xref: /titanic_51/usr/src/uts/sun4u/ml/trap_table.s (revision 19d61fc7991644175873937566d932d8cf52912a)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#if !defined(lint)
29#include "assym.h"
30#endif /* !lint */
31#include <sys/asm_linkage.h>
32#include <sys/privregs.h>
33#include <sys/sun4asi.h>
34#include <sys/spitregs.h>
35#include <sys/cheetahregs.h>
36#include <sys/machtrap.h>
37#include <sys/machthread.h>
38#include <sys/machbrand.h>
39#include <sys/pcb.h>
40#include <sys/pte.h>
41#include <sys/mmu.h>
42#include <sys/machpcb.h>
43#include <sys/async.h>
44#include <sys/intreg.h>
45#include <sys/scb.h>
46#include <sys/psr_compat.h>
47#include <sys/syscall.h>
48#include <sys/machparam.h>
49#include <sys/traptrace.h>
50#include <vm/hat_sfmmu.h>
51#include <sys/archsystm.h>
52#include <sys/utrap.h>
53#include <sys/clock.h>
54#include <sys/intr.h>
55#include <sys/fpu/fpu_simulator.h>
56#include <vm/seg_spt.h>
57
58/*
59 * WARNING: If you add a fast trap handler which can be invoked by a
60 * non-privileged user, you may have to use the FAST_TRAP_DONE macro
61 * instead of "done" instruction to return back to the user mode. See
62 * comments for the "fast_trap_done" entry point for more information.
63 *
64 * An alternate FAST_TRAP_DONE_CHK_INTR macro should be used for the
65 * cases where you always want to process any pending interrupts before
66 * returning back to the user mode.
67 */
68#define	FAST_TRAP_DONE		\
69	ba,a	fast_trap_done
70
71#define	FAST_TRAP_DONE_CHK_INTR	\
72	ba,a	fast_trap_done_chk_intr
73
74/*
75 * SPARC V9 Trap Table
76 *
77 * Most of the trap handlers are made from common building
78 * blocks, and some are instantiated multiple times within
79 * the trap table. So, I build a bunch of macros, then
80 * populate the table using only the macros.
81 *
82 * Many macros branch to sys_trap.  Its calling convention is:
83 *	%g1		kernel trap handler
84 *	%g2, %g3	args for above
85 *	%g4		desire %pil
86 */
87
88#ifdef	TRAPTRACE
89
90/*
91 * Tracing macro. Adds two instructions if TRAPTRACE is defined.
92 */
93#define	TT_TRACE(label)		\
94	ba	label		;\
95	rd	%pc, %g7
96#define	TT_TRACE_INS	2
97
98#define	TT_TRACE_L(label)	\
99	ba	label		;\
100	rd	%pc, %l4	;\
101	clr	%l4
102#define	TT_TRACE_L_INS	3
103
104#else
105
106#define	TT_TRACE(label)
107#define	TT_TRACE_INS	0
108
109#define	TT_TRACE_L(label)
110#define	TT_TRACE_L_INS	0
111
112#endif
113
114/*
115 * This first set are funneled to trap() with %tt as the type.
116 * Trap will then either panic or send the user a signal.
117 */
118/*
119 * NOT is used for traps that just shouldn't happen.
120 * It comes in both single and quadruple flavors.
121 */
122#if !defined(lint)
123	.global	trap
124#endif /* !lint */
125#define	NOT			\
126	TT_TRACE(trace_gen)	;\
127	set	trap, %g1	;\
128	rdpr	%tt, %g3	;\
129	ba,pt	%xcc, sys_trap	;\
130	sub	%g0, 1, %g4	;\
131	.align	32
132#define	NOT4	NOT; NOT; NOT; NOT
133/*
134 * RED is for traps that use the red mode handler.
135 * We should never see these either.
136 */
137#define	RED	NOT
138/*
139 * BAD is used for trap vectors we don't have a kernel
140 * handler for.
141 * It also comes in single and quadruple versions.
142 */
143#define	BAD	NOT
144#define	BAD4	NOT4
145
146#define	DONE			\
147	done;			\
148	.align	32
149
150/*
151 * TRAP vectors to the trap() function.
152 * It's main use is for user errors.
153 */
154#if !defined(lint)
155	.global	trap
156#endif /* !lint */
157#define	TRAP(arg)		\
158	TT_TRACE(trace_gen)	;\
159	set	trap, %g1	;\
160	mov	arg, %g3	;\
161	ba,pt	%xcc, sys_trap	;\
162	sub	%g0, 1, %g4	;\
163	.align	32
164
165/*
166 * SYSCALL is used for unsupported syscall interfaces (with 'which'
167 * set to 'nosys') and legacy support of old SunOS 4.x syscalls (with
168 * 'which' set to 'syscall_trap32').
169 *
170 * The SYSCALL_TRAP* macros are used for syscall entry points.
171 * SYSCALL_TRAP is used to support LP64 syscalls and SYSCALL_TRAP32
172 * is used to support ILP32.  Each macro can only be used once
173 * since they each define a symbol.  The symbols are used as hot patch
174 * points by the brand infrastructure to dynamically enable and disable
175 * brand syscall interposition.  See the comments around BRAND_CALLBACK
176 * and brand_plat_interposition_enable() for more information.
177 */
178#define	SYSCALL_NOTT(which)		\
179	set	(which), %g1		;\
180	ba,pt	%xcc, sys_trap		;\
181	sub	%g0, 1, %g4		;\
182	.align	32
183
184#define	SYSCALL(which)			\
185	TT_TRACE(trace_gen)		;\
186	SYSCALL_NOTT(which)
187
188#define	SYSCALL_TRAP32				\
189	TT_TRACE(trace_gen)			;\
190	ALTENTRY(syscall_trap32_patch_point)	\
191	SYSCALL_NOTT(syscall_trap32)
192
193#define	SYSCALL_TRAP				\
194	TT_TRACE(trace_gen)			;\
195	ALTENTRY(syscall_trap_patch_point)	\
196	SYSCALL_NOTT(syscall_trap)
197
198#define	FLUSHW()			\
199	set	trap, %g1		;\
200	mov	T_FLUSHW, %g3		;\
201	sub	%g0, 1, %g4		;\
202	save				;\
203	flushw				;\
204	restore				;\
205	FAST_TRAP_DONE			;\
206	.align	32
207
208/*
209 * GOTO just jumps to a label.
210 * It's used for things that can be fixed without going thru sys_trap.
211 */
212#define	GOTO(label)		\
213	.global	label		;\
214	ba,a	label		;\
215	.empty			;\
216	.align	32
217
218/*
219 * GOTO_TT just jumps to a label.
220 * correctable ECC error traps at  level 0 and 1 will use this macro.
221 * It's used for things that can be fixed without going thru sys_trap.
222 */
223#define	GOTO_TT(label, ttlabel)		\
224	.global	label		;\
225	TT_TRACE(ttlabel)	;\
226	ba,a	label		;\
227	.empty			;\
228	.align	32
229
230/*
231 * Privileged traps
232 * Takes breakpoint if privileged, calls trap() if not.
233 */
234#define	PRIV(label)			\
235	rdpr	%tstate, %g1		;\
236	btst	TSTATE_PRIV, %g1	;\
237	bnz	label			;\
238	rdpr	%tt, %g3		;\
239	set	trap, %g1		;\
240	ba,pt	%xcc, sys_trap		;\
241	sub	%g0, 1, %g4		;\
242	.align	32
243
244
245/*
246 * DTrace traps.
247 */
248#define	DTRACE_PID			\
249	.global dtrace_pid_probe				;\
250	set	dtrace_pid_probe, %g1				;\
251	ba,pt	%xcc, user_trap					;\
252	sub	%g0, 1, %g4					;\
253	.align	32
254
255#define	DTRACE_RETURN			\
256	.global dtrace_return_probe				;\
257	set	dtrace_return_probe, %g1			;\
258	ba,pt	%xcc, user_trap					;\
259	sub	%g0, 1, %g4					;\
260	.align	32
261
262/*
263 * REGISTER WINDOW MANAGEMENT MACROS
264 */
265
266/*
267 * various convenient units of padding
268 */
269#define	SKIP(n)	.skip 4*(n)
270
271/*
272 * CLEAN_WINDOW is the simple handler for cleaning a register window.
273 */
274#define	CLEAN_WINDOW						\
275	TT_TRACE_L(trace_win)					;\
276	rdpr %cleanwin, %l0; inc %l0; wrpr %l0, %cleanwin	;\
277	clr %l0; clr %l1; clr %l2; clr %l3			;\
278	clr %l4; clr %l5; clr %l6; clr %l7			;\
279	clr %o0; clr %o1; clr %o2; clr %o3			;\
280	clr %o4; clr %o5; clr %o6; clr %o7			;\
281	retry; .align 128
282
283#if !defined(lint)
284
285/*
286 * If we get an unresolved tlb miss while in a window handler, the fault
287 * handler will resume execution at the last instruction of the window
288 * hander, instead of delivering the fault to the kernel.  Spill handlers
289 * use this to spill windows into the wbuf.
290 *
291 * The mixed handler works by checking %sp, and branching to the correct
292 * handler.  This is done by branching back to label 1: for 32b frames,
293 * or label 2: for 64b frames; which implies the handler order is: 32b,
294 * 64b, mixed.  The 1: and 2: labels are offset into the routines to
295 * allow the branchs' delay slots to contain useful instructions.
296 */
297
298/*
299 * SPILL_32bit spills a 32-bit-wide kernel register window.  It
300 * assumes that the kernel context and the nucleus context are the
301 * same.  The stack pointer is required to be eight-byte aligned even
302 * though this code only needs it to be four-byte aligned.
303 */
304#define	SPILL_32bit(tail)					\
305	srl	%sp, 0, %sp					;\
3061:	st	%l0, [%sp + 0]					;\
307	st	%l1, [%sp + 4]					;\
308	st	%l2, [%sp + 8]					;\
309	st	%l3, [%sp + 12]					;\
310	st	%l4, [%sp + 16]					;\
311	st	%l5, [%sp + 20]					;\
312	st	%l6, [%sp + 24]					;\
313	st	%l7, [%sp + 28]					;\
314	st	%i0, [%sp + 32]					;\
315	st	%i1, [%sp + 36]					;\
316	st	%i2, [%sp + 40]					;\
317	st	%i3, [%sp + 44]					;\
318	st	%i4, [%sp + 48]					;\
319	st	%i5, [%sp + 52]					;\
320	st	%i6, [%sp + 56]					;\
321	st	%i7, [%sp + 60]					;\
322	TT_TRACE_L(trace_win)					;\
323	saved							;\
324	retry							;\
325	SKIP(31-19-TT_TRACE_L_INS)				;\
326	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
327	.empty
328
329/*
330 * SPILL_32bit_asi spills a 32-bit-wide register window into a 32-bit
331 * wide address space via the designated asi.  It is used to spill
332 * non-kernel windows.  The stack pointer is required to be eight-byte
333 * aligned even though this code only needs it to be four-byte
334 * aligned.
335 */
336#define	SPILL_32bit_asi(asi_num, tail)				\
337	srl	%sp, 0, %sp					;\
3381:	sta	%l0, [%sp + %g0]asi_num				;\
339	mov	4, %g1						;\
340	sta	%l1, [%sp + %g1]asi_num				;\
341	mov	8, %g2						;\
342	sta	%l2, [%sp + %g2]asi_num				;\
343	mov	12, %g3						;\
344	sta	%l3, [%sp + %g3]asi_num				;\
345	add	%sp, 16, %g4					;\
346	sta	%l4, [%g4 + %g0]asi_num				;\
347	sta	%l5, [%g4 + %g1]asi_num				;\
348	sta	%l6, [%g4 + %g2]asi_num				;\
349	sta	%l7, [%g4 + %g3]asi_num				;\
350	add	%g4, 16, %g4					;\
351	sta	%i0, [%g4 + %g0]asi_num				;\
352	sta	%i1, [%g4 + %g1]asi_num				;\
353	sta	%i2, [%g4 + %g2]asi_num				;\
354	sta	%i3, [%g4 + %g3]asi_num				;\
355	add	%g4, 16, %g4					;\
356	sta	%i4, [%g4 + %g0]asi_num				;\
357	sta	%i5, [%g4 + %g1]asi_num				;\
358	sta	%i6, [%g4 + %g2]asi_num				;\
359	sta	%i7, [%g4 + %g3]asi_num				;\
360	TT_TRACE_L(trace_win)					;\
361	saved							;\
362	retry							;\
363	SKIP(31-25-TT_TRACE_L_INS)				;\
364	ba,a,pt %xcc, fault_32bit_/**/tail			;\
365	.empty
366
367/*
368 * SPILL_32bit_tt1 spills a 32-bit-wide register window into a 32-bit
369 * wide address space via the designated asi.  It is used to spill
370 * windows at tl>1 where performance isn't the primary concern and
371 * where we don't want to use unnecessary registers.  The stack
372 * pointer is required to be eight-byte aligned even though this code
373 * only needs it to be four-byte aligned.
374 */
375#define	SPILL_32bit_tt1(asi_num, tail)				\
376	mov	asi_num, %asi					;\
3771:	srl	%sp, 0, %sp					;\
378	sta	%l0, [%sp + 0]%asi				;\
379	sta	%l1, [%sp + 4]%asi				;\
380	sta	%l2, [%sp + 8]%asi				;\
381	sta	%l3, [%sp + 12]%asi				;\
382	sta	%l4, [%sp + 16]%asi				;\
383	sta	%l5, [%sp + 20]%asi				;\
384	sta	%l6, [%sp + 24]%asi				;\
385	sta	%l7, [%sp + 28]%asi				;\
386	sta	%i0, [%sp + 32]%asi				;\
387	sta	%i1, [%sp + 36]%asi				;\
388	sta	%i2, [%sp + 40]%asi				;\
389	sta	%i3, [%sp + 44]%asi				;\
390	sta	%i4, [%sp + 48]%asi				;\
391	sta	%i5, [%sp + 52]%asi				;\
392	sta	%i6, [%sp + 56]%asi				;\
393	sta	%i7, [%sp + 60]%asi				;\
394	TT_TRACE_L(trace_win)					;\
395	saved							;\
396	retry							;\
397	SKIP(31-20-TT_TRACE_L_INS)				;\
398	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
399	.empty
400
401
402/*
403 * FILL_32bit fills a 32-bit-wide kernel register window.  It assumes
404 * that the kernel context and the nucleus context are the same.  The
405 * stack pointer is required to be eight-byte aligned even though this
406 * code only needs it to be four-byte aligned.
407 */
408#define	FILL_32bit(tail)					\
409	srl	%sp, 0, %sp					;\
4101:	TT_TRACE_L(trace_win)					;\
411	ld	[%sp + 0], %l0					;\
412	ld	[%sp + 4], %l1					;\
413	ld	[%sp + 8], %l2					;\
414	ld	[%sp + 12], %l3					;\
415	ld	[%sp + 16], %l4					;\
416	ld	[%sp + 20], %l5					;\
417	ld	[%sp + 24], %l6					;\
418	ld	[%sp + 28], %l7					;\
419	ld	[%sp + 32], %i0					;\
420	ld	[%sp + 36], %i1					;\
421	ld	[%sp + 40], %i2					;\
422	ld	[%sp + 44], %i3					;\
423	ld	[%sp + 48], %i4					;\
424	ld	[%sp + 52], %i5					;\
425	ld	[%sp + 56], %i6					;\
426	ld	[%sp + 60], %i7					;\
427	restored						;\
428	retry							;\
429	SKIP(31-19-TT_TRACE_L_INS)				;\
430	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
431	.empty
432
433/*
434 * FILL_32bit_asi fills a 32-bit-wide register window from a 32-bit
435 * wide address space via the designated asi.  It is used to fill
436 * non-kernel windows.  The stack pointer is required to be eight-byte
437 * aligned even though this code only needs it to be four-byte
438 * aligned.
439 */
440#define	FILL_32bit_asi(asi_num, tail)				\
441	srl	%sp, 0, %sp					;\
4421:	TT_TRACE_L(trace_win)					;\
443	mov	4, %g1						;\
444	lda	[%sp + %g0]asi_num, %l0				;\
445	mov	8, %g2						;\
446	lda	[%sp + %g1]asi_num, %l1				;\
447	mov	12, %g3						;\
448	lda	[%sp + %g2]asi_num, %l2				;\
449	lda	[%sp + %g3]asi_num, %l3				;\
450	add	%sp, 16, %g4					;\
451	lda	[%g4 + %g0]asi_num, %l4				;\
452	lda	[%g4 + %g1]asi_num, %l5				;\
453	lda	[%g4 + %g2]asi_num, %l6				;\
454	lda	[%g4 + %g3]asi_num, %l7				;\
455	add	%g4, 16, %g4					;\
456	lda	[%g4 + %g0]asi_num, %i0				;\
457	lda	[%g4 + %g1]asi_num, %i1				;\
458	lda	[%g4 + %g2]asi_num, %i2				;\
459	lda	[%g4 + %g3]asi_num, %i3				;\
460	add	%g4, 16, %g4					;\
461	lda	[%g4 + %g0]asi_num, %i4				;\
462	lda	[%g4 + %g1]asi_num, %i5				;\
463	lda	[%g4 + %g2]asi_num, %i6				;\
464	lda	[%g4 + %g3]asi_num, %i7				;\
465	restored						;\
466	retry							;\
467	SKIP(31-25-TT_TRACE_L_INS)				;\
468	ba,a,pt %xcc, fault_32bit_/**/tail			;\
469	.empty
470
471/*
472 * FILL_32bit_tt1 fills a 32-bit-wide register window from a 32-bit
473 * wide address space via the designated asi.  It is used to fill
474 * windows at tl>1 where performance isn't the primary concern and
475 * where we don't want to use unnecessary registers.  The stack
476 * pointer is required to be eight-byte aligned even though this code
477 * only needs it to be four-byte aligned.
478 */
479#define	FILL_32bit_tt1(asi_num, tail)				\
480	mov	asi_num, %asi					;\
4811:	srl	%sp, 0, %sp					;\
482	TT_TRACE_L(trace_win)					;\
483	lda	[%sp + 0]%asi, %l0				;\
484	lda	[%sp + 4]%asi, %l1				;\
485	lda	[%sp + 8]%asi, %l2				;\
486	lda	[%sp + 12]%asi, %l3				;\
487	lda	[%sp + 16]%asi, %l4				;\
488	lda	[%sp + 20]%asi, %l5				;\
489	lda	[%sp + 24]%asi, %l6				;\
490	lda	[%sp + 28]%asi, %l7				;\
491	lda	[%sp + 32]%asi, %i0				;\
492	lda	[%sp + 36]%asi, %i1				;\
493	lda	[%sp + 40]%asi, %i2				;\
494	lda	[%sp + 44]%asi, %i3				;\
495	lda	[%sp + 48]%asi, %i4				;\
496	lda	[%sp + 52]%asi, %i5				;\
497	lda	[%sp + 56]%asi, %i6				;\
498	lda	[%sp + 60]%asi, %i7				;\
499	restored						;\
500	retry							;\
501	SKIP(31-20-TT_TRACE_L_INS)				;\
502	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
503	.empty
504
505
506/*
507 * SPILL_64bit spills a 64-bit-wide kernel register window.  It
508 * assumes that the kernel context and the nucleus context are the
509 * same.  The stack pointer is required to be eight-byte aligned.
510 */
511#define	SPILL_64bit(tail)					\
5122:	stx	%l0, [%sp + V9BIAS64 + 0]			;\
513	stx	%l1, [%sp + V9BIAS64 + 8]			;\
514	stx	%l2, [%sp + V9BIAS64 + 16]			;\
515	stx	%l3, [%sp + V9BIAS64 + 24]			;\
516	stx	%l4, [%sp + V9BIAS64 + 32]			;\
517	stx	%l5, [%sp + V9BIAS64 + 40]			;\
518	stx	%l6, [%sp + V9BIAS64 + 48]			;\
519	stx	%l7, [%sp + V9BIAS64 + 56]			;\
520	stx	%i0, [%sp + V9BIAS64 + 64]			;\
521	stx	%i1, [%sp + V9BIAS64 + 72]			;\
522	stx	%i2, [%sp + V9BIAS64 + 80]			;\
523	stx	%i3, [%sp + V9BIAS64 + 88]			;\
524	stx	%i4, [%sp + V9BIAS64 + 96]			;\
525	stx	%i5, [%sp + V9BIAS64 + 104]			;\
526	stx	%i6, [%sp + V9BIAS64 + 112]			;\
527	stx	%i7, [%sp + V9BIAS64 + 120]			;\
528	TT_TRACE_L(trace_win)					;\
529	saved							;\
530	retry							;\
531	SKIP(31-18-TT_TRACE_L_INS)				;\
532	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
533	.empty
534
535/*
536 * SPILL_64bit_asi spills a 64-bit-wide register window into a 64-bit
537 * wide address space via the designated asi.  It is used to spill
538 * non-kernel windows.  The stack pointer is required to be eight-byte
539 * aligned.
540 */
541#define	SPILL_64bit_asi(asi_num, tail)				\
542	mov	0 + V9BIAS64, %g1				;\
5432:	stxa	%l0, [%sp + %g1]asi_num				;\
544	mov	8 + V9BIAS64, %g2				;\
545	stxa	%l1, [%sp + %g2]asi_num				;\
546	mov	16 + V9BIAS64, %g3				;\
547	stxa	%l2, [%sp + %g3]asi_num				;\
548	mov	24 + V9BIAS64, %g4				;\
549	stxa	%l3, [%sp + %g4]asi_num				;\
550	add	%sp, 32, %g5					;\
551	stxa	%l4, [%g5 + %g1]asi_num				;\
552	stxa	%l5, [%g5 + %g2]asi_num				;\
553	stxa	%l6, [%g5 + %g3]asi_num				;\
554	stxa	%l7, [%g5 + %g4]asi_num				;\
555	add	%g5, 32, %g5					;\
556	stxa	%i0, [%g5 + %g1]asi_num				;\
557	stxa	%i1, [%g5 + %g2]asi_num				;\
558	stxa	%i2, [%g5 + %g3]asi_num				;\
559	stxa	%i3, [%g5 + %g4]asi_num				;\
560	add	%g5, 32, %g5					;\
561	stxa	%i4, [%g5 + %g1]asi_num				;\
562	stxa	%i5, [%g5 + %g2]asi_num				;\
563	stxa	%i6, [%g5 + %g3]asi_num				;\
564	stxa	%i7, [%g5 + %g4]asi_num				;\
565	TT_TRACE_L(trace_win)					;\
566	saved							;\
567	retry							;\
568	SKIP(31-25-TT_TRACE_L_INS)				;\
569	ba,a,pt %xcc, fault_64bit_/**/tail			;\
570	.empty
571
572/*
573 * SPILL_64bit_tt1 spills a 64-bit-wide register window into a 64-bit
574 * wide address space via the designated asi.  It is used to spill
575 * windows at tl>1 where performance isn't the primary concern and
576 * where we don't want to use unnecessary registers.  The stack
577 * pointer is required to be eight-byte aligned.
578 */
579#define	SPILL_64bit_tt1(asi_num, tail)				\
580	mov	asi_num, %asi					;\
5812:	stxa	%l0, [%sp + V9BIAS64 + 0]%asi			;\
582	stxa	%l1, [%sp + V9BIAS64 + 8]%asi			;\
583	stxa	%l2, [%sp + V9BIAS64 + 16]%asi			;\
584	stxa	%l3, [%sp + V9BIAS64 + 24]%asi			;\
585	stxa	%l4, [%sp + V9BIAS64 + 32]%asi			;\
586	stxa	%l5, [%sp + V9BIAS64 + 40]%asi			;\
587	stxa	%l6, [%sp + V9BIAS64 + 48]%asi			;\
588	stxa	%l7, [%sp + V9BIAS64 + 56]%asi			;\
589	stxa	%i0, [%sp + V9BIAS64 + 64]%asi			;\
590	stxa	%i1, [%sp + V9BIAS64 + 72]%asi			;\
591	stxa	%i2, [%sp + V9BIAS64 + 80]%asi			;\
592	stxa	%i3, [%sp + V9BIAS64 + 88]%asi			;\
593	stxa	%i4, [%sp + V9BIAS64 + 96]%asi			;\
594	stxa	%i5, [%sp + V9BIAS64 + 104]%asi			;\
595	stxa	%i6, [%sp + V9BIAS64 + 112]%asi			;\
596	stxa	%i7, [%sp + V9BIAS64 + 120]%asi			;\
597	TT_TRACE_L(trace_win)					;\
598	saved							;\
599	retry							;\
600	SKIP(31-19-TT_TRACE_L_INS)				;\
601	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
602	.empty
603
604
605/*
606 * FILL_64bit fills a 64-bit-wide kernel register window.  It assumes
607 * that the kernel context and the nucleus context are the same.  The
608 * stack pointer is required to be eight-byte aligned.
609 */
610#define	FILL_64bit(tail)					\
6112:	TT_TRACE_L(trace_win)					;\
612	ldx	[%sp + V9BIAS64 + 0], %l0			;\
613	ldx	[%sp + V9BIAS64 + 8], %l1			;\
614	ldx	[%sp + V9BIAS64 + 16], %l2			;\
615	ldx	[%sp + V9BIAS64 + 24], %l3			;\
616	ldx	[%sp + V9BIAS64 + 32], %l4			;\
617	ldx	[%sp + V9BIAS64 + 40], %l5			;\
618	ldx	[%sp + V9BIAS64 + 48], %l6			;\
619	ldx	[%sp + V9BIAS64 + 56], %l7			;\
620	ldx	[%sp + V9BIAS64 + 64], %i0			;\
621	ldx	[%sp + V9BIAS64 + 72], %i1			;\
622	ldx	[%sp + V9BIAS64 + 80], %i2			;\
623	ldx	[%sp + V9BIAS64 + 88], %i3			;\
624	ldx	[%sp + V9BIAS64 + 96], %i4			;\
625	ldx	[%sp + V9BIAS64 + 104], %i5			;\
626	ldx	[%sp + V9BIAS64 + 112], %i6			;\
627	ldx	[%sp + V9BIAS64 + 120], %i7			;\
628	restored						;\
629	retry							;\
630	SKIP(31-18-TT_TRACE_L_INS)				;\
631	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
632	.empty
633
634/*
635 * FILL_64bit_asi fills a 64-bit-wide register window from a 64-bit
636 * wide address space via the designated asi.  It is used to fill
637 * non-kernel windows.  The stack pointer is required to be eight-byte
638 * aligned.
639 */
640#define	FILL_64bit_asi(asi_num, tail)				\
641	mov	V9BIAS64 + 0, %g1				;\
6422:	TT_TRACE_L(trace_win)					;\
643	ldxa	[%sp + %g1]asi_num, %l0				;\
644	mov	V9BIAS64 + 8, %g2				;\
645	ldxa	[%sp + %g2]asi_num, %l1				;\
646	mov	V9BIAS64 + 16, %g3				;\
647	ldxa	[%sp + %g3]asi_num, %l2				;\
648	mov	V9BIAS64 + 24, %g4				;\
649	ldxa	[%sp + %g4]asi_num, %l3				;\
650	add	%sp, 32, %g5					;\
651	ldxa	[%g5 + %g1]asi_num, %l4				;\
652	ldxa	[%g5 + %g2]asi_num, %l5				;\
653	ldxa	[%g5 + %g3]asi_num, %l6				;\
654	ldxa	[%g5 + %g4]asi_num, %l7				;\
655	add	%g5, 32, %g5					;\
656	ldxa	[%g5 + %g1]asi_num, %i0				;\
657	ldxa	[%g5 + %g2]asi_num, %i1				;\
658	ldxa	[%g5 + %g3]asi_num, %i2				;\
659	ldxa	[%g5 + %g4]asi_num, %i3				;\
660	add	%g5, 32, %g5					;\
661	ldxa	[%g5 + %g1]asi_num, %i4				;\
662	ldxa	[%g5 + %g2]asi_num, %i5				;\
663	ldxa	[%g5 + %g3]asi_num, %i6				;\
664	ldxa	[%g5 + %g4]asi_num, %i7				;\
665	restored						;\
666	retry							;\
667	SKIP(31-25-TT_TRACE_L_INS)				;\
668	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
669	.empty
670
671/*
672 * FILL_64bit_tt1 fills a 64-bit-wide register window from a 64-bit
673 * wide address space via the designated asi.  It is used to fill
674 * windows at tl>1 where performance isn't the primary concern and
675 * where we don't want to use unnecessary registers.  The stack
676 * pointer is required to be eight-byte aligned.
677 */
678#define	FILL_64bit_tt1(asi_num, tail)				\
679	mov	asi_num, %asi					;\
680	TT_TRACE_L(trace_win)					;\
681	ldxa	[%sp + V9BIAS64 + 0]%asi, %l0			;\
682	ldxa	[%sp + V9BIAS64 + 8]%asi, %l1			;\
683	ldxa	[%sp + V9BIAS64 + 16]%asi, %l2			;\
684	ldxa	[%sp + V9BIAS64 + 24]%asi, %l3			;\
685	ldxa	[%sp + V9BIAS64 + 32]%asi, %l4			;\
686	ldxa	[%sp + V9BIAS64 + 40]%asi, %l5			;\
687	ldxa	[%sp + V9BIAS64 + 48]%asi, %l6			;\
688	ldxa	[%sp + V9BIAS64 + 56]%asi, %l7			;\
689	ldxa	[%sp + V9BIAS64 + 64]%asi, %i0			;\
690	ldxa	[%sp + V9BIAS64 + 72]%asi, %i1			;\
691	ldxa	[%sp + V9BIAS64 + 80]%asi, %i2			;\
692	ldxa	[%sp + V9BIAS64 + 88]%asi, %i3			;\
693	ldxa	[%sp + V9BIAS64 + 96]%asi, %i4			;\
694	ldxa	[%sp + V9BIAS64 + 104]%asi, %i5			;\
695	ldxa	[%sp + V9BIAS64 + 112]%asi, %i6			;\
696	ldxa	[%sp + V9BIAS64 + 120]%asi, %i7			;\
697	restored						;\
698	retry							;\
699	SKIP(31-19-TT_TRACE_L_INS)				;\
700	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
701	.empty
702
703#endif /* !lint */
704
705/*
706 * SPILL_mixed spills either size window, depending on
707 * whether %sp is even or odd, to a 32-bit address space.
708 * This may only be used in conjunction with SPILL_32bit/
709 * SPILL_64bit. New versions of SPILL_mixed_{tt1,asi} would be
710 * needed for use with SPILL_{32,64}bit_{tt1,asi}.  Particular
711 * attention should be paid to the instructions that belong
712 * in the delay slots of the branches depending on the type
713 * of spill handler being branched to.
714 * Clear upper 32 bits of %sp if it is odd.
715 * We won't need to clear them in 64 bit kernel.
716 */
717#define	SPILL_mixed						\
718	btst	1, %sp						;\
719	bz,a,pt	%xcc, 1b					;\
720	srl	%sp, 0, %sp					;\
721	ba,pt	%xcc, 2b					;\
722	nop							;\
723	.align	128
724
725/*
726 * FILL_mixed(ASI) fills either size window, depending on
727 * whether %sp is even or odd, from a 32-bit address space.
728 * This may only be used in conjunction with FILL_32bit/
729 * FILL_64bit. New versions of FILL_mixed_{tt1,asi} would be
730 * needed for use with FILL_{32,64}bit_{tt1,asi}. Particular
731 * attention should be paid to the instructions that belong
732 * in the delay slots of the branches depending on the type
733 * of fill handler being branched to.
734 * Clear upper 32 bits of %sp if it is odd.
735 * We won't need to clear them in 64 bit kernel.
736 */
737#define	FILL_mixed						\
738	btst	1, %sp						;\
739	bz,a,pt	%xcc, 1b					;\
740	srl	%sp, 0, %sp					;\
741	ba,pt	%xcc, 2b					;\
742	nop							;\
743	.align	128
744
745
746/*
747 * SPILL_32clean/SPILL_64clean spill 32-bit and 64-bit register windows,
748 * respectively, into the address space via the designated asi.  The
749 * unbiased stack pointer is required to be eight-byte aligned (even for
750 * the 32-bit case even though this code does not require such strict
751 * alignment).
752 *
753 * With SPARC v9 the spill trap takes precedence over the cleanwin trap
754 * so when cansave == 0, canrestore == 6, and cleanwin == 6 the next save
755 * will cause cwp + 2 to be spilled but will not clean cwp + 1.  That
756 * window may contain kernel data so in user_rtt we set wstate to call
757 * these spill handlers on the first user spill trap.  These handler then
758 * spill the appropriate window but also back up a window and clean the
759 * window that didn't get a cleanwin trap.
760 */
761#define	SPILL_32clean(asi_num, tail)				\
762	srl	%sp, 0, %sp					;\
763	sta	%l0, [%sp + %g0]asi_num				;\
764	mov	4, %g1						;\
765	sta	%l1, [%sp + %g1]asi_num				;\
766	mov	8, %g2						;\
767	sta	%l2, [%sp + %g2]asi_num				;\
768	mov	12, %g3						;\
769	sta	%l3, [%sp + %g3]asi_num				;\
770	add	%sp, 16, %g4					;\
771	sta	%l4, [%g4 + %g0]asi_num				;\
772	sta	%l5, [%g4 + %g1]asi_num				;\
773	sta	%l6, [%g4 + %g2]asi_num				;\
774	sta	%l7, [%g4 + %g3]asi_num				;\
775	add	%g4, 16, %g4					;\
776	sta	%i0, [%g4 + %g0]asi_num				;\
777	sta	%i1, [%g4 + %g1]asi_num				;\
778	sta	%i2, [%g4 + %g2]asi_num				;\
779	sta	%i3, [%g4 + %g3]asi_num				;\
780	add	%g4, 16, %g4					;\
781	sta	%i4, [%g4 + %g0]asi_num				;\
782	sta	%i5, [%g4 + %g1]asi_num				;\
783	sta	%i6, [%g4 + %g2]asi_num				;\
784	sta	%i7, [%g4 + %g3]asi_num				;\
785	TT_TRACE_L(trace_win)					;\
786	b	.spill_clean					;\
787	  mov	WSTATE_USER32, %g7				;\
788	SKIP(31-25-TT_TRACE_L_INS)				;\
789	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
790	.empty
791
792#define	SPILL_64clean(asi_num, tail)				\
793	mov	0 + V9BIAS64, %g1				;\
794	stxa	%l0, [%sp + %g1]asi_num				;\
795	mov	8 + V9BIAS64, %g2				;\
796	stxa	%l1, [%sp + %g2]asi_num				;\
797	mov	16 + V9BIAS64, %g3				;\
798	stxa	%l2, [%sp + %g3]asi_num				;\
799	mov	24 + V9BIAS64, %g4				;\
800	stxa	%l3, [%sp + %g4]asi_num				;\
801	add	%sp, 32, %g5					;\
802	stxa	%l4, [%g5 + %g1]asi_num				;\
803	stxa	%l5, [%g5 + %g2]asi_num				;\
804	stxa	%l6, [%g5 + %g3]asi_num				;\
805	stxa	%l7, [%g5 + %g4]asi_num				;\
806	add	%g5, 32, %g5					;\
807	stxa	%i0, [%g5 + %g1]asi_num				;\
808	stxa	%i1, [%g5 + %g2]asi_num				;\
809	stxa	%i2, [%g5 + %g3]asi_num				;\
810	stxa	%i3, [%g5 + %g4]asi_num				;\
811	add	%g5, 32, %g5					;\
812	stxa	%i4, [%g5 + %g1]asi_num				;\
813	stxa	%i5, [%g5 + %g2]asi_num				;\
814	stxa	%i6, [%g5 + %g3]asi_num				;\
815	stxa	%i7, [%g5 + %g4]asi_num				;\
816	TT_TRACE_L(trace_win)					;\
817	b	.spill_clean					;\
818	  mov	WSTATE_USER64, %g7				;\
819	SKIP(31-25-TT_TRACE_L_INS)				;\
820	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
821	.empty
822
823
824/*
825 * Floating point disabled.
826 */
827#define	FP_DISABLED_TRAP		\
828	TT_TRACE(trace_gen)		;\
829	ba,pt	%xcc,.fp_disabled	;\
830	nop				;\
831	.align	32
832
833/*
834 * Floating point exceptions.
835 */
836#define	FP_IEEE_TRAP			\
837	TT_TRACE(trace_gen)		;\
838	ba,pt	%xcc,.fp_ieee_exception	;\
839	nop				;\
840	.align	32
841
842#define	FP_TRAP				\
843	TT_TRACE(trace_gen)		;\
844	ba,pt	%xcc,.fp_exception	;\
845	nop				;\
846	.align	32
847
848#if !defined(lint)
849/*
850 * asynchronous traps at level 0 and level 1
851 *
852 * The first instruction must be a membar for UltraSPARC-III
853 * to stop RED state entry if the store queue has many
854 * pending bad stores (PRM, Chapter 11).
855 */
856#define ASYNC_TRAP(ttype, ttlabel, table_name)\
857	.global	table_name	;\
858table_name:			;\
859	membar	#Sync		;\
860	TT_TRACE(ttlabel)	;\
861	ba	async_err	;\
862	mov	ttype, %g5	;\
863	.align	32
864
865/*
866 * Defaults to BAD entry, but establishes label to be used for
867 * architecture-specific overwrite of trap table entry.
868 */
869#define	LABELED_BAD(table_name)		\
870	.global	table_name		;\
871table_name:				;\
872	BAD
873
874#endif /* !lint */
875
876/*
877 * illegal instruction trap
878 */
879#define	ILLTRAP_INSTR			  \
880	membar	#Sync			  ;\
881	TT_TRACE(trace_gen)		  ;\
882	or	%g0, P_UTRAP4, %g2	  ;\
883	or	%g0, T_UNIMP_INSTR, %g3   ;\
884	sethi	%hi(.check_v9utrap), %g4  ;\
885	jmp	%g4 + %lo(.check_v9utrap) ;\
886	nop				  ;\
887	.align	32
888
889/*
890 * tag overflow trap
891 */
892#define	TAG_OVERFLOW			  \
893	TT_TRACE(trace_gen)		  ;\
894	or	%g0, P_UTRAP10, %g2	  ;\
895	or	%g0, T_TAG_OVERFLOW, %g3  ;\
896	sethi	%hi(.check_v9utrap), %g4  ;\
897	jmp	%g4 + %lo(.check_v9utrap) ;\
898	nop				  ;\
899	.align	32
900
901/*
902 * divide by zero trap
903 */
904#define	DIV_BY_ZERO			  \
905	TT_TRACE(trace_gen)		  ;\
906	or	%g0, P_UTRAP11, %g2	  ;\
907	or	%g0, T_IDIV0, %g3	  ;\
908	sethi	%hi(.check_v9utrap), %g4  ;\
909	jmp	%g4 + %lo(.check_v9utrap) ;\
910	nop				  ;\
911	.align	32
912
913/*
914 * trap instruction for V9 user trap handlers
915 */
916#define	TRAP_INSTR			  \
917	TT_TRACE(trace_gen)		  ;\
918	or	%g0, T_SOFTWARE_TRAP, %g3 ;\
919	sethi	%hi(.check_v9utrap), %g4  ;\
920	jmp	%g4 + %lo(.check_v9utrap) ;\
921	nop				  ;\
922	.align	32
923#define	TRP4	TRAP_INSTR; TRAP_INSTR; TRAP_INSTR; TRAP_INSTR
924
925/*
926 * LEVEL_INTERRUPT is for level N interrupts.
927 * VECTOR_INTERRUPT is for the vector trap.
928 */
929#define	LEVEL_INTERRUPT(level)		\
930	.global	tt_pil/**/level		;\
931tt_pil/**/level:			;\
932	ba,pt	%xcc, pil_interrupt	;\
933	mov	level, %g4		;\
934	.align	32
935
936#define	LEVEL14_INTERRUPT			\
937	ba	pil14_interrupt			;\
938	mov	PIL_14, %g4			;\
939	.align	32
940
941#define	VECTOR_INTERRUPT				\
942	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g1	;\
943	btst	IRSR_BUSY, %g1				;\
944	bnz,pt	%xcc, vec_interrupt			;\
945	nop						;\
946	ba,a,pt	%xcc, vec_intr_spurious			;\
947	.empty						;\
948	.align	32
949
950/*
951 * MMU Trap Handlers.
952 */
953#define	SWITCH_GLOBALS	/* mmu->alt, alt->mmu */			\
954	rdpr	%pstate, %g5						;\
955	wrpr	%g5, PSTATE_MG | PSTATE_AG, %pstate
956
957#define	IMMU_EXCEPTION							\
958	membar	#Sync							;\
959	SWITCH_GLOBALS							;\
960	wr	%g0, ASI_IMMU, %asi					;\
961	rdpr	%tpc, %g2						;\
962	ldxa	[MMU_SFSR]%asi, %g3					;\
963	ba,pt	%xcc, .mmu_exception_end				;\
964	mov	T_INSTR_EXCEPTION, %g1					;\
965	.align	32
966
967#define	DMMU_EXCEPTION							\
968	SWITCH_GLOBALS							;\
969	wr	%g0, ASI_DMMU, %asi					;\
970	ldxa	[MMU_TAG_ACCESS]%asi, %g2				;\
971	ldxa	[MMU_SFSR]%asi, %g3					;\
972	ba,pt	%xcc, .mmu_exception_end				;\
973	mov	T_DATA_EXCEPTION, %g1					;\
974	.align	32
975
976#define	DMMU_EXC_AG_PRIV						\
977	wr	%g0, ASI_DMMU, %asi					;\
978	ldxa	[MMU_SFAR]%asi, %g2					;\
979	ba,pt	%xcc, .mmu_priv_exception				;\
980	ldxa	[MMU_SFSR]%asi, %g3					;\
981	.align	32
982
983#define	DMMU_EXC_AG_NOT_ALIGNED						\
984	wr	%g0, ASI_DMMU, %asi					;\
985	ldxa	[MMU_SFAR]%asi, %g2					;\
986	ba,pt	%xcc, .mmu_exception_not_aligned			;\
987	ldxa	[MMU_SFSR]%asi, %g3					;\
988	.align	32
989
990/*
991 * SPARC V9 IMPL. DEP. #109(1) and (2) and #110(1) and (2)
992 */
993#define	DMMU_EXC_LDDF_NOT_ALIGNED					\
994	btst	1, %sp							;\
995	bnz,pt	%xcc, .lddf_exception_not_aligned			;\
996	wr	%g0, ASI_DMMU, %asi					;\
997	ldxa	[MMU_SFAR]%asi, %g2					;\
998	ba,pt	%xcc, .mmu_exception_not_aligned			;\
999	ldxa	[MMU_SFSR]%asi, %g3					;\
1000	.align	32
1001
1002#define	DMMU_EXC_STDF_NOT_ALIGNED					\
1003	btst	1, %sp							;\
1004	bnz,pt	%xcc, .stdf_exception_not_aligned			;\
1005	wr	%g0, ASI_DMMU, %asi					;\
1006	ldxa	[MMU_SFAR]%asi, %g2					;\
1007	ba,pt	%xcc, .mmu_exception_not_aligned			;\
1008	ldxa	[MMU_SFSR]%asi, %g3					;\
1009	.align	32
1010
1011/*
1012 * Flush the TLB using either the primary, secondary, or nucleus flush
1013 * operation based on whether the ctx from the tag access register matches
1014 * the primary or secondary context (flush the nucleus if neither matches).
1015 *
1016 * Requires a membar #Sync before next ld/st.
1017 * exits with:
1018 * g2 = tag access register
1019 * g3 = ctx number
1020 */
1021#if TAGACC_CTX_MASK != CTXREG_CTX_MASK
1022#error "TAGACC_CTX_MASK != CTXREG_CTX_MASK"
1023#endif
1024#define	DTLB_DEMAP_ENTRY						\
1025	mov	MMU_TAG_ACCESS, %g1					;\
1026	mov	MMU_PCONTEXT, %g5					;\
1027	ldxa	[%g1]ASI_DMMU, %g2					;\
1028	sethi	%hi(TAGACC_CTX_MASK), %g4				;\
1029	or	%g4, %lo(TAGACC_CTX_MASK), %g4				;\
1030	and	%g2, %g4, %g3			/* g3 = ctx */		;\
1031	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = primary ctx */	;\
1032	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1033	cmp	%g3, %g6						;\
1034	be,pt	%xcc, 1f						;\
1035	andn	%g2, %g4, %g1			/* ctx = primary */	;\
1036	mov	MMU_SCONTEXT, %g5					;\
1037	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = secondary ctx */ ;\
1038	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1039	cmp	%g3, %g6						;\
1040	be,a,pt	%xcc, 1f						;\
1041	  or	%g1, DEMAP_SECOND, %g1					;\
1042	or	%g1, DEMAP_NUCLEUS, %g1					;\
10431:	stxa	%g0, [%g1]ASI_DTLB_DEMAP	/* MMU_DEMAP_PAGE */	;\
1044	membar	#Sync
1045
1046#if defined(cscope)
1047/*
1048 * Define labels to direct cscope quickly to labels that
1049 * are generated by macro expansion of DTLB_MISS().
1050 */
1051	.global	tt0_dtlbmiss
1052tt0_dtlbmiss:
1053	.global	tt1_dtlbmiss
1054tt1_dtlbmiss:
1055	nop
1056#endif
1057
1058/*
1059 * Needs to be exactly 32 instructions
1060 *
1061 * UTLB NOTE: If we don't hit on the 8k pointer then we branch
1062 * to a special 4M tsb handler. It would be nice if that handler
1063 * could live in this file but currently it seems better to allow
1064 * it to fall thru to sfmmu_tsb_miss.
1065 */
1066#ifdef UTSB_PHYS
1067#define	DTLB_MISS(table_name)						;\
1068	.global	table_name/**/_dtlbmiss					;\
1069table_name/**/_dtlbmiss:						;\
1070	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1071	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1072	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1073	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1074	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1075	cmp	%g3, INVALID_CONTEXT					;\
1076	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1077	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1078	mov	SCRATCHPAD_UTSBREG, %g3				;\
1079	ldxa	[%g3]ASI_SCRATCHPAD, %g3	/* g3 = 2nd tsb reg */	;\
1080	brgez,pn %g3, sfmmu_udtlb_slowpath	/* branch if 2 TSBs */	;\
1081	  nop								;\
1082	ldda	[%g1]ASI_QUAD_LDD_PHYS, %g4	/* g4 = tag, %g5 data */;\
1083	cmp	%g4, %g7						;\
1084	bne,pn	%xcc, sfmmu_tsb_miss_tt		/* no 4M TSB, miss */	;\
1085	  mov	-1, %g3			/* set 4M tsbe ptr to -1 */	;\
1086	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1087	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1088	retry				/* in %g5 */			;\
1089	unimp	0							;\
1090	unimp	0							;\
1091	unimp	0							;\
1092	unimp	0							;\
1093	unimp	0							;\
1094	unimp	0							;\
1095	unimp	0							;\
1096	unimp	0							;\
1097	unimp	0							;\
1098        unimp   0                                                       ;\
1099        unimp   0                                                       ;\
1100        unimp   0                                                       ;\
1101	.align 128
1102#else /* UTSB_PHYS */
1103#define	DTLB_MISS(table_name)						;\
1104	.global	table_name/**/_dtlbmiss					;\
1105table_name/**/_dtlbmiss:						;\
1106	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1107	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1108	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1109	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1110	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1111	cmp	%g3, INVALID_CONTEXT					;\
1112	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1113	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1114	brlz,pn %g1, sfmmu_udtlb_slowpath				;\
1115	  nop								;\
1116	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, %g5 data */	;\
1117	cmp	%g4, %g7						;\
1118	bne,pn	%xcc, sfmmu_tsb_miss_tt		/* no 4M TSB, miss */	;\
1119	  mov	-1, %g3		/* set 4M tsbe ptr to -1 */		;\
1120	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1121	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1122	retry				/* in %g5 */			;\
1123	unimp	0							;\
1124	unimp	0							;\
1125	unimp	0							;\
1126	unimp	0							;\
1127	unimp	0							;\
1128        unimp   0                                                       ;\
1129        unimp   0                                                       ;\
1130        unimp   0                                                       ;\
1131        unimp   0                                                       ;\
1132        unimp   0                                                       ;\
1133        unimp   0                                                       ;\
1134        unimp   0                                                       ;\
1135        unimp   0                                                       ;\
1136        unimp   0                                                       ;\
1137	.align 128
1138#endif /* UTSB_PHYS */
1139
1140#if defined(cscope)
1141/*
1142 * Define labels to direct cscope quickly to labels that
1143 * are generated by macro expansion of ITLB_MISS().
1144 */
1145	.global	tt0_itlbmiss
1146tt0_itlbmiss:
1147	.global	tt1_itlbmiss
1148tt1_itlbmiss:
1149	nop
1150#endif
1151
1152/*
1153 * Instruction miss handler.
1154 * ldda instructions will have their ASI patched
1155 * by sfmmu_patch_ktsb at runtime.
1156 * MUST be EXACTLY 32 instructions or we'll break.
1157 */
1158#ifdef UTSB_PHYS
1159#define	ITLB_MISS(table_name)						 \
1160	.global	table_name/**/_itlbmiss					;\
1161table_name/**/_itlbmiss:						;\
1162	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1163	ldxa	[%g0]ASI_IMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1164	ldxa	[%g6]ASI_IMMU, %g2		/* g2 = tag access */	;\
1165	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1166	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1167	cmp	%g3, INVALID_CONTEXT					;\
1168	ble,pn	%xcc, sfmmu_kitlb_miss					;\
1169	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1170	mov	SCRATCHPAD_UTSBREG, %g3				;\
1171	ldxa	[%g3]ASI_SCRATCHPAD, %g3	/* g3 = 2nd tsb reg */	;\
1172	brgez,pn %g3, sfmmu_uitlb_slowpath	/* branch if 2 TSBs */	;\
1173	  nop								;\
1174	ldda	[%g1]ASI_QUAD_LDD_PHYS, %g4 /* g4 = tag, g5 = data */	;\
1175	cmp	%g4, %g7						;\
1176	bne,pn	%xcc, sfmmu_tsb_miss_tt	/* br if 8k ptr miss */		;\
1177	  mov	-1, %g3		/* set 4M TSB ptr to -1 */		;\
1178	andcc	%g5, TTE_EXECPRM_INT, %g0 /* check execute bit */	;\
1179	bz,pn	%icc, exec_fault					;\
1180	  nop								;\
1181	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1182	stxa	%g5, [%g0]ASI_ITLB_IN	/* trapstat expects %g5 */	;\
1183	retry								;\
1184	unimp	0							;\
1185	unimp	0							;\
1186	unimp	0							;\
1187	unimp	0							;\
1188	unimp	0							;\
1189	unimp	0							;\
1190        unimp   0                                                       ;\
1191        unimp   0                                                       ;\
1192        unimp   0                                                       ;\
1193	.align 128
1194#else /* UTSB_PHYS */
1195#define	ITLB_MISS(table_name)						 \
1196	.global	table_name/**/_itlbmiss					;\
1197table_name/**/_itlbmiss:						;\
1198	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1199	ldxa	[%g0]ASI_IMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1200	ldxa	[%g6]ASI_IMMU, %g2		/* g2 = tag access */	;\
1201	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1202	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1203	cmp	%g3, INVALID_CONTEXT					;\
1204	ble,pn	%xcc, sfmmu_kitlb_miss					;\
1205	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1206	brlz,pn	%g1, sfmmu_uitlb_slowpath	/* if >1 TSB branch */	;\
1207	  nop								;\
1208	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, g5 = data */	;\
1209	cmp	%g4, %g7						;\
1210	bne,pn	%xcc, sfmmu_tsb_miss_tt	/* br if 8k ptr miss */		;\
1211	  mov	-1, %g3		/* set 4M TSB ptr to -1 */		;\
1212	andcc	%g5, TTE_EXECPRM_INT, %g0 /* check execute bit */	;\
1213	bz,pn	%icc, exec_fault					;\
1214	  nop								;\
1215	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1216	stxa	%g5, [%g0]ASI_ITLB_IN	/* trapstat expects %g5 */	;\
1217	retry								;\
1218	unimp	0							;\
1219	unimp	0							;\
1220	unimp	0							;\
1221	unimp	0							;\
1222	unimp	0							;\
1223        unimp   0                                                       ;\
1224        unimp   0                                                       ;\
1225        unimp   0                                                       ;\
1226        unimp   0                                                       ;\
1227        unimp   0                                                       ;\
1228        unimp   0                                                       ;\
1229	.align 128
1230#endif /* UTSB_PHYS */
1231
1232
1233/*
1234 * This macro is the first level handler for fast protection faults.
1235 * It first demaps the tlb entry which generated the fault and then
1236 * attempts to set the modify bit on the hash.  It needs to be
1237 * exactly 32 instructions.
1238 */
1239#define	DTLB_PROT							 \
1240	DTLB_DEMAP_ENTRY		/* 20 instructions */		;\
1241	/*								;\
1242	 * At this point:						;\
1243	 *   g1 = ????							;\
1244	 *   g2 = tag access register					;\
1245	 *   g3 = ctx number						;\
1246	 *   g4 = ????							;\
1247	 */								;\
1248	TT_TRACE(trace_dataprot)	/* 2 instr ifdef TRAPTRACE */	;\
1249					/* clobbers g1 and g6 */	;\
1250	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1251	brnz,pt %g3, sfmmu_uprot_trap		/* user trap */		;\
1252	  nop								;\
1253	ba,a,pt	%xcc, sfmmu_kprot_trap		/* kernel trap */	;\
1254	unimp	0							;\
1255	unimp	0							;\
1256	unimp	0							;\
1257	unimp	0							;\
1258	unimp	0							;\
1259	unimp	0							;\
1260	.align 128
1261
1262#define	DMMU_EXCEPTION_TL1						;\
1263	SWITCH_GLOBALS							;\
1264	ba,a,pt	%xcc, mmu_trap_tl1					;\
1265	  nop								;\
1266	.align 32
1267
1268#define	MISALIGN_ADDR_TL1						;\
1269	ba,a,pt	%xcc, mmu_trap_tl1					;\
1270	  nop								;\
1271	.align 32
1272
1273/*
1274 * Trace a tsb hit
1275 * g1 = tsbe pointer (in/clobbered)
1276 * g2 = tag access register (in)
1277 * g3 - g4 = scratch (clobbered)
1278 * g5 = tsbe data (in)
1279 * g6 = scratch (clobbered)
1280 * g7 = pc we jumped here from (in)
1281 * ttextra = value to OR in to trap type (%tt) (in)
1282 */
1283#ifdef TRAPTRACE
1284#define TRACE_TSBHIT(ttextra)						 \
1285	membar	#Sync							;\
1286	sethi	%hi(FLUSH_ADDR), %g6					;\
1287	flush	%g6							;\
1288	TRACE_PTR(%g3, %g6)						;\
1289	GET_TRACE_TICK(%g6)						;\
1290	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi				;\
1291	stxa	%g2, [%g3 + TRAP_ENT_SP]%asi	/* tag access */	;\
1292	stxa	%g5, [%g3 + TRAP_ENT_F1]%asi	/* tsb data */		;\
1293	rdpr	%tnpc, %g6						;\
1294	stxa	%g6, [%g3 + TRAP_ENT_F2]%asi				;\
1295	stxa	%g1, [%g3 + TRAP_ENT_F3]%asi	/* tsb pointer */	;\
1296	stxa	%g0, [%g3 + TRAP_ENT_F4]%asi				;\
1297	rdpr	%tpc, %g6						;\
1298	stxa	%g6, [%g3 + TRAP_ENT_TPC]%asi				;\
1299	rdpr	%tl, %g6						;\
1300	stha	%g6, [%g3 + TRAP_ENT_TL]%asi				;\
1301	rdpr	%tt, %g6						;\
1302	or	%g6, (ttextra), %g6					;\
1303	stha	%g6, [%g3 + TRAP_ENT_TT]%asi				;\
1304	ldxa	[%g0]ASI_IMMU, %g1		/* tag target */	;\
1305	ldxa	[%g0]ASI_DMMU, %g4					;\
1306	cmp	%g6, FAST_IMMU_MISS_TT					;\
1307	movne	%icc, %g4, %g1						;\
1308	stxa	%g1, [%g3 + TRAP_ENT_TSTATE]%asi /* tsb tag */		;\
1309	stxa	%g0, [%g3 + TRAP_ENT_TR]%asi				;\
1310	TRACE_NEXT(%g3, %g4, %g6)
1311#else
1312#define TRACE_TSBHIT(ttextra)
1313#endif
1314
1315#if defined(lint)
1316
1317struct scb	trap_table;
1318struct scb	scb;		/* trap_table/scb are the same object */
1319
1320#else /* lint */
1321
1322/*
1323 * =======================================================================
1324 *		SPARC V9 TRAP TABLE
1325 *
1326 * The trap table is divided into two halves: the first half is used when
1327 * taking traps when TL=0; the second half is used when taking traps from
1328 * TL>0. Note that handlers in the second half of the table might not be able
1329 * to make the same assumptions as handlers in the first half of the table.
1330 *
1331 * Worst case trap nesting so far:
1332 *
1333 *	at TL=0 client issues software trap requesting service
1334 *	at TL=1 nucleus wants a register window
1335 *	at TL=2 register window clean/spill/fill takes a TLB miss
1336 *	at TL=3 processing TLB miss
1337 *	at TL=4 handle asynchronous error
1338 *
1339 * Note that a trap from TL=4 to TL=5 places Spitfire in "RED mode".
1340 *
1341 * =======================================================================
1342 */
1343	.section ".text"
1344	.align	4
1345	.global trap_table, scb, trap_table0, trap_table1, etrap_table
1346	.type	trap_table, #object
1347	.type	scb, #object
1348trap_table:
1349scb:
1350trap_table0:
1351	/* hardware traps */
1352	NOT;				/* 000	reserved */
1353	RED;				/* 001	power on reset */
1354	RED;				/* 002	watchdog reset */
1355	RED;				/* 003	externally initiated reset */
1356	RED;				/* 004	software initiated reset */
1357	RED;				/* 005	red mode exception */
1358	NOT; NOT;			/* 006 - 007 reserved */
1359	IMMU_EXCEPTION;			/* 008	instruction access exception */
1360	NOT;				/* 009	instruction access MMU miss */
1361	ASYNC_TRAP(T_INSTR_ERROR, trace_gen, tt0_iae);
1362					/* 00A	instruction access error */
1363	NOT; NOT4;			/* 00B - 00F reserved */
1364	ILLTRAP_INSTR;			/* 010	illegal instruction */
1365	TRAP(T_PRIV_INSTR);		/* 011	privileged opcode */
1366	NOT;				/* 012	unimplemented LDD */
1367	NOT;				/* 013	unimplemented STD */
1368	NOT4; NOT4; NOT4;		/* 014 - 01F reserved */
1369	FP_DISABLED_TRAP;		/* 020	fp disabled */
1370	FP_IEEE_TRAP;			/* 021	fp exception ieee 754 */
1371	FP_TRAP;			/* 022	fp exception other */
1372	TAG_OVERFLOW;			/* 023	tag overflow */
1373	CLEAN_WINDOW;			/* 024 - 027 clean window */
1374	DIV_BY_ZERO;			/* 028	division by zero */
1375	NOT;				/* 029	internal processor error */
1376	NOT; NOT; NOT4;			/* 02A - 02F reserved */
1377	DMMU_EXCEPTION;			/* 030	data access exception */
1378	NOT;				/* 031	data access MMU miss */
1379	ASYNC_TRAP(T_DATA_ERROR, trace_gen, tt0_dae);
1380					/* 032	data access error */
1381	NOT;				/* 033	data access protection */
1382	DMMU_EXC_AG_NOT_ALIGNED;	/* 034	mem address not aligned */
1383	DMMU_EXC_LDDF_NOT_ALIGNED;	/* 035	LDDF mem address not aligned */
1384	DMMU_EXC_STDF_NOT_ALIGNED;	/* 036	STDF mem address not aligned */
1385	DMMU_EXC_AG_PRIV;		/* 037	privileged action */
1386	NOT;				/* 038	LDQF mem address not aligned */
1387	NOT;				/* 039	STQF mem address not aligned */
1388	NOT; NOT; NOT4;			/* 03A - 03F reserved */
1389	LABELED_BAD(tt0_asdat);		/* 040	async data error */
1390	LEVEL_INTERRUPT(1);		/* 041	interrupt level 1 */
1391	LEVEL_INTERRUPT(2);		/* 042	interrupt level 2 */
1392	LEVEL_INTERRUPT(3);		/* 043	interrupt level 3 */
1393	LEVEL_INTERRUPT(4);		/* 044	interrupt level 4 */
1394	LEVEL_INTERRUPT(5);		/* 045	interrupt level 5 */
1395	LEVEL_INTERRUPT(6);		/* 046	interrupt level 6 */
1396	LEVEL_INTERRUPT(7);		/* 047	interrupt level 7 */
1397	LEVEL_INTERRUPT(8);		/* 048	interrupt level 8 */
1398	LEVEL_INTERRUPT(9);		/* 049	interrupt level 9 */
1399	LEVEL_INTERRUPT(10);		/* 04A	interrupt level 10 */
1400	LEVEL_INTERRUPT(11);		/* 04B	interrupt level 11 */
1401	LEVEL_INTERRUPT(12);		/* 04C	interrupt level 12 */
1402	LEVEL_INTERRUPT(13);		/* 04D	interrupt level 13 */
1403	LEVEL14_INTERRUPT;		/* 04E	interrupt level 14 */
1404	LEVEL_INTERRUPT(15);		/* 04F	interrupt level 15 */
1405	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F reserved */
1406	VECTOR_INTERRUPT;		/* 060	interrupt vector */
1407	GOTO(kmdb_trap);		/* 061	PA watchpoint */
1408	GOTO(kmdb_trap);		/* 062	VA watchpoint */
1409	GOTO_TT(ce_err, trace_gen);	/* 063	corrected ECC error */
1410	ITLB_MISS(tt0);			/* 064	instruction access MMU miss */
1411	DTLB_MISS(tt0);			/* 068	data access MMU miss */
1412	DTLB_PROT;			/* 06C	data access protection */
1413	LABELED_BAD(tt0_fecc);		/* 070  fast ecache ECC error */
1414	LABELED_BAD(tt0_dperr);		/* 071  Cheetah+ dcache parity error */
1415	LABELED_BAD(tt0_iperr);		/* 072  Cheetah+ icache parity error */
1416	NOT;				/* 073  reserved */
1417	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1418	NOT4;				/* 080	spill 0 normal */
1419	SPILL_32bit_asi(ASI_AIUP,sn0);	/* 084	spill 1 normal */
1420	SPILL_64bit_asi(ASI_AIUP,sn0);	/* 088	spill 2 normal */
1421	SPILL_32clean(ASI_AIUP,sn0);	/* 08C	spill 3 normal */
1422	SPILL_64clean(ASI_AIUP,sn0);	/* 090	spill 4 normal */
1423	SPILL_32bit(not);		/* 094	spill 5 normal */
1424	SPILL_64bit(not);		/* 098	spill 6 normal */
1425	SPILL_mixed;			/* 09C	spill 7 normal */
1426	NOT4;				/* 0A0	spill 0 other */
1427	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0A4	spill 1 other */
1428	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0A8	spill 2 other */
1429	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0AC	spill 3 other */
1430	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0B0	spill 4 other */
1431	NOT4;				/* 0B4	spill 5 other */
1432	NOT4;				/* 0B8	spill 6 other */
1433	NOT4;				/* 0BC	spill 7 other */
1434	NOT4;				/* 0C0	fill 0 normal */
1435	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0C4	fill 1 normal */
1436	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0C8	fill 2 normal */
1437	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0CC	fill 3 normal */
1438	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0D0	fill 4 normal */
1439	FILL_32bit(not);		/* 0D4	fill 5 normal */
1440	FILL_64bit(not);		/* 0D8	fill 6 normal */
1441	FILL_mixed;			/* 0DC	fill 7 normal */
1442	NOT4;				/* 0E0	fill 0 other */
1443	NOT4;				/* 0E4	fill 1 other */
1444	NOT4;				/* 0E8	fill 2 other */
1445	NOT4;				/* 0EC	fill 3 other */
1446	NOT4;				/* 0F0	fill 4 other */
1447	NOT4;				/* 0F4	fill 5 other */
1448	NOT4;				/* 0F8	fill 6 other */
1449	NOT4;				/* 0FC	fill 7 other */
1450	/* user traps */
1451	GOTO(syscall_trap_4x);		/* 100	old system call */
1452	TRAP(T_BREAKPOINT);		/* 101	user breakpoint */
1453	TRAP(T_DIV0);			/* 102	user divide by zero */
1454	FLUSHW();			/* 103	flush windows */
1455	GOTO(.clean_windows);		/* 104	clean windows */
1456	BAD;				/* 105	range check ?? */
1457	GOTO(.fix_alignment);		/* 106	do unaligned references */
1458	BAD;				/* 107	unused */
1459	SYSCALL_TRAP32;			/* 108	ILP32 system call on LP64 */
1460	GOTO(set_trap0_addr);		/* 109	set trap0 address */
1461	BAD; BAD; BAD4;			/* 10A - 10F unused */
1462	TRP4; TRP4; TRP4; TRP4;		/* 110 - 11F V9 user trap handlers */
1463	GOTO(.getcc);			/* 120	get condition codes */
1464	GOTO(.setcc);			/* 121	set condition codes */
1465	GOTO(.getpsr);			/* 122	get psr */
1466	GOTO(.setpsr);			/* 123	set psr (some fields) */
1467	GOTO(get_timestamp);		/* 124	get timestamp */
1468	GOTO(get_virtime);		/* 125	get lwp virtual time */
1469	PRIV(self_xcall);		/* 126	self xcall */
1470	GOTO(get_hrestime);		/* 127	get hrestime */
1471	BAD;				/* 128	ST_SETV9STACK */
1472	GOTO(.getlgrp);			/* 129  get lgrpid */
1473	BAD; BAD; BAD4;			/* 12A - 12F unused */
1474	BAD4; BAD4; 			/* 130 - 137 unused */
1475	DTRACE_PID;			/* 138  dtrace pid tracing provider */
1476	BAD;				/* 139  unused */
1477	DTRACE_RETURN;			/* 13A	dtrace pid return probe */
1478	BAD; BAD4;			/* 13B - 13F unused */
1479	SYSCALL_TRAP;			/* 140  LP64 system call */
1480	SYSCALL(nosys);			/* 141  unused system call trap */
1481#ifdef DEBUG_USER_TRAPTRACECTL
1482	GOTO(.traptrace_freeze);	/* 142  freeze traptrace */
1483	GOTO(.traptrace_unfreeze);	/* 143  unfreeze traptrace */
1484#else
1485	SYSCALL(nosys);			/* 142  unused system call trap */
1486	SYSCALL(nosys);			/* 143  unused system call trap */
1487#endif
1488	BAD4; BAD4; BAD4;		/* 144 - 14F unused */
1489	BAD4; BAD4; BAD4; BAD4;		/* 150 - 15F unused */
1490	BAD4; BAD4; BAD4; BAD4;		/* 160 - 16F unused */
1491	BAD;				/* 170 - unused */
1492	BAD;				/* 171 - unused */
1493	BAD; BAD;			/* 172 - 173 unused */
1494	BAD4; BAD4;			/* 174 - 17B unused */
1495#ifdef	PTL1_PANIC_DEBUG
1496	mov PTL1_BAD_DEBUG, %g1; GOTO(ptl1_panic);
1497					/* 17C	test ptl1_panic */
1498#else
1499	BAD;				/* 17C  unused */
1500#endif	/* PTL1_PANIC_DEBUG */
1501	PRIV(kmdb_trap);		/* 17D	kmdb enter (L1-A) */
1502	PRIV(kmdb_trap);		/* 17E	kmdb breakpoint */
1503	PRIV(kctx_obp_bpt);		/* 17F	obp breakpoint */
1504	/* reserved */
1505	NOT4; NOT4; NOT4; NOT4;		/* 180 - 18F reserved */
1506	NOT4; NOT4; NOT4; NOT4;		/* 190 - 19F reserved */
1507	NOT4; NOT4; NOT4; NOT4;		/* 1A0 - 1AF reserved */
1508	NOT4; NOT4; NOT4; NOT4;		/* 1B0 - 1BF reserved */
1509	NOT4; NOT4; NOT4; NOT4;		/* 1C0 - 1CF reserved */
1510	NOT4; NOT4; NOT4; NOT4;		/* 1D0 - 1DF reserved */
1511	NOT4; NOT4; NOT4; NOT4;		/* 1E0 - 1EF reserved */
1512	NOT4; NOT4; NOT4; NOT4;		/* 1F0 - 1FF reserved */
1513trap_table1:
1514	NOT4; NOT4; NOT; NOT;		/* 000 - 009 unused */
1515	ASYNC_TRAP(T_INSTR_ERROR + T_TL1, trace_gen, tt1_iae);
1516					/* 00A	instruction access error */
1517	NOT; NOT4;			/* 00B - 00F unused */
1518	NOT4; NOT4; NOT4; NOT4;		/* 010 - 01F unused */
1519	NOT4;				/* 020 - 023 unused */
1520	CLEAN_WINDOW;			/* 024 - 027 clean window */
1521	NOT4; NOT4;			/* 028 - 02F unused */
1522	DMMU_EXCEPTION_TL1;		/* 030 	data access exception */
1523	NOT;				/* 031 unused */
1524	ASYNC_TRAP(T_DATA_ERROR + T_TL1, trace_gen, tt1_dae);
1525					/* 032	data access error */
1526	NOT;				/* 033	unused */
1527	MISALIGN_ADDR_TL1;		/* 034	mem address not aligned */
1528	NOT; NOT; NOT; NOT4; NOT4	/* 035 - 03F unused */
1529	LABELED_BAD(tt1_asdat);		/* 040	async data error */
1530	NOT; NOT; NOT;			/* 041 - 043 unused */
1531	NOT4; NOT4; NOT4;		/* 044 - 04F unused */
1532	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F unused */
1533	NOT;				/* 060	unused */
1534	GOTO(kmdb_trap_tl1);		/* 061	PA watchpoint */
1535	GOTO(kmdb_trap_tl1);		/* 062	VA watchpoint */
1536	GOTO_TT(ce_err_tl1, trace_gen);	/* 063	corrected ECC error */
1537	ITLB_MISS(tt1);			/* 064	instruction access MMU miss */
1538	DTLB_MISS(tt1);			/* 068	data access MMU miss */
1539	DTLB_PROT;			/* 06C	data access protection */
1540	LABELED_BAD(tt1_fecc);		/* 070  fast ecache ECC error */
1541	LABELED_BAD(tt1_dperr);		/* 071  Cheetah+ dcache parity error */
1542	LABELED_BAD(tt1_iperr);		/* 072  Cheetah+ icache parity error */
1543	NOT;				/* 073  reserved */
1544	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1545	NOT4;				/* 080	spill 0 normal */
1546	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 084	spill 1 normal */
1547	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 088	spill 2 normal */
1548	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 08C	spill 3 normal */
1549	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 090	spill 4 normal */
1550	SPILL_32bit(not);		/* 094	spill 5 normal */
1551	SPILL_64bit(not);		/* 098	spill 6 normal */
1552	SPILL_mixed;			/* 09C	spill 7 normal */
1553	NOT4;				/* 0A0	spill 0 other */
1554	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0A4	spill 1 other */
1555	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0A8	spill 2 other */
1556	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0AC	spill 3 other */
1557	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0B0  spill 4 other */
1558	NOT4;				/* 0B4  spill 5 other */
1559	NOT4;				/* 0B8  spill 6 other */
1560	NOT4;				/* 0BC  spill 7 other */
1561	NOT4;				/* 0C0	fill 0 normal */
1562	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0C4	fill 1 normal */
1563	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0C8	fill 2 normal */
1564	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0CC	fill 3 normal */
1565	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0D0	fill 4 normal */
1566	FILL_32bit(not);		/* 0D4	fill 5 normal */
1567	FILL_64bit(not);		/* 0D8	fill 6 normal */
1568	FILL_mixed;			/* 0DC	fill 7 normal */
1569	NOT4; NOT4; NOT4; NOT4;		/* 0E0 - 0EF unused */
1570	NOT4; NOT4; NOT4; NOT4;		/* 0F0 - 0FF unused */
1571	LABELED_BAD(tt1_swtrap0);	/* 100  fast ecache ECC error (cont) */
1572	LABELED_BAD(tt1_swtrap1);	/* 101  Ch+ D$ parity error (cont) */
1573	LABELED_BAD(tt1_swtrap2);	/* 102  Ch+ I$ parity error (cont) */
1574	NOT;				/* 103  reserved */
1575/*
1576 * We only reserve the above four special case soft traps for code running
1577 * at TL>0, so we can truncate the trap table here.
1578 */
1579etrap_table:
1580	.size	trap_table, (.-trap_table)
1581	.size	scb, (.-scb)
1582
1583/*
1584 * We get to exec_fault in the case of an instruction miss and tte
1585 * has no execute bit set.  We go to tl0 to handle it.
1586 *
1587 * g1 = tsbe pointer (in/clobbered)
1588 * g2 = tag access register (in)
1589 * g3 - g4 = scratch (clobbered)
1590 * g5 = tsbe data (in)
1591 * g6 = scratch (clobbered)
1592 */
1593	ALTENTRY(exec_fault)
1594	TRACE_TSBHIT(0x200)
1595	SWITCH_GLOBALS
1596	mov	MMU_TAG_ACCESS, %g4
1597	ldxa	[%g4]ASI_IMMU, %g2			! arg1 = addr
1598	mov	T_INSTR_MMU_MISS, %g3			! arg2 = traptype
1599	set	trap, %g1
1600	ba,pt	%xcc, sys_trap
1601	  mov	-1, %g4
1602
1603.mmu_exception_not_aligned:
1604	rdpr	%tstate, %g1
1605	btst	TSTATE_PRIV, %g1
1606	bnz,pn	%icc, 2f
1607	nop
1608	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1609	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1610	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1611	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1612	brz,pt	%g5, 2f
1613	nop
1614	ldn	[%g5 + P_UTRAP15], %g5			! unaligned utrap?
1615	brz,pn	%g5, 2f
1616	nop
1617	btst	1, %sp
1618	bz,pt	%xcc, 1f				! 32 bit user program
1619	nop
1620	ba,pt	%xcc, .setup_v9utrap			! 64 bit user program
1621	nop
16221:
1623	ba,pt	%xcc, .setup_utrap
1624	or	%g2, %g0, %g7
16252:
1626	ba,pt	%xcc, .mmu_exception_end
1627	mov	T_ALIGNMENT, %g1
1628
1629.mmu_priv_exception:
1630	rdpr	%tstate, %g1
1631	btst	TSTATE_PRIV, %g1
1632	bnz,pn	%icc, 1f
1633	nop
1634	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1635	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1636	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1637	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1638	brz,pt	%g5, 1f
1639	nop
1640	ldn	[%g5 + P_UTRAP16], %g5
1641	brnz,pt	%g5, .setup_v9utrap
1642	nop
16431:
1644	mov	T_PRIV_INSTR, %g1
1645
1646.mmu_exception_end:
1647	CPU_INDEX(%g4, %g5)
1648	set	cpu_core, %g5
1649	sllx	%g4, CPU_CORE_SHIFT, %g4
1650	add	%g4, %g5, %g4
1651	lduh	[%g4 + CPUC_DTRACE_FLAGS], %g5
1652	andcc	%g5, CPU_DTRACE_NOFAULT, %g0
1653	bz	%xcc, .mmu_exception_tlb_chk
1654	or	%g5, CPU_DTRACE_BADADDR, %g5
1655	stuh	%g5, [%g4 + CPUC_DTRACE_FLAGS]
1656	done
1657
1658.mmu_exception_tlb_chk:
1659	GET_CPU_IMPL(%g5)			! check SFSR.FT to see if this
1660	cmp	%g5, PANTHER_IMPL		! is a TLB parity error. But
1661	bne	2f				! we only do this check while
1662	mov	1, %g4				! running on Panther CPUs
1663	sllx	%g4, PN_SFSR_PARITY_SHIFT, %g4	! since US-I/II use the same
1664	andcc	%g3, %g4, %g0			! bit for something else which
1665	bz	2f				! will be handled later.
1666	nop
1667.mmu_exception_is_tlb_parity:
1668	.weak itlb_parity_trap
1669	.weak dtlb_parity_trap
1670	set	itlb_parity_trap, %g4
1671	cmp	%g1, T_INSTR_EXCEPTION		! branch to the itlb or
1672	be	3f				! dtlb parity handler
1673	nop					! if this trap is due
1674	set	dtlb_parity_trap, %g4
1675	cmp	%g1, T_DATA_EXCEPTION		! to a IMMU exception
1676	be	3f				! or DMMU exception.
1677	nop
16782:
1679	sllx	%g3, 32, %g3
1680	or	%g3, %g1, %g3
1681	set	trap, %g1
1682	ba,pt	%xcc, sys_trap
1683	sub	%g0, 1, %g4
16843:
1685	jmp	%g4				! off to the appropriate
1686	nop					! TLB parity handler
1687
1688.fp_disabled:
1689	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1690	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1691#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1692	brz,a,pn %g1, 2f
1693	  nop
1694#endif
1695	rdpr	%tstate, %g4
1696	btst	TSTATE_PRIV, %g4
1697#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1698	bnz,pn %icc, 2f
1699	  nop
1700#else
1701	bnz,a,pn %icc, ptl1_panic
1702	  mov	PTL1_BAD_FPTRAP, %g1
1703#endif
1704	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1705	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1706	brz,a,pt %g5, 2f
1707	  nop
1708	ldn	[%g5 + P_UTRAP7], %g5			! fp_disabled utrap?
1709	brz,a,pn %g5, 2f
1710	  nop
1711	btst	1, %sp
1712	bz,a,pt	%xcc, 1f				! 32 bit user program
1713	  nop
1714	ba,a,pt	%xcc, .setup_v9utrap			! 64 bit user program
1715	  nop
17161:
1717	ba,pt	%xcc, .setup_utrap
1718	  or	%g0, %g0, %g7
17192:
1720	set	fp_disabled, %g1
1721	ba,pt	%xcc, sys_trap
1722	  sub	%g0, 1, %g4
1723
1724.fp_ieee_exception:
1725	rdpr	%tstate, %g1
1726	btst	TSTATE_PRIV, %g1
1727	bnz,a,pn %icc, ptl1_panic
1728	  mov	PTL1_BAD_FPTRAP, %g1
1729	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1730	stx	%fsr, [%g1 + CPU_TMP1]
1731	ldx	[%g1 + CPU_TMP1], %g2
1732	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1733	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1734	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1735	brz,a,pt %g5, 1f
1736	  nop
1737	ldn	[%g5 + P_UTRAP8], %g5
1738	brnz,a,pt %g5, .setup_v9utrap
1739	  nop
17401:
1741	set	_fp_ieee_exception, %g1
1742	ba,pt	%xcc, sys_trap
1743	  sub	%g0, 1, %g4
1744
1745/*
1746 * Register Inputs:
1747 *	%g5		user trap handler
1748 *	%g7		misaligned addr - for alignment traps only
1749 */
1750.setup_utrap:
1751	set	trap, %g1			! setup in case we go
1752	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1753	sub	%g0, 1, %g4			! the save instruction below
1754
1755	/*
1756	 * If the DTrace pid provider is single stepping a copied-out
1757	 * instruction, t->t_dtrace_step will be set. In that case we need
1758	 * to abort the single-stepping (since execution of the instruction
1759	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1760	 */
1761	save	%sp, -SA(MINFRAME32), %sp	! window for trap handler
1762	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1763	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1764	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1765	rdpr	%tnpc, %l2			! arg1 == tnpc
1766	brz,pt	%g2, 1f
1767	rdpr	%tpc, %l1			! arg0 == tpc
1768
1769	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1770	ldn	[%g1 + T_DTRACE_NPC], %l2	! arg1 = t->t_dtrace_npc (step)
1771	brz,pt	%g2, 1f
1772	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1773	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
17741:
1775	mov	%g7, %l3			! arg2 == misaligned address
1776
1777	rdpr	%tstate, %g1			! cwp for trap handler
1778	rdpr	%cwp, %g4
1779	bclr	TSTATE_CWP_MASK, %g1
1780	wrpr	%g1, %g4, %tstate
1781	wrpr	%g0, %g5, %tnpc			! trap handler address
1782	FAST_TRAP_DONE
1783	/* NOTREACHED */
1784
1785.check_v9utrap:
1786	rdpr	%tstate, %g1
1787	btst	TSTATE_PRIV, %g1
1788	bnz,a,pn %icc, 3f
1789	  nop
1790	CPU_ADDR(%g4, %g1)				! load CPU struct addr
1791	ldn	[%g4 + CPU_THREAD], %g5			! load thread pointer
1792	ldn	[%g5 + T_PROCP], %g5			! load proc pointer
1793	ldn	[%g5 + P_UTRAPS], %g5			! are there utraps?
1794
1795	cmp	%g3, T_SOFTWARE_TRAP
1796	bne,a,pt %icc, 1f
1797	  nop
1798
1799	brz,pt %g5, 3f			! if p_utraps == NULL goto trap()
1800	  rdpr	%tt, %g3		! delay - get actual hw trap type
1801
1802	sub	%g3, 254, %g1		! UT_TRAP_INSTRUCTION_16 = p_utraps[18]
1803	ba,pt	%icc, 2f
1804	  smul	%g1, CPTRSIZE, %g2
18051:
1806	brz,a,pt %g5, 3f		! if p_utraps == NULL goto trap()
1807	  nop
1808
1809	cmp	%g3, T_UNIMP_INSTR
1810	bne,a,pt %icc, 2f
1811	  nop
1812
1813	mov	1, %g1
1814	st	%g1, [%g4 + CPU_TL1_HDLR] ! set CPU_TL1_HDLR
1815	rdpr	%tpc, %g1		! ld trapping instruction using
1816	lduwa	[%g1]ASI_AIUP, %g1	! "AS IF USER" ASI which could fault
1817	st	%g0, [%g4 + CPU_TL1_HDLR] ! clr CPU_TL1_HDLR
1818
1819	sethi	%hi(0xc1c00000), %g4	! setup mask for illtrap instruction
1820	andcc	%g1, %g4, %g4		! and instruction with mask
1821	bnz,a,pt %icc, 3f		! if %g4 == zero, %g1 is an ILLTRAP
1822	  nop				! fall thru to setup
18232:
1824	ldn	[%g5 + %g2], %g5
1825	brnz,a,pt %g5, .setup_v9utrap
1826	  nop
18273:
1828	set	trap, %g1
1829	ba,pt	%xcc, sys_trap
1830	  sub	%g0, 1, %g4
1831	/* NOTREACHED */
1832
1833/*
1834 * Register Inputs:
1835 *	%g5		user trap handler
1836 */
1837.setup_v9utrap:
1838	set	trap, %g1			! setup in case we go
1839	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1840	sub	%g0, 1, %g4			! the save instruction below
1841
1842	/*
1843	 * If the DTrace pid provider is single stepping a copied-out
1844	 * instruction, t->t_dtrace_step will be set. In that case we need
1845	 * to abort the single-stepping (since execution of the instruction
1846	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1847	 */
1848	save	%sp, -SA(MINFRAME64), %sp	! window for trap handler
1849	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1850	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1851	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1852	rdpr	%tnpc, %l7			! arg1 == tnpc
1853	brz,pt	%g2, 1f
1854	rdpr	%tpc, %l6			! arg0 == tpc
1855
1856	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1857	ldn	[%g1 + T_DTRACE_NPC], %l7	! arg1 == t->t_dtrace_npc (step)
1858	brz,pt	%g2, 1f
1859	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1860	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
18611:
1862	rdpr	%tstate, %g2			! cwp for trap handler
1863	rdpr	%cwp, %g4
1864	bclr	TSTATE_CWP_MASK, %g2
1865	wrpr	%g2, %g4, %tstate
1866
1867	ldn	[%g1 + T_PROCP], %g4		! load proc pointer
1868	ldn	[%g4 + P_AS], %g4		! load as pointer
1869	ldn	[%g4 + A_USERLIMIT], %g4	! load as userlimit
1870	cmp	%l7, %g4			! check for single-step set
1871	bne,pt	%xcc, 4f
1872	  nop
1873	ldn	[%g1 + T_LWP], %g1		! load klwp pointer
1874	ld	[%g1 + PCB_STEP], %g4		! load single-step flag
1875	cmp	%g4, STEP_ACTIVE		! step flags set in pcb?
1876	bne,pt	%icc, 4f
1877	  nop
1878	stn	%g5, [%g1 + PCB_TRACEPC]	! save trap handler addr in pcb
1879	mov	%l7, %g4			! on entry to precise user trap
1880	add	%l6, 4, %l7			! handler, %l6 == pc, %l7 == npc
1881						! at time of trap
1882	wrpr	%g0, %g4, %tnpc			! generate FLTBOUNDS,
1883						! %g4 == userlimit
1884	FAST_TRAP_DONE
1885	/* NOTREACHED */
18864:
1887	wrpr	%g0, %g5, %tnpc			! trap handler address
1888	FAST_TRAP_DONE_CHK_INTR
1889	/* NOTREACHED */
1890
1891.fp_exception:
1892	CPU_ADDR(%g1, %g4)
1893	stx	%fsr, [%g1 + CPU_TMP1]
1894	ldx	[%g1 + CPU_TMP1], %g2
1895
1896	/*
1897	 * Cheetah takes unfinished_FPop trap for certain range of operands
1898	 * to the "fitos" instruction. Instead of going through the slow
1899	 * software emulation path, we try to simulate the "fitos" instruction
1900	 * via "fitod" and "fdtos" provided the following conditions are met:
1901	 *
1902	 *	fpu_exists is set (if DEBUG)
1903	 *	not in privileged mode
1904	 *	ftt is unfinished_FPop
1905	 *	NXM IEEE trap is not enabled
1906	 *	instruction at %tpc is "fitos"
1907	 *
1908	 *  Usage:
1909	 *	%g1	per cpu address
1910	 *	%g2	%fsr
1911	 *	%g6	user instruction
1912	 *
1913	 * Note that we can take a memory access related trap while trying
1914	 * to fetch the user instruction. Therefore, we set CPU_TL1_HDLR
1915	 * flag to catch those traps and let the SFMMU code deal with page
1916	 * fault and data access exception.
1917	 */
1918#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
1919	sethi	%hi(fpu_exists), %g7
1920	ld	[%g7 + %lo(fpu_exists)], %g7
1921	brz,pn %g7, .fp_exception_cont
1922	  nop
1923#endif
1924	rdpr	%tstate, %g7			! branch if in privileged mode
1925	btst	TSTATE_PRIV, %g7
1926	bnz,pn	%xcc, .fp_exception_cont
1927	srl	%g2, FSR_FTT_SHIFT, %g7		! extract ftt from %fsr
1928	and	%g7, (FSR_FTT>>FSR_FTT_SHIFT), %g7
1929	cmp	%g7, FTT_UNFIN
1930	set	FSR_TEM_NX, %g5
1931	bne,pn	%xcc, .fp_exception_cont	! branch if NOT unfinished_FPop
1932	  andcc	%g2, %g5, %g0
1933	bne,pn	%xcc, .fp_exception_cont	! branch if FSR_TEM_NX enabled
1934	  rdpr	%tpc, %g5			! get faulting PC
1935
1936	or	%g0, 1, %g7
1937	st	%g7, [%g1 + CPU_TL1_HDLR]	! set tl1_hdlr flag
1938	lda	[%g5]ASI_USER, %g6		! get user's instruction
1939	st	%g0, [%g1 + CPU_TL1_HDLR]	! clear tl1_hdlr flag
1940
1941	set	FITOS_INSTR_MASK, %g7
1942	and	%g6, %g7, %g7
1943	set	FITOS_INSTR, %g5
1944	cmp	%g7, %g5
1945	bne,pn	%xcc, .fp_exception_cont	! branch if not FITOS_INSTR
1946	 nop
1947
1948	/*
1949	 * This is unfinished FPops trap for "fitos" instruction. We
1950	 * need to simulate "fitos" via "fitod" and "fdtos" instruction
1951	 * sequence.
1952	 *
1953	 * We need a temporary FP register to do the conversion. Since
1954	 * both source and destination operands for the "fitos" instruction
1955	 * have to be within %f0-%f31, we use an FP register from the upper
1956	 * half to guarantee that it won't collide with the source or the
1957	 * dest operand. However, we do have to save and restore its value.
1958	 *
1959	 * We use %d62 as a temporary FP register for the conversion and
1960	 * branch to appropriate instruction within the conversion tables
1961	 * based upon the rs2 and rd values.
1962	 */
1963
1964	std	%d62, [%g1 + CPU_TMP1]		! save original value
1965
1966	srl	%g6, FITOS_RS2_SHIFT, %g7
1967	and	%g7, FITOS_REG_MASK, %g7
1968	set	_fitos_fitod_table, %g4
1969	sllx	%g7, 2, %g7
1970	jmp	%g4 + %g7
1971	  ba,pt	%xcc, _fitos_fitod_done
1972	.empty
1973
1974_fitos_fitod_table:
1975	  fitod	%f0, %d62
1976	  fitod	%f1, %d62
1977	  fitod	%f2, %d62
1978	  fitod	%f3, %d62
1979	  fitod	%f4, %d62
1980	  fitod	%f5, %d62
1981	  fitod	%f6, %d62
1982	  fitod	%f7, %d62
1983	  fitod	%f8, %d62
1984	  fitod	%f9, %d62
1985	  fitod	%f10, %d62
1986	  fitod	%f11, %d62
1987	  fitod	%f12, %d62
1988	  fitod	%f13, %d62
1989	  fitod	%f14, %d62
1990	  fitod	%f15, %d62
1991	  fitod	%f16, %d62
1992	  fitod	%f17, %d62
1993	  fitod	%f18, %d62
1994	  fitod	%f19, %d62
1995	  fitod	%f20, %d62
1996	  fitod	%f21, %d62
1997	  fitod	%f22, %d62
1998	  fitod	%f23, %d62
1999	  fitod	%f24, %d62
2000	  fitod	%f25, %d62
2001	  fitod	%f26, %d62
2002	  fitod	%f27, %d62
2003	  fitod	%f28, %d62
2004	  fitod	%f29, %d62
2005	  fitod	%f30, %d62
2006	  fitod	%f31, %d62
2007_fitos_fitod_done:
2008
2009	/*
2010	 * Now convert data back into single precision
2011	 */
2012	srl	%g6, FITOS_RD_SHIFT, %g7
2013	and	%g7, FITOS_REG_MASK, %g7
2014	set	_fitos_fdtos_table, %g4
2015	sllx	%g7, 2, %g7
2016	jmp	%g4 + %g7
2017	  ba,pt	%xcc, _fitos_fdtos_done
2018	.empty
2019
2020_fitos_fdtos_table:
2021	  fdtos	%d62, %f0
2022	  fdtos	%d62, %f1
2023	  fdtos	%d62, %f2
2024	  fdtos	%d62, %f3
2025	  fdtos	%d62, %f4
2026	  fdtos	%d62, %f5
2027	  fdtos	%d62, %f6
2028	  fdtos	%d62, %f7
2029	  fdtos	%d62, %f8
2030	  fdtos	%d62, %f9
2031	  fdtos	%d62, %f10
2032	  fdtos	%d62, %f11
2033	  fdtos	%d62, %f12
2034	  fdtos	%d62, %f13
2035	  fdtos	%d62, %f14
2036	  fdtos	%d62, %f15
2037	  fdtos	%d62, %f16
2038	  fdtos	%d62, %f17
2039	  fdtos	%d62, %f18
2040	  fdtos	%d62, %f19
2041	  fdtos	%d62, %f20
2042	  fdtos	%d62, %f21
2043	  fdtos	%d62, %f22
2044	  fdtos	%d62, %f23
2045	  fdtos	%d62, %f24
2046	  fdtos	%d62, %f25
2047	  fdtos	%d62, %f26
2048	  fdtos	%d62, %f27
2049	  fdtos	%d62, %f28
2050	  fdtos	%d62, %f29
2051	  fdtos	%d62, %f30
2052	  fdtos	%d62, %f31
2053_fitos_fdtos_done:
2054
2055	ldd	[%g1 + CPU_TMP1], %d62		! restore %d62
2056
2057#if DEBUG
2058	/*
2059	 * Update FPop_unfinished trap kstat
2060	 */
2061	set	fpustat+FPUSTAT_UNFIN_KSTAT, %g7
2062	ldx	[%g7], %g5
20631:
2064	add	%g5, 1, %g6
2065
2066	casxa	[%g7] ASI_N, %g5, %g6
2067	cmp	%g5, %g6
2068	bne,a,pn %xcc, 1b
2069	  or	%g0, %g6, %g5
2070
2071	/*
2072	 * Update fpu_sim_fitos kstat
2073	 */
2074	set	fpuinfo+FPUINFO_FITOS_KSTAT, %g7
2075	ldx	[%g7], %g5
20761:
2077	add	%g5, 1, %g6
2078
2079	casxa	[%g7] ASI_N, %g5, %g6
2080	cmp	%g5, %g6
2081	bne,a,pn %xcc, 1b
2082	  or	%g0, %g6, %g5
2083#endif /* DEBUG */
2084
2085	FAST_TRAP_DONE
2086
2087.fp_exception_cont:
2088	/*
2089	 * Let _fp_exception deal with simulating FPop instruction.
2090	 * Note that we need to pass %fsr in %g2 (already read above).
2091	 */
2092
2093	set	_fp_exception, %g1
2094	ba,pt	%xcc, sys_trap
2095	sub	%g0, 1, %g4
2096
2097.clean_windows:
2098	set	trap, %g1
2099	mov	T_FLUSH_PCB, %g3
2100	sub	%g0, 1, %g4
2101	save
2102	flushw
2103	restore
2104	wrpr	%g0, %g0, %cleanwin	! no clean windows
2105
2106	CPU_ADDR(%g4, %g5)
2107	ldn	[%g4 + CPU_MPCB], %g4
2108	brz,a,pn %g4, 1f
2109	  nop
2110	ld	[%g4 + MPCB_WSTATE], %g5
2111	add	%g5, WSTATE_CLEAN_OFFSET, %g5
2112	wrpr	%g0, %g5, %wstate
21131:	FAST_TRAP_DONE
2114
2115/*
2116 * .spill_clean: clean the previous window, restore the wstate, and
2117 * "done".
2118 *
2119 * Entry: %g7 contains new wstate
2120 */
2121.spill_clean:
2122	sethi	%hi(nwin_minus_one), %g5
2123	ld	[%g5 + %lo(nwin_minus_one)], %g5 ! %g5 = nwin - 1
2124	rdpr	%cwp, %g6			! %g6 = %cwp
2125	deccc	%g6				! %g6--
2126	movneg	%xcc, %g5, %g6			! if (%g6<0) %g6 = nwin-1
2127	wrpr	%g6, %cwp
2128	TT_TRACE_L(trace_win)
2129	clr	%l0
2130	clr	%l1
2131	clr	%l2
2132	clr	%l3
2133	clr	%l4
2134	clr	%l5
2135	clr	%l6
2136	clr	%l7
2137	wrpr	%g0, %g7, %wstate
2138	saved
2139	retry			! restores correct %cwp
2140
2141.fix_alignment:
2142	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2143	ldn	[%g1 + CPU_THREAD], %g1	! load thread pointer
2144	ldn	[%g1 + T_PROCP], %g1
2145	mov	1, %g2
2146	stb	%g2, [%g1 + P_FIXALIGNMENT]
2147	FAST_TRAP_DONE
2148
2149#define	STDF_REG(REG, ADDR, TMP)		\
2150	sll	REG, 3, REG			;\
2151mark1:	set	start1, TMP			;\
2152	jmp	REG + TMP			;\
2153	  nop					;\
2154start1:	ba,pt	%xcc, done1			;\
2155	  std	%f0, [ADDR + CPU_TMP1]		;\
2156	ba,pt	%xcc, done1			;\
2157	  std	%f32, [ADDR + CPU_TMP1]		;\
2158	ba,pt	%xcc, done1			;\
2159	  std	%f2, [ADDR + CPU_TMP1]		;\
2160	ba,pt	%xcc, done1			;\
2161	  std	%f34, [ADDR + CPU_TMP1]		;\
2162	ba,pt	%xcc, done1			;\
2163	  std	%f4, [ADDR + CPU_TMP1]		;\
2164	ba,pt	%xcc, done1			;\
2165	  std	%f36, [ADDR + CPU_TMP1]		;\
2166	ba,pt	%xcc, done1			;\
2167	  std	%f6, [ADDR + CPU_TMP1]		;\
2168	ba,pt	%xcc, done1			;\
2169	  std	%f38, [ADDR + CPU_TMP1]		;\
2170	ba,pt	%xcc, done1			;\
2171	  std	%f8, [ADDR + CPU_TMP1]		;\
2172	ba,pt	%xcc, done1			;\
2173	  std	%f40, [ADDR + CPU_TMP1]		;\
2174	ba,pt	%xcc, done1			;\
2175	  std	%f10, [ADDR + CPU_TMP1]		;\
2176	ba,pt	%xcc, done1			;\
2177	  std	%f42, [ADDR + CPU_TMP1]		;\
2178	ba,pt	%xcc, done1			;\
2179	  std	%f12, [ADDR + CPU_TMP1]		;\
2180	ba,pt	%xcc, done1			;\
2181	  std	%f44, [ADDR + CPU_TMP1]		;\
2182	ba,pt	%xcc, done1			;\
2183	  std	%f14, [ADDR + CPU_TMP1]		;\
2184	ba,pt	%xcc, done1			;\
2185	  std	%f46, [ADDR + CPU_TMP1]		;\
2186	ba,pt	%xcc, done1			;\
2187	  std	%f16, [ADDR + CPU_TMP1]		;\
2188	ba,pt	%xcc, done1			;\
2189	  std	%f48, [ADDR + CPU_TMP1]		;\
2190	ba,pt	%xcc, done1			;\
2191	  std	%f18, [ADDR + CPU_TMP1]		;\
2192	ba,pt	%xcc, done1			;\
2193	  std	%f50, [ADDR + CPU_TMP1]		;\
2194	ba,pt	%xcc, done1			;\
2195	  std	%f20, [ADDR + CPU_TMP1]		;\
2196	ba,pt	%xcc, done1			;\
2197	  std	%f52, [ADDR + CPU_TMP1]		;\
2198	ba,pt	%xcc, done1			;\
2199	  std	%f22, [ADDR + CPU_TMP1]		;\
2200	ba,pt	%xcc, done1			;\
2201	  std	%f54, [ADDR + CPU_TMP1]		;\
2202	ba,pt	%xcc, done1			;\
2203	  std	%f24, [ADDR + CPU_TMP1]		;\
2204	ba,pt	%xcc, done1			;\
2205	  std	%f56, [ADDR + CPU_TMP1]		;\
2206	ba,pt	%xcc, done1			;\
2207	  std	%f26, [ADDR + CPU_TMP1]		;\
2208	ba,pt	%xcc, done1			;\
2209	  std	%f58, [ADDR + CPU_TMP1]		;\
2210	ba,pt	%xcc, done1			;\
2211	  std	%f28, [ADDR + CPU_TMP1]		;\
2212	ba,pt	%xcc, done1			;\
2213	  std	%f60, [ADDR + CPU_TMP1]		;\
2214	ba,pt	%xcc, done1			;\
2215	  std	%f30, [ADDR + CPU_TMP1]		;\
2216	ba,pt	%xcc, done1			;\
2217	  std	%f62, [ADDR + CPU_TMP1]		;\
2218done1:
2219
2220#define	LDDF_REG(REG, ADDR, TMP)		\
2221	sll	REG, 3, REG			;\
2222mark2:	set	start2, TMP			;\
2223	jmp	REG + TMP			;\
2224	  nop					;\
2225start2:	ba,pt	%xcc, done2			;\
2226	  ldd	[ADDR + CPU_TMP1], %f0		;\
2227	ba,pt	%xcc, done2			;\
2228	  ldd	[ADDR + CPU_TMP1], %f32		;\
2229	ba,pt	%xcc, done2			;\
2230	  ldd	[ADDR + CPU_TMP1], %f2		;\
2231	ba,pt	%xcc, done2			;\
2232	  ldd	[ADDR + CPU_TMP1], %f34		;\
2233	ba,pt	%xcc, done2			;\
2234	  ldd	[ADDR + CPU_TMP1], %f4		;\
2235	ba,pt	%xcc, done2			;\
2236	  ldd	[ADDR + CPU_TMP1], %f36		;\
2237	ba,pt	%xcc, done2			;\
2238	  ldd	[ADDR + CPU_TMP1], %f6		;\
2239	ba,pt	%xcc, done2			;\
2240	  ldd	[ADDR + CPU_TMP1], %f38		;\
2241	ba,pt	%xcc, done2			;\
2242	  ldd	[ADDR + CPU_TMP1], %f8		;\
2243	ba,pt	%xcc, done2			;\
2244	  ldd	[ADDR + CPU_TMP1], %f40		;\
2245	ba,pt	%xcc, done2			;\
2246	  ldd	[ADDR + CPU_TMP1], %f10		;\
2247	ba,pt	%xcc, done2			;\
2248	  ldd	[ADDR + CPU_TMP1], %f42		;\
2249	ba,pt	%xcc, done2			;\
2250	  ldd	[ADDR + CPU_TMP1], %f12		;\
2251	ba,pt	%xcc, done2			;\
2252	  ldd	[ADDR + CPU_TMP1], %f44		;\
2253	ba,pt	%xcc, done2			;\
2254	  ldd	[ADDR + CPU_TMP1], %f14		;\
2255	ba,pt	%xcc, done2			;\
2256	  ldd	[ADDR + CPU_TMP1], %f46		;\
2257	ba,pt	%xcc, done2			;\
2258	  ldd	[ADDR + CPU_TMP1], %f16		;\
2259	ba,pt	%xcc, done2			;\
2260	  ldd	[ADDR + CPU_TMP1], %f48		;\
2261	ba,pt	%xcc, done2			;\
2262	  ldd	[ADDR + CPU_TMP1], %f18		;\
2263	ba,pt	%xcc, done2			;\
2264	  ldd	[ADDR + CPU_TMP1], %f50		;\
2265	ba,pt	%xcc, done2			;\
2266	  ldd	[ADDR + CPU_TMP1], %f20		;\
2267	ba,pt	%xcc, done2			;\
2268	  ldd	[ADDR + CPU_TMP1], %f52		;\
2269	ba,pt	%xcc, done2			;\
2270	  ldd	[ADDR + CPU_TMP1], %f22		;\
2271	ba,pt	%xcc, done2			;\
2272	  ldd	[ADDR + CPU_TMP1], %f54		;\
2273	ba,pt	%xcc, done2			;\
2274	  ldd	[ADDR + CPU_TMP1], %f24		;\
2275	ba,pt	%xcc, done2			;\
2276	  ldd	[ADDR + CPU_TMP1], %f56		;\
2277	ba,pt	%xcc, done2			;\
2278	  ldd	[ADDR + CPU_TMP1], %f26		;\
2279	ba,pt	%xcc, done2			;\
2280	  ldd	[ADDR + CPU_TMP1], %f58		;\
2281	ba,pt	%xcc, done2			;\
2282	  ldd	[ADDR + CPU_TMP1], %f28		;\
2283	ba,pt	%xcc, done2			;\
2284	  ldd	[ADDR + CPU_TMP1], %f60		;\
2285	ba,pt	%xcc, done2			;\
2286	  ldd	[ADDR + CPU_TMP1], %f30		;\
2287	ba,pt	%xcc, done2			;\
2288	  ldd	[ADDR + CPU_TMP1], %f62		;\
2289done2:
2290
2291.lddf_exception_not_aligned:
2292	/*
2293	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2294	 */
2295	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2296
2297#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2298	sethi	%hi(fpu_exists), %g2		! check fpu_exists
2299	ld	[%g2 + %lo(fpu_exists)], %g2
2300	brz,a,pn %g2, 4f
2301	  nop
2302#endif
2303	CPU_ADDR(%g1, %g4)
2304	or	%g0, 1, %g4
2305	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2306
2307	rdpr	%tpc, %g2
2308	lda	[%g2]ASI_AIUP, %g6	! get the user's lddf instruction
2309	srl	%g6, 23, %g1		! using ldda or not?
2310	and	%g1, 1, %g1
2311	brz,a,pt %g1, 2f		! check for ldda instruction
2312	  nop
2313	srl	%g6, 13, %g1		! check immflag
2314	and	%g1, 1, %g1
2315	rdpr	%tstate, %g2		! %tstate in %g2
2316	brnz,a,pn %g1, 1f
2317	  srl	%g2, 31, %g1		! get asi from %tstate
2318	srl	%g6, 5, %g1		! get asi from instruction
2319	and	%g1, 0xFF, %g1		! imm_asi field
23201:
2321	cmp	%g1, ASI_P		! primary address space
2322	be,a,pt %icc, 2f
2323	  nop
2324	cmp	%g1, ASI_PNF		! primary no fault address space
2325	be,a,pt %icc, 2f
2326	  nop
2327	cmp	%g1, ASI_S		! secondary address space
2328	be,a,pt %icc, 2f
2329	  nop
2330	cmp	%g1, ASI_SNF		! secondary no fault address space
2331	bne,a,pn %icc, 3f
2332	  nop
23332:
2334	lduwa	[%g5]ASI_USER, %g7	! get first half of misaligned data
2335	add	%g5, 4, %g5		! increment misaligned data address
2336	lduwa	[%g5]ASI_USER, %g5	! get second half of misaligned data
2337
2338	sllx	%g7, 32, %g7
2339	or	%g5, %g7, %g5		! combine data
2340	CPU_ADDR(%g7, %g1)		! save data on a per-cpu basis
2341	stx	%g5, [%g7 + CPU_TMP1]	! save in cpu_tmp1
2342
2343	srl	%g6, 25, %g3		! %g6 has the instruction
2344	and	%g3, 0x1F, %g3		! %g3 has rd
2345	LDDF_REG(%g3, %g7, %g4)
2346
2347	CPU_ADDR(%g1, %g4)
2348	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2349	FAST_TRAP_DONE
23503:
2351	CPU_ADDR(%g1, %g4)
2352	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
23534:
2354	set	T_USER, %g3		! trap type in %g3
2355	or	%g3, T_LDDF_ALIGN, %g3
2356	mov	%g5, %g2		! misaligned vaddr in %g2
2357	set	fpu_trap, %g1		! goto C for the little and
2358	ba,pt	%xcc, sys_trap		! no fault little asi's
2359	  sub	%g0, 1, %g4
2360
2361.stdf_exception_not_aligned:
2362	/*
2363	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2364	 */
2365	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2366
2367#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2368	sethi	%hi(fpu_exists), %g7		! check fpu_exists
2369	ld	[%g7 + %lo(fpu_exists)], %g3
2370	brz,a,pn %g3, 4f
2371	  nop
2372#endif
2373	CPU_ADDR(%g1, %g4)
2374	or	%g0, 1, %g4
2375	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2376
2377	rdpr	%tpc, %g2
2378	lda	[%g2]ASI_AIUP, %g6	! get the user's stdf instruction
2379
2380	srl	%g6, 23, %g1		! using stda or not?
2381	and	%g1, 1, %g1
2382	brz,a,pt %g1, 2f		! check for stda instruction
2383	  nop
2384	srl	%g6, 13, %g1		! check immflag
2385	and	%g1, 1, %g1
2386	rdpr	%tstate, %g2		! %tstate in %g2
2387	brnz,a,pn %g1, 1f
2388	  srl	%g2, 31, %g1		! get asi from %tstate
2389	srl	%g6, 5, %g1		! get asi from instruction
2390	and	%g1, 0xFF, %g1		! imm_asi field
23911:
2392	cmp	%g1, ASI_P		! primary address space
2393	be,a,pt %icc, 2f
2394	  nop
2395	cmp	%g1, ASI_S		! secondary address space
2396	bne,a,pn %icc, 3f
2397	  nop
23982:
2399	srl	%g6, 25, %g6
2400	and	%g6, 0x1F, %g6		! %g6 has rd
2401	CPU_ADDR(%g7, %g1)
2402	STDF_REG(%g6, %g7, %g4)		! STDF_REG(REG, ADDR, TMP)
2403
2404	ldx	[%g7 + CPU_TMP1], %g6
2405	srlx	%g6, 32, %g7
2406	stuwa	%g7, [%g5]ASI_USER	! first half
2407	add	%g5, 4, %g5		! increment misaligned data address
2408	stuwa	%g6, [%g5]ASI_USER	! second half
2409
2410	CPU_ADDR(%g1, %g4)
2411	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2412	FAST_TRAP_DONE
24133:
2414	CPU_ADDR(%g1, %g4)
2415	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
24164:
2417	set	T_USER, %g3		! trap type in %g3
2418	or	%g3, T_STDF_ALIGN, %g3
2419	mov	%g5, %g2		! misaligned vaddr in %g2
2420	set	fpu_trap, %g1		! goto C for the little and
2421	ba,pt	%xcc, sys_trap		! nofault little asi's
2422	  sub	%g0, 1, %g4
2423
2424#ifdef DEBUG_USER_TRAPTRACECTL
2425
2426.traptrace_freeze:
2427	mov	%l0, %g1 ; mov	%l1, %g2 ; mov	%l2, %g3 ; mov	%l4, %g4
2428	TT_TRACE_L(trace_win)
2429	mov	%g4, %l4 ; mov	%g3, %l2 ; mov	%g2, %l1 ; mov	%g1, %l0
2430	set	trap_freeze, %g1
2431	mov	1, %g2
2432	st	%g2, [%g1]
2433	FAST_TRAP_DONE
2434
2435.traptrace_unfreeze:
2436	set	trap_freeze, %g1
2437	st	%g0, [%g1]
2438	mov	%l0, %g1 ; mov	%l1, %g2 ; mov	%l2, %g3 ; mov	%l4, %g4
2439	TT_TRACE_L(trace_win)
2440	mov	%g4, %l4 ; mov	%g3, %l2 ; mov	%g2, %l1 ; mov	%g1, %l0
2441	FAST_TRAP_DONE
2442
2443#endif /* DEBUG_USER_TRAPTRACECTL */
2444
2445.getcc:
2446	CPU_ADDR(%g1, %g2)
2447	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2448	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2449	rdpr	%tstate, %g3			! get tstate
2450	srlx	%g3, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2451	set	PSR_ICC, %g2
2452	and	%o0, %g2, %o0			! mask out the rest
2453	srl	%o0, PSR_ICC_SHIFT, %o0		! right justify
2454	rdpr	%pstate, %o1
2455	wrpr	%o1, PSTATE_AG, %pstate		! get into normal globals
2456	mov	%o0, %g1			! move ccr to normal %g1
2457	wrpr	%g0, %o1, %pstate		! back into alternate globals
2458	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2459	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2460	FAST_TRAP_DONE
2461
2462.setcc:
2463	CPU_ADDR(%g1, %g2)
2464	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2465	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2466	rdpr	%pstate, %o0
2467	wrpr	%o0, PSTATE_AG, %pstate		! get into normal globals
2468	mov	%g1, %o1
2469	wrpr	%g0, %o0, %pstate		! back to alternates
2470	sll	%o1, PSR_ICC_SHIFT, %g2
2471	set	PSR_ICC, %g3
2472	and	%g2, %g3, %g2			! mask out rest
2473	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g2
2474	rdpr	%tstate, %g3			! get tstate
2475	srl	%g3, 0, %g3			! clear upper word
2476	or	%g3, %g2, %g3			! or in new bits
2477	wrpr	%g3, %tstate
2478	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2479	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2480	FAST_TRAP_DONE
2481
2482/*
2483 * getpsr(void)
2484 * Note that the xcc part of the ccr is not provided.
2485 * The V8 code shows why the V9 trap is not faster:
2486 * #define GETPSR_TRAP() \
2487 *      mov %psr, %i0; jmp %l2; rett %l2+4; nop;
2488 */
2489
2490	.type	.getpsr, #function
2491.getpsr:
2492	rdpr	%tstate, %g1			! get tstate
2493	srlx	%g1, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2494	set	PSR_ICC, %g2
2495	and	%o0, %g2, %o0			! mask out the rest
2496
2497	rd	%fprs, %g1			! get fprs
2498	and	%g1, FPRS_FEF, %g2		! mask out dirty upper/lower
2499	sllx	%g2, PSR_FPRS_FEF_SHIFT, %g2	! shift fef to V8 psr.ef
2500	or	%o0, %g2, %o0			! or result into psr.ef
2501
2502	set	V9_PSR_IMPLVER, %g2		! SI assigned impl/ver: 0xef
2503	or	%o0, %g2, %o0			! or psr.impl/ver
2504	FAST_TRAP_DONE
2505	SET_SIZE(.getpsr)
2506
2507/*
2508 * setpsr(newpsr)
2509 * Note that there is no support for ccr.xcc in the V9 code.
2510 */
2511
2512	.type	.setpsr, #function
2513.setpsr:
2514	rdpr	%tstate, %g1			! get tstate
2515!	setx	TSTATE_V8_UBITS, %g2
2516	or 	%g0, CCR_ICC, %g3
2517	sllx	%g3, TSTATE_CCR_SHIFT, %g2
2518
2519	andn	%g1, %g2, %g1			! zero current user bits
2520	set	PSR_ICC, %g2
2521	and	%g2, %o0, %g2			! clear all but psr.icc bits
2522	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g3	! shift to tstate.ccr.icc
2523	wrpr	%g1, %g3, %tstate		! write tstate
2524
2525	set	PSR_EF, %g2
2526	and	%g2, %o0, %g2			! clear all but fp enable bit
2527	srlx	%g2, PSR_FPRS_FEF_SHIFT, %g4	! shift ef to V9 fprs.fef
2528	wr	%g0, %g4, %fprs			! write fprs
2529
2530	CPU_ADDR(%g1, %g2)			! load CPU struct addr to %g1
2531	ldn	[%g1 + CPU_THREAD], %g2		! load thread pointer
2532	ldn	[%g2 + T_LWP], %g3		! load klwp pointer
2533	ldn	[%g3 + LWP_FPU], %g2		! get lwp_fpu pointer
2534	stuw	%g4, [%g2 + FPU_FPRS]		! write fef value to fpu_fprs
2535	srlx	%g4, 2, %g4			! shift fef value to bit 0
2536	stub	%g4, [%g2 + FPU_EN]		! write fef value to fpu_en
2537	FAST_TRAP_DONE
2538	SET_SIZE(.setpsr)
2539
2540/*
2541 * getlgrp
2542 * get home lgrpid on which the calling thread is currently executing.
2543 */
2544	.type	.getlgrp, #function
2545.getlgrp:
2546	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2547	ld	[%g1 + CPU_ID], %o0	! load cpu_id
2548	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2549	ldn	[%g2 + T_LPL], %g2	! load lpl pointer
2550	ld	[%g2 + LPL_LGRPID], %g1	! load lpl_lgrpid
2551	sra	%g1, 0, %o1
2552	FAST_TRAP_DONE
2553	SET_SIZE(.getlgrp)
2554
2555/*
2556 * Entry for old 4.x trap (trap 0).
2557 */
2558	ENTRY_NP(syscall_trap_4x)
2559	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2560	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2561	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2562	ld	[%g2 + PCB_TRAP0], %g2	! lwp->lwp_pcb.pcb_trap0addr
2563	brz,pn	%g2, 1f			! has it been set?
2564	st	%l0, [%g1 + CPU_TMP1]	! delay - save some locals
2565	st	%l1, [%g1 + CPU_TMP2]
2566	rdpr	%tnpc, %l1		! save old tnpc
2567	wrpr	%g0, %g2, %tnpc		! setup tnpc
2568
2569	rdpr	%pstate, %l0
2570	wrpr	%l0, PSTATE_AG, %pstate	! switch to normal globals
2571	mov	%l1, %g6		! pass tnpc to user code in %g6
2572	wrpr	%l0, %g0, %pstate	! switch back to alternate globals
2573
2574	! Note that %g1 still contains CPU struct addr
2575	ld	[%g1 + CPU_TMP2], %l1	! restore locals
2576	ld	[%g1 + CPU_TMP1], %l0
2577	FAST_TRAP_DONE_CHK_INTR
25781:
2579	mov	%g1, %l0
2580	st	%l1, [%g1 + CPU_TMP2]
2581	rdpr	%pstate, %l1
2582	wrpr	%l1, PSTATE_AG, %pstate
2583	!
2584	! check for old syscall mmap which is the only different one which
2585	! must be the same.  Others are handled in the compatibility library.
2586	!
2587	cmp	%g1, OSYS_mmap	! compare to old 4.x mmap
2588	movz	%icc, SYS_mmap, %g1
2589	wrpr	%g0, %l1, %pstate
2590	ld	[%l0 + CPU_TMP2], %l1	! restore locals
2591	ld	[%l0 + CPU_TMP1], %l0
2592	SYSCALL(syscall_trap32)
2593	SET_SIZE(syscall_trap_4x)
2594
2595/*
2596 * Handler for software trap 9.
2597 * Set trap0 emulation address for old 4.x system call trap.
2598 * XXX - this should be a system call.
2599 */
2600	ENTRY_NP(set_trap0_addr)
2601	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2602	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2603	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2604	st	%l0, [%g1 + CPU_TMP1]	! save some locals
2605	st	%l1, [%g1 + CPU_TMP2]
2606	rdpr	%pstate, %l0
2607	wrpr	%l0, PSTATE_AG, %pstate
2608	mov	%g1, %l1
2609	wrpr	%g0, %l0, %pstate
2610	andn	%l1, 3, %l1		! force alignment
2611	st	%l1, [%g2 + PCB_TRAP0]	! lwp->lwp_pcb.pcb_trap0addr
2612	ld	[%g1 + CPU_TMP1], %l0	! restore locals
2613	ld	[%g1 + CPU_TMP2], %l1
2614	FAST_TRAP_DONE
2615	SET_SIZE(set_trap0_addr)
2616
2617/*
2618 * mmu_trap_tl1
2619 * trap handler for unexpected mmu traps.
2620 * simply checks if the trap was a user lddf/stdf alignment trap, in which
2621 * case we go to fpu_trap or a user trap from the window handler, in which
2622 * case we go save the state on the pcb.  Otherwise, we go to ptl1_panic.
2623 */
2624	.type	mmu_trap_tl1, #function
2625mmu_trap_tl1:
2626#ifdef	TRAPTRACE
2627	TRACE_PTR(%g5, %g6)
2628	GET_TRACE_TICK(%g6)
2629	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2630	rdpr	%tl, %g6
2631	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2632	rdpr	%tt, %g6
2633	stha	%g6, [%g5 + TRAP_ENT_TT]%asi
2634	rdpr	%tstate, %g6
2635	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi
2636	stna	%sp, [%g5 + TRAP_ENT_SP]%asi
2637	stna	%g0, [%g5 + TRAP_ENT_TR]%asi
2638	rdpr	%tpc, %g6
2639	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2640	set	MMU_SFAR, %g6
2641	ldxa	[%g6]ASI_DMMU, %g6
2642	stxa	%g6, [%g5 + TRAP_ENT_F1]%asi
2643	CPU_PADDR(%g7, %g6);
2644	add	%g7, CPU_TL1_HDLR, %g7
2645	lda	[%g7]ASI_MEM, %g6
2646	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2647	set	0xdeadbeef, %g6
2648	stna	%g6, [%g5 + TRAP_ENT_F3]%asi
2649	stna	%g6, [%g5 + TRAP_ENT_F4]%asi
2650	TRACE_NEXT(%g5, %g6, %g7)
2651#endif /* TRAPTRACE */
2652
2653	GET_CPU_IMPL(%g5)
2654	cmp	%g5, PANTHER_IMPL
2655	bne	mmu_trap_tl1_4
2656	  nop
2657	rdpr	%tt, %g5
2658	cmp	%g5, T_DATA_EXCEPTION
2659	bne	mmu_trap_tl1_4
2660	  nop
2661	wr	%g0, ASI_DMMU, %asi
2662	ldxa	[MMU_SFSR]%asi, %g5
2663	mov	1, %g6
2664	sllx	%g6, PN_SFSR_PARITY_SHIFT, %g6
2665	andcc	%g5, %g6, %g0
2666	bz	mmu_trap_tl1_4
2667
2668	/*
2669	 * We are running on a Panther and have hit a DTLB parity error.
2670	 */
2671	ldxa	[MMU_TAG_ACCESS]%asi, %g2
2672	mov	%g5, %g3
2673	ba,pt	%xcc, .mmu_exception_is_tlb_parity
2674	mov	T_DATA_EXCEPTION, %g1
2675
2676mmu_trap_tl1_4:
2677	CPU_PADDR(%g7, %g6);
2678	add     %g7, CPU_TL1_HDLR, %g7		! %g7 = &cpu_m.tl1_hdlr (PA)
2679	/*
2680	 * AM is cleared on trap, so addresses are 64 bit
2681	 */
2682	lda     [%g7]ASI_MEM, %g6
2683	brz,a,pt %g6, 1f
2684	  nop
2685	/*
2686	 * We are going to update cpu_m.tl1_hdlr using physical address.
2687	 * Flush the D$ line, so that stale data won't be accessed later.
2688	 */
2689	CPU_ADDR(%g6, %g5)
2690	add     %g6, CPU_TL1_HDLR, %g6		! %g6 = &cpu_m.tl1_hdlr (VA)
2691	GET_CPU_IMPL(%g5)
2692	cmp	%g5, CHEETAH_IMPL
2693	bl,pt	%icc, 3f
2694	 cmp	%g5, SPITFIRE_IMPL
2695	stxa	%g0, [%g7]ASI_DC_INVAL
2696	membar	#Sync
2697	ba,pt	%xcc, 2f
2698	 nop
26993:
2700	bl,pt	%icc, 2f
2701	 sethi	%hi(dcache_line_mask), %g5
2702	ld	[%g5 + %lo(dcache_line_mask)], %g5
2703	and	%g6, %g5, %g5
2704	stxa	%g0, [%g5]ASI_DC_TAG
2705	membar	#Sync
27062:
2707	sta     %g0, [%g7]ASI_MEM
2708	SWITCH_GLOBALS				! back to mmu globals
2709	ba,a,pt	%xcc, sfmmu_mmu_trap		! handle page faults
27101:
2711	rdpr	%tt, %g5
2712	rdpr	%tl, %g7
2713	sub	%g7, 1, %g6
2714	wrpr	%g6, %tl
2715	rdpr	%tt, %g6
2716	wrpr	%g7, %tl
2717	and	%g6, WTRAP_TTMASK, %g6
2718	cmp	%g6, WTRAP_TYPE
2719	bne,a,pn %xcc, ptl1_panic
2720	mov	PTL1_BAD_MMUTRAP, %g1
2721	rdpr	%tpc, %g7
2722	/* tpc should be in the trap table */
2723	set	trap_table, %g6
2724	cmp	%g7, %g6
2725	blt,a,pn %xcc, ptl1_panic
2726	  mov	PTL1_BAD_MMUTRAP, %g1
2727	set	etrap_table, %g6
2728	cmp	%g7, %g6
2729	bge,a,pn %xcc, ptl1_panic
2730	  mov	PTL1_BAD_MMUTRAP, %g1
2731	cmp	%g5, T_ALIGNMENT
2732	move	%icc, MMU_SFAR, %g6
2733	movne	%icc, MMU_TAG_ACCESS, %g6
2734	ldxa	[%g6]ASI_DMMU, %g6
2735	andn	%g7, WTRAP_ALIGN, %g7	/* 128 byte aligned */
2736	add	%g7, WTRAP_FAULTOFF, %g7
2737	wrpr	%g0, %g7, %tnpc
2738	done
2739	SET_SIZE(mmu_trap_tl1)
2740
2741/*
2742 * Several traps use kmdb_trap and kmdb_trap_tl1 as their handlers.  These
2743 * traps are valid only when kmdb is loaded.  When the debugger is active,
2744 * the code below is rewritten to transfer control to the appropriate
2745 * debugger entry points.
2746 */
2747	.global	kmdb_trap
2748	.align	8
2749kmdb_trap:
2750	ba,a	trap_table0
2751	jmp	%g1 + 0
2752	nop
2753
2754	.global	kmdb_trap_tl1
2755	.align	8
2756kmdb_trap_tl1:
2757	ba,a	trap_table0
2758	jmp	%g1 + 0
2759	nop
2760
2761/*
2762 * This entry is copied from OBP's trap table during boot.
2763 */
2764	.global	obp_bpt
2765	.align	8
2766obp_bpt:
2767	NOT
2768
2769/*
2770 * if kernel, set PCONTEXT to 0 for debuggers
2771 * if user, clear nucleus page sizes
2772 */
2773	.global kctx_obp_bpt
2774kctx_obp_bpt:
2775	set	obp_bpt, %g2
27761:
2777	mov	MMU_PCONTEXT, %g1
2778	ldxa	[%g1]ASI_DMMU, %g1
2779	srlx	%g1, CTXREG_NEXT_SHIFT, %g3
2780	brz,pt	%g3, 3f			! nucleus pgsz is 0, no problem
2781	  sllx	%g3, CTXREG_NEXT_SHIFT, %g3
2782	set	CTXREG_CTX_MASK, %g4	! check Pcontext
2783	btst	%g4, %g1
2784	bz,a,pt	%xcc, 2f
2785	  clr	%g3			! kernel:  PCONTEXT=0
2786	xor	%g3, %g1, %g3		! user:	clr N_pgsz0/1 bits
27872:
2788	set	DEMAP_ALL_TYPE, %g1
2789	stxa	%g0, [%g1]ASI_DTLB_DEMAP
2790	stxa	%g0, [%g1]ASI_ITLB_DEMAP
2791	mov	MMU_PCONTEXT, %g1
2792	stxa	%g3, [%g1]ASI_DMMU
2793        membar  #Sync
2794	sethi	%hi(FLUSH_ADDR), %g1
2795	flush	%g1			! flush required by immu
27963:
2797	jmp	%g2
2798	  nop
2799
2800
2801#ifdef	TRAPTRACE
2802/*
2803 * TRAPTRACE support.
2804 * labels here are branched to with "rd %pc, %g7" in the delay slot.
2805 * Return is done by "jmp %g7 + 4".
2806 */
2807
2808trace_gen:
2809	TRACE_PTR(%g3, %g6)
2810	GET_TRACE_TICK(%g6)
2811	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi
2812	rdpr	%tl, %g6
2813	stha	%g6, [%g3 + TRAP_ENT_TL]%asi
2814	rdpr	%tt, %g6
2815	stha	%g6, [%g3 + TRAP_ENT_TT]%asi
2816	rdpr	%tstate, %g6
2817	stxa	%g6, [%g3 + TRAP_ENT_TSTATE]%asi
2818	stna	%sp, [%g3 + TRAP_ENT_SP]%asi
2819	rdpr	%tpc, %g6
2820	stna	%g6, [%g3 + TRAP_ENT_TPC]%asi
2821	TRACE_NEXT(%g3, %g4, %g5)
2822	jmp	%g7 + 4
2823	nop
2824
2825trace_win:
2826	TRACE_WIN_INFO(0, %l0, %l1, %l2)
2827	! Keep the locals as clean as possible, caller cleans %l4
2828	clr	%l2
2829	clr	%l1
2830	jmp	%l4 + 4
2831	  clr	%l0
2832
2833/*
2834 * Trace a tsb hit
2835 * g1 = tsbe pointer (in/clobbered)
2836 * g2 = tag access register (in)
2837 * g3 - g4 = scratch (clobbered)
2838 * g5 = tsbe data (in)
2839 * g6 = scratch (clobbered)
2840 * g7 = pc we jumped here from (in)
2841 */
2842
2843	! Do not disturb %g5, it will be used after the trace
2844	ALTENTRY(trace_tsbhit)
2845	TRACE_TSBHIT(0)
2846	jmp	%g7 + 4
2847	nop
2848
2849/*
2850 * Trace a TSB miss
2851 *
2852 * g1 = tsb8k pointer (in)
2853 * g2 = tag access register (in)
2854 * g3 = tsb4m pointer (in)
2855 * g4 = tsbe tag (in/clobbered)
2856 * g5 - g6 = scratch (clobbered)
2857 * g7 = pc we jumped here from (in)
2858 */
2859	.global	trace_tsbmiss
2860trace_tsbmiss:
2861	membar	#Sync
2862	sethi	%hi(FLUSH_ADDR), %g6
2863	flush	%g6
2864	TRACE_PTR(%g5, %g6)
2865	GET_TRACE_TICK(%g6)
2866	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2867	stxa	%g2, [%g5 + TRAP_ENT_SP]%asi		! tag access
2868	stxa	%g4, [%g5 + TRAP_ENT_F1]%asi		! tsb tag
2869	rdpr	%tnpc, %g6
2870	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2871	stna	%g1, [%g5 + TRAP_ENT_F3]%asi		! tsb8k pointer
2872	srlx	%g1, 32, %g6
2873	stna	%g6, [%g5 + TRAP_ENT_F4]%asi		! huh?
2874	rdpr	%tpc, %g6
2875	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2876	rdpr	%tl, %g6
2877	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2878	rdpr	%tt, %g6
2879	or	%g6, TT_MMU_MISS, %g4
2880	stha	%g4, [%g5 + TRAP_ENT_TT]%asi
2881	cmp	%g6, FAST_IMMU_MISS_TT
2882	be,a	%icc, 1f
2883	  ldxa	[%g0]ASI_IMMU, %g6
2884	ldxa	[%g0]ASI_DMMU, %g6
28851:	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi	! tag target
2886	stxa	%g3, [%g5 + TRAP_ENT_TR]%asi		! tsb4m pointer
2887	TRACE_NEXT(%g5, %g4, %g6)
2888	jmp	%g7 + 4
2889	nop
2890
2891/*
2892 * g2 = tag access register (in)
2893 * g3 = ctx number (in)
2894 */
2895trace_dataprot:
2896	membar	#Sync
2897	sethi	%hi(FLUSH_ADDR), %g6
2898	flush	%g6
2899	TRACE_PTR(%g1, %g6)
2900	GET_TRACE_TICK(%g6)
2901	stxa	%g6, [%g1 + TRAP_ENT_TICK]%asi
2902	rdpr	%tpc, %g6
2903	stna	%g6, [%g1 + TRAP_ENT_TPC]%asi
2904	rdpr	%tstate, %g6
2905	stxa	%g6, [%g1 + TRAP_ENT_TSTATE]%asi
2906	stxa	%g2, [%g1 + TRAP_ENT_SP]%asi		! tag access reg
2907	stxa	%g0, [%g1 + TRAP_ENT_TR]%asi
2908	stxa	%g0, [%g1 + TRAP_ENT_F1]%asi
2909	stxa	%g0, [%g1 + TRAP_ENT_F2]%asi
2910	stxa	%g0, [%g1 + TRAP_ENT_F3]%asi
2911	stxa	%g0, [%g1 + TRAP_ENT_F4]%asi
2912	rdpr	%tl, %g6
2913	stha	%g6, [%g1 + TRAP_ENT_TL]%asi
2914	rdpr	%tt, %g6
2915	stha	%g6, [%g1 + TRAP_ENT_TT]%asi
2916	TRACE_NEXT(%g1, %g4, %g5)
2917	jmp	%g7 + 4
2918	nop
2919
2920#endif /* TRAPTRACE */
2921
2922/*
2923 * fast_trap_done, fast_trap_done_chk_intr:
2924 *
2925 * Due to the design of UltraSPARC pipeline, pending interrupts are not
2926 * taken immediately after a RETRY or DONE instruction which causes IE to
2927 * go from 0 to 1. Instead, the instruction at %tpc or %tnpc is allowed
2928 * to execute first before taking any interrupts. If that instruction
2929 * results in other traps, and if the corresponding trap handler runs
2930 * entirely at TL=1 with interrupts disabled, then pending interrupts
2931 * won't be taken until after yet another instruction following the %tpc
2932 * or %tnpc.
2933 *
2934 * A malicious user program can use this feature to block out interrupts
2935 * for extended durations, which can result in send_mondo_timeout kernel
2936 * panic.
2937 *
2938 * This problem is addressed by servicing any pending interrupts via
2939 * sys_trap before returning back to the user mode from a fast trap
2940 * handler. The "done" instruction within a fast trap handler, which
2941 * runs entirely at TL=1 with interrupts disabled, is replaced with the
2942 * FAST_TRAP_DONE macro, which branches control to this fast_trap_done
2943 * entry point.
2944 *
2945 * We check for any pending interrupts here and force a sys_trap to
2946 * service those interrupts, if any. To minimize overhead, pending
2947 * interrupts are checked if the %tpc happens to be at 16K boundary,
2948 * which allows a malicious program to execute at most 4K consecutive
2949 * instructions before we service any pending interrupts. If a worst
2950 * case fast trap handler takes about 2 usec, then interrupts will be
2951 * blocked for at most 8 msec, less than a clock tick.
2952 *
2953 * For the cases where we don't know if the %tpc will cross a 16K
2954 * boundary, we can't use the above optimization and always process
2955 * any pending interrupts via fast_frap_done_chk_intr entry point.
2956 *
2957 * Entry Conditions:
2958 * 	%pstate		am:0 priv:1 ie:0
2959 * 			globals are AG (not normal globals)
2960 */
2961
2962	.global	fast_trap_done, fast_trap_done_chk_intr
2963fast_trap_done:
2964	rdpr	%tpc, %g5
2965	sethi	%hi(0xffffc000), %g6	! 1's complement of 0x3fff
2966	andncc	%g5, %g6, %g0		! check lower 14 bits of %tpc
2967	bz,a,pn	%icc, 1f		! branch if zero (lower 32 bits only)
2968	  ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2969	done
2970
2971	ALTENTRY(fast_trap_done_check_interrupts)
2972fast_trap_done_chk_intr:
2973	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2974
29751:	rd	SOFTINT, %g6
2976	and	%g5, IRSR_BUSY, %g5
2977	orcc	%g5, %g6, %g0
2978	bnz,pn	%xcc, 2f		! branch if any pending intr
2979	nop
2980	done
2981
29822:
2983	/*
2984	 * We get here if there are any pending interrupts.
2985	 * Adjust %tpc/%tnpc as we'll be resuming via "retry"
2986	 * instruction.
2987	 */
2988	rdpr	%tnpc, %g5
2989	wrpr	%g0, %g5, %tpc
2990	add	%g5, 4, %g5
2991	wrpr	%g0, %g5, %tnpc
2992
2993	/*
2994	 * Force a dummy sys_trap call so that interrupts can be serviced.
2995	 */
2996	set	fast_trap_dummy_call, %g1
2997	ba,pt	%xcc, sys_trap
2998	  mov	-1, %g4
2999
3000fast_trap_dummy_call:
3001	retl
3002	nop
3003
3004/*
3005 * Currently the brand syscall interposition code is not enabled by
3006 * default.  Instead, when a branded zone is first booted the brand
3007 * infrastructure will patch the trap table so that the syscall
3008 * entry points are redirected to syscall_wrapper32 and syscall_wrapper
3009 * for ILP32 and LP64 syscalls respectively.  this is done in
3010 * brand_plat_interposition_enable().  Note that the syscall wrappers
3011 * below do not collect any trap trace data since the syscall hot patch
3012 * points are reached after trap trace data has already been collected.
3013 */
3014#define	BRAND_CALLBACK(callback_id)					    \
3015	CPU_ADDR(%g2, %g1)		/* load CPU struct addr to %g2	*/ ;\
3016	ldn	[%g2 + CPU_THREAD], %g3	/* load thread pointer		*/ ;\
3017	ldn	[%g3 + T_PROCP], %g3	/* get proc pointer		*/ ;\
3018	ldn	[%g3 + P_BRAND], %g3	/* get brand pointer		*/ ;\
3019	brz	%g3, 1f			/* No brand?  No callback. 	*/ ;\
3020	nop 								   ;\
3021	ldn	[%g3 + B_MACHOPS], %g3	/* get machops list		*/ ;\
3022	ldn	[%g3 + (callback_id << 3)], %g3 			   ;\
3023	brz	%g3, 1f							   ;\
3024	/*								    \
3025	 * This isn't pretty.  We want a low-latency way for the callback   \
3026	 * routine to decline to do anything.  We just pass in an address   \
3027	 * the routine can directly jmp back to, pretending that nothing    \
3028	 * has happened.						    \
3029	 * 								    \
3030	 * %g1: return address (where the brand handler jumps back to)	    \
3031	 * %g2: address of CPU structure				    \
3032	 * %g3: address of brand handler (where we will jump to)	    \
3033	 */								    \
3034	mov	%pc, %g1						   ;\
3035	add	%g1, 16, %g1						   ;\
3036	jmp	%g3							   ;\
3037	nop								   ;\
30381:
3039
3040	ENTRY_NP(syscall_wrapper32)
3041	BRAND_CALLBACK(BRAND_CB_SYSCALL32)
3042	SYSCALL_NOTT(syscall_trap32)
3043	SET_SIZE(syscall_wrapper32)
3044
3045	ENTRY_NP(syscall_wrapper)
3046	BRAND_CALLBACK(BRAND_CB_SYSCALL)
3047	SYSCALL_NOTT(syscall_trap)
3048	SET_SIZE(syscall_wrapper)
3049
3050#endif	/* lint */
3051