2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 The Regents of the University of California.
6 * This code is derived from software contributed to Berkeley by
9 * Redistribution and use in source and binary forms, with or without
10 * 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 the
16 * documentation and/or other materials provided with the distribution.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #include <machine/asmacros.h>
37 #include <machine/specialreg.h>
40 #include "opt_sched.h"
42 /*****************************************************************************/
44 /*****************************************************************************/
54 #if defined(SCHED_ULE) && defined(SMP)
63 * This is the second half of cpu_switch(). It is used when the current
64 * thread is either a dummy or slated to die, and we no longer care
65 * about its state. This is only a slight optimization and is probably
66 * not worth it anymore. Note that we need to clear the pm_active bits so
67 * we do need the old proc if it still exists.
75 /* release bit from old pm_active */
76 movq PCPU(CURPMAP),%rdx
77 LK btrl %eax,PM_ACTIVE(%rdx) /* clear old */
79 movq TD_PCB(%rsi),%r8 /* newtd->td_pcb */
80 movq PCB_CR3(%r8),%rcx /* new address space */
85 * cpu_switch(old, new, mtx)
87 * Save the current thread state, then select the next thread to run
94 /* Switch to new thread. First, save context. */
96 orl $PCB_FULL_IRET,PCB_FLAGS(%r8)
98 movq (%rsp),%rax /* Hardware registers */
99 movq %r15,PCB_R15(%r8)
100 movq %r14,PCB_R14(%r8)
101 movq %r13,PCB_R13(%r8)
102 movq %r12,PCB_R12(%r8)
103 movq %rbp,PCB_RBP(%r8)
104 movq %rsp,PCB_RSP(%r8)
105 movq %rbx,PCB_RBX(%r8)
106 movq %rax,PCB_RIP(%r8)
108 testl $PCB_DBREGS,PCB_FLAGS(%r8)
109 jnz store_dr /* static predict not taken */
112 /* have we used fp, and need a save? */
113 cmpq %rdi,PCPU(FPCURTHREAD)
115 movq PCB_SAVEFPU(%r8),%r8
123 movl xsave_mask+4,%edx
124 .globl ctx_switch_xsave
126 /* This is patched to xsaveopt if supported, see fpuinit_bsp1() */
133 movq %rax,PCPU(FPCURTHREAD)
136 /* Save is done. Now fire up new thread. Leave old vmspace. */
137 movq TD_PCB(%rsi),%r8
139 /* switch address space */
140 movq PCB_CR3(%r8),%rcx
142 cmpq %rcx,%rax /* Same address space? */
144 SETLK %rdx, TD_LOCK(%rdi) /* Release the old thread */
147 movl PCPU(CPUID),%eax
148 /* Release bit from old pmap->pm_active */
149 movq PCPU(CURPMAP),%r12
150 LK btrl %eax,PM_ACTIVE(%r12) /* clear old */
151 SETLK %rdx,TD_LOCK(%rdi) /* Release the old thread */
153 /* Set bit in new pmap->pm_active */
154 movq TD_PROC(%rsi),%rdx /* newproc */
155 movq P_VMSPACE(%rdx), %rdx
157 cmpl $-1,PM_PCID(%rdx)
159 LK btsl %eax,PM_SAVE(%rdx)
161 btsq $63,%rcx /* CR3_PCID_SAVE */
162 incq PCPU(PM_SAVE_CNT)
164 movq %rcx,%cr3 /* new address space */
165 LK btsl %eax,PM_ACTIVE(%rdx) /* set new */
166 movq %rdx,PCPU(CURPMAP)
169 * We might lose the race and other CPU might have changed
170 * the pmap after we set our bit in pmap->pm_save. Recheck.
171 * Reload %cr3 with CR3_PCID_SAVE bit cleared if pmap was
172 * modified, causing TLB flush for this pcid.
176 LK btsl %eax,PM_SAVE(%rdx)
178 decq PCPU(PM_SAVE_CNT)
183 #if defined(SCHED_ULE) && defined(SMP)
184 /* Wait for the new thread to become unblocked */
185 movq $blocked_lock, %rdx
187 movq TD_LOCK(%rsi),%rcx
193 * At this point, we've switched address spaces and are ready
194 * to load up the rest of the next context.
197 /* Skip loading user fsbase/gsbase for kthreads */
198 testl $TDP_KTHREAD,TD_PFLAGS(%rsi)
204 movq TD_PROC(%rsi),%rcx
205 cmpq $0, P_MD+MD_LDT(%rcx)
210 /* Restore fs base in GDT */
211 movl PCB_FSBASE(%r8),%eax
212 movq PCPU(FS32P),%rdx
219 /* Restore gs base in GDT */
220 movl PCB_GSBASE(%r8),%eax
221 movq PCPU(GS32P),%rdx
229 /* Do we need to reload tss ? */
231 movq PCB_TSSP(%r8),%rdx
233 cmovzq PCPU(COMMONTSSP),%rdx
238 movq %r8,PCPU(CURPCB)
239 /* Update the TSS_RSP0 pointer for the next interrupt */
240 movq %r8,COMMON_TSS_RSP0(%rdx)
241 movq %rsi,PCPU(CURTHREAD) /* into next thread */
243 /* Test if debug registers should be restored. */
244 testl $PCB_DBREGS,PCB_FLAGS(%r8)
245 jnz load_dr /* static predict not taken */
248 /* Restore context. */
249 movq PCB_R15(%r8),%r15
250 movq PCB_R14(%r8),%r14
251 movq PCB_R13(%r8),%r13
252 movq PCB_R12(%r8),%r12
253 movq PCB_RBP(%r8),%rbp
254 movq PCB_RSP(%r8),%rsp
255 movq PCB_RBX(%r8),%rbx
256 movq PCB_RIP(%r8),%rax
261 * We order these strangely for several reasons.
262 * 1: I wanted to use static branch prediction hints
263 * 2: Most athlon64/opteron cpus don't have them. They define
264 * a forward branch as 'predict not taken'. Intel cores have
265 * the 'rep' prefix to invert this.
266 * So, to make it work on both forms of cpu we do the detour.
267 * We use jumps rather than call in order to avoid the stack.
271 movq %dr7,%rax /* yes, do the save */
277 movq %r15,PCB_DR0(%r8)
278 movq %r14,PCB_DR1(%r8)
279 movq %r13,PCB_DR2(%r8)
280 movq %r12,PCB_DR3(%r8)
281 movq %r11,PCB_DR6(%r8)
282 movq %rax,PCB_DR7(%r8)
283 andq $0x0000fc00, %rax /* disable all watchpoints */
289 movq PCB_DR0(%r8),%r15
290 movq PCB_DR1(%r8),%r14
291 movq PCB_DR2(%r8),%r13
292 movq PCB_DR3(%r8),%r12
293 movq PCB_DR6(%r8),%r11
294 movq PCB_DR7(%r8),%rcx
297 /* Preserve reserved bits in %dr7 */
298 andq $0x0000fc00,%rax
299 andq $~0x0000fc00,%rcx
307 do_tss: movq %rdx,PCPU(TSSP)
317 movb $0x89,5(%rax) /* unset busy */
322 do_ldt: movq PCPU(LDT),%rax
323 movq P_MD+MD_LDT_SD(%rcx),%rdx
325 movq P_MD+MD_LDT_SD+8(%rcx),%rdx
333 * Update pcb, saving current processor state.
336 /* Save caller's return address. */
338 movq %rax,PCB_RIP(%rdi)
340 movq %rbx,PCB_RBX(%rdi)
341 movq %rsp,PCB_RSP(%rdi)
342 movq %rbp,PCB_RBP(%rdi)
343 movq %r12,PCB_R12(%rdi)
344 movq %r13,PCB_R13(%rdi)
345 movq %r14,PCB_R14(%rdi)
346 movq %r15,PCB_R15(%rdi)
349 movq %rax,PCB_CR0(%rdi)
351 movq %rax,PCB_CR2(%rdi)
353 movq %rax,PCB_CR3(%rdi)
355 movq %rax,PCB_CR4(%rdi)
358 movq %rax,PCB_DR0(%rdi)
360 movq %rax,PCB_DR1(%rdi)
362 movq %rax,PCB_DR2(%rdi)
364 movq %rax,PCB_DR3(%rdi)
366 movq %rax,PCB_DR6(%rdi)
368 movq %rax,PCB_DR7(%rdi)
370 movl $MSR_FSBASE,%ecx
372 movl %eax,PCB_FSBASE(%rdi)
373 movl %edx,PCB_FSBASE+4(%rdi)
374 movl $MSR_GSBASE,%ecx
376 movl %eax,PCB_GSBASE(%rdi)
377 movl %edx,PCB_GSBASE+4(%rdi)
378 movl $MSR_KGSBASE,%ecx
380 movl %eax,PCB_KGSBASE(%rdi)
381 movl %edx,PCB_KGSBASE+4(%rdi)
384 movl %eax,PCB_EFER(%rdi)
385 movl %edx,PCB_EFER+4(%rdi)
388 movl %eax,PCB_STAR(%rdi)
389 movl %edx,PCB_STAR+4(%rdi)
392 movl %eax,PCB_LSTAR(%rdi)
393 movl %edx,PCB_LSTAR+4(%rdi)
396 movl %eax,PCB_CSTAR(%rdi)
397 movl %edx,PCB_CSTAR+4(%rdi)
398 movl $MSR_SF_MASK,%ecx
400 movl %eax,PCB_SFMASK(%rdi)
401 movl %edx,PCB_SFMASK+4(%rdi)
414 * Resuming processor state from pcb.
417 /* Switch to KPML4phys. */
421 /* Force kernel segment registers. */
431 movl $MSR_FSBASE,%ecx
432 movl PCB_FSBASE(%rdi),%eax
433 movl 4 + PCB_FSBASE(%rdi),%edx
435 movl $MSR_GSBASE,%ecx
436 movl PCB_GSBASE(%rdi),%eax
437 movl 4 + PCB_GSBASE(%rdi),%edx
439 movl $MSR_KGSBASE,%ecx
440 movl PCB_KGSBASE(%rdi),%eax
441 movl 4 + PCB_KGSBASE(%rdi),%edx
446 movl PCB_EFER(%rdi),%eax
449 /* Restore fast syscall stuff. */
451 movl PCB_STAR(%rdi),%eax
452 movl 4 + PCB_STAR(%rdi),%edx
455 movl PCB_LSTAR(%rdi),%eax
456 movl 4 + PCB_LSTAR(%rdi),%edx
459 movl PCB_CSTAR(%rdi),%eax
460 movl 4 + PCB_CSTAR(%rdi),%edx
462 movl $MSR_SF_MASK,%ecx
463 movl PCB_SFMASK(%rdi),%eax
466 /* Restore CR0, CR2, CR4 and CR3. */
467 movq PCB_CR0(%rdi),%rax
469 movq PCB_CR2(%rdi),%rax
471 movq PCB_CR4(%rdi),%rax
473 movq PCB_CR3(%rdi),%rax
476 /* Restore descriptor tables. */
481 #define SDT_SYSBSY 11
483 /* Clear "task busy" bit and reload TR. */
485 andb $(~SDT_SYSBSY | SDT_SYSTSS),5(%rax)
486 movw PCB_TR(%rdi),%ax
492 /* Restore debug registers. */
493 movq PCB_DR0(%rdi),%rax
495 movq PCB_DR1(%rdi),%rax
497 movq PCB_DR2(%rdi),%rax
499 movq PCB_DR3(%rdi),%rax
501 movq PCB_DR6(%rdi),%rax
503 movq PCB_DR7(%rdi),%rax
506 /* Restore other callee saved registers. */
507 movq PCB_R15(%rdi),%r15
508 movq PCB_R14(%rdi),%r14
509 movq PCB_R13(%rdi),%r13
510 movq PCB_R12(%rdi),%r12
511 movq PCB_RBP(%rdi),%rbp
512 movq PCB_RSP(%rdi),%rsp
513 movq PCB_RBX(%rdi),%rbx
515 /* Restore return address. */
516 movq PCB_RIP(%rdi),%rax