1/*
2 * Copyright (c) 2003,2004,2008 The DragonFly Project.  All rights reserved.
3 * Copyright (c) 2008 Jordan Gordeev.
4 *
5 * This code is derived from software contributed to The DragonFly Project
6 * by Matthew Dillon <dillon@backplane.com>
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in
16 *    the documentation and/or other materials provided with the
17 *    distribution.
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 *    contributors may be used to endorse or promote products derived
20 *    from this software without specific, prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * Copyright (c) 1990 The Regents of the University of California.
36 * All rights reserved.
37 *
38 * This code is derived from software contributed to Berkeley by
39 * William Jolitz.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 *    notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 *    notice, this list of conditions and the following disclaimer in the
48 *    documentation and/or other materials provided with the distribution.
49 * 3. All advertising materials mentioning features or use of this software
50 *    must display the following acknowledgement:
51 *	This product includes software developed by the University of
52 *	California, Berkeley and its contributors.
53 * 4. Neither the name of the University nor the names of its contributors
54 *    may be used to endorse or promote products derived from this software
55 *    without specific prior written permission.
56 *
57 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 * SUCH DAMAGE.
68 *
69 * $FreeBSD: src/sys/i386/i386/swtch.s,v 1.89.2.10 2003/01/23 03:36:24 ps Exp $
70 * $DragonFly: src/sys/platform/pc64/amd64/swtch.s,v 1.3 2008/08/29 17:07:10 dillon Exp $
71 */
72
73//#include "use_npx.h"
74
75#include <sys/rtprio.h>
76
77#include <machine/asmacros.h>
78#include <machine/segments.h>
79
80#include <machine/pmap.h>
81#include <machine/lock.h>
82
83#define CHECKNZ(expr, scratch_reg) \
84	movq expr, scratch_reg; testq scratch_reg, scratch_reg; jnz 7f; int $3; 7:
85
86#include "assym.s"
87
88#if defined(SMP)
89#define MPLOCKED        lock ;
90#else
91#define MPLOCKED
92#endif
93
94	.data
95
96	.globl	panic
97
98#if defined(SWTCH_OPTIM_STATS)
99	.globl	swtch_optim_stats, tlb_flush_count
100swtch_optim_stats:	.long	0		/* number of _swtch_optims */
101tlb_flush_count:	.long	0
102#endif
103
104	.text
105
106
107/*
108 * cpu_heavy_switch(struct thread *next_thread)
109 *
110 *	Switch from the current thread to a new thread.  This entry
111 *	is normally called via the thread->td_switch function, and will
112 *	only be called when the current thread is a heavy weight process.
113 *
114 *	Some instructions have been reordered to reduce pipeline stalls.
115 *
116 *	YYY disable interrupts once giant is removed.
117 */
118ENTRY(cpu_heavy_switch)
119	/*
120	 * Save RIP, RSP and callee-saved registers (RBX, RBP, R12-R15).
121	 */
122	movq	PCPU(curthread),%rcx
123	/* On top of the stack is the return adress. */
124	movq	(%rsp),%rax			/* (reorder optimization) */
125	movq	TD_PCB(%rcx),%rdx		/* RDX = PCB */
126	movq	%rax,PCB_RIP(%rdx)		/* return PC may be modified */
127	movq	%rbx,PCB_RBX(%rdx)
128	movq	%rsp,PCB_RSP(%rdx)
129	movq	%rbp,PCB_RBP(%rdx)
130	movq	%r12,PCB_R12(%rdx)
131	movq	%r13,PCB_R13(%rdx)
132	movq	%r14,PCB_R14(%rdx)
133	movq	%r15,PCB_R15(%rdx)
134
135	movq	%rcx,%rbx			/* RBX = curthread */
136	movq	TD_LWP(%rcx),%rcx
137	movl	PCPU(cpuid), %eax
138	movq	LWP_VMSPACE(%rcx), %rcx		/* RCX = vmspace */
139	MPLOCKED btrl	%eax, VM_PMAP+PM_ACTIVE(%rcx)
140
141	/*
142	 * Push the LWKT switch restore function, which resumes a heavy
143	 * weight process.  Note that the LWKT switcher is based on
144	 * TD_SP, while the heavy weight process switcher is based on
145	 * PCB_RSP.  TD_SP is usually two ints pushed relative to
146	 * PCB_RSP.  We push the flags for later restore by cpu_heavy_restore.
147	 */
148	pushfq
149	movq	$cpu_heavy_restore, %rax
150	pushq	%rax
151	movq	%rsp,TD_SP(%rbx)
152
153	/*
154	 * Save debug regs if necessary
155	 */
156	movq    PCB_FLAGS(%rdx),%rax
157	andq    $PCB_DBREGS,%rax
158	jz      1f                              /* no, skip over */
159	movq    %dr7,%rax                       /* yes, do the save */
160	movq    %rax,PCB_DR7(%rdx)
161	/* JG correct value? */
162	andq    $0x0000fc00, %rax               /* disable all watchpoints */
163	movq    %rax,%dr7
164	movq    %dr6,%rax
165	movq    %rax,PCB_DR6(%rdx)
166	movq    %dr3,%rax
167	movq    %rax,PCB_DR3(%rdx)
168	movq    %dr2,%rax
169	movq    %rax,PCB_DR2(%rdx)
170	movq    %dr1,%rax
171	movq    %rax,PCB_DR1(%rdx)
172	movq    %dr0,%rax
173	movq    %rax,PCB_DR0(%rdx)
1741:
175
176#if 1
177	/*
178	 * Save the FP state if we have used the FP.  Note that calling
179	 * npxsave will NULL out PCPU(npxthread).
180	 */
181	cmpq	%rbx,PCPU(npxthread)
182	jne	1f
183	movq	%rdi,%r12		/* save %rdi. %r12 is callee-saved */
184	movq	TD_SAVEFPU(%rbx),%rdi
185	call	npxsave			/* do it in a big C function */
186	movq	%r12,%rdi		/* restore %rdi */
1871:
188#endif
189
190	/*
191	 * Switch to the next thread, which was passed as an argument
192	 * to cpu_heavy_switch().  The argument is in %rdi.
193	 * Set the current thread, load the stack pointer,
194	 * and 'ret' into the switch-restore function.
195	 *
196	 * The switch restore function expects the new thread to be in %rax
197	 * and the old one to be in %rbx.
198	 *
199	 * There is a one-instruction window where curthread is the new
200	 * thread but %rsp still points to the old thread's stack, but
201	 * we are protected by a critical section so it is ok.
202	 */
203	movq	%rdi,%rax		/* RAX = newtd, RBX = oldtd */
204	movq	%rax,PCPU(curthread)
205	movq	TD_SP(%rax),%rsp
206	CHECKNZ((%rsp), %r9)
207	ret
208
209/*
210 *  cpu_exit_switch(struct thread *next)
211 *
212 *	The switch function is changed to this when a thread is going away
213 *	for good.  We have to ensure that the MMU state is not cached, and
214 *	we don't bother saving the existing thread state before switching.
215 *
216 *	At this point we are in a critical section and this cpu owns the
217 *	thread's token, which serves as an interlock until the switchout is
218 *	complete.
219 */
220ENTRY(cpu_exit_switch)
221	/*
222	 * Get us out of the vmspace
223	 */
224#if 0
225	movq	KPML4phys,%rcx
226	movq	%cr3,%rax
227	cmpq	%rcx,%rax
228	je	1f
229	/* JG no increment of statistics counters? see cpu_heavy_restore */
230	movq	%rcx,%cr3
2311:
232#endif
233	movq	PCPU(curthread),%rbx
234
235	/*
236	 * If this is a process/lwp, deactivate the pmap after we've
237	 * switched it out.
238	 */
239	movq	TD_LWP(%rbx),%rcx
240	testq	%rcx,%rcx
241	jz	2f
242	movl	PCPU(cpuid), %eax
243	movq	LWP_VMSPACE(%rcx), %rcx		/* RCX = vmspace */
244	MPLOCKED btrl	%eax, VM_PMAP+PM_ACTIVE(%rcx)
2452:
246	/*
247	 * Switch to the next thread.  RET into the restore function, which
248	 * expects the new thread in RAX and the old in RBX.
249	 *
250	 * There is a one-instruction window where curthread is the new
251	 * thread but %rsp still points to the old thread's stack, but
252	 * we are protected by a critical section so it is ok.
253	 */
254	movq	%rdi,%rax
255	movq	%rax,PCPU(curthread)
256	movq	TD_SP(%rax),%rsp
257	CHECKNZ((%rsp), %r9)
258	ret
259
260/*
261 * cpu_heavy_restore()	(current thread in %rax on entry)
262 *
263 *	Restore the thread after an LWKT switch.  This entry is normally
264 *	called via the LWKT switch restore function, which was pulled
265 *	off the thread stack and jumped to.
266 *
267 *	This entry is only called if the thread was previously saved
268 *	using cpu_heavy_switch() (the heavy weight process thread switcher),
269 *	or when a new process is initially scheduled.  The first thing we
270 *	do is clear the TDF_RUNNING bit in the old thread and set it in the
271 *	new thread.
272 *
273 *	NOTE: The lwp may be in any state, not necessarily LSRUN, because
274 *	a preemption switch may interrupt the process and then return via
275 *	cpu_heavy_restore.
276 *
277 *	YYY theoretically we do not have to restore everything here, a lot
278 *	of this junk can wait until we return to usermode.  But for now
279 *	we restore everything.
280 *
281 *	YYY the PCB crap is really crap, it makes startup a bitch because
282 *	we can't switch away.
283 *
284 *	YYY note: spl check is done in mi_switch when it splx()'s.
285 */
286
287ENTRY(cpu_heavy_restore)
288	popfq
289	movq	TD_PCB(%rax),%rdx		/* RDX = PCB */
290	movq	TD_LWP(%rax),%rcx
291
292#if defined(SWTCH_OPTIM_STATS)
293	incl	_swtch_optim_stats
294#endif
295	/*
296	 * Tell the pmap that our cpu is using the VMSPACE now.  We cannot
297	 * safely test/reload %cr3 until after we have set the bit in the
298	 * pmap (remember, we do not hold the MP lock in the switch code).
299	 */
300	movq	LWP_VMSPACE(%rcx), %rcx		/* RCX = vmspace */
301	movl	PCPU(cpuid), %esi
302	MPLOCKED btsl	%esi, VM_PMAP+PM_ACTIVE(%rcx)
303
304	/*
305	 * Restore the MMU address space.  If it is the same as the last
306	 * thread we don't have to invalidate the tlb (i.e. reload cr3).
307	 * YYY which naturally also means that the PM_ACTIVE bit had better
308	 * already have been set before we set it above, check? YYY
309	 */
310#if 0
311	movq	%cr3,%rsi
312	movq	PCB_CR3(%rdx),%rcx
313	cmpq	%rsi,%rcx
314	je	4f
315#if defined(SWTCH_OPTIM_STATS)
316	decl	_swtch_optim_stats
317	incl	_tlb_flush_count
318#endif
319	movq	%rcx,%cr3
3204:
321#endif
322	/*
323	 * Clear TDF_RUNNING flag in old thread only after cleaning up
324	 * %cr3.  The target thread is already protected by being TDF_RUNQ
325	 * so setting TDF_RUNNING isn't as big a deal.
326	 */
327	andl	$~TDF_RUNNING,TD_FLAGS(%rbx)
328	orl	$TDF_RUNNING,TD_FLAGS(%rax)
329
330#if 0
331	/*
332	 * Deal with the PCB extension, restore the private tss
333	 */
334	movq	PCB_EXT(%rdx),%rdi	/* check for a PCB extension */
335	/* JG cheaper than "movq $1,%rbx", right? */
336	/* JG what's that magic value $1? */
337	movl	$1,%ebx			/* maybe mark use of a private tss */
338	testq	%rdi,%rdi
339#if JG
340	jnz	2f
341#endif
342
343	/* JG
344	 * Going back to the common_tss.  We may need to update TSS_ESP0
345	 * which sets the top of the supervisor stack when entering from
346	 * usermode.  The PCB is at the top of the stack but we need another
347	 * 16 bytes to take vm86 into account.
348	 */
349	leaq	-16(%rdx),%rbx
350	movq	%rbx, PCPU(common_tss) + TSS_RSP0
351	movq	%rbx, PCPU(rsp0)
352
353#if JG
354	cmpl	$0,PCPU(private_tss)	/* don't have to reload if      */
355	je	3f			/* already using the common TSS */
356
357	/* JG? */
358	subl	%ebx,%ebx		/* unmark use of private tss */
359
360	/*
361	 * Get the address of the common TSS descriptor for the ltr.
362	 * There is no way to get the address of a segment-accessed variable
363	 * so we store a self-referential pointer at the base of the per-cpu
364	 * data area and add the appropriate offset.
365	 */
366	/* JG movl? */
367	movq	$gd_common_tssd, %rdi
368	/* JG name for "%gs:0"? */
369	addq	%gs:0, %rdi
370
371	/*
372	 * Move the correct TSS descriptor into the GDT slot, then reload
373	 * ltr.
374	 */
3752:
376	/* JG */
377	movl	%ebx,PCPU(private_tss)		/* mark/unmark private tss */
378	movq	PCPU(tss_gdt), %rbx		/* entry in GDT */
379	movq	0(%rdi), %rax
380	movq	%rax, 0(%rbx)
381	movl	$GPROC0_SEL*8, %esi		/* GSEL(entry, SEL_KPL) */
382	ltr	%si
383#endif
384
3853:
386#endif
387#if 0
388	/*
389	 * Restore the user %gs and %fs
390	 */
391	movq	PCB_FSBASE(%rdx),%r9
392	cmpq	PCPU(user_fs),%r9
393	je	4f
394	movq	%rdx,%r10
395	movq	%r9,PCPU(user_fs)
396	movl	$MSR_FSBASE,%ecx
397	movl	PCB_FSBASE(%r10),%eax
398	movl	PCB_FSBASE+4(%r10),%edx
399	wrmsr
400	movq	%r10,%rdx
4014:
402	movq	PCB_GSBASE(%rdx),%r9
403	cmpq	PCPU(user_gs),%r9
404	je	5f
405	movq	%rdx,%r10
406	movq	%r9,PCPU(user_gs)
407	movl	$MSR_KGSBASE,%ecx	/* later swapgs moves it to GSBASE */
408	movl	PCB_GSBASE(%r10),%eax
409	movl	PCB_GSBASE+4(%r10),%edx
410	wrmsr
411	movq	%r10,%rdx
4125:
413#endif
414
415	/*
416	 * Restore general registers.
417	 */
418	movq	PCB_RBX(%rdx), %rbx
419	movq	PCB_RSP(%rdx), %rsp
420	movq	PCB_RBP(%rdx), %rbp
421	movq	PCB_R12(%rdx), %r12
422	movq	PCB_R13(%rdx), %r13
423	movq	PCB_R14(%rdx), %r14
424	movq	PCB_R15(%rdx), %r15
425	movq	PCB_RIP(%rdx), %rax
426	movq	%rax, (%rsp)
427
428#if 0
429	/*
430	 * Restore the user LDT if we have one
431	 */
432	cmpl	$0, PCB_USERLDT(%edx)
433	jnz	1f
434	movl	_default_ldt,%eax
435	cmpl	PCPU(currentldt),%eax
436	je	2f
437	lldt	_default_ldt
438	movl	%eax,PCPU(currentldt)
439	jmp	2f
4401:	pushl	%edx
441	call	set_user_ldt
442	popl	%edx
4432:
444#endif
445#if 0
446	/*
447	 * Restore the user TLS if we have one
448	 */
449	pushl	%edx
450	call	set_user_TLS
451	popl	%edx
452#endif
453
454	/*
455	 * Restore the DEBUG register state if necessary.
456	 */
457	movq    PCB_FLAGS(%rdx),%rax
458	andq    $PCB_DBREGS,%rax
459	jz      1f                              /* no, skip over */
460	movq    PCB_DR6(%rdx),%rax              /* yes, do the restore */
461	movq    %rax,%dr6
462	movq    PCB_DR3(%rdx),%rax
463	movq    %rax,%dr3
464	movq    PCB_DR2(%rdx),%rax
465	movq    %rax,%dr2
466	movq    PCB_DR1(%rdx),%rax
467	movq    %rax,%dr1
468	movq    PCB_DR0(%rdx),%rax
469	movq    %rax,%dr0
470	movq	%dr7,%rax                /* load dr7 so as not to disturb */
471	/* JG correct value? */
472	andq    $0x0000fc00,%rax         /*   reserved bits               */
473	/* JG we've got more registers on amd64 */
474	pushq   %rbx
475	movq    PCB_DR7(%rdx),%rbx
476	/* JG correct value? */
477	andq	$~0x0000fc00,%rbx
478	orq     %rbx,%rax
479	popq	%rbx
480	movq    %rax,%dr7
4811:
482
483	CHECKNZ((%rsp), %r9)
484	ret
485
486/*
487 * savectx(struct pcb *pcb)
488 *
489 * Update pcb, saving current processor state.
490 */
491ENTRY(savectx)
492	/* fetch PCB */
493	/* JG use %rdi instead of %rcx everywhere? */
494	movq	%rdi,%rcx
495
496	/* caller's return address - child won't execute this routine */
497	movq	(%rsp),%rax
498	movq	%rax,PCB_RIP(%rcx)
499	movq	%rbx,PCB_RBX(%rcx)
500	movq	%rsp,PCB_RSP(%rcx)
501	movq	%rbp,PCB_RBP(%rcx)
502	movq	%r12,PCB_R12(%rcx)
503	movq	%r13,PCB_R13(%rcx)
504	movq	%r14,PCB_R14(%rcx)
505	movq	%r15,PCB_R15(%rcx)
506
507#if 1
508	/*
509	 * If npxthread == NULL, then the npx h/w state is irrelevant and the
510	 * state had better already be in the pcb.  This is true for forks
511	 * but not for dumps (the old book-keeping with FP flags in the pcb
512	 * always lost for dumps because the dump pcb has 0 flags).
513	 *
514	 * If npxthread != NULL, then we have to save the npx h/w state to
515	 * npxthread's pcb and copy it to the requested pcb, or save to the
516	 * requested pcb and reload.  Copying is easier because we would
517	 * have to handle h/w bugs for reloading.  We used to lose the
518	 * parent's npx state for forks by forgetting to reload.
519	 */
520	movq	PCPU(npxthread),%rax
521	testq	%rax,%rax
522	jz	1f
523
524	pushq	%rcx			/* target pcb */
525	movq	TD_SAVEFPU(%rax),%rax	/* originating savefpu area */
526	pushq	%rax
527
528	movq	%rax,%rdi
529	call	npxsave
530
531	popq	%rax
532	popq	%rcx
533
534	movq	$PCB_SAVEFPU_SIZE,%rdx
535	leaq    PCB_SAVEFPU(%rcx),%rcx
536	movq	%rcx,%rsi
537	movq	%rax,%rdi
538	call	bcopy
539#endif
540
5411:
542	CHECKNZ((%rsp), %r9)
543	ret
544
545/*
546 * cpu_idle_restore()	(current thread in %rax on entry) (one-time execution)
547 *
548 *	Don't bother setting up any regs other than %rbp so backtraces
549 *	don't die.  This restore function is used to bootstrap into the
550 *	cpu_idle() LWKT only, after that cpu_lwkt_*() will be used for
551 *	switching.
552 *
553 *	Clear TDF_RUNNING in old thread only after we've cleaned up %cr3.
554 *
555 *	If we are an AP we have to call ap_init() before jumping to
556 *	cpu_idle().  ap_init() will synchronize with the BP and finish
557 *	setting up various ncpu-dependant globaldata fields.  This may
558 *	happen on UP as well as SMP if we happen to be simulating multiple
559 *	cpus.
560 */
561ENTRY(cpu_idle_restore)
562	/* cli */
563	/* JG xor? */
564	movl	$0,%ebp
565	/* JG push RBP? */
566	pushq	$0
567	andl	$~TDF_RUNNING,TD_FLAGS(%rbx)
568	orl	$TDF_RUNNING,TD_FLAGS(%rax)
569#ifdef SMP
570	cmpl	$0,PCPU(cpuid)
571	je	1f
572	call	ap_init
5731:
574#endif
575	/* sti */
576	jmp	cpu_idle
577
578/*
579 * cpu_kthread_restore() (current thread is %rax on entry) (one-time execution)
580 *
581 *	Don't bother setting up any regs other then %rbp so backtraces
582 *	don't die.  This restore function is used to bootstrap into an
583 *	LWKT based kernel thread only.  cpu_lwkt_switch() will be used
584 *	after this.
585 *
586 *	Since all of our context is on the stack we are reentrant and
587 *	we can release our critical section and enable interrupts early.
588 */
589ENTRY(cpu_kthread_restore)
590	/*sti*/
591	movq	TD_PCB(%rax),%rdx
592	/* JG "movq $0, %rbp"? "xorq %rbp, %rbp"? */
593	movl	$0,%ebp
594	/* rax and rbx come from the switchout code */
595	andl	$~TDF_RUNNING,TD_FLAGS(%rbx)
596	orl	$TDF_RUNNING,TD_FLAGS(%rax)
597	subl	$TDPRI_CRIT,TD_PRI(%rax)
598	movq	PCB_R12(%rdx),%rdi	/* argument to RBX function */
599	movq	PCB_RBX(%rdx),%rax	/* thread function */
600	/* note: top of stack return address inherited by function */
601	CHECKNZ(%rax, %r9)
602	jmp	*%rax
603
604/*
605 * cpu_lwkt_switch(struct thread *)
606 *
607 *	Standard LWKT switching function.  Only non-scratch registers are
608 *	saved and we don't bother with the MMU state or anything else.
609 *
610 *	This function is always called while in a critical section.
611 *
612 *	There is a one-instruction window where curthread is the new
613 *	thread but %rsp still points to the old thread's stack, but
614 *	we are protected by a critical section so it is ok.
615 *
616 *	YYY BGL, SPL
617 */
618ENTRY(cpu_lwkt_switch)
619	pushq	%rbp	/* JG note: GDB hacked to locate ebp relative to td_sp */
620	/* JG we've got more registers on AMD64 */
621	pushq	%rbx
622	movq	PCPU(curthread),%rbx
623	pushq	%r12
624	pushq	%r13
625	pushq	%r14
626	pushq	%r15
627	pushfq
628
629#if 1
630	/*
631	 * Save the FP state if we have used the FP.  Note that calling
632	 * npxsave will NULL out PCPU(npxthread).
633	 *
634	 * We have to deal with the FP state for LWKT threads in case they
635	 * happen to get preempted or block while doing an optimized
636	 * bzero/bcopy/memcpy.
637	 */
638	cmpq	%rbx,PCPU(npxthread)
639	jne	1f
640	movq	%rdi,%r12		/* save %rdi. %r12 is callee-saved */
641	movq	TD_SAVEFPU(%rbx),%rdi
642	call	npxsave			/* do it in a big C function */
643	movq	%r12,%rdi		/* restore %rdi */
6441:
645#endif
646
647	movq	%rdi,%rax		/* switch to this thread */
648	pushq	$cpu_lwkt_restore
649	movq	%rsp,TD_SP(%rbx)
650	movq	%rax,PCPU(curthread)
651	movq	TD_SP(%rax),%rsp
652
653	/*
654	 * %rax contains new thread, %rbx contains old thread.
655	 */
656	CHECKNZ((%rsp), %r9)
657	ret
658
659/*
660 * cpu_lwkt_restore()	(current thread in %rax on entry)
661 *
662 *	Standard LWKT restore function.  This function is always called
663 *	while in a critical section.
664 *
665 *	Warning: due to preemption the restore function can be used to
666 *	'return' to the original thread.  Interrupt disablement must be
667 *	protected through the switch so we cannot run splz here.
668 */
669ENTRY(cpu_lwkt_restore)
670	andl	$~TDF_RUNNING,TD_FLAGS(%rbx)
671	orl	$TDF_RUNNING,TD_FLAGS(%rax)
672	popfq
673	popq	%r15
674	popq	%r14
675	popq	%r13
676	popq	%r12
677	popq	%rbx
678	popq	%rbp
679	ret
680
681/*
682 * bootstrap_idle()
683 *
684 * Make AP become the idle loop.
685 */
686ENTRY(bootstrap_idle)
687	movq	PCPU(curthread),%rax
688	movq	%rax,%rbx
689	movq	TD_SP(%rax),%rsp
690	ret
691