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 * 3. 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.
79 * cpu_switch(old, new, mtx)
81 * Save the current thread state, then select the next thread to run
88 /* Switch to new thread. First, save context. */
91 movq (%rsp),%rax /* Hardware registers */
92 movq %r15,PCB_R15(%r8)
93 movq %r14,PCB_R14(%r8)
94 movq %r13,PCB_R13(%r8)
95 movq %r12,PCB_R12(%r8)
96 movq %rbp,PCB_RBP(%r8)
97 movq %rsp,PCB_RSP(%r8)
98 movq %rbx,PCB_RBX(%r8)
99 movq %rax,PCB_RIP(%r8)
101 testl $PCB_FULL_IRET,PCB_FLAGS(%r8)
103 orl $PCB_FULL_IRET,PCB_FLAGS(%r8)
104 testl $TDP_KTHREAD,TD_PFLAGS(%rdi)
106 testb $CPUID_STDEXT_FSGSBASE,cpu_stdext_feature(%rip)
112 movq %rax,PCB_FSBASE(%r8)
117 movl $MSR_KGSBASE,%ecx /* Read user gs base */
121 movq %rax,PCB_GSBASE(%r8)
125 testl $PCB_DBREGS,PCB_FLAGS(%r8)
126 jnz store_dr /* static predict not taken */
129 /* have we used fp, and need a save? */
130 cmpq %rdi,PCPU(FPCURTHREAD)
132 movq PCB_SAVEFPU(%r8),%r8
140 movl xsave_mask+4,%edx
141 .globl ctx_switch_xsave
143 /* This is patched to xsaveopt if supported, see fpuinit_bsp1() */
150 movq %rax,PCPU(FPCURTHREAD)
152 /* Save is done. Now fire up new thread. Leave old vmspace. */
157 callq pmap_activate_sw
158 SETLK %r15,TD_LOCK(%r13) /* Release the old thread */
160 movq TD_PCB(%r12),%r8
161 #if defined(SCHED_ULE) && defined(SMP)
162 /* Wait for the new thread to become unblocked */
163 movq $blocked_lock, %rdx
165 movq TD_LOCK(%r12),%rcx
171 * At this point, we've switched address spaces and are ready
172 * to load up the rest of the next context.
175 /* Skip loading LDT and user fsbase/gsbase for kthreads */
176 testl $TDP_KTHREAD,TD_PFLAGS(%r12)
182 movq TD_PROC(%r12),%rcx
183 cmpq $0, P_MD+MD_LDT(%rcx)
188 /* Restore fs base in GDT */
189 movl PCB_FSBASE(%r8),%eax
190 movq PCPU(FS32P),%rdx
197 /* Restore gs base in GDT */
198 movl PCB_GSBASE(%r8),%eax
199 movq PCPU(GS32P),%rdx
207 /* Do we need to reload tss ? */
209 movq PCB_TSSP(%r8),%rdx
211 cmovzq PCPU(COMMONTSSP),%rdx
216 movq %r8,PCPU(CURPCB)
217 /* Update the TSS_RSP0 pointer for the next interrupt */
220 movq PCPU(PTI_RSP0),%rax
221 movq %rax,TSS_RSP0(%rdx)
223 1: movq %r8,TSS_RSP0(%rdx)
224 2: movq %r12,PCPU(CURTHREAD) /* into next thread */
226 /* Test if debug registers should be restored. */
227 testl $PCB_DBREGS,PCB_FLAGS(%r8)
228 jnz load_dr /* static predict not taken */
231 /* Restore context. */
232 movq PCB_R15(%r8),%r15
233 movq PCB_R14(%r8),%r14
234 movq PCB_R13(%r8),%r13
235 movq PCB_R12(%r8),%r12
236 movq PCB_RBP(%r8),%rbp
237 movq PCB_RSP(%r8),%rsp
238 movq PCB_RBX(%r8),%rbx
239 movq PCB_RIP(%r8),%rax
244 * We order these strangely for several reasons.
245 * 1: I wanted to use static branch prediction hints
246 * 2: Most athlon64/opteron cpus don't have them. They define
247 * a forward branch as 'predict not taken'. Intel cores have
248 * the 'rep' prefix to invert this.
249 * So, to make it work on both forms of cpu we do the detour.
250 * We use jumps rather than call in order to avoid the stack.
254 movq %dr7,%rax /* yes, do the save */
260 movq %r15,PCB_DR0(%r8)
261 movq %r14,PCB_DR1(%r8)
262 movq %r13,PCB_DR2(%r8)
263 movq %r12,PCB_DR3(%r8)
264 movq %r11,PCB_DR6(%r8)
265 movq %rax,PCB_DR7(%r8)
266 andq $0x0000fc00, %rax /* disable all watchpoints */
272 movq PCB_DR0(%r8),%r15
273 movq PCB_DR1(%r8),%r14
274 movq PCB_DR2(%r8),%r13
275 movq PCB_DR3(%r8),%r12
276 movq PCB_DR6(%r8),%r11
277 movq PCB_DR7(%r8),%rcx
280 /* Preserve reserved bits in %dr7 */
281 andq $0x0000fc00,%rax
282 andq $~0x0000fc00,%rcx
290 do_tss: movq %rdx,PCPU(TSSP)
300 movb $0x89,5(%rax) /* unset busy */
305 do_ldt: movq PCPU(LDT),%rax
306 movq P_MD+MD_LDT_SD(%rcx),%rdx
308 movq P_MD+MD_LDT_SD+8(%rcx),%rdx
316 * Update pcb, saving current processor state.
319 /* Save caller's return address. */
321 movq %rax,PCB_RIP(%rdi)
323 movq %rbx,PCB_RBX(%rdi)
324 movq %rsp,PCB_RSP(%rdi)
325 movq %rbp,PCB_RBP(%rdi)
326 movq %r12,PCB_R12(%rdi)
327 movq %r13,PCB_R13(%rdi)
328 movq %r14,PCB_R14(%rdi)
329 movq %r15,PCB_R15(%rdi)
332 movq %rax,PCB_CR0(%rdi)
334 movq %rax,PCB_CR2(%rdi)
336 movq %rax,PCB_CR3(%rdi)
338 movq %rax,PCB_CR4(%rdi)
341 movq %rax,PCB_DR0(%rdi)
343 movq %rax,PCB_DR1(%rdi)
345 movq %rax,PCB_DR2(%rdi)
347 movq %rax,PCB_DR3(%rdi)
349 movq %rax,PCB_DR6(%rdi)
351 movq %rax,PCB_DR7(%rdi)
353 movl $MSR_FSBASE,%ecx
355 movl %eax,PCB_FSBASE(%rdi)
356 movl %edx,PCB_FSBASE+4(%rdi)
357 movl $MSR_GSBASE,%ecx
359 movl %eax,PCB_GSBASE(%rdi)
360 movl %edx,PCB_GSBASE+4(%rdi)
361 movl $MSR_KGSBASE,%ecx
363 movl %eax,PCB_KGSBASE(%rdi)
364 movl %edx,PCB_KGSBASE+4(%rdi)
367 movl %eax,PCB_EFER(%rdi)
368 movl %edx,PCB_EFER+4(%rdi)
371 movl %eax,PCB_STAR(%rdi)
372 movl %edx,PCB_STAR+4(%rdi)
375 movl %eax,PCB_LSTAR(%rdi)
376 movl %edx,PCB_LSTAR+4(%rdi)
379 movl %eax,PCB_CSTAR(%rdi)
380 movl %edx,PCB_CSTAR+4(%rdi)
381 movl $MSR_SF_MASK,%ecx
383 movl %eax,PCB_SFMASK(%rdi)
384 movl %edx,PCB_SFMASK+4(%rdi)
397 * Resuming processor state from pcb.
400 /* Switch to KPML4phys. */
404 /* Force kernel segment registers. */
414 movl $MSR_FSBASE,%ecx
415 movl PCB_FSBASE(%rdi),%eax
416 movl 4 + PCB_FSBASE(%rdi),%edx
418 movl $MSR_GSBASE,%ecx
419 movl PCB_GSBASE(%rdi),%eax
420 movl 4 + PCB_GSBASE(%rdi),%edx
422 movl $MSR_KGSBASE,%ecx
423 movl PCB_KGSBASE(%rdi),%eax
424 movl 4 + PCB_KGSBASE(%rdi),%edx
427 /* Restore EFER one more time. */
429 movl PCB_EFER(%rdi),%eax
432 /* Restore fast syscall stuff. */
434 movl PCB_STAR(%rdi),%eax
435 movl 4 + PCB_STAR(%rdi),%edx
438 movl PCB_LSTAR(%rdi),%eax
439 movl 4 + PCB_LSTAR(%rdi),%edx
442 movl PCB_CSTAR(%rdi),%eax
443 movl 4 + PCB_CSTAR(%rdi),%edx
445 movl $MSR_SF_MASK,%ecx
446 movl PCB_SFMASK(%rdi),%eax
449 /* Restore CR0, CR2, CR4 and CR3. */
450 movq PCB_CR0(%rdi),%rax
452 movq PCB_CR2(%rdi),%rax
454 movq PCB_CR4(%rdi),%rax
456 movq PCB_CR3(%rdi),%rax
459 /* Restore descriptor tables. */
464 #define SDT_SYSBSY 11
466 /* Clear "task busy" bit and reload TR. */
468 andb $(~SDT_SYSBSY | SDT_SYSTSS),5(%rax)
469 movw PCB_TR(%rdi),%ax
475 /* Restore debug registers. */
476 movq PCB_DR0(%rdi),%rax
478 movq PCB_DR1(%rdi),%rax
480 movq PCB_DR2(%rdi),%rax
482 movq PCB_DR3(%rdi),%rax
484 movq PCB_DR6(%rdi),%rax
486 movq PCB_DR7(%rdi),%rax
489 /* Restore other callee saved registers. */
490 movq PCB_R15(%rdi),%r15
491 movq PCB_R14(%rdi),%r14
492 movq PCB_R13(%rdi),%r13
493 movq PCB_R12(%rdi),%r12
494 movq PCB_RBP(%rdi),%rbp
495 movq PCB_RSP(%rdi),%rsp
496 movq PCB_RBX(%rdi),%rbx
498 /* Restore return address. */
499 movq PCB_RIP(%rdi),%rax