xref: /illumos-gate/usr/src/uts/sun4u/ml/trap_table.S (revision 20a7641f9918de8574b8b3b47dbe35c4bfc78df1)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#include "assym.h"
27#include <sys/asm_linkage.h>
28#include <sys/privregs.h>
29#include <sys/sun4asi.h>
30#include <sys/spitregs.h>
31#include <sys/cheetahregs.h>
32#include <sys/machtrap.h>
33#include <sys/machthread.h>
34#include <sys/machbrand.h>
35#include <sys/pcb.h>
36#include <sys/pte.h>
37#include <sys/mmu.h>
38#include <sys/machpcb.h>
39#include <sys/async.h>
40#include <sys/intreg.h>
41#include <sys/scb.h>
42#include <sys/psr_compat.h>
43#include <sys/syscall.h>
44#include <sys/machparam.h>
45#include <sys/traptrace.h>
46#include <vm/hat_sfmmu.h>
47#include <sys/archsystm.h>
48#include <sys/utrap.h>
49#include <sys/clock.h>
50#include <sys/intr.h>
51#include <sys/fpu/fpu_simulator.h>
52#include <vm/seg_spt.h>
53
54/*
55 * WARNING: If you add a fast trap handler which can be invoked by a
56 * non-privileged user, you may have to use the FAST_TRAP_DONE macro
57 * instead of "done" instruction to return back to the user mode. See
58 * comments for the "fast_trap_done" entry point for more information.
59 *
60 * An alternate FAST_TRAP_DONE_CHK_INTR macro should be used for the
61 * cases where you always want to process any pending interrupts before
62 * returning back to the user mode.
63 */
64#define	FAST_TRAP_DONE		\
65	ba,a	fast_trap_done
66
67#define	FAST_TRAP_DONE_CHK_INTR	\
68	ba,a	fast_trap_done_chk_intr
69
70/*
71 * SPARC V9 Trap Table
72 *
73 * Most of the trap handlers are made from common building
74 * blocks, and some are instantiated multiple times within
75 * the trap table. So, I build a bunch of macros, then
76 * populate the table using only the macros.
77 *
78 * Many macros branch to sys_trap.  Its calling convention is:
79 *	%g1		kernel trap handler
80 *	%g2, %g3	args for above
81 *	%g4		desire %pil
82 */
83
84#ifdef	TRAPTRACE
85
86/*
87 * Tracing macro. Adds two instructions if TRAPTRACE is defined.
88 */
89#define	TT_TRACE(label)		\
90	ba	label		;\
91	rd	%pc, %g7
92#define	TT_TRACE_INS	2
93
94#define	TT_TRACE_L(label)	\
95	ba	label		;\
96	rd	%pc, %l4	;\
97	clr	%l4
98#define	TT_TRACE_L_INS	3
99
100#else
101
102#define	TT_TRACE(label)
103#define	TT_TRACE_INS	0
104
105#define	TT_TRACE_L(label)
106#define	TT_TRACE_L_INS	0
107
108#endif
109
110/*
111 * This first set are funneled to trap() with %tt as the type.
112 * Trap will then either panic or send the user a signal.
113 */
114/*
115 * NOT is used for traps that just shouldn't happen.
116 * It comes in both single and quadruple flavors.
117 */
118	.global	trap
119#define	NOT			\
120	TT_TRACE(trace_gen)	;\
121	set	trap, %g1	;\
122	rdpr	%tt, %g3	;\
123	ba,pt	%xcc, sys_trap	;\
124	sub	%g0, 1, %g4	;\
125	.align	32
126#define	NOT4	NOT; NOT; NOT; NOT
127/*
128 * RED is for traps that use the red mode handler.
129 * We should never see these either.
130 */
131#define	RED	NOT
132/*
133 * BAD is used for trap vectors we don't have a kernel
134 * handler for.
135 * It also comes in single and quadruple versions.
136 */
137#define	BAD	NOT
138#define	BAD4	NOT4
139
140#define	DONE			\
141	done;			\
142	.align	32
143
144/*
145 * TRAP vectors to the trap() function.
146 * It's main use is for user errors.
147 */
148	.global	trap
149#define	TRAP(arg)		\
150	TT_TRACE(trace_gen)	;\
151	set	trap, %g1	;\
152	mov	arg, %g3	;\
153	ba,pt	%xcc, sys_trap	;\
154	sub	%g0, 1, %g4	;\
155	.align	32
156
157/*
158 * SYSCALL is used for unsupported syscall interfaces (with 'which'
159 * set to 'nosys') and legacy support of old SunOS 4.x syscalls (with
160 * 'which' set to 'syscall_trap32').
161 *
162 * The SYSCALL_TRAP* macros are used for syscall entry points.
163 * SYSCALL_TRAP is used to support LP64 syscalls and SYSCALL_TRAP32
164 * is used to support ILP32.  Each macro can only be used once
165 * since they each define a symbol.  The symbols are used as hot patch
166 * points by the brand infrastructure to dynamically enable and disable
167 * brand syscall interposition.  See the comments around BRAND_CALLBACK
168 * and brand_plat_interposition_enable() for more information.
169 */
170#define	SYSCALL_NOTT(which)		\
171	set	(which), %g1		;\
172	ba,pt	%xcc, sys_trap		;\
173	sub	%g0, 1, %g4		;\
174	.align	32
175
176#define	SYSCALL(which)			\
177	TT_TRACE(trace_gen)		;\
178	SYSCALL_NOTT(which)
179
180#define	SYSCALL_TRAP32				\
181	TT_TRACE(trace_gen)			;\
182	ALTENTRY(syscall_trap32_patch_point)	\
183	SYSCALL_NOTT(syscall_trap32)
184
185#define	SYSCALL_TRAP				\
186	TT_TRACE(trace_gen)			;\
187	ALTENTRY(syscall_trap_patch_point)	\
188	SYSCALL_NOTT(syscall_trap)
189
190#define	FLUSHW(h_name)			\
191	.global h_name			;\
192h_name:					;\
193	set	trap, %g1		;\
194	mov	T_FLUSHW, %g3		;\
195	sub	%g0, 1, %g4		;\
196	save				;\
197	flushw				;\
198	restore				;\
199	FAST_TRAP_DONE			;\
200	.align	32
201
202/*
203 * GOTO just jumps to a label.
204 * It's used for things that can be fixed without going thru sys_trap.
205 */
206#define	GOTO(label)		\
207	.global	label		;\
208	ba,a	label		;\
209	.empty			;\
210	.align	32
211
212/*
213 * GOTO_TT just jumps to a label.
214 * correctable ECC error traps at  level 0 and 1 will use this macro.
215 * It's used for things that can be fixed without going thru sys_trap.
216 */
217#define	GOTO_TT(label, ttlabel)		\
218	.global	label		;\
219	TT_TRACE(ttlabel)	;\
220	ba,a	label		;\
221	.empty			;\
222	.align	32
223
224/*
225 * Privileged traps
226 * Takes breakpoint if privileged, calls trap() if not.
227 */
228#define	PRIV(label)			\
229	rdpr	%tstate, %g1		;\
230	btst	TSTATE_PRIV, %g1	;\
231	bnz	label			;\
232	rdpr	%tt, %g3		;\
233	set	trap, %g1		;\
234	ba,pt	%xcc, sys_trap		;\
235	sub	%g0, 1, %g4		;\
236	.align	32
237
238
239/*
240 * DTrace traps.
241 */
242#define	DTRACE_PID			\
243	.global dtrace_pid_probe				;\
244	set	dtrace_pid_probe, %g1				;\
245	ba,pt	%xcc, user_trap					;\
246	sub	%g0, 1, %g4					;\
247	.align	32
248
249#define	DTRACE_RETURN			\
250	.global dtrace_return_probe				;\
251	set	dtrace_return_probe, %g1			;\
252	ba,pt	%xcc, user_trap					;\
253	sub	%g0, 1, %g4					;\
254	.align	32
255
256/*
257 * REGISTER WINDOW MANAGEMENT MACROS
258 */
259
260/*
261 * various convenient units of padding
262 */
263#define	SKIP(n)	.skip 4*(n)
264
265/*
266 * CLEAN_WINDOW is the simple handler for cleaning a register window.
267 */
268#define	CLEAN_WINDOW						\
269	TT_TRACE_L(trace_win)					;\
270	rdpr %cleanwin, %l0; inc %l0; wrpr %l0, %cleanwin	;\
271	clr %l0; clr %l1; clr %l2; clr %l3			;\
272	clr %l4; clr %l5; clr %l6; clr %l7			;\
273	clr %o0; clr %o1; clr %o2; clr %o3			;\
274	clr %o4; clr %o5; clr %o6; clr %o7			;\
275	retry; .align 128
276
277/*
278 * If we get an unresolved tlb miss while in a window handler, the fault
279 * handler will resume execution at the last instruction of the window
280 * hander, instead of delivering the fault to the kernel.  Spill handlers
281 * use this to spill windows into the wbuf.
282 *
283 * The mixed handler works by checking %sp, and branching to the correct
284 * handler.  This is done by branching back to label 1: for 32b frames,
285 * or label 2: for 64b frames; which implies the handler order is: 32b,
286 * 64b, mixed.  The 1: and 2: labels are offset into the routines to
287 * allow the branchs' delay slots to contain useful instructions.
288 */
289
290/*
291 * SPILL_32bit spills a 32-bit-wide kernel register window.  It
292 * assumes that the kernel context and the nucleus context are the
293 * same.  The stack pointer is required to be eight-byte aligned even
294 * though this code only needs it to be four-byte aligned.
295 */
296#define	SPILL_32bit(tail)					\
297	srl	%sp, 0, %sp					;\
2981:	st	%l0, [%sp + 0]					;\
299	st	%l1, [%sp + 4]					;\
300	st	%l2, [%sp + 8]					;\
301	st	%l3, [%sp + 12]					;\
302	st	%l4, [%sp + 16]					;\
303	st	%l5, [%sp + 20]					;\
304	st	%l6, [%sp + 24]					;\
305	st	%l7, [%sp + 28]					;\
306	st	%i0, [%sp + 32]					;\
307	st	%i1, [%sp + 36]					;\
308	st	%i2, [%sp + 40]					;\
309	st	%i3, [%sp + 44]					;\
310	st	%i4, [%sp + 48]					;\
311	st	%i5, [%sp + 52]					;\
312	st	%i6, [%sp + 56]					;\
313	st	%i7, [%sp + 60]					;\
314	TT_TRACE_L(trace_win)					;\
315	saved							;\
316	retry							;\
317	SKIP(31-19-TT_TRACE_L_INS)				;\
318	ba,a,pt	%xcc, fault_32bit_##tail			;\
319	.empty
320
321/*
322 * SPILL_32bit_asi spills a 32-bit-wide register window into a 32-bit
323 * wide address space via the designated asi.  It is used to spill
324 * non-kernel windows.  The stack pointer is required to be eight-byte
325 * aligned even though this code only needs it to be four-byte
326 * aligned.
327 */
328#define	SPILL_32bit_asi(asi_num, tail)				\
329	srl	%sp, 0, %sp					;\
3301:	sta	%l0, [%sp + %g0]asi_num				;\
331	mov	4, %g1						;\
332	sta	%l1, [%sp + %g1]asi_num				;\
333	mov	8, %g2						;\
334	sta	%l2, [%sp + %g2]asi_num				;\
335	mov	12, %g3						;\
336	sta	%l3, [%sp + %g3]asi_num				;\
337	add	%sp, 16, %g4					;\
338	sta	%l4, [%g4 + %g0]asi_num				;\
339	sta	%l5, [%g4 + %g1]asi_num				;\
340	sta	%l6, [%g4 + %g2]asi_num				;\
341	sta	%l7, [%g4 + %g3]asi_num				;\
342	add	%g4, 16, %g4					;\
343	sta	%i0, [%g4 + %g0]asi_num				;\
344	sta	%i1, [%g4 + %g1]asi_num				;\
345	sta	%i2, [%g4 + %g2]asi_num				;\
346	sta	%i3, [%g4 + %g3]asi_num				;\
347	add	%g4, 16, %g4					;\
348	sta	%i4, [%g4 + %g0]asi_num				;\
349	sta	%i5, [%g4 + %g1]asi_num				;\
350	sta	%i6, [%g4 + %g2]asi_num				;\
351	sta	%i7, [%g4 + %g3]asi_num				;\
352	TT_TRACE_L(trace_win)					;\
353	saved							;\
354	retry							;\
355	SKIP(31-25-TT_TRACE_L_INS)				;\
356	ba,a,pt %xcc, fault_32bit_##tail			;\
357	.empty
358
359/*
360 * SPILL_32bit_tt1 spills a 32-bit-wide register window into a 32-bit
361 * wide address space via the designated asi.  It is used to spill
362 * windows at tl>1 where performance isn't the primary concern and
363 * where we don't want to use unnecessary registers.  The stack
364 * pointer is required to be eight-byte aligned even though this code
365 * only needs it to be four-byte aligned.
366 */
367#define	SPILL_32bit_tt1(asi_num, tail)				\
368	mov	asi_num, %asi					;\
3691:	srl	%sp, 0, %sp					;\
370	sta	%l0, [%sp + 0]%asi				;\
371	sta	%l1, [%sp + 4]%asi				;\
372	sta	%l2, [%sp + 8]%asi				;\
373	sta	%l3, [%sp + 12]%asi				;\
374	sta	%l4, [%sp + 16]%asi				;\
375	sta	%l5, [%sp + 20]%asi				;\
376	sta	%l6, [%sp + 24]%asi				;\
377	sta	%l7, [%sp + 28]%asi				;\
378	sta	%i0, [%sp + 32]%asi				;\
379	sta	%i1, [%sp + 36]%asi				;\
380	sta	%i2, [%sp + 40]%asi				;\
381	sta	%i3, [%sp + 44]%asi				;\
382	sta	%i4, [%sp + 48]%asi				;\
383	sta	%i5, [%sp + 52]%asi				;\
384	sta	%i6, [%sp + 56]%asi				;\
385	sta	%i7, [%sp + 60]%asi				;\
386	TT_TRACE_L(trace_win)					;\
387	saved							;\
388	retry							;\
389	SKIP(31-20-TT_TRACE_L_INS)				;\
390	ba,a,pt	%xcc, fault_32bit_##tail			;\
391	.empty
392
393
394/*
395 * FILL_32bit fills a 32-bit-wide kernel register window.  It assumes
396 * that the kernel context and the nucleus context are the same.  The
397 * stack pointer is required to be eight-byte aligned even though this
398 * code only needs it to be four-byte aligned.
399 */
400#define	FILL_32bit(tail)					\
401	srl	%sp, 0, %sp					;\
4021:	TT_TRACE_L(trace_win)					;\
403	ld	[%sp + 0], %l0					;\
404	ld	[%sp + 4], %l1					;\
405	ld	[%sp + 8], %l2					;\
406	ld	[%sp + 12], %l3					;\
407	ld	[%sp + 16], %l4					;\
408	ld	[%sp + 20], %l5					;\
409	ld	[%sp + 24], %l6					;\
410	ld	[%sp + 28], %l7					;\
411	ld	[%sp + 32], %i0					;\
412	ld	[%sp + 36], %i1					;\
413	ld	[%sp + 40], %i2					;\
414	ld	[%sp + 44], %i3					;\
415	ld	[%sp + 48], %i4					;\
416	ld	[%sp + 52], %i5					;\
417	ld	[%sp + 56], %i6					;\
418	ld	[%sp + 60], %i7					;\
419	restored						;\
420	retry							;\
421	SKIP(31-19-TT_TRACE_L_INS)				;\
422	ba,a,pt	%xcc, fault_32bit_##tail			;\
423	.empty
424
425/*
426 * FILL_32bit_asi fills a 32-bit-wide register window from a 32-bit
427 * wide address space via the designated asi.  It is used to fill
428 * non-kernel windows.  The stack pointer is required to be eight-byte
429 * aligned even though this code only needs it to be four-byte
430 * aligned.
431 */
432#define	FILL_32bit_asi(asi_num, tail)				\
433	srl	%sp, 0, %sp					;\
4341:	TT_TRACE_L(trace_win)					;\
435	mov	4, %g1						;\
436	lda	[%sp + %g0]asi_num, %l0				;\
437	mov	8, %g2						;\
438	lda	[%sp + %g1]asi_num, %l1				;\
439	mov	12, %g3						;\
440	lda	[%sp + %g2]asi_num, %l2				;\
441	lda	[%sp + %g3]asi_num, %l3				;\
442	add	%sp, 16, %g4					;\
443	lda	[%g4 + %g0]asi_num, %l4				;\
444	lda	[%g4 + %g1]asi_num, %l5				;\
445	lda	[%g4 + %g2]asi_num, %l6				;\
446	lda	[%g4 + %g3]asi_num, %l7				;\
447	add	%g4, 16, %g4					;\
448	lda	[%g4 + %g0]asi_num, %i0				;\
449	lda	[%g4 + %g1]asi_num, %i1				;\
450	lda	[%g4 + %g2]asi_num, %i2				;\
451	lda	[%g4 + %g3]asi_num, %i3				;\
452	add	%g4, 16, %g4					;\
453	lda	[%g4 + %g0]asi_num, %i4				;\
454	lda	[%g4 + %g1]asi_num, %i5				;\
455	lda	[%g4 + %g2]asi_num, %i6				;\
456	lda	[%g4 + %g3]asi_num, %i7				;\
457	restored						;\
458	retry							;\
459	SKIP(31-25-TT_TRACE_L_INS)				;\
460	ba,a,pt %xcc, fault_32bit_##tail			;\
461	.empty
462
463/*
464 * FILL_32bit_tt1 fills a 32-bit-wide register window from a 32-bit
465 * wide address space via the designated asi.  It is used to fill
466 * windows at tl>1 where performance isn't the primary concern and
467 * where we don't want to use unnecessary registers.  The stack
468 * pointer is required to be eight-byte aligned even though this code
469 * only needs it to be four-byte aligned.
470 */
471#define	FILL_32bit_tt1(asi_num, tail)				\
472	mov	asi_num, %asi					;\
4731:	srl	%sp, 0, %sp					;\
474	TT_TRACE_L(trace_win)					;\
475	lda	[%sp + 0]%asi, %l0				;\
476	lda	[%sp + 4]%asi, %l1				;\
477	lda	[%sp + 8]%asi, %l2				;\
478	lda	[%sp + 12]%asi, %l3				;\
479	lda	[%sp + 16]%asi, %l4				;\
480	lda	[%sp + 20]%asi, %l5				;\
481	lda	[%sp + 24]%asi, %l6				;\
482	lda	[%sp + 28]%asi, %l7				;\
483	lda	[%sp + 32]%asi, %i0				;\
484	lda	[%sp + 36]%asi, %i1				;\
485	lda	[%sp + 40]%asi, %i2				;\
486	lda	[%sp + 44]%asi, %i3				;\
487	lda	[%sp + 48]%asi, %i4				;\
488	lda	[%sp + 52]%asi, %i5				;\
489	lda	[%sp + 56]%asi, %i6				;\
490	lda	[%sp + 60]%asi, %i7				;\
491	restored						;\
492	retry							;\
493	SKIP(31-20-TT_TRACE_L_INS)				;\
494	ba,a,pt	%xcc, fault_32bit_##tail			;\
495	.empty
496
497
498/*
499 * SPILL_64bit spills a 64-bit-wide kernel register window.  It
500 * assumes that the kernel context and the nucleus context are the
501 * same.  The stack pointer is required to be eight-byte aligned.
502 */
503#define	SPILL_64bit(tail)					\
5042:	stx	%l0, [%sp + V9BIAS64 + 0]			;\
505	stx	%l1, [%sp + V9BIAS64 + 8]			;\
506	stx	%l2, [%sp + V9BIAS64 + 16]			;\
507	stx	%l3, [%sp + V9BIAS64 + 24]			;\
508	stx	%l4, [%sp + V9BIAS64 + 32]			;\
509	stx	%l5, [%sp + V9BIAS64 + 40]			;\
510	stx	%l6, [%sp + V9BIAS64 + 48]			;\
511	stx	%l7, [%sp + V9BIAS64 + 56]			;\
512	stx	%i0, [%sp + V9BIAS64 + 64]			;\
513	stx	%i1, [%sp + V9BIAS64 + 72]			;\
514	stx	%i2, [%sp + V9BIAS64 + 80]			;\
515	stx	%i3, [%sp + V9BIAS64 + 88]			;\
516	stx	%i4, [%sp + V9BIAS64 + 96]			;\
517	stx	%i5, [%sp + V9BIAS64 + 104]			;\
518	stx	%i6, [%sp + V9BIAS64 + 112]			;\
519	stx	%i7, [%sp + V9BIAS64 + 120]			;\
520	TT_TRACE_L(trace_win)					;\
521	saved							;\
522	retry							;\
523	SKIP(31-18-TT_TRACE_L_INS)				;\
524	ba,a,pt	%xcc, fault_64bit_##tail			;\
525	.empty
526
527/*
528 * SPILL_64bit_asi spills a 64-bit-wide register window into a 64-bit
529 * wide address space via the designated asi.  It is used to spill
530 * non-kernel windows.  The stack pointer is required to be eight-byte
531 * aligned.
532 */
533#define	SPILL_64bit_asi(asi_num, tail)				\
534	mov	0 + V9BIAS64, %g1				;\
5352:	stxa	%l0, [%sp + %g1]asi_num				;\
536	mov	8 + V9BIAS64, %g2				;\
537	stxa	%l1, [%sp + %g2]asi_num				;\
538	mov	16 + V9BIAS64, %g3				;\
539	stxa	%l2, [%sp + %g3]asi_num				;\
540	mov	24 + V9BIAS64, %g4				;\
541	stxa	%l3, [%sp + %g4]asi_num				;\
542	add	%sp, 32, %g5					;\
543	stxa	%l4, [%g5 + %g1]asi_num				;\
544	stxa	%l5, [%g5 + %g2]asi_num				;\
545	stxa	%l6, [%g5 + %g3]asi_num				;\
546	stxa	%l7, [%g5 + %g4]asi_num				;\
547	add	%g5, 32, %g5					;\
548	stxa	%i0, [%g5 + %g1]asi_num				;\
549	stxa	%i1, [%g5 + %g2]asi_num				;\
550	stxa	%i2, [%g5 + %g3]asi_num				;\
551	stxa	%i3, [%g5 + %g4]asi_num				;\
552	add	%g5, 32, %g5					;\
553	stxa	%i4, [%g5 + %g1]asi_num				;\
554	stxa	%i5, [%g5 + %g2]asi_num				;\
555	stxa	%i6, [%g5 + %g3]asi_num				;\
556	stxa	%i7, [%g5 + %g4]asi_num				;\
557	TT_TRACE_L(trace_win)					;\
558	saved							;\
559	retry							;\
560	SKIP(31-25-TT_TRACE_L_INS)				;\
561	ba,a,pt %xcc, fault_64bit_##tail			;\
562	.empty
563
564/*
565 * SPILL_64bit_tt1 spills a 64-bit-wide register window into a 64-bit
566 * wide address space via the designated asi.  It is used to spill
567 * windows at tl>1 where performance isn't the primary concern and
568 * where we don't want to use unnecessary registers.  The stack
569 * pointer is required to be eight-byte aligned.
570 */
571#define	SPILL_64bit_tt1(asi_num, tail)				\
572	mov	asi_num, %asi					;\
5732:	stxa	%l0, [%sp + V9BIAS64 + 0]%asi			;\
574	stxa	%l1, [%sp + V9BIAS64 + 8]%asi			;\
575	stxa	%l2, [%sp + V9BIAS64 + 16]%asi			;\
576	stxa	%l3, [%sp + V9BIAS64 + 24]%asi			;\
577	stxa	%l4, [%sp + V9BIAS64 + 32]%asi			;\
578	stxa	%l5, [%sp + V9BIAS64 + 40]%asi			;\
579	stxa	%l6, [%sp + V9BIAS64 + 48]%asi			;\
580	stxa	%l7, [%sp + V9BIAS64 + 56]%asi			;\
581	stxa	%i0, [%sp + V9BIAS64 + 64]%asi			;\
582	stxa	%i1, [%sp + V9BIAS64 + 72]%asi			;\
583	stxa	%i2, [%sp + V9BIAS64 + 80]%asi			;\
584	stxa	%i3, [%sp + V9BIAS64 + 88]%asi			;\
585	stxa	%i4, [%sp + V9BIAS64 + 96]%asi			;\
586	stxa	%i5, [%sp + V9BIAS64 + 104]%asi			;\
587	stxa	%i6, [%sp + V9BIAS64 + 112]%asi			;\
588	stxa	%i7, [%sp + V9BIAS64 + 120]%asi			;\
589	TT_TRACE_L(trace_win)					;\
590	saved							;\
591	retry							;\
592	SKIP(31-19-TT_TRACE_L_INS)				;\
593	ba,a,pt	%xcc, fault_64bit_##tail			;\
594	.empty
595
596
597/*
598 * FILL_64bit fills a 64-bit-wide kernel register window.  It assumes
599 * that the kernel context and the nucleus context are the same.  The
600 * stack pointer is required to be eight-byte aligned.
601 */
602#define	FILL_64bit(tail)					\
6032:	TT_TRACE_L(trace_win)					;\
604	ldx	[%sp + V9BIAS64 + 0], %l0			;\
605	ldx	[%sp + V9BIAS64 + 8], %l1			;\
606	ldx	[%sp + V9BIAS64 + 16], %l2			;\
607	ldx	[%sp + V9BIAS64 + 24], %l3			;\
608	ldx	[%sp + V9BIAS64 + 32], %l4			;\
609	ldx	[%sp + V9BIAS64 + 40], %l5			;\
610	ldx	[%sp + V9BIAS64 + 48], %l6			;\
611	ldx	[%sp + V9BIAS64 + 56], %l7			;\
612	ldx	[%sp + V9BIAS64 + 64], %i0			;\
613	ldx	[%sp + V9BIAS64 + 72], %i1			;\
614	ldx	[%sp + V9BIAS64 + 80], %i2			;\
615	ldx	[%sp + V9BIAS64 + 88], %i3			;\
616	ldx	[%sp + V9BIAS64 + 96], %i4			;\
617	ldx	[%sp + V9BIAS64 + 104], %i5			;\
618	ldx	[%sp + V9BIAS64 + 112], %i6			;\
619	ldx	[%sp + V9BIAS64 + 120], %i7			;\
620	restored						;\
621	retry							;\
622	SKIP(31-18-TT_TRACE_L_INS)				;\
623	ba,a,pt	%xcc, fault_64bit_##tail			;\
624	.empty
625
626/*
627 * FILL_64bit_asi fills a 64-bit-wide register window from a 64-bit
628 * wide address space via the designated asi.  It is used to fill
629 * non-kernel windows.  The stack pointer is required to be eight-byte
630 * aligned.
631 */
632#define	FILL_64bit_asi(asi_num, tail)				\
633	mov	V9BIAS64 + 0, %g1				;\
6342:	TT_TRACE_L(trace_win)					;\
635	ldxa	[%sp + %g1]asi_num, %l0				;\
636	mov	V9BIAS64 + 8, %g2				;\
637	ldxa	[%sp + %g2]asi_num, %l1				;\
638	mov	V9BIAS64 + 16, %g3				;\
639	ldxa	[%sp + %g3]asi_num, %l2				;\
640	mov	V9BIAS64 + 24, %g4				;\
641	ldxa	[%sp + %g4]asi_num, %l3				;\
642	add	%sp, 32, %g5					;\
643	ldxa	[%g5 + %g1]asi_num, %l4				;\
644	ldxa	[%g5 + %g2]asi_num, %l5				;\
645	ldxa	[%g5 + %g3]asi_num, %l6				;\
646	ldxa	[%g5 + %g4]asi_num, %l7				;\
647	add	%g5, 32, %g5					;\
648	ldxa	[%g5 + %g1]asi_num, %i0				;\
649	ldxa	[%g5 + %g2]asi_num, %i1				;\
650	ldxa	[%g5 + %g3]asi_num, %i2				;\
651	ldxa	[%g5 + %g4]asi_num, %i3				;\
652	add	%g5, 32, %g5					;\
653	ldxa	[%g5 + %g1]asi_num, %i4				;\
654	ldxa	[%g5 + %g2]asi_num, %i5				;\
655	ldxa	[%g5 + %g3]asi_num, %i6				;\
656	ldxa	[%g5 + %g4]asi_num, %i7				;\
657	restored						;\
658	retry							;\
659	SKIP(31-25-TT_TRACE_L_INS)				;\
660	ba,a,pt	%xcc, fault_64bit_##tail			;\
661	.empty
662
663/*
664 * FILL_64bit_tt1 fills a 64-bit-wide register window from a 64-bit
665 * wide address space via the designated asi.  It is used to fill
666 * windows at tl>1 where performance isn't the primary concern and
667 * where we don't want to use unnecessary registers.  The stack
668 * pointer is required to be eight-byte aligned.
669 */
670#define	FILL_64bit_tt1(asi_num, tail)				\
671	mov	asi_num, %asi					;\
672	TT_TRACE_L(trace_win)					;\
673	ldxa	[%sp + V9BIAS64 + 0]%asi, %l0			;\
674	ldxa	[%sp + V9BIAS64 + 8]%asi, %l1			;\
675	ldxa	[%sp + V9BIAS64 + 16]%asi, %l2			;\
676	ldxa	[%sp + V9BIAS64 + 24]%asi, %l3			;\
677	ldxa	[%sp + V9BIAS64 + 32]%asi, %l4			;\
678	ldxa	[%sp + V9BIAS64 + 40]%asi, %l5			;\
679	ldxa	[%sp + V9BIAS64 + 48]%asi, %l6			;\
680	ldxa	[%sp + V9BIAS64 + 56]%asi, %l7			;\
681	ldxa	[%sp + V9BIAS64 + 64]%asi, %i0			;\
682	ldxa	[%sp + V9BIAS64 + 72]%asi, %i1			;\
683	ldxa	[%sp + V9BIAS64 + 80]%asi, %i2			;\
684	ldxa	[%sp + V9BIAS64 + 88]%asi, %i3			;\
685	ldxa	[%sp + V9BIAS64 + 96]%asi, %i4			;\
686	ldxa	[%sp + V9BIAS64 + 104]%asi, %i5			;\
687	ldxa	[%sp + V9BIAS64 + 112]%asi, %i6			;\
688	ldxa	[%sp + V9BIAS64 + 120]%asi, %i7			;\
689	restored						;\
690	retry							;\
691	SKIP(31-19-TT_TRACE_L_INS)				;\
692	ba,a,pt	%xcc, fault_64bit_##tail			;\
693	.empty
694
695/*
696 * SPILL_mixed spills either size window, depending on
697 * whether %sp is even or odd, to a 32-bit address space.
698 * This may only be used in conjunction with SPILL_32bit/
699 * SPILL_64bit. New versions of SPILL_mixed_{tt1,asi} would be
700 * needed for use with SPILL_{32,64}bit_{tt1,asi}.  Particular
701 * attention should be paid to the instructions that belong
702 * in the delay slots of the branches depending on the type
703 * of spill handler being branched to.
704 * Clear upper 32 bits of %sp if it is odd.
705 * We won't need to clear them in 64 bit kernel.
706 */
707#define	SPILL_mixed						\
708	btst	1, %sp						;\
709	bz,a,pt	%xcc, 1b					;\
710	srl	%sp, 0, %sp					;\
711	ba,pt	%xcc, 2b					;\
712	nop							;\
713	.align	128
714
715/*
716 * FILL_mixed(ASI) fills either size window, depending on
717 * whether %sp is even or odd, from a 32-bit address space.
718 * This may only be used in conjunction with FILL_32bit/
719 * FILL_64bit. New versions of FILL_mixed_{tt1,asi} would be
720 * needed for use with FILL_{32,64}bit_{tt1,asi}. Particular
721 * attention should be paid to the instructions that belong
722 * in the delay slots of the branches depending on the type
723 * of fill handler being branched to.
724 * Clear upper 32 bits of %sp if it is odd.
725 * We won't need to clear them in 64 bit kernel.
726 */
727#define	FILL_mixed						\
728	btst	1, %sp						;\
729	bz,a,pt	%xcc, 1b					;\
730	srl	%sp, 0, %sp					;\
731	ba,pt	%xcc, 2b					;\
732	nop							;\
733	.align	128
734
735
736/*
737 * SPILL_32clean/SPILL_64clean spill 32-bit and 64-bit register windows,
738 * respectively, into the address space via the designated asi.  The
739 * unbiased stack pointer is required to be eight-byte aligned (even for
740 * the 32-bit case even though this code does not require such strict
741 * alignment).
742 *
743 * With SPARC v9 the spill trap takes precedence over the cleanwin trap
744 * so when cansave == 0, canrestore == 6, and cleanwin == 6 the next save
745 * will cause cwp + 2 to be spilled but will not clean cwp + 1.  That
746 * window may contain kernel data so in user_rtt we set wstate to call
747 * these spill handlers on the first user spill trap.  These handler then
748 * spill the appropriate window but also back up a window and clean the
749 * window that didn't get a cleanwin trap.
750 */
751#define	SPILL_32clean(asi_num, tail)				\
752	srl	%sp, 0, %sp					;\
753	sta	%l0, [%sp + %g0]asi_num				;\
754	mov	4, %g1						;\
755	sta	%l1, [%sp + %g1]asi_num				;\
756	mov	8, %g2						;\
757	sta	%l2, [%sp + %g2]asi_num				;\
758	mov	12, %g3						;\
759	sta	%l3, [%sp + %g3]asi_num				;\
760	add	%sp, 16, %g4					;\
761	sta	%l4, [%g4 + %g0]asi_num				;\
762	sta	%l5, [%g4 + %g1]asi_num				;\
763	sta	%l6, [%g4 + %g2]asi_num				;\
764	sta	%l7, [%g4 + %g3]asi_num				;\
765	add	%g4, 16, %g4					;\
766	sta	%i0, [%g4 + %g0]asi_num				;\
767	sta	%i1, [%g4 + %g1]asi_num				;\
768	sta	%i2, [%g4 + %g2]asi_num				;\
769	sta	%i3, [%g4 + %g3]asi_num				;\
770	add	%g4, 16, %g4					;\
771	sta	%i4, [%g4 + %g0]asi_num				;\
772	sta	%i5, [%g4 + %g1]asi_num				;\
773	sta	%i6, [%g4 + %g2]asi_num				;\
774	sta	%i7, [%g4 + %g3]asi_num				;\
775	TT_TRACE_L(trace_win)					;\
776	b	.spill_clean					;\
777	  mov	WSTATE_USER32, %g7				;\
778	SKIP(31-25-TT_TRACE_L_INS)				;\
779	ba,a,pt	%xcc, fault_32bit_##tail			;\
780	.empty
781
782#define	SPILL_64clean(asi_num, tail)				\
783	mov	0 + V9BIAS64, %g1				;\
784	stxa	%l0, [%sp + %g1]asi_num				;\
785	mov	8 + V9BIAS64, %g2				;\
786	stxa	%l1, [%sp + %g2]asi_num				;\
787	mov	16 + V9BIAS64, %g3				;\
788	stxa	%l2, [%sp + %g3]asi_num				;\
789	mov	24 + V9BIAS64, %g4				;\
790	stxa	%l3, [%sp + %g4]asi_num				;\
791	add	%sp, 32, %g5					;\
792	stxa	%l4, [%g5 + %g1]asi_num				;\
793	stxa	%l5, [%g5 + %g2]asi_num				;\
794	stxa	%l6, [%g5 + %g3]asi_num				;\
795	stxa	%l7, [%g5 + %g4]asi_num				;\
796	add	%g5, 32, %g5					;\
797	stxa	%i0, [%g5 + %g1]asi_num				;\
798	stxa	%i1, [%g5 + %g2]asi_num				;\
799	stxa	%i2, [%g5 + %g3]asi_num				;\
800	stxa	%i3, [%g5 + %g4]asi_num				;\
801	add	%g5, 32, %g5					;\
802	stxa	%i4, [%g5 + %g1]asi_num				;\
803	stxa	%i5, [%g5 + %g2]asi_num				;\
804	stxa	%i6, [%g5 + %g3]asi_num				;\
805	stxa	%i7, [%g5 + %g4]asi_num				;\
806	TT_TRACE_L(trace_win)					;\
807	b	.spill_clean					;\
808	  mov	WSTATE_USER64, %g7				;\
809	SKIP(31-25-TT_TRACE_L_INS)				;\
810	ba,a,pt	%xcc, fault_64bit_##tail			;\
811	.empty
812
813
814/*
815 * Floating point disabled.
816 */
817#define	FP_DISABLED_TRAP		\
818	TT_TRACE(trace_gen)		;\
819	ba,pt	%xcc,.fp_disabled	;\
820	nop				;\
821	.align	32
822
823/*
824 * Floating point exceptions.
825 */
826#define	FP_IEEE_TRAP			\
827	TT_TRACE(trace_gen)		;\
828	ba,pt	%xcc,.fp_ieee_exception	;\
829	nop				;\
830	.align	32
831
832#define	FP_TRAP				\
833	TT_TRACE(trace_gen)		;\
834	ba,pt	%xcc,.fp_exception	;\
835	nop				;\
836	.align	32
837
838/*
839 * asynchronous traps at level 0 and level 1
840 *
841 * The first instruction must be a membar for UltraSPARC-III
842 * to stop RED state entry if the store queue has many
843 * pending bad stores (PRM, Chapter 11).
844 */
845#define ASYNC_TRAP(ttype, ttlabel, table_name)\
846	.global	table_name	;\
847table_name:			;\
848	membar	#Sync		;\
849	TT_TRACE(ttlabel)	;\
850	ba	async_err	;\
851	mov	ttype, %g5	;\
852	.align	32
853
854/*
855 * Defaults to BAD entry, but establishes label to be used for
856 * architecture-specific overwrite of trap table entry.
857 */
858#define	LABELED_BAD(table_name)		\
859	.global	table_name		;\
860table_name:				;\
861	BAD
862
863/*
864 * illegal instruction trap
865 */
866#define	ILLTRAP_INSTR			  \
867	membar	#Sync			  ;\
868	TT_TRACE(trace_gen)		  ;\
869	or	%g0, P_UTRAP4, %g2	  ;\
870	or	%g0, T_UNIMP_INSTR, %g3   ;\
871	sethi	%hi(.check_v9utrap), %g4  ;\
872	jmp	%g4 + %lo(.check_v9utrap) ;\
873	nop				  ;\
874	.align	32
875
876/*
877 * tag overflow trap
878 */
879#define	TAG_OVERFLOW			  \
880	TT_TRACE(trace_gen)		  ;\
881	or	%g0, P_UTRAP10, %g2	  ;\
882	or	%g0, T_TAG_OVERFLOW, %g3  ;\
883	sethi	%hi(.check_v9utrap), %g4  ;\
884	jmp	%g4 + %lo(.check_v9utrap) ;\
885	nop				  ;\
886	.align	32
887
888/*
889 * divide by zero trap
890 */
891#define	DIV_BY_ZERO			  \
892	TT_TRACE(trace_gen)		  ;\
893	or	%g0, P_UTRAP11, %g2	  ;\
894	or	%g0, T_IDIV0, %g3	  ;\
895	sethi	%hi(.check_v9utrap), %g4  ;\
896	jmp	%g4 + %lo(.check_v9utrap) ;\
897	nop				  ;\
898	.align	32
899
900/*
901 * trap instruction for V9 user trap handlers
902 */
903#define	TRAP_INSTR			  \
904	TT_TRACE(trace_gen)		  ;\
905	or	%g0, T_SOFTWARE_TRAP, %g3 ;\
906	sethi	%hi(.check_v9utrap), %g4  ;\
907	jmp	%g4 + %lo(.check_v9utrap) ;\
908	nop				  ;\
909	.align	32
910#define	TRP4	TRAP_INSTR; TRAP_INSTR; TRAP_INSTR; TRAP_INSTR
911
912/*
913 * LEVEL_INTERRUPT is for level N interrupts.
914 * VECTOR_INTERRUPT is for the vector trap.
915 */
916#define	LEVEL_INTERRUPT(level)		\
917	.global	tt_pil##level		;\
918tt_pil##level:			;\
919	ba,pt	%xcc, pil_interrupt	;\
920	mov	level, %g4		;\
921	.align	32
922
923#define	LEVEL14_INTERRUPT			\
924	ba	pil14_interrupt			;\
925	mov	PIL_14, %g4			;\
926	.align	32
927
928#define        LEVEL15_INTERRUPT                       \
929       ba      pil15_interrupt                 ;\
930       mov     PIL_15, %g4                     ;\
931       .align  32
932
933#define	VECTOR_INTERRUPT				\
934	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g1	;\
935	btst	IRSR_BUSY, %g1				;\
936	bnz,pt	%xcc, vec_interrupt			;\
937	nop						;\
938	ba,a,pt	%xcc, vec_intr_spurious			;\
939	.empty						;\
940	.align	32
941
942/*
943 * MMU Trap Handlers.
944 */
945#define	SWITCH_GLOBALS	/* mmu->alt, alt->mmu */			\
946	rdpr	%pstate, %g5						;\
947	wrpr	%g5, PSTATE_MG | PSTATE_AG, %pstate
948
949#define	IMMU_EXCEPTION							\
950	membar	#Sync							;\
951	SWITCH_GLOBALS							;\
952	wr	%g0, ASI_IMMU, %asi					;\
953	rdpr	%tpc, %g2						;\
954	ldxa	[MMU_SFSR]%asi, %g3					;\
955	ba,pt	%xcc, .mmu_exception_end				;\
956	mov	T_INSTR_EXCEPTION, %g1					;\
957	.align	32
958
959#define	DMMU_EXCEPTION							\
960	SWITCH_GLOBALS							;\
961	wr	%g0, ASI_DMMU, %asi					;\
962	ldxa	[MMU_TAG_ACCESS]%asi, %g2				;\
963	ldxa	[MMU_SFSR]%asi, %g3					;\
964	ba,pt	%xcc, .mmu_exception_end				;\
965	mov	T_DATA_EXCEPTION, %g1					;\
966	.align	32
967
968#define	DMMU_EXC_AG_PRIV						\
969	wr	%g0, ASI_DMMU, %asi					;\
970	ldxa	[MMU_SFAR]%asi, %g2					;\
971	ba,pt	%xcc, .mmu_priv_exception				;\
972	ldxa	[MMU_SFSR]%asi, %g3					;\
973	.align	32
974
975#define	DMMU_EXC_AG_NOT_ALIGNED						\
976	wr	%g0, ASI_DMMU, %asi					;\
977	ldxa	[MMU_SFAR]%asi, %g2					;\
978	ba,pt	%xcc, .mmu_exception_not_aligned			;\
979	ldxa	[MMU_SFSR]%asi, %g3					;\
980	.align	32
981
982/*
983 * SPARC V9 IMPL. DEP. #109(1) and (2) and #110(1) and (2)
984 */
985#define	DMMU_EXC_LDDF_NOT_ALIGNED					\
986	btst	1, %sp							;\
987	bnz,pt	%xcc, .lddf_exception_not_aligned			;\
988	wr	%g0, ASI_DMMU, %asi					;\
989	ldxa	[MMU_SFAR]%asi, %g2					;\
990	ba,pt	%xcc, .mmu_exception_not_aligned			;\
991	ldxa	[MMU_SFSR]%asi, %g3					;\
992	.align	32
993
994#define	DMMU_EXC_STDF_NOT_ALIGNED					\
995	btst	1, %sp							;\
996	bnz,pt	%xcc, .stdf_exception_not_aligned			;\
997	wr	%g0, ASI_DMMU, %asi					;\
998	ldxa	[MMU_SFAR]%asi, %g2					;\
999	ba,pt	%xcc, .mmu_exception_not_aligned			;\
1000	ldxa	[MMU_SFSR]%asi, %g3					;\
1001	.align	32
1002
1003/*
1004 * Flush the TLB using either the primary, secondary, or nucleus flush
1005 * operation based on whether the ctx from the tag access register matches
1006 * the primary or secondary context (flush the nucleus if neither matches).
1007 *
1008 * Requires a membar #Sync before next ld/st.
1009 * exits with:
1010 * g2 = tag access register
1011 * g3 = ctx number
1012 */
1013#if TAGACC_CTX_MASK != CTXREG_CTX_MASK
1014#error "TAGACC_CTX_MASK != CTXREG_CTX_MASK"
1015#endif
1016#define	DTLB_DEMAP_ENTRY						\
1017	mov	MMU_TAG_ACCESS, %g1					;\
1018	mov	MMU_PCONTEXT, %g5					;\
1019	ldxa	[%g1]ASI_DMMU, %g2					;\
1020	sethi	%hi(TAGACC_CTX_MASK), %g4				;\
1021	or	%g4, %lo(TAGACC_CTX_MASK), %g4				;\
1022	and	%g2, %g4, %g3			/* g3 = ctx */		;\
1023	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = primary ctx */	;\
1024	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1025	cmp	%g3, %g6						;\
1026	be,pt	%xcc, 1f						;\
1027	andn	%g2, %g4, %g1			/* ctx = primary */	;\
1028	mov	MMU_SCONTEXT, %g5					;\
1029	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = secondary ctx */ ;\
1030	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1031	cmp	%g3, %g6						;\
1032	be,a,pt	%xcc, 1f						;\
1033	  or	%g1, DEMAP_SECOND, %g1					;\
1034	or	%g1, DEMAP_NUCLEUS, %g1					;\
10351:	stxa	%g0, [%g1]ASI_DTLB_DEMAP	/* MMU_DEMAP_PAGE */	;\
1036	membar	#Sync
1037
1038#if defined(cscope)
1039/*
1040 * Define labels to direct cscope quickly to labels that
1041 * are generated by macro expansion of DTLB_MISS().
1042 */
1043	.global	tt0_dtlbmiss
1044tt0_dtlbmiss:
1045	.global	tt1_dtlbmiss
1046tt1_dtlbmiss:
1047	nop
1048#endif
1049
1050/*
1051 * Needs to be exactly 32 instructions
1052 *
1053 * UTLB NOTE: If we don't hit on the 8k pointer then we branch
1054 * to a special 4M tsb handler. It would be nice if that handler
1055 * could live in this file but currently it seems better to allow
1056 * it to fall thru to sfmmu_tsb_miss.
1057 */
1058#ifdef UTSB_PHYS
1059#define	DTLB_MISS(table_name)						;\
1060	.global	table_name##_dtlbmiss					;\
1061table_name##_dtlbmiss:						;\
1062	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1063	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1064	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1065	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1066	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1067	cmp	%g3, INVALID_CONTEXT					;\
1068	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1069	  srax	%g2, PREDISM_BASESHIFT, %g6  /* g6 > 0 ISM predicted */ ;\
1070	brgz,pn %g6, sfmmu_udtlb_slowpath_ismpred			;\
1071	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1072	ldda	[%g1]ASI_QUAD_LDD_PHYS, %g4	/* g4 = tag, %g5 data */;\
1073	cmp	%g4, %g7						;\
1074	bne,pn %xcc, sfmmu_udtlb_slowpath_noismpred			;\
1075	  nop								;\
1076	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1077	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1078	retry				/* in %g5 */			;\
1079	unimp	0							;\
1080	unimp	0							;\
1081	unimp	0							;\
1082	unimp	0							;\
1083	unimp	0							;\
1084	unimp	0							;\
1085	unimp	0							;\
1086	unimp	0							;\
1087	unimp	0							;\
1088	unimp	0							;\
1089	unimp	0							;\
1090	unimp	0							;\
1091	unimp	0							;\
1092	unimp	0							;\
1093	.align 128
1094
1095#else /* UTSB_PHYS */
1096#define	DTLB_MISS(table_name)						;\
1097	.global	table_name##_dtlbmiss					;\
1098table_name##_dtlbmiss:						;\
1099	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1100	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1101	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1102	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1103	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1104	cmp	%g3, INVALID_CONTEXT					;\
1105	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1106	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1107	brlz,pn %g1, sfmmu_udtlb_slowpath				;\
1108	  nop								;\
1109	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, %g5 data */	;\
1110	cmp	%g4, %g7						;\
1111	bne,pn	%xcc, sfmmu_tsb_miss_tt		/* no 4M TSB, miss */	;\
1112	  mov	-1, %g3		/* set 4M tsbe ptr to -1 */		;\
1113	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1114	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1115	retry				/* in %g5 */			;\
1116	unimp	0							;\
1117	unimp	0							;\
1118	unimp	0							;\
1119	unimp	0							;\
1120	unimp	0							;\
1121        unimp   0                                                       ;\
1122        unimp   0                                                       ;\
1123        unimp   0                                                       ;\
1124        unimp   0                                                       ;\
1125        unimp   0                                                       ;\
1126        unimp   0                                                       ;\
1127        unimp   0                                                       ;\
1128        unimp   0                                                       ;\
1129        unimp   0                                                       ;\
1130	.align 128
1131#endif /* UTSB_PHYS */
1132
1133#if defined(cscope)
1134/*
1135 * Define labels to direct cscope quickly to labels that
1136 * are generated by macro expansion of ITLB_MISS().
1137 */
1138	.global	tt0_itlbmiss
1139tt0_itlbmiss:
1140	.global	tt1_itlbmiss
1141tt1_itlbmiss:
1142	nop
1143#endif
1144
1145/*
1146 * Instruction miss handler.
1147 * ldda instructions will have their ASI patched
1148 * by sfmmu_patch_ktsb at runtime.
1149 * MUST be EXACTLY 32 instructions or we'll break.
1150 */
1151#ifdef UTSB_PHYS
1152#define ITLB_MISS(table_name)                                            \
1153        .global table_name##_itlbmiss                                 ;\
1154table_name##_itlbmiss:                                                ;\
1155        mov     MMU_TAG_ACCESS, %g6             /* select tag acc */    ;\
1156        ldxa    [%g0]ASI_IMMU_TSB_8K, %g1       /* g1 = tsbe ptr */     ;\
1157        ldxa    [%g6]ASI_IMMU, %g2              /* g2 = tag access */   ;\
1158        sllx    %g2, TAGACC_CTX_LSHIFT, %g3                             ;\
1159        srlx    %g3, TAGACC_CTX_LSHIFT, %g3     /* g3 = ctx */          ;\
1160        cmp     %g3, INVALID_CONTEXT                                    ;\
1161        ble,pn  %xcc, sfmmu_kitlb_miss                                  ;\
1162          srlx  %g2, TAG_VALO_SHIFT, %g7        /* g7 = tsb tag */      ;\
1163        ldda    [%g1]ASI_QUAD_LDD_PHYS, %g4 /* g4 = tag, g5 = data */   ;\
1164        cmp     %g4, %g7                                                ;\
1165        bne,pn  %xcc, sfmmu_uitlb_slowpath                              ;\
1166          andcc %g5, TTE_EXECPRM_INT, %g0 /* check execute bit */       ;\
1167        bz,pn   %icc, exec_fault                                        ;\
1168          nop                                                           ;\
1169        TT_TRACE(trace_tsbhit)          /* 2 instr ifdef TRAPTRACE */   ;\
1170        stxa    %g5, [%g0]ASI_ITLB_IN   /* trapstat expects %g5 */      ;\
1171        retry                                                           ;\
1172        unimp   0                                                       ;\
1173        unimp   0                                                       ;\
1174        unimp   0                                                       ;\
1175        unimp   0                                                       ;\
1176        unimp   0                                                       ;\
1177        unimp   0                                                       ;\
1178        unimp   0                                                       ;\
1179        unimp   0                                                       ;\
1180        unimp   0                                                       ;\
1181        unimp   0                                                       ;\
1182        unimp   0                                                       ;\
1183        unimp   0                                                       ;\
1184        unimp   0                                                       ;\
1185        unimp   0                                                       ;\
1186        .align 128
1187
1188#else /* UTSB_PHYS */
1189#define	ITLB_MISS(table_name)						 \
1190	.global	table_name##_itlbmiss					;\
1191table_name##_itlbmiss:						;\
1192	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1193	ldxa	[%g0]ASI_IMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1194	ldxa	[%g6]ASI_IMMU, %g2		/* g2 = tag access */	;\
1195	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1196	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1197	cmp	%g3, INVALID_CONTEXT					;\
1198	ble,pn	%xcc, sfmmu_kitlb_miss					;\
1199	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1200	brlz,pn	%g1, sfmmu_uitlb_slowpath	/* if >1 TSB branch */	;\
1201	  nop								;\
1202	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, g5 = data */	;\
1203	cmp	%g4, %g7						;\
1204	bne,pn	%xcc, sfmmu_tsb_miss_tt	/* br if 8k ptr miss */		;\
1205	  mov	-1, %g3		/* set 4M TSB ptr to -1 */		;\
1206	andcc	%g5, TTE_EXECPRM_INT, %g0 /* check execute bit */	;\
1207	bz,pn	%icc, exec_fault					;\
1208	  nop								;\
1209	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1210	stxa	%g5, [%g0]ASI_ITLB_IN	/* trapstat expects %g5 */	;\
1211	retry								;\
1212	unimp	0							;\
1213	unimp	0							;\
1214	unimp	0							;\
1215	unimp	0							;\
1216	unimp	0							;\
1217        unimp   0                                                       ;\
1218        unimp   0                                                       ;\
1219        unimp   0                                                       ;\
1220        unimp   0                                                       ;\
1221        unimp   0                                                       ;\
1222        unimp   0                                                       ;\
1223	.align 128
1224#endif /* UTSB_PHYS */
1225
1226
1227/*
1228 * This macro is the first level handler for fast protection faults.
1229 * It first demaps the tlb entry which generated the fault and then
1230 * attempts to set the modify bit on the hash.  It needs to be
1231 * exactly 32 instructions.
1232 */
1233#define	DTLB_PROT							 \
1234	DTLB_DEMAP_ENTRY		/* 20 instructions */		;\
1235	/*								;\
1236	 * At this point:						;\
1237	 *   g1 = ????							;\
1238	 *   g2 = tag access register					;\
1239	 *   g3 = ctx number						;\
1240	 *   g4 = ????							;\
1241	 */								;\
1242	TT_TRACE(trace_dataprot)	/* 2 instr ifdef TRAPTRACE */	;\
1243					/* clobbers g1 and g6 */	;\
1244	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1245	brnz,pt %g3, sfmmu_uprot_trap		/* user trap */		;\
1246	  nop								;\
1247	ba,a,pt	%xcc, sfmmu_kprot_trap		/* kernel trap */	;\
1248	unimp	0							;\
1249	unimp	0							;\
1250	unimp	0							;\
1251	unimp	0							;\
1252	unimp	0							;\
1253	unimp	0							;\
1254	.align 128
1255
1256#define	DMMU_EXCEPTION_TL1						;\
1257	SWITCH_GLOBALS							;\
1258	ba,a,pt	%xcc, mmu_trap_tl1					;\
1259	  nop								;\
1260	.align 32
1261
1262#define	MISALIGN_ADDR_TL1						;\
1263	ba,a,pt	%xcc, mmu_trap_tl1					;\
1264	  nop								;\
1265	.align 32
1266
1267/*
1268 * Trace a tsb hit
1269 * g1 = tsbe pointer (in/clobbered)
1270 * g2 = tag access register (in)
1271 * g3 - g4 = scratch (clobbered)
1272 * g5 = tsbe data (in)
1273 * g6 = scratch (clobbered)
1274 * g7 = pc we jumped here from (in)
1275 * ttextra = value to OR in to trap type (%tt) (in)
1276 */
1277#ifdef TRAPTRACE
1278#define TRACE_TSBHIT(ttextra)						 \
1279	membar	#Sync							;\
1280	sethi	%hi(FLUSH_ADDR), %g6					;\
1281	flush	%g6							;\
1282	TRACE_PTR(%g3, %g6)						;\
1283	GET_TRACE_TICK(%g6, %g4)					;\
1284	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi				;\
1285	stxa	%g2, [%g3 + TRAP_ENT_SP]%asi	/* tag access */	;\
1286	stxa	%g5, [%g3 + TRAP_ENT_F1]%asi	/* tsb data */		;\
1287	rdpr	%tnpc, %g6						;\
1288	stxa	%g6, [%g3 + TRAP_ENT_F2]%asi				;\
1289	stxa	%g1, [%g3 + TRAP_ENT_F3]%asi	/* tsb pointer */	;\
1290	stxa	%g0, [%g3 + TRAP_ENT_F4]%asi				;\
1291	rdpr	%tpc, %g6						;\
1292	stxa	%g6, [%g3 + TRAP_ENT_TPC]%asi				;\
1293	rdpr	%tl, %g6						;\
1294	stha	%g6, [%g3 + TRAP_ENT_TL]%asi				;\
1295	rdpr	%tt, %g6						;\
1296	or	%g6, (ttextra), %g6					;\
1297	stha	%g6, [%g3 + TRAP_ENT_TT]%asi				;\
1298	ldxa	[%g0]ASI_IMMU, %g1		/* tag target */	;\
1299	ldxa	[%g0]ASI_DMMU, %g4					;\
1300	cmp	%g6, FAST_IMMU_MISS_TT					;\
1301	movne	%icc, %g4, %g1						;\
1302	stxa	%g1, [%g3 + TRAP_ENT_TSTATE]%asi /* tsb tag */		;\
1303	stxa	%g0, [%g3 + TRAP_ENT_TR]%asi				;\
1304	TRACE_NEXT(%g3, %g4, %g6)
1305#else
1306#define TRACE_TSBHIT(ttextra)
1307#endif
1308
1309/*
1310 * =======================================================================
1311 *		SPARC V9 TRAP TABLE
1312 *
1313 * The trap table is divided into two halves: the first half is used when
1314 * taking traps when TL=0; the second half is used when taking traps from
1315 * TL>0. Note that handlers in the second half of the table might not be able
1316 * to make the same assumptions as handlers in the first half of the table.
1317 *
1318 * Worst case trap nesting so far:
1319 *
1320 *	at TL=0 client issues software trap requesting service
1321 *	at TL=1 nucleus wants a register window
1322 *	at TL=2 register window clean/spill/fill takes a TLB miss
1323 *	at TL=3 processing TLB miss
1324 *	at TL=4 handle asynchronous error
1325 *
1326 * Note that a trap from TL=4 to TL=5 places Spitfire in "RED mode".
1327 *
1328 * =======================================================================
1329 */
1330	.section ".text"
1331	.align	4
1332	.global trap_table, scb, trap_table0, trap_table1, etrap_table
1333	.type	trap_table, #object
1334	.type	scb, #object
1335trap_table:
1336scb:
1337trap_table0:
1338	/* hardware traps */
1339	NOT;				/* 000	reserved */
1340	RED;				/* 001	power on reset */
1341	RED;				/* 002	watchdog reset */
1342	RED;				/* 003	externally initiated reset */
1343	RED;				/* 004	software initiated reset */
1344	RED;				/* 005	red mode exception */
1345	NOT; NOT;			/* 006 - 007 reserved */
1346	IMMU_EXCEPTION;			/* 008	instruction access exception */
1347	NOT;				/* 009	instruction access MMU miss */
1348	ASYNC_TRAP(T_INSTR_ERROR, trace_gen, tt0_iae);
1349					/* 00A	instruction access error */
1350	NOT; NOT4;			/* 00B - 00F reserved */
1351	ILLTRAP_INSTR;			/* 010	illegal instruction */
1352	TRAP(T_PRIV_INSTR);		/* 011	privileged opcode */
1353	NOT;				/* 012	unimplemented LDD */
1354	NOT;				/* 013	unimplemented STD */
1355	NOT4; NOT4; NOT4;		/* 014 - 01F reserved */
1356	FP_DISABLED_TRAP;		/* 020	fp disabled */
1357	FP_IEEE_TRAP;			/* 021	fp exception ieee 754 */
1358	FP_TRAP;			/* 022	fp exception other */
1359	TAG_OVERFLOW;			/* 023	tag overflow */
1360	CLEAN_WINDOW;			/* 024 - 027 clean window */
1361	DIV_BY_ZERO;			/* 028	division by zero */
1362	NOT;				/* 029	internal processor error */
1363	NOT; NOT; NOT4;			/* 02A - 02F reserved */
1364	DMMU_EXCEPTION;			/* 030	data access exception */
1365	NOT;				/* 031	data access MMU miss */
1366	ASYNC_TRAP(T_DATA_ERROR, trace_gen, tt0_dae);
1367					/* 032	data access error */
1368	NOT;				/* 033	data access protection */
1369	DMMU_EXC_AG_NOT_ALIGNED;	/* 034	mem address not aligned */
1370	DMMU_EXC_LDDF_NOT_ALIGNED;	/* 035	LDDF mem address not aligned */
1371	DMMU_EXC_STDF_NOT_ALIGNED;	/* 036	STDF mem address not aligned */
1372	DMMU_EXC_AG_PRIV;		/* 037	privileged action */
1373	NOT;				/* 038	LDQF mem address not aligned */
1374	NOT;				/* 039	STQF mem address not aligned */
1375	NOT; NOT; NOT4;			/* 03A - 03F reserved */
1376	LABELED_BAD(tt0_asdat);		/* 040	async data error */
1377	LEVEL_INTERRUPT(1);		/* 041	interrupt level 1 */
1378	LEVEL_INTERRUPT(2);		/* 042	interrupt level 2 */
1379	LEVEL_INTERRUPT(3);		/* 043	interrupt level 3 */
1380	LEVEL_INTERRUPT(4);		/* 044	interrupt level 4 */
1381	LEVEL_INTERRUPT(5);		/* 045	interrupt level 5 */
1382	LEVEL_INTERRUPT(6);		/* 046	interrupt level 6 */
1383	LEVEL_INTERRUPT(7);		/* 047	interrupt level 7 */
1384	LEVEL_INTERRUPT(8);		/* 048	interrupt level 8 */
1385	LEVEL_INTERRUPT(9);		/* 049	interrupt level 9 */
1386	LEVEL_INTERRUPT(10);		/* 04A	interrupt level 10 */
1387	LEVEL_INTERRUPT(11);		/* 04B	interrupt level 11 */
1388	LEVEL_INTERRUPT(12);		/* 04C	interrupt level 12 */
1389	LEVEL_INTERRUPT(13);		/* 04D	interrupt level 13 */
1390	LEVEL14_INTERRUPT;		/* 04E	interrupt level 14 */
1391	LEVEL15_INTERRUPT;		/* 04F	interrupt level 15 */
1392	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F reserved */
1393	VECTOR_INTERRUPT;		/* 060	interrupt vector */
1394	GOTO(kmdb_trap);		/* 061	PA watchpoint */
1395	GOTO(kmdb_trap);		/* 062	VA watchpoint */
1396	GOTO_TT(ce_err, trace_gen);	/* 063	corrected ECC error */
1397	ITLB_MISS(tt0);			/* 064	instruction access MMU miss */
1398	DTLB_MISS(tt0);			/* 068	data access MMU miss */
1399	DTLB_PROT;			/* 06C	data access protection */
1400	LABELED_BAD(tt0_fecc);		/* 070  fast ecache ECC error */
1401	LABELED_BAD(tt0_dperr);		/* 071  Cheetah+ dcache parity error */
1402	LABELED_BAD(tt0_iperr);		/* 072  Cheetah+ icache parity error */
1403	NOT;				/* 073  reserved */
1404	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1405	NOT4;				/* 080	spill 0 normal */
1406	SPILL_32bit_asi(ASI_AIUP,sn0);	/* 084	spill 1 normal */
1407	SPILL_64bit_asi(ASI_AIUP,sn0);	/* 088	spill 2 normal */
1408	SPILL_32clean(ASI_AIUP,sn0);	/* 08C	spill 3 normal */
1409	SPILL_64clean(ASI_AIUP,sn0);	/* 090	spill 4 normal */
1410	SPILL_32bit(not);		/* 094	spill 5 normal */
1411	SPILL_64bit(not);		/* 098	spill 6 normal */
1412	SPILL_mixed;			/* 09C	spill 7 normal */
1413	NOT4;				/* 0A0	spill 0 other */
1414	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0A4	spill 1 other */
1415	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0A8	spill 2 other */
1416	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0AC	spill 3 other */
1417	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0B0	spill 4 other */
1418	NOT4;				/* 0B4	spill 5 other */
1419	NOT4;				/* 0B8	spill 6 other */
1420	NOT4;				/* 0BC	spill 7 other */
1421	NOT4;				/* 0C0	fill 0 normal */
1422	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0C4	fill 1 normal */
1423	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0C8	fill 2 normal */
1424	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0CC	fill 3 normal */
1425	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0D0	fill 4 normal */
1426	FILL_32bit(not);		/* 0D4	fill 5 normal */
1427	FILL_64bit(not);		/* 0D8	fill 6 normal */
1428	FILL_mixed;			/* 0DC	fill 7 normal */
1429	NOT4;				/* 0E0	fill 0 other */
1430	NOT4;				/* 0E4	fill 1 other */
1431	NOT4;				/* 0E8	fill 2 other */
1432	NOT4;				/* 0EC	fill 3 other */
1433	NOT4;				/* 0F0	fill 4 other */
1434	NOT4;				/* 0F4	fill 5 other */
1435	NOT4;				/* 0F8	fill 6 other */
1436	NOT4;				/* 0FC	fill 7 other */
1437	/* user traps */
1438	GOTO(syscall_trap_4x);		/* 100	old system call */
1439	TRAP(T_BREAKPOINT);		/* 101	user breakpoint */
1440	TRAP(T_DIV0);			/* 102	user divide by zero */
1441	FLUSHW(tt0_flushw);		/* 103	flush windows */
1442	GOTO(.clean_windows);		/* 104	clean windows */
1443	BAD;				/* 105	range check ?? */
1444	GOTO(.fix_alignment);		/* 106	do unaligned references */
1445	BAD;				/* 107	unused */
1446	SYSCALL_TRAP32;			/* 108	ILP32 system call on LP64 */
1447	GOTO(set_trap0_addr);		/* 109	set trap0 address */
1448	BAD; BAD; BAD4;			/* 10A - 10F unused */
1449	TRP4; TRP4; TRP4; TRP4;		/* 110 - 11F V9 user trap handlers */
1450	GOTO(.getcc);			/* 120	get condition codes */
1451	GOTO(.setcc);			/* 121	set condition codes */
1452	GOTO(.getpsr);			/* 122	get psr */
1453	GOTO(.setpsr);			/* 123	set psr (some fields) */
1454	GOTO(get_timestamp);		/* 124	get timestamp */
1455	GOTO(get_virtime);		/* 125	get lwp virtual time */
1456	PRIV(self_xcall);		/* 126	self xcall */
1457	GOTO(get_hrestime);		/* 127	get hrestime */
1458	BAD;				/* 128	ST_SETV9STACK */
1459	GOTO(.getlgrp);			/* 129  get lgrpid */
1460	BAD; BAD; BAD4;			/* 12A - 12F unused */
1461	BAD4; BAD4; 			/* 130 - 137 unused */
1462	DTRACE_PID;			/* 138  dtrace pid tracing provider */
1463	BAD;				/* 139  unused */
1464	DTRACE_RETURN;			/* 13A	dtrace pid return probe */
1465	BAD; BAD4;			/* 13B - 13F unused */
1466	SYSCALL_TRAP;			/* 140  LP64 system call */
1467	SYSCALL(nosys);			/* 141  unused system call trap */
1468#ifdef DEBUG_USER_TRAPTRACECTL
1469	GOTO(.traptrace_freeze);	/* 142  freeze traptrace */
1470	GOTO(.traptrace_unfreeze);	/* 143  unfreeze traptrace */
1471#else
1472	SYSCALL(nosys);			/* 142  unused system call trap */
1473	SYSCALL(nosys);			/* 143  unused system call trap */
1474#endif
1475	BAD4; BAD4; BAD4;		/* 144 - 14F unused */
1476	BAD4; BAD4; BAD4; BAD4;		/* 150 - 15F unused */
1477	BAD4; BAD4; BAD4; BAD4;		/* 160 - 16F unused */
1478	BAD;				/* 170 - unused */
1479	BAD;				/* 171 - unused */
1480	BAD; BAD;			/* 172 - 173 unused */
1481	BAD4; BAD4;			/* 174 - 17B unused */
1482#ifdef	PTL1_PANIC_DEBUG
1483	mov PTL1_BAD_DEBUG, %g1; GOTO(ptl1_panic);
1484					/* 17C	test ptl1_panic */
1485#else
1486	BAD;				/* 17C  unused */
1487#endif	/* PTL1_PANIC_DEBUG */
1488	PRIV(kmdb_trap);		/* 17D	kmdb enter (L1-A) */
1489	PRIV(kmdb_trap);		/* 17E	kmdb breakpoint */
1490	PRIV(kctx_obp_bpt);		/* 17F	obp breakpoint */
1491	/* reserved */
1492	NOT4; NOT4; NOT4; NOT4;		/* 180 - 18F reserved */
1493	NOT4; NOT4; NOT4; NOT4;		/* 190 - 19F reserved */
1494	NOT4; NOT4; NOT4; NOT4;		/* 1A0 - 1AF reserved */
1495	NOT4; NOT4; NOT4; NOT4;		/* 1B0 - 1BF reserved */
1496	NOT4; NOT4; NOT4; NOT4;		/* 1C0 - 1CF reserved */
1497	NOT4; NOT4; NOT4; NOT4;		/* 1D0 - 1DF reserved */
1498	NOT4; NOT4; NOT4; NOT4;		/* 1E0 - 1EF reserved */
1499	NOT4; NOT4; NOT4; NOT4;		/* 1F0 - 1FF reserved */
1500trap_table1:
1501	NOT4; NOT4; NOT; NOT;		/* 000 - 009 unused */
1502	ASYNC_TRAP(T_INSTR_ERROR + T_TL1, trace_gen, tt1_iae);
1503					/* 00A	instruction access error */
1504	NOT; NOT4;			/* 00B - 00F unused */
1505	NOT4; NOT4; NOT4; NOT4;		/* 010 - 01F unused */
1506	NOT4;				/* 020 - 023 unused */
1507	CLEAN_WINDOW;			/* 024 - 027 clean window */
1508	NOT4; NOT4;			/* 028 - 02F unused */
1509	DMMU_EXCEPTION_TL1;		/* 030 	data access exception */
1510	NOT;				/* 031 unused */
1511	ASYNC_TRAP(T_DATA_ERROR + T_TL1, trace_gen, tt1_dae);
1512					/* 032	data access error */
1513	NOT;				/* 033	unused */
1514	MISALIGN_ADDR_TL1;		/* 034	mem address not aligned */
1515	NOT; NOT; NOT; NOT4; NOT4	/* 035 - 03F unused */
1516	LABELED_BAD(tt1_asdat);		/* 040	async data error */
1517	NOT; NOT; NOT;			/* 041 - 043 unused */
1518	NOT4; NOT4; NOT4;		/* 044 - 04F unused */
1519	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F unused */
1520	NOT;				/* 060	unused */
1521	GOTO(kmdb_trap_tl1);		/* 061	PA watchpoint */
1522	GOTO(kmdb_trap_tl1);		/* 062	VA watchpoint */
1523	GOTO_TT(ce_err_tl1, trace_gen);	/* 063	corrected ECC error */
1524	ITLB_MISS(tt1);			/* 064	instruction access MMU miss */
1525	DTLB_MISS(tt1);			/* 068	data access MMU miss */
1526	DTLB_PROT;			/* 06C	data access protection */
1527	LABELED_BAD(tt1_fecc);		/* 070  fast ecache ECC error */
1528	LABELED_BAD(tt1_dperr);		/* 071  Cheetah+ dcache parity error */
1529	LABELED_BAD(tt1_iperr);		/* 072  Cheetah+ icache parity error */
1530	NOT;				/* 073  reserved */
1531	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1532	NOT4;				/* 080	spill 0 normal */
1533	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 084	spill 1 normal */
1534	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 088	spill 2 normal */
1535	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 08C	spill 3 normal */
1536	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 090	spill 4 normal */
1537	SPILL_32bit(not);		/* 094	spill 5 normal */
1538	SPILL_64bit(not);		/* 098	spill 6 normal */
1539	SPILL_mixed;			/* 09C	spill 7 normal */
1540	NOT4;				/* 0A0	spill 0 other */
1541	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0A4	spill 1 other */
1542	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0A8	spill 2 other */
1543	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0AC	spill 3 other */
1544	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0B0  spill 4 other */
1545	NOT4;				/* 0B4  spill 5 other */
1546	NOT4;				/* 0B8  spill 6 other */
1547	NOT4;				/* 0BC  spill 7 other */
1548	NOT4;				/* 0C0	fill 0 normal */
1549	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0C4	fill 1 normal */
1550	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0C8	fill 2 normal */
1551	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0CC	fill 3 normal */
1552	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0D0	fill 4 normal */
1553	FILL_32bit(not);		/* 0D4	fill 5 normal */
1554	FILL_64bit(not);		/* 0D8	fill 6 normal */
1555	FILL_mixed;			/* 0DC	fill 7 normal */
1556	NOT4; NOT4; NOT4; NOT4;		/* 0E0 - 0EF unused */
1557	NOT4; NOT4; NOT4; NOT4;		/* 0F0 - 0FF unused */
1558	LABELED_BAD(tt1_swtrap0);	/* 100  fast ecache ECC error (cont) */
1559	LABELED_BAD(tt1_swtrap1);	/* 101  Ch+ D$ parity error (cont) */
1560	LABELED_BAD(tt1_swtrap2);	/* 102  Ch+ I$ parity error (cont) */
1561	NOT;				/* 103  reserved */
1562/*
1563 * We only reserve the above four special case soft traps for code running
1564 * at TL>0, so we can truncate the trap table here.
1565 */
1566etrap_table:
1567	.size	trap_table, (.-trap_table)
1568	.size	scb, (.-scb)
1569
1570/*
1571 * We get to exec_fault in the case of an instruction miss and tte
1572 * has no execute bit set.  We go to tl0 to handle it.
1573 *
1574 * g1 = tsbe pointer (in/clobbered)
1575 * g2 = tag access register (in)
1576 * g3 - g4 = scratch (clobbered)
1577 * g5 = tsbe data (in)
1578 * g6 = scratch (clobbered)
1579 */
1580	ALTENTRY(exec_fault)
1581	TRACE_TSBHIT(0x200)
1582	SWITCH_GLOBALS
1583	mov	MMU_TAG_ACCESS, %g4
1584	ldxa	[%g4]ASI_IMMU, %g2			! arg1 = addr
1585	mov	T_INSTR_MMU_MISS, %g3			! arg2 = traptype
1586	set	trap, %g1
1587	ba,pt	%xcc, sys_trap
1588	  mov	-1, %g4
1589
1590.mmu_exception_not_aligned:
1591	rdpr	%tstate, %g1
1592	btst	TSTATE_PRIV, %g1
1593	bnz,pn	%icc, 2f
1594	nop
1595	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1596	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1597	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1598	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1599	brz,pt	%g5, 2f
1600	nop
1601	ldn	[%g5 + P_UTRAP15], %g5			! unaligned utrap?
1602	brz,pn	%g5, 2f
1603	nop
1604	btst	1, %sp
1605	bz,pt	%xcc, 1f				! 32 bit user program
1606	nop
1607	ba,pt	%xcc, .setup_v9utrap			! 64 bit user program
1608	nop
16091:
1610	ba,pt	%xcc, .setup_utrap
1611	or	%g2, %g0, %g7
16122:
1613	ba,pt	%xcc, .mmu_exception_end
1614	mov	T_ALIGNMENT, %g1
1615
1616.mmu_priv_exception:
1617	rdpr	%tstate, %g1
1618	btst	TSTATE_PRIV, %g1
1619	bnz,pn	%icc, 1f
1620	nop
1621	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1622	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1623	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1624	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1625	brz,pt	%g5, 1f
1626	nop
1627	ldn	[%g5 + P_UTRAP16], %g5
1628	brnz,pt	%g5, .setup_v9utrap
1629	nop
16301:
1631	mov	T_PRIV_INSTR, %g1
1632
1633.mmu_exception_end:
1634	CPU_INDEX(%g4, %g5)
1635	set	cpu_core, %g5
1636	sllx	%g4, CPU_CORE_SHIFT, %g4
1637	add	%g4, %g5, %g4
1638	lduh	[%g4 + CPUC_DTRACE_FLAGS], %g5
1639	andcc	%g5, CPU_DTRACE_NOFAULT, %g0
1640	bz	%xcc, .mmu_exception_tlb_chk
1641	or	%g5, CPU_DTRACE_BADADDR, %g5
1642	stuh	%g5, [%g4 + CPUC_DTRACE_FLAGS]
1643	done
1644
1645.mmu_exception_tlb_chk:
1646	GET_CPU_IMPL(%g5)			! check SFSR.FT to see if this
1647	cmp	%g5, PANTHER_IMPL		! is a TLB parity error. But
1648	bne	2f				! we only do this check while
1649	mov	1, %g4				! running on Panther CPUs
1650	sllx	%g4, PN_SFSR_PARITY_SHIFT, %g4	! since US-I/II use the same
1651	andcc	%g3, %g4, %g0			! bit for something else which
1652	bz	2f				! will be handled later.
1653	nop
1654.mmu_exception_is_tlb_parity:
1655	.weak itlb_parity_trap
1656	.weak dtlb_parity_trap
1657	set	itlb_parity_trap, %g4
1658	cmp	%g1, T_INSTR_EXCEPTION		! branch to the itlb or
1659	be	3f				! dtlb parity handler
1660	nop					! if this trap is due
1661	set	dtlb_parity_trap, %g4
1662	cmp	%g1, T_DATA_EXCEPTION		! to a IMMU exception
1663	be	3f				! or DMMU exception.
1664	nop
16652:
1666	sllx	%g3, 32, %g3
1667	or	%g3, %g1, %g3
1668	set	trap, %g1
1669	ba,pt	%xcc, sys_trap
1670	sub	%g0, 1, %g4
16713:
1672	jmp	%g4				! off to the appropriate
1673	nop					! TLB parity handler
1674
1675.fp_disabled:
1676	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1677	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1678#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1679	brz,a,pn %g1, 2f
1680	  nop
1681#endif
1682	rdpr	%tstate, %g4
1683	btst	TSTATE_PRIV, %g4
1684#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1685	bnz,pn %icc, 2f
1686	  nop
1687#else
1688	bnz,a,pn %icc, ptl1_panic
1689	  mov	PTL1_BAD_FPTRAP, %g1
1690#endif
1691	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1692	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1693	brz,a,pt %g5, 2f
1694	  nop
1695	ldn	[%g5 + P_UTRAP7], %g5			! fp_disabled utrap?
1696	brz,a,pn %g5, 2f
1697	  nop
1698	btst	1, %sp
1699	bz,a,pt	%xcc, 1f				! 32 bit user program
1700	  nop
1701	ba,a,pt	%xcc, .setup_v9utrap			! 64 bit user program
1702	  nop
17031:
1704	ba,pt	%xcc, .setup_utrap
1705	  or	%g0, %g0, %g7
17062:
1707	set	fp_disabled, %g1
1708	ba,pt	%xcc, sys_trap
1709	  sub	%g0, 1, %g4
1710
1711.fp_ieee_exception:
1712	rdpr	%tstate, %g1
1713	btst	TSTATE_PRIV, %g1
1714	bnz,a,pn %icc, ptl1_panic
1715	  mov	PTL1_BAD_FPTRAP, %g1
1716	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1717	stx	%fsr, [%g1 + CPU_TMP1]
1718	ldx	[%g1 + CPU_TMP1], %g2
1719	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1720	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1721	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1722	brz,a,pt %g5, 1f
1723	  nop
1724	ldn	[%g5 + P_UTRAP8], %g5
1725	brnz,a,pt %g5, .setup_v9utrap
1726	  nop
17271:
1728	set	_fp_ieee_exception, %g1
1729	ba,pt	%xcc, sys_trap
1730	  sub	%g0, 1, %g4
1731
1732/*
1733 * Register Inputs:
1734 *	%g5		user trap handler
1735 *	%g7		misaligned addr - for alignment traps only
1736 */
1737.setup_utrap:
1738	set	trap, %g1			! setup in case we go
1739	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1740	sub	%g0, 1, %g4			! the save instruction below
1741
1742	/*
1743	 * If the DTrace pid provider is single stepping a copied-out
1744	 * instruction, t->t_dtrace_step will be set. In that case we need
1745	 * to abort the single-stepping (since execution of the instruction
1746	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1747	 */
1748	save	%sp, -SA(MINFRAME32), %sp	! window for trap handler
1749	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1750	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1751	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1752	rdpr	%tnpc, %l2			! arg1 == tnpc
1753	brz,pt	%g2, 1f
1754	rdpr	%tpc, %l1			! arg0 == tpc
1755
1756	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1757	ldn	[%g1 + T_DTRACE_NPC], %l2	! arg1 = t->t_dtrace_npc (step)
1758	brz,pt	%g2, 1f
1759	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1760	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
17611:
1762	mov	%g7, %l3			! arg2 == misaligned address
1763
1764	rdpr	%tstate, %g1			! cwp for trap handler
1765	rdpr	%cwp, %g4
1766	bclr	TSTATE_CWP_MASK, %g1
1767	wrpr	%g1, %g4, %tstate
1768	wrpr	%g0, %g5, %tnpc			! trap handler address
1769	FAST_TRAP_DONE
1770	/* NOTREACHED */
1771
1772.check_v9utrap:
1773	rdpr	%tstate, %g1
1774	btst	TSTATE_PRIV, %g1
1775	bnz,a,pn %icc, 3f
1776	  nop
1777	CPU_ADDR(%g4, %g1)				! load CPU struct addr
1778	ldn	[%g4 + CPU_THREAD], %g5			! load thread pointer
1779	ldn	[%g5 + T_PROCP], %g5			! load proc pointer
1780	ldn	[%g5 + P_UTRAPS], %g5			! are there utraps?
1781
1782	cmp	%g3, T_SOFTWARE_TRAP
1783	bne,a,pt %icc, 1f
1784	  nop
1785
1786	brz,pt %g5, 3f			! if p_utraps == NULL goto trap()
1787	  rdpr	%tt, %g3		! delay - get actual hw trap type
1788
1789	sub	%g3, 254, %g1		! UT_TRAP_INSTRUCTION_16 = p_utraps[18]
1790	ba,pt	%icc, 2f
1791	  smul	%g1, CPTRSIZE, %g2
17921:
1793	brz,a,pt %g5, 3f		! if p_utraps == NULL goto trap()
1794	  nop
1795
1796	cmp	%g3, T_UNIMP_INSTR
1797	bne,a,pt %icc, 2f
1798	  nop
1799
1800	mov	1, %g1
1801	st	%g1, [%g4 + CPU_TL1_HDLR] ! set CPU_TL1_HDLR
1802	rdpr	%tpc, %g1		! ld trapping instruction using
1803	lduwa	[%g1]ASI_AIUP, %g1	! "AS IF USER" ASI which could fault
1804	st	%g0, [%g4 + CPU_TL1_HDLR] ! clr CPU_TL1_HDLR
1805
1806	sethi	%hi(0xc1c00000), %g4	! setup mask for illtrap instruction
1807	andcc	%g1, %g4, %g4		! and instruction with mask
1808	bnz,a,pt %icc, 3f		! if %g4 == zero, %g1 is an ILLTRAP
1809	  nop				! fall thru to setup
18102:
1811	ldn	[%g5 + %g2], %g5
1812	brnz,a,pt %g5, .setup_v9utrap
1813	  nop
18143:
1815	set	trap, %g1
1816	ba,pt	%xcc, sys_trap
1817	  sub	%g0, 1, %g4
1818	/* NOTREACHED */
1819
1820/*
1821 * Register Inputs:
1822 *	%g5		user trap handler
1823 */
1824.setup_v9utrap:
1825	set	trap, %g1			! setup in case we go
1826	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1827	sub	%g0, 1, %g4			! the save instruction below
1828
1829	/*
1830	 * If the DTrace pid provider is single stepping a copied-out
1831	 * instruction, t->t_dtrace_step will be set. In that case we need
1832	 * to abort the single-stepping (since execution of the instruction
1833	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1834	 */
1835	save	%sp, -SA(MINFRAME64), %sp	! window for trap handler
1836	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1837	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1838	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1839	rdpr	%tnpc, %l7			! arg1 == tnpc
1840	brz,pt	%g2, 1f
1841	rdpr	%tpc, %l6			! arg0 == tpc
1842
1843	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1844	ldn	[%g1 + T_DTRACE_NPC], %l7	! arg1 == t->t_dtrace_npc (step)
1845	brz,pt	%g2, 1f
1846	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1847	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
18481:
1849	rdpr	%tstate, %g2			! cwp for trap handler
1850	rdpr	%cwp, %g4
1851	bclr	TSTATE_CWP_MASK, %g2
1852	wrpr	%g2, %g4, %tstate
1853
1854	ldn	[%g1 + T_PROCP], %g4		! load proc pointer
1855	ldn	[%g4 + P_AS], %g4		! load as pointer
1856	ldn	[%g4 + A_USERLIMIT], %g4	! load as userlimit
1857	cmp	%l7, %g4			! check for single-step set
1858	bne,pt	%xcc, 4f
1859	  nop
1860	ldn	[%g1 + T_LWP], %g1		! load klwp pointer
1861	ld	[%g1 + PCB_STEP], %g4		! load single-step flag
1862	cmp	%g4, STEP_ACTIVE		! step flags set in pcb?
1863	bne,pt	%icc, 4f
1864	  nop
1865	stn	%g5, [%g1 + PCB_TRACEPC]	! save trap handler addr in pcb
1866	mov	%l7, %g4			! on entry to precise user trap
1867	add	%l6, 4, %l7			! handler, %l6 == pc, %l7 == npc
1868						! at time of trap
1869	wrpr	%g0, %g4, %tnpc			! generate FLTBOUNDS,
1870						! %g4 == userlimit
1871	FAST_TRAP_DONE
1872	/* NOTREACHED */
18734:
1874	wrpr	%g0, %g5, %tnpc			! trap handler address
1875	FAST_TRAP_DONE_CHK_INTR
1876	/* NOTREACHED */
1877
1878.fp_exception:
1879	CPU_ADDR(%g1, %g4)
1880	stx	%fsr, [%g1 + CPU_TMP1]
1881	ldx	[%g1 + CPU_TMP1], %g2
1882
1883	/*
1884	 * Cheetah takes unfinished_FPop trap for certain range of operands
1885	 * to the "fitos" instruction. Instead of going through the slow
1886	 * software emulation path, we try to simulate the "fitos" instruction
1887	 * via "fitod" and "fdtos" provided the following conditions are met:
1888	 *
1889	 *	fpu_exists is set (if DEBUG)
1890	 *	not in privileged mode
1891	 *	ftt is unfinished_FPop
1892	 *	NXM IEEE trap is not enabled
1893	 *	instruction at %tpc is "fitos"
1894	 *
1895	 *  Usage:
1896	 *	%g1	per cpu address
1897	 *	%g2	%fsr
1898	 *	%g6	user instruction
1899	 *
1900	 * Note that we can take a memory access related trap while trying
1901	 * to fetch the user instruction. Therefore, we set CPU_TL1_HDLR
1902	 * flag to catch those traps and let the SFMMU code deal with page
1903	 * fault and data access exception.
1904	 */
1905#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
1906	sethi	%hi(fpu_exists), %g7
1907	ld	[%g7 + %lo(fpu_exists)], %g7
1908	brz,pn %g7, .fp_exception_cont
1909	  nop
1910#endif
1911	rdpr	%tstate, %g7			! branch if in privileged mode
1912	btst	TSTATE_PRIV, %g7
1913	bnz,pn	%xcc, .fp_exception_cont
1914	srl	%g2, FSR_FTT_SHIFT, %g7		! extract ftt from %fsr
1915	and	%g7, (FSR_FTT>>FSR_FTT_SHIFT), %g7
1916	cmp	%g7, FTT_UNFIN
1917	set	FSR_TEM_NX, %g5
1918	bne,pn	%xcc, .fp_exception_cont	! branch if NOT unfinished_FPop
1919	  andcc	%g2, %g5, %g0
1920	bne,pn	%xcc, .fp_exception_cont	! branch if FSR_TEM_NX enabled
1921	  rdpr	%tpc, %g5			! get faulting PC
1922
1923	or	%g0, 1, %g7
1924	st	%g7, [%g1 + CPU_TL1_HDLR]	! set tl1_hdlr flag
1925	lda	[%g5]ASI_USER, %g6		! get user's instruction
1926	st	%g0, [%g1 + CPU_TL1_HDLR]	! clear tl1_hdlr flag
1927
1928	set	FITOS_INSTR_MASK, %g7
1929	and	%g6, %g7, %g7
1930	set	FITOS_INSTR, %g5
1931	cmp	%g7, %g5
1932	bne,pn	%xcc, .fp_exception_cont	! branch if not FITOS_INSTR
1933	 nop
1934
1935	/*
1936	 * This is unfinished FPops trap for "fitos" instruction. We
1937	 * need to simulate "fitos" via "fitod" and "fdtos" instruction
1938	 * sequence.
1939	 *
1940	 * We need a temporary FP register to do the conversion. Since
1941	 * both source and destination operands for the "fitos" instruction
1942	 * have to be within %f0-%f31, we use an FP register from the upper
1943	 * half to guarantee that it won't collide with the source or the
1944	 * dest operand. However, we do have to save and restore its value.
1945	 *
1946	 * We use %d62 as a temporary FP register for the conversion and
1947	 * branch to appropriate instruction within the conversion tables
1948	 * based upon the rs2 and rd values.
1949	 */
1950
1951	std	%d62, [%g1 + CPU_TMP1]		! save original value
1952
1953	srl	%g6, FITOS_RS2_SHIFT, %g7
1954	and	%g7, FITOS_REG_MASK, %g7
1955	set	_fitos_fitod_table, %g4
1956	sllx	%g7, 2, %g7
1957	jmp	%g4 + %g7
1958	  ba,pt	%xcc, _fitos_fitod_done
1959	.empty
1960
1961_fitos_fitod_table:
1962	  fitod	%f0, %d62
1963	  fitod	%f1, %d62
1964	  fitod	%f2, %d62
1965	  fitod	%f3, %d62
1966	  fitod	%f4, %d62
1967	  fitod	%f5, %d62
1968	  fitod	%f6, %d62
1969	  fitod	%f7, %d62
1970	  fitod	%f8, %d62
1971	  fitod	%f9, %d62
1972	  fitod	%f10, %d62
1973	  fitod	%f11, %d62
1974	  fitod	%f12, %d62
1975	  fitod	%f13, %d62
1976	  fitod	%f14, %d62
1977	  fitod	%f15, %d62
1978	  fitod	%f16, %d62
1979	  fitod	%f17, %d62
1980	  fitod	%f18, %d62
1981	  fitod	%f19, %d62
1982	  fitod	%f20, %d62
1983	  fitod	%f21, %d62
1984	  fitod	%f22, %d62
1985	  fitod	%f23, %d62
1986	  fitod	%f24, %d62
1987	  fitod	%f25, %d62
1988	  fitod	%f26, %d62
1989	  fitod	%f27, %d62
1990	  fitod	%f28, %d62
1991	  fitod	%f29, %d62
1992	  fitod	%f30, %d62
1993	  fitod	%f31, %d62
1994_fitos_fitod_done:
1995
1996	/*
1997	 * Now convert data back into single precision
1998	 */
1999	srl	%g6, FITOS_RD_SHIFT, %g7
2000	and	%g7, FITOS_REG_MASK, %g7
2001	set	_fitos_fdtos_table, %g4
2002	sllx	%g7, 2, %g7
2003	jmp	%g4 + %g7
2004	  ba,pt	%xcc, _fitos_fdtos_done
2005	.empty
2006
2007_fitos_fdtos_table:
2008	  fdtos	%d62, %f0
2009	  fdtos	%d62, %f1
2010	  fdtos	%d62, %f2
2011	  fdtos	%d62, %f3
2012	  fdtos	%d62, %f4
2013	  fdtos	%d62, %f5
2014	  fdtos	%d62, %f6
2015	  fdtos	%d62, %f7
2016	  fdtos	%d62, %f8
2017	  fdtos	%d62, %f9
2018	  fdtos	%d62, %f10
2019	  fdtos	%d62, %f11
2020	  fdtos	%d62, %f12
2021	  fdtos	%d62, %f13
2022	  fdtos	%d62, %f14
2023	  fdtos	%d62, %f15
2024	  fdtos	%d62, %f16
2025	  fdtos	%d62, %f17
2026	  fdtos	%d62, %f18
2027	  fdtos	%d62, %f19
2028	  fdtos	%d62, %f20
2029	  fdtos	%d62, %f21
2030	  fdtos	%d62, %f22
2031	  fdtos	%d62, %f23
2032	  fdtos	%d62, %f24
2033	  fdtos	%d62, %f25
2034	  fdtos	%d62, %f26
2035	  fdtos	%d62, %f27
2036	  fdtos	%d62, %f28
2037	  fdtos	%d62, %f29
2038	  fdtos	%d62, %f30
2039	  fdtos	%d62, %f31
2040_fitos_fdtos_done:
2041
2042	ldd	[%g1 + CPU_TMP1], %d62		! restore %d62
2043
2044#if DEBUG
2045	/*
2046	 * Update FPop_unfinished trap kstat
2047	 */
2048	set	fpustat+FPUSTAT_UNFIN_KSTAT, %g7
2049	ldx	[%g7], %g5
20501:
2051	add	%g5, 1, %g6
2052
2053	casxa	[%g7] ASI_N, %g5, %g6
2054	cmp	%g5, %g6
2055	bne,a,pn %xcc, 1b
2056	  or	%g0, %g6, %g5
2057
2058	/*
2059	 * Update fpu_sim_fitos kstat
2060	 */
2061	set	fpuinfo+FPUINFO_FITOS_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#endif /* DEBUG */
2071
2072	FAST_TRAP_DONE
2073
2074.fp_exception_cont:
2075	/*
2076	 * Let _fp_exception deal with simulating FPop instruction.
2077	 * Note that we need to pass %fsr in %g2 (already read above).
2078	 */
2079
2080	set	_fp_exception, %g1
2081	ba,pt	%xcc, sys_trap
2082	sub	%g0, 1, %g4
2083
2084	.global opl_cleanw_patch
2085opl_cleanw_patch:
2086.clean_windows:
2087	set	trap, %g1
2088	mov	T_FLUSH_PCB, %g3
2089	sub	%g0, 1, %g4
2090	save
2091	flushw
2092	restore
2093	wrpr	%g0, %g0, %cleanwin	! no clean windows
2094
2095	CPU_ADDR(%g4, %g5)
2096	ldn	[%g4 + CPU_MPCB], %g4
2097	brz,a,pn %g4, 1f
2098	  nop
2099	ld	[%g4 + MPCB_WSTATE], %g5
2100	add	%g5, WSTATE_CLEAN_OFFSET, %g5
2101	wrpr	%g0, %g5, %wstate
21021:	FAST_TRAP_DONE
2103
2104/*
2105 * .spill_clean: clean the previous window, restore the wstate, and
2106 * "done".
2107 *
2108 * Entry: %g7 contains new wstate
2109 */
2110.spill_clean:
2111	sethi	%hi(nwin_minus_one), %g5
2112	ld	[%g5 + %lo(nwin_minus_one)], %g5 ! %g5 = nwin - 1
2113	rdpr	%cwp, %g6			! %g6 = %cwp
2114	deccc	%g6				! %g6--
2115	movneg	%xcc, %g5, %g6			! if (%g6<0) %g6 = nwin-1
2116	wrpr	%g6, %cwp
2117	TT_TRACE_L(trace_win)
2118	clr	%l0
2119	clr	%l1
2120	clr	%l2
2121	clr	%l3
2122	clr	%l4
2123	clr	%l5
2124	clr	%l6
2125	clr	%l7
2126	wrpr	%g0, %g7, %wstate
2127	saved
2128	retry			! restores correct %cwp
2129
2130.fix_alignment:
2131	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2132	ldn	[%g1 + CPU_THREAD], %g1	! load thread pointer
2133	ldn	[%g1 + T_PROCP], %g1
2134	mov	1, %g2
2135	stb	%g2, [%g1 + P_FIXALIGNMENT]
2136	FAST_TRAP_DONE
2137
2138#define	STDF_REG(REG, ADDR, TMP)		\
2139	sll	REG, 3, REG			;\
2140mark1:	set	start1, TMP			;\
2141	jmp	REG + TMP			;\
2142	  nop					;\
2143start1:	ba,pt	%xcc, done1			;\
2144	  std	%f0, [ADDR + CPU_TMP1]		;\
2145	ba,pt	%xcc, done1			;\
2146	  std	%f32, [ADDR + CPU_TMP1]		;\
2147	ba,pt	%xcc, done1			;\
2148	  std	%f2, [ADDR + CPU_TMP1]		;\
2149	ba,pt	%xcc, done1			;\
2150	  std	%f34, [ADDR + CPU_TMP1]		;\
2151	ba,pt	%xcc, done1			;\
2152	  std	%f4, [ADDR + CPU_TMP1]		;\
2153	ba,pt	%xcc, done1			;\
2154	  std	%f36, [ADDR + CPU_TMP1]		;\
2155	ba,pt	%xcc, done1			;\
2156	  std	%f6, [ADDR + CPU_TMP1]		;\
2157	ba,pt	%xcc, done1			;\
2158	  std	%f38, [ADDR + CPU_TMP1]		;\
2159	ba,pt	%xcc, done1			;\
2160	  std	%f8, [ADDR + CPU_TMP1]		;\
2161	ba,pt	%xcc, done1			;\
2162	  std	%f40, [ADDR + CPU_TMP1]		;\
2163	ba,pt	%xcc, done1			;\
2164	  std	%f10, [ADDR + CPU_TMP1]		;\
2165	ba,pt	%xcc, done1			;\
2166	  std	%f42, [ADDR + CPU_TMP1]		;\
2167	ba,pt	%xcc, done1			;\
2168	  std	%f12, [ADDR + CPU_TMP1]		;\
2169	ba,pt	%xcc, done1			;\
2170	  std	%f44, [ADDR + CPU_TMP1]		;\
2171	ba,pt	%xcc, done1			;\
2172	  std	%f14, [ADDR + CPU_TMP1]		;\
2173	ba,pt	%xcc, done1			;\
2174	  std	%f46, [ADDR + CPU_TMP1]		;\
2175	ba,pt	%xcc, done1			;\
2176	  std	%f16, [ADDR + CPU_TMP1]		;\
2177	ba,pt	%xcc, done1			;\
2178	  std	%f48, [ADDR + CPU_TMP1]		;\
2179	ba,pt	%xcc, done1			;\
2180	  std	%f18, [ADDR + CPU_TMP1]		;\
2181	ba,pt	%xcc, done1			;\
2182	  std	%f50, [ADDR + CPU_TMP1]		;\
2183	ba,pt	%xcc, done1			;\
2184	  std	%f20, [ADDR + CPU_TMP1]		;\
2185	ba,pt	%xcc, done1			;\
2186	  std	%f52, [ADDR + CPU_TMP1]		;\
2187	ba,pt	%xcc, done1			;\
2188	  std	%f22, [ADDR + CPU_TMP1]		;\
2189	ba,pt	%xcc, done1			;\
2190	  std	%f54, [ADDR + CPU_TMP1]		;\
2191	ba,pt	%xcc, done1			;\
2192	  std	%f24, [ADDR + CPU_TMP1]		;\
2193	ba,pt	%xcc, done1			;\
2194	  std	%f56, [ADDR + CPU_TMP1]		;\
2195	ba,pt	%xcc, done1			;\
2196	  std	%f26, [ADDR + CPU_TMP1]		;\
2197	ba,pt	%xcc, done1			;\
2198	  std	%f58, [ADDR + CPU_TMP1]		;\
2199	ba,pt	%xcc, done1			;\
2200	  std	%f28, [ADDR + CPU_TMP1]		;\
2201	ba,pt	%xcc, done1			;\
2202	  std	%f60, [ADDR + CPU_TMP1]		;\
2203	ba,pt	%xcc, done1			;\
2204	  std	%f30, [ADDR + CPU_TMP1]		;\
2205	ba,pt	%xcc, done1			;\
2206	  std	%f62, [ADDR + CPU_TMP1]		;\
2207done1:
2208
2209#define	LDDF_REG(REG, ADDR, TMP)		\
2210	sll	REG, 3, REG			;\
2211mark2:	set	start2, TMP			;\
2212	jmp	REG + TMP			;\
2213	  nop					;\
2214start2:	ba,pt	%xcc, done2			;\
2215	  ldd	[ADDR + CPU_TMP1], %f0		;\
2216	ba,pt	%xcc, done2			;\
2217	  ldd	[ADDR + CPU_TMP1], %f32		;\
2218	ba,pt	%xcc, done2			;\
2219	  ldd	[ADDR + CPU_TMP1], %f2		;\
2220	ba,pt	%xcc, done2			;\
2221	  ldd	[ADDR + CPU_TMP1], %f34		;\
2222	ba,pt	%xcc, done2			;\
2223	  ldd	[ADDR + CPU_TMP1], %f4		;\
2224	ba,pt	%xcc, done2			;\
2225	  ldd	[ADDR + CPU_TMP1], %f36		;\
2226	ba,pt	%xcc, done2			;\
2227	  ldd	[ADDR + CPU_TMP1], %f6		;\
2228	ba,pt	%xcc, done2			;\
2229	  ldd	[ADDR + CPU_TMP1], %f38		;\
2230	ba,pt	%xcc, done2			;\
2231	  ldd	[ADDR + CPU_TMP1], %f8		;\
2232	ba,pt	%xcc, done2			;\
2233	  ldd	[ADDR + CPU_TMP1], %f40		;\
2234	ba,pt	%xcc, done2			;\
2235	  ldd	[ADDR + CPU_TMP1], %f10		;\
2236	ba,pt	%xcc, done2			;\
2237	  ldd	[ADDR + CPU_TMP1], %f42		;\
2238	ba,pt	%xcc, done2			;\
2239	  ldd	[ADDR + CPU_TMP1], %f12		;\
2240	ba,pt	%xcc, done2			;\
2241	  ldd	[ADDR + CPU_TMP1], %f44		;\
2242	ba,pt	%xcc, done2			;\
2243	  ldd	[ADDR + CPU_TMP1], %f14		;\
2244	ba,pt	%xcc, done2			;\
2245	  ldd	[ADDR + CPU_TMP1], %f46		;\
2246	ba,pt	%xcc, done2			;\
2247	  ldd	[ADDR + CPU_TMP1], %f16		;\
2248	ba,pt	%xcc, done2			;\
2249	  ldd	[ADDR + CPU_TMP1], %f48		;\
2250	ba,pt	%xcc, done2			;\
2251	  ldd	[ADDR + CPU_TMP1], %f18		;\
2252	ba,pt	%xcc, done2			;\
2253	  ldd	[ADDR + CPU_TMP1], %f50		;\
2254	ba,pt	%xcc, done2			;\
2255	  ldd	[ADDR + CPU_TMP1], %f20		;\
2256	ba,pt	%xcc, done2			;\
2257	  ldd	[ADDR + CPU_TMP1], %f52		;\
2258	ba,pt	%xcc, done2			;\
2259	  ldd	[ADDR + CPU_TMP1], %f22		;\
2260	ba,pt	%xcc, done2			;\
2261	  ldd	[ADDR + CPU_TMP1], %f54		;\
2262	ba,pt	%xcc, done2			;\
2263	  ldd	[ADDR + CPU_TMP1], %f24		;\
2264	ba,pt	%xcc, done2			;\
2265	  ldd	[ADDR + CPU_TMP1], %f56		;\
2266	ba,pt	%xcc, done2			;\
2267	  ldd	[ADDR + CPU_TMP1], %f26		;\
2268	ba,pt	%xcc, done2			;\
2269	  ldd	[ADDR + CPU_TMP1], %f58		;\
2270	ba,pt	%xcc, done2			;\
2271	  ldd	[ADDR + CPU_TMP1], %f28		;\
2272	ba,pt	%xcc, done2			;\
2273	  ldd	[ADDR + CPU_TMP1], %f60		;\
2274	ba,pt	%xcc, done2			;\
2275	  ldd	[ADDR + CPU_TMP1], %f30		;\
2276	ba,pt	%xcc, done2			;\
2277	  ldd	[ADDR + CPU_TMP1], %f62		;\
2278done2:
2279
2280.lddf_exception_not_aligned:
2281	/*
2282	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2283	 */
2284	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2285
2286#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2287	sethi	%hi(fpu_exists), %g2		! check fpu_exists
2288	ld	[%g2 + %lo(fpu_exists)], %g2
2289	brz,a,pn %g2, 4f
2290	  nop
2291#endif
2292	CPU_ADDR(%g1, %g4)
2293	or	%g0, 1, %g4
2294	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2295
2296	rdpr	%tpc, %g2
2297	lda	[%g2]ASI_AIUP, %g6	! get the user's lddf instruction
2298	srl	%g6, 23, %g1		! using ldda or not?
2299	and	%g1, 1, %g1
2300	brz,a,pt %g1, 2f		! check for ldda instruction
2301	  nop
2302	srl	%g6, 13, %g1		! check immflag
2303	and	%g1, 1, %g1
2304	rdpr	%tstate, %g2		! %tstate in %g2
2305	brnz,a,pn %g1, 1f
2306	  srl	%g2, 31, %g1		! get asi from %tstate
2307	srl	%g6, 5, %g1		! get asi from instruction
2308	and	%g1, 0xFF, %g1		! imm_asi field
23091:
2310	cmp	%g1, ASI_P		! primary address space
2311	be,a,pt %icc, 2f
2312	  nop
2313	cmp	%g1, ASI_PNF		! primary no fault address space
2314	be,a,pt %icc, 2f
2315	  nop
2316	cmp	%g1, ASI_S		! secondary address space
2317	be,a,pt %icc, 2f
2318	  nop
2319	cmp	%g1, ASI_SNF		! secondary no fault address space
2320	bne,a,pn %icc, 3f
2321	  nop
23222:
2323	lduwa	[%g5]ASI_USER, %g7	! get first half of misaligned data
2324	add	%g5, 4, %g5		! increment misaligned data address
2325	lduwa	[%g5]ASI_USER, %g5	! get second half of misaligned data
2326
2327	sllx	%g7, 32, %g7
2328	or	%g5, %g7, %g5		! combine data
2329	CPU_ADDR(%g7, %g1)		! save data on a per-cpu basis
2330	stx	%g5, [%g7 + CPU_TMP1]	! save in cpu_tmp1
2331
2332	srl	%g6, 25, %g3		! %g6 has the instruction
2333	and	%g3, 0x1F, %g3		! %g3 has rd
2334	LDDF_REG(%g3, %g7, %g4)
2335
2336	CPU_ADDR(%g1, %g4)
2337	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2338	FAST_TRAP_DONE
23393:
2340	CPU_ADDR(%g1, %g4)
2341	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
23424:
2343	set	T_USER, %g3		! trap type in %g3
2344	or	%g3, T_LDDF_ALIGN, %g3
2345	mov	%g5, %g2		! misaligned vaddr in %g2
2346	set	fpu_trap, %g1		! goto C for the little and
2347	ba,pt	%xcc, sys_trap		! no fault little asi's
2348	  sub	%g0, 1, %g4
2349
2350.stdf_exception_not_aligned:
2351	/*
2352	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2353	 */
2354	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2355
2356#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2357	sethi	%hi(fpu_exists), %g7		! check fpu_exists
2358	ld	[%g7 + %lo(fpu_exists)], %g3
2359	brz,a,pn %g3, 4f
2360	  nop
2361#endif
2362	CPU_ADDR(%g1, %g4)
2363	or	%g0, 1, %g4
2364	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2365
2366	rdpr	%tpc, %g2
2367	lda	[%g2]ASI_AIUP, %g6	! get the user's stdf instruction
2368
2369	srl	%g6, 23, %g1		! using stda or not?
2370	and	%g1, 1, %g1
2371	brz,a,pt %g1, 2f		! check for stda instruction
2372	  nop
2373	srl	%g6, 13, %g1		! check immflag
2374	and	%g1, 1, %g1
2375	rdpr	%tstate, %g2		! %tstate in %g2
2376	brnz,a,pn %g1, 1f
2377	  srl	%g2, 31, %g1		! get asi from %tstate
2378	srl	%g6, 5, %g1		! get asi from instruction
2379	and	%g1, 0xFF, %g1		! imm_asi field
23801:
2381	cmp	%g1, ASI_P		! primary address space
2382	be,a,pt %icc, 2f
2383	  nop
2384	cmp	%g1, ASI_S		! secondary address space
2385	bne,a,pn %icc, 3f
2386	  nop
23872:
2388	srl	%g6, 25, %g6
2389	and	%g6, 0x1F, %g6		! %g6 has rd
2390	CPU_ADDR(%g7, %g1)
2391	STDF_REG(%g6, %g7, %g4)		! STDF_REG(REG, ADDR, TMP)
2392
2393	ldx	[%g7 + CPU_TMP1], %g6
2394	srlx	%g6, 32, %g7
2395	stuwa	%g7, [%g5]ASI_USER	! first half
2396	add	%g5, 4, %g5		! increment misaligned data address
2397	stuwa	%g6, [%g5]ASI_USER	! second half
2398
2399	CPU_ADDR(%g1, %g4)
2400	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2401	FAST_TRAP_DONE
24023:
2403	CPU_ADDR(%g1, %g4)
2404	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
24054:
2406	set	T_USER, %g3		! trap type in %g3
2407	or	%g3, T_STDF_ALIGN, %g3
2408	mov	%g5, %g2		! misaligned vaddr in %g2
2409	set	fpu_trap, %g1		! goto C for the little and
2410	ba,pt	%xcc, sys_trap		! nofault little asi's
2411	  sub	%g0, 1, %g4
2412
2413#ifdef DEBUG_USER_TRAPTRACECTL
2414
2415.traptrace_freeze:
2416	mov	%l0, %g1 ; mov	%l1, %g2 ; mov	%l2, %g3 ; mov	%l4, %g4
2417	TT_TRACE_L(trace_win)
2418	mov	%g4, %l4 ; mov	%g3, %l2 ; mov	%g2, %l1 ; mov	%g1, %l0
2419	set	trap_freeze, %g1
2420	mov	1, %g2
2421	st	%g2, [%g1]
2422	FAST_TRAP_DONE
2423
2424.traptrace_unfreeze:
2425	set	trap_freeze, %g1
2426	st	%g0, [%g1]
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	FAST_TRAP_DONE
2431
2432#endif /* DEBUG_USER_TRAPTRACECTL */
2433
2434.getcc:
2435	CPU_ADDR(%g1, %g2)
2436	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2437	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2438	rdpr	%tstate, %g3			! get tstate
2439	srlx	%g3, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2440	set	PSR_ICC, %g2
2441	and	%o0, %g2, %o0			! mask out the rest
2442	srl	%o0, PSR_ICC_SHIFT, %o0		! right justify
2443	rdpr	%pstate, %o1
2444	wrpr	%o1, PSTATE_AG, %pstate		! get into normal globals
2445	mov	%o0, %g1			! move ccr to normal %g1
2446	wrpr	%g0, %o1, %pstate		! back into alternate globals
2447	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2448	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2449	FAST_TRAP_DONE
2450
2451.setcc:
2452	CPU_ADDR(%g1, %g2)
2453	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2454	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2455	rdpr	%pstate, %o0
2456	wrpr	%o0, PSTATE_AG, %pstate		! get into normal globals
2457	mov	%g1, %o1
2458	wrpr	%g0, %o0, %pstate		! back to alternates
2459	sll	%o1, PSR_ICC_SHIFT, %g2
2460	set	PSR_ICC, %g3
2461	and	%g2, %g3, %g2			! mask out rest
2462	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g2
2463	rdpr	%tstate, %g3			! get tstate
2464	srl	%g3, 0, %g3			! clear upper word
2465	or	%g3, %g2, %g3			! or in new bits
2466	wrpr	%g3, %tstate
2467	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2468	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2469	FAST_TRAP_DONE
2470
2471/*
2472 * getpsr(void)
2473 * Note that the xcc part of the ccr is not provided.
2474 * The V8 code shows why the V9 trap is not faster:
2475 * #define GETPSR_TRAP() \
2476 *      mov %psr, %i0; jmp %l2; rett %l2+4; nop;
2477 */
2478
2479	.type	.getpsr, #function
2480.getpsr:
2481	rdpr	%tstate, %g1			! get tstate
2482	srlx	%g1, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2483	set	PSR_ICC, %g2
2484	and	%o0, %g2, %o0			! mask out the rest
2485
2486	rd	%fprs, %g1			! get fprs
2487	and	%g1, FPRS_FEF, %g2		! mask out dirty upper/lower
2488	sllx	%g2, PSR_FPRS_FEF_SHIFT, %g2	! shift fef to V8 psr.ef
2489	or	%o0, %g2, %o0			! or result into psr.ef
2490
2491	set	V9_PSR_IMPLVER, %g2		! SI assigned impl/ver: 0xef
2492	or	%o0, %g2, %o0			! or psr.impl/ver
2493	FAST_TRAP_DONE
2494	SET_SIZE(.getpsr)
2495
2496/*
2497 * setpsr(newpsr)
2498 * Note that there is no support for ccr.xcc in the V9 code.
2499 */
2500
2501	.type	.setpsr, #function
2502.setpsr:
2503	rdpr	%tstate, %g1			! get tstate
2504!	setx	TSTATE_V8_UBITS, %g2
2505	or 	%g0, CCR_ICC, %g3
2506	sllx	%g3, TSTATE_CCR_SHIFT, %g2
2507
2508	andn	%g1, %g2, %g1			! zero current user bits
2509	set	PSR_ICC, %g2
2510	and	%g2, %o0, %g2			! clear all but psr.icc bits
2511	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g3	! shift to tstate.ccr.icc
2512	wrpr	%g1, %g3, %tstate		! write tstate
2513
2514	set	PSR_EF, %g2
2515	and	%g2, %o0, %g2			! clear all but fp enable bit
2516	srlx	%g2, PSR_FPRS_FEF_SHIFT, %g4	! shift ef to V9 fprs.fef
2517	wr	%g0, %g4, %fprs			! write fprs
2518
2519	CPU_ADDR(%g1, %g2)			! load CPU struct addr to %g1
2520	ldn	[%g1 + CPU_THREAD], %g2		! load thread pointer
2521	ldn	[%g2 + T_LWP], %g3		! load klwp pointer
2522	ldn	[%g3 + LWP_FPU], %g2		! get lwp_fpu pointer
2523	stuw	%g4, [%g2 + FPU_FPRS]		! write fef value to fpu_fprs
2524	srlx	%g4, 2, %g4			! shift fef value to bit 0
2525	stub	%g4, [%g2 + FPU_EN]		! write fef value to fpu_en
2526	FAST_TRAP_DONE
2527	SET_SIZE(.setpsr)
2528
2529/*
2530 * getlgrp
2531 * get home lgrpid on which the calling thread is currently executing.
2532 */
2533	.type	.getlgrp, #function
2534.getlgrp:
2535	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2536	ld	[%g1 + CPU_ID], %o0	! load cpu_id
2537	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2538	ldn	[%g2 + T_LPL], %g2	! load lpl pointer
2539	ld	[%g2 + LPL_LGRPID], %g1	! load lpl_lgrpid
2540	sra	%g1, 0, %o1
2541	FAST_TRAP_DONE
2542	SET_SIZE(.getlgrp)
2543
2544/*
2545 * Entry for old 4.x trap (trap 0).
2546 */
2547	ENTRY_NP(syscall_trap_4x)
2548	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2549	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2550	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2551	ld	[%g2 + PCB_TRAP0], %g2	! lwp->lwp_pcb.pcb_trap0addr
2552	brz,pn	%g2, 1f			! has it been set?
2553	st	%l0, [%g1 + CPU_TMP1]	! delay - save some locals
2554	st	%l1, [%g1 + CPU_TMP2]
2555	rdpr	%tnpc, %l1		! save old tnpc
2556	wrpr	%g0, %g2, %tnpc		! setup tnpc
2557
2558	rdpr	%pstate, %l0
2559	wrpr	%l0, PSTATE_AG, %pstate	! switch to normal globals
2560	mov	%l1, %g6		! pass tnpc to user code in %g6
2561	wrpr	%l0, %g0, %pstate	! switch back to alternate globals
2562
2563	! Note that %g1 still contains CPU struct addr
2564	ld	[%g1 + CPU_TMP2], %l1	! restore locals
2565	ld	[%g1 + CPU_TMP1], %l0
2566	FAST_TRAP_DONE_CHK_INTR
25671:
2568	mov	%g1, %l0
2569	st	%l1, [%g1 + CPU_TMP2]
2570	rdpr	%pstate, %l1
2571	wrpr	%l1, PSTATE_AG, %pstate
2572	!
2573	! check for old syscall mmap which is the only different one which
2574	! must be the same.  Others are handled in the compatibility library.
2575	!
2576	cmp	%g1, OSYS_mmap	! compare to old 4.x mmap
2577	movz	%icc, SYS_mmap, %g1
2578	wrpr	%g0, %l1, %pstate
2579	ld	[%l0 + CPU_TMP2], %l1	! restore locals
2580	ld	[%l0 + CPU_TMP1], %l0
2581	SYSCALL(syscall_trap32)
2582	SET_SIZE(syscall_trap_4x)
2583
2584/*
2585 * Handler for software trap 9.
2586 * Set trap0 emulation address for old 4.x system call trap.
2587 * XXX - this should be a system call.
2588 */
2589	ENTRY_NP(set_trap0_addr)
2590	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2591	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2592	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2593	st	%l0, [%g1 + CPU_TMP1]	! save some locals
2594	st	%l1, [%g1 + CPU_TMP2]
2595	rdpr	%pstate, %l0
2596	wrpr	%l0, PSTATE_AG, %pstate
2597	mov	%g1, %l1
2598	wrpr	%g0, %l0, %pstate
2599	andn	%l1, 3, %l1		! force alignment
2600	st	%l1, [%g2 + PCB_TRAP0]	! lwp->lwp_pcb.pcb_trap0addr
2601	ld	[%g1 + CPU_TMP1], %l0	! restore locals
2602	ld	[%g1 + CPU_TMP2], %l1
2603	FAST_TRAP_DONE
2604	SET_SIZE(set_trap0_addr)
2605
2606/*
2607 * mmu_trap_tl1
2608 * trap handler for unexpected mmu traps.
2609 * simply checks if the trap was a user lddf/stdf alignment trap, in which
2610 * case we go to fpu_trap or a user trap from the window handler, in which
2611 * case we go save the state on the pcb.  Otherwise, we go to ptl1_panic.
2612 */
2613	.type	mmu_trap_tl1, #function
2614mmu_trap_tl1:
2615#ifdef	TRAPTRACE
2616	TRACE_PTR(%g5, %g6)
2617	GET_TRACE_TICK(%g6, %g7)
2618	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2619	rdpr	%tl, %g6
2620	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2621	rdpr	%tt, %g6
2622	stha	%g6, [%g5 + TRAP_ENT_TT]%asi
2623	rdpr	%tstate, %g6
2624	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi
2625	stna	%sp, [%g5 + TRAP_ENT_SP]%asi
2626	stna	%g0, [%g5 + TRAP_ENT_TR]%asi
2627	rdpr	%tpc, %g6
2628	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2629	set	MMU_SFAR, %g6
2630	ldxa	[%g6]ASI_DMMU, %g6
2631	stxa	%g6, [%g5 + TRAP_ENT_F1]%asi
2632	CPU_PADDR(%g7, %g6);
2633	add	%g7, CPU_TL1_HDLR, %g7
2634	lda	[%g7]ASI_MEM, %g6
2635	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2636	set	0xdeadbeef, %g6
2637	stna	%g6, [%g5 + TRAP_ENT_F3]%asi
2638	stna	%g6, [%g5 + TRAP_ENT_F4]%asi
2639	TRACE_NEXT(%g5, %g6, %g7)
2640#endif /* TRAPTRACE */
2641
2642	GET_CPU_IMPL(%g5)
2643	cmp	%g5, PANTHER_IMPL
2644	bne	mmu_trap_tl1_4
2645	  nop
2646	rdpr	%tt, %g5
2647	cmp	%g5, T_DATA_EXCEPTION
2648	bne	mmu_trap_tl1_4
2649	  nop
2650	wr	%g0, ASI_DMMU, %asi
2651	ldxa	[MMU_SFSR]%asi, %g5
2652	mov	1, %g6
2653	sllx	%g6, PN_SFSR_PARITY_SHIFT, %g6
2654	andcc	%g5, %g6, %g0
2655	bz	mmu_trap_tl1_4
2656
2657	/*
2658	 * We are running on a Panther and have hit a DTLB parity error.
2659	 */
2660	ldxa	[MMU_TAG_ACCESS]%asi, %g2
2661	mov	%g5, %g3
2662	ba,pt	%xcc, .mmu_exception_is_tlb_parity
2663	mov	T_DATA_EXCEPTION, %g1
2664
2665mmu_trap_tl1_4:
2666	CPU_PADDR(%g7, %g6);
2667	add     %g7, CPU_TL1_HDLR, %g7		! %g7 = &cpu_m.tl1_hdlr (PA)
2668	/*
2669	 * AM is cleared on trap, so addresses are 64 bit
2670	 */
2671	lda     [%g7]ASI_MEM, %g6
2672	brz,a,pt %g6, 1f
2673	  nop
2674	/*
2675	 * We are going to update cpu_m.tl1_hdlr using physical address.
2676	 * Flush the D$ line, so that stale data won't be accessed later.
2677	 */
2678	CPU_ADDR(%g6, %g5)
2679	add     %g6, CPU_TL1_HDLR, %g6		! %g6 = &cpu_m.tl1_hdlr (VA)
2680	GET_CPU_IMPL(%g5)
2681	cmp	%g5, CHEETAH_IMPL
2682	bl,pt	%icc, 3f
2683	 cmp	%g5, SPITFIRE_IMPL
2684	stxa	%g0, [%g7]ASI_DC_INVAL
2685	membar	#Sync
2686	ba,pt	%xcc, 2f
2687	 nop
26883:
2689	bl,pt	%icc, 2f
2690	 sethi	%hi(dcache_line_mask), %g5
2691	ld	[%g5 + %lo(dcache_line_mask)], %g5
2692	and	%g6, %g5, %g5
2693	stxa	%g0, [%g5]ASI_DC_TAG
2694	membar	#Sync
26952:
2696	sta     %g0, [%g7]ASI_MEM
2697	SWITCH_GLOBALS				! back to mmu globals
2698	ba,a,pt	%xcc, sfmmu_mmu_trap		! handle page faults
26991:
2700	rdpr	%tt, %g5
2701	rdpr	%tl, %g7
2702	sub	%g7, 1, %g6
2703	wrpr	%g6, %tl
2704	rdpr	%tt, %g6
2705	wrpr	%g7, %tl
2706	and	%g6, WTRAP_TTMASK, %g6
2707	cmp	%g6, WTRAP_TYPE
2708	bne,a,pn %xcc, ptl1_panic
2709	mov	PTL1_BAD_MMUTRAP, %g1
2710	rdpr	%tpc, %g7
2711	/* tpc should be in the trap table */
2712	set	trap_table, %g6
2713	cmp	%g7, %g6
2714	blt,a,pn %xcc, ptl1_panic
2715	  mov	PTL1_BAD_MMUTRAP, %g1
2716	set	etrap_table, %g6
2717	cmp	%g7, %g6
2718	bge,a,pn %xcc, ptl1_panic
2719	  mov	PTL1_BAD_MMUTRAP, %g1
2720	cmp	%g5, T_ALIGNMENT
2721	move	%icc, MMU_SFAR, %g6
2722	movne	%icc, MMU_TAG_ACCESS, %g6
2723	ldxa	[%g6]ASI_DMMU, %g6
2724	andn	%g7, WTRAP_ALIGN, %g7	/* 128 byte aligned */
2725	add	%g7, WTRAP_FAULTOFF, %g7
2726	wrpr	%g0, %g7, %tnpc
2727	done
2728	SET_SIZE(mmu_trap_tl1)
2729
2730/*
2731 * Several traps use kmdb_trap and kmdb_trap_tl1 as their handlers.  These
2732 * traps are valid only when kmdb is loaded.  When the debugger is active,
2733 * the code below is rewritten to transfer control to the appropriate
2734 * debugger entry points.
2735 */
2736	.global	kmdb_trap
2737	.align	8
2738kmdb_trap:
2739	ba,a	trap_table0
2740	jmp	%g1 + 0
2741	nop
2742
2743	.global	kmdb_trap_tl1
2744	.align	8
2745kmdb_trap_tl1:
2746	ba,a	trap_table0
2747	jmp	%g1 + 0
2748	nop
2749
2750/*
2751 * This entry is copied from OBP's trap table during boot.
2752 */
2753	.global	obp_bpt
2754	.align	8
2755obp_bpt:
2756	NOT
2757
2758/*
2759 * if kernel, set PCONTEXT to 0 for debuggers
2760 * if user, clear nucleus page sizes
2761 */
2762	.global kctx_obp_bpt
2763kctx_obp_bpt:
2764	set	obp_bpt, %g2
27651:
2766#ifndef _OPL
2767	mov	MMU_PCONTEXT, %g1
2768	ldxa	[%g1]ASI_DMMU, %g1
2769	srlx	%g1, CTXREG_NEXT_SHIFT, %g3
2770	brz,pt	%g3, 3f			! nucleus pgsz is 0, no problem
2771	  sllx	%g3, CTXREG_NEXT_SHIFT, %g3
2772	set	CTXREG_CTX_MASK, %g4	! check Pcontext
2773	btst	%g4, %g1
2774	bz,a,pt	%xcc, 2f
2775	  clr	%g3			! kernel:  PCONTEXT=0
2776	xor	%g3, %g1, %g3		! user:	clr N_pgsz0/1 bits
27772:
2778	set	DEMAP_ALL_TYPE, %g1
2779	stxa	%g0, [%g1]ASI_DTLB_DEMAP
2780	stxa	%g0, [%g1]ASI_ITLB_DEMAP
2781	mov	MMU_PCONTEXT, %g1
2782	stxa	%g3, [%g1]ASI_DMMU
2783        membar  #Sync
2784	sethi	%hi(FLUSH_ADDR), %g1
2785	flush	%g1			! flush required by immu
2786#endif /* _OPL */
27873:
2788	jmp	%g2
2789	  nop
2790
2791
2792#ifdef	TRAPTRACE
2793/*
2794 * TRAPTRACE support.
2795 * labels here are branched to with "rd %pc, %g7" in the delay slot.
2796 * Return is done by "jmp %g7 + 4".
2797 */
2798
2799trace_gen:
2800	TRACE_PTR(%g3, %g6)
2801	GET_TRACE_TICK(%g6, %g4)
2802	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi
2803	rdpr	%tl, %g6
2804	stha	%g6, [%g3 + TRAP_ENT_TL]%asi
2805	rdpr	%tt, %g6
2806	stha	%g6, [%g3 + TRAP_ENT_TT]%asi
2807	rdpr	%tstate, %g6
2808	stxa	%g6, [%g3 + TRAP_ENT_TSTATE]%asi
2809	stna	%sp, [%g3 + TRAP_ENT_SP]%asi
2810	rdpr	%tpc, %g6
2811	stna	%g6, [%g3 + TRAP_ENT_TPC]%asi
2812	TRACE_NEXT(%g3, %g4, %g5)
2813	jmp	%g7 + 4
2814	nop
2815
2816trace_win:
2817	TRACE_WIN_INFO(0, %l0, %l1, %l2)
2818	! Keep the locals as clean as possible, caller cleans %l4
2819	clr	%l2
2820	clr	%l1
2821	jmp	%l4 + 4
2822	  clr	%l0
2823
2824/*
2825 * Trace a tsb hit
2826 * g1 = tsbe pointer (in/clobbered)
2827 * g2 = tag access register (in)
2828 * g3 - g4 = scratch (clobbered)
2829 * g5 = tsbe data (in)
2830 * g6 = scratch (clobbered)
2831 * g7 = pc we jumped here from (in)
2832 */
2833
2834	! Do not disturb %g5, it will be used after the trace
2835	ALTENTRY(trace_tsbhit)
2836	TRACE_TSBHIT(0)
2837	jmp	%g7 + 4
2838	nop
2839
2840/*
2841 * Trace a TSB miss
2842 *
2843 * g1 = tsb8k pointer (in)
2844 * g2 = tag access register (in)
2845 * g3 = tsb4m pointer (in)
2846 * g4 = tsbe tag (in/clobbered)
2847 * g5 - g6 = scratch (clobbered)
2848 * g7 = pc we jumped here from (in)
2849 */
2850	.global	trace_tsbmiss
2851trace_tsbmiss:
2852	membar	#Sync
2853	sethi	%hi(FLUSH_ADDR), %g6
2854	flush	%g6
2855	TRACE_PTR(%g5, %g6)
2856	stxa	%g2, [%g5 + TRAP_ENT_SP]%asi		! tag access
2857	stxa	%g4, [%g5 + TRAP_ENT_F1]%asi		! tsb tag
2858	GET_TRACE_TICK(%g6, %g4)
2859	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2860	rdpr	%tnpc, %g6
2861	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2862	stna	%g1, [%g5 + TRAP_ENT_F3]%asi		! tsb8k pointer
2863	srlx	%g1, 32, %g6
2864	stna	%g6, [%g5 + TRAP_ENT_F4]%asi		! huh?
2865	rdpr	%tpc, %g6
2866	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2867	rdpr	%tl, %g6
2868	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2869	rdpr	%tt, %g6
2870	or	%g6, TT_MMU_MISS, %g4
2871	stha	%g4, [%g5 + TRAP_ENT_TT]%asi
2872	cmp	%g6, FAST_IMMU_MISS_TT
2873	be,a	%icc, 1f
2874	  ldxa	[%g0]ASI_IMMU, %g6
2875	ldxa	[%g0]ASI_DMMU, %g6
28761:	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi	! tag target
2877	stxa	%g3, [%g5 + TRAP_ENT_TR]%asi		! tsb4m pointer
2878	TRACE_NEXT(%g5, %g4, %g6)
2879	jmp	%g7 + 4
2880	nop
2881
2882/*
2883 * g2 = tag access register (in)
2884 * g3 = ctx number (in)
2885 */
2886trace_dataprot:
2887	membar	#Sync
2888	sethi	%hi(FLUSH_ADDR), %g6
2889	flush	%g6
2890	TRACE_PTR(%g1, %g6)
2891	GET_TRACE_TICK(%g6, %g5)
2892	stxa	%g6, [%g1 + TRAP_ENT_TICK]%asi
2893	rdpr	%tpc, %g6
2894	stna	%g6, [%g1 + TRAP_ENT_TPC]%asi
2895	rdpr	%tstate, %g6
2896	stxa	%g6, [%g1 + TRAP_ENT_TSTATE]%asi
2897	stxa	%g2, [%g1 + TRAP_ENT_SP]%asi		! tag access reg
2898	stxa	%g0, [%g1 + TRAP_ENT_TR]%asi
2899	stxa	%g0, [%g1 + TRAP_ENT_F1]%asi
2900	stxa	%g0, [%g1 + TRAP_ENT_F2]%asi
2901	stxa	%g0, [%g1 + TRAP_ENT_F3]%asi
2902	stxa	%g0, [%g1 + TRAP_ENT_F4]%asi
2903	rdpr	%tl, %g6
2904	stha	%g6, [%g1 + TRAP_ENT_TL]%asi
2905	rdpr	%tt, %g6
2906	stha	%g6, [%g1 + TRAP_ENT_TT]%asi
2907	TRACE_NEXT(%g1, %g4, %g5)
2908	jmp	%g7 + 4
2909	nop
2910
2911#endif /* TRAPTRACE */
2912
2913       .align  32
2914       .global pil15_epilogue
2915pil15_epilogue:
2916       ba      pil_interrupt_common
2917       nop
2918       .align  32
2919
2920/*
2921 * fast_trap_done, fast_trap_done_chk_intr:
2922 *
2923 * Due to the design of UltraSPARC pipeline, pending interrupts are not
2924 * taken immediately after a RETRY or DONE instruction which causes IE to
2925 * go from 0 to 1. Instead, the instruction at %tpc or %tnpc is allowed
2926 * to execute first before taking any interrupts. If that instruction
2927 * results in other traps, and if the corresponding trap handler runs
2928 * entirely at TL=1 with interrupts disabled, then pending interrupts
2929 * won't be taken until after yet another instruction following the %tpc
2930 * or %tnpc.
2931 *
2932 * A malicious user program can use this feature to block out interrupts
2933 * for extended durations, which can result in send_mondo_timeout kernel
2934 * panic.
2935 *
2936 * This problem is addressed by servicing any pending interrupts via
2937 * sys_trap before returning back to the user mode from a fast trap
2938 * handler. The "done" instruction within a fast trap handler, which
2939 * runs entirely at TL=1 with interrupts disabled, is replaced with the
2940 * FAST_TRAP_DONE macro, which branches control to this fast_trap_done
2941 * entry point.
2942 *
2943 * We check for any pending interrupts here and force a sys_trap to
2944 * service those interrupts, if any. To minimize overhead, pending
2945 * interrupts are checked if the %tpc happens to be at 16K boundary,
2946 * which allows a malicious program to execute at most 4K consecutive
2947 * instructions before we service any pending interrupts. If a worst
2948 * case fast trap handler takes about 2 usec, then interrupts will be
2949 * blocked for at most 8 msec, less than a clock tick.
2950 *
2951 * For the cases where we don't know if the %tpc will cross a 16K
2952 * boundary, we can't use the above optimization and always process
2953 * any pending interrupts via fast_frap_done_chk_intr entry point.
2954 *
2955 * Entry Conditions:
2956 * 	%pstate		am:0 priv:1 ie:0
2957 * 			globals are AG (not normal globals)
2958 */
2959
2960	.global	fast_trap_done, fast_trap_done_chk_intr
2961fast_trap_done:
2962	rdpr	%tpc, %g5
2963	sethi	%hi(0xffffc000), %g6	! 1's complement of 0x3fff
2964	andncc	%g5, %g6, %g0		! check lower 14 bits of %tpc
2965	bz,a,pn	%icc, 1f		! branch if zero (lower 32 bits only)
2966	  ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2967	done
2968
2969	ALTENTRY(fast_trap_done_check_interrupts)
2970fast_trap_done_chk_intr:
2971	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2972
29731:	rd	SOFTINT, %g6
2974	and	%g5, IRSR_BUSY, %g5
2975	orcc	%g5, %g6, %g0
2976	bnz,pn	%xcc, 2f		! branch if any pending intr
2977	nop
2978	done
2979
29802:
2981	/*
2982	 * We get here if there are any pending interrupts.
2983	 * Adjust %tpc/%tnpc as we'll be resuming via "retry"
2984	 * instruction.
2985	 */
2986	rdpr	%tnpc, %g5
2987	wrpr	%g0, %g5, %tpc
2988	add	%g5, 4, %g5
2989	wrpr	%g0, %g5, %tnpc
2990
2991	/*
2992	 * Force a dummy sys_trap call so that interrupts can be serviced.
2993	 */
2994	set	fast_trap_dummy_call, %g1
2995	ba,pt	%xcc, sys_trap
2996	  mov	-1, %g4
2997
2998fast_trap_dummy_call:
2999	retl
3000	nop
3001
3002/*
3003 * Currently the brand syscall interposition code is not enabled by
3004 * default.  Instead, when a branded zone is first booted the brand
3005 * infrastructure will patch the trap table so that the syscall
3006 * entry points are redirected to syscall_wrapper32 and syscall_wrapper
3007 * for ILP32 and LP64 syscalls respectively.  this is done in
3008 * brand_plat_interposition_enable().  Note that the syscall wrappers
3009 * below do not collect any trap trace data since the syscall hot patch
3010 * points are reached after trap trace data has already been collected.
3011 */
3012#define	BRAND_CALLBACK(callback_id)					    \
3013	CPU_ADDR(%g2, %g1)		/* load CPU struct addr to %g2	*/ ;\
3014	ldn	[%g2 + CPU_THREAD], %g3	/* load thread pointer		*/ ;\
3015	ldn	[%g3 + T_PROCP], %g3	/* get proc pointer		*/ ;\
3016	ldn	[%g3 + P_BRAND], %g3	/* get brand pointer		*/ ;\
3017	brz	%g3, 1f			/* No brand?  No callback. 	*/ ;\
3018	nop 								   ;\
3019	ldn	[%g3 + B_MACHOPS], %g3	/* get machops list		*/ ;\
3020	ldn	[%g3 + (callback_id << 3)], %g3 			   ;\
3021	brz	%g3, 1f							   ;\
3022	/*								    \
3023	 * This isn't pretty.  We want a low-latency way for the callback   \
3024	 * routine to decline to do anything.  We just pass in an address   \
3025	 * the routine can directly jmp back to, pretending that nothing    \
3026	 * has happened.						    \
3027	 * 								    \
3028	 * %g1: return address (where the brand handler jumps back to)	    \
3029	 * %g2: address of CPU structure				    \
3030	 * %g3: address of brand handler (where we will jump to)	    \
3031	 */								    \
3032	mov	%pc, %g1						   ;\
3033	add	%g1, 16, %g1						   ;\
3034	jmp	%g3							   ;\
3035	nop								   ;\
30361:
3037
3038	ENTRY_NP(syscall_wrapper32)
3039	BRAND_CALLBACK(BRAND_CB_SYSCALL32)
3040	SYSCALL_NOTT(syscall_trap32)
3041	SET_SIZE(syscall_wrapper32)
3042
3043	ENTRY_NP(syscall_wrapper)
3044	BRAND_CALLBACK(BRAND_CB_SYSCALL)
3045	SYSCALL_NOTT(syscall_trap)
3046	SET_SIZE(syscall_wrapper)
3047