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:
88 #define MPLOCKED lock ;
96 .globl lwkt_switch_return
98 #if defined(SWTCH_OPTIM_STATS)
99 .globl swtch_optim_stats, tlb_flush_count
100 swtch_optim_stats: .long 0 /* number of _swtch_optims */
101 tlb_flush_count: .long 0
108 * cpu_heavy_switch(struct thread *next_thread)
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.
114 * Some instructions have been reordered to reduce pipeline stalls.
116 * YYY disable interrupts once giant is removed.
118 ENTRY(cpu_heavy_switch)
120 * Save RIP, RSP and callee-saved registers (RBX, RBP, R12-R15).
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)
135 movq %rcx,%rbx /* RBX = curthread */
136 movq TD_LWP(%rcx),%rcx
137 movslq PCPU(cpuid), %rax
138 movq LWP_VMSPACE(%rcx), %rcx /* RCX = vmspace */
139 MPLOCKED btrq %rax, VM_PMAP+PM_ACTIVE(%rcx)
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.
149 movq $cpu_heavy_restore, %rax
151 movq %rsp,TD_SP(%rbx)
154 * Save debug regs if necessary
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 */
165 movq %rax,PCB_DR6(%rdx)
167 movq %rax,PCB_DR3(%rdx)
169 movq %rax,PCB_DR2(%rdx)
171 movq %rax,PCB_DR1(%rdx)
173 movq %rax,PCB_DR0(%rdx)
178 * Save the FP state if we have used the FP. Note that calling
179 * npxsave will NULL out PCPU(npxthread).
181 cmpq %rbx,PCPU(npxthread)
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 */
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.
196 * The switch restore function expects the new thread to be in %rax
197 * and the old one to be in %rbx.
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.
203 movq %rdi,%rax /* RAX = newtd, RBX = oldtd */
204 movq %rax,PCPU(curthread)
205 movq TD_SP(%rax),%rsp
210 * cpu_exit_switch(struct thread *next)
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.
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
220 ENTRY(cpu_exit_switch)
222 * Get us out of the vmspace
229 /* JG no increment of statistics counters? see cpu_heavy_restore */
233 movq PCPU(curthread),%rbx
236 * If this is a process/lwp, deactivate the pmap after we've
239 movq TD_LWP(%rbx),%rcx
242 movslq PCPU(cpuid), %rax
243 movq LWP_VMSPACE(%rcx), %rcx /* RCX = vmspace */
244 MPLOCKED btrq %rax, VM_PMAP+PM_ACTIVE(%rcx)
247 * Switch to the next thread. RET into the restore function, which
248 * expects the new thread in RAX and the old in RBX.
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.
255 movq %rax,PCPU(curthread)
256 movq TD_SP(%rax),%rsp
261 * cpu_heavy_restore() (current thread in %rax on entry)
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.
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.
271 * NOTE: The lwp may be in any state, not necessarily LSRUN, because
272 * a preemption switch may interrupt the process and then return via
275 * YYY theoretically we do not have to restore everything here, a lot
276 * of this junk can wait until we return to usermode. But for now
277 * we restore everything.
279 * YYY the PCB crap is really crap, it makes startup a bitch because
280 * we can't switch away.
282 * YYY note: spl check is done in mi_switch when it splx()'s.
285 ENTRY(cpu_heavy_restore)
287 movq TD_PCB(%rax),%rdx /* RDX = PCB */
288 movq TD_LWP(%rax),%rcx
290 #if defined(SWTCH_OPTIM_STATS)
291 incl _swtch_optim_stats
294 * Tell the pmap that our cpu is using the VMSPACE now. We cannot
295 * safely test/reload %cr3 until after we have set the bit in the
296 * pmap (remember, we do not hold the MP lock in the switch code).
298 movq LWP_VMSPACE(%rcx), %rcx /* RCX = vmspace */
299 movslq PCPU(cpuid), %rsi
300 MPLOCKED btsq %rsi, VM_PMAP+PM_ACTIVE(%rcx)
303 * Restore the MMU address space. If it is the same as the last
304 * thread we don't have to invalidate the tlb (i.e. reload cr3).
305 * YYY which naturally also means that the PM_ACTIVE bit had better
306 * already have been set before we set it above, check? YYY
310 movq PCB_CR3(%rdx),%rcx
313 #if defined(SWTCH_OPTIM_STATS)
314 decl _swtch_optim_stats
315 incl _tlb_flush_count
321 * NOTE: %rbx is the previous thread and %eax is the new thread.
322 * %rbx is retained throughout so we can return it.
324 * lwkt_switch[_return] is responsible for handling TDF_RUNNING.
329 * Deal with the PCB extension, restore the private tss
331 movq PCB_EXT(%rdx),%rdi /* check for a PCB extension */
332 movq $1,%rcx /* maybe mark use of a private tss */
339 * Going back to the common_tss. We may need to update TSS_ESP0
340 * which sets the top of the supervisor stack when entering from
341 * usermode. The PCB is at the top of the stack but we need another
342 * 16 bytes to take vm86 into account.
345 movq %rcx, PCPU(common_tss) + TSS_RSP0
346 movq %rcx, PCPU(rsp0)
349 cmpl $0,PCPU(private_tss) /* don't have to reload if */
350 je 3f /* already using the common TSS */
353 subq %rcx,%rcx /* unmark use of private tss */
356 * Get the address of the common TSS descriptor for the ltr.
357 * There is no way to get the address of a segment-accessed variable
358 * so we store a self-referential pointer at the base of the per-cpu
359 * data area and add the appropriate offset.
362 movq $gd_common_tssd, %rdi
363 /* JG name for "%gs:0"? */
367 * Move the correct TSS descriptor into the GDT slot, then reload
372 movl %rcx,PCPU(private_tss) /* mark/unmark private tss */
373 movq PCPU(tss_gdt), %rcx /* entry in GDT */
376 movl $GPROC0_SEL*8, %esi /* GSEL(entry, SEL_KPL) */
384 * Restore the user %gs and %fs
386 movq PCB_FSBASE(%rdx),%r9
387 cmpq PCPU(user_fs),%r9
390 movq %r9,PCPU(user_fs)
391 movl $MSR_FSBASE,%ecx
392 movl PCB_FSBASE(%r10),%eax
393 movl PCB_FSBASE+4(%r10),%edx
397 movq PCB_GSBASE(%rdx),%r9
398 cmpq PCPU(user_gs),%r9
401 movq %r9,PCPU(user_gs)
402 movl $MSR_KGSBASE,%ecx /* later swapgs moves it to GSBASE */
403 movl PCB_GSBASE(%r10),%eax
404 movl PCB_GSBASE+4(%r10),%edx
411 * Restore general registers. %rbx is restored later.
413 movq PCB_RSP(%rdx), %rsp
414 movq PCB_RBP(%rdx), %rbp
415 movq PCB_R12(%rdx), %r12
416 movq PCB_R13(%rdx), %r13
417 movq PCB_R14(%rdx), %r14
418 movq PCB_R15(%rdx), %r15
419 movq PCB_RIP(%rdx), %rax
424 * Restore the user LDT if we have one
426 cmpl $0, PCB_USERLDT(%edx)
428 movl _default_ldt,%eax
429 cmpl PCPU(currentldt),%eax
432 movl %eax,PCPU(currentldt)
441 * Restore the user TLS if we have one
449 * Restore the DEBUG register state if necessary.
451 movq PCB_FLAGS(%rdx),%rax
452 andq $PCB_DBREGS,%rax
453 jz 1f /* no, skip over */
454 movq PCB_DR6(%rdx),%rax /* yes, do the restore */
456 movq PCB_DR3(%rdx),%rax
458 movq PCB_DR2(%rdx),%rax
460 movq PCB_DR1(%rdx),%rax
462 movq PCB_DR0(%rdx),%rax
464 movq %dr7,%rax /* load dr7 so as not to disturb */
465 /* JG correct value? */
466 andq $0x0000fc00,%rax /* reserved bits */
467 /* JG we've got more registers on x86_64 */
468 movq PCB_DR7(%rdx),%rcx
469 /* JG correct value? */
470 andq $~0x0000fc00,%rcx
475 movq PCB_RBX(%rdx),%rbx
481 * savectx(struct pcb *pcb)
483 * Update pcb, saving current processor state.
487 /* JG use %rdi instead of %rcx everywhere? */
490 /* caller's return address - child won't execute this routine */
492 movq %rax,PCB_RIP(%rcx)
493 movq %rbx,PCB_RBX(%rcx)
494 movq %rsp,PCB_RSP(%rcx)
495 movq %rbp,PCB_RBP(%rcx)
496 movq %r12,PCB_R12(%rcx)
497 movq %r13,PCB_R13(%rcx)
498 movq %r14,PCB_R14(%rcx)
499 movq %r15,PCB_R15(%rcx)
503 * If npxthread == NULL, then the npx h/w state is irrelevant and the
504 * state had better already be in the pcb. This is true for forks
505 * but not for dumps (the old book-keeping with FP flags in the pcb
506 * always lost for dumps because the dump pcb has 0 flags).
508 * If npxthread != NULL, then we have to save the npx h/w state to
509 * npxthread's pcb and copy it to the requested pcb, or save to the
510 * requested pcb and reload. Copying is easier because we would
511 * have to handle h/w bugs for reloading. We used to lose the
512 * parent's npx state for forks by forgetting to reload.
514 movq PCPU(npxthread),%rax
518 pushq %rcx /* target pcb */
519 movq TD_SAVEFPU(%rax),%rax /* originating savefpu area */
528 movq $PCB_SAVEFPU_SIZE,%rdx
529 leaq PCB_SAVEFPU(%rcx),%rcx
540 * cpu_idle_restore() (current thread in %rax on entry) (one-time execution)
542 * Don't bother setting up any regs other than %rbp so backtraces
543 * don't die. This restore function is used to bootstrap into the
544 * cpu_idle() LWKT only, after that cpu_lwkt_*() will be used for
547 * Clear TDF_RUNNING in old thread only after we've cleaned up %cr3.
548 * This only occurs during system boot so no special handling is
549 * required for migration.
551 * If we are an AP we have to call ap_init() before jumping to
552 * cpu_idle(). ap_init() will synchronize with the BP and finish
553 * setting up various ncpu-dependant globaldata fields. This may
554 * happen on UP as well as SMP if we happen to be simulating multiple
557 ENTRY(cpu_idle_restore)
563 andl $~TDF_RUNNING,TD_FLAGS(%rbx)
564 orl $TDF_RUNNING,TD_FLAGS(%rax) /* manual, no switch_return */
575 * cpu_kthread_restore() (current thread is %rax on entry) (one-time execution)
577 * Don't bother setting up any regs other then %rbp so backtraces
578 * don't die. This restore function is used to bootstrap into an
579 * LWKT based kernel thread only. cpu_lwkt_switch() will be used
582 * Because this switch target does not 'return' to lwkt_switch()
583 * we have to call lwkt_switch_return(otd) to clean up otd.
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.
589 ENTRY(cpu_kthread_restore)
591 movq TD_PCB(%rax),%r13
595 * rax and rbx come from the switchout code. Call
596 * lwkt_switch_return(otd).
598 * NOTE: unlike i386, %rsi and %rdi are not call-saved regs.
602 call lwkt_switch_return
604 decl TD_CRITCOUNT(%rax)
605 movq PCB_R12(%r13),%rdi /* argument to RBX function */
606 movq PCB_RBX(%r13),%rax /* thread function */
607 /* note: top of stack return address inherited by function */
612 * cpu_lwkt_switch(struct thread *)
614 * Standard LWKT switching function. Only non-scratch registers are
615 * saved and we don't bother with the MMU state or anything else.
617 * This function is always called while in a critical section.
619 * There is a one-instruction window where curthread is the new
620 * thread but %rsp still points to the old thread's stack, but
621 * we are protected by a critical section so it is ok.
625 ENTRY(cpu_lwkt_switch)
626 pushq %rbp /* JG note: GDB hacked to locate ebp relative to td_sp */
627 /* JG we've got more registers on x86_64 */
629 movq PCPU(curthread),%rbx
638 * Save the FP state if we have used the FP. Note that calling
639 * npxsave will NULL out PCPU(npxthread).
641 * We have to deal with the FP state for LWKT threads in case they
642 * happen to get preempted or block while doing an optimized
643 * bzero/bcopy/memcpy.
645 cmpq %rbx,PCPU(npxthread)
647 movq %rdi,%r12 /* save %rdi. %r12 is callee-saved */
648 movq TD_SAVEFPU(%rbx),%rdi
649 call npxsave /* do it in a big C function */
650 movq %r12,%rdi /* restore %rdi */
654 movq %rdi,%rax /* switch to this thread */
655 pushq $cpu_lwkt_restore
656 movq %rsp,TD_SP(%rbx)
657 movq %rax,PCPU(curthread)
658 movq TD_SP(%rax),%rsp
661 * %rax contains new thread, %rbx contains old thread.
667 * cpu_lwkt_restore() (current thread in %rax on entry)
669 * Standard LWKT restore function. This function is always called
670 * while in a critical section.
672 * Warning: due to preemption the restore function can be used to
673 * 'return' to the original thread. Interrupt disablement must be
674 * protected through the switch so we cannot run splz here.
676 ENTRY(cpu_lwkt_restore)
678 * NOTE: %rbx is the previous thread and %eax is the new thread.
679 * %rbx is retained throughout so we can return it.
681 * lwkt_switch[_return] is responsible for handling TDF_RUNNING.
696 * Make AP become the idle loop.
698 ENTRY(bootstrap_idle)
699 movq PCPU(curthread),%rax
701 movq TD_SP(%rax),%rsp