/7					;\
2374	  ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2375									;\
2376	/*								;\
2377 	 * We found an invalid 8K tte in shme.				;\
2378	 * it may not belong to shme's region since			;\
2379	 * region size/alignment granularity is 8K but different	;\
2380	 * regions don't share hmeblks. Continue the search.		;\
2381	 */								;\
2382	sub	hmeblkpa, hatid, hmeblkpa				;\
2383	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2384	srlx	tagacc, hmeshift, tte					;\
2385	add	hmeblkpa, HMEBLK_NEXTPA, hmeblkpa			;\
2386	ldxa	[hmeblkpa]ASI_MEM, hmeblkpa				;\
2387	MAKE_HASHTAG(tte, hatid, hmeshift, hashno, hmemisc)		;\
2388	ba,a,pt	%xcc, label/**/8					;\
2389label/**/6:								;\
2390	GET_SCDSHMERMAP(tsbarea, hmeblkpa, hatid, hmemisc)		;\
2391	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HMEBP)], hatid 	;\
2392label/**/7:								;\
2393	set	TTE_SUSPEND, hatid					;\
2394	TTE_SUSPEND_INT_SHIFT(hatid)					;\
2395	btst	tte, hatid						;\
2396	bz,pt	%xcc, foundlabel					;\
2397	ldn	[tsbarea + (TSBMISS_SCRATCH + TSBMISS_HATID)], hatid	;\
2398									;\
2399	/*								;\
2400	 * Mapping is suspended, so goto suspend label.			;\
2401	 */								;\
2402	ba,pt	%xcc, suspendlabel					;\
2403	  nop
2404
2405	/*
2406	 * KERNEL PROTECTION HANDLER
2407	 *
2408	 * g1 = tsb8k pointer register (clobbered)
2409	 * g2 = tag access register (ro)
2410	 * g3 - g7 = scratch registers
2411	 *
2412	 * Note: This function is patched at runtime for performance reasons.
2413	 * 	 Any changes here require sfmmu_patch_ktsb fixed.
2414	 */
2415	ENTRY_NP(sfmmu_kprot_trap)
2416	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
2417sfmmu_kprot_patch_ktsb_base:
2418	RUNTIME_PATCH_SETX(%g1, %g6)
2419	/* %g1 = contents of ktsb_base or ktsb_pbase */
2420sfmmu_kprot_patch_ktsb_szcode:
2421	or	%g0, RUNTIME_PATCH, %g3	! ktsb_szcode (hot patched)
2422
2423	GET_TSBE_POINTER(MMU_PAGESHIFT, %g1, %g7, %g3, %g5)
2424	! %g1 = First TSB entry pointer, as TSB miss handler expects
2425
2426	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
2427sfmmu_kprot_patch_ktsb4m_base:
2428	RUNTIME_PATCH_SETX(%g3, %g6)
2429	/* %g3 = contents of ktsb4m_base or ktsb4m_pbase */
2430sfmmu_kprot_patch_ktsb4m_szcode:
2431	or	%g0, RUNTIME_PATCH, %g6	! ktsb4m_szcode (hot patched)
2432
2433	GET_TSBE_POINTER(MMU_PAGESHIFT4M, %g3, %g7, %g6, %g5)
2434	! %g3 = 4M tsb entry pointer, as TSB miss handler expects
2435
2436        CPU_TSBMISS_AREA(%g6, %g7)
2437        HAT_PERCPU_STAT16(%g6, TSBMISS_KPROTS, %g7)
2438	ba,pt	%xcc, sfmmu_tsb_miss_tt
2439	  nop
2440
2441	/*
2442	 * USER PROTECTION HANDLER
2443	 *
2444	 * g1 = tsb8k pointer register (ro)
2445	 * g2 = tag access register (ro)
2446	 * g3 = faulting context (clobbered, currently not used)
2447	 * g4 - g7 = scratch registers
2448	 */
2449	ALTENTRY(sfmmu_uprot_trap)
2450#ifdef sun4v
2451	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
2452	/* %g1 = first TSB entry ptr now, %g2 preserved */
2453
2454	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)	/* get 2nd utsbreg */
2455	brlz,pt %g3, 9f				/* check for 2nd TSB */
2456	  nop
2457
2458	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2459	/* %g3 = second TSB entry ptr now, %g2 preserved */
2460
2461#else /* sun4v */
2462#ifdef UTSB_PHYS
2463	/* g1 = first TSB entry ptr */
2464	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)
2465	brlz,pt %g3, 9f			/* check for 2nd TSB */
2466	  nop
2467
2468	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2469	/* %g3 = second TSB entry ptr now, %g2 preserved */
2470#else /* UTSB_PHYS */
2471	brgez,pt %g1, 9f		/* check for 2nd TSB */
2472	  mov	-1, %g3			/* set second tsbe ptr to -1 */
2473
2474	mov	%g2, %g7
2475	GET_2ND_TSBE_PTR(%g7, %g1, %g3, %g4, %g5, sfmmu_uprot)
2476	/* %g3 = second TSB entry ptr now, %g7 clobbered */
2477	mov	%g1, %g7
2478	GET_1ST_TSBE_PTR(%g7, %g1, %g5, sfmmu_uprot)
2479#endif /* UTSB_PHYS */
2480#endif /* sun4v */
24819:
2482	CPU_TSBMISS_AREA(%g6, %g7)
2483	HAT_PERCPU_STAT16(%g6, TSBMISS_UPROTS, %g7)
2484	ba,pt	%xcc, sfmmu_tsb_miss_tt		/* branch TSB miss handler */
2485	  nop
2486
2487	/*
2488	 * Kernel 8K page iTLB miss.  We also get here if we took a
2489	 * fast instruction access mmu miss trap while running in
2490	 * invalid context.
2491	 *
2492	 * %g1 = 8K TSB pointer register (not used, clobbered)
2493	 * %g2 = tag access register (used)
2494	 * %g3 = faulting context id (used)
2495	 * %g7 = TSB tag to match (used)
2496	 */
2497	.align	64
2498	ALTENTRY(sfmmu_kitlb_miss)
2499	brnz,pn %g3, tsb_tl0_noctxt
2500	  nop
2501
2502	/* kernel miss */
2503	/* get kernel tsb pointer */
2504	/* we patch the next set of instructions at run time */
2505	/* NOTE: any changes here require sfmmu_patch_ktsb fixed */
2506iktsbbase:
2507	RUNTIME_PATCH_SETX(%g4, %g5)
2508	/* %g4 = contents of ktsb_base or ktsb_pbase */
2509
2510iktsb:	sllx	%g2, 64-(TAGACC_SHIFT + TSB_START_SIZE + RUNTIME_PATCH), %g1
2511	srlx	%g1, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g1
2512	or	%g4, %g1, %g1			! form tsb ptr
2513	ldda	[%g1]RUNTIME_PATCH, %g4		! %g4 = tag, %g5 = data
2514	cmp	%g4, %g7
2515	bne,pn	%xcc, iktsb4mbase		! check 4m ktsb
2516	  srlx    %g2, MMU_PAGESHIFT4M, %g3	! use 4m virt-page as TSB index
2517
2518	andcc %g5, TTE_EXECPRM_INT, %g0		! check exec bit
2519	bz,pn	%icc, exec_fault
2520	  nop
2521	TT_TRACE(trace_tsbhit)			! 2 instr traptrace
2522	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2523	retry
2524
2525iktsb4mbase:
2526        RUNTIME_PATCH_SETX(%g4, %g6)
2527        /* %g4 = contents of ktsb4m_base or ktsb4m_pbase */
2528iktsb4m:
2529	sllx    %g3, 64-(TSB_START_SIZE + RUNTIME_PATCH), %g3
2530        srlx    %g3, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g3
2531	add	%g4, %g3, %g3			! %g3 = 4m tsbe ptr
2532	ldda	[%g3]RUNTIME_PATCH, %g4		! %g4 = tag, %g5 = data
2533	cmp	%g4, %g7
2534	bne,pn	%xcc, sfmmu_tsb_miss_tt		! branch on miss
2535	  andcc %g5, TTE_EXECPRM_INT, %g0		! check exec bit
2536	bz,pn	%icc, exec_fault
2537	  nop
2538	TT_TRACE(trace_tsbhit)			! 2 instr traptrace
2539	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2540	retry
2541
2542	/*
2543	 * Kernel dTLB miss.  We also get here if we took a fast data
2544	 * access mmu miss trap while running in invalid context.
2545	 *
2546	 * Note: for now we store kpm TTEs in the kernel TSB as usual.
2547	 *	We select the TSB miss handler to branch to depending on
2548	 *	the virtual address of the access.  In the future it may
2549	 *	be desirable to separate kpm TTEs into their own TSB,
2550	 *	in which case all that needs to be done is to set
2551	 *	kpm_tsbbase/kpm_tsbsz to point to the new TSB and branch
2552	 *	early in the miss if we detect a kpm VA to a new handler.
2553	 *
2554	 * %g1 = 8K TSB pointer register (not used, clobbered)
2555	 * %g2 = tag access register (used)
2556	 * %g3 = faulting context id (used)
2557	 */
2558	.align	64
2559	ALTENTRY(sfmmu_kdtlb_miss)
2560	brnz,pn	%g3, tsb_tl0_noctxt		/* invalid context? */
2561	  nop
2562
2563	/* Gather some stats for kpm misses in the TLB. */
2564	/* KPM_TLBMISS_STAT_INCR(tagacc, val, tsbma, tmp1, label) */
2565	KPM_TLBMISS_STAT_INCR(%g2, %g4, %g5, %g6, kpmtlbm_stat_out)
2566
2567	/*
2568	 * Get first TSB offset and look for 8K/64K/512K mapping
2569	 * using the 8K virtual page as the index.
2570	 *
2571	 * We patch the next set of instructions at run time;
2572	 * any changes here require sfmmu_patch_ktsb changes too.
2573	 */
2574dktsbbase:
2575	RUNTIME_PATCH_SETX(%g7, %g6)
2576	/* %g7 = contents of ktsb_base or ktsb_pbase */
2577
2578dktsb:	sllx	%g2, 64-(TAGACC_SHIFT + TSB_START_SIZE + RUNTIME_PATCH), %g1
2579	srlx	%g1, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g1
2580
2581	/*
2582	 * At this point %g1 is our index into the TSB.
2583	 * We just masked off enough bits of the VA depending
2584	 * on our TSB size code.
2585	 */
2586	ldda	[%g7 + %g1]RUNTIME_PATCH, %g4	! %g4 = tag, %g5 = data
2587	srlx	%g2, TAG_VALO_SHIFT, %g6	! make tag to compare
2588	cmp	%g6, %g4			! compare tag
2589	bne,pn	%xcc, dktsb4m_kpmcheck_small
2590	  add	%g7, %g1, %g1			/* form tsb ptr */
2591	TT_TRACE(trace_tsbhit)
2592	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2593	/* trapstat expects tte in %g5 */
2594	retry
2595
2596	/*
2597	 * If kpm is using large pages, the following instruction needs
2598	 * to be patched to a nop at boot time (by sfmmu_kpm_patch_tsbm)
2599	 * so that we will probe the 4M TSB regardless of the VA.  In
2600	 * the case kpm is using small pages, we know no large kernel
2601	 * mappings are located above 0x80000000.00000000 so we skip the
2602	 * probe as an optimization.
2603	 */
2604dktsb4m_kpmcheck_small:
2605	brlz,pn %g2, sfmmu_kpm_dtsb_miss_small
2606	  /* delay slot safe, below */
2607
2608	/*
2609	 * Get second TSB offset and look for 4M mapping
2610	 * using 4M virtual page as the TSB index.
2611	 *
2612	 * Here:
2613	 * %g1 = 8K TSB pointer.  Don't squash it.
2614	 * %g2 = tag access register (we still need it)
2615	 */
2616	srlx	%g2, MMU_PAGESHIFT4M, %g3
2617
2618	/*
2619	 * We patch the next set of instructions at run time;
2620	 * any changes here require sfmmu_patch_ktsb changes too.
2621	 */
2622dktsb4mbase:
2623	RUNTIME_PATCH_SETX(%g7, %g6)
2624	/* %g7 = contents of ktsb4m_base or ktsb4m_pbase */
2625dktsb4m:
2626	sllx	%g3, 64-(TSB_START_SIZE + RUNTIME_PATCH), %g3
2627	srlx	%g3, 64-(TSB_START_SIZE + TSB_ENTRY_SHIFT + RUNTIME_PATCH), %g3
2628
2629	/*
2630	 * At this point %g3 is our index into the TSB.
2631	 * We just masked off enough bits of the VA depending
2632	 * on our TSB size code.
2633	 */
2634	ldda	[%g7 + %g3]RUNTIME_PATCH, %g4	! %g4 = tag, %g5 = data
2635	srlx	%g2, TAG_VALO_SHIFT, %g6	! make tag to compare
2636	cmp	%g6, %g4			! compare tag
2637
2638dktsb4m_tsbmiss:
2639	bne,pn	%xcc, dktsb4m_kpmcheck
2640	  add	%g7, %g3, %g3			! %g3 = kernel second TSB ptr
2641	TT_TRACE(trace_tsbhit)
2642	/* we don't check TTE size here since we assume 4M TSB is separate */
2643	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
2644	/* trapstat expects tte in %g5 */
2645	retry
2646
2647	/*
2648	 * So, we failed to find a valid TTE to match the faulting
2649	 * address in either TSB.  There are a few cases that could land
2650	 * us here:
2651	 *
2652	 * 1) This is a kernel VA below 0x80000000.00000000.  We branch
2653	 *    to sfmmu_tsb_miss_tt to handle the miss.
2654	 * 2) We missed on a kpm VA, and we didn't find the mapping in the
2655	 *    4M TSB.  Let segkpm handle it.
2656	 *
2657	 * Note that we shouldn't land here in the case of a kpm VA when
2658	 * kpm_smallpages is active -- we handled that case earlier at
2659	 * dktsb4m_kpmcheck_small.
2660	 *
2661	 * At this point:
2662	 *  g1 = 8K-indexed primary TSB pointer
2663	 *  g2 = tag access register
2664	 *  g3 = 4M-indexed secondary TSB pointer
2665	 */
2666dktsb4m_kpmcheck:
2667	cmp	%g2, %g0
2668	bl,pn	%xcc, sfmmu_kpm_dtsb_miss
2669	  nop
2670	ba,a,pt	%icc, sfmmu_tsb_miss_tt
2671	  nop
2672
2673#ifdef sun4v
2674	/*
2675	 * User instruction miss w/ single TSB.
2676	 * The first probe covers 8K, 64K, and 512K page sizes,
2677	 * because 64K and 512K mappings are replicated off 8K
2678	 * pointer.
2679	 *
2680	 * g1 = tsb8k pointer register
2681	 * g2 = tag access register
2682	 * g3 - g6 = scratch registers
2683	 * g7 = TSB tag to match
2684	 */
2685	.align	64
2686	ALTENTRY(sfmmu_uitlb_fastpath)
2687
2688	PROBE_1ST_ITSB(%g1, %g7, uitlb_fast_8k_probefail)
2689	/* g4 - g5 = clobbered by PROBE_1ST_ITSB */
2690	ba,pn	%xcc, sfmmu_tsb_miss_tt
2691	  mov	-1, %g3
2692
2693	/*
2694	 * User data miss w/ single TSB.
2695	 * The first probe covers 8K, 64K, and 512K page sizes,
2696	 * because 64K and 512K mappings are replicated off 8K
2697	 * pointer.
2698	 *
2699	 * g1 = tsb8k pointer register
2700	 * g2 = tag access register
2701	 * g3 - g6 = scratch registers
2702	 * g7 = TSB tag to match
2703	 */
2704	.align 64
2705	ALTENTRY(sfmmu_udtlb_fastpath)
2706
2707	PROBE_1ST_DTSB(%g1, %g7, udtlb_fast_8k_probefail)
2708	/* g4 - g5 = clobbered by PROBE_1ST_DTSB */
2709	ba,pn	%xcc, sfmmu_tsb_miss_tt
2710	  mov	-1, %g3
2711
2712	/*
2713	 * User instruction miss w/ multiple TSBs (sun4v).
2714	 * The first probe covers 8K, 64K, and 512K page sizes,
2715	 * because 64K and 512K mappings are replicated off 8K
2716	 * pointer.  Second probe covers 4M page size only.
2717	 *
2718	 * Just like sfmmu_udtlb_slowpath, except:
2719	 *   o Uses ASI_ITLB_IN
2720	 *   o checks for execute permission
2721	 *   o No ISM prediction.
2722	 *
2723	 * g1 = tsb8k pointer register
2724	 * g2 = tag access register
2725	 * g3 - g6 = scratch registers
2726	 * g7 = TSB tag to match
2727	 */
2728	.align	64
2729	ALTENTRY(sfmmu_uitlb_slowpath)
2730
2731	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
2732	PROBE_1ST_ITSB(%g1, %g7, uitlb_8k_probefail)
2733	/* g4 - g5 = clobbered here */
2734
2735	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2736	/* g1 = first TSB pointer, g3 = second TSB pointer */
2737	srlx	%g2, TAG_VALO_SHIFT, %g7
2738	PROBE_2ND_ITSB(%g3, %g7)
2739	/* NOT REACHED */
2740
2741#else /* sun4v */
2742
2743	/*
2744	 * User instruction miss w/ multiple TSBs (sun4u).
2745	 * The first probe covers 8K, 64K, and 512K page sizes,
2746	 * because 64K and 512K mappings are replicated off 8K
2747	 * pointer.  Probe of 1st TSB has already been done prior to entry
2748	 * into this routine. For the UTSB_PHYS case we probe up to 3
2749	 * valid other TSBs in the following order:
2750	 * 1) shared TSB for 4M-256M pages
2751	 * 2) private TSB for 4M-256M pages
2752	 * 3) shared TSB for 8K-512K pages
2753	 *
2754	 * For the non UTSB_PHYS case we probe the 2nd TSB here that backs
2755	 * 4M-256M pages.
2756	 *
2757	 * Just like sfmmu_udtlb_slowpath, except:
2758	 *   o Uses ASI_ITLB_IN
2759	 *   o checks for execute permission
2760	 *   o No ISM prediction.
2761	 *
2762	 * g1 = tsb8k pointer register
2763	 * g2 = tag access register
2764	 * g4 - g6 = scratch registers
2765	 * g7 = TSB tag to match
2766	 */
2767	.align	64
2768	ALTENTRY(sfmmu_uitlb_slowpath)
2769
2770#ifdef UTSB_PHYS
2771
2772       GET_UTSBREG(SCRATCHPAD_UTSBREG4, %g6)
2773        brlz,pt %g6, 1f
2774          nop
2775        GET_4TH_TSBE_PTR(%g2, %g6, %g4, %g5)
2776        PROBE_4TH_ITSB(%g6, %g7, uitlb_4m_scd_probefail)
27771:
2778        GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)
2779        brlz,pt %g3, 2f
2780          nop
2781        GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2782        PROBE_2ND_ITSB(%g3, %g7, uitlb_4m_probefail)
27832:
2784        GET_UTSBREG(SCRATCHPAD_UTSBREG3, %g6)
2785        brlz,pt %g6, sfmmu_tsb_miss_tt
2786          nop
2787        GET_3RD_TSBE_PTR(%g2, %g6, %g4, %g5)
2788        PROBE_3RD_ITSB(%g6, %g7, uitlb_8K_scd_probefail)
2789        ba,pn   %xcc, sfmmu_tsb_miss_tt
2790          nop
2791
2792#else /* UTSB_PHYS */
2793	mov	%g1, %g3	/* save tsb8k reg in %g3 */
2794	GET_1ST_TSBE_PTR(%g3, %g1, %g5, sfmmu_uitlb)
2795	PROBE_1ST_ITSB(%g1, %g7, uitlb_8k_probefail)
2796	mov	%g2, %g6	/* GET_2ND_TSBE_PTR clobbers tagacc */
2797	mov	%g3, %g7	/* copy tsb8k reg in %g7 */
2798	GET_2ND_TSBE_PTR(%g6, %g7, %g3, %g4, %g5, sfmmu_uitlb)
2799       /* g1 = first TSB pointer, g3 = second TSB pointer */
2800        srlx    %g2, TAG_VALO_SHIFT, %g7
2801        PROBE_2ND_ITSB(%g3, %g7, isynth)
2802	ba,pn	%xcc, sfmmu_tsb_miss_tt
2803	  nop
2804
2805#endif /* UTSB_PHYS */
2806#endif /* sun4v */
2807
2808#if defined(sun4u) && defined(UTSB_PHYS)
2809
2810        /*
2811	 * We come here for ism predict DTLB_MISS case or if
2812	 * if probe in first TSB failed.
2813         */
2814
2815        .align 64
2816        ALTENTRY(sfmmu_udtlb_slowpath_noismpred)
2817
2818	/*
2819         * g1 = tsb8k pointer register
2820         * g2 = tag access register
2821         * g4 - %g6 = scratch registers
2822         * g7 = TSB tag to match
2823	 */
2824
2825	/*
2826	 * ISM non-predict probe order
2827         * probe 1ST_TSB (8K index)
2828         * probe 2ND_TSB (4M index)
2829         * probe 4TH_TSB (4M index)
2830         * probe 3RD_TSB (8K index)
2831	 *
2832	 * We already probed first TSB in DTLB_MISS handler.
2833	 */
2834
2835        /*
2836         * Private 2ND TSB 4M-256 pages
2837         */
2838	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)
2839	brlz,pt %g3, 1f
2840	  nop
2841        GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2842        PROBE_2ND_DTSB(%g3, %g7, udtlb_4m_probefail)
2843
2844	/*
2845	 * Shared Context 4TH TSB 4M-256 pages
2846	 */
28471:
2848	GET_UTSBREG(SCRATCHPAD_UTSBREG4, %g6)
2849	brlz,pt %g6, 2f
2850	  nop
2851        GET_4TH_TSBE_PTR(%g2, %g6, %g4, %g5)
2852        PROBE_4TH_DTSB(%g6, %g7, udtlb_4m_shctx_probefail)
2853
2854        /*
2855         * Shared Context 3RD TSB 8K-512K pages
2856         */
28572:
2858	GET_UTSBREG(SCRATCHPAD_UTSBREG3, %g6)
2859	brlz,pt %g6, sfmmu_tsb_miss_tt
2860	  nop
2861        GET_3RD_TSBE_PTR(%g2, %g6, %g4, %g5)
2862        PROBE_3RD_DTSB(%g6, %g7, udtlb_8k_shctx_probefail)
2863	ba,pn	%xcc, sfmmu_tsb_miss_tt
2864	  nop
2865
2866	.align 64
2867        ALTENTRY(sfmmu_udtlb_slowpath_ismpred)
2868
2869	/*
2870         * g1 = tsb8k pointer register
2871         * g2 = tag access register
2872         * g4 - g6 = scratch registers
2873         * g7 = TSB tag to match
2874	 */
2875
2876	/*
2877	 * ISM predict probe order
2878	 * probe 4TH_TSB (4M index)
2879	 * probe 2ND_TSB (4M index)
2880	 * probe 1ST_TSB (8K index)
2881	 * probe 3RD_TSB (8K index)
2882
2883	/*
2884	 * Shared Context 4TH TSB 4M-256 pages
2885	 */
2886	GET_UTSBREG(SCRATCHPAD_UTSBREG4, %g6)
2887	brlz,pt %g6, 4f
2888	  nop
2889        GET_4TH_TSBE_PTR(%g2, %g6, %g4, %g5)
2890        PROBE_4TH_DTSB(%g6, %g7, udtlb_4m_shctx_probefail2)
2891
2892        /*
2893         * Private 2ND TSB 4M-256 pages
2894         */
28954:
2896	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)
2897	brlz,pt %g3, 5f
2898	  nop
2899        GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2900        PROBE_2ND_DTSB(%g3, %g7, udtlb_4m_probefail2)
2901
29025:
2903        PROBE_1ST_DTSB(%g1, %g7, udtlb_8k_first_probefail2)
2904
2905        /*
2906         * Shared Context 3RD TSB 8K-512K pages
2907         */
2908	GET_UTSBREG(SCRATCHPAD_UTSBREG3, %g6)
2909	brlz,pt %g6, 6f
2910	  nop
2911        GET_3RD_TSBE_PTR(%g2, %g6, %g4, %g5)
2912        PROBE_3RD_DTSB(%g6, %g7, udtlb_8k_shctx_probefail2)
29136:
2914	ba,pn	%xcc, sfmmu_tsb_miss_tt /* ISM Predict and ISM non-predict path */
2915	  nop
2916
2917#else /* sun4u && UTSB_PHYS */
2918
2919       .align 64
2920        ALTENTRY(sfmmu_udtlb_slowpath)
2921
2922	srax	%g2, PREDISM_BASESHIFT, %g6	/* g6 > 0 : ISM predicted */
2923	brgz,pn %g6, udtlb_miss_probesecond	/* check for ISM */
2924	  mov	%g1, %g3
2925
2926udtlb_miss_probefirst:
2927	/*
2928	 * g1 = 8K TSB pointer register
2929	 * g2 = tag access register
2930	 * g3 = (potentially) second TSB entry ptr
2931	 * g6 = ism pred.
2932	 * g7 = vpg_4m
2933	 */
2934#ifdef sun4v
2935	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
2936	PROBE_1ST_DTSB(%g1, %g7, udtlb_first_probefail)
2937
2938	/*
2939	 * Here:
2940	 *   g1 = first TSB pointer
2941	 *   g2 = tag access reg
2942	 *   g3 = second TSB ptr IFF ISM pred. (else don't care)
2943	 */
2944	brgz,pn	%g6, sfmmu_tsb_miss_tt
2945	  nop
2946#else /* sun4v */
2947	mov	%g1, %g4
2948	GET_1ST_TSBE_PTR(%g4, %g1, %g5, sfmmu_udtlb)
2949	PROBE_1ST_DTSB(%g1, %g7, udtlb_first_probefail)
2950
2951	/*
2952	 * Here:
2953	 *   g1 = first TSB pointer
2954	 *   g2 = tag access reg
2955	 *   g3 = second TSB ptr IFF ISM pred. (else don't care)
2956	 */
2957	brgz,pn	%g6, sfmmu_tsb_miss_tt
2958	  nop
2959	ldxa	[%g0]ASI_DMMU_TSB_8K, %g3
2960	/* fall through in 8K->4M probe order */
2961#endif /* sun4v */
2962
2963udtlb_miss_probesecond:
2964	/*
2965	 * Look in the second TSB for the TTE
2966	 * g1 = First TSB entry ptr if !ISM pred, TSB8K ptr reg if ISM pred.
2967	 * g2 = tag access reg
2968	 * g3 = 8K TSB pointer register
2969	 * g6 = ism pred.
2970	 * g7 = vpg_4m
2971	 */
2972#ifdef sun4v
2973	/* GET_2ND_TSBE_PTR(tagacc, tsbe_ptr, tmp1, tmp2) */
2974	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
2975	/* %g2 is okay, no need to reload, %g3 = second tsbe ptr */
2976#else /* sun4v */
2977	mov	%g3, %g7
2978	GET_2ND_TSBE_PTR(%g2, %g7, %g3, %g4, %g5, sfmmu_udtlb)
2979	/* %g2 clobbered, %g3 =second tsbe ptr */
2980	mov	MMU_TAG_ACCESS, %g2
2981	ldxa	[%g2]ASI_DMMU, %g2
2982#endif /* sun4v */
2983
2984	srlx	%g2, TAG_VALO_SHIFT, %g7
2985	PROBE_2ND_DTSB(%g3, %g7, udtlb_4m_probefail)
2986	/* g4 - g5 = clobbered here; %g7 still vpg_4m at this point */
2987	brgz,pn	%g6, udtlb_miss_probefirst
2988	  nop
2989
2990	/* fall through to sfmmu_tsb_miss_tt */
2991#endif /* sun4u && UTSB_PHYS */
2992
2993
2994	ALTENTRY(sfmmu_tsb_miss_tt)
2995	TT_TRACE(trace_tsbmiss)
2996	/*
2997	 * We get here if there is a TSB miss OR a write protect trap.
2998	 *
2999	 * g1 = First TSB entry pointer
3000	 * g2 = tag access register
3001	 * g3 = 4M TSB entry pointer; -1 if no 2nd TSB
3002	 * g4 - g7 = scratch registers
3003	 */
3004
3005	ALTENTRY(sfmmu_tsb_miss)
3006
3007	/*
3008	 * If trapstat is running, we need to shift the %tpc and %tnpc to
3009	 * point to trapstat's TSB miss return code (note that trapstat
3010	 * itself will patch the correct offset to add).
3011	 */
3012	rdpr	%tl, %g7
3013	cmp	%g7, 1
3014	ble,pt	%xcc, 0f
3015	  sethi	%hi(KERNELBASE), %g6
3016	rdpr	%tpc, %g7
3017	or	%g6, %lo(KERNELBASE), %g6
3018	cmp	%g7, %g6
3019	bgeu,pt	%xcc, 0f
3020	/* delay slot safe */
3021
3022	ALTENTRY(tsbmiss_trapstat_patch_point)
3023	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
3024	wrpr	%g7, %tpc
3025	add	%g7, 4, %g7
3026	wrpr	%g7, %tnpc
30270:
3028	CPU_TSBMISS_AREA(%g6, %g7)
3029	stn	%g1, [%g6 + TSBMISS_TSBPTR]	/* save 1ST tsb pointer */
3030	stn	%g3, [%g6 + TSBMISS_TSBPTR4M]	/* save 2ND tsb pointer */
3031
3032	sllx	%g2, TAGACC_CTX_LSHIFT, %g3
3033	brz,a,pn %g3, 1f			/* skip ahead if kernel */
3034	  ldn	[%g6 + TSBMISS_KHATID], %g7
3035	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctxnum */
3036	ldn	[%g6 + TSBMISS_UHATID], %g7     /* g7 = hatid */
3037
3038	HAT_PERCPU_STAT32(%g6, TSBMISS_UTSBMISS, %g5)
3039
3040	cmp	%g3, INVALID_CONTEXT
3041	be,pn	%icc, tsb_tl0_noctxt		/* no ctx miss exception */
3042	  stn	%g7, [%g6 + (TSBMISS_SCRATCH + TSBMISS_HATID)]
3043
3044#if defined(sun4v) || defined(UTSB_PHYS)
3045        ldub    [%g6 + TSBMISS_URTTEFLAGS], %g7	/* clear ctx1 flag set from */
3046        andn    %g7, HAT_CHKCTX1_FLAG, %g7	/* the previous tsb miss    */
3047        stub    %g7, [%g6 + TSBMISS_URTTEFLAGS]
3048#endif /* sun4v || UTSB_PHYS */
3049
3050	ISM_CHECK(%g2, %g6, %g3, %g4, %g5, %g7, %g1, tsb_l1, tsb_ism)
3051	/*
3052	 * The miss wasn't in an ISM segment.
3053	 *
3054	 * %g1 %g3, %g4, %g5, %g7 all clobbered
3055	 * %g2 = (pseudo) tag access
3056	 */
3057
3058	ba,pt	%icc, 2f
3059	  ldn	[%g6 + (TSBMISS_SCRATCH + TSBMISS_HATID)], %g7
3060
30611:
3062	HAT_PERCPU_STAT32(%g6, TSBMISS_KTSBMISS, %g5)
3063	/*
3064	 * 8K and 64K hash.
3065	 */
30662:
3067
3068	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3069		MMU_PAGESHIFT64K, TTE64K, %g5, tsb_l8K, tsb_checktte,
3070		sfmmu_suspend_tl, tsb_512K)
3071	/* NOT REACHED */
3072
3073tsb_512K:
3074	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3075	brz,pn	%g5, 3f
3076	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3077	and	%g4, HAT_512K_FLAG, %g5
3078
3079	/*
3080	 * Note that there is a small window here where we may have
3081	 * a 512k page in the hash list but have not set the HAT_512K_FLAG
3082	 * flag yet, so we will skip searching the 512k hash list.
3083	 * In this case we will end up in pagefault which will find
3084	 * the mapping and return.  So, in this instance we will end up
3085	 * spending a bit more time resolving this TSB miss, but it can
3086	 * only happen once per process and even then, the chances of that
3087	 * are very small, so it's not worth the extra overhead it would
3088	 * take to close this window.
3089	 */
3090	brz,pn	%g5, tsb_4M
3091	  nop
30923:
3093	/*
3094	 * 512K hash
3095	 */
3096
3097	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3098		MMU_PAGESHIFT512K, TTE512K, %g5, tsb_l512K, tsb_checktte,
3099		sfmmu_suspend_tl, tsb_4M)
3100	/* NOT REACHED */
3101
3102tsb_4M:
3103	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3104	brz,pn	%g5, 4f
3105	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3106	and	%g4, HAT_4M_FLAG, %g5
3107	brz,pn	%g5, tsb_32M
3108	  nop
31094:
3110	/*
3111	 * 4M hash
3112	 */
3113
3114	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3115		MMU_PAGESHIFT4M, TTE4M, %g5, tsb_l4M, tsb_checktte,
3116		sfmmu_suspend_tl, tsb_32M)
3117	/* NOT REACHED */
3118
3119tsb_32M:
3120	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3121#ifdef	sun4v
3122        brz,pn	%g5, 6f
3123#else
3124	brz,pn  %g5, tsb_pagefault
3125#endif
3126	  ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3127	and	%g4, HAT_32M_FLAG, %g5
3128	brz,pn	%g5, tsb_256M
3129	  nop
31305:
3131	/*
3132	 * 32M hash
3133	 */
3134
3135	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3136		MMU_PAGESHIFT32M, TTE32M, %g5, tsb_l32M, tsb_checktte,
3137		sfmmu_suspend_tl, tsb_256M)
3138	/* NOT REACHED */
3139
3140#if defined(sun4u) && !defined(UTSB_PHYS)
3141#define tsb_shme        tsb_pagefault
3142#endif
3143tsb_256M:
3144	ldub	[%g6 + TSBMISS_UTTEFLAGS], %g4
3145	and	%g4, HAT_256M_FLAG, %g5
3146	brz,pn	%g5, tsb_shme
3147	  nop
31486:
3149	/*
3150	 * 256M hash
3151	 */
3152
3153	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3154	    MMU_PAGESHIFT256M, TTE256M, %g5, tsb_l256M, tsb_checktte,
3155	    sfmmu_suspend_tl, tsb_shme)
3156	/* NOT REACHED */
3157
3158tsb_checktte:
3159	/*
3160	 * g1 = hblk_misc
3161	 * g2 = tagacc
3162	 * g3 = tte
3163	 * g4 = tte pa
3164	 * g6 = tsbmiss area
3165	 * g7 = hatid
3166	 */
3167	brlz,a,pt %g3, tsb_validtte
3168	  rdpr	%tt, %g7
3169
3170#if defined(sun4u) && !defined(UTSB_PHYS)
3171#undef tsb_shme
3172	ba      tsb_pagefault
3173	  nop
3174#else /* sun4u && !UTSB_PHYS */
3175
3176tsb_shme:
3177	/*
3178	 * g2 = tagacc
3179	 * g6 = tsbmiss area
3180	 */
3181	sllx	%g2, TAGACC_CTX_LSHIFT, %g5
3182	brz,pn	%g5, tsb_pagefault
3183	  nop
3184	ldx	[%g6 + TSBMISS_SHARED_UHATID], %g7	/* g7 = srdp */
3185	brz,pn	%g7, tsb_pagefault
3186	  nop
3187
3188	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3189		MMU_PAGESHIFT64K, TTE64K, %g5, tsb_shme_l8K, tsb_shme_checktte,
3190		sfmmu_suspend_tl, tsb_shme_512K)
3191	/* NOT REACHED */
3192
3193tsb_shme_512K:
3194	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3195	and	%g4, HAT_512K_FLAG, %g5
3196	brz,pn	%g5, tsb_shme_4M
3197	  nop
3198
3199	/*
3200	 * 512K hash
3201	 */
3202
3203	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3204		MMU_PAGESHIFT512K, TTE512K, %g5, tsb_shme_l512K, tsb_shme_checktte,
3205		sfmmu_suspend_tl, tsb_shme_4M)
3206	/* NOT REACHED */
3207
3208tsb_shme_4M:
3209	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3210	and	%g4, HAT_4M_FLAG, %g5
3211	brz,pn	%g5, tsb_shme_32M
3212	  nop
32134:
3214	/*
3215	 * 4M hash
3216	 */
3217	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3218		MMU_PAGESHIFT4M, TTE4M, %g5, tsb_shme_l4M, tsb_shme_checktte,
3219		sfmmu_suspend_tl, tsb_shme_32M)
3220	/* NOT REACHED */
3221
3222tsb_shme_32M:
3223	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3224	and	%g4, HAT_32M_FLAG, %g5
3225	brz,pn	%g5, tsb_shme_256M
3226	  nop
3227
3228	/*
3229	 * 32M hash
3230	 */
3231
3232	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3233		MMU_PAGESHIFT32M, TTE32M, %g5, tsb_shme_l32M, tsb_shme_checktte,
3234		sfmmu_suspend_tl, tsb_shme_256M)
3235	/* NOT REACHED */
3236
3237tsb_shme_256M:
3238	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g4
3239	and	%g4, HAT_256M_FLAG, %g5
3240	brz,pn	%g5, tsb_pagefault
3241	  nop
3242
3243	/*
3244	 * 256M hash
3245	 */
3246
3247	GET_SHME_TTE(%g2, %g7, %g3, %g4, %g6, %g1,
3248	    MMU_PAGESHIFT256M, TTE256M, %g5, tsb_shme_l256M, tsb_shme_checktte,
3249	    sfmmu_suspend_tl, tsb_pagefault)
3250	/* NOT REACHED */
3251
3252tsb_shme_checktte:
3253
3254	brgez,pn %g3, tsb_pagefault
3255	  rdpr	%tt, %g7
3256	/*
3257	 * g1 = ctx1 flag
3258	 * g3 = tte
3259	 * g4 = tte pa
3260	 * g6 = tsbmiss area
3261	 * g7 = tt
3262	 */
3263
3264	brz,pt  %g1, tsb_validtte
3265	  nop
3266	ldub    [%g6 + TSBMISS_URTTEFLAGS], %g1
3267	  or	%g1, HAT_CHKCTX1_FLAG, %g1
3268	stub    %g1, [%g6 + TSBMISS_URTTEFLAGS]
3269
3270	SAVE_CTX1(%g7, %g2, %g1, tsb_shmel)
3271#endif /* sun4u && !UTSB_PHYS */
3272
3273tsb_validtte:
3274	/*
3275	 * g3 = tte
3276	 * g4 = tte pa
3277	 * g6 = tsbmiss area
3278	 * g7 = tt
3279	 */
3280
3281	/*
3282	 * Set ref/mod bits if this is a prot trap.  Usually, it isn't.
3283	 */
3284	cmp	%g7, FAST_PROT_TT
3285	bne,pt	%icc, 4f
3286	  nop
3287
3288	TTE_SET_REFMOD_ML(%g3, %g4, %g6, %g7, %g5, tsb_lset_refmod,
3289	    tsb_protfault)
3290
3291	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
3292#ifdef sun4v
3293	MMU_FAULT_STATUS_AREA(%g7)
3294	ldx	[%g7 + MMFSA_D_ADDR], %g5	/* load fault addr for later */
3295#else /* sun4v */
3296	mov     MMU_TAG_ACCESS, %g5
3297	ldxa    [%g5]ASI_DMMU, %g5
3298#endif /* sun4v */
3299	ba,pt	%xcc, tsb_update_tl1
3300	  nop
33014:
3302	/*
3303	 * If ITLB miss check exec bit.
3304	 * If not set treat as invalid TTE.
3305	 */
3306	cmp     %g7, T_INSTR_MMU_MISS
3307	be,pn	%icc, 5f
3308	  andcc   %g3, TTE_EXECPRM_INT, %g0	/* check execute bit is set */
3309	cmp     %g7, FAST_IMMU_MISS_TT
3310	bne,pt %icc, 3f
3311	  andcc   %g3, TTE_EXECPRM_INT, %g0	/* check execute bit is set */
33125:
3313	bz,pn %icc, tsb_protfault
3314	  nop
3315
33163:
3317	/*
3318	 * Set reference bit if not already set
3319	 */
3320	TTE_SET_REF_ML(%g3, %g4, %g6, %g7, %g5, tsb_lset_ref)
3321
3322	/*
3323	 * Now, load into TSB/TLB.  At this point:
3324	 * g3 = tte
3325	 * g4 = patte
3326	 * g6 = tsbmiss area
3327	 */
3328	rdpr	%tt, %g7
3329#ifdef sun4v
3330	MMU_FAULT_STATUS_AREA(%g2)
3331	cmp	%g7, T_INSTR_MMU_MISS
3332	be,a,pt	%icc, 9f
3333	  nop
3334	cmp	%g7, FAST_IMMU_MISS_TT
3335	be,a,pt	%icc, 9f
3336	  nop
3337	add	%g2, MMFSA_D_, %g2
33389:
3339	ldx	[%g2 + MMFSA_CTX_], %g7
3340	sllx	%g7, TTARGET_CTX_SHIFT, %g7
3341	ldx	[%g2 + MMFSA_ADDR_], %g2
3342	mov	%g2, %g5		! load the fault addr for later use
3343	srlx	%g2, TTARGET_VA_SHIFT, %g2
3344	or	%g2, %g7, %g2
3345#else /* sun4v */
3346	mov     MMU_TAG_ACCESS, %g5
3347	cmp     %g7, FAST_IMMU_MISS_TT
3348	be,a,pt %icc, 9f
3349	   ldxa  [%g0]ASI_IMMU, %g2
3350	ldxa    [%g0]ASI_DMMU, %g2
3351	ba,pt   %icc, tsb_update_tl1
3352	   ldxa  [%g5]ASI_DMMU, %g5
33539:
3354	ldxa    [%g5]ASI_IMMU, %g5
3355#endif /* sun4v */
3356
3357tsb_update_tl1:
3358	srlx	%g2, TTARGET_CTX_SHIFT, %g7
3359	brz,pn	%g7, tsb_kernel
3360#ifdef sun4v
3361	  and	%g3, TTE_SZ_BITS, %g7	! assumes TTE_SZ_SHFT is 0
3362#else  /* sun4v */
3363	  srlx	%g3, TTE_SZ_SHFT, %g7
3364#endif /* sun4v */
3365
3366tsb_user:
3367#ifdef sun4v
3368	cmp	%g7, TTE4M
3369	bge,pn	%icc, tsb_user4m
3370	  nop
3371#else /* sun4v */
3372	cmp	%g7, TTESZ_VALID | TTE4M
3373	be,pn	%icc, tsb_user4m
3374	  srlx	%g3, TTE_SZ2_SHFT, %g7
3375	andcc	%g7, TTE_SZ2_BITS, %g7		! check 32/256MB
3376#ifdef ITLB_32M_256M_SUPPORT
3377	bnz,pn	%icc, tsb_user4m
3378	  nop
3379#else /* ITLB_32M_256M_SUPPORT */
3380	bnz,a,pn %icc, tsb_user_pn_synth
3381	 nop
3382#endif /* ITLB_32M_256M_SUPPORT */
3383#endif /* sun4v */
3384
3385tsb_user8k:
3386#if defined(sun4v) || defined(UTSB_PHYS)
3387	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g7
3388	and	%g7, HAT_CHKCTX1_FLAG, %g1
3389	brz,a,pn %g1, 1f
3390	  ldn	[%g6 + TSBMISS_TSBPTR], %g1		! g1 = 1ST TSB ptr
3391	GET_UTSBREG_SHCTX(%g6, TSBMISS_TSBSCDPTR, %g1)
3392	brlz,a,pn %g1, ptl1_panic			! if no shared 3RD tsb
3393	  mov PTL1_NO_SCDTSB8K, %g1			! panic
3394        GET_3RD_TSBE_PTR(%g5, %g1, %g6, %g7)
33951:
3396#else /* defined(sun4v) || defined(UTSB_PHYS) */
3397	ldn   [%g6 + TSBMISS_TSBPTR], %g1             ! g1 = 1ST TSB ptr
3398#endif /* defined(sun4v) || defined(UTSB_PHYS) */
3399
3400#ifndef UTSB_PHYS
3401	mov	ASI_N, %g7	! user TSBs accessed by VA
3402	mov	%g7, %asi
3403#endif /* !UTSB_PHYS */
3404
3405	TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, locked_tsb_l3)
3406
3407	rdpr    %tt, %g5
3408#ifdef sun4v
3409	cmp	%g5, T_INSTR_MMU_MISS
3410	be,a,pn	%xcc, 9f
3411	  mov	%g3, %g5
3412#endif /* sun4v */
3413	cmp	%g5, FAST_IMMU_MISS_TT
3414	be,pn	%xcc, 9f
3415	  mov	%g3, %g5
3416
3417	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3418	! trapstat wants TTE in %g5
3419	retry
34209:
3421	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3422	! trapstat wants TTE in %g5
3423	retry
3424
3425tsb_user4m:
3426#if defined(sun4v) || defined(UTSB_PHYS)
3427	ldub	[%g6 + TSBMISS_URTTEFLAGS], %g7
3428	and	%g7, HAT_CHKCTX1_FLAG, %g1
3429	brz,a,pn %g1, 4f
3430	  ldn	[%g6 + TSBMISS_TSBPTR4M], %g1		! g1 = 2ND TSB ptr
3431	GET_UTSBREG_SHCTX(%g6, TSBMISS_TSBSCDPTR4M, %g1)! g1 = 4TH TSB ptr
3432	brlz,a,pn %g1, 5f				! if no shared 4TH TSB
3433	  nop
3434        GET_4TH_TSBE_PTR(%g5, %g1, %g6, %g7)
3435
3436#else /* defined(sun4v) || defined(UTSB_PHYS) */
3437	ldn   [%g6 + TSBMISS_TSBPTR4M], %g1             ! g1 = 2ND TSB ptr
3438#endif /* defined(sun4v) || defined(UTSB_PHYS) */
34394:
3440	brlz,pn %g1, 5f	/* Check to see if we have 2nd TSB programmed */
3441	  nop
3442
3443#ifndef UTSB_PHYS
3444	mov	ASI_N, %g7	! user TSBs accessed by VA
3445	mov	%g7, %asi
3446#endif /* UTSB_PHYS */
3447
3448        TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, locked_tsb_l4)
3449
34505:
3451	rdpr    %tt, %g5
3452#ifdef sun4v
3453        cmp     %g5, T_INSTR_MMU_MISS
3454        be,a,pn %xcc, 9f
3455          mov   %g3, %g5
3456#endif /* sun4v */
3457        cmp     %g5, FAST_IMMU_MISS_TT
3458        be,pn   %xcc, 9f
3459        mov     %g3, %g5
3460
3461        DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3462        ! trapstat wants TTE in %g5
3463        retry
34649:
3465        ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3466        ! trapstat wants TTE in %g5
3467        retry
3468
3469#if !defined(sun4v) && !defined(ITLB_32M_256M_SUPPORT)
3470	/*
3471	 * Panther ITLB synthesis.
3472	 * The Panther 32M and 256M ITLB code simulates these two large page
3473	 * sizes with 4M pages, to provide support for programs, for example
3474	 * Java, that may copy instructions into a 32M or 256M data page and
3475	 * then execute them. The code below generates the 4M pfn bits and
3476	 * saves them in the modified 32M/256M ttes in the TSB. If the tte is
3477	 * stored in the DTLB to map a 32M/256M page, the 4M pfn offset bits
3478	 * are ignored by the hardware.
3479	 *
3480	 * Now, load into TSB/TLB.  At this point:
3481	 * g2 = tagtarget
3482	 * g3 = tte
3483	 * g4 = patte
3484	 * g5 = tt
3485	 * g6 = tsbmiss area
3486	 */
3487tsb_user_pn_synth:
3488	rdpr %tt, %g5
3489	cmp    %g5, FAST_IMMU_MISS_TT
3490	be,pt	%xcc, tsb_user_itlb_synth	/* ITLB miss */
3491	  andcc %g3, TTE_EXECPRM_INT, %g0	/* is execprm bit set */
3492	bz,pn %icc, 4b				/* if not, been here before */
3493	  ldn	[%g6 + TSBMISS_TSBPTR4M], %g1	/* g1 = tsbp */
3494	brlz,a,pn %g1, 5f			/* no 2nd tsb */
3495	  mov	%g3, %g5
3496
3497	mov	MMU_TAG_ACCESS, %g7
3498	ldxa	[%g7]ASI_DMMU, %g6		/* get tag access va */
3499	GET_4M_PFN_OFF(%g3, %g6, %g5, %g7, 1)	/* make 4M pfn offset */
3500
3501	mov	ASI_N, %g7	/* user TSBs always accessed by VA */
3502	mov	%g7, %asi
3503	TSB_UPDATE_TL_PN(%g1, %g5, %g2, %g4, %g7, %g3, locked_tsb_l5) /* update TSB */
35045:
3505        DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3506        retry
3507
3508tsb_user_itlb_synth:
3509	ldn	[%g6 + TSBMISS_TSBPTR4M], %g1		/* g1 =  2ND TSB */
3510
3511	mov	MMU_TAG_ACCESS, %g7
3512	ldxa	[%g7]ASI_IMMU, %g6		/* get tag access va */
3513	GET_4M_PFN_OFF(%g3, %g6, %g5, %g7, 2)	/* make 4M pfn offset */
3514	brlz,a,pn %g1, 7f	/* Check to see if we have 2nd TSB programmed */
3515	  or	%g5, %g3, %g5			/* add 4M bits to TTE */
3516
3517	mov	ASI_N, %g7	/* user TSBs always accessed by VA */
3518	mov	%g7, %asi
3519	TSB_UPDATE_TL_PN(%g1, %g5, %g2, %g4, %g7, %g3, locked_tsb_l6) /* update TSB */
35207:
3521	SET_TTE4M_PN(%g5, %g7)			/* add TTE4M pagesize to TTE */
3522        ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3523        retry
3524#endif /* sun4v && ITLB_32M_256M_SUPPORT */
3525
3526tsb_kernel:
3527	rdpr	%tt, %g5
3528#ifdef sun4v
3529	cmp	%g7, TTE4M
3530	bge,pn	%icc, 5f
3531#else
3532	cmp	%g7, TTESZ_VALID | TTE4M	! no 32M or 256M support
3533	be,pn	%icc, 5f
3534#endif /* sun4v */
3535	  nop
3536	ldn	[%g6 + TSBMISS_TSBPTR], %g1	! g1 = 8K TSB ptr
3537	ba,pt	%xcc, 6f
3538	  nop
35395:
3540	ldn	[%g6 + TSBMISS_TSBPTR4M], %g1	! g1 = 4M TSB ptr
3541	brlz,pn	%g1, 3f		/* skip programming if 4M TSB ptr is -1 */
3542	  nop
35436:
3544#ifndef sun4v
3545tsb_kernel_patch_asi:
3546	or	%g0, RUNTIME_PATCH, %g6
3547	mov	%g6, %asi	! XXX avoid writing to %asi !!
3548#endif
3549	TSB_UPDATE_TL(%g1, %g3, %g2, %g4, %g7, %g6, locked_tsb_l7)
35503:
3551#ifdef sun4v
3552	cmp	%g5, T_INSTR_MMU_MISS
3553	be,a,pn	%icc, 1f
3554	  mov	%g3, %g5			! trapstat wants TTE in %g5
3555#endif /* sun4v */
3556	cmp	%g5, FAST_IMMU_MISS_TT
3557	be,pn	%icc, 1f
3558	  mov	%g3, %g5			! trapstat wants TTE in %g5
3559	DTLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3560	! trapstat wants TTE in %g5
3561	retry
35621:
3563	ITLB_STUFF(%g5, %g1, %g2, %g3, %g4)
3564	! trapstat wants TTE in %g5
3565	retry
3566
3567tsb_ism:
3568	/*
3569	 * This is an ISM [i|d]tlb miss.  We optimize for largest
3570	 * page size down to smallest.
3571	 *
3572	 * g2 = vaddr + ctx(or ctxtype (sun4v)) aka (pseudo-)tag access
3573	 *	register
3574	 * g3 = ismmap->ism_seg
3575	 * g4 = physical address of ismmap->ism_sfmmu
3576	 * g6 = tsbmiss area
3577	 */
3578	ldna	[%g4]ASI_MEM, %g7		/* g7 = ism hatid */
3579	brz,a,pn %g7, ptl1_panic		/* if zero jmp ahead */
3580	  mov	PTL1_BAD_ISM, %g1
3581						/* g5 = pa of imap_vb_shift */
3582	sub	%g4, (IMAP_ISMHAT - IMAP_VB_SHIFT), %g5
3583	lduba	[%g5]ASI_MEM, %g4		/* g4 = imap_vb_shift */
3584	srlx	%g3, %g4, %g3			/* clr size field */
3585	set	TAGACC_CTX_MASK, %g1		/* mask off ctx number */
3586	sllx    %g3, %g4, %g3                   /* g3 = ism vbase */
3587	and     %g2, %g1, %g4                   /* g4 = ctx number */
3588	andn    %g2, %g1, %g1                   /* g1 = tlb miss vaddr */
3589	sub     %g1, %g3, %g2                   /* g2 = offset in ISM seg */
3590	or      %g2, %g4, %g2                   /* g2 = (pseudo-)tagacc */
3591	sub     %g5, (IMAP_VB_SHIFT - IMAP_HATFLAGS), %g5
3592	lduha   [%g5]ASI_MEM, %g4               /* g5 = pa of imap_hatflags */
3593#if defined(sun4v) || defined(UTSB_PHYS)
3594	and     %g4, HAT_CTX1_FLAG, %g5         /* g5 = imap_hatflags */
3595	brz,pt %g5, tsb_chk4M_ism
3596	  nop
3597	ldub    [%g6 + TSBMISS_URTTEFLAGS], %g5
3598	or      %g5, HAT_CHKCTX1_FLAG, %g5
3599	stub    %g5, [%g6 + TSBMISS_URTTEFLAGS]
3600	rdpr    %tt, %g5
3601	SAVE_CTX1(%g5, %g3, %g1, tsb_shctxl)
3602#endif /* defined(sun4v) || defined(UTSB_PHYS) */
3603
3604	/*
3605	 * ISM pages are always locked down.
3606	 * If we can't find the tte then pagefault
3607	 * and let the spt segment driver resolve it.
3608	 *
3609	 * g2 = tagacc w/ISM vaddr (offset in ISM seg)
3610	 * g4 = imap_hatflags
3611	 * g6 = tsb miss area
3612	 * g7 = ISM hatid
3613	 */
3614
3615tsb_chk4M_ism:
3616	and	%g4, HAT_4M_FLAG, %g5		/* g4 = imap_hatflags */
3617	brnz,pt	%g5, tsb_ism_4M			/* branch if 4M pages */
3618	  nop
3619
3620tsb_ism_32M:
3621	and	%g4, HAT_32M_FLAG, %g5		/* check default 32M next */
3622	brz,pn	%g5, tsb_ism_256M
3623	  nop
3624
3625	/*
3626	 * 32M hash.
3627	 */
3628
3629	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1, MMU_PAGESHIFT32M,
3630	    TTE32M, %g5, tsb_ism_l32M, tsb_ism_32M_found, sfmmu_suspend_tl,
3631	    tsb_ism_4M)
3632	/* NOT REACHED */
3633
3634tsb_ism_32M_found:
3635	brlz,a,pt %g3, tsb_validtte
3636	  rdpr	%tt, %g7
3637	ba,pt	%xcc, tsb_ism_4M
3638	  nop
3639
3640tsb_ism_256M:
3641	and	%g4, HAT_256M_FLAG, %g5		/* 256M is last resort */
3642	brz,a,pn %g5, ptl1_panic
3643	  mov	PTL1_BAD_ISM, %g1
3644
3645	/*
3646	 * 256M hash.
3647	 */
3648	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1, MMU_PAGESHIFT256M,
3649	    TTE256M, %g5, tsb_ism_l256M, tsb_ism_256M_found, sfmmu_suspend_tl,
3650	    tsb_ism_4M)
3651
3652tsb_ism_256M_found:
3653	brlz,a,pt %g3, tsb_validtte
3654	  rdpr	%tt, %g7
3655
3656tsb_ism_4M:
3657	/*
3658	 * 4M hash.
3659	 */
3660	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1, MMU_PAGESHIFT4M,
3661	    TTE4M, %g5, tsb_ism_l4M, tsb_ism_4M_found, sfmmu_suspend_tl,
3662	    tsb_ism_8K)
3663	/* NOT REACHED */
3664
3665tsb_ism_4M_found:
3666	brlz,a,pt %g3, tsb_validtte
3667	  rdpr	%tt, %g7
3668
3669tsb_ism_8K:
3670	/*
3671	 * 8K and 64K hash.
3672	 */
3673
3674	GET_TTE(%g2, %g7, %g3, %g4, %g6, %g1, MMU_PAGESHIFT64K,
3675	    TTE64K, %g5, tsb_ism_l8K, tsb_ism_8K_found, sfmmu_suspend_tl,
3676	    tsb_pagefault)
3677	/* NOT REACHED */
3678
3679tsb_ism_8K_found:
3680	brlz,a,pt %g3, tsb_validtte
3681	  rdpr	%tt, %g7
3682
3683tsb_pagefault:
3684	rdpr	%tt, %g7
3685	cmp	%g7, FAST_PROT_TT
3686	be,a,pn	%icc, tsb_protfault
3687	  wrpr	%g0, FAST_DMMU_MISS_TT, %tt
3688
3689tsb_protfault:
3690	/*
3691	 * we get here if we couldn't find a valid tte in the hash.
3692	 *
3693	 * If user and we are at tl>1 we go to window handling code.
3694	 *
3695	 * If kernel and the fault is on the same page as our stack
3696	 * pointer, then we know the stack is bad and the trap handler
3697	 * will fail, so we call ptl1_panic with PTL1_BAD_STACK.
3698	 *
3699	 * If this is a kernel trap and tl>1, panic.
3700	 *
3701	 * Otherwise we call pagefault.
3702	 */
3703	cmp	%g7, FAST_IMMU_MISS_TT
3704#ifdef sun4v
3705	MMU_FAULT_STATUS_AREA(%g4)
3706	ldx	[%g4 + MMFSA_I_CTX], %g5
3707	ldx	[%g4 + MMFSA_D_CTX], %g4
3708	move	%icc, %g5, %g4
3709	cmp	%g7, T_INSTR_MMU_MISS
3710	move	%icc, %g5, %g4
3711#else
3712	mov	MMU_TAG_ACCESS, %g4
3713	ldxa	[%g4]ASI_DMMU, %g2
3714	ldxa	[%g4]ASI_IMMU, %g5
3715	move	%icc, %g5, %g2
3716	cmp	%g7, T_INSTR_MMU_MISS
3717	move	%icc, %g5, %g2
3718	sllx	%g2, TAGACC_CTX_LSHIFT, %g4
3719#endif /* sun4v */
3720	brnz,pn	%g4, 3f				/* skip if not kernel */
3721	  rdpr	%tl, %g5
3722
3723	add	%sp, STACK_BIAS, %g3
3724	srlx	%g3, MMU_PAGESHIFT, %g3
3725	srlx	%g2, MMU_PAGESHIFT, %g4
3726	cmp	%g3, %g4
3727	be,a,pn	%icc, ptl1_panic		/* panic if bad %sp */
3728	  mov	PTL1_BAD_STACK, %g1
3729
3730	cmp	%g5, 1
3731	ble,pt	%icc, 2f
3732	  nop
3733	TSTAT_CHECK_TL1(2f, %g1, %g2)
3734	rdpr	%tt, %g2
3735	cmp	%g2, FAST_PROT_TT
3736	mov	PTL1_BAD_KPROT_FAULT, %g1
3737	movne	%icc, PTL1_BAD_KMISS, %g1
3738	ba,pt	%icc, ptl1_panic
3739	  nop
3740
37412:
3742	/*
3743	 * We are taking a pagefault in the kernel on a kernel address.  If
3744	 * CPU_DTRACE_NOFAULT is set in the cpuc_dtrace_flags, we don't actually
3745	 * want to call sfmmu_pagefault -- we will instead note that a fault
3746	 * has occurred by setting CPU_DTRACE_BADADDR and issue a "done"
3747	 * (instead of a "retry").  This will step over the faulting
3748	 * instruction.
3749	 */
3750	CPU_INDEX(%g1, %g2)
3751	set	cpu_core, %g2
3752	sllx	%g1, CPU_CORE_SHIFT, %g1
3753	add	%g1, %g2, %g1
3754	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
3755	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
3756	bz	sfmmu_pagefault
3757	or	%g2, CPU_DTRACE_BADADDR, %g2
3758	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
3759	GET_MMU_D_ADDR(%g3, %g4)
3760	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
3761	done
3762
37633:
3764	cmp	%g5, 1
3765	ble,pt	%icc, 4f
3766	  nop
3767	TSTAT_CHECK_TL1(4f, %g1, %g2)
3768	ba,pt	%icc, sfmmu_window_trap
3769	  nop
3770
37714:
3772	/*
3773	 * We are taking a pagefault on a non-kernel address.  If we are in
3774	 * the kernel (e.g., due to a copyin()), we will check cpuc_dtrace_flags
3775	 * and (if CPU_DTRACE_NOFAULT is set) will proceed as outlined above.
3776	 */
3777	CPU_INDEX(%g1, %g2)
3778	set	cpu_core, %g2
3779	sllx	%g1, CPU_CORE_SHIFT, %g1
3780	add	%g1, %g2, %g1
3781	lduh	[%g1 + CPUC_DTRACE_FLAGS], %g2
3782	andcc	%g2, CPU_DTRACE_NOFAULT, %g0
3783	bz	sfmmu_mmu_trap
3784	or	%g2, CPU_DTRACE_BADADDR, %g2
3785	stuh	%g2, [%g1 + CPUC_DTRACE_FLAGS]
3786	GET_MMU_D_ADDR(%g3, %g4)
3787	stx	%g3, [%g1 + CPUC_DTRACE_ILLVAL]
3788
3789	/*
3790	 * Be sure that we're actually taking this miss from the kernel --
3791	 * otherwise we have managed to return to user-level with
3792	 * CPU_DTRACE_NOFAULT set in cpuc_dtrace_flags.
3793	 */
3794	rdpr	%tstate, %g2
3795	btst	TSTATE_PRIV, %g2
3796	bz,a	ptl1_panic
3797	  mov	PTL1_BAD_DTRACE_FLAGS, %g1
3798	done
3799
3800	ALTENTRY(tsb_tl0_noctxt)
3801	/*
3802	 * If we have no context, check to see if CPU_DTRACE_NOFAULT is set;
3803	 * if it is, indicated that we have faulted and issue a done.
3804	 */
3805	CPU_INDEX(%g5, %g6)
3806	set	cpu_core, %g6
3807	sllx	%g5, CPU_CORE_SHIFT, %g5
3808	add	%g5, %g6, %g5
3809	lduh	[%g5 + CPUC_DTRACE_FLAGS], %g6
3810	andcc	%g6, CPU_DTRACE_NOFAULT, %g0
3811	bz	1f
3812	or	%g6, CPU_DTRACE_BADADDR, %g6
3813	stuh	%g6, [%g5 + CPUC_DTRACE_FLAGS]
3814	GET_MMU_D_ADDR(%g3, %g4)
3815	stx	%g3, [%g5 + CPUC_DTRACE_ILLVAL]
3816
3817	/*
3818	 * Be sure that we're actually taking this miss from the kernel --
3819	 * otherwise we have managed to return to user-level with
3820	 * CPU_DTRACE_NOFAULT set in cpuc_dtrace_flags.
3821	 */
3822	rdpr	%tstate, %g5
3823	btst	TSTATE_PRIV, %g5
3824	bz,a	ptl1_panic
3825	  mov	PTL1_BAD_DTRACE_FLAGS, %g1
3826	TSTAT_CHECK_TL1(2f, %g1, %g2);
38272:
3828	done
3829
38301:
3831	rdpr	%tt, %g5
3832	cmp	%g5, FAST_IMMU_MISS_TT
3833#ifdef sun4v
3834	MMU_FAULT_STATUS_AREA(%g2)
3835	be,a,pt	%icc, 2f
3836	  ldx	[%g2 + MMFSA_I_CTX], %g3
3837	cmp	%g5, T_INSTR_MMU_MISS
3838	be,a,pt	%icc, 2f
3839	  ldx	[%g2 + MMFSA_I_CTX], %g3
3840	ldx	[%g2 + MMFSA_D_CTX], %g3
38412:
3842#else
3843	mov	MMU_TAG_ACCESS, %g2
3844	be,a,pt	%icc, 2f
3845	  ldxa	[%g2]ASI_IMMU, %g3
3846	ldxa	[%g2]ASI_DMMU, %g3
38472:	sllx	%g3, TAGACC_CTX_LSHIFT, %g3
3848#endif /* sun4v */
3849	brz,a,pn %g3, ptl1_panic		! panic if called for kernel
3850	  mov	PTL1_BAD_CTX_STEAL, %g1		! since kernel ctx was stolen
3851	rdpr	%tl, %g5
3852	cmp	%g5, 1
3853	ble,pt	%icc, sfmmu_mmu_trap
3854	  nop
3855	TSTAT_CHECK_TL1(sfmmu_mmu_trap, %g1, %g2)
3856	ba,pt	%icc, sfmmu_window_trap
3857	  nop
3858	SET_SIZE(sfmmu_tsb_miss)
3859#endif  /* lint */
3860
3861#if defined (lint)
3862/*
3863 * This routine will look for a user or kernel vaddr in the hash
3864 * structure.  It returns a valid pfn or PFN_INVALID.  It doesn't
3865 * grab any locks.  It should only be used by other sfmmu routines.
3866 */
3867/* ARGSUSED */
3868pfn_t
3869sfmmu_vatopfn(caddr_t vaddr, sfmmu_t *sfmmup, tte_t *ttep)
3870{
3871	return(0);
3872}
3873
3874/* ARGSUSED */
3875pfn_t
3876sfmmu_kvaszc2pfn(caddr_t vaddr, int hashno)
3877{
3878	return(0);
3879}
3880
3881#else /* lint */
3882
3883	ENTRY_NP(sfmmu_vatopfn)
3884 	/*
3885 	 * disable interrupts
3886 	 */
3887 	rdpr	%pstate, %o3
3888#ifdef DEBUG
3889	PANIC_IF_INTR_DISABLED_PSTR(%o3, sfmmu_di_l5, %g1)
3890#endif
3891	/*
3892	 * disable interrupts to protect the TSBMISS area
3893	 */
3894	andn    %o3, PSTATE_IE, %o5
3895	wrpr    %o5, 0, %pstate
3896
3897	/*
3898	 * o0 = vaddr
3899	 * o1 = sfmmup
3900	 * o2 = ttep
3901	 */
3902	CPU_TSBMISS_AREA(%g1, %o5)
3903	ldn	[%g1 + TSBMISS_KHATID], %o4
3904	cmp	%o4, %o1
3905	bne,pn	%ncc, vatopfn_nokernel
3906	  mov	TTE64K, %g5			/* g5 = rehash # */
3907	mov %g1,%o5				/* o5 = tsbmiss_area */
3908	/*
3909	 * o0 = vaddr
3910	 * o1 & o4 = hatid
3911	 * o2 = ttep
3912	 * o5 = tsbmiss area
3913	 */
3914	mov	HBLK_RANGE_SHIFT, %g6
39151:
3916
3917	/*
3918	 * o0 = vaddr
3919	 * o1 = sfmmup
3920	 * o2 = ttep
3921	 * o3 = old %pstate
3922	 * o4 = hatid
3923	 * o5 = tsbmiss
3924	 * g5 = rehash #
3925	 * g6 = hmeshift
3926	 *
3927	 * The first arg to GET_TTE is actually tagaccess register
3928	 * not just vaddr. Since this call is for kernel we need to clear
3929	 * any lower vaddr bits that would be interpreted as ctx bits.
3930	 */
3931	set     TAGACC_CTX_MASK, %g1
3932	andn    %o0, %g1, %o0
3933	GET_TTE(%o0, %o4, %g1, %g2, %o5, %g4, %g6, %g5, %g3,
3934		vatopfn_l1, kvtop_hblk_found, tsb_suspend, kvtop_nohblk)
3935
3936kvtop_hblk_found:
3937	/*
3938	 * o0 = vaddr
3939	 * o1 = sfmmup
3940	 * o2 = ttep
3941	 * g1 = tte
3942	 * g2 = tte pa
3943	 * g3 = scratch
3944	 * o2 = tsbmiss area
3945	 * o1 = hat id
3946	 */
3947	brgez,a,pn %g1, 6f			/* if tte invalid goto tl0 */
3948	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
3949	stx %g1,[%o2]				/* put tte into *ttep */
3950	TTETOPFN(%g1, %o0, vatopfn_l2, %g2, %g3, %g4)
3951	/*
3952	 * o0 = vaddr
3953	 * o1 = sfmmup
3954	 * o2 = ttep
3955	 * g1 = pfn
3956	 */
3957	ba,pt	%xcc, 6f
3958	  mov	%g1, %o0
3959
3960kvtop_nohblk:
3961	/*
3962	 * we get here if we couldn't find valid hblk in hash.  We rehash
3963	 * if neccesary.
3964	 */
3965	ldn	[%o5 + (TSBMISS_SCRATCH + TSB_TAGACC)], %o0
3966#ifdef sun4v
3967	cmp	%g5, MAX_HASHCNT
3968#else
3969	cmp	%g5, DEFAULT_MAX_HASHCNT	/* no 32/256M kernel pages */
3970#endif /* sun4v */
3971	be,a,pn	%icc, 6f
3972	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
3973	mov	%o1, %o4			/* restore hatid */
3974#ifdef sun4v
3975        add	%g5, 2, %g5
3976	cmp	%g5, 3
3977	move	%icc, MMU_PAGESHIFT4M, %g6
3978	ba,pt	%icc, 1b
3979	movne	%icc, MMU_PAGESHIFT256M, %g6
3980#else
3981        inc	%g5
3982	cmp	%g5, 2
3983	move	%icc, MMU_PAGESHIFT512K, %g6
3984	ba,pt	%icc, 1b
3985	movne	%icc, MMU_PAGESHIFT4M, %g6
3986#endif /* sun4v */
39876:
3988	retl
3989 	  wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
3990
3991tsb_suspend:
3992	/*
3993	 * o0 = vaddr
3994	 * o1 = sfmmup
3995	 * o2 = ttep
3996	 * g1 = tte
3997	 * g2 = tte pa
3998	 * g3 = tte va
3999	 * o2 = tsbmiss area  use o5 instead of o2 for tsbmiss
4000	 */
4001	stx %g1,[%o2]				/* put tte into *ttep */
4002	brgez,a,pn %g1, 8f			/* if tte invalid goto 8: */
4003	  sub	%g0, 1, %o0			/* output = PFN_INVALID */
4004	sub	%g0, 2, %o0			/* output = PFN_SUSPENDED */
40058:
4006	retl
4007	 wrpr	%g0, %o3, %pstate		/* enable interrupts */
4008
4009vatopfn_nokernel:
4010	/*
4011	 * This routine does NOT support user addresses
4012	 * There is a routine in C that supports this.
4013	 * The only reason why we don't have the C routine
4014	 * support kernel addresses as well is because
4015	 * we do va_to_pa while holding the hashlock.
4016	 */
4017 	wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
4018	save	%sp, -SA(MINFRAME), %sp
4019	sethi	%hi(sfmmu_panic3), %o0
4020	call	panic
4021	 or	%o0, %lo(sfmmu_panic3), %o0
4022
4023	SET_SIZE(sfmmu_vatopfn)
4024
4025	/*
4026	 * %o0 = vaddr
4027	 * %o1 = hashno (aka szc)
4028	 *
4029	 *
4030	 * This routine is similar to sfmmu_vatopfn() but will only look for
4031	 * a kernel vaddr in the hash structure for the specified rehash value.
4032	 * It's just an optimization for the case when pagesize for a given
4033	 * va range is already known (e.g. large page heap) and we don't want
4034	 * to start the search with rehash value 1 as sfmmu_vatopfn() does.
4035	 *
4036	 * Returns valid pfn or PFN_INVALID if
4037	 * tte for specified rehash # is not found, invalid or suspended.
4038	 */
4039	ENTRY_NP(sfmmu_kvaszc2pfn)
4040 	/*
4041 	 * disable interrupts
4042 	 */
4043 	rdpr	%pstate, %o3
4044#ifdef DEBUG
4045	PANIC_IF_INTR_DISABLED_PSTR(%o3, sfmmu_di_l6, %g1)
4046#endif
4047	/*
4048	 * disable interrupts to protect the TSBMISS area
4049	 */
4050	andn    %o3, PSTATE_IE, %o5
4051	wrpr    %o5, 0, %pstate
4052
4053	CPU_TSBMISS_AREA(%g1, %o5)
4054	ldn	[%g1 + TSBMISS_KHATID], %o4
4055	sll	%o1, 1, %g6
4056	add	%g6, %o1, %g6
4057	add	%g6, MMU_PAGESHIFT, %g6
4058	/*
4059	 * %o0 = vaddr
4060	 * %o1 = hashno
4061	 * %o3 = old %pstate
4062	 * %o4 = ksfmmup
4063	 * %g1 = tsbmiss area
4064	 * %g6 = hmeshift
4065	 */
4066
4067	/*
4068	 * The first arg to GET_TTE is actually tagaccess register
4069	 * not just vaddr. Since this call is for kernel we need to clear
4070	 * any lower vaddr bits that would be interpreted as ctx bits.
4071	 */
4072	srlx	%o0, MMU_PAGESHIFT, %o0
4073	sllx	%o0, MMU_PAGESHIFT, %o0
4074	GET_TTE(%o0, %o4, %g3, %g4, %g1, %o5, %g6, %o1, %g5,
4075		kvaszc2pfn_l1, kvaszc2pfn_hblk_found, kvaszc2pfn_nohblk,
4076		kvaszc2pfn_nohblk)
4077
4078kvaszc2pfn_hblk_found:
4079	/*
4080	 * %g3 = tte
4081	 * %o0 = vaddr
4082	 */
4083	brgez,a,pn %g3, 1f			/* check if tte is invalid */
4084	  mov	-1, %o0				/* output = -1 (PFN_INVALID) */
4085	TTETOPFN(%g3, %o0, kvaszc2pfn_l2, %g2, %g4, %g5)
4086	/*
4087	 * g3 = pfn
4088	 */
4089	ba,pt	%xcc, 1f
4090	  mov	%g3, %o0
4091
4092kvaszc2pfn_nohblk:
4093	mov	-1, %o0
4094
40951:
4096	retl
4097 	  wrpr	%g0, %o3, %pstate		/* re-enable interrupts */
4098
4099	SET_SIZE(sfmmu_kvaszc2pfn)
4100
4101#endif /* lint */
4102
4103
4104
4105#if !defined(lint)
4106
4107/*
4108 * kpm lock used between trap level tsbmiss handler and kpm C level.
4109 */
4110#define KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi)			\
4111	mov     0xff, tmp1						;\
4112label1:									;\
4113	casa    [kpmlckp]asi, %g0, tmp1					;\
4114	brnz,pn tmp1, label1						;\
4115	mov     0xff, tmp1						;\
4116	membar  #LoadLoad
4117
4118#define KPMLOCK_EXIT(kpmlckp, asi)					\
4119	membar  #LoadStore|#StoreStore					;\
4120	sta     %g0, [kpmlckp]asi
4121
4122/*
4123 * Lookup a memseg for a given pfn and if found, return the physical
4124 * address of the corresponding struct memseg in mseg, otherwise
4125 * return MSEG_NULLPTR_PA. The kpmtsbm pointer must be provided in
4126 * tsbmp, %asi is assumed to be ASI_MEM.
4127 * This lookup is done by strictly traversing only the physical memseg
4128 * linkage. The more generic approach, to check the virtual linkage
4129 * before using the physical (used e.g. with hmehash buckets), cannot
4130 * be used here. Memory DR operations can run in parallel to this
4131 * lookup w/o any locks and updates of the physical and virtual linkage
4132 * cannot be done atomically wrt. to each other. Because physical
4133 * address zero can be valid physical address, MSEG_NULLPTR_PA acts
4134 * as "physical NULL" pointer.
4135 */
4136#define	PAGE_NUM2MEMSEG_NOLOCK_PA(pfn, mseg, tsbmp, tmp1, tmp2, tmp3, label) \
4137	sethi	%hi(mhash_per_slot), tmp3 /* no tsbmp use due to DR */	;\
4138	ldx	[tmp3 + %lo(mhash_per_slot)], mseg			;\
4139	udivx	pfn, mseg, mseg						;\
4140	ldx	[tsbmp + KPMTSBM_MSEGPHASHPA], tmp1			;\
4141	and	mseg, SFMMU_N_MEM_SLOTS - 1, mseg			;\
4142	sllx	mseg, SFMMU_MEM_HASH_ENTRY_SHIFT, mseg			;\
4143	add	tmp1, mseg, tmp1					;\
4144	ldxa	[tmp1]%asi, mseg					;\
4145	cmp	mseg, MSEG_NULLPTR_PA					;\
4146	be,pn	%xcc, label/**/1		/* if not found */	;\
4147	  nop								;\
4148	ldxa	[mseg + MEMSEG_PAGES_BASE]%asi, tmp1			;\
4149	cmp	pfn, tmp1			/* pfn - pages_base */	;\
4150	blu,pn	%xcc, label/**/1					;\
4151	  ldxa	[mseg + MEMSEG_PAGES_END]%asi, tmp2			;\
4152	cmp	pfn, tmp2			/* pfn - pages_end */	;\
4153	bgeu,pn	%xcc, label/**/1					;\
4154	  sub	pfn, tmp1, tmp1			/* pfn - pages_base */	;\
4155	mulx	tmp1, PAGE_SIZE, tmp1					;\
4156	ldxa	[mseg + MEMSEG_PAGESPA]%asi, tmp2	/* pages */	;\
4157	add	tmp2, tmp1, tmp1			/* pp */	;\
4158	lduwa	[tmp1 + PAGE_PAGENUM]%asi, tmp2				;\
4159	cmp	tmp2, pfn						;\
4160	be,pt	%xcc, label/**/_ok			/* found */	;\
4161label/**/1:								;\
4162	/* brute force lookup */					;\
4163	sethi	%hi(memsegspa), tmp3 /* no tsbmp use due to DR */	;\
4164	ldx	[tmp3 + %lo(memsegspa)], mseg				;\
4165label/**/2:								;\
4166	cmp	mseg, MSEG_NULLPTR_PA					;\
4167	be,pn	%xcc, label/**/_ok		/* if not found */	;\
4168	  nop								;\
4169	ldxa	[mseg + MEMSEG_PAGES_BASE]%asi, tmp1			;\
4170	cmp	pfn, tmp1			/* pfn - pages_base */	;\
4171	blu,a,pt %xcc, label/**/2					;\
4172	  ldxa	[mseg + MEMSEG_NEXTPA]%asi, mseg			;\
4173	ldxa	[mseg + MEMSEG_PAGES_END]%asi, tmp2			;\
4174	cmp	pfn, tmp2			/* pfn - pages_end */	;\
4175	bgeu,a,pt %xcc, label/**/2					;\
4176	  ldxa	[mseg + MEMSEG_NEXTPA]%asi, mseg			;\
4177label/**/_ok:
4178
4179	/*
4180	 * kpm tsb miss handler large pages
4181	 * g1 = 8K kpm TSB entry pointer
4182	 * g2 = tag access register
4183	 * g3 = 4M kpm TSB entry pointer
4184	 */
4185	ALTENTRY(sfmmu_kpm_dtsb_miss)
4186	TT_TRACE(trace_tsbmiss)
4187
4188	CPU_INDEX(%g7, %g6)
4189	sethi	%hi(kpmtsbm_area), %g6
4190	sllx	%g7, KPMTSBM_SHIFT, %g7
4191	or	%g6, %lo(kpmtsbm_area), %g6
4192	add	%g6, %g7, %g6			/* g6 = kpmtsbm ptr */
4193
4194	/* check enable flag */
4195	ldub	[%g6 + KPMTSBM_FLAGS], %g4
4196	and	%g4, KPMTSBM_ENABLE_FLAG, %g5
4197	brz,pn	%g5, sfmmu_tsb_miss		/* if kpm not enabled */
4198	  nop
4199
4200	/* VA range check */
4201	ldx	[%g6 + KPMTSBM_VBASE], %g7
4202	cmp	%g2, %g7
4203	blu,pn	%xcc, sfmmu_tsb_miss
4204	  ldx	[%g6 + KPMTSBM_VEND], %g5
4205	cmp	%g2, %g5
4206	bgeu,pn	%xcc, sfmmu_tsb_miss
4207	  stx	%g3, [%g6 + KPMTSBM_TSBPTR]
4208
4209	/*
4210	 * check TL tsbmiss handling flag
4211	 * bump tsbmiss counter
4212	 */
4213	lduw	[%g6 + KPMTSBM_TSBMISS], %g5
4214#ifdef	DEBUG
4215	and	%g4, KPMTSBM_TLTSBM_FLAG, %g3
4216	inc	%g5
4217	brz,pn	%g3, sfmmu_kpm_exception
4218	  st	%g5, [%g6 + KPMTSBM_TSBMISS]
4219#else
4220	inc	%g5
4221	st	%g5, [%g6 + KPMTSBM_TSBMISS]
4222#endif
4223	/*
4224	 * At this point:
4225	 *  g1 = 8K kpm TSB pointer (not used)
4226	 *  g2 = tag access register
4227	 *  g3 = clobbered
4228	 *  g6 = per-CPU kpm tsbmiss area
4229	 *  g7 = kpm_vbase
4230	 */
4231
4232	/* vaddr2pfn */
4233	ldub	[%g6 + KPMTSBM_SZSHIFT], %g3
4234	sub	%g2, %g7, %g4			/* paddr = vaddr-kpm_vbase */
4235	srax    %g4, %g3, %g2			/* which alias range (r) */
4236	brnz,pn	%g2, sfmmu_kpm_exception	/* if (r != 0) goto C handler */
4237	  srlx	%g4, MMU_PAGESHIFT, %g2		/* %g2 = pfn */
4238
4239	/*
4240	 * Setup %asi
4241	 * mseg_pa = page_numtomemseg_nolock(pfn)
4242	 * if (mseg_pa == NULL) sfmmu_kpm_exception
4243	 * g2=pfn
4244	 */
4245	mov	ASI_MEM, %asi
4246	PAGE_NUM2MEMSEG_NOLOCK_PA(%g2, %g3, %g6, %g4, %g5, %g7, kpmtsbmp2m)
4247	cmp	%g3, MSEG_NULLPTR_PA
4248	be,pn	%xcc, sfmmu_kpm_exception	/* if mseg not found */
4249	  nop
4250
4251	/*
4252	 * inx = ptokpmp((kpmptop((ptopkpmp(pfn))) - mseg_pa->kpm_pbase));
4253	 * g2=pfn g3=mseg_pa
4254	 */
4255	ldub	[%g6 + KPMTSBM_KPMP2PSHFT], %g5
4256	ldxa	[%g3 + MEMSEG_KPM_PBASE]%asi, %g7
4257	srlx	%g2, %g5, %g4
4258	sllx	%g4, %g5, %g4
4259	sub	%g4, %g7, %g4
4260	srlx	%g4, %g5, %g4
4261
4262	/*
4263	 * Validate inx value
4264	 * g2=pfn g3=mseg_pa g4=inx
4265	 */
4266#ifdef	DEBUG
4267	ldxa	[%g3 + MEMSEG_KPM_NKPMPGS]%asi, %g5
4268	cmp	%g4, %g5			/* inx - nkpmpgs */
4269	bgeu,pn	%xcc, sfmmu_kpm_exception	/* if out of range */
4270	  ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4271#else
4272	ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4273#endif
4274	/*
4275	 * kp = &mseg_pa->kpm_pages[inx]
4276	 */
4277	sllx	%g4, KPMPAGE_SHIFT, %g4		/* kpm_pages offset */
4278	ldxa	[%g3 + MEMSEG_KPM_PAGES]%asi, %g5 /* kpm_pages */
4279	add	%g5, %g4, %g5			/* kp */
4280
4281	/*
4282	 * KPMP_HASH(kp)
4283	 * g2=pfn g3=mseg_pa g4=offset g5=kp g7=kpmp_table_sz
4284	 */
4285	ldub	[%g6 + KPMTSBM_KPMPSHIFT], %g1	/* kpmp_shift */
4286	sub	%g7, 1, %g7			/* mask */
4287	srlx	%g5, %g1, %g1			/* x = ksp >> kpmp_shift */
4288	add	%g5, %g1, %g5			/* y = ksp + x */
4289	and 	%g5, %g7, %g5			/* hashinx = y & mask */
4290
4291	/*
4292	 * Calculate physical kpm_page pointer
4293	 * g2=pfn g3=mseg_pa g4=offset g5=hashinx
4294	 */
4295	ldxa	[%g3 + MEMSEG_KPM_PAGESPA]%asi, %g1 /* kpm_pagespa */
4296	add	%g1, %g4, %g1			/* kp_pa */
4297
4298	/*
4299	 * Calculate physical hash lock address
4300	 * g1=kp_refcntc_pa g2=pfn g5=hashinx
4301	 */
4302	ldx	[%g6 + KPMTSBM_KPMPTABLEPA], %g4 /* kpmp_tablepa */
4303	sllx	%g5, KPMHLK_SHIFT, %g5
4304	add	%g4, %g5, %g3
4305	add	%g3, KPMHLK_LOCK, %g3		/* hlck_pa */
4306
4307	/*
4308	 * Assemble tte
4309	 * g1=kp_pa g2=pfn g3=hlck_pa
4310	 */
4311#ifdef sun4v
4312	sethi	%hi(TTE_VALID_INT), %g5		/* upper part */
4313	sllx	%g5, 32, %g5
4314	mov	(TTE_CP_INT|TTE_CV_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4315	or	%g4, TTE4M, %g4
4316	or	%g5, %g4, %g5
4317#else
4318	sethi	%hi(TTE_VALID_INT), %g4
4319	mov	TTE4M, %g5
4320	sllx	%g5, TTE_SZ_SHFT_INT, %g5
4321	or	%g5, %g4, %g5			/* upper part */
4322	sllx	%g5, 32, %g5
4323	mov	(TTE_CP_INT|TTE_CV_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4324	or	%g5, %g4, %g5
4325#endif
4326	sllx	%g2, MMU_PAGESHIFT, %g4
4327	or	%g5, %g4, %g5			/* tte */
4328	ldx	[%g6 + KPMTSBM_TSBPTR], %g4
4329	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
4330
4331	/*
4332	 * tsb dropin
4333	 * g1=kp_pa g2=ttarget g3=hlck_pa g4=kpmtsbp4m g5=tte g6=kpmtsbm_area
4334	 */
4335
4336	/* KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi) */
4337	KPMLOCK_ENTER(%g3, %g7, kpmtsbmhdlr1, ASI_MEM)
4338
4339	/* use C-handler if there's no go for dropin */
4340	ldsha	[%g1 + KPMPAGE_REFCNTC]%asi, %g7 /* kp_refcntc */
4341	cmp	%g7, -1
4342	bne,pn	%xcc, 5f	/* use C-handler if there's no go for dropin */
4343	  nop
4344
4345#ifdef	DEBUG
4346	/* double check refcnt */
4347	ldsha	[%g1 + KPMPAGE_REFCNT]%asi, %g7
4348	brz,pn	%g7, 5f			/* let C-handler deal with this */
4349	  nop
4350#endif
4351
4352#ifndef sun4v
4353	ldub	[%g6 + KPMTSBM_FLAGS], %g7
4354	mov	ASI_N, %g1
4355	andcc	%g7, KPMTSBM_TSBPHYS_FLAG, %g0
4356	movnz	%icc, ASI_MEM, %g1
4357	mov	%g1, %asi
4358#endif
4359
4360	/*
4361	 * TSB_LOCK_ENTRY(tsbp, tmp1, tmp2, label) (needs %asi set)
4362	 * If we fail to lock the TSB entry then just load the tte into the
4363	 * TLB.
4364	 */
4365	TSB_LOCK_ENTRY(%g4, %g1, %g7, locked_tsb_l1)
4366
4367	/* TSB_INSERT_UNLOCK_ENTRY(tsbp, tte, tagtarget, tmp) */
4368	TSB_INSERT_UNLOCK_ENTRY(%g4, %g5, %g2, %g7)
4369locked_tsb_l1:
4370	DTLB_STUFF(%g5, %g1, %g2, %g4, %g6)
4371
4372	/* KPMLOCK_EXIT(kpmlckp, asi) */
4373	KPMLOCK_EXIT(%g3, ASI_MEM)
4374
4375	/*
4376	 * If trapstat is running, we need to shift the %tpc and %tnpc to
4377	 * point to trapstat's TSB miss return code (note that trapstat
4378	 * itself will patch the correct offset to add).
4379	 * Note: TTE is expected in %g5 (allows per pagesize reporting).
4380	 */
4381	rdpr	%tl, %g7
4382	cmp	%g7, 1
4383	ble	%icc, 0f
4384	sethi	%hi(KERNELBASE), %g6
4385	rdpr	%tpc, %g7
4386	or	%g6, %lo(KERNELBASE), %g6
4387	cmp	%g7, %g6
4388	bgeu	%xcc, 0f
4389	ALTENTRY(tsbmiss_trapstat_patch_point_kpm)
4390	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
4391	wrpr	%g7, %tpc
4392	add	%g7, 4, %g7
4393	wrpr	%g7, %tnpc
43940:
4395	retry
43965:
4397	/* g3=hlck_pa */
4398	KPMLOCK_EXIT(%g3, ASI_MEM)
4399	ba,pt	%icc, sfmmu_kpm_exception
4400	  nop
4401	SET_SIZE(sfmmu_kpm_dtsb_miss)
4402
4403	/*
4404	 * kpm tsbmiss handler for smallpages
4405	 * g1 = 8K kpm TSB pointer
4406	 * g2 = tag access register
4407	 * g3 = 4M kpm TSB pointer
4408	 */
4409	ALTENTRY(sfmmu_kpm_dtsb_miss_small)
4410	TT_TRACE(trace_tsbmiss)
4411	CPU_INDEX(%g7, %g6)
4412	sethi	%hi(kpmtsbm_area), %g6
4413	sllx	%g7, KPMTSBM_SHIFT, %g7
4414	or	%g6, %lo(kpmtsbm_area), %g6
4415	add	%g6, %g7, %g6			/* g6 = kpmtsbm ptr */
4416
4417	/* check enable flag */
4418	ldub	[%g6 + KPMTSBM_FLAGS], %g4
4419	and	%g4, KPMTSBM_ENABLE_FLAG, %g5
4420	brz,pn	%g5, sfmmu_tsb_miss		/* if kpm not enabled */
4421	  nop
4422
4423	/*
4424	 * VA range check
4425	 * On fail: goto sfmmu_tsb_miss
4426	 */
4427	ldx	[%g6 + KPMTSBM_VBASE], %g7
4428	cmp	%g2, %g7
4429	blu,pn	%xcc, sfmmu_tsb_miss
4430	  ldx	[%g6 + KPMTSBM_VEND], %g5
4431	cmp	%g2, %g5
4432	bgeu,pn	%xcc, sfmmu_tsb_miss
4433	  stx	%g1, [%g6 + KPMTSBM_TSBPTR]	/* save 8K kpm TSB pointer */
4434
4435	/*
4436	 * check TL tsbmiss handling flag
4437	 * bump tsbmiss counter
4438	 */
4439	lduw	[%g6 + KPMTSBM_TSBMISS], %g5
4440#ifdef	DEBUG
4441	and	%g4, KPMTSBM_TLTSBM_FLAG, %g1
4442	inc	%g5
4443	brz,pn	%g1, sfmmu_kpm_exception
4444	  st	%g5, [%g6 + KPMTSBM_TSBMISS]
4445#else
4446	inc	%g5
4447	st	%g5, [%g6 + KPMTSBM_TSBMISS]
4448#endif
4449	/*
4450	 * At this point:
4451	 *  g1 = clobbered
4452	 *  g2 = tag access register
4453	 *  g3 = 4M kpm TSB pointer (not used)
4454	 *  g6 = per-CPU kpm tsbmiss area
4455	 *  g7 = kpm_vbase
4456	 */
4457
4458	/*
4459	 * Assembly implementation of SFMMU_KPM_VTOP(vaddr, paddr)
4460	 * which is defined in mach_kpm.h. Any changes in that macro
4461	 * should also be ported back to this assembly code.
4462	 */
4463	ldub	[%g6 + KPMTSBM_SZSHIFT], %g3	/* g3 = kpm_size_shift */
4464	sub	%g2, %g7, %g4			/* paddr = vaddr-kpm_vbase */
4465	srax    %g4, %g3, %g7			/* which alias range (r) */
4466	brz,pt	%g7, 2f
4467	  sethi   %hi(vac_colors_mask), %g5
4468	ld	[%g5 + %lo(vac_colors_mask)], %g5
4469
4470	srlx	%g2, MMU_PAGESHIFT, %g1		/* vaddr >> MMU_PAGESHIFT */
4471	and	%g1, %g5, %g1			/* g1 = v */
4472	sllx	%g7, %g3, %g5			/* g5 = r << kpm_size_shift */
4473	cmp	%g7, %g1			/* if (r > v) */
4474	bleu,pn %xcc, 1f
4475	  sub   %g4, %g5, %g4			/* paddr -= r << kpm_size_shift */
4476	sub	%g7, %g1, %g5			/* g5 = r - v */
4477	sllx	%g5, MMU_PAGESHIFT, %g7		/* (r-v) << MMU_PAGESHIFT */
4478	add	%g4, %g7, %g4			/* paddr += (r-v)<<MMU_PAGESHIFT */
4479	ba	2f
4480	  nop
44811:
4482	sllx	%g7, MMU_PAGESHIFT, %g5		/* else */
4483	sub	%g4, %g5, %g4			/* paddr -= r << MMU_PAGESHIFT */
4484
4485	/*
4486	 * paddr2pfn
4487	 *  g1 = vcolor (not used)
4488	 *  g2 = tag access register
4489	 *  g3 = clobbered
4490	 *  g4 = paddr
4491	 *  g5 = clobbered
4492	 *  g6 = per-CPU kpm tsbmiss area
4493	 *  g7 = clobbered
4494	 */
44952:
4496	srlx	%g4, MMU_PAGESHIFT, %g2		/* g2 = pfn */
4497
4498	/*
4499	 * Setup %asi
4500	 * mseg_pa = page_numtomemseg_nolock_pa(pfn)
4501	 * if (mseg not found) sfmmu_kpm_exception
4502	 * g2=pfn g6=per-CPU kpm tsbmiss area
4503	 * g4 g5 g7 for scratch use.
4504	 */
4505	mov	ASI_MEM, %asi
4506	PAGE_NUM2MEMSEG_NOLOCK_PA(%g2, %g3, %g6, %g4, %g5, %g7, kpmtsbmsp2m)
4507	cmp	%g3, MSEG_NULLPTR_PA
4508	be,pn	%xcc, sfmmu_kpm_exception	/* if mseg not found */
4509	  nop
4510
4511	/*
4512	 * inx = pfn - mseg_pa->kpm_pbase
4513	 * g2=pfn  g3=mseg_pa  g6=per-CPU kpm tsbmiss area
4514	 */
4515	ldxa	[%g3 + MEMSEG_KPM_PBASE]%asi, %g7
4516	sub	%g2, %g7, %g4
4517
4518#ifdef	DEBUG
4519	/*
4520	 * Validate inx value
4521	 * g2=pfn g3=mseg_pa g4=inx g6=per-CPU tsbmiss area
4522	 */
4523	ldxa	[%g3 + MEMSEG_KPM_NKPMPGS]%asi, %g5
4524	cmp	%g4, %g5			/* inx - nkpmpgs */
4525	bgeu,pn	%xcc, sfmmu_kpm_exception	/* if out of range */
4526	  ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4527#else
4528	ld	[%g6 + KPMTSBM_KPMPTABLESZ], %g7
4529#endif
4530	/* ksp = &mseg_pa->kpm_spages[inx] */
4531	ldxa	[%g3 + MEMSEG_KPM_SPAGES]%asi, %g5
4532	add	%g5, %g4, %g5			/* ksp */
4533
4534	/*
4535	 * KPMP_SHASH(kp)
4536	 * g2=pfn g3=mseg_pa g4=inx g5=ksp
4537	 * g6=per-CPU kpm tsbmiss area  g7=kpmp_stable_sz
4538	 */
4539	ldub	[%g6 + KPMTSBM_KPMPSHIFT], %g1	/* kpmp_shift */
4540	sub	%g7, 1, %g7			/* mask */
4541	sllx	%g5, %g1, %g1			/* x = ksp << kpmp_shift */
4542	add	%g5, %g1, %g5			/* y = ksp + x */
4543	and 	%g5, %g7, %g5			/* hashinx = y & mask */
4544
4545	/*
4546	 * Calculate physical kpm_spage pointer
4547	 * g2=pfn g3=mseg_pa g4=offset g5=hashinx
4548	 * g6=per-CPU kpm tsbmiss area
4549	 */
4550	ldxa	[%g3 + MEMSEG_KPM_PAGESPA]%asi, %g1 /* kpm_spagespa */
4551	add	%g1, %g4, %g1			/* ksp_pa */
4552
4553	/*
4554	 * Calculate physical hash lock address.
4555	 * Note: Changes in kpm_shlk_t must be reflected here.
4556	 * g1=ksp_pa g2=pfn g5=hashinx
4557	 * g6=per-CPU kpm tsbmiss area
4558	 */
4559	ldx	[%g6 + KPMTSBM_KPMPTABLEPA], %g4 /* kpmp_stablepa */
4560	sllx	%g5, KPMSHLK_SHIFT, %g5
4561	add	%g4, %g5, %g3			/* hlck_pa */
4562
4563	/*
4564	 * Assemble non-cacheable tte initially
4565	 * g1=ksp_pa g2=pfn g3=hlck_pa
4566	 * g6=per-CPU kpm tsbmiss area
4567	 */
4568	sethi	%hi(TTE_VALID_INT), %g5		/* upper part */
4569	sllx	%g5, 32, %g5
4570	mov	(TTE_CP_INT|TTE_PRIV_INT|TTE_HWWR_INT), %g4
4571	or	%g5, %g4, %g5
4572	sllx	%g2, MMU_PAGESHIFT, %g4
4573	or	%g5, %g4, %g5			/* tte */
4574	ldx	[%g6 + KPMTSBM_TSBPTR], %g4
4575	GET_MMU_D_TTARGET(%g2, %g7)		/* %g2 = ttarget */
4576
4577	/*
4578	 * tsb dropin
4579	 * g1=ksp_pa g2=ttarget g3=hlck_pa g4=ktsbp g5=tte (non-cacheable)
4580	 * g6=per-CPU kpm tsbmiss area  g7=scratch register
4581	 */
4582
4583	/* KPMLOCK_ENTER(kpmlckp, tmp1, label1, asi) */
4584	KPMLOCK_ENTER(%g3, %g7, kpmtsbsmlock, ASI_MEM)
4585
4586	/* use C-handler if there's no go for dropin */
4587	ldsba	[%g1 + KPMSPAGE_MAPPED]%asi, %g7	/* kp_mapped */
4588	andcc	%g7, KPM_MAPPED_GO, %g0			/* go or no go ? */
4589	bz,pt	%icc, 5f				/* no go */
4590	  nop
4591	and	%g7, KPM_MAPPED_MASK, %g7		/* go */
4592	cmp	%g7, KPM_MAPPEDS			/* cacheable ? */
4593	be,a,pn	%xcc, 3f
4594	  or	%g5, TTE_CV_INT, %g5			/* cacheable */
45953:
4596#ifndef sun4v
4597	ldub	[%g6 + KPMTSBM_FLAGS], %g7
4598	mov	ASI_N, %g1
4599	andcc	%g7, KPMTSBM_TSBPHYS_FLAG, %g0
4600	movnz	%icc, ASI_MEM, %g1
4601	mov	%g1, %asi
4602#endif
4603
4604	/*
4605	 * TSB_LOCK_ENTRY(tsbp, tmp1, tmp2, label) (needs %asi set)
4606	 * If we fail to lock the TSB entry then just load the tte into the
4607	 * TLB.
4608	 */
4609	TSB_LOCK_ENTRY(%g4, %g1, %g7, locked_tsb_l2)
4610
4611	/* TSB_INSERT_UNLOCK_ENTRY(tsbp, tte, tagtarget, tmp) */
4612	TSB_INSERT_UNLOCK_ENTRY(%g4, %g5, %g2, %g7)
4613locked_tsb_l2:
4614	DTLB_STUFF(%g5, %g2, %g4, %g5, %g6)
4615
4616	/* KPMLOCK_EXIT(kpmlckp, asi) */
4617	KPMLOCK_EXIT(%g3, ASI_MEM)
4618
4619	/*
4620	 * If trapstat is running, we need to shift the %tpc and %tnpc to
4621	 * point to trapstat's TSB miss return code (note that trapstat
4622	 * itself will patch the correct offset to add).
4623	 * Note: TTE is expected in %g5 (allows per pagesize reporting).
4624	 */
4625	rdpr	%tl, %g7
4626	cmp	%g7, 1
4627	ble	%icc, 0f
4628	sethi	%hi(KERNELBASE), %g6
4629	rdpr	%tpc, %g7
4630	or	%g6, %lo(KERNELBASE), %g6
4631	cmp	%g7, %g6
4632	bgeu	%xcc, 0f
4633	ALTENTRY(tsbmiss_trapstat_patch_point_kpm_small)
4634	add	%g7, RUNTIME_PATCH, %g7	/* must match TSTAT_TSBMISS_INSTR */
4635	wrpr	%g7, %tpc
4636	add	%g7, 4, %g7
4637	wrpr	%g7, %tnpc
46380:
4639	retry
46405:
4641	/* g3=hlck_pa */
4642	KPMLOCK_EXIT(%g3, ASI_MEM)
4643	ba,pt	%icc, sfmmu_kpm_exception
4644	  nop
4645	SET_SIZE(sfmmu_kpm_dtsb_miss_small)
4646
4647#if (1<< KPMTSBM_SHIFT) != KPMTSBM_SIZE
4648#error - KPMTSBM_SHIFT does not correspond to size of kpmtsbm struct
4649#endif
4650
4651#endif /* lint */
4652
4653#ifdef	lint
4654/*
4655 * Enable/disable tsbmiss handling at trap level for a kpm (large) page.
4656 * Called from C-level, sets/clears "go" indication for trap level handler.
4657 * khl_lock is a low level spin lock to protect the kp_tsbmtl field.
4658 * Assumed that &kp->kp_refcntc is checked for zero or -1 at C-level.
4659 * Assumes khl_mutex is held when called from C-level.
4660 */
4661/* ARGSUSED */
4662void
4663sfmmu_kpm_tsbmtl(short *kp_refcntc, uint_t *khl_lock, int cmd)
4664{
4665}
4666
4667/*
4668 * kpm_smallpages: stores val to byte at address mapped within
4669 * low level lock brackets. The old value is returned.
4670 * Called from C-level.
4671 */
4672/* ARGSUSED */
4673int
4674sfmmu_kpm_stsbmtl(uchar_t *mapped, uint_t *kshl_lock, int val)
4675{
4676	return (0);
4677}
4678
4679#else /* lint */
4680
4681	.seg	".data"
4682sfmmu_kpm_tsbmtl_panic:
4683	.ascii	"sfmmu_kpm_tsbmtl: interrupts disabled"
4684	.byte	0
4685sfmmu_kpm_stsbmtl_panic:
4686	.ascii	"sfmmu_kpm_stsbmtl: interrupts disabled"
4687	.byte	0
4688	.align	4
4689	.seg	".text"
4690
4691	ENTRY_NP(sfmmu_kpm_tsbmtl)
4692	rdpr	%pstate, %o3
4693	/*
4694	 * %o0 = &kp_refcntc
4695	 * %o1 = &khl_lock
4696	 * %o2 = 0/1 (off/on)
4697	 * %o3 = pstate save
4698	 */
4699#ifdef DEBUG
4700	andcc	%o3, PSTATE_IE, %g0		/* if interrupts already */
4701	bnz,pt %icc, 1f				/* disabled, panic	 */
4702	  nop
4703	save	%sp, -SA(MINFRAME), %sp
4704	sethi	%hi(sfmmu_kpm_tsbmtl_panic), %o0
4705	call	panic
4706	 or	%o0, %lo(sfmmu_kpm_tsbmtl_panic), %o0
4707	ret
4708	restore
47091:
4710#endif /* DEBUG */
4711	wrpr	%o3, PSTATE_IE, %pstate		/* disable interrupts */
4712
4713	KPMLOCK_ENTER(%o1, %o4, kpmtsbmtl1, ASI_N)
4714	mov	-1, %o5
4715	brz,a	%o2, 2f
4716	  mov	0, %o5
47172:
4718	sth	%o5, [%o0]
4719	KPMLOCK_EXIT(%o1, ASI_N)
4720
4721	retl
4722	  wrpr	%g0, %o3, %pstate		/* enable interrupts */
4723	SET_SIZE(sfmmu_kpm_tsbmtl)
4724
4725	ENTRY_NP(sfmmu_kpm_stsbmtl)
4726	rdpr	%pstate, %o3
4727	/*
4728	 * %o0 = &mapped
4729	 * %o1 = &kshl_lock
4730	 * %o2 = val
4731	 * %o3 = pstate save
4732	 */
4733#ifdef DEBUG
4734	andcc	%o3, PSTATE_IE, %g0		/* if interrupts already */
4735	bnz,pt %icc, 1f				/* disabled, panic	 */
4736	  nop
4737	save	%sp, -SA(MINFRAME), %sp
4738	sethi	%hi(sfmmu_kpm_stsbmtl_panic), %o0
4739	call	panic
4740	  or	%o0, %lo(sfmmu_kpm_stsbmtl_panic), %o0
4741	ret
4742	restore
47431:
4744#endif /* DEBUG */
4745	wrpr	%o3, PSTATE_IE, %pstate		/* disable interrupts */
4746
4747	KPMLOCK_ENTER(%o1, %o4, kpmstsbmtl1, ASI_N)
4748	ldsb	[%o0], %o5
4749	stb	%o2, [%o0]
4750	KPMLOCK_EXIT(%o1, ASI_N)
4751
4752	and	%o5, KPM_MAPPED_MASK, %o0	/* return old val */
4753	retl
4754	  wrpr	%g0, %o3, %pstate		/* enable interrupts */
4755	SET_SIZE(sfmmu_kpm_stsbmtl)
4756
4757#endif /* lint */
4758
4759#ifndef lint
4760#ifdef sun4v
4761	/*
4762	 * User/kernel data miss w// multiple TSBs
4763	 * The first probe covers 8K, 64K, and 512K page sizes,
4764	 * because 64K and 512K mappings are replicated off 8K
4765	 * pointer.  Second probe covers 4M page size only.
4766	 *
4767	 * MMU fault area contains miss address and context.
4768	 */
4769	ALTENTRY(sfmmu_slow_dmmu_miss)
4770	GET_MMU_D_PTAGACC_CTXTYPE(%g2, %g3)	! %g2 = ptagacc, %g3 = ctx type
4771
4772slow_miss_common:
4773	/*
4774	 *  %g2 = tagacc register (needed for sfmmu_tsb_miss_tt)
4775	 *  %g3 = ctx (cannot be INVALID_CONTEXT)
4776	 */
4777	brnz,pt	%g3, 8f			! check for user context
4778	  nop
4779
4780	/*
4781	 * Kernel miss
4782	 * Get 8K and 4M TSB pointers in %g1 and %g3 and
4783	 * branch to sfmmu_tsb_miss_tt to handle it.
4784	 */
4785	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
4786sfmmu_dslow_patch_ktsb_base:
4787	RUNTIME_PATCH_SETX(%g1, %g6)	! %g1 = contents of ktsb_pbase
4788sfmmu_dslow_patch_ktsb_szcode:
4789	or	%g0, RUNTIME_PATCH, %g3	! ktsb_szcode (hot patched)
4790
4791	GET_TSBE_POINTER(MMU_PAGESHIFT, %g1, %g7, %g3, %g5)
4792	! %g1 = First TSB entry pointer, as TSB miss handler expects
4793
4794	mov	%g2, %g7		! TSB pointer macro clobbers tagacc
4795sfmmu_dslow_patch_ktsb4m_base:
4796	RUNTIME_PATCH_SETX(%g3, %g6)	! %g3 = contents of ktsb4m_pbase
4797sfmmu_dslow_patch_ktsb4m_szcode:
4798	or	%g0, RUNTIME_PATCH, %g6	! ktsb4m_szcode (hot patched)
4799
4800	GET_TSBE_POINTER(MMU_PAGESHIFT4M, %g3, %g7, %g6, %g5)
4801	! %g3 = 4M tsb entry pointer, as TSB miss handler expects
4802	ba,a,pt	%xcc, sfmmu_tsb_miss_tt
4803	.empty
4804
48058:
4806	/*
4807	 * User miss
4808	 * Get first TSB pointer in %g1
4809	 * Get second TSB pointer (or NULL if no second TSB) in %g3
4810	 * Branch to sfmmu_tsb_miss_tt to handle it
4811	 */
4812	GET_1ST_TSBE_PTR(%g2, %g1, %g4, %g5)
4813	/* %g1 = first TSB entry ptr now, %g2 preserved */
4814
4815	GET_UTSBREG(SCRATCHPAD_UTSBREG2, %g3)	/* get 2nd utsbreg */
4816	brlz,pt %g3, sfmmu_tsb_miss_tt		/* done if no 2nd TSB */
4817	  nop
4818
4819	GET_2ND_TSBE_PTR(%g2, %g3, %g4, %g5)
4820	/* %g3 = second TSB entry ptr now, %g2 preserved */
48219:
4822	ba,a,pt	%xcc, sfmmu_tsb_miss_tt
4823	.empty
4824	SET_SIZE(sfmmu_slow_dmmu_miss)
4825
4826
4827	/*
4828	 * User/kernel instruction miss w/ multiple TSBs
4829	 * The first probe covers 8K, 64K, and 512K page sizes,
4830	 * because 64K and 512K mappings are replicated off 8K
4831	 * pointer.  Second probe covers 4M page size only.
4832	 *
4833	 * MMU fault area contains miss address and context.
4834	 */
4835	ALTENTRY(sfmmu_slow_immu_miss)
4836	GET_MMU_I_PTAGACC_CTXTYPE(%g2, %g3)
4837	ba,a,pt	%xcc, slow_miss_common
4838	SET_SIZE(sfmmu_slow_immu_miss)
4839
4840#endif /* sun4v */
4841#endif	/* lint */
4842
4843#ifndef lint
4844
4845/*
4846 * Per-CPU tsbmiss areas to avoid cache misses in TSB miss handlers.
4847 */
4848	.seg	".data"
4849	.align	64
4850	.global tsbmiss_area
4851tsbmiss_area:
4852	.skip	(TSBMISS_SIZE * NCPU)
4853
4854	.align	64
4855	.global kpmtsbm_area
4856kpmtsbm_area:
4857	.skip	(KPMTSBM_SIZE * NCPU)
4858#endif	/* lint */
4859