xref: /titanic_52/usr/src/uts/intel/ia32/ml/i86_subr.s (revision 84ab085a13f931bc78e7415e7ce921dbaa14fcb3)
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, Version 1.0 only
6 * (the "License").  You may not use this file except in compliance
7 * with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or http://www.opensolaris.org/os/licensing.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22/*
23 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 *  Copyright (c) 1990, 1991 UNIX System Laboratories, Inc.
29 *  Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T
30 *    All Rights Reserved
31 */
32
33#pragma ident	"%Z%%M%	%I%	%E% SMI"
34
35/*
36 * General assembly language routines.
37 * It is the intent of this file to contain routines that are
38 * independent of the specific kernel architecture, and those that are
39 * common across kernel architectures.
40 * As architectures diverge, and implementations of specific
41 * architecture-dependent routines change, the routines should be moved
42 * from this file into the respective ../`arch -k`/subr.s file.
43 */
44
45#include <sys/asm_linkage.h>
46#include <sys/asm_misc.h>
47#include <sys/panic.h>
48#include <sys/ontrap.h>
49#include <sys/regset.h>
50#include <sys/privregs.h>
51#include <sys/reboot.h>
52#include <sys/psw.h>
53#include <sys/x86_archext.h>
54
55#if defined(__lint)
56#include <sys/types.h>
57#include <sys/systm.h>
58#include <sys/thread.h>
59#include <sys/archsystm.h>
60#include <sys/byteorder.h>
61#include <sys/dtrace.h>
62#else	/* __lint */
63#include "assym.h"
64#endif	/* __lint */
65#include <sys/dditypes.h>
66
67/*
68 * on_fault()
69 * Catch lofault faults. Like setjmp except it returns one
70 * if code following causes uncorrectable fault. Turned off
71 * by calling no_fault().
72 */
73
74#if defined(__lint)
75
76/* ARGSUSED */
77int
78on_fault(label_t *ljb)
79{ return (0); }
80
81void
82no_fault(void)
83{}
84
85#else	/* __lint */
86
87#if defined(__amd64)
88
89	ENTRY(on_fault)
90	movq	%gs:CPU_THREAD, %rsi
91	leaq	catch_fault(%rip), %rdx
92	movq	%rdi, T_ONFAULT(%rsi)		/* jumpbuf in t_onfault */
93	movq	%rdx, T_LOFAULT(%rsi)		/* catch_fault in t_lofault */
94	jmp	setjmp				/* let setjmp do the rest */
95
96catch_fault:
97	movq	%gs:CPU_THREAD, %rsi
98	movq	T_ONFAULT(%rsi), %rdi		/* address of save area */
99	xorl	%eax, %eax
100	movq	%rax, T_ONFAULT(%rsi)		/* turn off onfault */
101	movq	%rax, T_LOFAULT(%rsi)		/* turn off lofault */
102	jmp	longjmp				/* let longjmp do the rest */
103	SET_SIZE(on_fault)
104
105	ENTRY(no_fault)
106	movq	%gs:CPU_THREAD, %rsi
107	xorl	%eax, %eax
108	movq	%rax, T_ONFAULT(%rsi)		/* turn off onfault */
109	movq	%rax, T_LOFAULT(%rsi)		/* turn off lofault */
110	ret
111	SET_SIZE(no_fault)
112
113#elif defined(__i386)
114
115	ENTRY(on_fault)
116	movl	%gs:CPU_THREAD, %edx
117	movl	4(%esp), %eax			/* jumpbuf address */
118	leal	catch_fault, %ecx
119	movl	%eax, T_ONFAULT(%edx)		/* jumpbuf in t_onfault */
120	movl	%ecx, T_LOFAULT(%edx)		/* catch_fault in t_lofault */
121	jmp	setjmp				/* let setjmp do the rest */
122
123catch_fault:
124	movl	%gs:CPU_THREAD, %edx
125	xorl	%eax, %eax
126	movl	T_ONFAULT(%edx), %ecx		/* address of save area */
127	movl	%eax, T_ONFAULT(%edx)		/* turn off onfault */
128	movl	%eax, T_LOFAULT(%edx)		/* turn off lofault */
129	pushl	%ecx
130	call	longjmp				/* let longjmp do the rest */
131	SET_SIZE(on_fault)
132
133	ENTRY(no_fault)
134	movl	%gs:CPU_THREAD, %edx
135	xorl	%eax, %eax
136	movl	%eax, T_ONFAULT(%edx)		/* turn off onfault */
137	movl	%eax, T_LOFAULT(%edx)		/* turn off lofault */
138	ret
139	SET_SIZE(no_fault)
140
141#endif	/* __i386 */
142#endif	/* __lint */
143
144/*
145 * Default trampoline code for on_trap() (see <sys/ontrap.h>).  We just
146 * do a longjmp(&curthread->t_ontrap->ot_jmpbuf) if this is ever called.
147 */
148
149#if defined(lint)
150
151void
152on_trap_trampoline(void)
153{}
154
155#else	/* __lint */
156
157#if defined(__amd64)
158
159	ENTRY(on_trap_trampoline)
160	movq	%gs:CPU_THREAD, %rsi
161	movq	T_ONTRAP(%rsi), %rdi
162	addq	$OT_JMPBUF, %rdi
163	jmp	longjmp
164	SET_SIZE(on_trap_trampoline)
165
166#elif defined(__i386)
167
168	ENTRY(on_trap_trampoline)
169	movl	%gs:CPU_THREAD, %eax
170	movl	T_ONTRAP(%eax), %eax
171	addl	$OT_JMPBUF, %eax
172	pushl	%eax
173	call	longjmp
174	SET_SIZE(on_trap_trampoline)
175
176#endif	/* __i386 */
177#endif	/* __lint */
178
179/*
180 * Push a new element on to the t_ontrap stack.  Refer to <sys/ontrap.h> for
181 * more information about the on_trap() mechanism.  If the on_trap_data is the
182 * same as the topmost stack element, we just modify that element.
183 */
184#if defined(lint)
185
186/*ARGSUSED*/
187int
188on_trap(on_trap_data_t *otp, uint_t prot)
189{ return (0); }
190
191#else	/* __lint */
192
193#if defined(__amd64)
194
195	ENTRY(on_trap)
196	movw	%si, OT_PROT(%rdi)		/* ot_prot = prot */
197	movw	$0, OT_TRAP(%rdi)		/* ot_trap = 0 */
198	leaq	on_trap_trampoline(%rip), %rdx	/* rdx = &on_trap_trampoline */
199	movq	%rdx, OT_TRAMPOLINE(%rdi)	/* ot_trampoline = rdx */
200	xorl	%ecx, %ecx
201	movq	%rcx, OT_HANDLE(%rdi)		/* ot_handle = NULL */
202	movq	%rcx, OT_PAD1(%rdi)		/* ot_pad1 = NULL */
203	movq	%gs:CPU_THREAD, %rdx		/* rdx = curthread */
204	movq	T_ONTRAP(%rdx), %rcx		/* rcx = curthread->t_ontrap */
205	cmpq	%rdi, %rcx			/* if (otp == %rcx)	*/
206	je	0f				/*	don't modify t_ontrap */
207
208	movq	%rcx, OT_PREV(%rdi)		/* ot_prev = t_ontrap */
209	movq	%rdi, T_ONTRAP(%rdx)		/* curthread->t_ontrap = otp */
210
2110:	addq	$OT_JMPBUF, %rdi		/* &ot_jmpbuf */
212	jmp	setjmp
213	SET_SIZE(on_trap)
214
215#elif defined(__i386)
216
217	ENTRY(on_trap)
218	movl	4(%esp), %eax			/* %eax = otp */
219	movl	8(%esp), %edx			/* %edx = prot */
220
221	movw	%dx, OT_PROT(%eax)		/* ot_prot = prot */
222	movw	$0, OT_TRAP(%eax)		/* ot_trap = 0 */
223	leal	on_trap_trampoline, %edx	/* %edx = &on_trap_trampoline */
224	movl	%edx, OT_TRAMPOLINE(%eax)	/* ot_trampoline = %edx */
225	movl	$0, OT_HANDLE(%eax)		/* ot_handle = NULL */
226	movl	$0, OT_PAD1(%eax)		/* ot_pad1 = NULL */
227	movl	%gs:CPU_THREAD, %edx		/* %edx = curthread */
228	movl	T_ONTRAP(%edx), %ecx		/* %ecx = curthread->t_ontrap */
229	cmpl	%eax, %ecx			/* if (otp == %ecx) */
230	je	0f				/*    don't modify t_ontrap */
231
232	movl	%ecx, OT_PREV(%eax)		/* ot_prev = t_ontrap */
233	movl	%eax, T_ONTRAP(%edx)		/* curthread->t_ontrap = otp */
234
2350:	addl	$OT_JMPBUF, %eax		/* %eax = &ot_jmpbuf */
236	movl	%eax, 4(%esp)			/* put %eax back on the stack */
237	jmp	setjmp				/* let setjmp do the rest */
238	SET_SIZE(on_trap)
239
240#endif	/* __i386 */
241#endif	/* __lint */
242
243/*
244 * Setjmp and longjmp implement non-local gotos using state vectors
245 * type label_t.
246 */
247
248#if defined(__lint)
249
250/* ARGSUSED */
251int
252setjmp(label_t *lp)
253{ return (0); }
254
255/* ARGSUSED */
256void
257longjmp(label_t *lp)
258{}
259
260#else	/* __lint */
261
262#if LABEL_PC != 0
263#error LABEL_PC MUST be defined as 0 for setjmp/longjmp to work as coded
264#endif	/* LABEL_PC != 0 */
265
266#if defined(__amd64)
267
268	ENTRY(setjmp)
269	movq	%rsp, LABEL_SP(%rdi)
270	movq	%rbp, LABEL_RBP(%rdi)
271	movq	%rbx, LABEL_RBX(%rdi)
272	movq	%r12, LABEL_R12(%rdi)
273	movq	%r13, LABEL_R13(%rdi)
274	movq	%r14, LABEL_R14(%rdi)
275	movq	%r15, LABEL_R15(%rdi)
276	movq	(%rsp), %rdx		/* return address */
277	movq	%rdx, (%rdi)		/* LABEL_PC is 0 */
278	xorl	%eax, %eax		/* return 0 */
279	ret
280	SET_SIZE(setjmp)
281
282	ENTRY(longjmp)
283	movq	LABEL_SP(%rdi), %rsp
284	movq	LABEL_RBP(%rdi), %rbp
285	movq	LABEL_RBX(%rdi), %rbx
286	movq	LABEL_R12(%rdi), %r12
287	movq	LABEL_R13(%rdi), %r13
288	movq	LABEL_R14(%rdi), %r14
289	movq	LABEL_R15(%rdi), %r15
290	movq	(%rdi), %rdx		/* return address; LABEL_PC is 0 */
291	movq	%rdx, (%rsp)
292	xorl	%eax, %eax
293	incl	%eax			/* return 1 */
294	ret
295	SET_SIZE(longjmp)
296
297#elif defined(__i386)
298
299	ENTRY(setjmp)
300	movl	4(%esp), %edx		/* address of save area */
301	movl	%ebp, LABEL_EBP(%edx)
302	movl	%ebx, LABEL_EBX(%edx)
303	movl	%esi, LABEL_ESI(%edx)
304	movl	%edi, LABEL_EDI(%edx)
305	movl	%esp, 4(%edx)
306	movl	(%esp), %ecx		/* %eip (return address) */
307	movl	%ecx, (%edx)		/* LABEL_PC is 0 */
308	subl	%eax, %eax		/* return 0 */
309	ret
310	SET_SIZE(setjmp)
311
312	ENTRY(longjmp)
313	movl	4(%esp), %edx		/* address of save area */
314	movl	LABEL_EBP(%edx), %ebp
315	movl	LABEL_EBX(%edx), %ebx
316	movl	LABEL_ESI(%edx), %esi
317	movl	LABEL_EDI(%edx), %edi
318	movl	4(%edx), %esp
319	movl	(%edx), %ecx		/* %eip (return addr); LABEL_PC is 0 */
320	movl	$1, %eax
321	addl	$4, %esp		/* pop ret adr */
322	jmp	*%ecx			/* indirect */
323	SET_SIZE(longjmp)
324
325#endif	/* __i386 */
326#endif	/* __lint */
327
328/*
329 * if a() calls b() calls caller(),
330 * caller() returns return address in a().
331 * (Note: We assume a() and b() are C routines which do the normal entry/exit
332 *  sequence.)
333 */
334
335#if defined(__lint)
336
337caddr_t
338caller(void)
339{ return (0); }
340
341#else	/* __lint */
342
343#if defined(__amd64)
344
345	ENTRY(caller)
346	movq	8(%rbp), %rax		/* b()'s return pc, in a() */
347	ret
348	SET_SIZE(caller)
349
350#elif defined(__i386)
351
352	ENTRY(caller)
353	movl	4(%ebp), %eax		/* b()'s return pc, in a() */
354	ret
355	SET_SIZE(caller)
356
357#endif	/* __i386 */
358#endif	/* __lint */
359
360/*
361 * if a() calls callee(), callee() returns the
362 * return address in a();
363 */
364
365#if defined(__lint)
366
367caddr_t
368callee(void)
369{ return (0); }
370
371#else	/* __lint */
372
373#if defined(__amd64)
374
375	ENTRY(callee)
376	movq	(%rsp), %rax		/* callee()'s return pc, in a() */
377	ret
378	SET_SIZE(callee)
379
380#elif defined(__i386)
381
382	ENTRY(callee)
383	movl	(%esp), %eax		/* callee()'s return pc, in a() */
384	ret
385	SET_SIZE(callee)
386
387#endif	/* __i386 */
388#endif	/* __lint */
389
390/*
391 * return the current frame pointer
392 */
393
394#if defined(__lint)
395
396greg_t
397getfp(void)
398{ return (0); }
399
400#else	/* __lint */
401
402#if defined(__amd64)
403
404	ENTRY(getfp)
405	movq	%rbp, %rax
406	ret
407	SET_SIZE(getfp)
408
409#elif defined(__i386)
410
411	ENTRY(getfp)
412	movl	%ebp, %eax
413	ret
414	SET_SIZE(getfp)
415
416#endif	/* __i386 */
417#endif	/* __lint */
418
419/*
420 * Invalidate a single page table entry in the TLB
421 */
422
423#if defined(__lint)
424
425/* ARGSUSED */
426void
427mmu_tlbflush_entry(caddr_t m)
428{}
429
430#else	/* __lint */
431
432#if defined(__amd64)
433
434	ENTRY(mmu_tlbflush_entry)
435	invlpg	(%rdi)
436	ret
437	SET_SIZE(mmu_tlbflush_entry)
438
439#elif defined(__i386)
440
441	ENTRY(mmu_tlbflush_entry)
442	movl	4(%esp), %eax
443	invlpg	(%eax)
444	ret
445	SET_SIZE(mmu_tlbflush_entry)
446
447#endif	/* __i386 */
448#endif	/* __lint */
449
450
451/*
452 * Get/Set the value of various control registers
453 */
454
455#if defined(__lint)
456
457ulong_t
458getcr0(void)
459{ return (0); }
460
461/* ARGSUSED */
462void
463setcr0(ulong_t value)
464{}
465
466ulong_t
467getcr2(void)
468{ return (0); }
469
470ulong_t
471getcr3(void)
472{ return (0); }
473
474/* ARGSUSED */
475void
476setcr3(ulong_t val)
477{}
478
479void
480reload_cr3(void)
481{}
482
483ulong_t
484getcr4(void)
485{ return (0); }
486
487/* ARGSUSED */
488void
489setcr4(ulong_t val)
490{}
491
492#if defined(__amd64)
493
494ulong_t
495getcr8(void)
496{ return (0); }
497
498/* ARGSUSED */
499void
500setcr8(ulong_t val)
501{}
502
503#endif	/* __amd64 */
504
505#else	/* __lint */
506
507#if defined(__amd64)
508
509	ENTRY(getcr0)
510	movq	%cr0, %rax
511	ret
512	SET_SIZE(getcr0)
513
514	ENTRY(setcr0)
515	movq	%rdi, %cr0
516	ret
517	SET_SIZE(setcr0)
518
519	ENTRY(getcr2)
520	movq	%cr2, %rax
521	ret
522	SET_SIZE(getcr2)
523
524	ENTRY(getcr3)
525	movq	%cr3, %rax
526	ret
527	SET_SIZE(getcr3)
528
529	ENTRY(setcr3)
530	movq	%rdi, %cr3
531	ret
532	SET_SIZE(setcr3)
533
534	ENTRY(reload_cr3)
535	movq	%cr3, %rdi
536	movq	%rdi, %cr3
537	ret
538	SET_SIZE(reload_cr3)
539
540	ENTRY(getcr4)
541	movq	%cr4, %rax
542	ret
543	SET_SIZE(getcr4)
544
545	ENTRY(setcr4)
546	movq	%rdi, %cr4
547	ret
548	SET_SIZE(setcr4)
549
550	ENTRY(getcr8)
551	movq	%cr8, %rax
552	ret
553	SET_SIZE(getcr8)
554
555	ENTRY(setcr8)
556	movq	%rdi, %cr8
557	ret
558	SET_SIZE(setcr8)
559
560#elif defined(__i386)
561
562        ENTRY(getcr0)
563        movl    %cr0, %eax
564        ret
565	SET_SIZE(getcr0)
566
567        ENTRY(setcr0)
568        movl    4(%esp), %eax
569        movl    %eax, %cr0
570        ret
571	SET_SIZE(setcr0)
572
573        ENTRY(getcr2)
574        movl    %cr2, %eax
575        ret
576	SET_SIZE(getcr2)
577
578	ENTRY(getcr3)
579	movl    %cr3, %eax
580	ret
581	SET_SIZE(getcr3)
582
583        ENTRY(setcr3)
584        movl    4(%esp), %eax
585        movl    %eax, %cr3
586        ret
587	SET_SIZE(setcr3)
588
589	ENTRY(reload_cr3)
590	movl    %cr3, %eax
591	movl    %eax, %cr3
592	ret
593	SET_SIZE(reload_cr3)
594
595	ENTRY(getcr4)
596	movl    %cr4, %eax
597	ret
598	SET_SIZE(getcr4)
599
600        ENTRY(setcr4)
601        movl    4(%esp), %eax
602        movl    %eax, %cr4
603        ret
604	SET_SIZE(setcr4)
605
606#endif	/* __i386 */
607#endif	/* __lint */
608
609#if defined(__lint)
610
611/*ARGSUSED*/
612uint32_t
613__cpuid_insn(uint32_t eax, uint32_t *ebxp, uint32_t *ecxp, uint32_t *edxp)
614{ return (0); }
615
616#else	/* __lint */
617
618#if defined(__amd64)
619
620	ENTRY(__cpuid_insn)
621	movq	%rbx, %r11
622	movq	%rdx, %r8	/* r8 = ecxp */
623	movq	%rcx, %r9	/* r9 = edxp */
624	movl	%edi, %eax
625	cpuid
626	movl	%ebx, (%rsi)
627	movl	%ecx, (%r8)
628	movl	%edx, (%r9)
629	movq	%r11, %rbx
630	ret
631	SET_SIZE(__cpuid_insn)
632
633#elif defined(__i386)
634
635        ENTRY(__cpuid_insn)
636	pushl	%ebp
637	movl	%esp, %ebp
638	pushl	%ebx
639	movl	8(%ebp), %eax
640	cpuid
641	pushl	%eax
642	movl	0x0c(%ebp), %eax
643	movl	%ebx, (%eax)
644	movl	0x10(%ebp), %eax
645	movl	%ecx, (%eax)
646	movl	0x14(%ebp), %eax
647	movl	%edx, (%eax)
648	popl	%eax
649	popl	%ebx
650	popl	%ebp
651	ret
652	SET_SIZE(__cpuid_insn)
653
654#endif	/* __i386 */
655#endif	/* __lint */
656
657/*
658 * Insert entryp after predp in a doubly linked list.
659 */
660
661#if defined(__lint)
662
663/*ARGSUSED*/
664void
665_insque(caddr_t entryp, caddr_t predp)
666{}
667
668#else	/* __lint */
669
670#if defined(__amd64)
671
672	ENTRY(_insque)
673	movq	(%rsi), %rax		/* predp->forw 			*/
674	movq	%rsi, CPTRSIZE(%rdi)	/* entryp->back = predp		*/
675	movq	%rax, (%rdi)		/* entryp->forw = predp->forw	*/
676	movq	%rdi, (%rsi)		/* predp->forw = entryp		*/
677	movq	%rdi, CPTRSIZE(%rax)	/* predp->forw->back = entryp	*/
678	ret
679	SET_SIZE(_insque)
680
681#elif defined(__i386)
682
683	ENTRY(_insque)
684	movl	8(%esp), %edx
685	movl	4(%esp), %ecx
686	movl	(%edx), %eax		/* predp->forw			*/
687	movl	%edx, CPTRSIZE(%ecx)	/* entryp->back = predp		*/
688	movl	%eax, (%ecx)		/* entryp->forw = predp->forw	*/
689	movl	%ecx, (%edx)		/* predp->forw = entryp		*/
690	movl	%ecx, CPTRSIZE(%eax)	/* predp->forw->back = entryp	*/
691	ret
692	SET_SIZE(_insque)
693
694#endif	/* __i386 */
695#endif	/* __lint */
696
697/*
698 * Remove entryp from a doubly linked list
699 */
700
701#if defined(__lint)
702
703/*ARGSUSED*/
704void
705_remque(caddr_t entryp)
706{}
707
708#else	/* __lint */
709
710#if defined(__amd64)
711
712	ENTRY(_remque)
713	movq	(%rdi), %rax		/* entry->forw */
714	movq	CPTRSIZE(%rdi), %rdx	/* entry->back */
715	movq	%rax, (%rdx)		/* entry->back->forw = entry->forw */
716	movq	%rdx, CPTRSIZE(%rax)	/* entry->forw->back = entry->back */
717	ret
718	SET_SIZE(_remque)
719
720#elif defined(__i386)
721
722	ENTRY(_remque)
723	movl	4(%esp), %ecx
724	movl	(%ecx), %eax		/* entry->forw */
725	movl	CPTRSIZE(%ecx), %edx	/* entry->back */
726	movl	%eax, (%edx)		/* entry->back->forw = entry->forw */
727	movl	%edx, CPTRSIZE(%eax)	/* entry->forw->back = entry->back */
728	ret
729	SET_SIZE(_remque)
730
731#endif	/* __i386 */
732#endif	/* __lint */
733
734/*
735 * Returns the number of
736 * non-NULL bytes in string argument.
737 */
738
739#if defined(__lint)
740
741/* ARGSUSED */
742size_t
743strlen(const char *str)
744{ return (0); }
745
746#else	/* __lint */
747
748#if defined(__amd64)
749
750/*
751 * This is close to a simple transliteration of a C version of this
752 * routine.  We should either just -make- this be a C version, or
753 * justify having it in assembler by making it significantly faster.
754 *
755 * size_t
756 * strlen(const char *s)
757 * {
758 *	const char *s0;
759 * #if defined(DEBUG)
760 *	if ((uintptr_t)s < KERNELBASE)
761 *		panic(.str_panic_msg);
762 * #endif
763 *	for (s0 = s; *s; s++)
764 *		;
765 *	return (s - s0);
766 * }
767 */
768
769	ENTRY(strlen)
770#ifdef DEBUG
771	movq	kernelbase(%rip), %rax
772	cmpq	%rax, %rdi
773	jae	str_valid
774	pushq	%rbp
775	movq	%rsp, %rbp
776	leaq	.str_panic_msg(%rip), %rdi
777	xorl	%eax, %eax
778	call	panic
779#endif	/* DEBUG */
780str_valid:
781	cmpb	$0, (%rdi)
782	movq	%rdi, %rax
783	je	.null_found
784	.align	4
785.strlen_loop:
786	incq	%rdi
787	cmpb	$0, (%rdi)
788	jne	.strlen_loop
789.null_found:
790	subq	%rax, %rdi
791	movq	%rdi, %rax
792	ret
793	SET_SIZE(strlen)
794
795#elif defined(__i386)
796
797	ENTRY(strlen)
798#ifdef DEBUG
799	movl	kernelbase, %eax
800	cmpl	%eax, 4(%esp)
801	jae	str_valid
802	pushl	%ebp
803	movl	%esp, %ebp
804	pushl	$.str_panic_msg
805	call	panic
806#endif /* DEBUG */
807
808str_valid:
809	movl	4(%esp), %eax		/* %eax = string address */
810	testl	$3, %eax		/* if %eax not word aligned */
811	jnz	.not_word_aligned	/* goto .not_word_aligned */
812	.align	4
813.word_aligned:
814	movl	(%eax), %edx		/* move 1 word from (%eax) to %edx */
815	movl	$0x7f7f7f7f, %ecx
816	andl	%edx, %ecx		/* %ecx = %edx & 0x7f7f7f7f */
817	addl	$4, %eax		/* next word */
818	addl	$0x7f7f7f7f, %ecx	/* %ecx += 0x7f7f7f7f */
819	orl	%edx, %ecx		/* %ecx |= %edx */
820	andl	$0x80808080, %ecx	/* %ecx &= 0x80808080 */
821	cmpl	$0x80808080, %ecx	/* if no null byte in this word */
822	je	.word_aligned		/* goto .word_aligned */
823	subl	$4, %eax		/* post-incremented */
824.not_word_aligned:
825	cmpb	$0, (%eax)		/* if a byte in (%eax) is null */
826	je	.null_found		/* goto .null_found */
827	incl	%eax			/* next byte */
828	testl	$3, %eax		/* if %eax not word aligned */
829	jnz	.not_word_aligned	/* goto .not_word_aligned */
830	jmp	.word_aligned		/* goto .word_aligned */
831	.align	4
832.null_found:
833	subl	4(%esp), %eax		/* %eax -= string address */
834	ret
835	SET_SIZE(strlen)
836
837#endif	/* __i386 */
838
839#ifdef DEBUG
840	.text
841.str_panic_msg:
842	.string "strlen: argument below kernelbase"
843#endif /* DEBUG */
844
845#endif	/* __lint */
846
847	/*
848	 * Berkley 4.3 introduced symbolically named interrupt levels
849	 * as a way deal with priority in a machine independent fashion.
850	 * Numbered priorities are machine specific, and should be
851	 * discouraged where possible.
852	 *
853	 * Note, for the machine specific priorities there are
854	 * examples listed for devices that use a particular priority.
855	 * It should not be construed that all devices of that
856	 * type should be at that priority.  It is currently were
857	 * the current devices fit into the priority scheme based
858	 * upon time criticalness.
859	 *
860	 * The underlying assumption of these assignments is that
861	 * IPL 10 is the highest level from which a device
862	 * routine can call wakeup.  Devices that interrupt from higher
863	 * levels are restricted in what they can do.  If they need
864	 * kernels services they should schedule a routine at a lower
865	 * level (via software interrupt) to do the required
866	 * processing.
867	 *
868	 * Examples of this higher usage:
869	 *	Level	Usage
870	 *	14	Profiling clock (and PROM uart polling clock)
871	 *	12	Serial ports
872	 *
873	 * The serial ports request lower level processing on level 6.
874	 *
875	 * Also, almost all splN routines (where N is a number or a
876	 * mnemonic) will do a RAISE(), on the assumption that they are
877	 * never used to lower our priority.
878	 * The exceptions are:
879	 *	spl8()		Because you can't be above 15 to begin with!
880	 *	splzs()		Because this is used at boot time to lower our
881	 *			priority, to allow the PROM to poll the uart.
882	 *	spl0()		Used to lower priority to 0.
883	 */
884
885#if defined(__lint)
886
887int spl0(void)		{ return (0); }
888int spl6(void)		{ return (0); }
889int spl7(void)		{ return (0); }
890int spl8(void)		{ return (0); }
891int splhigh(void)	{ return (0); }
892int splhi(void)		{ return (0); }
893int splzs(void)		{ return (0); }
894
895#else	/* __lint */
896
897/* reg = cpu->cpu_m.cpu_pri; */
898#define	GETIPL_NOGS(reg, cpup)	\
899	movl	CPU_PRI(cpup), reg;
900
901/* cpu->cpu_m.cpu_pri; */
902#define	SETIPL_NOGS(val, cpup)	\
903	movl	val, CPU_PRI(cpup);
904
905/* reg = cpu->cpu_m.cpu_pri; */
906#define	GETIPL(reg)	\
907	movl	%gs:CPU_PRI, reg;
908
909/* cpu->cpu_m.cpu_pri; */
910#define	SETIPL(val)	\
911	movl	val, %gs:CPU_PRI;
912
913/*
914 * Macro to raise processor priority level.
915 * Avoid dropping processor priority if already at high level.
916 * Also avoid going below CPU->cpu_base_spl, which could've just been set by
917 * a higher-level interrupt thread that just blocked.
918 */
919#if defined(__amd64)
920
921#define	RAISE(level) \
922	cli;			\
923	LOADCPU(%rcx);		\
924	movl	$/**/level, %edi;\
925	GETIPL_NOGS(%eax, %rcx);\
926	cmpl 	%eax, %edi;	\
927	jg	spl;		\
928	jmp	setsplhisti
929
930#elif defined(__i386)
931
932#define	RAISE(level) \
933	cli;			\
934	LOADCPU(%ecx);		\
935	movl	$/**/level, %edx;\
936	GETIPL_NOGS(%eax, %ecx);\
937	cmpl 	%eax, %edx;	\
938	jg	spl;		\
939	jmp	setsplhisti
940
941#endif	/* __i386 */
942
943/*
944 * Macro to set the priority to a specified level.
945 * Avoid dropping the priority below CPU->cpu_base_spl.
946 */
947#if defined(__amd64)
948
949#define	SETPRI(level) \
950	cli;				\
951	LOADCPU(%rcx);			\
952	movl	$/**/level, %edi;	\
953	jmp	spl
954
955#elif defined(__i386)
956
957#define SETPRI(level) \
958	cli;				\
959	LOADCPU(%ecx);			\
960	movl	$/**/level, %edx;	\
961	jmp	spl
962
963#endif	/* __i386 */
964
965	/* locks out all interrupts, including memory errors */
966	ENTRY(spl8)
967	SETPRI(15)
968	SET_SIZE(spl8)
969
970	/* just below the level that profiling runs */
971	ENTRY(spl7)
972	RAISE(13)
973	SET_SIZE(spl7)
974
975	/* sun specific - highest priority onboard serial i/o asy ports */
976	ENTRY(splzs)
977	SETPRI(12)	/* Can't be a RAISE, as it's used to lower us */
978	SET_SIZE(splzs)
979
980	/*
981	 * should lock out clocks and all interrupts,
982	 * as you can see, there are exceptions
983	 */
984
985#if defined(__amd64)
986
987	.align	16
988	ENTRY(splhi)
989	ALTENTRY(splhigh)
990	ALTENTRY(spl6)
991	ALTENTRY(i_ddi_splhigh)
992	cli
993	LOADCPU(%rcx)
994	movl	$DISP_LEVEL, %edi
995	movl	CPU_PRI(%rcx), %eax
996	cmpl	%eax, %edi
997	jle	setsplhisti
998	SETIPL_NOGS(%edi, %rcx)
999	/*
1000	 * If we aren't using cr8 to control ipl then we patch this
1001	 * with a jump to slow_setsplhi
1002	 */
1003	ALTENTRY(setsplhi_patch)
1004	movq	CPU_PRI_DATA(%rcx), %r11 /* get pri data ptr */
1005	movzb	(%r11, %rdi, 1), %rdx	/* get apic mask for this ipl */
1006	movq	%rdx, %cr8		/* set new apic priority */
1007	/*
1008	 * enable interrupts
1009	 */
1010setsplhisti:
1011	nop	/* patch this to a sti when a proper setspl routine appears */
1012	ret
1013
1014	ALTENTRY(slow_setsplhi)
1015	pushq	%rbp
1016	movq	%rsp, %rbp
1017	subq	$16, %rsp
1018	movl	%eax, -4(%rbp)		/* save old ipl */
1019	call	*setspl(%rip)
1020	movl	-4(%rbp), %eax		/* return old ipl */
1021	leave
1022	jmp	setsplhisti
1023
1024	SET_SIZE(i_ddi_splhigh)
1025	SET_SIZE(spl6)
1026	SET_SIZE(splhigh)
1027	SET_SIZE(splhi)
1028
1029#elif defined(__i386)
1030
1031	.align	16
1032	ENTRY(splhi)
1033	ALTENTRY(splhigh)
1034	ALTENTRY(spl6)
1035	ALTENTRY(i_ddi_splhigh)
1036	cli
1037	LOADCPU(%ecx)
1038	movl	$DISP_LEVEL, %edx
1039	movl	CPU_PRI(%ecx), %eax
1040	cmpl	%eax, %edx
1041	jle	setsplhisti
1042	SETIPL_NOGS(%edx, %ecx)		/* set new ipl */
1043
1044	pushl   %eax                    /* save old ipl */
1045	pushl	%edx			/* pass new ipl */
1046	call	*setspl
1047	popl	%ecx			/* dummy pop */
1048	popl    %eax                    /* return old ipl */
1049	/*
1050	 * enable interrupts
1051	 *
1052	 * (we patch this to an sti once a proper setspl routine
1053	 * is installed)
1054	 */
1055setsplhisti:
1056	nop	/* patch this to a sti when a proper setspl routine appears */
1057	ret
1058	SET_SIZE(i_ddi_splhigh)
1059	SET_SIZE(spl6)
1060	SET_SIZE(splhigh)
1061	SET_SIZE(splhi)
1062
1063#endif	/* __i386 */
1064
1065	/* allow all interrupts */
1066	ENTRY(spl0)
1067	SETPRI(0)
1068	SET_SIZE(spl0)
1069
1070#endif	/* __lint */
1071
1072/*
1073 * splr is like splx but will only raise the priority and never drop it
1074 */
1075#if defined(__lint)
1076
1077/* ARGSUSED */
1078int
1079splr(int level)
1080{ return (0); }
1081
1082#else	/* __lint */
1083
1084#if defined(__amd64)
1085
1086	ENTRY(splr)
1087	cli
1088	LOADCPU(%rcx)
1089	GETIPL_NOGS(%eax, %rcx)
1090	cmpl	%eax, %edi		/* if new level > current level */
1091	jg	spl			/* then set ipl to new level */
1092splr_setsti:
1093	nop	/* patch this to a sti when a proper setspl routine appears */
1094	ret				/* else return the current level */
1095	SET_SIZE(splr)
1096
1097#elif defined(__i386)
1098
1099	ENTRY(splr)
1100	cli
1101	LOADCPU(%ecx)
1102	movl	4(%esp), %edx		/* get new spl level */
1103	GETIPL_NOGS(%eax, %ecx)
1104	cmpl 	%eax, %edx		/* if new level > current level */
1105	jg	spl			/* then set ipl to new level */
1106splr_setsti:
1107	nop	/* patch this to a sti when a proper setspl routine appears */
1108	ret				/* else return the current level */
1109	SET_SIZE(splr)
1110
1111#endif	/* __i386 */
1112#endif	/* __lint */
1113
1114
1115
1116/*
1117 * splx - set PIL back to that indicated by the level passed as an argument,
1118 * or to the CPU's base priority, whichever is higher.
1119 * Needs to be fall through to spl to save cycles.
1120 * Algorithm for spl:
1121 *
1122 *      turn off interrupts
1123 *
1124 *	if (CPU->cpu_base_spl > newipl)
1125 *		newipl = CPU->cpu_base_spl;
1126 *      oldipl = CPU->cpu_pridata->c_ipl;
1127 *      CPU->cpu_pridata->c_ipl = newipl;
1128 *
1129 *	/indirectly call function to set spl values (usually setpicmasks)
1130 *      setspl();  // load new masks into pics
1131 *
1132 * Be careful not to set priority lower than CPU->cpu_base_pri,
1133 * even though it seems we're raising the priority, it could be set
1134 * higher at any time by an interrupt routine, so we must block interrupts
1135 * and look at CPU->cpu_base_pri
1136 */
1137#if defined(__lint)
1138
1139/* ARGSUSED */
1140void
1141splx(int level)
1142{}
1143
1144#else	/* __lint */
1145
1146#if defined(__amd64)
1147
1148	ENTRY(splx)
1149	ALTENTRY(i_ddi_splx)
1150	cli				/* disable interrupts */
1151	LOADCPU(%rcx)
1152	/*FALLTHRU*/
1153	.align	4
1154spl:
1155	/*
1156	 * New priority level is in %edi, cpu struct pointer is in %rcx
1157	 */
1158	GETIPL_NOGS(%eax, %rcx)		/* get current ipl */
1159	cmpl   %edi, CPU_BASE_SPL(%rcx) /* if (base spl > new ipl) */
1160	ja     set_to_base_spl		/* then use base_spl */
1161
1162setprilev:
1163	SETIPL_NOGS(%edi, %rcx)		/* set new ipl */
1164	/*
1165	 * If we aren't using cr8 to control ipl then we patch this
1166	 * with a jump to slow_spl
1167	 */
1168	ALTENTRY(spl_patch)
1169	movq	CPU_PRI_DATA(%rcx), %r11 /* get pri data ptr */
1170	movzb	(%r11, %rdi, 1), %rdx	/* get apic mask for this ipl */
1171	movq	%rdx, %cr8		/* set new apic priority */
1172	xorl	%edx, %edx
1173	bsrl	CPU_SOFTINFO(%rcx), %edx /* fls(cpu->cpu_softinfo.st_pending) */
1174	cmpl	%edi, %edx		/* new ipl vs. st_pending */
1175	jle	setsplsti
1176
1177	pushq	%rbp
1178	movq	%rsp, %rbp
1179	/* stack now 16-byte aligned */
1180	pushq	%rax			/* save old spl */
1181	pushq	%rdi			/* save new ipl too */
1182	jmp	fakesoftint
1183
1184setsplsti:
1185	nop	/* patch this to a sti when a proper setspl routine appears */
1186	ret
1187
1188	ALTENTRY(slow_spl)
1189	pushq	%rbp
1190	movq	%rsp, %rbp
1191	/* stack now 16-byte aligned */
1192
1193	pushq	%rax			/* save old spl */
1194	pushq	%rdi			/* save new ipl too */
1195
1196	call	*setspl(%rip)
1197
1198	LOADCPU(%rcx)
1199	movl	CPU_SOFTINFO(%rcx), %eax
1200	orl	%eax, %eax
1201	jz	slow_setsplsti
1202
1203	bsrl	%eax, %edx		/* fls(cpu->cpu_softinfo.st_pending) */
1204	cmpl	0(%rsp), %edx		/* new ipl vs. st_pending */
1205	jg	fakesoftint
1206
1207	ALTENTRY(fakesoftint_return)
1208	/*
1209	 * enable interrupts
1210	 */
1211slow_setsplsti:
1212	nop	/* patch this to a sti when a proper setspl routine appears */
1213	popq	%rdi
1214	popq	%rax			/* return old ipl */
1215	leave
1216	ret
1217	SET_SIZE(fakesoftint_return)
1218
1219set_to_base_spl:
1220	movl	CPU_BASE_SPL(%rcx), %edi
1221	jmp	setprilev
1222	SET_SIZE(spl)
1223	SET_SIZE(i_ddi_splx)
1224	SET_SIZE(splx)
1225
1226#elif defined(__i386)
1227
1228	ENTRY(splx)
1229	ALTENTRY(i_ddi_splx)
1230	cli                             /* disable interrupts */
1231	LOADCPU(%ecx)
1232	movl	4(%esp), %edx		/* get new spl level */
1233	/*FALLTHRU*/
1234
1235	.align	4
1236	ALTENTRY(spl)
1237	/*
1238	 * New priority level is in %edx
1239	 * (doing this early to avoid an AGI in the next instruction)
1240	 */
1241	GETIPL_NOGS(%eax, %ecx)		/* get current ipl */
1242	cmpl	%edx, CPU_BASE_SPL(%ecx) /* if ( base spl > new ipl) */
1243	ja	set_to_base_spl		/* then use base_spl */
1244
1245setprilev:
1246	SETIPL_NOGS(%edx, %ecx)		/* set new ipl */
1247
1248	pushl   %eax                    /* save old ipl */
1249	pushl	%edx			/* pass new ipl */
1250	call	*setspl
1251
1252	LOADCPU(%ecx)
1253	movl	CPU_SOFTINFO(%ecx), %eax
1254	orl	%eax, %eax
1255	jz	setsplsti
1256
1257	/*
1258	 * Before dashing off, check that setsplsti has been patched.
1259	 */
1260	cmpl	$NOP_INSTR, setsplsti
1261	je	setsplsti
1262
1263	bsrl	%eax, %edx
1264	cmpl	0(%esp), %edx
1265	jg	fakesoftint
1266
1267	ALTENTRY(fakesoftint_return)
1268	/*
1269	 * enable interrupts
1270	 */
1271setsplsti:
1272	nop	/* patch this to a sti when a proper setspl routine appears */
1273	popl	%eax
1274	popl    %eax			/ return old ipl
1275	ret
1276	SET_SIZE(fakesoftint_return)
1277
1278set_to_base_spl:
1279	movl	CPU_BASE_SPL(%ecx), %edx
1280	jmp	setprilev
1281	SET_SIZE(spl)
1282	SET_SIZE(i_ddi_splx)
1283	SET_SIZE(splx)
1284
1285#endif	/* __i386 */
1286#endif	/* __lint */
1287
1288#if defined(__lint)
1289
1290void
1291install_spl(void)
1292{}
1293
1294#else	/* __lint */
1295
1296#if defined(__amd64)
1297
1298	ENTRY_NP(install_spl)
1299	movq	%cr0, %rax
1300	movq	%rax, %rdx
1301	movl	$_BITNOT(CR0_WP), %ecx
1302	movslq	%ecx, %rcx
1303	andq	%rcx, %rax		/* we don't want to take a fault */
1304	movq	%rax, %cr0
1305	jmp	1f
13061:	movb	$STI_INSTR, setsplsti(%rip)
1307	movb	$STI_INSTR, slow_setsplsti(%rip)
1308	movb	$STI_INSTR, setsplhisti(%rip)
1309	movb	$STI_INSTR, splr_setsti(%rip)
1310	testl	$1, intpri_use_cr8(%rip)	/* are using %cr8 ? */
1311	jz	2f				/* no, go patch more */
1312	movq	%rdx, %cr0
1313	ret
13142:
1315	/*
1316	 * Patch spl functions to use slow spl method
1317	 */
1318	leaq	setsplhi_patch(%rip), %rdi	/* get patch point addr */
1319	leaq	slow_setsplhi(%rip), %rax	/* jmp target */
1320	subq	%rdi, %rax			/* calculate jmp distance */
1321	subq	$2, %rax			/* minus size of jmp instr */
1322	shlq	$8, %rax			/* construct jmp instr */
1323	addq	$JMP_INSTR, %rax
1324	movw	%ax, setsplhi_patch(%rip)	/* patch in the jmp */
1325	leaq	spl_patch(%rip), %rdi		/* get patch point addr */
1326	leaq	slow_spl(%rip), %rax		/* jmp target */
1327	subq	%rdi, %rax			/* calculate jmp distance */
1328	subq	$2, %rax			/* minus size of jmp instr */
1329	shlq	$8, %rax			/* construct jmp instr */
1330	addq	$JMP_INSTR, %rax
1331	movw	%ax, spl_patch(%rip)		/* patch in the jmp */
1332	/*
1333	 * Ensure %cr8 is zero since we aren't using it
1334	 */
1335	xorl	%eax, %eax
1336	movq	%rax, %cr8
1337	movq	%rdx, %cr0
1338	ret
1339	SET_SIZE(install_spl)
1340
1341#elif defined(__i386)
1342
1343	ENTRY_NP(install_spl)
1344	movl	%cr0, %eax
1345	movl	%eax, %edx
1346	andl	$_BITNOT(CR0_WP), %eax	/* we don't want to take a fault */
1347	movl	%eax, %cr0
1348	jmp	1f
13491:	movb	$STI_INSTR, setsplsti
1350	movb	$STI_INSTR, setsplhisti
1351	movb	$STI_INSTR, splr_setsti
1352	movl	%edx, %cr0
1353	ret
1354	SET_SIZE(install_spl)
1355
1356#endif	/* __i386 */
1357#endif	/* __lint */
1358
1359
1360/*
1361 * Get current processor interrupt level
1362 */
1363
1364#if defined(__lint)
1365
1366int
1367getpil(void)
1368{ return (0); }
1369
1370#else	/* __lint */
1371
1372#if defined(__amd64)
1373
1374	ENTRY(getpil)
1375	GETIPL(%eax)			/* priority level into %eax */
1376	ret
1377	SET_SIZE(getpil)
1378
1379#elif defined(__i386)
1380
1381	ENTRY(getpil)
1382	GETIPL(%eax)			/* priority level into %eax */
1383	ret
1384	SET_SIZE(getpil)
1385
1386#endif	/* __i386 */
1387#endif	/* __lint */
1388
1389#if defined(__i386)
1390
1391/*
1392 * Read and write the %gs register
1393 */
1394
1395#if defined(__lint)
1396
1397/*ARGSUSED*/
1398uint16_t
1399getgs(void)
1400{ return (0); }
1401
1402/*ARGSUSED*/
1403void
1404setgs(uint16_t sel)
1405{}
1406
1407#else	/* __lint */
1408
1409	ENTRY(getgs)
1410	clr	%eax
1411	movw	%gs, %ax
1412	ret
1413	SET_SIZE(getgs)
1414
1415	ENTRY(setgs)
1416	movw	4(%esp), %gs
1417	ret
1418	SET_SIZE(setgs)
1419
1420#endif	/* __lint */
1421#endif	/* __i386 */
1422
1423#if defined(__lint)
1424
1425void
1426pc_reset(void)
1427{}
1428
1429#else	/* __lint */
1430
1431	ENTRY(pc_reset)
1432	movw	$0x64, %dx
1433	movb	$0xfe, %al
1434	outb	(%dx)
1435	hlt
1436	/*NOTREACHED*/
1437	SET_SIZE(pc_reset)
1438
1439#endif	/* __lint */
1440
1441/*
1442 * C callable in and out routines
1443 */
1444
1445#if defined(__lint)
1446
1447/* ARGSUSED */
1448void
1449outl(int port_address, uint32_t val)
1450{}
1451
1452#else	/* __lint */
1453
1454#if defined(__amd64)
1455
1456	ENTRY(outl)
1457	movw	%di, %dx
1458	movl	%esi, %eax
1459	outl	(%dx)
1460	ret
1461	SET_SIZE(outl)
1462
1463#elif defined(__i386)
1464
1465	.set	PORT, 4
1466	.set	VAL, 8
1467
1468	ENTRY(outl)
1469	movw	PORT(%esp), %dx
1470	movl	VAL(%esp), %eax
1471	outl	(%dx)
1472	ret
1473	SET_SIZE(outl)
1474
1475#endif	/* __i386 */
1476#endif	/* __lint */
1477
1478#if defined(__lint)
1479
1480/* ARGSUSED */
1481void
1482outw(int port_address, uint16_t val)
1483{}
1484
1485#else	/* __lint */
1486
1487#if defined(__amd64)
1488
1489	ENTRY(outw)
1490	movw	%di, %dx
1491	movw	%si, %ax
1492	D16 outl (%dx)		/* XX64 why not outw? */
1493	ret
1494	SET_SIZE(outw)
1495
1496#elif defined(__i386)
1497
1498	ENTRY(outw)
1499	movw	PORT(%esp), %dx
1500	movw	VAL(%esp), %ax
1501	D16 outl (%dx)
1502	ret
1503	SET_SIZE(outw)
1504
1505#endif	/* __i386 */
1506#endif	/* __lint */
1507
1508#if defined(__lint)
1509
1510/* ARGSUSED */
1511void
1512outb(int port_address, uint8_t val)
1513{}
1514
1515#else	/* __lint */
1516
1517#if defined(__amd64)
1518
1519	ENTRY(outb)
1520	movw	%di, %dx
1521	movb	%sil, %al
1522	outb	(%dx)
1523	ret
1524	SET_SIZE(outb)
1525
1526#elif defined(__i386)
1527
1528	ENTRY(outb)
1529	movw	PORT(%esp), %dx
1530	movb	VAL(%esp), %al
1531	outb	(%dx)
1532	ret
1533	SET_SIZE(outb)
1534
1535#endif	/* __i386 */
1536#endif	/* __lint */
1537
1538#if defined(__lint)
1539
1540/* ARGSUSED */
1541uint32_t
1542inl(int port_address)
1543{ return (0); }
1544
1545#else	/* __lint */
1546
1547#if defined(__amd64)
1548
1549	ENTRY(inl)
1550	xorl	%eax, %eax
1551	movw	%di, %dx
1552	inl	(%dx)
1553	ret
1554	SET_SIZE(inl)
1555
1556#elif defined(__i386)
1557
1558	ENTRY(inl)
1559	movw	PORT(%esp), %dx
1560	inl	(%dx)
1561	ret
1562	SET_SIZE(inl)
1563
1564#endif	/* __i386 */
1565#endif	/* __lint */
1566
1567#if defined(__lint)
1568
1569/* ARGSUSED */
1570uint16_t
1571inw(int port_address)
1572{ return (0); }
1573
1574#else	/* __lint */
1575
1576#if defined(__amd64)
1577
1578	ENTRY(inw)
1579	xorl	%eax, %eax
1580	movw	%di, %dx
1581	D16 inl	(%dx)
1582	ret
1583	SET_SIZE(inw)
1584
1585#elif defined(__i386)
1586
1587	ENTRY(inw)
1588	subl	%eax, %eax
1589	movw	PORT(%esp), %dx
1590	D16 inl	(%dx)
1591	ret
1592	SET_SIZE(inw)
1593
1594#endif	/* __i386 */
1595#endif	/* __lint */
1596
1597
1598#if defined(__lint)
1599
1600/* ARGSUSED */
1601uint8_t
1602inb(int port_address)
1603{ return (0); }
1604
1605#else	/* __lint */
1606
1607#if defined(__amd64)
1608
1609	ENTRY(inb)
1610	xorl	%eax, %eax
1611	movw	%di, %dx
1612	inb	(%dx)
1613	ret
1614	SET_SIZE(inb)
1615
1616#elif defined(__i386)
1617
1618	ENTRY(inb)
1619	subl    %eax, %eax
1620	movw	PORT(%esp), %dx
1621	inb	(%dx)
1622	ret
1623	SET_SIZE(inb)
1624
1625#endif	/* __i386 */
1626#endif	/* __lint */
1627
1628
1629#if defined(__lint)
1630
1631/* ARGSUSED */
1632void
1633repoutsw(int port, uint16_t *addr, int cnt)
1634{}
1635
1636#else	/* __lint */
1637
1638#if defined(__amd64)
1639
1640	ENTRY(repoutsw)
1641	movl	%edx, %ecx
1642	movw	%di, %dx
1643	rep
1644	  D16 outsl
1645	ret
1646	SET_SIZE(repoutsw)
1647
1648#elif defined(__i386)
1649
1650	/*
1651	 * The arguments and saved registers are on the stack in the
1652	 *  following order:
1653	 *      |  cnt  |  +16
1654	 *      | *addr |  +12
1655	 *      | port  |  +8
1656	 *      |  eip  |  +4
1657	 *      |  esi  |  <-- %esp
1658	 * If additional values are pushed onto the stack, make sure
1659	 * to adjust the following constants accordingly.
1660	 */
1661	.set	PORT, 8
1662	.set	ADDR, 12
1663	.set	COUNT, 16
1664
1665	ENTRY(repoutsw)
1666	pushl	%esi
1667	movl	PORT(%esp), %edx
1668	movl	ADDR(%esp), %esi
1669	movl	COUNT(%esp), %ecx
1670	rep
1671	  D16 outsl
1672	popl	%esi
1673	ret
1674	SET_SIZE(repoutsw)
1675
1676#endif	/* __i386 */
1677#endif	/* __lint */
1678
1679
1680#if defined(__lint)
1681
1682/* ARGSUSED */
1683void
1684repinsw(int port_addr, uint16_t *addr, int cnt)
1685{}
1686
1687#else	/* __lint */
1688
1689#if defined(__amd64)
1690
1691	ENTRY(repinsw)
1692	movl	%edx, %ecx
1693	movw	%di, %dx
1694	rep
1695	  D16 insl
1696	ret
1697	SET_SIZE(repinsw)
1698
1699#elif defined(__i386)
1700
1701	ENTRY(repinsw)
1702	pushl	%edi
1703	movl	PORT(%esp), %edx
1704	movl	ADDR(%esp), %edi
1705	movl	COUNT(%esp), %ecx
1706	rep
1707	  D16 insl
1708	popl	%edi
1709	ret
1710	SET_SIZE(repinsw)
1711
1712#endif	/* __i386 */
1713#endif	/* __lint */
1714
1715
1716#if defined(__lint)
1717
1718/* ARGSUSED */
1719void
1720repinsb(int port, uint8_t *addr, int count)
1721{}
1722
1723#else	/* __lint */
1724
1725#if defined(__amd64)
1726
1727	ENTRY(repinsb)
1728	movl	%edx, %ecx
1729	movw	%di, %dx
1730	movq	%rsi, %rdi
1731	rep
1732	  insb
1733	ret
1734	SET_SIZE(repinsb)
1735
1736#elif defined(__i386)
1737
1738	/*
1739	 * The arguments and saved registers are on the stack in the
1740	 *  following order:
1741	 *      |  cnt  |  +16
1742	 *      | *addr |  +12
1743	 *      | port  |  +8
1744	 *      |  eip  |  +4
1745	 *      |  esi  |  <-- %esp
1746	 * If additional values are pushed onto the stack, make sure
1747	 * to adjust the following constants accordingly.
1748	 */
1749	.set	IO_PORT, 8
1750	.set	IO_ADDR, 12
1751	.set	IO_COUNT, 16
1752
1753	ENTRY(repinsb)
1754	pushl	%edi
1755	movl	IO_ADDR(%esp), %edi
1756	movl	IO_COUNT(%esp), %ecx
1757	movl	IO_PORT(%esp), %edx
1758	rep
1759	  insb
1760	popl	%edi
1761	ret
1762	SET_SIZE(repinsb)
1763
1764#endif	/* __i386 */
1765#endif	/* __lint */
1766
1767
1768/*
1769 * Input a stream of 32-bit words.
1770 * NOTE: count is a DWORD count.
1771 */
1772#if defined(__lint)
1773
1774/* ARGSUSED */
1775void
1776repinsd(int port, uint32_t *addr, int count)
1777{}
1778
1779#else	/* __lint */
1780
1781#if defined(__amd64)
1782
1783	ENTRY(repinsd)
1784	movl	%edx, %ecx
1785	movw	%di, %dx
1786	movq	%rsi, %rdi
1787	rep
1788	  insl
1789	ret
1790	SET_SIZE(repinsd)
1791
1792#elif defined(__i386)
1793
1794	ENTRY(repinsd)
1795	pushl	%edi
1796	movl	IO_ADDR(%esp), %edi
1797	movl	IO_COUNT(%esp), %ecx
1798	movl	IO_PORT(%esp), %edx
1799	rep
1800	  insl
1801	popl	%edi
1802	ret
1803	SET_SIZE(repinsd)
1804
1805#endif	/* __i386 */
1806#endif	/* __lint */
1807
1808/*
1809 * Output a stream of bytes
1810 * NOTE: count is a byte count
1811 */
1812#if defined(__lint)
1813
1814/* ARGSUSED */
1815void
1816repoutsb(int port, uint8_t *addr, int count)
1817{}
1818
1819#else	/* __lint */
1820
1821#if defined(__amd64)
1822
1823	ENTRY(repoutsb)
1824	movl	%edx, %ecx
1825	movw	%di, %dx
1826	rep
1827	  outsb
1828	ret
1829	SET_SIZE(repoutsb)
1830
1831#elif defined(__i386)
1832
1833	ENTRY(repoutsb)
1834	pushl	%esi
1835	movl	IO_ADDR(%esp), %esi
1836	movl	IO_COUNT(%esp), %ecx
1837	movl	IO_PORT(%esp), %edx
1838	rep
1839	  outsb
1840	popl	%esi
1841	ret
1842	SET_SIZE(repoutsb)
1843
1844#endif	/* __i386 */
1845#endif	/* __lint */
1846
1847/*
1848 * Output a stream of 32-bit words
1849 * NOTE: count is a DWORD count
1850 */
1851#if defined(__lint)
1852
1853/* ARGSUSED */
1854void
1855repoutsd(int port, uint32_t *addr, int count)
1856{}
1857
1858#else	/* __lint */
1859
1860#if defined(__amd64)
1861
1862	ENTRY(repoutsd)
1863	movl	%edx, %ecx
1864	movw	%di, %dx
1865	rep
1866	  outsl
1867	ret
1868	SET_SIZE(repoutsd)
1869
1870#elif defined(__i386)
1871
1872	ENTRY(repoutsd)
1873	pushl	%esi
1874	movl	IO_ADDR(%esp), %esi
1875	movl	IO_COUNT(%esp), %ecx
1876	movl	IO_PORT(%esp), %edx
1877	rep
1878	  outsl
1879	popl	%esi
1880	ret
1881	SET_SIZE(repoutsd)
1882
1883#endif	/* __i386 */
1884#endif	/* __lint */
1885
1886/*
1887 * void int20(void)
1888 */
1889
1890#if defined(__lint)
1891
1892void
1893int20(void)
1894{}
1895
1896#else	/* __lint */
1897
1898	ENTRY(int20)
1899	movl	boothowto, %eax
1900	andl	$RB_DEBUG, %eax
1901	jz	1f
1902
1903	int	$20
19041:
1905	ret
1906	SET_SIZE(int20)
1907
1908#endif	/* __lint */
1909
1910#if defined(__lint)
1911
1912/* ARGSUSED */
1913int
1914scanc(size_t size, uchar_t *cp, uchar_t *table, uchar_t mask)
1915{ return (0); }
1916
1917#else	/* __lint */
1918
1919#if defined(__amd64)
1920
1921	ENTRY(scanc)
1922					/* rdi == size */
1923					/* rsi == cp */
1924					/* rdx == table */
1925					/* rcx == mask */
1926	addq	%rsi, %rdi		/* end = &cp[size] */
1927.scanloop:
1928	cmpq	%rdi, %rsi		/* while (cp < end */
1929	jnb	.scandone
1930	movzbq	(%rsi), %r8		/* %r8 = *cp */
1931	incq	%rsi			/* cp++ */
1932	testb	%cl, (%r8, %rdx)
1933	jz	.scanloop		/*  && (table[*cp] & mask) == 0) */
1934	decq	%rsi			/* (fix post-increment) */
1935.scandone:
1936	movl	%edi, %eax
1937	subl	%esi, %eax		/* return (end - cp) */
1938	ret
1939	SET_SIZE(scanc)
1940
1941#elif defined(__i386)
1942
1943	ENTRY(scanc)
1944	pushl	%edi
1945	pushl	%esi
1946	movb	24(%esp), %cl		/* mask = %cl */
1947	movl	16(%esp), %esi		/* cp = %esi */
1948	movl	20(%esp), %edx		/* table = %edx */
1949	movl	%esi, %edi
1950	addl	12(%esp), %edi		/* end = &cp[size]; */
1951.scanloop:
1952	cmpl	%edi, %esi		/* while (cp < end */
1953	jnb	.scandone
1954	movzbl	(%esi),  %eax		/* %al = *cp */
1955	incl	%esi			/* cp++ */
1956	movb	(%edx,  %eax), %al	/* %al = table[*cp] */
1957	testb	%al, %cl
1958	jz	.scanloop		/*   && (table[*cp] & mask) == 0) */
1959	dec	%esi			/* post-incremented */
1960.scandone:
1961	movl	%edi, %eax
1962	subl	%esi, %eax		/* return (end - cp) */
1963	popl	%esi
1964	popl	%edi
1965	ret
1966	SET_SIZE(scanc)
1967
1968#endif	/* __i386 */
1969#endif	/* __lint */
1970
1971/*
1972 * Replacement functions for ones that are normally inlined.
1973 * In addition to the copy in i86.il, they are defined here just in case.
1974 */
1975
1976#if defined(__lint)
1977
1978int
1979intr_clear(void)
1980{ return 0; }
1981
1982int
1983clear_int_flag(void)
1984{ return 0; }
1985
1986#else	/* __lint */
1987
1988#if defined(__amd64)
1989
1990	ENTRY(intr_clear)
1991	ENTRY(clear_int_flag)
1992	pushfq
1993	cli
1994	popq	%rax
1995	ret
1996	SET_SIZE(clear_int_flag)
1997	SET_SIZE(intr_clear)
1998
1999#elif defined(__i386)
2000
2001	ENTRY(intr_clear)
2002	ENTRY(clear_int_flag)
2003	pushfl
2004	cli
2005	popl	%eax
2006	ret
2007	SET_SIZE(clear_int_flag)
2008	SET_SIZE(intr_clear)
2009
2010#endif	/* __i386 */
2011#endif	/* __lint */
2012
2013#if defined(__lint)
2014
2015struct cpu *
2016curcpup(void)
2017{ return 0; }
2018
2019#else	/* __lint */
2020
2021#if defined(__amd64)
2022
2023	ENTRY(curcpup)
2024	movq	%gs:CPU_SELF, %rax
2025	ret
2026	SET_SIZE(curcpup)
2027
2028#elif defined(__i386)
2029
2030	ENTRY(curcpup)
2031	movl	%gs:CPU_SELF, %eax
2032	ret
2033	SET_SIZE(curcpup)
2034
2035#endif	/* __i386 */
2036#endif	/* __lint */
2037
2038#if defined(__lint)
2039
2040/* ARGSUSED */
2041uint32_t
2042htonl(uint32_t i)
2043{ return (0); }
2044
2045/* ARGSUSED */
2046uint32_t
2047ntohl(uint32_t i)
2048{ return (0); }
2049
2050#else	/* __lint */
2051
2052#if defined(__amd64)
2053
2054	/* XX64 there must be shorter sequences for this */
2055	ENTRY(htonl)
2056	ALTENTRY(ntohl)
2057	movl	%edi, %eax
2058	bswap	%eax
2059	ret
2060	SET_SIZE(ntohl)
2061	SET_SIZE(htonl)
2062
2063#elif defined(__i386)
2064
2065	ENTRY(htonl)
2066	ALTENTRY(ntohl)
2067	movl	4(%esp), %eax
2068	bswap	%eax
2069	ret
2070	SET_SIZE(ntohl)
2071	SET_SIZE(htonl)
2072
2073#endif	/* __i386 */
2074#endif	/* __lint */
2075
2076#if defined(__lint)
2077
2078/* ARGSUSED */
2079uint16_t
2080htons(uint16_t i)
2081{ return (0); }
2082
2083/* ARGSUSED */
2084uint16_t
2085ntohs(uint16_t i)
2086{ return (0); }
2087
2088
2089#else	/* __lint */
2090
2091#if defined(__amd64)
2092
2093	/* XX64 there must be better sequences for this */
2094	ENTRY(htons)
2095	ALTENTRY(ntohs)
2096	movl	%edi, %eax
2097	bswap	%eax
2098	shrl	$16, %eax
2099	ret
2100	SET_SIZE(ntohs)
2101	SET_SIZE(htons)
2102
2103#elif defined(__i386)
2104
2105	ENTRY(htons)
2106	ALTENTRY(ntohs)
2107	movl	4(%esp), %eax
2108	bswap	%eax
2109	shrl	$16, %eax
2110	ret
2111	SET_SIZE(ntohs)
2112	SET_SIZE(htons)
2113
2114#endif	/* __i386 */
2115#endif	/* __lint */
2116
2117
2118#if defined(__lint)
2119
2120/* ARGSUSED */
2121void
2122intr_restore(uint_t i)
2123{ return; }
2124
2125/* ARGSUSED */
2126void
2127restore_int_flag(int i)
2128{ return; }
2129
2130#else	/* __lint */
2131
2132#if defined(__amd64)
2133
2134	ENTRY(intr_restore)
2135	ENTRY(restore_int_flag)
2136	pushq	%rdi
2137	popfq
2138	ret
2139	SET_SIZE(restore_int_flag)
2140	SET_SIZE(intr_restore)
2141
2142#elif defined(__i386)
2143
2144	ENTRY(intr_restore)
2145	ENTRY(restore_int_flag)
2146	pushl	4(%esp)
2147	popfl
2148	ret
2149	SET_SIZE(restore_int_flag)
2150	SET_SIZE(intr_restore)
2151
2152#endif	/* __i386 */
2153#endif	/* __lint */
2154
2155#if defined(__lint)
2156
2157void
2158sti(void)
2159{}
2160
2161#else	/* __lint */
2162
2163	ENTRY(sti)
2164	sti
2165	ret
2166	SET_SIZE(sti)
2167
2168#endif	/* __lint */
2169
2170#if defined(__lint)
2171
2172dtrace_icookie_t
2173dtrace_interrupt_disable(void)
2174{ return (0); }
2175
2176#else   /* __lint */
2177
2178#if defined(__amd64)
2179
2180	ENTRY(dtrace_interrupt_disable)
2181	pushfq
2182	popq	%rax
2183	cli
2184	ret
2185	SET_SIZE(dtrace_interrupt_disable)
2186
2187#elif defined(__i386)
2188
2189	ENTRY(dtrace_interrupt_disable)
2190	pushfl
2191	popl	%eax
2192	cli
2193	ret
2194	SET_SIZE(dtrace_interrupt_disable)
2195
2196#endif	/* __i386 */
2197#endif	/* __lint */
2198
2199#if defined(__lint)
2200
2201/*ARGSUSED*/
2202void
2203dtrace_interrupt_enable(dtrace_icookie_t cookie)
2204{}
2205
2206#else	/* __lint */
2207
2208#if defined(__amd64)
2209
2210	ENTRY(dtrace_interrupt_enable)
2211	pushq	%rdi
2212	popfq
2213	ret
2214	SET_SIZE(dtrace_interrupt_enable)
2215
2216#elif defined(__i386)
2217
2218	ENTRY(dtrace_interrupt_enable)
2219	movl	4(%esp), %eax
2220	pushl	%eax
2221	popfl
2222	ret
2223	SET_SIZE(dtrace_interrupt_enable)
2224
2225#endif	/* __i386 */
2226#endif	/* __lint */
2227
2228
2229#if defined(lint)
2230
2231void
2232dtrace_membar_producer(void)
2233{}
2234
2235void
2236dtrace_membar_consumer(void)
2237{}
2238
2239#else	/* __lint */
2240
2241	ENTRY(dtrace_membar_producer)
2242	ret
2243	SET_SIZE(dtrace_membar_producer)
2244
2245	ENTRY(dtrace_membar_consumer)
2246	ret
2247	SET_SIZE(dtrace_membar_consumer)
2248
2249#endif	/* __lint */
2250
2251#if defined(__lint)
2252
2253kthread_id_t
2254threadp(void)
2255{ return ((kthread_id_t)0); }
2256
2257#else	/* __lint */
2258
2259#if defined(__amd64)
2260
2261	ENTRY(threadp)
2262	movq	%gs:CPU_THREAD, %rax
2263	ret
2264	SET_SIZE(threadp)
2265
2266#elif defined(__i386)
2267
2268	ENTRY(threadp)
2269	movl	%gs:CPU_THREAD, %eax
2270	ret
2271	SET_SIZE(threadp)
2272
2273#endif	/* __i386 */
2274#endif	/* __lint */
2275
2276/*
2277 *   Checksum routine for Internet Protocol Headers
2278 */
2279
2280#if defined(__lint)
2281
2282/* ARGSUSED */
2283unsigned int
2284ip_ocsum(
2285	ushort_t *address,	/* ptr to 1st message buffer */
2286	int halfword_count,	/* length of data */
2287	unsigned int sum)	/* partial checksum */
2288{
2289	int		i;
2290	unsigned int	psum = 0;	/* partial sum */
2291
2292	for (i = 0; i < halfword_count; i++, address++) {
2293		psum += *address;
2294	}
2295
2296	while ((psum >> 16) != 0) {
2297		psum = (psum & 0xffff) + (psum >> 16);
2298	}
2299
2300	psum += sum;
2301
2302	while ((psum >> 16) != 0) {
2303		psum = (psum & 0xffff) + (psum >> 16);
2304	}
2305
2306	return (psum);
2307}
2308
2309#else	/* __lint */
2310
2311#if defined(__amd64)
2312
2313	ENTRY(ip_ocsum)
2314	pushq	%rbp
2315	movq	%rsp, %rbp
2316#ifdef DEBUG
2317	movq	kernelbase(%rip), %rax
2318	cmpq	%rax, %rdi
2319	jnb	1f
2320	xorl	%eax, %eax
2321	movq	%rdi, %rsi
2322	leaq	.ip_ocsum_panic_msg(%rip), %rdi
2323	call	panic
2324	/*NOTREACHED*/
2325.ip_ocsum_panic_msg:
2326	.string	"ip_ocsum: address 0x%p below kernelbase\n"
23271:
2328#endif
2329	movl	%esi, %ecx	/* halfword_count */
2330	movq	%rdi, %rsi	/* address */
2331				/* partial sum in %edx */
2332	xorl	%eax, %eax
2333	testl	%ecx, %ecx
2334	jz	.ip_ocsum_done
2335	testq	$3, %rsi
2336	jnz	.ip_csum_notaligned
2337.ip_csum_aligned:	/* XX64 opportunities for 8-byte operations? */
2338.next_iter:
2339	/* XX64 opportunities for prefetch? */
2340	/* XX64 compute csum with 64 bit quantities? */
2341	subl	$32, %ecx
2342	jl	.less_than_32
2343
2344	addl	0(%rsi), %edx
2345.only60:
2346	adcl	4(%rsi), %eax
2347.only56:
2348	adcl	8(%rsi), %edx
2349.only52:
2350	adcl	12(%rsi), %eax
2351.only48:
2352	adcl	16(%rsi), %edx
2353.only44:
2354	adcl	20(%rsi), %eax
2355.only40:
2356	adcl	24(%rsi), %edx
2357.only36:
2358	adcl	28(%rsi), %eax
2359.only32:
2360	adcl	32(%rsi), %edx
2361.only28:
2362	adcl	36(%rsi), %eax
2363.only24:
2364	adcl	40(%rsi), %edx
2365.only20:
2366	adcl	44(%rsi), %eax
2367.only16:
2368	adcl	48(%rsi), %edx
2369.only12:
2370	adcl	52(%rsi), %eax
2371.only8:
2372	adcl	56(%rsi), %edx
2373.only4:
2374	adcl	60(%rsi), %eax	/* could be adding -1 and -1 with a carry */
2375.only0:
2376	adcl	$0, %eax	/* could be adding -1 in eax with a carry */
2377	adcl	$0, %eax
2378
2379	addq	$64, %rsi
2380	testl	%ecx, %ecx
2381	jnz	.next_iter
2382
2383.ip_ocsum_done:
2384	addl	%eax, %edx
2385	adcl	$0, %edx
2386	movl	%edx, %eax	/* form a 16 bit checksum by */
2387	shrl	$16, %eax	/* adding two halves of 32 bit checksum */
2388	addw	%dx, %ax
2389	adcw	$0, %ax
2390	andl	$0xffff, %eax
2391	leave
2392	ret
2393
2394.ip_csum_notaligned:
2395	xorl	%edi, %edi
2396	movw	(%rsi), %di
2397	addl	%edi, %edx
2398	adcl	$0, %edx
2399	addq	$2, %rsi
2400	decl	%ecx
2401	jmp	.ip_csum_aligned
2402
2403.less_than_32:
2404	addl	$32, %ecx
2405	testl	$1, %ecx
2406	jz	.size_aligned
2407	andl	$0xfe, %ecx
2408	movzwl	(%rsi, %rcx, 2), %edi
2409	addl	%edi, %edx
2410	adcl	$0, %edx
2411.size_aligned:
2412	movl	%ecx, %edi
2413	shrl	$1, %ecx
2414	shl	$1, %edi
2415	subq	$64, %rdi
2416	addq	%rdi, %rsi
2417	leaq    .ip_ocsum_jmptbl(%rip), %rdi
2418	leaq	(%rdi, %rcx, 8), %rdi
2419	xorl	%ecx, %ecx
2420	clc
2421	jmp 	*(%rdi)
2422
2423	.align	8
2424.ip_ocsum_jmptbl:
2425	.quad	.only0, .only4, .only8, .only12, .only16, .only20
2426	.quad	.only24, .only28, .only32, .only36, .only40, .only44
2427	.quad	.only48, .only52, .only56, .only60
2428	SET_SIZE(ip_ocsum)
2429
2430#elif defined(__i386)
2431
2432	ENTRY(ip_ocsum)
2433	pushl	%ebp
2434	movl	%esp, %ebp
2435	pushl	%ebx
2436	pushl	%esi
2437	pushl	%edi
2438	movl	12(%ebp), %ecx	/* count of half words */
2439	movl	16(%ebp), %edx	/* partial checksum */
2440	movl	8(%ebp), %esi
2441	xorl	%eax, %eax
2442	testl	%ecx, %ecx
2443	jz	.ip_ocsum_done
2444
2445	testl	$3, %esi
2446	jnz	.ip_csum_notaligned
2447.ip_csum_aligned:
2448.next_iter:
2449	subl	$32, %ecx
2450	jl	.less_than_32
2451
2452	addl	0(%esi), %edx
2453.only60:
2454	adcl	4(%esi), %eax
2455.only56:
2456	adcl	8(%esi), %edx
2457.only52:
2458	adcl	12(%esi), %eax
2459.only48:
2460	adcl	16(%esi), %edx
2461.only44:
2462	adcl	20(%esi), %eax
2463.only40:
2464	adcl	24(%esi), %edx
2465.only36:
2466	adcl	28(%esi), %eax
2467.only32:
2468	adcl	32(%esi), %edx
2469.only28:
2470	adcl	36(%esi), %eax
2471.only24:
2472	adcl	40(%esi), %edx
2473.only20:
2474	adcl	44(%esi), %eax
2475.only16:
2476	adcl	48(%esi), %edx
2477.only12:
2478	adcl	52(%esi), %eax
2479.only8:
2480	adcl	56(%esi), %edx
2481.only4:
2482	adcl	60(%esi), %eax	/* We could be adding -1 and -1 with a carry */
2483.only0:
2484	adcl	$0, %eax	/* we could be adding -1 in eax with a carry */
2485	adcl	$0, %eax
2486
2487	addl	$64, %esi
2488	andl	%ecx, %ecx
2489	jnz	.next_iter
2490
2491.ip_ocsum_done:
2492	addl	%eax, %edx
2493	adcl	$0, %edx
2494	movl	%edx, %eax	/* form a 16 bit checksum by */
2495	shrl	$16, %eax	/* adding two halves of 32 bit checksum */
2496	addw	%dx, %ax
2497	adcw	$0, %ax
2498	andl	$0xffff, %eax
2499	popl	%edi		/* restore registers */
2500	popl	%esi
2501	popl	%ebx
2502	leave
2503	ret
2504
2505.ip_csum_notaligned:
2506	xorl	%edi, %edi
2507	movw	(%esi), %di
2508	addl	%edi, %edx
2509	adcl	$0, %edx
2510	addl	$2, %esi
2511	decl	%ecx
2512	jmp	.ip_csum_aligned
2513
2514.less_than_32:
2515	addl	$32, %ecx
2516	testl	$1, %ecx
2517	jz	.size_aligned
2518	andl	$0xfe, %ecx
2519	movzwl	(%esi, %ecx, 2), %edi
2520	addl	%edi, %edx
2521	adcl	$0, %edx
2522.size_aligned:
2523	movl	%ecx, %edi
2524	shrl	$1, %ecx
2525	shl	$1, %edi
2526	subl	$64, %edi
2527	addl	%edi, %esi
2528	movl	$.ip_ocsum_jmptbl, %edi
2529	lea	(%edi, %ecx, 4), %edi
2530	xorl	%ecx, %ecx
2531	clc
2532	jmp 	*(%edi)
2533	SET_SIZE(ip_ocsum)
2534
2535	.data
2536	.align	4
2537
2538.ip_ocsum_jmptbl:
2539	.long	.only0, .only4, .only8, .only12, .only16, .only20
2540	.long	.only24, .only28, .only32, .only36, .only40, .only44
2541	.long	.only48, .only52, .only56, .only60
2542
2543
2544#endif	/* __i386 */
2545#endif	/* __lint */
2546
2547/*
2548 * multiply two long numbers and yield a u_longlong_t result, callable from C.
2549 * Provided to manipulate hrtime_t values.
2550 */
2551#if defined(__lint)
2552
2553/* result = a * b; */
2554
2555/* ARGSUSED */
2556unsigned long long
2557mul32(uint_t a, uint_t b)
2558{ return (0); }
2559
2560#else	/* __lint */
2561
2562#if defined(__amd64)
2563
2564	ENTRY(mul32)
2565	xorl	%edx, %edx	/* XX64 joe, paranoia? */
2566	movl	%edi, %eax
2567	mull	%esi
2568	shlq	$32, %rdx
2569	orq	%rdx, %rax
2570	ret
2571	SET_SIZE(mul32)
2572
2573#elif defined(__i386)
2574
2575	ENTRY(mul32)
2576	movl	8(%esp), %eax
2577	movl	4(%esp), %ecx
2578	mull	%ecx
2579	ret
2580	SET_SIZE(mul32)
2581
2582#endif	/* __i386 */
2583#endif	/* __lint */
2584
2585#if defined(notused)
2586#if defined(__lint)
2587/* ARGSUSED */
2588void
2589load_pte64(uint64_t *pte, uint64_t pte_value)
2590{}
2591#else	/* __lint */
2592	.globl load_pte64
2593load_pte64:
2594	movl	4(%esp), %eax
2595	movl	8(%esp), %ecx
2596	movl	12(%esp), %edx
2597	movl	%edx, 4(%eax)
2598	movl	%ecx, (%eax)
2599	ret
2600#endif	/* __lint */
2601#endif	/* notused */
2602
2603#if defined(__lint)
2604
2605/*ARGSUSED*/
2606void
2607scan_memory(caddr_t addr, size_t size)
2608{}
2609
2610#else	/* __lint */
2611
2612#if defined(__amd64)
2613
2614	ENTRY(scan_memory)
2615	shrq	$3, %rsi	/* convert %rsi from byte to quadword count */
2616	jz	.scanm_done
2617	movq	%rsi, %rcx	/* move count into rep control register */
2618	movq	%rdi, %rsi	/* move addr into lodsq control reg. */
2619	rep lodsq		/* scan the memory range */
2620.scanm_done:
2621	ret
2622	SET_SIZE(scan_memory)
2623
2624#elif defined(__i386)
2625
2626	ENTRY(scan_memory)
2627	pushl	%ecx
2628	pushl	%esi
2629	movl	16(%esp), %ecx	/* move 2nd arg into rep control register */
2630	shrl	$2, %ecx	/* convert from byte count to word count */
2631	jz	.scanm_done
2632	movl	12(%esp), %esi	/* move 1st arg into lodsw control register */
2633	.byte	0xf3		/* rep prefix.  lame assembler.  sigh. */
2634	lodsl
2635.scanm_done:
2636	popl	%esi
2637	popl	%ecx
2638	ret
2639	SET_SIZE(scan_memory)
2640
2641#endif	/* __i386 */
2642#endif	/* __lint */
2643
2644
2645#if defined(__lint)
2646
2647/*ARGSUSED */
2648int
2649lowbit(ulong_t i)
2650{ return (0); }
2651
2652#else	/* __lint */
2653
2654#if defined(__amd64)
2655
2656	ENTRY(lowbit)
2657	movl	$-1, %eax
2658	bsfq	%rdi, %rax
2659	incl	%eax
2660	ret
2661	SET_SIZE(lowbit)
2662
2663#elif defined(__i386)
2664
2665	ENTRY(lowbit)
2666	movl	$-1, %eax
2667	bsfl	4(%esp), %eax
2668	incl	%eax
2669	ret
2670	SET_SIZE(lowbit)
2671
2672#endif	/* __i386 */
2673#endif	/* __lint */
2674
2675#if defined(__lint)
2676
2677/*ARGSUSED*/
2678int
2679highbit(ulong_t i)
2680{ return (0); }
2681
2682#else	/* __lint */
2683
2684#if defined(__amd64)
2685
2686	ENTRY(highbit)
2687	movl	$-1, %eax
2688	bsrq	%rdi, %rax
2689	incl	%eax
2690	ret
2691	SET_SIZE(highbit)
2692
2693#elif defined(__i386)
2694
2695	ENTRY(highbit)
2696	movl	$-1, %eax
2697	bsrl	4(%esp), %eax
2698	incl	%eax
2699	ret
2700	SET_SIZE(highbit)
2701
2702#endif	/* __i386 */
2703#endif	/* __lint */
2704
2705#if defined(__lint)
2706
2707/*ARGSUSED*/
2708uint64_t
2709rdmsr(uint_t r, uint64_t *mtr)
2710{ return (0); }
2711
2712/*ARGSUSED*/
2713void
2714wrmsr(uint_t r, const uint64_t *mtr)
2715{}
2716
2717void
2718invalidate_cache(void)
2719{}
2720
2721#else  /* __lint */
2722
2723#if defined(__amd64)
2724
2725	ENTRY(rdmsr)
2726	movl	%edi, %ecx
2727	rdmsr
2728	movl	%eax, (%rsi)
2729	movl	%edx, 4(%rsi)
2730	shlq	$32, %rdx
2731	orq	%rdx, %rax
2732	ret
2733	SET_SIZE(rdmsr)
2734
2735	ENTRY(wrmsr)
2736	movl	(%rsi), %eax
2737	movl	4(%rsi), %edx
2738	movl	%edi, %ecx
2739	wrmsr
2740	ret
2741	SET_SIZE(wrmsr)
2742
2743#elif defined(__i386)
2744
2745	ENTRY(rdmsr)
2746	movl	4(%esp), %ecx
2747	rdmsr
2748	movl	8(%esp), %ecx
2749	movl	%eax, (%ecx)
2750	movl	%edx, 4(%ecx)
2751	ret
2752	SET_SIZE(rdmsr)
2753
2754	ENTRY(wrmsr)
2755	movl	8(%esp), %ecx
2756	movl	(%ecx), %eax
2757	movl	4(%ecx), %edx
2758	movl	4(%esp), %ecx
2759	wrmsr
2760	ret
2761	SET_SIZE(wrmsr)
2762
2763#endif	/* __i386 */
2764
2765	ENTRY(invalidate_cache)
2766	wbinvd
2767	ret
2768	SET_SIZE(invalidate_cache)
2769
2770#endif	/* __lint */
2771
2772#if defined(__lint)
2773
2774/*ARGSUSED*/
2775void getcregs(struct cregs *crp)
2776{}
2777
2778#else	/* __lint */
2779
2780#if defined(__amd64)
2781
2782#define	GETMSR(r, off, d)	\
2783	movl	$r, %ecx;	\
2784	rdmsr;			\
2785	movl	%eax, off(d);	\
2786	movl	%edx, off+4(d)
2787
2788	ENTRY_NP(getcregs)
2789	xorl	%eax, %eax
2790	movq	%rax, CREG_GDT+8(%rdi)
2791	sgdt	CREG_GDT(%rdi)		/* 10 bytes */
2792	movq	%rax, CREG_IDT+8(%rdi)
2793	sidt	CREG_IDT(%rdi)		/* 10 bytes */
2794	movq	%rax, CREG_LDT(%rdi)
2795	sldt	CREG_LDT(%rdi)		/* 2 bytes */
2796	movq	%rax, CREG_TASKR(%rdi)
2797	str	CREG_TASKR(%rdi)	/* 2 bytes */
2798	movq	%cr0, %rax
2799	movq	%rax, CREG_CR0(%rdi)	/* cr0 */
2800	movq	%cr2, %rax
2801	movq	%rax, CREG_CR2(%rdi)	/* cr2 */
2802	movq	%cr3, %rax
2803	movq	%rax, CREG_CR3(%rdi)	/* cr3 */
2804	movq	%cr4, %rax
2805	movq	%rax, CREG_CR8(%rdi)	/* cr4 */
2806	movq	%cr8, %rax
2807	movq	%rax, CREG_CR8(%rdi)	/* cr8 */
2808	GETMSR(MSR_AMD_KGSBASE, CREG_KGSBASE, %rdi)
2809	GETMSR(MSR_AMD_EFER, CREG_EFER, %rdi)
2810	SET_SIZE(getcregs)
2811
2812#undef GETMSR
2813
2814#elif defined(__i386)
2815
2816	ENTRY_NP(getcregs)
2817	movl	4(%esp), %edx
2818	movw	$0, CREG_GDT+6(%edx)
2819	movw	$0, CREG_IDT+6(%edx)
2820	sgdt	CREG_GDT(%edx)		/* gdt */
2821	sidt	CREG_IDT(%edx)		/* idt */
2822	sldt	CREG_LDT(%edx)		/* ldt */
2823	str	CREG_TASKR(%edx)	/* task */
2824	movl	%cr0, %eax
2825	movl	%eax, CREG_CR0(%edx)	/* cr0 */
2826	movl	%cr2, %eax
2827	movl	%eax, CREG_CR2(%edx)	/* cr2 */
2828	movl	%cr3, %eax
2829	movl	%eax, CREG_CR3(%edx)	/* cr3 */
2830	testl	$X86_LARGEPAGE, x86_feature
2831	jz	.nocr4
2832	movl	%cr4, %eax
2833	movl	%eax, CREG_CR4(%edx)	/* cr4 */
2834	jmp	.skip
2835.nocr4:
2836	movl	$0, CREG_CR4(%edx)
2837.skip:
2838	ret
2839	SET_SIZE(getcregs)
2840
2841#endif	/* __i386 */
2842#endif	/* __lint */
2843
2844
2845/*
2846 * A panic trigger is a word which is updated atomically and can only be set
2847 * once.  We atomically store 0xDEFACEDD and load the old value.  If the
2848 * previous value was 0, we succeed and return 1; otherwise return 0.
2849 * This allows a partially corrupt trigger to still trigger correctly.  DTrace
2850 * has its own version of this function to allow it to panic correctly from
2851 * probe context.
2852 */
2853#if defined(__lint)
2854
2855/*ARGSUSED*/
2856int
2857panic_trigger(int *tp)
2858{ return (0); }
2859
2860/*ARGSUSED*/
2861int
2862dtrace_panic_trigger(int *tp)
2863{ return (0); }
2864
2865#else	/* __lint */
2866
2867#if defined(__amd64)
2868
2869	ENTRY_NP(panic_trigger)
2870	xorl	%eax, %eax
2871	movl	$0xdefacedd, %edx
2872	lock
2873	  xchgl	%edx, (%rdi)
2874	cmpl	$0, %edx
2875	je	0f
2876	movl	$0, %eax
2877	ret
28780:	movl	$1, %eax
2879	ret
2880	SET_SIZE(panic_trigger)
2881
2882	ENTRY_NP(dtrace_panic_trigger)
2883	xorl	%eax, %eax
2884	movl	$0xdefacedd, %edx
2885	lock
2886	  xchgl	%edx, (%rdi)
2887	cmpl	$0, %edx
2888	je	0f
2889	movl	$0, %eax
2890	ret
28910:	movl	$1, %eax
2892	ret
2893	SET_SIZE(dtrace_panic_trigger)
2894
2895#elif defined(__i386)
2896
2897	ENTRY_NP(panic_trigger)
2898	movl	4(%esp), %edx		/ %edx = address of trigger
2899	movl	$0xdefacedd, %eax	/ %eax = 0xdefacedd
2900	lock				/ assert lock
2901	xchgl %eax, (%edx)		/ exchange %eax and the trigger
2902	cmpl	$0, %eax		/ if (%eax == 0x0)
2903	je	0f			/   return (1);
2904	movl	$0, %eax		/ else
2905	ret				/   return (0);
29060:	movl	$1, %eax
2907	ret
2908	SET_SIZE(panic_trigger)
2909
2910	ENTRY_NP(dtrace_panic_trigger)
2911	movl	4(%esp), %edx		/ %edx = address of trigger
2912	movl	$0xdefacedd, %eax	/ %eax = 0xdefacedd
2913	lock				/ assert lock
2914	xchgl %eax, (%edx)		/ exchange %eax and the trigger
2915	cmpl	$0, %eax		/ if (%eax == 0x0)
2916	je	0f			/   return (1);
2917	movl	$0, %eax		/ else
2918	ret				/   return (0);
29190:	movl	$1, %eax
2920	ret
2921	SET_SIZE(dtrace_panic_trigger)
2922
2923#endif	/* __i386 */
2924#endif	/* __lint */
2925
2926/*
2927 * The panic() and cmn_err() functions invoke vpanic() as a common entry point
2928 * into the panic code implemented in panicsys().  vpanic() is responsible
2929 * for passing through the format string and arguments, and constructing a
2930 * regs structure on the stack into which it saves the current register
2931 * values.  If we are not dying due to a fatal trap, these registers will
2932 * then be preserved in panicbuf as the current processor state.  Before
2933 * invoking panicsys(), vpanic() activates the first panic trigger (see
2934 * common/os/panic.c) and switches to the panic_stack if successful.  Note that
2935 * DTrace takes a slightly different panic path if it must panic from probe
2936 * context.  Instead of calling panic, it calls into dtrace_vpanic(), which
2937 * sets up the initial stack as vpanic does, calls dtrace_panic_trigger(), and
2938 * branches back into vpanic().
2939 */
2940#if defined(__lint)
2941
2942/*ARGSUSED*/
2943void
2944vpanic(const char *format, va_list alist)
2945{}
2946
2947/*ARGSUSED*/
2948void
2949dtrace_vpanic(const char *format, va_list alist)
2950{}
2951
2952#else	/* __lint */
2953
2954#if defined(__amd64)
2955
2956	ENTRY_NP(vpanic)			/* Initial stack layout: */
2957
2958	pushq	%rbp				/* | %rip | 	0x60	*/
2959	movq	%rsp, %rbp			/* | %rbp |	0x58	*/
2960	pushfq					/* | rfl  |	0x50	*/
2961	pushq	%r11				/* | %r11 |	0x48	*/
2962	pushq	%r10				/* | %r10 |	0x40	*/
2963	pushq	%rbx				/* | %rbx |	0x38	*/
2964	pushq	%rax				/* | %rax |	0x30	*/
2965	pushq	%r9				/* | %r9  |	0x28	*/
2966	pushq	%r8				/* | %r8  |	0x20	*/
2967	pushq	%rcx				/* | %rcx |	0x18	*/
2968	pushq	%rdx				/* | %rdx |	0x10	*/
2969	pushq	%rsi				/* | %rsi |	0x8 alist */
2970	pushq	%rdi				/* | %rdi |	0x0 format */
2971
2972	movq	%rsp, %rbx			/* %rbx = current %rsp */
2973
2974	leaq	panic_quiesce(%rip), %rdi	/* %rdi = &panic_quiesce */
2975	call	panic_trigger			/* %eax = panic_trigger() */
2976
2977vpanic_common:
2978	cmpl	$0, %eax
2979	je	0f
2980
2981	/*
2982	 * If panic_trigger() was successful, we are the first to initiate a
2983	 * panic: we now switch to the reserved panic_stack before continuing.
2984	 */
2985	leaq	panic_stack(%rip), %rsp
2986	addq	$PANICSTKSIZE, %rsp
29870:	subq	$REGSIZE, %rsp
2988	/*
2989	 * Now that we've got everything set up, store the register values as
2990	 * they were when we entered vpanic() to the designated location in
2991	 * the regs structure we allocated on the stack.
2992	 */
2993	movq	0x0(%rbx), %rcx
2994	movq	%rcx, REGOFF_RDI(%rsp)
2995	movq	0x8(%rbx), %rcx
2996	movq	%rcx, REGOFF_RSI(%rsp)
2997	movq	0x10(%rbx), %rcx
2998	movq	%rcx, REGOFF_RDX(%rsp)
2999	movq	0x18(%rbx), %rcx
3000	movq	%rcx, REGOFF_RCX(%rsp)
3001	movq	0x20(%rbx), %rcx
3002
3003	movq	%rcx, REGOFF_R8(%rsp)
3004	movq	0x28(%rbx), %rcx
3005	movq	%rcx, REGOFF_R9(%rsp)
3006	movq	0x30(%rbx), %rcx
3007	movq	%rcx, REGOFF_RAX(%rsp)
3008	movq	0x38(%rbx), %rcx
3009	movq	%rbx, REGOFF_RBX(%rsp)
3010	movq	0x58(%rbx), %rcx
3011
3012	movq	%rcx, REGOFF_RBP(%rsp)
3013	movq	0x40(%rbx), %rcx
3014	movq	%rcx, REGOFF_R10(%rsp)
3015	movq	0x48(%rbx), %rcx
3016	movq	%rcx, REGOFF_R11(%rsp)
3017	movq	%r12, REGOFF_R12(%rsp)
3018
3019	movq	%r13, REGOFF_R13(%rsp)
3020	movq	%r14, REGOFF_R14(%rsp)
3021	movq	%r15, REGOFF_R15(%rsp)
3022
3023	movl	$MSR_AMD_FSBASE, %ecx
3024	rdmsr
3025	movl	%eax, REGOFF_FSBASE(%rsp)
3026	movl	%edx, REGOFF_FSBASE+4(%rsp)
3027
3028	movl	$MSR_AMD_GSBASE, %ecx
3029	rdmsr
3030	movl	%eax, REGOFF_GSBASE(%rsp)
3031	movl	%edx, REGOFF_GSBASE+4(%rsp)
3032
3033	xorl	%ecx, %ecx
3034	movw	%ds, %cx
3035	movq	%rcx, REGOFF_DS(%rsp)
3036	movw	%es, %cx
3037	movq	%rcx, REGOFF_ES(%rsp)
3038	movw	%fs, %cx
3039	movq	%rcx, REGOFF_FS(%rsp)
3040	movw	%gs, %cx
3041	movq	%rcx, REGOFF_GS(%rsp)
3042
3043	movq	$0, REGOFF_TRAPNO(%rsp)
3044
3045	movq	$0, REGOFF_ERR(%rsp)
3046	leaq	vpanic(%rip), %rcx
3047	movq	%rcx, REGOFF_RIP(%rsp)
3048	movw	%cs, %cx
3049	movzwq	%cx, %rcx
3050	movq	%rcx, REGOFF_CS(%rsp)
3051	movq	0x50(%rbx), %rcx
3052	movq	%rcx, REGOFF_RFL(%rsp)
3053	movq	%rbx, %rcx
3054	addq	$0x60, %rcx
3055	movq	%rcx, REGOFF_RSP(%rsp)
3056	movw	%ss, %cx
3057	movzwq	%cx, %rcx
3058	movq	%rcx, REGOFF_SS(%rsp)
3059
3060	/*
3061	 * panicsys(format, alist, rp, on_panic_stack)
3062	 */
3063	movq	REGOFF_RDI(%rsp), %rdi		/* format */
3064	movq	REGOFF_RSI(%rsp), %rsi		/* alist */
3065	movq	%rsp, %rdx			/* struct regs */
3066	movl	%eax, %ecx			/* on_panic_stack */
3067	call	panicsys
3068	addq	$REGSIZE, %rsp
3069	popq	%rdi
3070	popq	%rsi
3071	popq	%rdx
3072	popq	%rcx
3073	popq	%r8
3074	popq	%r9
3075	popq	%rax
3076	popq	%rbx
3077	popq	%r10
3078	popq	%r11
3079	popfq
3080	leave
3081	ret
3082	SET_SIZE(vpanic)
3083
3084	ENTRY_NP(dtrace_vpanic)			/* Initial stack layout: */
3085
3086	pushq	%rbp				/* | %rip | 	0x60	*/
3087	movq	%rsp, %rbp			/* | %rbp |	0x58	*/
3088	pushfq					/* | rfl  |	0x50	*/
3089	pushq	%r11				/* | %r11 |	0x48	*/
3090	pushq	%r10				/* | %r10 |	0x40	*/
3091	pushq	%rbx				/* | %rbx |	0x38	*/
3092	pushq	%rax				/* | %rax |	0x30	*/
3093	pushq	%r9				/* | %r9  |	0x28	*/
3094	pushq	%r8				/* | %r8  |	0x20	*/
3095	pushq	%rcx				/* | %rcx |	0x18	*/
3096	pushq	%rdx				/* | %rdx |	0x10	*/
3097	pushq	%rsi				/* | %rsi |	0x8 alist */
3098	pushq	%rdi				/* | %rdi |	0x0 format */
3099
3100	movq	%rsp, %rbx			/* %rbx = current %rsp */
3101
3102	leaq	panic_quiesce(%rip), %rdi	/* %rdi = &panic_quiesce */
3103	call	dtrace_panic_trigger	/* %eax = dtrace_panic_trigger() */
3104	jmp	vpanic_common
3105
3106	SET_SIZE(dtrace_vpanic)
3107
3108#elif defined(__i386)
3109
3110	ENTRY_NP(vpanic)			/ Initial stack layout:
3111
3112	pushl	%ebp				/ | %eip | 20
3113	movl	%esp, %ebp			/ | %ebp | 16
3114	pushl	%eax				/ | %eax | 12
3115	pushl	%ebx				/ | %ebx |  8
3116	pushl	%ecx				/ | %ecx |  4
3117	pushl	%edx				/ | %edx |  0
3118
3119	movl	%esp, %ebx			/ %ebx = current stack pointer
3120
3121	lea	panic_quiesce, %eax		/ %eax = &panic_quiesce
3122	pushl	%eax				/ push &panic_quiesce
3123	call	panic_trigger			/ %eax = panic_trigger()
3124	addl	$4, %esp			/ reset stack pointer
3125
3126vpanic_common:
3127	cmpl	$0, %eax			/ if (%eax == 0)
3128	je	0f				/   goto 0f;
3129
3130	/*
3131	 * If panic_trigger() was successful, we are the first to initiate a
3132	 * panic: we now switch to the reserved panic_stack before continuing.
3133	 */
3134	lea	panic_stack, %esp		/ %esp  = panic_stack
3135	addl	$PANICSTKSIZE, %esp		/ %esp += PANICSTKSIZE
3136
31370:	subl	$REGSIZE, %esp			/ allocate struct regs
3138
3139	/*
3140	 * Now that we've got everything set up, store the register values as
3141	 * they were when we entered vpanic() to the designated location in
3142	 * the regs structure we allocated on the stack.
3143	 */
3144#if !defined(__GNUC_AS__)
3145	movw	%gs, %edx
3146	movl	%edx, REGOFF_GS(%esp)
3147	movw	%fs, %edx
3148	movl	%edx, REGOFF_FS(%esp)
3149	movw	%es, %edx
3150	movl	%edx, REGOFF_ES(%esp)
3151	movw	%ds, %edx
3152	movl	%edx, REGOFF_DS(%esp)
3153#else	/* __GNUC_AS__ */
3154	mov	%gs, %edx
3155	mov	%edx, REGOFF_GS(%esp)
3156	mov	%fs, %edx
3157	mov	%edx, REGOFF_FS(%esp)
3158	mov	%es, %edx
3159	mov	%edx, REGOFF_ES(%esp)
3160	mov	%ds, %edx
3161	mov	%edx, REGOFF_DS(%esp)
3162#endif	/* __GNUC_AS__ */
3163	movl	%edi, REGOFF_EDI(%esp)
3164	movl	%esi, REGOFF_ESI(%esp)
3165	movl	16(%ebx), %ecx
3166	movl	%ecx, REGOFF_EBP(%esp)
3167	movl	%ebx, %ecx
3168	addl	$20, %ecx
3169	movl	%ecx, REGOFF_ESP(%esp)
3170	movl	8(%ebx), %ecx
3171	movl	%ecx, REGOFF_EBX(%esp)
3172	movl	0(%ebx), %ecx
3173	movl	%ecx, REGOFF_EDX(%esp)
3174	movl	4(%ebx), %ecx
3175	movl	%ecx, REGOFF_ECX(%esp)
3176	movl	12(%ebx), %ecx
3177	movl	%ecx, REGOFF_EAX(%esp)
3178	movl	$0, REGOFF_TRAPNO(%esp)
3179	movl	$0, REGOFF_ERR(%esp)
3180	lea	vpanic, %ecx
3181	movl	%ecx, REGOFF_EIP(%esp)
3182#if !defined(__GNUC_AS__)
3183	movw	%cs, %edx
3184#else	/* __GNUC_AS__ */
3185	mov	%cs, %edx
3186#endif	/* __GNUC_AS__ */
3187	movl	%edx, REGOFF_CS(%esp)
3188	pushfl
3189	popl	%ecx
3190	movl	%ecx, REGOFF_EFL(%esp)
3191	movl	$0, REGOFF_UESP(%esp)
3192#if !defined(__GNUC_AS__)
3193	movw	%ss, %edx
3194#else	/* __GNUC_AS__ */
3195	mov	%ss, %edx
3196#endif	/* __GNUC_AS__ */
3197	movl	%edx, REGOFF_SS(%esp)
3198
3199	movl	%esp, %ecx			/ %ecx = &regs
3200	pushl	%eax				/ push on_panic_stack
3201	pushl	%ecx				/ push &regs
3202	movl	12(%ebp), %ecx			/ %ecx = alist
3203	pushl	%ecx				/ push alist
3204	movl	8(%ebp), %ecx			/ %ecx = format
3205	pushl	%ecx				/ push format
3206	call	panicsys			/ panicsys();
3207	addl	$16, %esp			/ pop arguments
3208
3209	addl	$REGSIZE, %esp
3210	popl	%edx
3211	popl	%ecx
3212	popl	%ebx
3213	popl	%eax
3214	leave
3215	ret
3216	SET_SIZE(vpanic)
3217
3218	ENTRY_NP(dtrace_vpanic)			/ Initial stack layout:
3219
3220	pushl	%ebp				/ | %eip | 20
3221	movl	%esp, %ebp			/ | %ebp | 16
3222	pushl	%eax				/ | %eax | 12
3223	pushl	%ebx				/ | %ebx |  8
3224	pushl	%ecx				/ | %ecx |  4
3225	pushl	%edx				/ | %edx |  0
3226
3227	movl	%esp, %ebx			/ %ebx = current stack pointer
3228
3229	lea	panic_quiesce, %eax		/ %eax = &panic_quiesce
3230	pushl	%eax				/ push &panic_quiesce
3231	call	dtrace_panic_trigger		/ %eax = dtrace_panic_trigger()
3232	addl	$4, %esp			/ reset stack pointer
3233	jmp	vpanic_common			/ jump back to common code
3234
3235	SET_SIZE(dtrace_vpanic)
3236
3237#endif	/* __i386 */
3238#endif	/* __lint */
3239
3240#if defined(__lint)
3241
3242void
3243hres_tick(void)
3244{}
3245
3246int64_t timedelta;
3247hrtime_t hres_last_tick;
3248timestruc_t hrestime;
3249int64_t hrestime_adj;
3250volatile int hres_lock;
3251uint_t nsec_scale;
3252hrtime_t hrtime_base;
3253
3254#else	/* __lint */
3255
3256	DGDEF3(hrestime, _MUL(2, CLONGSIZE), 8)
3257	.NWORD	0, 0
3258
3259	DGDEF3(hrestime_adj, 8, 8)
3260	.long	0, 0
3261
3262	DGDEF3(hres_last_tick, 8, 8)
3263	.long	0, 0
3264
3265	DGDEF3(timedelta, 8, 8)
3266	.long	0, 0
3267
3268	DGDEF3(hres_lock, 4, 8)
3269	.long	0
3270
3271	/*
3272	 * initialized to a non zero value to make pc_gethrtime()
3273	 * work correctly even before clock is initialized
3274	 */
3275	DGDEF3(hrtime_base, 8, 8)
3276	.long	_MUL(NSEC_PER_CLOCK_TICK, 6), 0
3277
3278	DGDEF3(adj_shift, 4, 4)
3279	.long	ADJ_SHIFT
3280
3281#if defined(__amd64)
3282
3283	ENTRY_NP(hres_tick)
3284	pushq	%rbp
3285	movq	%rsp, %rbp
3286
3287	/*
3288	 * We need to call *gethrtimef before picking up CLOCK_LOCK (obviously,
3289	 * hres_last_tick can only be modified while holding CLOCK_LOCK).
3290	 * At worst, performing this now instead of under CLOCK_LOCK may
3291	 * introduce some jitter in pc_gethrestime().
3292	 */
3293	call	*gethrtimef(%rip)
3294	movq	%rax, %r8
3295
3296	leaq	hres_lock(%rip), %rax
3297	movb	$-1, %dl
3298.CL1:
3299	xchgb	%dl, (%rax)
3300	testb	%dl, %dl
3301	jz	.CL3			/* got it */
3302.CL2:
3303	cmpb	$0, (%rax)		/* possible to get lock? */
3304	pause
3305	jne	.CL2
3306	jmp	.CL1			/* yes, try again */
3307.CL3:
3308	/*
3309	 * compute the interval since last time hres_tick was called
3310	 * and adjust hrtime_base and hrestime accordingly
3311	 * hrtime_base is an 8 byte value (in nsec), hrestime is
3312	 * a timestruc_t (sec, nsec)
3313	 */
3314	leaq	hres_last_tick(%rip), %rax
3315	movq	%r8, %r11
3316	subq	(%rax), %r8
3317	addq	%r8, hrtime_base(%rip)	/* add interval to hrtime_base */
3318	addq	%r8, hrestime+8(%rip)	/* add interval to hrestime.tv_nsec */
3319	/*
3320	 * Now that we have CLOCK_LOCK, we can update hres_last_tick
3321	 */
3322	movq	%r11, (%rax)
3323
3324	call	__adj_hrestime
3325
3326	/*
3327	 * release the hres_lock
3328	 */
3329	incl	hres_lock(%rip)
3330	leave
3331	ret
3332	SET_SIZE(hres_tick)
3333
3334#elif defined(__i386)
3335
3336	ENTRY_NP(hres_tick)
3337	pushl	%ebp
3338	movl	%esp, %ebp
3339	pushl	%esi
3340	pushl	%ebx
3341
3342	/*
3343	 * We need to call *gethrtimef before picking up CLOCK_LOCK (obviously,
3344	 * hres_last_tick can only be modified while holding CLOCK_LOCK).
3345	 * At worst, performing this now instead of under CLOCK_LOCK may
3346	 * introduce some jitter in pc_gethrestime().
3347	 */
3348	call	*gethrtimef
3349	movl	%eax, %ebx
3350	movl	%edx, %esi
3351
3352	movl	$hres_lock, %eax
3353	movl	$-1, %edx
3354.CL1:
3355	xchgb	%dl, (%eax)
3356	testb	%dl, %dl
3357	jz	.CL3			/ got it
3358.CL2:
3359	cmpb	$0, (%eax)		/ possible to get lock?
3360	pause
3361	jne	.CL2
3362	jmp	.CL1			/ yes, try again
3363.CL3:
3364	/*
3365	 * compute the interval since last time hres_tick was called
3366	 * and adjust hrtime_base and hrestime accordingly
3367	 * hrtime_base is an 8 byte value (in nsec), hrestime is
3368	 * timestruc_t (sec, nsec)
3369	 */
3370
3371	lea	hres_last_tick, %eax
3372
3373	movl	%ebx, %edx
3374	movl	%esi, %ecx
3375
3376	subl 	(%eax), %edx
3377	sbbl 	4(%eax), %ecx
3378
3379	addl	%edx, hrtime_base	/ add interval to hrtime_base
3380	adcl	%ecx, hrtime_base+4
3381
3382	addl 	%edx, hrestime+4	/ add interval to hrestime.tv_nsec
3383
3384	/
3385	/ Now that we have CLOCK_LOCK, we can update hres_last_tick.
3386	/
3387	movl	%ebx, (%eax)
3388	movl	%esi,  4(%eax)
3389
3390	/ get hrestime at this moment. used as base for pc_gethrestime
3391	/
3392	/ Apply adjustment, if any
3393	/
3394	/ #define HRES_ADJ	(NSEC_PER_CLOCK_TICK >> ADJ_SHIFT)
3395	/ (max_hres_adj)
3396	/
3397	/ void
3398	/ adj_hrestime()
3399	/ {
3400	/	long long adj;
3401	/
3402	/	if (hrestime_adj == 0)
3403	/		adj = 0;
3404	/	else if (hrestime_adj > 0) {
3405	/		if (hrestime_adj < HRES_ADJ)
3406	/			adj = hrestime_adj;
3407	/		else
3408	/			adj = HRES_ADJ;
3409	/	}
3410	/	else {
3411	/		if (hrestime_adj < -(HRES_ADJ))
3412	/			adj = -(HRES_ADJ);
3413	/		else
3414	/			adj = hrestime_adj;
3415	/	}
3416	/
3417	/	timedelta -= adj;
3418	/	hrestime_adj = timedelta;
3419	/	hrestime.tv_nsec += adj;
3420	/
3421	/	while (hrestime.tv_nsec >= NANOSEC) {
3422	/		one_sec++;
3423	/		hrestime.tv_sec++;
3424	/		hrestime.tv_nsec -= NANOSEC;
3425	/	}
3426	/ }
3427__adj_hrestime:
3428	movl	hrestime_adj, %esi	/ if (hrestime_adj == 0)
3429	movl	hrestime_adj+4, %edx
3430	andl	%esi, %esi
3431	jne	.CL4			/ no
3432	andl	%edx, %edx
3433	jne	.CL4			/ no
3434	subl	%ecx, %ecx		/ yes, adj = 0;
3435	subl	%edx, %edx
3436	jmp	.CL5
3437.CL4:
3438	subl	%ecx, %ecx
3439	subl	%eax, %eax
3440	subl	%esi, %ecx
3441	sbbl	%edx, %eax
3442	andl	%eax, %eax		/ if (hrestime_adj > 0)
3443	jge	.CL6
3444
3445	/ In the following comments, HRES_ADJ is used, while in the code
3446	/ max_hres_adj is used.
3447	/
3448	/ The test for "hrestime_adj < HRES_ADJ" is complicated because
3449	/ hrestime_adj is 64-bits, while HRES_ADJ is 32-bits.  We rely
3450	/ on the logical equivalence of:
3451	/
3452	/	!(hrestime_adj < HRES_ADJ)
3453	/
3454	/ and the two step sequence:
3455	/
3456	/	(HRES_ADJ - lsw(hrestime_adj)) generates a Borrow/Carry
3457	/
3458	/ which computes whether or not the least significant 32-bits
3459	/ of hrestime_adj is greater than HRES_ADJ, followed by:
3460	/
3461	/	Previous Borrow/Carry + -1 + msw(hrestime_adj) generates a Carry
3462	/
3463	/ which generates a carry whenever step 1 is true or the most
3464	/ significant long of the longlong hrestime_adj is non-zero.
3465
3466	movl	max_hres_adj, %ecx	/ hrestime_adj is positive
3467	subl	%esi, %ecx
3468	movl	%edx, %eax
3469	adcl	$-1, %eax
3470	jnc	.CL7
3471	movl	max_hres_adj, %ecx	/ adj = HRES_ADJ;
3472	subl	%edx, %edx
3473	jmp	.CL5
3474
3475	/ The following computation is similar to the one above.
3476	/
3477	/ The test for "hrestime_adj < -(HRES_ADJ)" is complicated because
3478	/ hrestime_adj is 64-bits, while HRES_ADJ is 32-bits.  We rely
3479	/ on the logical equivalence of:
3480	/
3481	/	(hrestime_adj > -HRES_ADJ)
3482	/
3483	/ and the two step sequence:
3484	/
3485	/	(HRES_ADJ + lsw(hrestime_adj)) generates a Carry
3486	/
3487	/ which means the least significant 32-bits of hrestime_adj is
3488	/ greater than -HRES_ADJ, followed by:
3489	/
3490	/	Previous Carry + 0 + msw(hrestime_adj) generates a Carry
3491	/
3492	/ which generates a carry only when step 1 is true and the most
3493	/ significant long of the longlong hrestime_adj is -1.
3494
3495.CL6:					/ hrestime_adj is negative
3496	movl	%esi, %ecx
3497	addl	max_hres_adj, %ecx
3498	movl	%edx, %eax
3499	adcl	$0, %eax
3500	jc	.CL7
3501	xor	%ecx, %ecx
3502	subl	max_hres_adj, %ecx	/ adj = -(HRES_ADJ);
3503	movl	$-1, %edx
3504	jmp	.CL5
3505.CL7:
3506	movl	%esi, %ecx		/ adj = hrestime_adj;
3507.CL5:
3508	movl	timedelta, %esi
3509	subl	%ecx, %esi
3510	movl	timedelta+4, %eax
3511	sbbl	%edx, %eax
3512	movl	%esi, timedelta
3513	movl	%eax, timedelta+4	/ timedelta -= adj;
3514	movl	%esi, hrestime_adj
3515	movl	%eax, hrestime_adj+4	/ hrestime_adj = timedelta;
3516	addl	hrestime+4, %ecx
3517
3518	movl	%ecx, %eax		/ eax = tv_nsec
35191:
3520	cmpl	$NANOSEC, %eax		/ if ((unsigned long)tv_nsec >= NANOSEC)
3521	jb	.CL8			/ no
3522	incl	one_sec			/ yes,  one_sec++;
3523	incl	hrestime		/ hrestime.tv_sec++;
3524	addl	$-NANOSEC, %eax		/ tv_nsec -= NANOSEC
3525	jmp	1b			/ check for more seconds
3526
3527.CL8:
3528	movl	%eax, hrestime+4	/ store final into hrestime.tv_nsec
3529	incl	hres_lock		/ release the hres_lock
3530
3531	popl	%ebx
3532	popl	%esi
3533	leave
3534	ret
3535	SET_SIZE(hres_tick)
3536
3537#endif	/* __i386 */
3538#endif	/* __lint */
3539
3540/*
3541 * void prefetch_smap_w(void *)
3542 *
3543 * Prefetch ahead within a linear list of smap structures.
3544 * Not implemented for ia32.  Stub for compatibility.
3545 */
3546
3547#if defined(__lint)
3548
3549/*ARGSUSED*/
3550void prefetch_smap_w(void *smp)
3551{}
3552
3553#else	/* __lint */
3554
3555	ENTRY(prefetch_smap_w)
3556	ret
3557	SET_SIZE(prefetch_smap_w)
3558
3559#endif	/* __lint */
3560
3561/*
3562 * prefetch_page_r(page_t *)
3563 * issue prefetch instructions for a page_t
3564 */
3565#if defined(__lint)
3566
3567/*ARGSUSED*/
3568void
3569prefetch_page_r(void *pp)
3570{}
3571
3572#else	/* __lint */
3573
3574	ENTRY(prefetch_page_r)
3575	ret
3576	SET_SIZE(prefetch_page_r)
3577
3578#endif	/* __lint */
3579
3580#if defined(__lint)
3581
3582/*ARGSUSED*/
3583int
3584bcmp(const void *s1, const void *s2, size_t count)
3585{ return (0); }
3586
3587#else   /* __lint */
3588
3589#if defined(__amd64)
3590
3591	ENTRY(bcmp)
3592	pushq	%rbp
3593	movq	%rsp, %rbp
3594#ifdef DEBUG
3595	movq	kernelbase(%rip), %r11
3596	cmpq	%r11, %rdi
3597	jb	0f
3598	cmpq	%r11, %rsi
3599	jnb	1f
36000:	leaq	.bcmp_panic_msg(%rip), %rdi
3601	xorl	%eax, %eax
3602	call	panic
36031:
3604#endif	/* DEBUG */
3605	call	memcmp
3606	testl	%eax, %eax
3607	setne	%dl
3608	leave
3609	movzbl	%dl, %eax
3610	ret
3611	SET_SIZE(bcmp)
3612
3613#elif defined(__i386)
3614
3615#define	ARG_S1		8
3616#define	ARG_S2		12
3617#define	ARG_LENGTH	16
3618
3619	ENTRY(bcmp)
3620#ifdef DEBUG
3621	pushl   %ebp
3622	movl    %esp, %ebp
3623	movl    kernelbase, %eax
3624	cmpl    %eax, ARG_S1(%ebp)
3625	jb	0f
3626	cmpl    %eax, ARG_S2(%ebp)
3627	jnb	1f
36280:	pushl   $.bcmp_panic_msg
3629	call    panic
36301:	popl    %ebp
3631#endif	/* DEBUG */
3632
3633	pushl	%edi		/ save register variable
3634	movl	ARG_S1(%esp), %eax	/ %eax = address of string 1
3635	movl	ARG_S2(%esp), %ecx	/ %ecx = address of string 2
3636	cmpl	%eax, %ecx	/ if the same string
3637	je	.equal		/ goto .equal
3638	movl	ARG_LENGTH(%esp), %edi	/ %edi = length in bytes
3639	cmpl	$4, %edi	/ if %edi < 4
3640	jb	.byte_check	/ goto .byte_check
3641	.align	4
3642.word_loop:
3643	movl	(%ecx), %edx	/ move 1 word from (%ecx) to %edx
3644	leal	-4(%edi), %edi	/ %edi -= 4
3645	cmpl	(%eax), %edx	/ compare 1 word from (%eax) with %edx
3646	jne	.word_not_equal	/ if not equal, goto .word_not_equal
3647	leal	4(%ecx), %ecx	/ %ecx += 4 (next word)
3648	leal	4(%eax), %eax	/ %eax += 4 (next word)
3649	cmpl	$4, %edi	/ if %edi >= 4
3650	jae	.word_loop	/ goto .word_loop
3651.byte_check:
3652	cmpl	$0, %edi	/ if %edi == 0
3653	je	.equal		/ goto .equal
3654	jmp	.byte_loop	/ goto .byte_loop (checks in bytes)
3655.word_not_equal:
3656	leal	4(%edi), %edi	/ %edi += 4 (post-decremented)
3657	.align	4
3658.byte_loop:
3659	movb	(%ecx),	%dl	/ move 1 byte from (%ecx) to %dl
3660	cmpb	%dl, (%eax)	/ compare %dl with 1 byte from (%eax)
3661	jne	.not_equal	/ if not equal, goto .not_equal
3662	incl	%ecx		/ %ecx++ (next byte)
3663	incl	%eax		/ %eax++ (next byte)
3664	decl	%edi		/ %edi--
3665	jnz	.byte_loop	/ if not zero, goto .byte_loop
3666.equal:
3667	xorl	%eax, %eax	/ %eax = 0
3668	popl	%edi		/ restore register variable
3669	ret			/ return (NULL)
3670	.align	4
3671.not_equal:
3672	movl	$1, %eax	/ return 1
3673	popl	%edi		/ restore register variable
3674	ret			/ return (NULL)
3675	SET_SIZE(bcmp)
3676
3677#endif	/* __i386 */
3678
3679#ifdef DEBUG
3680	.text
3681.bcmp_panic_msg:
3682	.string "bcmp: arguments below kernelbase"
3683#endif	/* DEBUG */
3684
3685#endif	/* __lint */
3686