xref: /titanic_51/usr/src/uts/sun4u/ml/trap_table.s (revision 8a40a695ee676a322b094e9afe5375567bfb51e3)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#if !defined(lint)
29#include "assym.h"
30#endif /* !lint */
31#include <sys/asm_linkage.h>
32#include <sys/privregs.h>
33#include <sys/sun4asi.h>
34#include <sys/spitregs.h>
35#include <sys/cheetahregs.h>
36#include <sys/machtrap.h>
37#include <sys/machthread.h>
38#include <sys/machbrand.h>
39#include <sys/pcb.h>
40#include <sys/pte.h>
41#include <sys/mmu.h>
42#include <sys/machpcb.h>
43#include <sys/async.h>
44#include <sys/intreg.h>
45#include <sys/scb.h>
46#include <sys/psr_compat.h>
47#include <sys/syscall.h>
48#include <sys/machparam.h>
49#include <sys/traptrace.h>
50#include <vm/hat_sfmmu.h>
51#include <sys/archsystm.h>
52#include <sys/utrap.h>
53#include <sys/clock.h>
54#include <sys/intr.h>
55#include <sys/fpu/fpu_simulator.h>
56#include <vm/seg_spt.h>
57
58/*
59 * WARNING: If you add a fast trap handler which can be invoked by a
60 * non-privileged user, you may have to use the FAST_TRAP_DONE macro
61 * instead of "done" instruction to return back to the user mode. See
62 * comments for the "fast_trap_done" entry point for more information.
63 *
64 * An alternate FAST_TRAP_DONE_CHK_INTR macro should be used for the
65 * cases where you always want to process any pending interrupts before
66 * returning back to the user mode.
67 */
68#define	FAST_TRAP_DONE		\
69	ba,a	fast_trap_done
70
71#define	FAST_TRAP_DONE_CHK_INTR	\
72	ba,a	fast_trap_done_chk_intr
73
74/*
75 * SPARC V9 Trap Table
76 *
77 * Most of the trap handlers are made from common building
78 * blocks, and some are instantiated multiple times within
79 * the trap table. So, I build a bunch of macros, then
80 * populate the table using only the macros.
81 *
82 * Many macros branch to sys_trap.  Its calling convention is:
83 *	%g1		kernel trap handler
84 *	%g2, %g3	args for above
85 *	%g4		desire %pil
86 */
87
88#ifdef	TRAPTRACE
89
90/*
91 * Tracing macro. Adds two instructions if TRAPTRACE is defined.
92 */
93#define	TT_TRACE(label)		\
94	ba	label		;\
95	rd	%pc, %g7
96#define	TT_TRACE_INS	2
97
98#define	TT_TRACE_L(label)	\
99	ba	label		;\
100	rd	%pc, %l4	;\
101	clr	%l4
102#define	TT_TRACE_L_INS	3
103
104#else
105
106#define	TT_TRACE(label)
107#define	TT_TRACE_INS	0
108
109#define	TT_TRACE_L(label)
110#define	TT_TRACE_L_INS	0
111
112#endif
113
114/*
115 * This macro is used to update per cpu mmu stats in perf critical
116 * paths. It is only enabled in debug kernels or if SFMMU_STAT_GATHER
117 * is defined.
118 */
119#if defined(DEBUG) || defined(SFMMU_STAT_GATHER)
120#define	HAT_PERCPU_DBSTAT(stat)			\
121	mov	stat, %g1			;\
122	ba	stat_mmu			;\
123	rd	%pc, %g7
124#else
125#define	HAT_PERCPU_DBSTAT(stat)
126#endif /* DEBUG || SFMMU_STAT_GATHER */
127
128/*
129 * This first set are funneled to trap() with %tt as the type.
130 * Trap will then either panic or send the user a signal.
131 */
132/*
133 * NOT is used for traps that just shouldn't happen.
134 * It comes in both single and quadruple flavors.
135 */
136#if !defined(lint)
137	.global	trap
138#endif /* !lint */
139#define	NOT			\
140	TT_TRACE(trace_gen)	;\
141	set	trap, %g1	;\
142	rdpr	%tt, %g3	;\
143	ba,pt	%xcc, sys_trap	;\
144	sub	%g0, 1, %g4	;\
145	.align	32
146#define	NOT4	NOT; NOT; NOT; NOT
147/*
148 * RED is for traps that use the red mode handler.
149 * We should never see these either.
150 */
151#define	RED	NOT
152/*
153 * BAD is used for trap vectors we don't have a kernel
154 * handler for.
155 * It also comes in single and quadruple versions.
156 */
157#define	BAD	NOT
158#define	BAD4	NOT4
159
160#define	DONE			\
161	done;			\
162	.align	32
163
164/*
165 * TRAP vectors to the trap() function.
166 * It's main use is for user errors.
167 */
168#if !defined(lint)
169	.global	trap
170#endif /* !lint */
171#define	TRAP(arg)		\
172	TT_TRACE(trace_gen)	;\
173	set	trap, %g1	;\
174	mov	arg, %g3	;\
175	ba,pt	%xcc, sys_trap	;\
176	sub	%g0, 1, %g4	;\
177	.align	32
178
179/*
180 * SYSCALL is used for system calls on both ILP32 and LP64 kernels
181 * depending on the "which" parameter (should be syscall_trap,
182 * syscall_trap32, or nosys for unused system call traps).
183 */
184#define	SYSCALL(which)			\
185	TT_TRACE(trace_gen)		;\
186	set	(which), %g1		;\
187	ba,pt	%xcc, sys_trap		;\
188	sub	%g0, 1, %g4		;\
189	.align	32
190
191#define	FLUSHW()			\
192	set	trap, %g1		;\
193	mov	T_FLUSHW, %g3		;\
194	sub	%g0, 1, %g4		;\
195	save				;\
196	flushw				;\
197	restore				;\
198	FAST_TRAP_DONE			;\
199	.align	32
200
201/*
202 * GOTO just jumps to a label.
203 * It's used for things that can be fixed without going thru sys_trap.
204 */
205#define	GOTO(label)		\
206	.global	label		;\
207	ba,a	label		;\
208	.empty			;\
209	.align	32
210
211/*
212 * GOTO_TT just jumps to a label.
213 * correctable ECC error traps at  level 0 and 1 will use this macro.
214 * It's used for things that can be fixed without going thru sys_trap.
215 */
216#define	GOTO_TT(label, ttlabel)		\
217	.global	label		;\
218	TT_TRACE(ttlabel)	;\
219	ba,a	label		;\
220	.empty			;\
221	.align	32
222
223/*
224 * Privileged traps
225 * Takes breakpoint if privileged, calls trap() if not.
226 */
227#define	PRIV(label)			\
228	rdpr	%tstate, %g1		;\
229	btst	TSTATE_PRIV, %g1	;\
230	bnz	label			;\
231	rdpr	%tt, %g3		;\
232	set	trap, %g1		;\
233	ba,pt	%xcc, sys_trap		;\
234	sub	%g0, 1, %g4		;\
235	.align	32
236
237
238/*
239 * DTrace traps.
240 */
241#define	DTRACE_PID			\
242	.global dtrace_pid_probe				;\
243	set	dtrace_pid_probe, %g1				;\
244	ba,pt	%xcc, user_trap					;\
245	sub	%g0, 1, %g4					;\
246	.align	32
247
248#define	DTRACE_RETURN			\
249	.global dtrace_return_probe				;\
250	set	dtrace_return_probe, %g1			;\
251	ba,pt	%xcc, user_trap					;\
252	sub	%g0, 1, %g4					;\
253	.align	32
254
255/*
256 * REGISTER WINDOW MANAGEMENT MACROS
257 */
258
259/*
260 * various convenient units of padding
261 */
262#define	SKIP(n)	.skip 4*(n)
263
264/*
265 * CLEAN_WINDOW is the simple handler for cleaning a register window.
266 */
267#define	CLEAN_WINDOW						\
268	TT_TRACE_L(trace_win)					;\
269	rdpr %cleanwin, %l0; inc %l0; wrpr %l0, %cleanwin	;\
270	clr %l0; clr %l1; clr %l2; clr %l3			;\
271	clr %l4; clr %l5; clr %l6; clr %l7			;\
272	clr %o0; clr %o1; clr %o2; clr %o3			;\
273	clr %o4; clr %o5; clr %o6; clr %o7			;\
274	retry; .align 128
275
276#if !defined(lint)
277
278/*
279 * If we get an unresolved tlb miss while in a window handler, the fault
280 * handler will resume execution at the last instruction of the window
281 * hander, instead of delivering the fault to the kernel.  Spill handlers
282 * use this to spill windows into the wbuf.
283 *
284 * The mixed handler works by checking %sp, and branching to the correct
285 * handler.  This is done by branching back to label 1: for 32b frames,
286 * or label 2: for 64b frames; which implies the handler order is: 32b,
287 * 64b, mixed.  The 1: and 2: labels are offset into the routines to
288 * allow the branchs' delay slots to contain useful instructions.
289 */
290
291/*
292 * SPILL_32bit spills a 32-bit-wide kernel register window.  It
293 * assumes that the kernel context and the nucleus context are the
294 * same.  The stack pointer is required to be eight-byte aligned even
295 * though this code only needs it to be four-byte aligned.
296 */
297#define	SPILL_32bit(tail)					\
298	srl	%sp, 0, %sp					;\
2991:	st	%l0, [%sp + 0]					;\
300	st	%l1, [%sp + 4]					;\
301	st	%l2, [%sp + 8]					;\
302	st	%l3, [%sp + 12]					;\
303	st	%l4, [%sp + 16]					;\
304	st	%l5, [%sp + 20]					;\
305	st	%l6, [%sp + 24]					;\
306	st	%l7, [%sp + 28]					;\
307	st	%i0, [%sp + 32]					;\
308	st	%i1, [%sp + 36]					;\
309	st	%i2, [%sp + 40]					;\
310	st	%i3, [%sp + 44]					;\
311	st	%i4, [%sp + 48]					;\
312	st	%i5, [%sp + 52]					;\
313	st	%i6, [%sp + 56]					;\
314	st	%i7, [%sp + 60]					;\
315	TT_TRACE_L(trace_win)					;\
316	saved							;\
317	retry							;\
318	SKIP(31-19-TT_TRACE_L_INS)				;\
319	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
320	.empty
321
322/*
323 * SPILL_32bit_asi spills a 32-bit-wide register window into a 32-bit
324 * wide address space via the designated asi.  It is used to spill
325 * non-kernel windows.  The stack pointer is required to be eight-byte
326 * aligned even though this code only needs it to be four-byte
327 * aligned.
328 */
329#define	SPILL_32bit_asi(asi_num, tail)				\
330	srl	%sp, 0, %sp					;\
3311:	sta	%l0, [%sp + %g0]asi_num				;\
332	mov	4, %g1						;\
333	sta	%l1, [%sp + %g1]asi_num				;\
334	mov	8, %g2						;\
335	sta	%l2, [%sp + %g2]asi_num				;\
336	mov	12, %g3						;\
337	sta	%l3, [%sp + %g3]asi_num				;\
338	add	%sp, 16, %g4					;\
339	sta	%l4, [%g4 + %g0]asi_num				;\
340	sta	%l5, [%g4 + %g1]asi_num				;\
341	sta	%l6, [%g4 + %g2]asi_num				;\
342	sta	%l7, [%g4 + %g3]asi_num				;\
343	add	%g4, 16, %g4					;\
344	sta	%i0, [%g4 + %g0]asi_num				;\
345	sta	%i1, [%g4 + %g1]asi_num				;\
346	sta	%i2, [%g4 + %g2]asi_num				;\
347	sta	%i3, [%g4 + %g3]asi_num				;\
348	add	%g4, 16, %g4					;\
349	sta	%i4, [%g4 + %g0]asi_num				;\
350	sta	%i5, [%g4 + %g1]asi_num				;\
351	sta	%i6, [%g4 + %g2]asi_num				;\
352	sta	%i7, [%g4 + %g3]asi_num				;\
353	TT_TRACE_L(trace_win)					;\
354	saved							;\
355	retry							;\
356	SKIP(31-25-TT_TRACE_L_INS)				;\
357	ba,a,pt %xcc, fault_32bit_/**/tail			;\
358	.empty
359
360/*
361 * SPILL_32bit_tt1 spills a 32-bit-wide register window into a 32-bit
362 * wide address space via the designated asi.  It is used to spill
363 * windows at tl>1 where performance isn't the primary concern and
364 * where we don't want to use unnecessary registers.  The stack
365 * pointer is required to be eight-byte aligned even though this code
366 * only needs it to be four-byte aligned.
367 */
368#define	SPILL_32bit_tt1(asi_num, tail)				\
369	mov	asi_num, %asi					;\
3701:	srl	%sp, 0, %sp					;\
371	sta	%l0, [%sp + 0]%asi				;\
372	sta	%l1, [%sp + 4]%asi				;\
373	sta	%l2, [%sp + 8]%asi				;\
374	sta	%l3, [%sp + 12]%asi				;\
375	sta	%l4, [%sp + 16]%asi				;\
376	sta	%l5, [%sp + 20]%asi				;\
377	sta	%l6, [%sp + 24]%asi				;\
378	sta	%l7, [%sp + 28]%asi				;\
379	sta	%i0, [%sp + 32]%asi				;\
380	sta	%i1, [%sp + 36]%asi				;\
381	sta	%i2, [%sp + 40]%asi				;\
382	sta	%i3, [%sp + 44]%asi				;\
383	sta	%i4, [%sp + 48]%asi				;\
384	sta	%i5, [%sp + 52]%asi				;\
385	sta	%i6, [%sp + 56]%asi				;\
386	sta	%i7, [%sp + 60]%asi				;\
387	TT_TRACE_L(trace_win)					;\
388	saved							;\
389	retry							;\
390	SKIP(31-20-TT_TRACE_L_INS)				;\
391	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
392	.empty
393
394
395/*
396 * FILL_32bit fills a 32-bit-wide kernel register window.  It assumes
397 * that the kernel context and the nucleus context are the same.  The
398 * stack pointer is required to be eight-byte aligned even though this
399 * code only needs it to be four-byte aligned.
400 */
401#define	FILL_32bit(tail)					\
402	srl	%sp, 0, %sp					;\
4031:	TT_TRACE_L(trace_win)					;\
404	ld	[%sp + 0], %l0					;\
405	ld	[%sp + 4], %l1					;\
406	ld	[%sp + 8], %l2					;\
407	ld	[%sp + 12], %l3					;\
408	ld	[%sp + 16], %l4					;\
409	ld	[%sp + 20], %l5					;\
410	ld	[%sp + 24], %l6					;\
411	ld	[%sp + 28], %l7					;\
412	ld	[%sp + 32], %i0					;\
413	ld	[%sp + 36], %i1					;\
414	ld	[%sp + 40], %i2					;\
415	ld	[%sp + 44], %i3					;\
416	ld	[%sp + 48], %i4					;\
417	ld	[%sp + 52], %i5					;\
418	ld	[%sp + 56], %i6					;\
419	ld	[%sp + 60], %i7					;\
420	restored						;\
421	retry							;\
422	SKIP(31-19-TT_TRACE_L_INS)				;\
423	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
424	.empty
425
426/*
427 * FILL_32bit_asi fills a 32-bit-wide register window from a 32-bit
428 * wide address space via the designated asi.  It is used to fill
429 * non-kernel windows.  The stack pointer is required to be eight-byte
430 * aligned even though this code only needs it to be four-byte
431 * aligned.
432 */
433#define	FILL_32bit_asi(asi_num, tail)				\
434	srl	%sp, 0, %sp					;\
4351:	TT_TRACE_L(trace_win)					;\
436	mov	4, %g1						;\
437	lda	[%sp + %g0]asi_num, %l0				;\
438	mov	8, %g2						;\
439	lda	[%sp + %g1]asi_num, %l1				;\
440	mov	12, %g3						;\
441	lda	[%sp + %g2]asi_num, %l2				;\
442	lda	[%sp + %g3]asi_num, %l3				;\
443	add	%sp, 16, %g4					;\
444	lda	[%g4 + %g0]asi_num, %l4				;\
445	lda	[%g4 + %g1]asi_num, %l5				;\
446	lda	[%g4 + %g2]asi_num, %l6				;\
447	lda	[%g4 + %g3]asi_num, %l7				;\
448	add	%g4, 16, %g4					;\
449	lda	[%g4 + %g0]asi_num, %i0				;\
450	lda	[%g4 + %g1]asi_num, %i1				;\
451	lda	[%g4 + %g2]asi_num, %i2				;\
452	lda	[%g4 + %g3]asi_num, %i3				;\
453	add	%g4, 16, %g4					;\
454	lda	[%g4 + %g0]asi_num, %i4				;\
455	lda	[%g4 + %g1]asi_num, %i5				;\
456	lda	[%g4 + %g2]asi_num, %i6				;\
457	lda	[%g4 + %g3]asi_num, %i7				;\
458	restored						;\
459	retry							;\
460	SKIP(31-25-TT_TRACE_L_INS)				;\
461	ba,a,pt %xcc, fault_32bit_/**/tail			;\
462	.empty
463
464/*
465 * FILL_32bit_tt1 fills a 32-bit-wide register window from a 32-bit
466 * wide address space via the designated asi.  It is used to fill
467 * windows at tl>1 where performance isn't the primary concern and
468 * where we don't want to use unnecessary registers.  The stack
469 * pointer is required to be eight-byte aligned even though this code
470 * only needs it to be four-byte aligned.
471 */
472#define	FILL_32bit_tt1(asi_num, tail)				\
473	mov	asi_num, %asi					;\
4741:	srl	%sp, 0, %sp					;\
475	TT_TRACE_L(trace_win)					;\
476	lda	[%sp + 0]%asi, %l0				;\
477	lda	[%sp + 4]%asi, %l1				;\
478	lda	[%sp + 8]%asi, %l2				;\
479	lda	[%sp + 12]%asi, %l3				;\
480	lda	[%sp + 16]%asi, %l4				;\
481	lda	[%sp + 20]%asi, %l5				;\
482	lda	[%sp + 24]%asi, %l6				;\
483	lda	[%sp + 28]%asi, %l7				;\
484	lda	[%sp + 32]%asi, %i0				;\
485	lda	[%sp + 36]%asi, %i1				;\
486	lda	[%sp + 40]%asi, %i2				;\
487	lda	[%sp + 44]%asi, %i3				;\
488	lda	[%sp + 48]%asi, %i4				;\
489	lda	[%sp + 52]%asi, %i5				;\
490	lda	[%sp + 56]%asi, %i6				;\
491	lda	[%sp + 60]%asi, %i7				;\
492	restored						;\
493	retry							;\
494	SKIP(31-20-TT_TRACE_L_INS)				;\
495	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
496	.empty
497
498
499/*
500 * SPILL_64bit spills a 64-bit-wide kernel register window.  It
501 * assumes that the kernel context and the nucleus context are the
502 * same.  The stack pointer is required to be eight-byte aligned.
503 */
504#define	SPILL_64bit(tail)					\
5052:	stx	%l0, [%sp + V9BIAS64 + 0]			;\
506	stx	%l1, [%sp + V9BIAS64 + 8]			;\
507	stx	%l2, [%sp + V9BIAS64 + 16]			;\
508	stx	%l3, [%sp + V9BIAS64 + 24]			;\
509	stx	%l4, [%sp + V9BIAS64 + 32]			;\
510	stx	%l5, [%sp + V9BIAS64 + 40]			;\
511	stx	%l6, [%sp + V9BIAS64 + 48]			;\
512	stx	%l7, [%sp + V9BIAS64 + 56]			;\
513	stx	%i0, [%sp + V9BIAS64 + 64]			;\
514	stx	%i1, [%sp + V9BIAS64 + 72]			;\
515	stx	%i2, [%sp + V9BIAS64 + 80]			;\
516	stx	%i3, [%sp + V9BIAS64 + 88]			;\
517	stx	%i4, [%sp + V9BIAS64 + 96]			;\
518	stx	%i5, [%sp + V9BIAS64 + 104]			;\
519	stx	%i6, [%sp + V9BIAS64 + 112]			;\
520	stx	%i7, [%sp + V9BIAS64 + 120]			;\
521	TT_TRACE_L(trace_win)					;\
522	saved							;\
523	retry							;\
524	SKIP(31-18-TT_TRACE_L_INS)				;\
525	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
526	.empty
527
528/*
529 * SPILL_64bit_asi spills a 64-bit-wide register window into a 64-bit
530 * wide address space via the designated asi.  It is used to spill
531 * non-kernel windows.  The stack pointer is required to be eight-byte
532 * aligned.
533 */
534#define	SPILL_64bit_asi(asi_num, tail)				\
535	mov	0 + V9BIAS64, %g1				;\
5362:	stxa	%l0, [%sp + %g1]asi_num				;\
537	mov	8 + V9BIAS64, %g2				;\
538	stxa	%l1, [%sp + %g2]asi_num				;\
539	mov	16 + V9BIAS64, %g3				;\
540	stxa	%l2, [%sp + %g3]asi_num				;\
541	mov	24 + V9BIAS64, %g4				;\
542	stxa	%l3, [%sp + %g4]asi_num				;\
543	add	%sp, 32, %g5					;\
544	stxa	%l4, [%g5 + %g1]asi_num				;\
545	stxa	%l5, [%g5 + %g2]asi_num				;\
546	stxa	%l6, [%g5 + %g3]asi_num				;\
547	stxa	%l7, [%g5 + %g4]asi_num				;\
548	add	%g5, 32, %g5					;\
549	stxa	%i0, [%g5 + %g1]asi_num				;\
550	stxa	%i1, [%g5 + %g2]asi_num				;\
551	stxa	%i2, [%g5 + %g3]asi_num				;\
552	stxa	%i3, [%g5 + %g4]asi_num				;\
553	add	%g5, 32, %g5					;\
554	stxa	%i4, [%g5 + %g1]asi_num				;\
555	stxa	%i5, [%g5 + %g2]asi_num				;\
556	stxa	%i6, [%g5 + %g3]asi_num				;\
557	stxa	%i7, [%g5 + %g4]asi_num				;\
558	TT_TRACE_L(trace_win)					;\
559	saved							;\
560	retry							;\
561	SKIP(31-25-TT_TRACE_L_INS)				;\
562	ba,a,pt %xcc, fault_64bit_/**/tail			;\
563	.empty
564
565/*
566 * SPILL_64bit_tt1 spills a 64-bit-wide register window into a 64-bit
567 * wide address space via the designated asi.  It is used to spill
568 * windows at tl>1 where performance isn't the primary concern and
569 * where we don't want to use unnecessary registers.  The stack
570 * pointer is required to be eight-byte aligned.
571 */
572#define	SPILL_64bit_tt1(asi_num, tail)				\
573	mov	asi_num, %asi					;\
5742:	stxa	%l0, [%sp + V9BIAS64 + 0]%asi			;\
575	stxa	%l1, [%sp + V9BIAS64 + 8]%asi			;\
576	stxa	%l2, [%sp + V9BIAS64 + 16]%asi			;\
577	stxa	%l3, [%sp + V9BIAS64 + 24]%asi			;\
578	stxa	%l4, [%sp + V9BIAS64 + 32]%asi			;\
579	stxa	%l5, [%sp + V9BIAS64 + 40]%asi			;\
580	stxa	%l6, [%sp + V9BIAS64 + 48]%asi			;\
581	stxa	%l7, [%sp + V9BIAS64 + 56]%asi			;\
582	stxa	%i0, [%sp + V9BIAS64 + 64]%asi			;\
583	stxa	%i1, [%sp + V9BIAS64 + 72]%asi			;\
584	stxa	%i2, [%sp + V9BIAS64 + 80]%asi			;\
585	stxa	%i3, [%sp + V9BIAS64 + 88]%asi			;\
586	stxa	%i4, [%sp + V9BIAS64 + 96]%asi			;\
587	stxa	%i5, [%sp + V9BIAS64 + 104]%asi			;\
588	stxa	%i6, [%sp + V9BIAS64 + 112]%asi			;\
589	stxa	%i7, [%sp + V9BIAS64 + 120]%asi			;\
590	TT_TRACE_L(trace_win)					;\
591	saved							;\
592	retry							;\
593	SKIP(31-19-TT_TRACE_L_INS)				;\
594	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
595	.empty
596
597
598/*
599 * FILL_64bit fills a 64-bit-wide kernel register window.  It assumes
600 * that the kernel context and the nucleus context are the same.  The
601 * stack pointer is required to be eight-byte aligned.
602 */
603#define	FILL_64bit(tail)					\
6042:	TT_TRACE_L(trace_win)					;\
605	ldx	[%sp + V9BIAS64 + 0], %l0			;\
606	ldx	[%sp + V9BIAS64 + 8], %l1			;\
607	ldx	[%sp + V9BIAS64 + 16], %l2			;\
608	ldx	[%sp + V9BIAS64 + 24], %l3			;\
609	ldx	[%sp + V9BIAS64 + 32], %l4			;\
610	ldx	[%sp + V9BIAS64 + 40], %l5			;\
611	ldx	[%sp + V9BIAS64 + 48], %l6			;\
612	ldx	[%sp + V9BIAS64 + 56], %l7			;\
613	ldx	[%sp + V9BIAS64 + 64], %i0			;\
614	ldx	[%sp + V9BIAS64 + 72], %i1			;\
615	ldx	[%sp + V9BIAS64 + 80], %i2			;\
616	ldx	[%sp + V9BIAS64 + 88], %i3			;\
617	ldx	[%sp + V9BIAS64 + 96], %i4			;\
618	ldx	[%sp + V9BIAS64 + 104], %i5			;\
619	ldx	[%sp + V9BIAS64 + 112], %i6			;\
620	ldx	[%sp + V9BIAS64 + 120], %i7			;\
621	restored						;\
622	retry							;\
623	SKIP(31-18-TT_TRACE_L_INS)				;\
624	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
625	.empty
626
627/*
628 * FILL_64bit_asi fills a 64-bit-wide register window from a 64-bit
629 * wide address space via the designated asi.  It is used to fill
630 * non-kernel windows.  The stack pointer is required to be eight-byte
631 * aligned.
632 */
633#define	FILL_64bit_asi(asi_num, tail)				\
634	mov	V9BIAS64 + 0, %g1				;\
6352:	TT_TRACE_L(trace_win)					;\
636	ldxa	[%sp + %g1]asi_num, %l0				;\
637	mov	V9BIAS64 + 8, %g2				;\
638	ldxa	[%sp + %g2]asi_num, %l1				;\
639	mov	V9BIAS64 + 16, %g3				;\
640	ldxa	[%sp + %g3]asi_num, %l2				;\
641	mov	V9BIAS64 + 24, %g4				;\
642	ldxa	[%sp + %g4]asi_num, %l3				;\
643	add	%sp, 32, %g5					;\
644	ldxa	[%g5 + %g1]asi_num, %l4				;\
645	ldxa	[%g5 + %g2]asi_num, %l5				;\
646	ldxa	[%g5 + %g3]asi_num, %l6				;\
647	ldxa	[%g5 + %g4]asi_num, %l7				;\
648	add	%g5, 32, %g5					;\
649	ldxa	[%g5 + %g1]asi_num, %i0				;\
650	ldxa	[%g5 + %g2]asi_num, %i1				;\
651	ldxa	[%g5 + %g3]asi_num, %i2				;\
652	ldxa	[%g5 + %g4]asi_num, %i3				;\
653	add	%g5, 32, %g5					;\
654	ldxa	[%g5 + %g1]asi_num, %i4				;\
655	ldxa	[%g5 + %g2]asi_num, %i5				;\
656	ldxa	[%g5 + %g3]asi_num, %i6				;\
657	ldxa	[%g5 + %g4]asi_num, %i7				;\
658	restored						;\
659	retry							;\
660	SKIP(31-25-TT_TRACE_L_INS)				;\
661	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
662	.empty
663
664/*
665 * FILL_64bit_tt1 fills a 64-bit-wide register window from a 64-bit
666 * wide address space via the designated asi.  It is used to fill
667 * windows at tl>1 where performance isn't the primary concern and
668 * where we don't want to use unnecessary registers.  The stack
669 * pointer is required to be eight-byte aligned.
670 */
671#define	FILL_64bit_tt1(asi_num, tail)				\
672	mov	asi_num, %asi					;\
673	TT_TRACE_L(trace_win)					;\
674	ldxa	[%sp + V9BIAS64 + 0]%asi, %l0			;\
675	ldxa	[%sp + V9BIAS64 + 8]%asi, %l1			;\
676	ldxa	[%sp + V9BIAS64 + 16]%asi, %l2			;\
677	ldxa	[%sp + V9BIAS64 + 24]%asi, %l3			;\
678	ldxa	[%sp + V9BIAS64 + 32]%asi, %l4			;\
679	ldxa	[%sp + V9BIAS64 + 40]%asi, %l5			;\
680	ldxa	[%sp + V9BIAS64 + 48]%asi, %l6			;\
681	ldxa	[%sp + V9BIAS64 + 56]%asi, %l7			;\
682	ldxa	[%sp + V9BIAS64 + 64]%asi, %i0			;\
683	ldxa	[%sp + V9BIAS64 + 72]%asi, %i1			;\
684	ldxa	[%sp + V9BIAS64 + 80]%asi, %i2			;\
685	ldxa	[%sp + V9BIAS64 + 88]%asi, %i3			;\
686	ldxa	[%sp + V9BIAS64 + 96]%asi, %i4			;\
687	ldxa	[%sp + V9BIAS64 + 104]%asi, %i5			;\
688	ldxa	[%sp + V9BIAS64 + 112]%asi, %i6			;\
689	ldxa	[%sp + V9BIAS64 + 120]%asi, %i7			;\
690	restored						;\
691	retry							;\
692	SKIP(31-19-TT_TRACE_L_INS)				;\
693	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
694	.empty
695
696#endif /* !lint */
697
698/*
699 * SPILL_mixed spills either size window, depending on
700 * whether %sp is even or odd, to a 32-bit address space.
701 * This may only be used in conjunction with SPILL_32bit/
702 * SPILL_64bit. New versions of SPILL_mixed_{tt1,asi} would be
703 * needed for use with SPILL_{32,64}bit_{tt1,asi}.  Particular
704 * attention should be paid to the instructions that belong
705 * in the delay slots of the branches depending on the type
706 * of spill handler being branched to.
707 * Clear upper 32 bits of %sp if it is odd.
708 * We won't need to clear them in 64 bit kernel.
709 */
710#define	SPILL_mixed						\
711	btst	1, %sp						;\
712	bz,a,pt	%xcc, 1b					;\
713	srl	%sp, 0, %sp					;\
714	ba,pt	%xcc, 2b					;\
715	nop							;\
716	.align	128
717
718/*
719 * FILL_mixed(ASI) fills either size window, depending on
720 * whether %sp is even or odd, from a 32-bit address space.
721 * This may only be used in conjunction with FILL_32bit/
722 * FILL_64bit. New versions of FILL_mixed_{tt1,asi} would be
723 * needed for use with FILL_{32,64}bit_{tt1,asi}. Particular
724 * attention should be paid to the instructions that belong
725 * in the delay slots of the branches depending on the type
726 * of fill handler being branched to.
727 * Clear upper 32 bits of %sp if it is odd.
728 * We won't need to clear them in 64 bit kernel.
729 */
730#define	FILL_mixed						\
731	btst	1, %sp						;\
732	bz,a,pt	%xcc, 1b					;\
733	srl	%sp, 0, %sp					;\
734	ba,pt	%xcc, 2b					;\
735	nop							;\
736	.align	128
737
738
739/*
740 * SPILL_32clean/SPILL_64clean spill 32-bit and 64-bit register windows,
741 * respectively, into the address space via the designated asi.  The
742 * unbiased stack pointer is required to be eight-byte aligned (even for
743 * the 32-bit case even though this code does not require such strict
744 * alignment).
745 *
746 * With SPARC v9 the spill trap takes precedence over the cleanwin trap
747 * so when cansave == 0, canrestore == 6, and cleanwin == 6 the next save
748 * will cause cwp + 2 to be spilled but will not clean cwp + 1.  That
749 * window may contain kernel data so in user_rtt we set wstate to call
750 * these spill handlers on the first user spill trap.  These handler then
751 * spill the appropriate window but also back up a window and clean the
752 * window that didn't get a cleanwin trap.
753 */
754#define	SPILL_32clean(asi_num, tail)				\
755	srl	%sp, 0, %sp					;\
756	sta	%l0, [%sp + %g0]asi_num				;\
757	mov	4, %g1						;\
758	sta	%l1, [%sp + %g1]asi_num				;\
759	mov	8, %g2						;\
760	sta	%l2, [%sp + %g2]asi_num				;\
761	mov	12, %g3						;\
762	sta	%l3, [%sp + %g3]asi_num				;\
763	add	%sp, 16, %g4					;\
764	sta	%l4, [%g4 + %g0]asi_num				;\
765	sta	%l5, [%g4 + %g1]asi_num				;\
766	sta	%l6, [%g4 + %g2]asi_num				;\
767	sta	%l7, [%g4 + %g3]asi_num				;\
768	add	%g4, 16, %g4					;\
769	sta	%i0, [%g4 + %g0]asi_num				;\
770	sta	%i1, [%g4 + %g1]asi_num				;\
771	sta	%i2, [%g4 + %g2]asi_num				;\
772	sta	%i3, [%g4 + %g3]asi_num				;\
773	add	%g4, 16, %g4					;\
774	sta	%i4, [%g4 + %g0]asi_num				;\
775	sta	%i5, [%g4 + %g1]asi_num				;\
776	sta	%i6, [%g4 + %g2]asi_num				;\
777	sta	%i7, [%g4 + %g3]asi_num				;\
778	TT_TRACE_L(trace_win)					;\
779	b	.spill_clean					;\
780	  mov	WSTATE_USER32, %g7				;\
781	SKIP(31-25-TT_TRACE_L_INS)				;\
782	ba,a,pt	%xcc, fault_32bit_/**/tail			;\
783	.empty
784
785#define	SPILL_64clean(asi_num, tail)				\
786	mov	0 + V9BIAS64, %g1				;\
787	stxa	%l0, [%sp + %g1]asi_num				;\
788	mov	8 + V9BIAS64, %g2				;\
789	stxa	%l1, [%sp + %g2]asi_num				;\
790	mov	16 + V9BIAS64, %g3				;\
791	stxa	%l2, [%sp + %g3]asi_num				;\
792	mov	24 + V9BIAS64, %g4				;\
793	stxa	%l3, [%sp + %g4]asi_num				;\
794	add	%sp, 32, %g5					;\
795	stxa	%l4, [%g5 + %g1]asi_num				;\
796	stxa	%l5, [%g5 + %g2]asi_num				;\
797	stxa	%l6, [%g5 + %g3]asi_num				;\
798	stxa	%l7, [%g5 + %g4]asi_num				;\
799	add	%g5, 32, %g5					;\
800	stxa	%i0, [%g5 + %g1]asi_num				;\
801	stxa	%i1, [%g5 + %g2]asi_num				;\
802	stxa	%i2, [%g5 + %g3]asi_num				;\
803	stxa	%i3, [%g5 + %g4]asi_num				;\
804	add	%g5, 32, %g5					;\
805	stxa	%i4, [%g5 + %g1]asi_num				;\
806	stxa	%i5, [%g5 + %g2]asi_num				;\
807	stxa	%i6, [%g5 + %g3]asi_num				;\
808	stxa	%i7, [%g5 + %g4]asi_num				;\
809	TT_TRACE_L(trace_win)					;\
810	b	.spill_clean					;\
811	  mov	WSTATE_USER64, %g7				;\
812	SKIP(31-25-TT_TRACE_L_INS)				;\
813	ba,a,pt	%xcc, fault_64bit_/**/tail			;\
814	.empty
815
816
817/*
818 * Floating point disabled.
819 */
820#define	FP_DISABLED_TRAP		\
821	TT_TRACE(trace_gen)		;\
822	ba,pt	%xcc,.fp_disabled	;\
823	nop				;\
824	.align	32
825
826/*
827 * Floating point exceptions.
828 */
829#define	FP_IEEE_TRAP			\
830	TT_TRACE(trace_gen)		;\
831	ba,pt	%xcc,.fp_ieee_exception	;\
832	nop				;\
833	.align	32
834
835#define	FP_TRAP				\
836	TT_TRACE(trace_gen)		;\
837	ba,pt	%xcc,.fp_exception	;\
838	nop				;\
839	.align	32
840
841#if !defined(lint)
842/*
843 * asynchronous traps at level 0 and level 1
844 *
845 * The first instruction must be a membar for UltraSPARC-III
846 * to stop RED state entry if the store queue has many
847 * pending bad stores (PRM, Chapter 11).
848 */
849#define ASYNC_TRAP(ttype, ttlabel, table_name)\
850	.global	table_name	;\
851table_name:			;\
852	membar	#Sync		;\
853	TT_TRACE(ttlabel)	;\
854	ba	async_err	;\
855	mov	ttype, %g5	;\
856	.align	32
857
858/*
859 * Defaults to BAD entry, but establishes label to be used for
860 * architecture-specific overwrite of trap table entry.
861 */
862#define	LABELED_BAD(table_name)		\
863	.global	table_name		;\
864table_name:				;\
865	BAD
866
867#endif /* !lint */
868
869/*
870 * illegal instruction trap
871 */
872#define	ILLTRAP_INSTR			  \
873	membar	#Sync			  ;\
874	TT_TRACE(trace_gen)		  ;\
875	or	%g0, P_UTRAP4, %g2	  ;\
876	or	%g0, T_UNIMP_INSTR, %g3   ;\
877	sethi	%hi(.check_v9utrap), %g4  ;\
878	jmp	%g4 + %lo(.check_v9utrap) ;\
879	nop				  ;\
880	.align	32
881
882/*
883 * tag overflow trap
884 */
885#define	TAG_OVERFLOW			  \
886	TT_TRACE(trace_gen)		  ;\
887	or	%g0, P_UTRAP10, %g2	  ;\
888	or	%g0, T_TAG_OVERFLOW, %g3  ;\
889	sethi	%hi(.check_v9utrap), %g4  ;\
890	jmp	%g4 + %lo(.check_v9utrap) ;\
891	nop				  ;\
892	.align	32
893
894/*
895 * divide by zero trap
896 */
897#define	DIV_BY_ZERO			  \
898	TT_TRACE(trace_gen)		  ;\
899	or	%g0, P_UTRAP11, %g2	  ;\
900	or	%g0, T_IDIV0, %g3	  ;\
901	sethi	%hi(.check_v9utrap), %g4  ;\
902	jmp	%g4 + %lo(.check_v9utrap) ;\
903	nop				  ;\
904	.align	32
905
906/*
907 * trap instruction for V9 user trap handlers
908 */
909#define	TRAP_INSTR			  \
910	TT_TRACE(trace_gen)		  ;\
911	or	%g0, T_SOFTWARE_TRAP, %g3 ;\
912	sethi	%hi(.check_v9utrap), %g4  ;\
913	jmp	%g4 + %lo(.check_v9utrap) ;\
914	nop				  ;\
915	.align	32
916#define	TRP4	TRAP_INSTR; TRAP_INSTR; TRAP_INSTR; TRAP_INSTR
917
918/*
919 * LEVEL_INTERRUPT is for level N interrupts.
920 * VECTOR_INTERRUPT is for the vector trap.
921 */
922#define	LEVEL_INTERRUPT(level)		\
923	.global	tt_pil/**/level		;\
924tt_pil/**/level:			;\
925	ba,pt	%xcc, pil_interrupt	;\
926	mov	level, %g4		;\
927	.align	32
928
929#define	LEVEL14_INTERRUPT			\
930	ba	pil14_interrupt			;\
931	mov	PIL_14, %g4			;\
932	.align	32
933
934#define	VECTOR_INTERRUPT				\
935	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g1	;\
936	btst	IRSR_BUSY, %g1				;\
937	bnz,pt	%xcc, vec_interrupt			;\
938	nop						;\
939	ba,a,pt	%xcc, vec_intr_spurious			;\
940	.empty						;\
941	.align	32
942
943/*
944 * MMU Trap Handlers.
945 */
946#define	SWITCH_GLOBALS	/* mmu->alt, alt->mmu */			\
947	rdpr	%pstate, %g5						;\
948	wrpr	%g5, PSTATE_MG | PSTATE_AG, %pstate
949
950#define	IMMU_EXCEPTION							\
951	membar	#Sync							;\
952	SWITCH_GLOBALS							;\
953	wr	%g0, ASI_IMMU, %asi					;\
954	rdpr	%tpc, %g2						;\
955	ldxa	[MMU_SFSR]%asi, %g3					;\
956	ba,pt	%xcc, .mmu_exception_end				;\
957	mov	T_INSTR_EXCEPTION, %g1					;\
958	.align	32
959
960#define	DMMU_EXCEPTION							\
961	SWITCH_GLOBALS							;\
962	wr	%g0, ASI_DMMU, %asi					;\
963	ldxa	[MMU_TAG_ACCESS]%asi, %g2				;\
964	ldxa	[MMU_SFSR]%asi, %g3					;\
965	ba,pt	%xcc, .mmu_exception_end				;\
966	mov	T_DATA_EXCEPTION, %g1					;\
967	.align	32
968
969#define	DMMU_EXC_AG_PRIV						\
970	wr	%g0, ASI_DMMU, %asi					;\
971	ldxa	[MMU_SFAR]%asi, %g2					;\
972	ba,pt	%xcc, .mmu_priv_exception				;\
973	ldxa	[MMU_SFSR]%asi, %g3					;\
974	.align	32
975
976#define	DMMU_EXC_AG_NOT_ALIGNED						\
977	wr	%g0, ASI_DMMU, %asi					;\
978	ldxa	[MMU_SFAR]%asi, %g2					;\
979	ba,pt	%xcc, .mmu_exception_not_aligned			;\
980	ldxa	[MMU_SFSR]%asi, %g3					;\
981	.align	32
982
983/*
984 * SPARC V9 IMPL. DEP. #109(1) and (2) and #110(1) and (2)
985 */
986#define	DMMU_EXC_LDDF_NOT_ALIGNED					\
987	btst	1, %sp							;\
988	bnz,pt	%xcc, .lddf_exception_not_aligned			;\
989	wr	%g0, ASI_DMMU, %asi					;\
990	ldxa	[MMU_SFAR]%asi, %g2					;\
991	ba,pt	%xcc, .mmu_exception_not_aligned			;\
992	ldxa	[MMU_SFSR]%asi, %g3					;\
993	.align	32
994
995#define	DMMU_EXC_STDF_NOT_ALIGNED					\
996	btst	1, %sp							;\
997	bnz,pt	%xcc, .stdf_exception_not_aligned			;\
998	wr	%g0, ASI_DMMU, %asi					;\
999	ldxa	[MMU_SFAR]%asi, %g2					;\
1000	ba,pt	%xcc, .mmu_exception_not_aligned			;\
1001	ldxa	[MMU_SFSR]%asi, %g3					;\
1002	.align	32
1003
1004/*
1005 * Flush the TLB using either the primary, secondary, or nucleus flush
1006 * operation based on whether the ctx from the tag access register matches
1007 * the primary or secondary context (flush the nucleus if neither matches).
1008 *
1009 * Requires a membar #Sync before next ld/st.
1010 * exits with:
1011 * g2 = tag access register
1012 * g3 = ctx number
1013 */
1014#if TAGACC_CTX_MASK != CTXREG_CTX_MASK
1015#error "TAGACC_CTX_MASK != CTXREG_CTX_MASK"
1016#endif
1017#define	DTLB_DEMAP_ENTRY						\
1018	mov	MMU_TAG_ACCESS, %g1					;\
1019	mov	MMU_PCONTEXT, %g5					;\
1020	ldxa	[%g1]ASI_DMMU, %g2					;\
1021	sethi	%hi(TAGACC_CTX_MASK), %g4				;\
1022	or	%g4, %lo(TAGACC_CTX_MASK), %g4				;\
1023	and	%g2, %g4, %g3			/* g3 = ctx */		;\
1024	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = primary ctx */	;\
1025	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1026	cmp	%g3, %g6						;\
1027	be,pt	%xcc, 1f						;\
1028	andn	%g2, %g4, %g1			/* ctx = primary */	;\
1029	mov	MMU_SCONTEXT, %g5					;\
1030	ldxa	[%g5]ASI_DMMU, %g6		/* g6 = secondary ctx */ ;\
1031	and	%g6, %g4, %g6			/* &= CTXREG_CTX_MASK */ ;\
1032	cmp	%g3, %g6						;\
1033	be,a,pt	%xcc, 1f						;\
1034	  or	%g1, DEMAP_SECOND, %g1					;\
1035	or	%g1, DEMAP_NUCLEUS, %g1					;\
10361:	stxa	%g0, [%g1]ASI_DTLB_DEMAP	/* MMU_DEMAP_PAGE */	;\
1037	membar	#Sync
1038
1039#if defined(cscope)
1040/*
1041 * Define labels to direct cscope quickly to labels that
1042 * are generated by macro expansion of DTLB_MISS().
1043 */
1044	.global	tt0_dtlbmiss
1045tt0_dtlbmiss:
1046	.global	tt1_dtlbmiss
1047tt1_dtlbmiss:
1048	nop
1049#endif
1050
1051/*
1052 * Needs to be exactly 32 instructions
1053 *
1054 * UTLB NOTE: If we don't hit on the 8k pointer then we branch
1055 * to a special 4M tsb handler. It would be nice if that handler
1056 * could live in this file but currently it seems better to allow
1057 * it to fall thru to sfmmu_tsb_miss.
1058 */
1059#ifdef UTSB_PHYS
1060#define	DTLB_MISS(table_name)						;\
1061	.global	table_name/**/_dtlbmiss					;\
1062table_name/**/_dtlbmiss:						;\
1063	HAT_PERCPU_DBSTAT(TSBMISS_DTLBMISS) /* 3 instr ifdef DEBUG */	;\
1064	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1065	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1066	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1067	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1068	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1069	cmp	%g3, INVALID_CONTEXT					;\
1070	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1071	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1072	mov	SCRATCHPAD_UTSBREG, %g3				;\
1073	ldxa	[%g3]ASI_SCRATCHPAD, %g3	/* g3 = 2nd tsb reg */	;\
1074	brgez,pn %g3, sfmmu_udtlb_slowpath	/* branch if 2 TSBs */	;\
1075	  nop								;\
1076	ldda	[%g1]ASI_QUAD_LDD_PHYS, %g4	/* g4 = tag, %g5 data */;\
1077	cmp	%g4, %g7						;\
1078	bne,pn	%xcc, sfmmu_tsb_miss_tt		/* no 4M TSB, miss */	;\
1079	  mov	%g0, %g3			/* clear 4M tsbe ptr */	;\
1080	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1081	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1082	retry				/* in %g5 */			;\
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	.align 128
1093#else /* UTSB_PHYS */
1094#define	DTLB_MISS(table_name)						;\
1095	.global	table_name/**/_dtlbmiss					;\
1096table_name/**/_dtlbmiss:						;\
1097	HAT_PERCPU_DBSTAT(TSBMISS_DTLBMISS) /* 3 instr ifdef DEBUG */	;\
1098	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1099	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1100	ldxa	[%g6]ASI_DMMU, %g2		/* g2 = tag access */	;\
1101	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1102	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1103	cmp	%g3, INVALID_CONTEXT					;\
1104	ble,pn	%xcc, sfmmu_kdtlb_miss					;\
1105	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1106	brlz,pn %g1, sfmmu_udtlb_slowpath				;\
1107	  nop								;\
1108	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, %g5 data */	;\
1109	cmp	%g4, %g7						;\
1110	bne,pn	%xcc, sfmmu_tsb_miss_tt		/* no 4M TSB, miss */	;\
1111	  mov	%g0, %g3		/* clear 4M tsbe ptr */		;\
1112	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1113	stxa	%g5, [%g0]ASI_DTLB_IN	/* trapstat expects TTE */	;\
1114	retry				/* in %g5 */			;\
1115	unimp	0							;\
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	.align 128
1127#endif /* UTSB_PHYS */
1128
1129#if defined(cscope)
1130/*
1131 * Define labels to direct cscope quickly to labels that
1132 * are generated by macro expansion of ITLB_MISS().
1133 */
1134	.global	tt0_itlbmiss
1135tt0_itlbmiss:
1136	.global	tt1_itlbmiss
1137tt1_itlbmiss:
1138	nop
1139#endif
1140
1141/*
1142 * Instruction miss handler.
1143 * ldda instructions will have their ASI patched
1144 * by sfmmu_patch_ktsb at runtime.
1145 * MUST be EXACTLY 32 instructions or we'll break.
1146 */
1147#ifdef UTSB_PHYS
1148#define	ITLB_MISS(table_name)						 \
1149	.global	table_name/**/_itlbmiss					;\
1150table_name/**/_itlbmiss:						;\
1151	HAT_PERCPU_DBSTAT(TSBMISS_ITLBMISS) /* 3 instr ifdef DEBUG */	;\
1152	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1153	ldxa	[%g0]ASI_IMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1154	ldxa	[%g6]ASI_IMMU, %g2		/* g2 = tag access */	;\
1155	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1156	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1157	cmp	%g3, INVALID_CONTEXT					;\
1158	ble,pn	%xcc, sfmmu_kitlb_miss					;\
1159	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1160	mov	SCRATCHPAD_UTSBREG, %g3				;\
1161	ldxa	[%g3]ASI_SCRATCHPAD, %g3	/* g3 = 2nd tsb reg */	;\
1162	brgez,pn %g3, sfmmu_uitlb_slowpath	/* branch if 2 TSBs */	;\
1163	  nop								;\
1164	ldda	[%g1]ASI_QUAD_LDD_PHYS, %g4 /* g4 = tag, g5 = data */	;\
1165	cmp	%g4, %g7						;\
1166	bne,pn	%xcc, sfmmu_tsb_miss_tt	/* br if 8k ptr miss */		;\
1167	  mov	%g0, %g3		/* no 4M TSB */			;\
1168	andcc	%g5, TTE_EXECPRM_INT, %g0 /* check execute bit */	;\
1169	bz,pn	%icc, exec_fault					;\
1170	  nop								;\
1171	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1172	stxa	%g5, [%g0]ASI_ITLB_IN	/* trapstat expects %g5 */	;\
1173	retry								;\
1174	unimp	0							;\
1175	unimp	0							;\
1176	unimp	0							;\
1177	unimp	0							;\
1178	unimp	0							;\
1179	unimp	0							;\
1180	.align 128
1181#else /* UTSB_PHYS */
1182#define	ITLB_MISS(table_name)						 \
1183	.global	table_name/**/_itlbmiss					;\
1184table_name/**/_itlbmiss:						;\
1185	HAT_PERCPU_DBSTAT(TSBMISS_ITLBMISS) /* 3 instr ifdef DEBUG */	;\
1186	mov	MMU_TAG_ACCESS, %g6		/* select tag acc */	;\
1187	ldxa	[%g0]ASI_IMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1188	ldxa	[%g6]ASI_IMMU, %g2		/* g2 = tag access */	;\
1189	sllx	%g2, TAGACC_CTX_LSHIFT, %g3				;\
1190	srlx	%g3, TAGACC_CTX_LSHIFT, %g3	/* g3 = ctx */		;\
1191	cmp	%g3, INVALID_CONTEXT					;\
1192	ble,pn	%xcc, sfmmu_kitlb_miss					;\
1193	  srlx	%g2, TAG_VALO_SHIFT, %g7	/* g7 = tsb tag */	;\
1194	brlz,pn	%g1, sfmmu_uitlb_slowpath	/* if >1 TSB branch */	;\
1195	  nop								;\
1196	ldda	[%g1]ASI_NQUAD_LD, %g4	/* g4 = tag, g5 = data */	;\
1197	cmp	%g4, %g7						;\
1198	bne,pn	%xcc, sfmmu_tsb_miss_tt	/* br if 8k ptr miss */		;\
1199	  mov	%g0, %g3		/* no 4M TSB */			;\
1200	andcc	%g5, TTE_EXECPRM_INT, %g0 /* check execute bit */	;\
1201	bz,pn	%icc, exec_fault					;\
1202	  nop								;\
1203	TT_TRACE(trace_tsbhit)		/* 2 instr ifdef TRAPTRACE */	;\
1204	stxa	%g5, [%g0]ASI_ITLB_IN	/* trapstat expects %g5 */	;\
1205	retry								;\
1206	unimp	0							;\
1207	unimp	0							;\
1208	unimp	0							;\
1209	unimp	0							;\
1210	unimp	0							;\
1211	unimp	0							;\
1212	unimp	0							;\
1213	unimp	0							;\
1214	.align 128
1215#endif /* UTSB_PHYS */
1216
1217
1218/*
1219 * This macro is the first level handler for fast protection faults.
1220 * It first demaps the tlb entry which generated the fault and then
1221 * attempts to set the modify bit on the hash.  It needs to be
1222 * exactly 32 instructions.
1223 */
1224#define	DTLB_PROT							 \
1225	DTLB_DEMAP_ENTRY		/* 20 instructions */		;\
1226	/*								;\
1227	 * At this point:						;\
1228	 *   g1 = ????							;\
1229	 *   g2 = tag access register					;\
1230	 *   g3 = ctx number						;\
1231	 *   g4 = ????							;\
1232	 */								;\
1233	TT_TRACE(trace_dataprot)	/* 2 instr ifdef TRAPTRACE */	;\
1234					/* clobbers g1 and g6 */	;\
1235	ldxa	[%g0]ASI_DMMU_TSB_8K, %g1	/* g1 = tsbe ptr */	;\
1236	brnz,pt %g3, sfmmu_uprot_trap		/* user trap */		;\
1237	  nop								;\
1238	ba,a,pt	%xcc, sfmmu_kprot_trap		/* kernel trap */	;\
1239	unimp	0							;\
1240	unimp	0							;\
1241	unimp	0							;\
1242	unimp	0							;\
1243	unimp	0							;\
1244	unimp	0							;\
1245	.align 128
1246
1247#define	DMMU_EXCEPTION_TL1						;\
1248	SWITCH_GLOBALS							;\
1249	ba,a,pt	%xcc, mmu_trap_tl1					;\
1250	  nop								;\
1251	.align 32
1252
1253#define	MISALIGN_ADDR_TL1						;\
1254	ba,a,pt	%xcc, mmu_trap_tl1					;\
1255	  nop								;\
1256	.align 32
1257
1258/*
1259 * Trace a tsb hit
1260 * g1 = tsbe pointer (in/clobbered)
1261 * g2 = tag access register (in)
1262 * g3 - g4 = scratch (clobbered)
1263 * g5 = tsbe data (in)
1264 * g6 = scratch (clobbered)
1265 * g7 = pc we jumped here from (in)
1266 * ttextra = value to OR in to trap type (%tt) (in)
1267 */
1268#ifdef TRAPTRACE
1269#define TRACE_TSBHIT(ttextra)						 \
1270	membar	#Sync							;\
1271	sethi	%hi(FLUSH_ADDR), %g6					;\
1272	flush	%g6							;\
1273	TRACE_PTR(%g3, %g6)						;\
1274	GET_TRACE_TICK(%g6)						;\
1275	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi				;\
1276	stxa	%g2, [%g3 + TRAP_ENT_SP]%asi	/* tag access */	;\
1277	stxa	%g5, [%g3 + TRAP_ENT_F1]%asi	/* tsb data */		;\
1278	rdpr	%tnpc, %g6						;\
1279	stxa	%g6, [%g3 + TRAP_ENT_F2]%asi				;\
1280	stxa	%g1, [%g3 + TRAP_ENT_F3]%asi	/* tsb pointer */	;\
1281	stxa	%g0, [%g3 + TRAP_ENT_F4]%asi				;\
1282	rdpr	%tpc, %g6						;\
1283	stxa	%g6, [%g3 + TRAP_ENT_TPC]%asi				;\
1284	rdpr	%tl, %g6						;\
1285	stha	%g6, [%g3 + TRAP_ENT_TL]%asi				;\
1286	rdpr	%tt, %g6						;\
1287	or	%g6, (ttextra), %g6					;\
1288	stha	%g6, [%g3 + TRAP_ENT_TT]%asi				;\
1289	ldxa	[%g0]ASI_IMMU, %g1		/* tag target */	;\
1290	ldxa	[%g0]ASI_DMMU, %g4					;\
1291	cmp	%g6, FAST_IMMU_MISS_TT					;\
1292	movne	%icc, %g4, %g1						;\
1293	stxa	%g1, [%g3 + TRAP_ENT_TSTATE]%asi /* tsb tag */		;\
1294	stxa	%g0, [%g3 + TRAP_ENT_TR]%asi				;\
1295	TRACE_NEXT(%g3, %g4, %g6)
1296#else
1297#define TRACE_TSBHIT(ttextra)
1298#endif
1299
1300#if defined(lint)
1301
1302struct scb	trap_table;
1303struct scb	scb;		/* trap_table/scb are the same object */
1304
1305#else /* lint */
1306
1307/*
1308 * =======================================================================
1309 *		SPARC V9 TRAP TABLE
1310 *
1311 * The trap table is divided into two halves: the first half is used when
1312 * taking traps when TL=0; the second half is used when taking traps from
1313 * TL>0. Note that handlers in the second half of the table might not be able
1314 * to make the same assumptions as handlers in the first half of the table.
1315 *
1316 * Worst case trap nesting so far:
1317 *
1318 *	at TL=0 client issues software trap requesting service
1319 *	at TL=1 nucleus wants a register window
1320 *	at TL=2 register window clean/spill/fill takes a TLB miss
1321 *	at TL=3 processing TLB miss
1322 *	at TL=4 handle asynchronous error
1323 *
1324 * Note that a trap from TL=4 to TL=5 places Spitfire in "RED mode".
1325 *
1326 * =======================================================================
1327 */
1328	.section ".text"
1329	.align	4
1330	.global trap_table, scb, trap_table0, trap_table1, etrap_table
1331	.type	trap_table, #object
1332	.type	scb, #object
1333trap_table:
1334scb:
1335trap_table0:
1336	/* hardware traps */
1337	NOT;				/* 000	reserved */
1338	RED;				/* 001	power on reset */
1339	RED;				/* 002	watchdog reset */
1340	RED;				/* 003	externally initiated reset */
1341	RED;				/* 004	software initiated reset */
1342	RED;				/* 005	red mode exception */
1343	NOT; NOT;			/* 006 - 007 reserved */
1344	IMMU_EXCEPTION;			/* 008	instruction access exception */
1345	NOT;				/* 009	instruction access MMU miss */
1346	ASYNC_TRAP(T_INSTR_ERROR, trace_gen, tt0_iae);
1347					/* 00A	instruction access error */
1348	NOT; NOT4;			/* 00B - 00F reserved */
1349	ILLTRAP_INSTR;			/* 010	illegal instruction */
1350	TRAP(T_PRIV_INSTR);		/* 011	privileged opcode */
1351	NOT;				/* 012	unimplemented LDD */
1352	NOT;				/* 013	unimplemented STD */
1353	NOT4; NOT4; NOT4;		/* 014 - 01F reserved */
1354	FP_DISABLED_TRAP;		/* 020	fp disabled */
1355	FP_IEEE_TRAP;			/* 021	fp exception ieee 754 */
1356	FP_TRAP;			/* 022	fp exception other */
1357	TAG_OVERFLOW;			/* 023	tag overflow */
1358	CLEAN_WINDOW;			/* 024 - 027 clean window */
1359	DIV_BY_ZERO;			/* 028	division by zero */
1360	NOT;				/* 029	internal processor error */
1361	NOT; NOT; NOT4;			/* 02A - 02F reserved */
1362	DMMU_EXCEPTION;			/* 030	data access exception */
1363	NOT;				/* 031	data access MMU miss */
1364	ASYNC_TRAP(T_DATA_ERROR, trace_gen, tt0_dae);
1365					/* 032	data access error */
1366	NOT;				/* 033	data access protection */
1367	DMMU_EXC_AG_NOT_ALIGNED;	/* 034	mem address not aligned */
1368	DMMU_EXC_LDDF_NOT_ALIGNED;	/* 035	LDDF mem address not aligned */
1369	DMMU_EXC_STDF_NOT_ALIGNED;	/* 036	STDF mem address not aligned */
1370	DMMU_EXC_AG_PRIV;		/* 037	privileged action */
1371	NOT;				/* 038	LDQF mem address not aligned */
1372	NOT;				/* 039	STQF mem address not aligned */
1373	NOT; NOT; NOT4;			/* 03A - 03F reserved */
1374	LABELED_BAD(tt0_asdat);		/* 040	async data error */
1375	LEVEL_INTERRUPT(1);		/* 041	interrupt level 1 */
1376	LEVEL_INTERRUPT(2);		/* 042	interrupt level 2 */
1377	LEVEL_INTERRUPT(3);		/* 043	interrupt level 3 */
1378	LEVEL_INTERRUPT(4);		/* 044	interrupt level 4 */
1379	LEVEL_INTERRUPT(5);		/* 045	interrupt level 5 */
1380	LEVEL_INTERRUPT(6);		/* 046	interrupt level 6 */
1381	LEVEL_INTERRUPT(7);		/* 047	interrupt level 7 */
1382	LEVEL_INTERRUPT(8);		/* 048	interrupt level 8 */
1383	LEVEL_INTERRUPT(9);		/* 049	interrupt level 9 */
1384	LEVEL_INTERRUPT(10);		/* 04A	interrupt level 10 */
1385	LEVEL_INTERRUPT(11);		/* 04B	interrupt level 11 */
1386	LEVEL_INTERRUPT(12);		/* 04C	interrupt level 12 */
1387	LEVEL_INTERRUPT(13);		/* 04D	interrupt level 13 */
1388	LEVEL14_INTERRUPT;		/* 04E	interrupt level 14 */
1389	LEVEL_INTERRUPT(15);		/* 04F	interrupt level 15 */
1390	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F reserved */
1391	VECTOR_INTERRUPT;		/* 060	interrupt vector */
1392	GOTO(kmdb_trap);		/* 061	PA watchpoint */
1393	GOTO(kmdb_trap);		/* 062	VA watchpoint */
1394	GOTO_TT(ce_err, trace_gen);	/* 063	corrected ECC error */
1395	ITLB_MISS(tt0);			/* 064	instruction access MMU miss */
1396	DTLB_MISS(tt0);			/* 068	data access MMU miss */
1397	DTLB_PROT;			/* 06C	data access protection */
1398	LABELED_BAD(tt0_fecc);		/* 070  fast ecache ECC error */
1399	LABELED_BAD(tt0_dperr);		/* 071  Cheetah+ dcache parity error */
1400	LABELED_BAD(tt0_iperr);		/* 072  Cheetah+ icache parity error */
1401	NOT;				/* 073  reserved */
1402	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1403	NOT4;				/* 080	spill 0 normal */
1404	SPILL_32bit_asi(ASI_AIUP,sn0);	/* 084	spill 1 normal */
1405	SPILL_64bit_asi(ASI_AIUP,sn0);	/* 088	spill 2 normal */
1406	SPILL_32clean(ASI_AIUP,sn0);	/* 08C	spill 3 normal */
1407	SPILL_64clean(ASI_AIUP,sn0);	/* 090	spill 4 normal */
1408	SPILL_32bit(not);		/* 094	spill 5 normal */
1409	SPILL_64bit(not);		/* 098	spill 6 normal */
1410	SPILL_mixed;			/* 09C	spill 7 normal */
1411	NOT4;				/* 0A0	spill 0 other */
1412	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0A4	spill 1 other */
1413	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0A8	spill 2 other */
1414	SPILL_32bit_asi(ASI_AIUS,so0);	/* 0AC	spill 3 other */
1415	SPILL_64bit_asi(ASI_AIUS,so0);	/* 0B0	spill 4 other */
1416	NOT4;				/* 0B4	spill 5 other */
1417	NOT4;				/* 0B8	spill 6 other */
1418	NOT4;				/* 0BC	spill 7 other */
1419	NOT4;				/* 0C0	fill 0 normal */
1420	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0C4	fill 1 normal */
1421	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0C8	fill 2 normal */
1422	FILL_32bit_asi(ASI_AIUP,fn0);	/* 0CC	fill 3 normal */
1423	FILL_64bit_asi(ASI_AIUP,fn0);	/* 0D0	fill 4 normal */
1424	FILL_32bit(not);		/* 0D4	fill 5 normal */
1425	FILL_64bit(not);		/* 0D8	fill 6 normal */
1426	FILL_mixed;			/* 0DC	fill 7 normal */
1427	NOT4;				/* 0E0	fill 0 other */
1428	NOT4;				/* 0E4	fill 1 other */
1429	NOT4;				/* 0E8	fill 2 other */
1430	NOT4;				/* 0EC	fill 3 other */
1431	NOT4;				/* 0F0	fill 4 other */
1432	NOT4;				/* 0F4	fill 5 other */
1433	NOT4;				/* 0F8	fill 6 other */
1434	NOT4;				/* 0FC	fill 7 other */
1435	/* user traps */
1436	GOTO(syscall_trap_4x);		/* 100	old system call */
1437	TRAP(T_BREAKPOINT);		/* 101	user breakpoint */
1438	TRAP(T_DIV0);			/* 102	user divide by zero */
1439	FLUSHW();			/* 103	flush windows */
1440	GOTO(.clean_windows);		/* 104	clean windows */
1441	BAD;				/* 105	range check ?? */
1442	GOTO(.fix_alignment);		/* 106	do unaligned references */
1443	BAD;				/* 107	unused */
1444#ifdef	DEBUG
1445	GOTO(syscall_wrapper32)		/* 108	ILP32 system call on LP64 */
1446#else
1447	SYSCALL(syscall_trap32)		/* 108	ILP32 system call on LP64 */
1448#endif
1449	GOTO(set_trap0_addr);		/* 109	set trap0 address */
1450	BAD; BAD; BAD4;			/* 10A - 10F unused */
1451	TRP4; TRP4; TRP4; TRP4;		/* 110 - 11F V9 user trap handlers */
1452	GOTO(.getcc);			/* 120	get condition codes */
1453	GOTO(.setcc);			/* 121	set condition codes */
1454	GOTO(.getpsr);			/* 122	get psr */
1455	GOTO(.setpsr);			/* 123	set psr (some fields) */
1456	GOTO(get_timestamp);		/* 124	get timestamp */
1457	GOTO(get_virtime);		/* 125	get lwp virtual time */
1458	PRIV(self_xcall);		/* 126	self xcall */
1459	GOTO(get_hrestime);		/* 127	get hrestime */
1460	BAD;				/* 128	ST_SETV9STACK */
1461	GOTO(.getlgrp);			/* 129  get lgrpid */
1462	BAD; BAD; BAD4;			/* 12A - 12F unused */
1463	BAD4; BAD4; 			/* 130 - 137 unused */
1464	DTRACE_PID;			/* 138  dtrace pid tracing provider */
1465	BAD;				/* 139  unused */
1466	DTRACE_RETURN;			/* 13A	dtrace pid return probe */
1467	BAD; BAD4;			/* 13B - 13F unused */
1468#ifdef	DEBUG
1469	GOTO(syscall_wrapper)		/* 140  LP64 system call */
1470#else
1471	SYSCALL(syscall_trap)		/* 140  LP64 system call */
1472#endif
1473	SYSCALL(nosys);			/* 141  unused system call trap */
1474#ifdef DEBUG_USER_TRAPTRACECTL
1475	GOTO(.traptrace_freeze);	/* 142  freeze traptrace */
1476	GOTO(.traptrace_unfreeze);	/* 143  unfreeze traptrace */
1477#else
1478	SYSCALL(nosys);			/* 142  unused system call trap */
1479	SYSCALL(nosys);			/* 143  unused system call trap */
1480#endif
1481	BAD4; BAD4; BAD4;		/* 144 - 14F unused */
1482	BAD4; BAD4; BAD4; BAD4;		/* 150 - 15F unused */
1483	BAD4; BAD4; BAD4; BAD4;		/* 160 - 16F unused */
1484	BAD;				/* 170 - unused */
1485	BAD;				/* 171 - unused */
1486	BAD; BAD;			/* 172 - 173 unused */
1487	BAD4; BAD4;			/* 174 - 17B unused */
1488#ifdef	PTL1_PANIC_DEBUG
1489	mov PTL1_BAD_DEBUG, %g1; GOTO(ptl1_panic);
1490					/* 17C	test ptl1_panic */
1491#else
1492	BAD;				/* 17C  unused */
1493#endif	/* PTL1_PANIC_DEBUG */
1494	PRIV(kmdb_trap);		/* 17D	kmdb enter (L1-A) */
1495	PRIV(kmdb_trap);		/* 17E	kmdb breakpoint */
1496	PRIV(kctx_obp_bpt);		/* 17F	obp breakpoint */
1497	/* reserved */
1498	NOT4; NOT4; NOT4; NOT4;		/* 180 - 18F reserved */
1499	NOT4; NOT4; NOT4; NOT4;		/* 190 - 19F reserved */
1500	NOT4; NOT4; NOT4; NOT4;		/* 1A0 - 1AF reserved */
1501	NOT4; NOT4; NOT4; NOT4;		/* 1B0 - 1BF reserved */
1502	NOT4; NOT4; NOT4; NOT4;		/* 1C0 - 1CF reserved */
1503	NOT4; NOT4; NOT4; NOT4;		/* 1D0 - 1DF reserved */
1504	NOT4; NOT4; NOT4; NOT4;		/* 1E0 - 1EF reserved */
1505	NOT4; NOT4; NOT4; NOT4;		/* 1F0 - 1FF reserved */
1506trap_table1:
1507	NOT4; NOT4; NOT; NOT;		/* 000 - 009 unused */
1508	ASYNC_TRAP(T_INSTR_ERROR + T_TL1, trace_gen, tt1_iae);
1509					/* 00A	instruction access error */
1510	NOT; NOT4;			/* 00B - 00F unused */
1511	NOT4; NOT4; NOT4; NOT4;		/* 010 - 01F unused */
1512	NOT4;				/* 020 - 023 unused */
1513	CLEAN_WINDOW;			/* 024 - 027 clean window */
1514	NOT4; NOT4;			/* 028 - 02F unused */
1515	DMMU_EXCEPTION_TL1;		/* 030 	data access exception */
1516	NOT;				/* 031 unused */
1517	ASYNC_TRAP(T_DATA_ERROR + T_TL1, trace_gen, tt1_dae);
1518					/* 032	data access error */
1519	NOT;				/* 033	unused */
1520	MISALIGN_ADDR_TL1;		/* 034	mem address not aligned */
1521	NOT; NOT; NOT; NOT4; NOT4	/* 035 - 03F unused */
1522	LABELED_BAD(tt1_asdat);		/* 040	async data error */
1523	NOT; NOT; NOT;			/* 041 - 043 unused */
1524	NOT4; NOT4; NOT4;		/* 044 - 04F unused */
1525	NOT4; NOT4; NOT4; NOT4;		/* 050 - 05F unused */
1526	NOT;				/* 060	unused */
1527	GOTO(kmdb_trap_tl1);		/* 061	PA watchpoint */
1528	GOTO(kmdb_trap_tl1);		/* 062	VA watchpoint */
1529	GOTO_TT(ce_err_tl1, trace_gen);	/* 063	corrected ECC error */
1530	ITLB_MISS(tt1);			/* 064	instruction access MMU miss */
1531	DTLB_MISS(tt1);			/* 068	data access MMU miss */
1532	DTLB_PROT;			/* 06C	data access protection */
1533	LABELED_BAD(tt1_fecc);		/* 070  fast ecache ECC error */
1534	LABELED_BAD(tt1_dperr);		/* 071  Cheetah+ dcache parity error */
1535	LABELED_BAD(tt1_iperr);		/* 072  Cheetah+ icache parity error */
1536	NOT;				/* 073  reserved */
1537	NOT4; NOT4; NOT4;		/* 074 - 07F reserved */
1538	NOT4;				/* 080	spill 0 normal */
1539	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 084	spill 1 normal */
1540	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 088	spill 2 normal */
1541	SPILL_32bit_tt1(ASI_AIUP,sn1);	/* 08C	spill 3 normal */
1542	SPILL_64bit_tt1(ASI_AIUP,sn1);	/* 090	spill 4 normal */
1543	SPILL_32bit(not);		/* 094	spill 5 normal */
1544	SPILL_64bit(not);		/* 098	spill 6 normal */
1545	SPILL_mixed;			/* 09C	spill 7 normal */
1546	NOT4;				/* 0A0	spill 0 other */
1547	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0A4	spill 1 other */
1548	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0A8	spill 2 other */
1549	SPILL_32bit_tt1(ASI_AIUS,so1);	/* 0AC	spill 3 other */
1550	SPILL_64bit_tt1(ASI_AIUS,so1);	/* 0B0  spill 4 other */
1551	NOT4;				/* 0B4  spill 5 other */
1552	NOT4;				/* 0B8  spill 6 other */
1553	NOT4;				/* 0BC  spill 7 other */
1554	NOT4;				/* 0C0	fill 0 normal */
1555	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0C4	fill 1 normal */
1556	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0C8	fill 2 normal */
1557	FILL_32bit_tt1(ASI_AIUP,fn1);	/* 0CC	fill 3 normal */
1558	FILL_64bit_tt1(ASI_AIUP,fn1);	/* 0D0	fill 4 normal */
1559	FILL_32bit(not);		/* 0D4	fill 5 normal */
1560	FILL_64bit(not);		/* 0D8	fill 6 normal */
1561	FILL_mixed;			/* 0DC	fill 7 normal */
1562	NOT4; NOT4; NOT4; NOT4;		/* 0E0 - 0EF unused */
1563	NOT4; NOT4; NOT4; NOT4;		/* 0F0 - 0FF unused */
1564	LABELED_BAD(tt1_swtrap0);	/* 100  fast ecache ECC error (cont) */
1565	LABELED_BAD(tt1_swtrap1);	/* 101  Ch+ D$ parity error (cont) */
1566	LABELED_BAD(tt1_swtrap2);	/* 102  Ch+ I$ parity error (cont) */
1567	NOT;				/* 103  reserved */
1568/*
1569 * We only reserve the above four special case soft traps for code running
1570 * at TL>0, so we can truncate the trap table here.
1571 */
1572etrap_table:
1573	.size	trap_table, (.-trap_table)
1574	.size	scb, (.-scb)
1575
1576/*
1577 * We get to exec_fault in the case of an instruction miss and tte
1578 * has no execute bit set.  We go to tl0 to handle it.
1579 *
1580 * g1 = tsbe pointer (in/clobbered)
1581 * g2 = tag access register (in)
1582 * g3 - g4 = scratch (clobbered)
1583 * g5 = tsbe data (in)
1584 * g6 = scratch (clobbered)
1585 */
1586	ALTENTRY(exec_fault)
1587	TRACE_TSBHIT(0x200)
1588	SWITCH_GLOBALS
1589	mov	MMU_TAG_ACCESS, %g4
1590	ldxa	[%g4]ASI_IMMU, %g2			! arg1 = addr
1591	mov	T_INSTR_MMU_MISS, %g3			! arg2 = traptype
1592	set	trap, %g1
1593	ba,pt	%xcc, sys_trap
1594	  mov	-1, %g4
1595
1596.mmu_exception_not_aligned:
1597	rdpr	%tstate, %g1
1598	btst	TSTATE_PRIV, %g1
1599	bnz,pn	%icc, 2f
1600	nop
1601	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1602	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1603	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1604	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1605	brz,pt	%g5, 2f
1606	nop
1607	ldn	[%g5 + P_UTRAP15], %g5			! unaligned utrap?
1608	brz,pn	%g5, 2f
1609	nop
1610	btst	1, %sp
1611	bz,pt	%xcc, 1f				! 32 bit user program
1612	nop
1613	ba,pt	%xcc, .setup_v9utrap			! 64 bit user program
1614	nop
16151:
1616	ba,pt	%xcc, .setup_utrap
1617	or	%g2, %g0, %g7
16182:
1619	ba,pt	%xcc, .mmu_exception_end
1620	mov	T_ALIGNMENT, %g1
1621
1622.mmu_priv_exception:
1623	rdpr	%tstate, %g1
1624	btst	TSTATE_PRIV, %g1
1625	bnz,pn	%icc, 1f
1626	nop
1627	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1628	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1629	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1630	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1631	brz,pt	%g5, 1f
1632	nop
1633	ldn	[%g5 + P_UTRAP16], %g5
1634	brnz,pt	%g5, .setup_v9utrap
1635	nop
16361:
1637	mov	T_PRIV_INSTR, %g1
1638
1639.mmu_exception_end:
1640	CPU_INDEX(%g4, %g5)
1641	set	cpu_core, %g5
1642	sllx	%g4, CPU_CORE_SHIFT, %g4
1643	add	%g4, %g5, %g4
1644	lduh	[%g4 + CPUC_DTRACE_FLAGS], %g5
1645	andcc	%g5, CPU_DTRACE_NOFAULT, %g0
1646	bz	%xcc, .mmu_exception_tlb_chk
1647	or	%g5, CPU_DTRACE_BADADDR, %g5
1648	stuh	%g5, [%g4 + CPUC_DTRACE_FLAGS]
1649	done
1650
1651.mmu_exception_tlb_chk:
1652	GET_CPU_IMPL(%g5)			! check SFSR.FT to see if this
1653	cmp	%g5, PANTHER_IMPL		! is a TLB parity error. But
1654	bne	2f				! we only do this check while
1655	mov	1, %g4				! running on Panther CPUs
1656	sllx	%g4, PN_SFSR_PARITY_SHIFT, %g4	! since US-I/II use the same
1657	andcc	%g3, %g4, %g0			! bit for something else which
1658	bz	2f				! will be handled later.
1659	nop
1660.mmu_exception_is_tlb_parity:
1661	.weak itlb_parity_trap
1662	.weak dtlb_parity_trap
1663	set	itlb_parity_trap, %g4
1664	cmp	%g1, T_INSTR_EXCEPTION		! branch to the itlb or
1665	be	3f				! dtlb parity handler
1666	nop					! if this trap is due
1667	set	dtlb_parity_trap, %g4
1668	cmp	%g1, T_DATA_EXCEPTION		! to a IMMU exception
1669	be	3f				! or DMMU exception.
1670	nop
16712:
1672	sllx	%g3, 32, %g3
1673	or	%g3, %g1, %g3
1674	set	trap, %g1
1675	ba,pt	%xcc, sys_trap
1676	sub	%g0, 1, %g4
16773:
1678	jmp	%g4				! off to the appropriate
1679	nop					! TLB parity handler
1680
1681.fp_disabled:
1682	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1683	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1684#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1685	brz,a,pn %g1, 2f
1686	  nop
1687#endif
1688	rdpr	%tstate, %g4
1689	btst	TSTATE_PRIV, %g4
1690#ifdef SF_ERRATA_30 /* call causes fp-disabled */
1691	bnz,pn %icc, 2f
1692	  nop
1693#else
1694	bnz,a,pn %icc, ptl1_panic
1695	  mov	PTL1_BAD_FPTRAP, %g1
1696#endif
1697	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1698	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1699	brz,a,pt %g5, 2f
1700	  nop
1701	ldn	[%g5 + P_UTRAP7], %g5			! fp_disabled utrap?
1702	brz,a,pn %g5, 2f
1703	  nop
1704	btst	1, %sp
1705	bz,a,pt	%xcc, 1f				! 32 bit user program
1706	  nop
1707	ba,a,pt	%xcc, .setup_v9utrap			! 64 bit user program
1708	  nop
17091:
1710	ba,pt	%xcc, .setup_utrap
1711	  or	%g0, %g0, %g7
17122:
1713	set	fp_disabled, %g1
1714	ba,pt	%xcc, sys_trap
1715	  sub	%g0, 1, %g4
1716
1717.fp_ieee_exception:
1718	rdpr	%tstate, %g1
1719	btst	TSTATE_PRIV, %g1
1720	bnz,a,pn %icc, ptl1_panic
1721	  mov	PTL1_BAD_FPTRAP, %g1
1722	CPU_ADDR(%g1, %g4)				! load CPU struct addr
1723	stx	%fsr, [%g1 + CPU_TMP1]
1724	ldx	[%g1 + CPU_TMP1], %g2
1725	ldn	[%g1 + CPU_THREAD], %g1			! load thread pointer
1726	ldn	[%g1 + T_PROCP], %g1			! load proc pointer
1727	ldn	[%g1 + P_UTRAPS], %g5			! are there utraps?
1728	brz,a,pt %g5, 1f
1729	  nop
1730	ldn	[%g5 + P_UTRAP8], %g5
1731	brnz,a,pt %g5, .setup_v9utrap
1732	  nop
17331:
1734	set	_fp_ieee_exception, %g1
1735	ba,pt	%xcc, sys_trap
1736	  sub	%g0, 1, %g4
1737
1738/*
1739 * Register Inputs:
1740 *	%g5		user trap handler
1741 *	%g7		misaligned addr - for alignment traps only
1742 */
1743.setup_utrap:
1744	set	trap, %g1			! setup in case we go
1745	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1746	sub	%g0, 1, %g4			! the save instruction below
1747
1748	/*
1749	 * If the DTrace pid provider is single stepping a copied-out
1750	 * instruction, t->t_dtrace_step will be set. In that case we need
1751	 * to abort the single-stepping (since execution of the instruction
1752	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1753	 */
1754	save	%sp, -SA(MINFRAME32), %sp	! window for trap handler
1755	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1756	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1757	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1758	rdpr	%tnpc, %l2			! arg1 == tnpc
1759	brz,pt	%g2, 1f
1760	rdpr	%tpc, %l1			! arg0 == tpc
1761
1762	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1763	ldn	[%g1 + T_DTRACE_NPC], %l2	! arg1 = t->t_dtrace_npc (step)
1764	brz,pt	%g2, 1f
1765	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1766	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
17671:
1768	mov	%g7, %l3			! arg2 == misaligned address
1769
1770	rdpr	%tstate, %g1			! cwp for trap handler
1771	rdpr	%cwp, %g4
1772	bclr	TSTATE_CWP_MASK, %g1
1773	wrpr	%g1, %g4, %tstate
1774	wrpr	%g0, %g5, %tnpc			! trap handler address
1775	FAST_TRAP_DONE
1776	/* NOTREACHED */
1777
1778.check_v9utrap:
1779	rdpr	%tstate, %g1
1780	btst	TSTATE_PRIV, %g1
1781	bnz,a,pn %icc, 3f
1782	  nop
1783	CPU_ADDR(%g4, %g1)				! load CPU struct addr
1784	ldn	[%g4 + CPU_THREAD], %g5			! load thread pointer
1785	ldn	[%g5 + T_PROCP], %g5			! load proc pointer
1786	ldn	[%g5 + P_UTRAPS], %g5			! are there utraps?
1787
1788	cmp	%g3, T_SOFTWARE_TRAP
1789	bne,a,pt %icc, 1f
1790	  nop
1791
1792	brz,pt %g5, 3f			! if p_utraps == NULL goto trap()
1793	  rdpr	%tt, %g3		! delay - get actual hw trap type
1794
1795	sub	%g3, 254, %g1		! UT_TRAP_INSTRUCTION_16 = p_utraps[18]
1796	ba,pt	%icc, 2f
1797	  smul	%g1, CPTRSIZE, %g2
17981:
1799	brz,a,pt %g5, 3f		! if p_utraps == NULL goto trap()
1800	  nop
1801
1802	cmp	%g3, T_UNIMP_INSTR
1803	bne,a,pt %icc, 2f
1804	  nop
1805
1806	mov	1, %g1
1807	st	%g1, [%g4 + CPU_TL1_HDLR] ! set CPU_TL1_HDLR
1808	rdpr	%tpc, %g1		! ld trapping instruction using
1809	lduwa	[%g1]ASI_AIUP, %g1	! "AS IF USER" ASI which could fault
1810	st	%g0, [%g4 + CPU_TL1_HDLR] ! clr CPU_TL1_HDLR
1811
1812	sethi	%hi(0xc1c00000), %g4	! setup mask for illtrap instruction
1813	andcc	%g1, %g4, %g4		! and instruction with mask
1814	bnz,a,pt %icc, 3f		! if %g4 == zero, %g1 is an ILLTRAP
1815	  nop				! fall thru to setup
18162:
1817	ldn	[%g5 + %g2], %g5
1818	brnz,a,pt %g5, .setup_v9utrap
1819	  nop
18203:
1821	set	trap, %g1
1822	ba,pt	%xcc, sys_trap
1823	  sub	%g0, 1, %g4
1824	/* NOTREACHED */
1825
1826/*
1827 * Register Inputs:
1828 *	%g5		user trap handler
1829 */
1830.setup_v9utrap:
1831	set	trap, %g1			! setup in case we go
1832	mov	T_FLUSH_PCB, %g3		! through sys_trap on
1833	sub	%g0, 1, %g4			! the save instruction below
1834
1835	/*
1836	 * If the DTrace pid provider is single stepping a copied-out
1837	 * instruction, t->t_dtrace_step will be set. In that case we need
1838	 * to abort the single-stepping (since execution of the instruction
1839	 * was interrupted) and use the value of t->t_dtrace_npc as the %npc.
1840	 */
1841	save	%sp, -SA(MINFRAME64), %sp	! window for trap handler
1842	CPU_ADDR(%g1, %g4)			! load CPU struct addr
1843	ldn	[%g1 + CPU_THREAD], %g1		! load thread pointer
1844	ldub	[%g1 + T_DTRACE_STEP], %g2	! load t->t_dtrace_step
1845	rdpr	%tnpc, %l7			! arg1 == tnpc
1846	brz,pt	%g2, 1f
1847	rdpr	%tpc, %l6			! arg0 == tpc
1848
1849	ldub	[%g1 + T_DTRACE_AST], %g2	! load t->t_dtrace_ast
1850	ldn	[%g1 + T_DTRACE_NPC], %l7	! arg1 == t->t_dtrace_npc (step)
1851	brz,pt	%g2, 1f
1852	st	%g0, [%g1 + T_DTRACE_FT]	! zero all pid provider flags
1853	stub	%g2, [%g1 + T_ASTFLAG]		! aston(t) if t->t_dtrace_ast
18541:
1855	rdpr	%tstate, %g2			! cwp for trap handler
1856	rdpr	%cwp, %g4
1857	bclr	TSTATE_CWP_MASK, %g2
1858	wrpr	%g2, %g4, %tstate
1859
1860	ldn	[%g1 + T_PROCP], %g4		! load proc pointer
1861	ldn	[%g4 + P_AS], %g4		! load as pointer
1862	ldn	[%g4 + A_USERLIMIT], %g4	! load as userlimit
1863	cmp	%l7, %g4			! check for single-step set
1864	bne,pt	%xcc, 4f
1865	  nop
1866	ldn	[%g1 + T_LWP], %g1		! load klwp pointer
1867	ld	[%g1 + PCB_STEP], %g4		! load single-step flag
1868	cmp	%g4, STEP_ACTIVE		! step flags set in pcb?
1869	bne,pt	%icc, 4f
1870	  nop
1871	stn	%g5, [%g1 + PCB_TRACEPC]	! save trap handler addr in pcb
1872	mov	%l7, %g4			! on entry to precise user trap
1873	add	%l6, 4, %l7			! handler, %l6 == pc, %l7 == npc
1874						! at time of trap
1875	wrpr	%g0, %g4, %tnpc			! generate FLTBOUNDS,
1876						! %g4 == userlimit
1877	FAST_TRAP_DONE
1878	/* NOTREACHED */
18794:
1880	wrpr	%g0, %g5, %tnpc			! trap handler address
1881	FAST_TRAP_DONE_CHK_INTR
1882	/* NOTREACHED */
1883
1884.fp_exception:
1885	CPU_ADDR(%g1, %g4)
1886	stx	%fsr, [%g1 + CPU_TMP1]
1887	ldx	[%g1 + CPU_TMP1], %g2
1888
1889	/*
1890	 * Cheetah takes unfinished_FPop trap for certain range of operands
1891	 * to the "fitos" instruction. Instead of going through the slow
1892	 * software emulation path, we try to simulate the "fitos" instruction
1893	 * via "fitod" and "fdtos" provided the following conditions are met:
1894	 *
1895	 *	fpu_exists is set (if DEBUG)
1896	 *	not in privileged mode
1897	 *	ftt is unfinished_FPop
1898	 *	NXM IEEE trap is not enabled
1899	 *	instruction at %tpc is "fitos"
1900	 *
1901	 *  Usage:
1902	 *	%g1	per cpu address
1903	 *	%g2	%fsr
1904	 *	%g6	user instruction
1905	 *
1906	 * Note that we can take a memory access related trap while trying
1907	 * to fetch the user instruction. Therefore, we set CPU_TL1_HDLR
1908	 * flag to catch those traps and let the SFMMU code deal with page
1909	 * fault and data access exception.
1910	 */
1911#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
1912	sethi	%hi(fpu_exists), %g7
1913	ld	[%g7 + %lo(fpu_exists)], %g7
1914	brz,pn %g7, .fp_exception_cont
1915	  nop
1916#endif
1917	rdpr	%tstate, %g7			! branch if in privileged mode
1918	btst	TSTATE_PRIV, %g7
1919	bnz,pn	%xcc, .fp_exception_cont
1920	srl	%g2, FSR_FTT_SHIFT, %g7		! extract ftt from %fsr
1921	and	%g7, (FSR_FTT>>FSR_FTT_SHIFT), %g7
1922	cmp	%g7, FTT_UNFIN
1923	set	FSR_TEM_NX, %g5
1924	bne,pn	%xcc, .fp_exception_cont	! branch if NOT unfinished_FPop
1925	  andcc	%g2, %g5, %g0
1926	bne,pn	%xcc, .fp_exception_cont	! branch if FSR_TEM_NX enabled
1927	  rdpr	%tpc, %g5			! get faulting PC
1928
1929	or	%g0, 1, %g7
1930	st	%g7, [%g1 + CPU_TL1_HDLR]	! set tl1_hdlr flag
1931	lda	[%g5]ASI_USER, %g6		! get user's instruction
1932	st	%g0, [%g1 + CPU_TL1_HDLR]	! clear tl1_hdlr flag
1933
1934	set	FITOS_INSTR_MASK, %g7
1935	and	%g6, %g7, %g7
1936	set	FITOS_INSTR, %g5
1937	cmp	%g7, %g5
1938	bne,pn	%xcc, .fp_exception_cont	! branch if not FITOS_INSTR
1939	 nop
1940
1941	/*
1942	 * This is unfinished FPops trap for "fitos" instruction. We
1943	 * need to simulate "fitos" via "fitod" and "fdtos" instruction
1944	 * sequence.
1945	 *
1946	 * We need a temporary FP register to do the conversion. Since
1947	 * both source and destination operands for the "fitos" instruction
1948	 * have to be within %f0-%f31, we use an FP register from the upper
1949	 * half to guarantee that it won't collide with the source or the
1950	 * dest operand. However, we do have to save and restore its value.
1951	 *
1952	 * We use %d62 as a temporary FP register for the conversion and
1953	 * branch to appropriate instruction within the conversion tables
1954	 * based upon the rs2 and rd values.
1955	 */
1956
1957	std	%d62, [%g1 + CPU_TMP1]		! save original value
1958
1959	srl	%g6, FITOS_RS2_SHIFT, %g7
1960	and	%g7, FITOS_REG_MASK, %g7
1961	set	_fitos_fitod_table, %g4
1962	sllx	%g7, 2, %g7
1963	jmp	%g4 + %g7
1964	  ba,pt	%xcc, _fitos_fitod_done
1965	.empty
1966
1967_fitos_fitod_table:
1968	  fitod	%f0, %d62
1969	  fitod	%f1, %d62
1970	  fitod	%f2, %d62
1971	  fitod	%f3, %d62
1972	  fitod	%f4, %d62
1973	  fitod	%f5, %d62
1974	  fitod	%f6, %d62
1975	  fitod	%f7, %d62
1976	  fitod	%f8, %d62
1977	  fitod	%f9, %d62
1978	  fitod	%f10, %d62
1979	  fitod	%f11, %d62
1980	  fitod	%f12, %d62
1981	  fitod	%f13, %d62
1982	  fitod	%f14, %d62
1983	  fitod	%f15, %d62
1984	  fitod	%f16, %d62
1985	  fitod	%f17, %d62
1986	  fitod	%f18, %d62
1987	  fitod	%f19, %d62
1988	  fitod	%f20, %d62
1989	  fitod	%f21, %d62
1990	  fitod	%f22, %d62
1991	  fitod	%f23, %d62
1992	  fitod	%f24, %d62
1993	  fitod	%f25, %d62
1994	  fitod	%f26, %d62
1995	  fitod	%f27, %d62
1996	  fitod	%f28, %d62
1997	  fitod	%f29, %d62
1998	  fitod	%f30, %d62
1999	  fitod	%f31, %d62
2000_fitos_fitod_done:
2001
2002	/*
2003	 * Now convert data back into single precision
2004	 */
2005	srl	%g6, FITOS_RD_SHIFT, %g7
2006	and	%g7, FITOS_REG_MASK, %g7
2007	set	_fitos_fdtos_table, %g4
2008	sllx	%g7, 2, %g7
2009	jmp	%g4 + %g7
2010	  ba,pt	%xcc, _fitos_fdtos_done
2011	.empty
2012
2013_fitos_fdtos_table:
2014	  fdtos	%d62, %f0
2015	  fdtos	%d62, %f1
2016	  fdtos	%d62, %f2
2017	  fdtos	%d62, %f3
2018	  fdtos	%d62, %f4
2019	  fdtos	%d62, %f5
2020	  fdtos	%d62, %f6
2021	  fdtos	%d62, %f7
2022	  fdtos	%d62, %f8
2023	  fdtos	%d62, %f9
2024	  fdtos	%d62, %f10
2025	  fdtos	%d62, %f11
2026	  fdtos	%d62, %f12
2027	  fdtos	%d62, %f13
2028	  fdtos	%d62, %f14
2029	  fdtos	%d62, %f15
2030	  fdtos	%d62, %f16
2031	  fdtos	%d62, %f17
2032	  fdtos	%d62, %f18
2033	  fdtos	%d62, %f19
2034	  fdtos	%d62, %f20
2035	  fdtos	%d62, %f21
2036	  fdtos	%d62, %f22
2037	  fdtos	%d62, %f23
2038	  fdtos	%d62, %f24
2039	  fdtos	%d62, %f25
2040	  fdtos	%d62, %f26
2041	  fdtos	%d62, %f27
2042	  fdtos	%d62, %f28
2043	  fdtos	%d62, %f29
2044	  fdtos	%d62, %f30
2045	  fdtos	%d62, %f31
2046_fitos_fdtos_done:
2047
2048	ldd	[%g1 + CPU_TMP1], %d62		! restore %d62
2049
2050#if DEBUG
2051	/*
2052	 * Update FPop_unfinished trap kstat
2053	 */
2054	set	fpustat+FPUSTAT_UNFIN_KSTAT, %g7
2055	ldx	[%g7], %g5
20561:
2057	add	%g5, 1, %g6
2058
2059	casxa	[%g7] ASI_N, %g5, %g6
2060	cmp	%g5, %g6
2061	bne,a,pn %xcc, 1b
2062	  or	%g0, %g6, %g5
2063
2064	/*
2065	 * Update fpu_sim_fitos kstat
2066	 */
2067	set	fpuinfo+FPUINFO_FITOS_KSTAT, %g7
2068	ldx	[%g7], %g5
20691:
2070	add	%g5, 1, %g6
2071
2072	casxa	[%g7] ASI_N, %g5, %g6
2073	cmp	%g5, %g6
2074	bne,a,pn %xcc, 1b
2075	  or	%g0, %g6, %g5
2076#endif /* DEBUG */
2077
2078	FAST_TRAP_DONE
2079
2080.fp_exception_cont:
2081	/*
2082	 * Let _fp_exception deal with simulating FPop instruction.
2083	 * Note that we need to pass %fsr in %g2 (already read above).
2084	 */
2085
2086	set	_fp_exception, %g1
2087	ba,pt	%xcc, sys_trap
2088	sub	%g0, 1, %g4
2089
2090.clean_windows:
2091	set	trap, %g1
2092	mov	T_FLUSH_PCB, %g3
2093	sub	%g0, 1, %g4
2094	save
2095	flushw
2096	restore
2097	wrpr	%g0, %g0, %cleanwin	! no clean windows
2098
2099	CPU_ADDR(%g4, %g5)
2100	ldn	[%g4 + CPU_MPCB], %g4
2101	brz,a,pn %g4, 1f
2102	  nop
2103	ld	[%g4 + MPCB_WSTATE], %g5
2104	add	%g5, WSTATE_CLEAN_OFFSET, %g5
2105	wrpr	%g0, %g5, %wstate
21061:	FAST_TRAP_DONE
2107
2108/*
2109 * .spill_clean: clean the previous window, restore the wstate, and
2110 * "done".
2111 *
2112 * Entry: %g7 contains new wstate
2113 */
2114.spill_clean:
2115	sethi	%hi(nwin_minus_one), %g5
2116	ld	[%g5 + %lo(nwin_minus_one)], %g5 ! %g5 = nwin - 1
2117	rdpr	%cwp, %g6			! %g6 = %cwp
2118	deccc	%g6				! %g6--
2119	movneg	%xcc, %g5, %g6			! if (%g6<0) %g6 = nwin-1
2120	wrpr	%g6, %cwp
2121	TT_TRACE_L(trace_win)
2122	clr	%l0
2123	clr	%l1
2124	clr	%l2
2125	clr	%l3
2126	clr	%l4
2127	clr	%l5
2128	clr	%l6
2129	clr	%l7
2130	wrpr	%g0, %g7, %wstate
2131	saved
2132	retry			! restores correct %cwp
2133
2134.fix_alignment:
2135	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2136	ldn	[%g1 + CPU_THREAD], %g1	! load thread pointer
2137	ldn	[%g1 + T_PROCP], %g1
2138	mov	1, %g2
2139	stb	%g2, [%g1 + P_FIXALIGNMENT]
2140	FAST_TRAP_DONE
2141
2142#define	STDF_REG(REG, ADDR, TMP)		\
2143	sll	REG, 3, REG			;\
2144mark1:	set	start1, TMP			;\
2145	jmp	REG + TMP			;\
2146	  nop					;\
2147start1:	ba,pt	%xcc, done1			;\
2148	  std	%f0, [ADDR + CPU_TMP1]		;\
2149	ba,pt	%xcc, done1			;\
2150	  std	%f32, [ADDR + CPU_TMP1]		;\
2151	ba,pt	%xcc, done1			;\
2152	  std	%f2, [ADDR + CPU_TMP1]		;\
2153	ba,pt	%xcc, done1			;\
2154	  std	%f34, [ADDR + CPU_TMP1]		;\
2155	ba,pt	%xcc, done1			;\
2156	  std	%f4, [ADDR + CPU_TMP1]		;\
2157	ba,pt	%xcc, done1			;\
2158	  std	%f36, [ADDR + CPU_TMP1]		;\
2159	ba,pt	%xcc, done1			;\
2160	  std	%f6, [ADDR + CPU_TMP1]		;\
2161	ba,pt	%xcc, done1			;\
2162	  std	%f38, [ADDR + CPU_TMP1]		;\
2163	ba,pt	%xcc, done1			;\
2164	  std	%f8, [ADDR + CPU_TMP1]		;\
2165	ba,pt	%xcc, done1			;\
2166	  std	%f40, [ADDR + CPU_TMP1]		;\
2167	ba,pt	%xcc, done1			;\
2168	  std	%f10, [ADDR + CPU_TMP1]		;\
2169	ba,pt	%xcc, done1			;\
2170	  std	%f42, [ADDR + CPU_TMP1]		;\
2171	ba,pt	%xcc, done1			;\
2172	  std	%f12, [ADDR + CPU_TMP1]		;\
2173	ba,pt	%xcc, done1			;\
2174	  std	%f44, [ADDR + CPU_TMP1]		;\
2175	ba,pt	%xcc, done1			;\
2176	  std	%f14, [ADDR + CPU_TMP1]		;\
2177	ba,pt	%xcc, done1			;\
2178	  std	%f46, [ADDR + CPU_TMP1]		;\
2179	ba,pt	%xcc, done1			;\
2180	  std	%f16, [ADDR + CPU_TMP1]		;\
2181	ba,pt	%xcc, done1			;\
2182	  std	%f48, [ADDR + CPU_TMP1]		;\
2183	ba,pt	%xcc, done1			;\
2184	  std	%f18, [ADDR + CPU_TMP1]		;\
2185	ba,pt	%xcc, done1			;\
2186	  std	%f50, [ADDR + CPU_TMP1]		;\
2187	ba,pt	%xcc, done1			;\
2188	  std	%f20, [ADDR + CPU_TMP1]		;\
2189	ba,pt	%xcc, done1			;\
2190	  std	%f52, [ADDR + CPU_TMP1]		;\
2191	ba,pt	%xcc, done1			;\
2192	  std	%f22, [ADDR + CPU_TMP1]		;\
2193	ba,pt	%xcc, done1			;\
2194	  std	%f54, [ADDR + CPU_TMP1]		;\
2195	ba,pt	%xcc, done1			;\
2196	  std	%f24, [ADDR + CPU_TMP1]		;\
2197	ba,pt	%xcc, done1			;\
2198	  std	%f56, [ADDR + CPU_TMP1]		;\
2199	ba,pt	%xcc, done1			;\
2200	  std	%f26, [ADDR + CPU_TMP1]		;\
2201	ba,pt	%xcc, done1			;\
2202	  std	%f58, [ADDR + CPU_TMP1]		;\
2203	ba,pt	%xcc, done1			;\
2204	  std	%f28, [ADDR + CPU_TMP1]		;\
2205	ba,pt	%xcc, done1			;\
2206	  std	%f60, [ADDR + CPU_TMP1]		;\
2207	ba,pt	%xcc, done1			;\
2208	  std	%f30, [ADDR + CPU_TMP1]		;\
2209	ba,pt	%xcc, done1			;\
2210	  std	%f62, [ADDR + CPU_TMP1]		;\
2211done1:
2212
2213#define	LDDF_REG(REG, ADDR, TMP)		\
2214	sll	REG, 3, REG			;\
2215mark2:	set	start2, TMP			;\
2216	jmp	REG + TMP			;\
2217	  nop					;\
2218start2:	ba,pt	%xcc, done2			;\
2219	  ldd	[ADDR + CPU_TMP1], %f0		;\
2220	ba,pt	%xcc, done2			;\
2221	  ldd	[ADDR + CPU_TMP1], %f32		;\
2222	ba,pt	%xcc, done2			;\
2223	  ldd	[ADDR + CPU_TMP1], %f2		;\
2224	ba,pt	%xcc, done2			;\
2225	  ldd	[ADDR + CPU_TMP1], %f34		;\
2226	ba,pt	%xcc, done2			;\
2227	  ldd	[ADDR + CPU_TMP1], %f4		;\
2228	ba,pt	%xcc, done2			;\
2229	  ldd	[ADDR + CPU_TMP1], %f36		;\
2230	ba,pt	%xcc, done2			;\
2231	  ldd	[ADDR + CPU_TMP1], %f6		;\
2232	ba,pt	%xcc, done2			;\
2233	  ldd	[ADDR + CPU_TMP1], %f38		;\
2234	ba,pt	%xcc, done2			;\
2235	  ldd	[ADDR + CPU_TMP1], %f8		;\
2236	ba,pt	%xcc, done2			;\
2237	  ldd	[ADDR + CPU_TMP1], %f40		;\
2238	ba,pt	%xcc, done2			;\
2239	  ldd	[ADDR + CPU_TMP1], %f10		;\
2240	ba,pt	%xcc, done2			;\
2241	  ldd	[ADDR + CPU_TMP1], %f42		;\
2242	ba,pt	%xcc, done2			;\
2243	  ldd	[ADDR + CPU_TMP1], %f12		;\
2244	ba,pt	%xcc, done2			;\
2245	  ldd	[ADDR + CPU_TMP1], %f44		;\
2246	ba,pt	%xcc, done2			;\
2247	  ldd	[ADDR + CPU_TMP1], %f14		;\
2248	ba,pt	%xcc, done2			;\
2249	  ldd	[ADDR + CPU_TMP1], %f46		;\
2250	ba,pt	%xcc, done2			;\
2251	  ldd	[ADDR + CPU_TMP1], %f16		;\
2252	ba,pt	%xcc, done2			;\
2253	  ldd	[ADDR + CPU_TMP1], %f48		;\
2254	ba,pt	%xcc, done2			;\
2255	  ldd	[ADDR + CPU_TMP1], %f18		;\
2256	ba,pt	%xcc, done2			;\
2257	  ldd	[ADDR + CPU_TMP1], %f50		;\
2258	ba,pt	%xcc, done2			;\
2259	  ldd	[ADDR + CPU_TMP1], %f20		;\
2260	ba,pt	%xcc, done2			;\
2261	  ldd	[ADDR + CPU_TMP1], %f52		;\
2262	ba,pt	%xcc, done2			;\
2263	  ldd	[ADDR + CPU_TMP1], %f22		;\
2264	ba,pt	%xcc, done2			;\
2265	  ldd	[ADDR + CPU_TMP1], %f54		;\
2266	ba,pt	%xcc, done2			;\
2267	  ldd	[ADDR + CPU_TMP1], %f24		;\
2268	ba,pt	%xcc, done2			;\
2269	  ldd	[ADDR + CPU_TMP1], %f56		;\
2270	ba,pt	%xcc, done2			;\
2271	  ldd	[ADDR + CPU_TMP1], %f26		;\
2272	ba,pt	%xcc, done2			;\
2273	  ldd	[ADDR + CPU_TMP1], %f58		;\
2274	ba,pt	%xcc, done2			;\
2275	  ldd	[ADDR + CPU_TMP1], %f28		;\
2276	ba,pt	%xcc, done2			;\
2277	  ldd	[ADDR + CPU_TMP1], %f60		;\
2278	ba,pt	%xcc, done2			;\
2279	  ldd	[ADDR + CPU_TMP1], %f30		;\
2280	ba,pt	%xcc, done2			;\
2281	  ldd	[ADDR + CPU_TMP1], %f62		;\
2282done2:
2283
2284.lddf_exception_not_aligned:
2285	/*
2286	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2287	 */
2288	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2289
2290#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2291	sethi	%hi(fpu_exists), %g2		! check fpu_exists
2292	ld	[%g2 + %lo(fpu_exists)], %g2
2293	brz,a,pn %g2, 4f
2294	  nop
2295#endif
2296	CPU_ADDR(%g1, %g4)
2297	or	%g0, 1, %g4
2298	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2299
2300	rdpr	%tpc, %g2
2301	lda	[%g2]ASI_AIUP, %g6	! get the user's lddf instruction
2302	srl	%g6, 23, %g1		! using ldda or not?
2303	and	%g1, 1, %g1
2304	brz,a,pt %g1, 2f		! check for ldda instruction
2305	  nop
2306	srl	%g6, 13, %g1		! check immflag
2307	and	%g1, 1, %g1
2308	rdpr	%tstate, %g2		! %tstate in %g2
2309	brnz,a,pn %g1, 1f
2310	  srl	%g2, 31, %g1		! get asi from %tstate
2311	srl	%g6, 5, %g1		! get asi from instruction
2312	and	%g1, 0xFF, %g1		! imm_asi field
23131:
2314	cmp	%g1, ASI_P		! primary address space
2315	be,a,pt %icc, 2f
2316	  nop
2317	cmp	%g1, ASI_PNF		! primary no fault address space
2318	be,a,pt %icc, 2f
2319	  nop
2320	cmp	%g1, ASI_S		! secondary address space
2321	be,a,pt %icc, 2f
2322	  nop
2323	cmp	%g1, ASI_SNF		! secondary no fault address space
2324	bne,a,pn %icc, 3f
2325	  nop
23262:
2327	lduwa	[%g5]ASI_USER, %g7	! get first half of misaligned data
2328	add	%g5, 4, %g5		! increment misaligned data address
2329	lduwa	[%g5]ASI_USER, %g5	! get second half of misaligned data
2330
2331	sllx	%g7, 32, %g7
2332	or	%g5, %g7, %g5		! combine data
2333	CPU_ADDR(%g7, %g1)		! save data on a per-cpu basis
2334	stx	%g5, [%g7 + CPU_TMP1]	! save in cpu_tmp1
2335
2336	srl	%g6, 25, %g3		! %g6 has the instruction
2337	and	%g3, 0x1F, %g3		! %g3 has rd
2338	LDDF_REG(%g3, %g7, %g4)
2339
2340	CPU_ADDR(%g1, %g4)
2341	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2342	FAST_TRAP_DONE
23433:
2344	CPU_ADDR(%g1, %g4)
2345	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
23464:
2347	set	T_USER, %g3		! trap type in %g3
2348	or	%g3, T_LDDF_ALIGN, %g3
2349	mov	%g5, %g2		! misaligned vaddr in %g2
2350	set	fpu_trap, %g1		! goto C for the little and
2351	ba,pt	%xcc, sys_trap		! no fault little asi's
2352	  sub	%g0, 1, %g4
2353
2354.stdf_exception_not_aligned:
2355	/*
2356	 * Cheetah overwrites SFAR on a DTLB miss, hence read it now.
2357	 */
2358	ldxa	[MMU_SFAR]%asi, %g5	! misaligned vaddr in %g5
2359
2360#if defined(DEBUG) || defined(NEED_FPU_EXISTS)
2361	sethi	%hi(fpu_exists), %g7		! check fpu_exists
2362	ld	[%g7 + %lo(fpu_exists)], %g3
2363	brz,a,pn %g3, 4f
2364	  nop
2365#endif
2366	CPU_ADDR(%g1, %g4)
2367	or	%g0, 1, %g4
2368	st	%g4, [%g1 + CPU_TL1_HDLR] ! set tl1_hdlr flag
2369
2370	rdpr	%tpc, %g2
2371	lda	[%g2]ASI_AIUP, %g6	! get the user's stdf instruction
2372
2373	srl	%g6, 23, %g1		! using stda or not?
2374	and	%g1, 1, %g1
2375	brz,a,pt %g1, 2f		! check for stda instruction
2376	  nop
2377	srl	%g6, 13, %g1		! check immflag
2378	and	%g1, 1, %g1
2379	rdpr	%tstate, %g2		! %tstate in %g2
2380	brnz,a,pn %g1, 1f
2381	  srl	%g2, 31, %g1		! get asi from %tstate
2382	srl	%g6, 5, %g1		! get asi from instruction
2383	and	%g1, 0xFF, %g1		! imm_asi field
23841:
2385	cmp	%g1, ASI_P		! primary address space
2386	be,a,pt %icc, 2f
2387	  nop
2388	cmp	%g1, ASI_S		! secondary address space
2389	bne,a,pn %icc, 3f
2390	  nop
23912:
2392	srl	%g6, 25, %g6
2393	and	%g6, 0x1F, %g6		! %g6 has rd
2394	CPU_ADDR(%g7, %g1)
2395	STDF_REG(%g6, %g7, %g4)		! STDF_REG(REG, ADDR, TMP)
2396
2397	ldx	[%g7 + CPU_TMP1], %g6
2398	srlx	%g6, 32, %g7
2399	stuwa	%g7, [%g5]ASI_USER	! first half
2400	add	%g5, 4, %g5		! increment misaligned data address
2401	stuwa	%g6, [%g5]ASI_USER	! second half
2402
2403	CPU_ADDR(%g1, %g4)
2404	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
2405	FAST_TRAP_DONE
24063:
2407	CPU_ADDR(%g1, %g4)
2408	st	%g0, [%g1 + CPU_TL1_HDLR] ! clear tl1_hdlr flag
24094:
2410	set	T_USER, %g3		! trap type in %g3
2411	or	%g3, T_STDF_ALIGN, %g3
2412	mov	%g5, %g2		! misaligned vaddr in %g2
2413	set	fpu_trap, %g1		! goto C for the little and
2414	ba,pt	%xcc, sys_trap		! nofault little asi's
2415	  sub	%g0, 1, %g4
2416
2417#ifdef DEBUG_USER_TRAPTRACECTL
2418
2419.traptrace_freeze:
2420	mov	%l0, %g1 ; mov	%l1, %g2 ; mov	%l2, %g3 ; mov	%l4, %g4
2421	TT_TRACE_L(trace_win)
2422	mov	%g4, %l4 ; mov	%g3, %l2 ; mov	%g2, %l1 ; mov	%g1, %l0
2423	set	trap_freeze, %g1
2424	mov	1, %g2
2425	st	%g2, [%g1]
2426	FAST_TRAP_DONE
2427
2428.traptrace_unfreeze:
2429	set	trap_freeze, %g1
2430	st	%g0, [%g1]
2431	mov	%l0, %g1 ; mov	%l1, %g2 ; mov	%l2, %g3 ; mov	%l4, %g4
2432	TT_TRACE_L(trace_win)
2433	mov	%g4, %l4 ; mov	%g3, %l2 ; mov	%g2, %l1 ; mov	%g1, %l0
2434	FAST_TRAP_DONE
2435
2436#endif /* DEBUG_USER_TRAPTRACECTL */
2437
2438.getcc:
2439	CPU_ADDR(%g1, %g2)
2440	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2441	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2442	rdpr	%tstate, %g3			! get tstate
2443	srlx	%g3, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2444	set	PSR_ICC, %g2
2445	and	%o0, %g2, %o0			! mask out the rest
2446	srl	%o0, PSR_ICC_SHIFT, %o0		! right justify
2447	rdpr	%pstate, %o1
2448	wrpr	%o1, PSTATE_AG, %pstate		! get into normal globals
2449	mov	%o0, %g1			! move ccr to normal %g1
2450	wrpr	%g0, %o1, %pstate		! back into alternate globals
2451	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2452	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2453	FAST_TRAP_DONE
2454
2455.setcc:
2456	CPU_ADDR(%g1, %g2)
2457	stx	%o0, [%g1 + CPU_TMP1]		! save %o0
2458	stx	%o1, [%g1 + CPU_TMP2]		! save %o1
2459	rdpr	%pstate, %o0
2460	wrpr	%o0, PSTATE_AG, %pstate		! get into normal globals
2461	mov	%g1, %o1
2462	wrpr	%g0, %o0, %pstate		! back to alternates
2463	sll	%o1, PSR_ICC_SHIFT, %g2
2464	set	PSR_ICC, %g3
2465	and	%g2, %g3, %g2			! mask out rest
2466	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g2
2467	rdpr	%tstate, %g3			! get tstate
2468	srl	%g3, 0, %g3			! clear upper word
2469	or	%g3, %g2, %g3			! or in new bits
2470	wrpr	%g3, %tstate
2471	ldx	[%g1 + CPU_TMP1], %o0		! restore %o0
2472	ldx	[%g1 + CPU_TMP2], %o1		! restore %o1
2473	FAST_TRAP_DONE
2474
2475/*
2476 * getpsr(void)
2477 * Note that the xcc part of the ccr is not provided.
2478 * The V8 code shows why the V9 trap is not faster:
2479 * #define GETPSR_TRAP() \
2480 *      mov %psr, %i0; jmp %l2; rett %l2+4; nop;
2481 */
2482
2483	.type	.getpsr, #function
2484.getpsr:
2485	rdpr	%tstate, %g1			! get tstate
2486	srlx	%g1, PSR_TSTATE_CC_SHIFT, %o0	! shift ccr to V8 psr
2487	set	PSR_ICC, %g2
2488	and	%o0, %g2, %o0			! mask out the rest
2489
2490	rd	%fprs, %g1			! get fprs
2491	and	%g1, FPRS_FEF, %g2		! mask out dirty upper/lower
2492	sllx	%g2, PSR_FPRS_FEF_SHIFT, %g2	! shift fef to V8 psr.ef
2493	or	%o0, %g2, %o0			! or result into psr.ef
2494
2495	set	V9_PSR_IMPLVER, %g2		! SI assigned impl/ver: 0xef
2496	or	%o0, %g2, %o0			! or psr.impl/ver
2497	FAST_TRAP_DONE
2498	SET_SIZE(.getpsr)
2499
2500/*
2501 * setpsr(newpsr)
2502 * Note that there is no support for ccr.xcc in the V9 code.
2503 */
2504
2505	.type	.setpsr, #function
2506.setpsr:
2507	rdpr	%tstate, %g1			! get tstate
2508!	setx	TSTATE_V8_UBITS, %g2
2509	or 	%g0, CCR_ICC, %g3
2510	sllx	%g3, TSTATE_CCR_SHIFT, %g2
2511
2512	andn	%g1, %g2, %g1			! zero current user bits
2513	set	PSR_ICC, %g2
2514	and	%g2, %o0, %g2			! clear all but psr.icc bits
2515	sllx	%g2, PSR_TSTATE_CC_SHIFT, %g3	! shift to tstate.ccr.icc
2516	wrpr	%g1, %g3, %tstate		! write tstate
2517
2518	set	PSR_EF, %g2
2519	and	%g2, %o0, %g2			! clear all but fp enable bit
2520	srlx	%g2, PSR_FPRS_FEF_SHIFT, %g4	! shift ef to V9 fprs.fef
2521	wr	%g0, %g4, %fprs			! write fprs
2522
2523	CPU_ADDR(%g1, %g2)			! load CPU struct addr to %g1
2524	ldn	[%g1 + CPU_THREAD], %g2		! load thread pointer
2525	ldn	[%g2 + T_LWP], %g3		! load klwp pointer
2526	ldn	[%g3 + LWP_FPU], %g2		! get lwp_fpu pointer
2527	stuw	%g4, [%g2 + FPU_FPRS]		! write fef value to fpu_fprs
2528	srlx	%g4, 2, %g4			! shift fef value to bit 0
2529	stub	%g4, [%g2 + FPU_EN]		! write fef value to fpu_en
2530	FAST_TRAP_DONE
2531	SET_SIZE(.setpsr)
2532
2533/*
2534 * getlgrp
2535 * get home lgrpid on which the calling thread is currently executing.
2536 */
2537	.type	.getlgrp, #function
2538.getlgrp:
2539	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2540	ld	[%g1 + CPU_ID], %o0	! load cpu_id
2541	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2542	ldn	[%g2 + T_LPL], %g2	! load lpl pointer
2543	ld	[%g2 + LPL_LGRPID], %g1	! load lpl_lgrpid
2544	sra	%g1, 0, %o1
2545	FAST_TRAP_DONE
2546	SET_SIZE(.getlgrp)
2547
2548/*
2549 * Entry for old 4.x trap (trap 0).
2550 */
2551	ENTRY_NP(syscall_trap_4x)
2552	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2553	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2554	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2555	ld	[%g2 + PCB_TRAP0], %g2	! lwp->lwp_pcb.pcb_trap0addr
2556	brz,pn	%g2, 1f			! has it been set?
2557	st	%l0, [%g1 + CPU_TMP1]	! delay - save some locals
2558	st	%l1, [%g1 + CPU_TMP2]
2559	rdpr	%tnpc, %l1		! save old tnpc
2560	wrpr	%g0, %g2, %tnpc		! setup tnpc
2561
2562	rdpr	%pstate, %l0
2563	wrpr	%l0, PSTATE_AG, %pstate	! switch to normal globals
2564	mov	%l1, %g6		! pass tnpc to user code in %g6
2565	wrpr	%l0, %g0, %pstate	! switch back to alternate globals
2566
2567	! Note that %g1 still contains CPU struct addr
2568	ld	[%g1 + CPU_TMP2], %l1	! restore locals
2569	ld	[%g1 + CPU_TMP1], %l0
2570	FAST_TRAP_DONE_CHK_INTR
25711:
2572	mov	%g1, %l0
2573	st	%l1, [%g1 + CPU_TMP2]
2574	rdpr	%pstate, %l1
2575	wrpr	%l1, PSTATE_AG, %pstate
2576	!
2577	! check for old syscall mmap which is the only different one which
2578	! must be the same.  Others are handled in the compatibility library.
2579	!
2580	cmp	%g1, OSYS_mmap	! compare to old 4.x mmap
2581	movz	%icc, SYS_mmap, %g1
2582	wrpr	%g0, %l1, %pstate
2583	ld	[%l0 + CPU_TMP2], %l1	! restore locals
2584	ld	[%l0 + CPU_TMP1], %l0
2585	SYSCALL(syscall_trap32)
2586	SET_SIZE(syscall_trap_4x)
2587
2588/*
2589 * Handler for software trap 9.
2590 * Set trap0 emulation address for old 4.x system call trap.
2591 * XXX - this should be a system call.
2592 */
2593	ENTRY_NP(set_trap0_addr)
2594	CPU_ADDR(%g1, %g2)		! load CPU struct addr to %g1 using %g2
2595	ldn	[%g1 + CPU_THREAD], %g2	! load thread pointer
2596	ldn	[%g2 + T_LWP], %g2	! load klwp pointer
2597	st	%l0, [%g1 + CPU_TMP1]	! save some locals
2598	st	%l1, [%g1 + CPU_TMP2]
2599	rdpr	%pstate, %l0
2600	wrpr	%l0, PSTATE_AG, %pstate
2601	mov	%g1, %l1
2602	wrpr	%g0, %l0, %pstate
2603	andn	%l1, 3, %l1		! force alignment
2604	st	%l1, [%g2 + PCB_TRAP0]	! lwp->lwp_pcb.pcb_trap0addr
2605	ld	[%g1 + CPU_TMP1], %l0	! restore locals
2606	ld	[%g1 + CPU_TMP2], %l1
2607	FAST_TRAP_DONE
2608	SET_SIZE(set_trap0_addr)
2609
2610/*
2611 * mmu_trap_tl1
2612 * trap handler for unexpected mmu traps.
2613 * simply checks if the trap was a user lddf/stdf alignment trap, in which
2614 * case we go to fpu_trap or a user trap from the window handler, in which
2615 * case we go save the state on the pcb.  Otherwise, we go to ptl1_panic.
2616 */
2617	.type	mmu_trap_tl1, #function
2618mmu_trap_tl1:
2619#ifdef	TRAPTRACE
2620	TRACE_PTR(%g5, %g6)
2621	GET_TRACE_TICK(%g6)
2622	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2623	rdpr	%tl, %g6
2624	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2625	rdpr	%tt, %g6
2626	stha	%g6, [%g5 + TRAP_ENT_TT]%asi
2627	rdpr	%tstate, %g6
2628	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi
2629	stna	%sp, [%g5 + TRAP_ENT_SP]%asi
2630	stna	%g0, [%g5 + TRAP_ENT_TR]%asi
2631	rdpr	%tpc, %g6
2632	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2633	set	MMU_SFAR, %g6
2634	ldxa	[%g6]ASI_DMMU, %g6
2635	stxa	%g6, [%g5 + TRAP_ENT_F1]%asi
2636	CPU_PADDR(%g7, %g6);
2637	add	%g7, CPU_TL1_HDLR, %g7
2638	lda	[%g7]ASI_MEM, %g6
2639	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2640	set	0xdeadbeef, %g6
2641	stna	%g6, [%g5 + TRAP_ENT_F3]%asi
2642	stna	%g6, [%g5 + TRAP_ENT_F4]%asi
2643	TRACE_NEXT(%g5, %g6, %g7)
2644#endif /* TRAPTRACE */
2645
2646	GET_CPU_IMPL(%g5)
2647	cmp	%g5, PANTHER_IMPL
2648	bne	mmu_trap_tl1_4
2649	  nop
2650	rdpr	%tt, %g5
2651	cmp	%g5, T_DATA_EXCEPTION
2652	bne	mmu_trap_tl1_4
2653	  nop
2654	wr	%g0, ASI_DMMU, %asi
2655	ldxa	[MMU_SFSR]%asi, %g5
2656	mov	1, %g6
2657	sllx	%g6, PN_SFSR_PARITY_SHIFT, %g6
2658	andcc	%g5, %g6, %g0
2659	bz	mmu_trap_tl1_4
2660
2661	/*
2662	 * We are running on a Panther and have hit a DTLB parity error.
2663	 */
2664	ldxa	[MMU_TAG_ACCESS]%asi, %g2
2665	mov	%g5, %g3
2666	ba,pt	%xcc, .mmu_exception_is_tlb_parity
2667	mov	T_DATA_EXCEPTION, %g1
2668
2669mmu_trap_tl1_4:
2670	CPU_PADDR(%g7, %g6);
2671	add     %g7, CPU_TL1_HDLR, %g7		! %g7 = &cpu_m.tl1_hdlr (PA)
2672	/*
2673	 * AM is cleared on trap, so addresses are 64 bit
2674	 */
2675	lda     [%g7]ASI_MEM, %g6
2676	brz,a,pt %g6, 1f
2677	  nop
2678	/*
2679	 * We are going to update cpu_m.tl1_hdlr using physical address.
2680	 * Flush the D$ line, so that stale data won't be accessed later.
2681	 */
2682	CPU_ADDR(%g6, %g5)
2683	add     %g6, CPU_TL1_HDLR, %g6		! %g6 = &cpu_m.tl1_hdlr (VA)
2684	GET_CPU_IMPL(%g5)
2685	cmp	%g5, CHEETAH_IMPL
2686	bl,pt	%icc, 3f
2687	 cmp	%g5, SPITFIRE_IMPL
2688	stxa	%g0, [%g7]ASI_DC_INVAL
2689	membar	#Sync
2690	ba,pt	%xcc, 2f
2691	 nop
26923:
2693	bl,pt	%icc, 2f
2694	 sethi	%hi(dcache_line_mask), %g5
2695	ld	[%g5 + %lo(dcache_line_mask)], %g5
2696	and	%g6, %g5, %g5
2697	stxa	%g0, [%g5]ASI_DC_TAG
2698	membar	#Sync
26992:
2700	sta     %g0, [%g7]ASI_MEM
2701	SWITCH_GLOBALS				! back to mmu globals
2702	ba,a,pt	%xcc, sfmmu_mmu_trap		! handle page faults
27031:
2704	rdpr	%tt, %g5
2705	rdpr	%tl, %g7
2706	sub	%g7, 1, %g6
2707	wrpr	%g6, %tl
2708	rdpr	%tt, %g6
2709	wrpr	%g7, %tl
2710	and	%g6, WTRAP_TTMASK, %g6
2711	cmp	%g6, WTRAP_TYPE
2712	bne,a,pn %xcc, ptl1_panic
2713	mov	PTL1_BAD_MMUTRAP, %g1
2714	rdpr	%tpc, %g7
2715	/* tpc should be in the trap table */
2716	set	trap_table, %g6
2717	cmp	%g7, %g6
2718	blt,a,pn %xcc, ptl1_panic
2719	  mov	PTL1_BAD_MMUTRAP, %g1
2720	set	etrap_table, %g6
2721	cmp	%g7, %g6
2722	bge,a,pn %xcc, ptl1_panic
2723	  mov	PTL1_BAD_MMUTRAP, %g1
2724	cmp	%g5, T_ALIGNMENT
2725	move	%icc, MMU_SFAR, %g6
2726	movne	%icc, MMU_TAG_ACCESS, %g6
2727	ldxa	[%g6]ASI_DMMU, %g6
2728	andn	%g7, WTRAP_ALIGN, %g7	/* 128 byte aligned */
2729	add	%g7, WTRAP_FAULTOFF, %g7
2730	wrpr	%g0, %g7, %tnpc
2731	done
2732	SET_SIZE(mmu_trap_tl1)
2733
2734/*
2735 * Several traps use kmdb_trap and kmdb_trap_tl1 as their handlers.  These
2736 * traps are valid only when kmdb is loaded.  When the debugger is active,
2737 * the code below is rewritten to transfer control to the appropriate
2738 * debugger entry points.
2739 */
2740	.global	kmdb_trap
2741	.align	8
2742kmdb_trap:
2743	ba,a	trap_table0
2744	jmp	%g1 + 0
2745	nop
2746
2747	.global	kmdb_trap_tl1
2748	.align	8
2749kmdb_trap_tl1:
2750	ba,a	trap_table0
2751	jmp	%g1 + 0
2752	nop
2753
2754/*
2755 * This entry is copied from OBP's trap table during boot.
2756 */
2757	.global	obp_bpt
2758	.align	8
2759obp_bpt:
2760	NOT
2761
2762/*
2763 * if kernel, set PCONTEXT to 0 for debuggers
2764 * if user, clear nucleus page sizes
2765 */
2766	.global kctx_obp_bpt
2767kctx_obp_bpt:
2768	set	obp_bpt, %g2
27691:
2770	mov	MMU_PCONTEXT, %g1
2771	ldxa	[%g1]ASI_DMMU, %g1
2772	srlx	%g1, CTXREG_NEXT_SHIFT, %g3
2773	brz,pt	%g3, 3f			! nucleus pgsz is 0, no problem
2774	  sllx	%g3, CTXREG_NEXT_SHIFT, %g3
2775	set	CTXREG_CTX_MASK, %g4	! check Pcontext
2776	btst	%g4, %g1
2777	bz,a,pt	%xcc, 2f
2778	  clr	%g3			! kernel:  PCONTEXT=0
2779	xor	%g3, %g1, %g3		! user:	clr N_pgsz0/1 bits
27802:
2781	set	DEMAP_ALL_TYPE, %g1
2782	stxa	%g0, [%g1]ASI_DTLB_DEMAP
2783	stxa	%g0, [%g1]ASI_ITLB_DEMAP
2784	mov	MMU_PCONTEXT, %g1
2785	stxa	%g3, [%g1]ASI_DMMU
2786        membar  #Sync
2787	sethi	%hi(FLUSH_ADDR), %g1
2788	flush	%g1			! flush required by immu
27893:
2790	jmp	%g2
2791	  nop
2792
2793
2794#ifdef	TRAPTRACE
2795/*
2796 * TRAPTRACE support.
2797 * labels here are branched to with "rd %pc, %g7" in the delay slot.
2798 * Return is done by "jmp %g7 + 4".
2799 */
2800
2801trace_gen:
2802	TRACE_PTR(%g3, %g6)
2803	GET_TRACE_TICK(%g6)
2804	stxa	%g6, [%g3 + TRAP_ENT_TICK]%asi
2805	rdpr	%tl, %g6
2806	stha	%g6, [%g3 + TRAP_ENT_TL]%asi
2807	rdpr	%tt, %g6
2808	stha	%g6, [%g3 + TRAP_ENT_TT]%asi
2809	rdpr	%tstate, %g6
2810	stxa	%g6, [%g3 + TRAP_ENT_TSTATE]%asi
2811	stna	%sp, [%g3 + TRAP_ENT_SP]%asi
2812	rdpr	%tpc, %g6
2813	stna	%g6, [%g3 + TRAP_ENT_TPC]%asi
2814	TRACE_NEXT(%g3, %g4, %g5)
2815	jmp	%g7 + 4
2816	nop
2817
2818trace_win:
2819	TRACE_WIN_INFO(0, %l0, %l1, %l2)
2820	! Keep the locals as clean as possible, caller cleans %l4
2821	clr	%l2
2822	clr	%l1
2823	jmp	%l4 + 4
2824	  clr	%l0
2825
2826/*
2827 * Trace a tsb hit
2828 * g1 = tsbe pointer (in/clobbered)
2829 * g2 = tag access register (in)
2830 * g3 - g4 = scratch (clobbered)
2831 * g5 = tsbe data (in)
2832 * g6 = scratch (clobbered)
2833 * g7 = pc we jumped here from (in)
2834 */
2835
2836	! Do not disturb %g5, it will be used after the trace
2837	ALTENTRY(trace_tsbhit)
2838	TRACE_TSBHIT(0)
2839	jmp	%g7 + 4
2840	nop
2841
2842/*
2843 * Trace a TSB miss
2844 *
2845 * g1 = tsb8k pointer (in)
2846 * g2 = tag access register (in)
2847 * g3 = tsb4m pointer (in)
2848 * g4 = tsbe tag (in/clobbered)
2849 * g5 - g6 = scratch (clobbered)
2850 * g7 = pc we jumped here from (in)
2851 */
2852	.global	trace_tsbmiss
2853trace_tsbmiss:
2854	membar	#Sync
2855	sethi	%hi(FLUSH_ADDR), %g6
2856	flush	%g6
2857	TRACE_PTR(%g5, %g6)
2858	GET_TRACE_TICK(%g6)
2859	stxa	%g6, [%g5 + TRAP_ENT_TICK]%asi
2860	stxa	%g2, [%g5 + TRAP_ENT_SP]%asi		! tag access
2861	stxa	%g4, [%g5 + TRAP_ENT_F1]%asi		! tsb tag
2862	rdpr	%tnpc, %g6
2863	stxa	%g6, [%g5 + TRAP_ENT_F2]%asi
2864	stna	%g1, [%g5 + TRAP_ENT_F3]%asi		! tsb8k pointer
2865	srlx	%g1, 32, %g6
2866	stna	%g6, [%g5 + TRAP_ENT_F4]%asi		! huh?
2867	rdpr	%tpc, %g6
2868	stna	%g6, [%g5 + TRAP_ENT_TPC]%asi
2869	rdpr	%tl, %g6
2870	stha	%g6, [%g5 + TRAP_ENT_TL]%asi
2871	rdpr	%tt, %g6
2872	or	%g6, TT_MMU_MISS, %g4
2873	stha	%g4, [%g5 + TRAP_ENT_TT]%asi
2874	cmp	%g6, FAST_IMMU_MISS_TT
2875	be,a	%icc, 1f
2876	  ldxa	[%g0]ASI_IMMU, %g6
2877	ldxa	[%g0]ASI_DMMU, %g6
28781:	stxa	%g6, [%g5 + TRAP_ENT_TSTATE]%asi	! tag target
2879	stxa	%g3, [%g5 + TRAP_ENT_TR]%asi		! tsb4m pointer
2880	TRACE_NEXT(%g5, %g4, %g6)
2881	jmp	%g7 + 4
2882	nop
2883
2884/*
2885 * g2 = tag access register (in)
2886 * g3 = ctx number (in)
2887 */
2888trace_dataprot:
2889	membar	#Sync
2890	sethi	%hi(FLUSH_ADDR), %g6
2891	flush	%g6
2892	TRACE_PTR(%g1, %g6)
2893	GET_TRACE_TICK(%g6)
2894	stxa	%g6, [%g1 + TRAP_ENT_TICK]%asi
2895	rdpr	%tpc, %g6
2896	stna	%g6, [%g1 + TRAP_ENT_TPC]%asi
2897	rdpr	%tstate, %g6
2898	stxa	%g6, [%g1 + TRAP_ENT_TSTATE]%asi
2899	stxa	%g2, [%g1 + TRAP_ENT_SP]%asi		! tag access reg
2900	stxa	%g0, [%g1 + TRAP_ENT_TR]%asi
2901	stxa	%g0, [%g1 + TRAP_ENT_F1]%asi
2902	stxa	%g0, [%g1 + TRAP_ENT_F2]%asi
2903	stxa	%g0, [%g1 + TRAP_ENT_F3]%asi
2904	stxa	%g0, [%g1 + TRAP_ENT_F4]%asi
2905	rdpr	%tl, %g6
2906	stha	%g6, [%g1 + TRAP_ENT_TL]%asi
2907	rdpr	%tt, %g6
2908	stha	%g6, [%g1 + TRAP_ENT_TT]%asi
2909	TRACE_NEXT(%g1, %g4, %g5)
2910	jmp	%g7 + 4
2911	nop
2912
2913#endif /* TRAPTRACE */
2914
2915/*
2916 * expects offset into tsbmiss area in %g1 and return pc in %g7
2917 */
2918stat_mmu:
2919	CPU_INDEX(%g5, %g6)
2920	sethi	%hi(tsbmiss_area), %g6
2921	sllx	%g5, TSBMISS_SHIFT, %g5
2922	or	%g6, %lo(tsbmiss_area), %g6
2923	add	%g6, %g5, %g6		/* g6 = tsbmiss area */
2924	ld	[%g6 + %g1], %g5
2925	add	%g5, 1, %g5
2926	jmp	%g7 + 4
2927	st	%g5, [%g6 + %g1]
2928
2929
2930/*
2931 * fast_trap_done, fast_trap_done_chk_intr:
2932 *
2933 * Due to the design of UltraSPARC pipeline, pending interrupts are not
2934 * taken immediately after a RETRY or DONE instruction which causes IE to
2935 * go from 0 to 1. Instead, the instruction at %tpc or %tnpc is allowed
2936 * to execute first before taking any interrupts. If that instruction
2937 * results in other traps, and if the corresponding trap handler runs
2938 * entirely at TL=1 with interrupts disabled, then pending interrupts
2939 * won't be taken until after yet another instruction following the %tpc
2940 * or %tnpc.
2941 *
2942 * A malicious user program can use this feature to block out interrupts
2943 * for extended durations, which can result in send_mondo_timeout kernel
2944 * panic.
2945 *
2946 * This problem is addressed by servicing any pending interrupts via
2947 * sys_trap before returning back to the user mode from a fast trap
2948 * handler. The "done" instruction within a fast trap handler, which
2949 * runs entirely at TL=1 with interrupts disabled, is replaced with the
2950 * FAST_TRAP_DONE macro, which branches control to this fast_trap_done
2951 * entry point.
2952 *
2953 * We check for any pending interrupts here and force a sys_trap to
2954 * service those interrupts, if any. To minimize overhead, pending
2955 * interrupts are checked if the %tpc happens to be at 16K boundary,
2956 * which allows a malicious program to execute at most 4K consecutive
2957 * instructions before we service any pending interrupts. If a worst
2958 * case fast trap handler takes about 2 usec, then interrupts will be
2959 * blocked for at most 8 msec, less than a clock tick.
2960 *
2961 * For the cases where we don't know if the %tpc will cross a 16K
2962 * boundary, we can't use the above optimization and always process
2963 * any pending interrupts via fast_frap_done_chk_intr entry point.
2964 *
2965 * Entry Conditions:
2966 * 	%pstate		am:0 priv:1 ie:0
2967 * 			globals are AG (not normal globals)
2968 */
2969
2970	.global	fast_trap_done, fast_trap_done_chk_intr
2971fast_trap_done:
2972	rdpr	%tpc, %g5
2973	sethi	%hi(0xffffc000), %g6	! 1's complement of 0x3fff
2974	andncc	%g5, %g6, %g0		! check lower 14 bits of %tpc
2975	bz,a,pn	%icc, 1f		! branch if zero (lower 32 bits only)
2976	  ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2977	done
2978
2979	ALTENTRY(fast_trap_done_check_interrupts)
2980fast_trap_done_chk_intr:
2981	ldxa	[%g0]ASI_INTR_RECEIVE_STATUS, %g5
2982
29831:	rd	SOFTINT, %g6
2984	and	%g5, IRSR_BUSY, %g5
2985	orcc	%g5, %g6, %g0
2986	bnz,pn	%xcc, 2f		! branch if any pending intr
2987	nop
2988	done
2989
29902:
2991	/*
2992	 * We get here if there are any pending interrupts.
2993	 * Adjust %tpc/%tnpc as we'll be resuming via "retry"
2994	 * instruction.
2995	 */
2996	rdpr	%tnpc, %g5
2997	wrpr	%g0, %g5, %tpc
2998	add	%g5, 4, %g5
2999	wrpr	%g0, %g5, %tnpc
3000
3001	/*
3002	 * Force a dummy sys_trap call so that interrupts can be serviced.
3003	 */
3004	set	fast_trap_dummy_call, %g1
3005	ba,pt	%xcc, sys_trap
3006	  mov	-1, %g4
3007
3008fast_trap_dummy_call:
3009	retl
3010	nop
3011
3012#ifdef	DEBUG
3013/*
3014 * Currently we only support syscall interposition for branded zones on
3015 * DEBUG kernels.  The only brand that makes use of this functionality is
3016 * the fake Solaris 10 brand.  Since this brand is only used for exercising
3017 * the framework, we don't want this overhead incurred on production
3018 * systems.
3019 */
3020#define	BRAND_CALLBACK(callback_id)					    \
3021	CPU_ADDR(%g1, %g2)		/* load CPU struct addr to %g1	*/ ;\
3022	ldn	[%g1 + CPU_THREAD], %g2	/* load thread pointer		*/ ;\
3023	ldn	[%g2 + T_PROCP], %g2	/* get proc pointer		*/ ;\
3024	ldn	[%g2 + P_BRAND], %g2	/* get brand pointer		*/ ;\
3025	brz	%g2, 1f			/* No brand?  No callback. 	*/ ;\
3026	nop 								   ;\
3027	ldn	[%g2 + B_MACHOPS], %g2	/* get machops list		*/ ;\
3028	ldn	[%g2 + (callback_id << 3)], %g2 			   ;\
3029	brz	%g2, 1f							   ;\
3030	/*								    \
3031	 * This isn't pretty.  We want a low-latency way for the callback   \
3032	 * routine to decline to do anything.  We just pass in an address   \
3033	 * the routine can directly jmp back to, pretending that nothing    \
3034	 * has happened.						    \
3035	 */								    \
3036	mov	%pc, %g1						   ;\
3037	add	%g1, 16, %g1						   ;\
3038	jmp	%g2							   ;\
3039	nop								   ;\
30401:
3041
3042	ENTRY_NP(syscall_wrapper32)
3043	BRAND_CALLBACK(BRAND_CB_SYSCALL32)
3044	SYSCALL(syscall_trap32)
3045	SET_SIZE(syscall_wrapper32)
3046
3047	ENTRY_NP(syscall_wrapper)
3048	BRAND_CALLBACK(BRAND_CB_SYSCALL)
3049	SYSCALL(syscall_trap)
3050	SET_SIZE(syscall_wrapper)
3051
3052#endif	/* DEBUG */
3053
3054#endif	/* lint */
3055