2 * Copyright (c) 2003,2004,2008 The DragonFly Project. All rights reserved.
3 * Copyright (c) 2008 Jordan Gordeev.
5 * This code is derived from software contributed to The DragonFly Project
6 * by Matthew Dillon <dillon@backplane.com>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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
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.
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
35 * Copyright (c) 1990 The Regents of the University of California.
36 * All rights reserved.
38 * This code is derived from software contributed to Berkeley by
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
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.
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
69 * $FreeBSD: src/sys/i386/i386/swtch.s,v 1.89.2.10 2003/01/23 03:36:24 ps Exp $
72 //#include "use_npx.h"
74 #include <sys/rtprio.h>
76 #include <machine/asmacros.h>
77 #include <machine/segments.h>
79 #include <machine/pmap.h>
80 #include <machine/lock.h>
82 #define CHECKNZ(expr, scratch_reg) \
83 movq expr, scratch_reg; testq scratch_reg, scratch_reg; jnz 7f; int $3; 7:
87 #define MPLOCKED lock ;
92 .globl lwkt_switch_return
94 #if defined(SWTCH_OPTIM_STATS)
95 .globl swtch_optim_stats, tlb_flush_count
96 swtch_optim_stats: .long 0 /* number of _swtch_optims */
97 tlb_flush_count: .long 0
104 * cpu_heavy_switch(struct thread *next_thread)
106 * Switch from the current thread to a new thread. This entry
107 * is normally called via the thread->td_switch function, and will
108 * only be called when the current thread is a heavy weight process.
110 * Some instructions have been reordered to reduce pipeline stalls.
112 * YYY disable interrupts once giant is removed.
114 ENTRY(cpu_heavy_switch)
116 * Save RIP, RSP and callee-saved registers (RBX, RBP, R12-R15).
118 movq PCPU(curthread),%rcx
119 /* On top of the stack is the return adress. */
120 movq (%rsp),%rax /* (reorder optimization) */
121 movq TD_PCB(%rcx),%rdx /* RDX = PCB */
122 movq %rax,PCB_RIP(%rdx) /* return PC may be modified */
123 movq %rbx,PCB_RBX(%rdx)
124 movq %rsp,PCB_RSP(%rdx)
125 movq %rbp,PCB_RBP(%rdx)
126 movq %r12,PCB_R12(%rdx)
127 movq %r13,PCB_R13(%rdx)
128 movq %r14,PCB_R14(%rdx)
129 movq %r15,PCB_R15(%rdx)
132 * Clear the cpu bit in the pmap active mask. The restore
133 * function will set the bit in the pmap active mask.
135 * Special case: when switching between threads sharing the
136 * same vmspace if we avoid clearing the bit we do not have
137 * to reload %cr3 (if we clear the bit we could race page
138 * table ops done by other threads and would have to reload
139 * %cr3, because those ops will not know to IPI us).
141 movq %rcx,%rbx /* RBX = oldthread */
142 movq TD_LWP(%rcx),%rcx /* RCX = oldlwp */
143 movq TD_LWP(%rdi),%r13 /* R13 = newlwp */
144 movq LWP_VMSPACE(%rcx), %rcx /* RCX = oldvmspace */
145 testq %r13,%r13 /* might not be a heavy */
147 cmpq LWP_VMSPACE(%r13),%rcx /* same vmspace? */
150 movq PCPU(other_cpus)+0, %rax
151 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+0(%rcx)
152 movq PCPU(other_cpus)+8, %rax
153 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+8(%rcx)
154 movq PCPU(other_cpus)+16, %rax
155 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+16(%rcx)
156 movq PCPU(other_cpus)+24, %rax
157 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+24(%rcx)
161 * Push the LWKT switch restore function, which resumes a heavy
162 * weight process. Note that the LWKT switcher is based on
163 * TD_SP, while the heavy weight process switcher is based on
164 * PCB_RSP. TD_SP is usually two ints pushed relative to
165 * PCB_RSP. We push the flags for later restore by cpu_heavy_restore.
168 movq $cpu_heavy_restore, %rax
170 movq %rsp,TD_SP(%rbx)
173 * Save debug regs if necessary
175 movq PCB_FLAGS(%rdx),%rax
176 andq $PCB_DBREGS,%rax
177 jz 1f /* no, skip over */
178 movq %dr7,%rax /* yes, do the save */
179 movq %rax,PCB_DR7(%rdx)
180 /* JG correct value? */
181 andq $0x0000fc00, %rax /* disable all watchpoints */
184 movq %rax,PCB_DR6(%rdx)
186 movq %rax,PCB_DR3(%rdx)
188 movq %rax,PCB_DR2(%rdx)
190 movq %rax,PCB_DR1(%rdx)
192 movq %rax,PCB_DR0(%rdx)
197 * Save the FP state if we have used the FP. Note that calling
198 * npxsave will NULL out PCPU(npxthread).
200 cmpq %rbx,PCPU(npxthread)
202 movq %rdi,%r12 /* save %rdi. %r12 is callee-saved */
203 movq TD_SAVEFPU(%rbx),%rdi
204 call npxsave /* do it in a big C function */
205 movq %r12,%rdi /* restore %rdi */
210 * Switch to the next thread, which was passed as an argument
211 * to cpu_heavy_switch(). The argument is in %rdi.
212 * Set the current thread, load the stack pointer,
213 * and 'ret' into the switch-restore function.
215 * The switch restore function expects the new thread to be in %rax
216 * and the old one to be in %rbx.
218 * There is a one-instruction window where curthread is the new
219 * thread but %rsp still points to the old thread's stack, but
220 * we are protected by a critical section so it is ok.
222 movq %rdi,%rax /* RAX = newtd, RBX = oldtd */
223 movq %rax,PCPU(curthread)
224 movq TD_SP(%rax),%rsp
229 * cpu_exit_switch(struct thread *next)
231 * The switch function is changed to this when a thread is going away
232 * for good. We have to ensure that the MMU state is not cached, and
233 * we don't bother saving the existing thread state before switching.
235 * At this point we are in a critical section and this cpu owns the
236 * thread's token, which serves as an interlock until the switchout is
239 ENTRY(cpu_exit_switch)
241 * Get us out of the vmspace
248 /* JG no increment of statistics counters? see cpu_heavy_restore */
252 movq PCPU(curthread),%rbx
255 * If this is a process/lwp, deactivate the pmap after we've
258 movq TD_LWP(%rbx),%rcx
261 movq LWP_VMSPACE(%rcx), %rcx /* RCX = vmspace */
262 movq PCPU(other_cpus)+0, %rax
263 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+0(%rcx)
264 movq PCPU(other_cpus)+8, %rax
265 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+8(%rcx)
266 movq PCPU(other_cpus)+16, %rax
267 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+16(%rcx)
268 movq PCPU(other_cpus)+24, %rax
269 MPLOCKED andq %rax, VM_PMAP+PM_ACTIVE+24(%rcx)
272 * Switch to the next thread. RET into the restore function, which
273 * expects the new thread in RAX and the old in RBX.
275 * There is a one-instruction window where curthread is the new
276 * thread but %rsp still points to the old thread's stack, but
277 * we are protected by a critical section so it is ok.
280 movq %rax,PCPU(curthread)
281 movq TD_SP(%rax),%rsp
286 * cpu_heavy_restore() (current thread in %rax on entry)
288 * Restore the thread after an LWKT switch. This entry is normally
289 * called via the LWKT switch restore function, which was pulled
290 * off the thread stack and jumped to.
292 * This entry is only called if the thread was previously saved
293 * using cpu_heavy_switch() (the heavy weight process thread switcher),
294 * or when a new process is initially scheduled.
296 * NOTE: The lwp may be in any state, not necessarily LSRUN, because
297 * a preemption switch may interrupt the process and then return via
300 * YYY theoretically we do not have to restore everything here, a lot
301 * of this junk can wait until we return to usermode. But for now
302 * we restore everything.
304 * YYY the PCB crap is really crap, it makes startup a bitch because
305 * we can't switch away.
307 * YYY note: spl check is done in mi_switch when it splx()'s.
310 ENTRY(cpu_heavy_restore)
312 movq TD_PCB(%rax),%rdx /* RDX = PCB */
314 #if defined(SWTCH_OPTIM_STATS)
315 incl _swtch_optim_stats
318 * Tell the pmap that our cpu is using the VMSPACE now. We cannot
319 * safely test/reload %cr3 until after we have set the bit in the
320 * pmap (remember, we do not hold the MP lock in the switch code).
322 movq TD_LWP(%rax),%rcx
323 movq LWP_VMSPACE(%rcx), %rcx /* RCX = vmspace */
325 movq PCPU(other_cpus)+0, %rsi
326 MPLOCKED orq %rsi, VM_PMAP+PM_ACTIVE+0(%rcx)
327 movq PCPU(other_cpus)+8, %rsi
328 MPLOCKED orq %rsi, VM_PMAP+PM_ACTIVE+8(%rcx)
329 movq PCPU(other_cpus)+16, %rsi
330 MPLOCKED orq %rsi, VM_PMAP+PM_ACTIVE+16(%rcx)
331 movq PCPU(other_cpus)+24, %rsi
332 MPLOCKED orq %rsi, VM_PMAP+PM_ACTIVE+24(%rcx)
334 movl VM_PMAP+PM_ACTIVE_LOCK(%rcx),%esi
335 testl $CPULOCK_EXCL,%esi
338 movq %rax,%r12 /* save newthread ptr */
339 movq %rcx,%rdi /* (found to be set) */
340 call pmap_interlock_wait /* pmap_interlock_wait(%rdi:vm) */
342 movq TD_PCB(%rax),%rdx /* RDX = PCB */
345 * Restore the MMU address space. If it is the same as the last
346 * thread we don't have to invalidate the tlb (i.e. reload cr3).
347 * YYY which naturally also means that the PM_ACTIVE bit had better
348 * already have been set before we set it above, check? YYY
352 movq PCB_CR3(%rdx),%rcx
355 #if defined(SWTCH_OPTIM_STATS)
356 decl _swtch_optim_stats
357 incl _tlb_flush_count
363 * NOTE: %rbx is the previous thread and %rax is the new thread.
364 * %rbx is retained throughout so we can return it.
366 * lwkt_switch[_return] is responsible for handling TDF_RUNNING.
371 * Deal with the PCB extension, restore the private tss
373 movq PCB_EXT(%rdx),%rdi /* check for a PCB extension */
374 movq $1,%rcx /* maybe mark use of a private tss */
381 * Going back to the common_tss. We may need to update TSS_ESP0
382 * which sets the top of the supervisor stack when entering from
383 * usermode. The PCB is at the top of the stack but we need another
384 * 16 bytes to take vm86 into account.
387 movq %rcx, PCPU(common_tss) + TSS_RSP0
388 movq %rcx, PCPU(rsp0)
391 cmpl $0,PCPU(private_tss) /* don't have to reload if */
392 je 3f /* already using the common TSS */
395 subq %rcx,%rcx /* unmark use of private tss */
398 * Get the address of the common TSS descriptor for the ltr.
399 * There is no way to get the address of a segment-accessed variable
400 * so we store a self-referential pointer at the base of the per-cpu
401 * data area and add the appropriate offset.
404 movq $gd_common_tssd, %rdi
405 /* JG name for "%gs:0"? */
409 * Move the correct TSS descriptor into the GDT slot, then reload
414 movl %rcx,PCPU(private_tss) /* mark/unmark private tss */
415 movq PCPU(tss_gdt), %rcx /* entry in GDT */
418 movl $GPROC0_SEL*8, %esi /* GSEL(entry, SEL_KPL) */
426 * Restore the user %gs and %fs
428 movq PCB_FSBASE(%rdx),%r9
429 cmpq PCPU(user_fs),%r9
432 movq %r9,PCPU(user_fs)
433 movl $MSR_FSBASE,%ecx
434 movl PCB_FSBASE(%r10),%eax
435 movl PCB_FSBASE+4(%r10),%edx
439 movq PCB_GSBASE(%rdx),%r9
440 cmpq PCPU(user_gs),%r9
443 movq %r9,PCPU(user_gs)
444 movl $MSR_KGSBASE,%ecx /* later swapgs moves it to GSBASE */
445 movl PCB_GSBASE(%r10),%eax
446 movl PCB_GSBASE+4(%r10),%edx
453 * Restore general registers. %rbx is restored later.
455 movq PCB_RSP(%rdx), %rsp
456 movq PCB_RBP(%rdx), %rbp
457 movq PCB_R12(%rdx), %r12
458 movq PCB_R13(%rdx), %r13
459 movq PCB_R14(%rdx), %r14
460 movq PCB_R15(%rdx), %r15
461 movq PCB_RIP(%rdx), %rax
466 * Restore the user LDT if we have one
468 cmpl $0, PCB_USERLDT(%edx)
470 movl _default_ldt,%eax
471 cmpl PCPU(currentldt),%eax
474 movl %eax,PCPU(currentldt)
483 * Restore the user TLS if we have one
491 * Restore the DEBUG register state if necessary.
493 movq PCB_FLAGS(%rdx),%rax
494 andq $PCB_DBREGS,%rax
495 jz 1f /* no, skip over */
496 movq PCB_DR6(%rdx),%rax /* yes, do the restore */
498 movq PCB_DR3(%rdx),%rax
500 movq PCB_DR2(%rdx),%rax
502 movq PCB_DR1(%rdx),%rax
504 movq PCB_DR0(%rdx),%rax
506 movq %dr7,%rax /* load dr7 so as not to disturb */
507 /* JG correct value? */
508 andq $0x0000fc00,%rax /* reserved bits */
509 /* JG we've got more registers on x86_64 */
510 movq PCB_DR7(%rdx),%rcx
511 /* JG correct value? */
512 andq $~0x0000fc00,%rcx
517 movq PCB_RBX(%rdx),%rbx
523 * savectx(struct pcb *pcb)
525 * Update pcb, saving current processor state.
529 /* JG use %rdi instead of %rcx everywhere? */
532 /* caller's return address - child won't execute this routine */
534 movq %rax,PCB_RIP(%rcx)
535 movq %rbx,PCB_RBX(%rcx)
536 movq %rsp,PCB_RSP(%rcx)
537 movq %rbp,PCB_RBP(%rcx)
538 movq %r12,PCB_R12(%rcx)
539 movq %r13,PCB_R13(%rcx)
540 movq %r14,PCB_R14(%rcx)
541 movq %r15,PCB_R15(%rcx)
545 * If npxthread == NULL, then the npx h/w state is irrelevant and the
546 * state had better already be in the pcb. This is true for forks
547 * but not for dumps (the old book-keeping with FP flags in the pcb
548 * always lost for dumps because the dump pcb has 0 flags).
550 * If npxthread != NULL, then we have to save the npx h/w state to
551 * npxthread's pcb and copy it to the requested pcb, or save to the
552 * requested pcb and reload. Copying is easier because we would
553 * have to handle h/w bugs for reloading. We used to lose the
554 * parent's npx state for forks by forgetting to reload.
556 movq PCPU(npxthread),%rax
560 pushq %rcx /* target pcb */
561 movq TD_SAVEFPU(%rax),%rax /* originating savefpu area */
570 movq $PCB_SAVEFPU_SIZE,%rdx
571 leaq PCB_SAVEFPU(%rcx),%rcx
582 * cpu_idle_restore() (current thread in %rax on entry) (one-time execution)
584 * Don't bother setting up any regs other than %rbp so backtraces
585 * don't die. This restore function is used to bootstrap into the
586 * cpu_idle() LWKT only, after that cpu_lwkt_*() will be used for
589 * Clear TDF_RUNNING in old thread only after we've cleaned up %cr3.
590 * This only occurs during system boot so no special handling is
591 * required for migration.
593 * If we are an AP we have to call ap_init() before jumping to
594 * cpu_idle(). ap_init() will synchronize with the BP and finish
595 * setting up various ncpu-dependant globaldata fields. This may
596 * happen on UP as well as SMP if we happen to be simulating multiple
599 ENTRY(cpu_idle_restore)
607 andl $~TDF_RUNNING,TD_FLAGS(%rbx)
608 orl $TDF_RUNNING,TD_FLAGS(%rax) /* manual, no switch_return */
614 * cpu 0's idle thread entry for the first time must use normal
615 * lwkt_switch_return() semantics or a pending cpu migration on
616 * thread0 will deadlock.
621 call lwkt_switch_return
626 * cpu_kthread_restore() (current thread is %rax on entry) (one-time execution)
628 * Don't bother setting up any regs other then %rbp so backtraces
629 * don't die. This restore function is used to bootstrap into an
630 * LWKT based kernel thread only. cpu_lwkt_switch() will be used
633 * Because this switch target does not 'return' to lwkt_switch()
634 * we have to call lwkt_switch_return(otd) to clean up otd.
637 * Since all of our context is on the stack we are reentrant and
638 * we can release our critical section and enable interrupts early.
640 ENTRY(cpu_kthread_restore)
642 movq TD_PCB(%rax),%r13
646 * rax and rbx come from the switchout code. Call
647 * lwkt_switch_return(otd).
649 * NOTE: unlike i386, %rsi and %rdi are not call-saved regs.
653 call lwkt_switch_return
655 decl TD_CRITCOUNT(%rax)
656 movq PCB_R12(%r13),%rdi /* argument to RBX function */
657 movq PCB_RBX(%r13),%rax /* thread function */
658 /* note: top of stack return address inherited by function */
663 * cpu_lwkt_switch(struct thread *)
665 * Standard LWKT switching function. Only non-scratch registers are
666 * saved and we don't bother with the MMU state or anything else.
668 * This function is always called while in a critical section.
670 * There is a one-instruction window where curthread is the new
671 * thread but %rsp still points to the old thread's stack, but
672 * we are protected by a critical section so it is ok.
676 ENTRY(cpu_lwkt_switch)
677 pushq %rbp /* JG note: GDB hacked to locate ebp relative to td_sp */
678 /* JG we've got more registers on x86_64 */
680 movq PCPU(curthread),%rbx
689 * Save the FP state if we have used the FP. Note that calling
690 * npxsave will NULL out PCPU(npxthread).
692 * We have to deal with the FP state for LWKT threads in case they
693 * happen to get preempted or block while doing an optimized
694 * bzero/bcopy/memcpy.
696 cmpq %rbx,PCPU(npxthread)
698 movq %rdi,%r12 /* save %rdi. %r12 is callee-saved */
699 movq TD_SAVEFPU(%rbx),%rdi
700 call npxsave /* do it in a big C function */
701 movq %r12,%rdi /* restore %rdi */
705 movq %rdi,%rax /* switch to this thread */
706 pushq $cpu_lwkt_restore
707 movq %rsp,TD_SP(%rbx)
708 movq %rax,PCPU(curthread)
709 movq TD_SP(%rax),%rsp
712 * %rax contains new thread, %rbx contains old thread.
718 * cpu_lwkt_restore() (current thread in %rax on entry)
720 * Standard LWKT restore function. This function is always called
721 * while in a critical section.
723 * Warning: due to preemption the restore function can be used to
724 * 'return' to the original thread. Interrupt disablement must be
725 * protected through the switch so we cannot run splz here.
727 ENTRY(cpu_lwkt_restore)
729 * NOTE: %rbx is the previous thread and %eax is the new thread.
730 * %rbx is retained throughout so we can return it.
732 * lwkt_switch[_return] is responsible for handling TDF_RUNNING.
747 * Make AP become the idle loop.
749 ENTRY(bootstrap_idle)
750 movq PCPU(curthread),%rax
752 movq TD_SP(%rax),%rsp