1/*- 2 * Copyright (c) 2003 Peter Wemm. 3 * Copyright (c) 1990 The Regents of the University of California. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 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. 20 * 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 31 * SUCH DAMAGE. 32 * 33 * $FreeBSD$ 34 */ 35 36#include <machine/asmacros.h> 37#include <machine/specialreg.h> 38 39#include "assym.s" 40#include "opt_sched.h" 41 42/*****************************************************************************/ 43/* Scheduling */ 44/*****************************************************************************/ 45 46 .text 47 48#ifdef SMP 49#define LK lock ; 50#else 51#define LK 52#endif 53 54#if defined(SCHED_ULE) && defined(SMP) 55#define SETLK xchgq 56#else 57#define SETLK movq 58#endif 59 60/* 61 * cpu_throw() 62 * 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. 68 * %rdi = oldtd 69 * %rsi = newtd 70 */ 71ENTRY(cpu_throw) 72 movl PCPU(CPUID),%eax 73 testq %rdi,%rdi 74 jz 1f 75 /* release bit from old pm_active */ 76 movq PCPU(CURPMAP),%rdx 77 LK btrl %eax,PM_ACTIVE(%rdx) /* clear old */ 781: 79 movq TD_PCB(%rsi),%r8 /* newtd->td_pcb */ 80 movq PCB_CR3(%r8),%rcx /* new address space */ 81 jmp swact 82END(cpu_throw) 83 84/* 85 * cpu_switch(old, new, mtx) 86 * 87 * Save the current thread state, then select the next thread to run 88 * and load its state. 89 * %rdi = oldtd 90 * %rsi = newtd 91 * %rdx = mtx 92 */ 93ENTRY(cpu_switch) 94 /* Switch to new thread. First, save context. */ 95 movq TD_PCB(%rdi),%r8 96 orl $PCB_FULL_IRET,PCB_FLAGS(%r8) 97 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) 107 108 testl $PCB_DBREGS,PCB_FLAGS(%r8) 109 jnz store_dr /* static predict not taken */ 110done_store_dr: 111 112 /* have we used fp, and need a save? */ 113 cmpq %rdi,PCPU(FPCURTHREAD) 114 jne 3f 115 movq PCB_SAVEFPU(%r8),%r8 116 clts 117 cmpl $0,use_xsave 118 jne 1f 119 fxsave (%r8) 120 jmp 2f 1211: movq %rdx,%rcx 122 movl xsave_mask,%eax 123 movl xsave_mask+4,%edx 124 .globl ctx_switch_xsave 125ctx_switch_xsave: 126 /* This is patched to xsaveopt if supported, see fpuinit_bsp1() */ 127 xsave (%r8) 128 movq %rcx,%rdx 1292: smsw %ax 130 orb $CR0_TS,%al 131 lmsw %ax 132 xorl %eax,%eax 133 movq %rax,PCPU(FPCURTHREAD) 1343: 135 136 /* Save is done. Now fire up new thread. Leave old vmspace. */ 137 movq TD_PCB(%rsi),%r8 138 139 /* switch address space */ 140 movq PCB_CR3(%r8),%rcx 141 movq %cr3,%rax 142 cmpq %rcx,%rax /* Same address space? */ 143 jne swinact 144 SETLK %rdx, TD_LOCK(%rdi) /* Release the old thread */ 145 jmp sw1 146swinact: 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 */ 152swact: 153 /* Set bit in new pmap->pm_active */ 154 movq TD_PROC(%rsi),%rdx /* newproc */ 155 movq P_VMSPACE(%rdx), %rdx 156 addq $VM_PMAP,%rdx 157 cmpl $-1,PM_PCID(%rdx) 158 je 1f 159 LK btsl %eax,PM_SAVE(%rdx) 160 jnc 1f 161 btsq $63,%rcx /* CR3_PCID_SAVE */ 162 incq PCPU(PM_SAVE_CNT) 1631: 164 movq %rcx,%cr3 /* new address space */ 165 LK btsl %eax,PM_ACTIVE(%rdx) /* set new */ 166 movq %rdx,PCPU(CURPMAP) 167 168 /* 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. 173 */ 174 btrq $63,%rcx 175 jnc 1f 176 LK btsl %eax,PM_SAVE(%rdx) 177 jc 1f 178 decq PCPU(PM_SAVE_CNT) 179 movq %rcx,%cr3 1801: 181 182sw1: 183#if defined(SCHED_ULE) && defined(SMP) 184 /* Wait for the new thread to become unblocked */ 185 movq $blocked_lock, %rdx 1861: 187 movq TD_LOCK(%rsi),%rcx 188 cmpq %rcx, %rdx 189 pause 190 je 1b 191#endif 192 /* 193 * At this point, we've switched address spaces and are ready 194 * to load up the rest of the next context. 195 */ 196 197 /* Skip loading user fsbase/gsbase for kthreads */ 198 testl $TDP_KTHREAD,TD_PFLAGS(%rsi) 199 jnz do_kthread 200 201 /* 202 * Load ldt register 203 */ 204 movq TD_PROC(%rsi),%rcx 205 cmpq $0, P_MD+MD_LDT(%rcx) 206 jne do_ldt 207 xorl %eax,%eax 208ld_ldt: lldt %ax 209 210 /* Restore fs base in GDT */ 211 movl PCB_FSBASE(%r8),%eax 212 movq PCPU(FS32P),%rdx 213 movw %ax,2(%rdx) 214 shrl $16,%eax 215 movb %al,4(%rdx) 216 shrl $8,%eax 217 movb %al,7(%rdx) 218 219 /* Restore gs base in GDT */ 220 movl PCB_GSBASE(%r8),%eax 221 movq PCPU(GS32P),%rdx 222 movw %ax,2(%rdx) 223 shrl $16,%eax 224 movb %al,4(%rdx) 225 shrl $8,%eax 226 movb %al,7(%rdx) 227 228do_kthread: 229 /* Do we need to reload tss ? */ 230 movq PCPU(TSSP),%rax 231 movq PCB_TSSP(%r8),%rdx 232 testq %rdx,%rdx 233 cmovzq PCPU(COMMONTSSP),%rdx 234 cmpq %rax,%rdx 235 jne do_tss 236done_tss: 237 movq %r8,PCPU(RSP0) 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 */ 242 243 /* Test if debug registers should be restored. */ 244 testl $PCB_DBREGS,PCB_FLAGS(%r8) 245 jnz load_dr /* static predict not taken */ 246done_load_dr: 247 248 /* 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 257 movq %rax,(%rsp) 258 ret 259 260 /* 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. 268 */ 269 270store_dr: 271 movq %dr7,%rax /* yes, do the save */ 272 movq %dr0,%r15 273 movq %dr1,%r14 274 movq %dr2,%r13 275 movq %dr3,%r12 276 movq %dr6,%r11 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 */ 284 movq %rax,%dr7 285 jmp done_store_dr 286 287load_dr: 288 movq %dr7,%rax 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 295 movq %r15,%dr0 296 movq %r14,%dr1 297 /* Preserve reserved bits in %dr7 */ 298 andq $0x0000fc00,%rax 299 andq $~0x0000fc00,%rcx 300 movq %r13,%dr2 301 movq %r12,%dr3 302 orq %rcx,%rax 303 movq %r11,%dr6 304 movq %rax,%dr7 305 jmp done_load_dr 306 307do_tss: movq %rdx,PCPU(TSSP) 308 movq %rdx,%rcx 309 movq PCPU(TSS),%rax 310 movw %cx,2(%rax) 311 shrq $16,%rcx 312 movb %cl,4(%rax) 313 shrq $8,%rcx 314 movb %cl,7(%rax) 315 shrq $8,%rcx 316 movl %ecx,8(%rax) 317 movb $0x89,5(%rax) /* unset busy */ 318 movl $TSSSEL,%eax 319 ltr %ax 320 jmp done_tss 321 322do_ldt: movq PCPU(LDT),%rax 323 movq P_MD+MD_LDT_SD(%rcx),%rdx 324 movq %rdx,(%rax) 325 movq P_MD+MD_LDT_SD+8(%rcx),%rdx 326 movq %rdx,8(%rax) 327 movl $LDTSEL,%eax 328 jmp ld_ldt 329END(cpu_switch) 330 331/* 332 * savectx(pcb) 333 * Update pcb, saving current processor state. 334 */ 335ENTRY(savectx) 336 /* Save caller's return address. */ 337 movq (%rsp),%rax 338 movq %rax,PCB_RIP(%rdi) 339 340 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) 347 348 movq %cr0,%rsi 349 movq %rsi,PCB_CR0(%rdi) 350 movq %cr2,%rax 351 movq %rax,PCB_CR2(%rdi) 352 movq %cr3,%rax 353 movq %rax,PCB_CR3(%rdi) 354 movq %cr4,%rax 355 movq %rax,PCB_CR4(%rdi) 356 357 movq %dr0,%rax 358 movq %rax,PCB_DR0(%rdi) 359 movq %dr1,%rax 360 movq %rax,PCB_DR1(%rdi) 361 movq %dr2,%rax 362 movq %rax,PCB_DR2(%rdi) 363 movq %dr3,%rax 364 movq %rax,PCB_DR3(%rdi) 365 movq %dr6,%rax 366 movq %rax,PCB_DR6(%rdi) 367 movq %dr7,%rax 368 movq %rax,PCB_DR7(%rdi) 369 370 movl $MSR_FSBASE,%ecx 371 rdmsr 372 movl %eax,PCB_FSBASE(%rdi) 373 movl %edx,PCB_FSBASE+4(%rdi) 374 movl $MSR_GSBASE,%ecx 375 rdmsr 376 movl %eax,PCB_GSBASE(%rdi) 377 movl %edx,PCB_GSBASE+4(%rdi) 378 movl $MSR_KGSBASE,%ecx 379 rdmsr 380 movl %eax,PCB_KGSBASE(%rdi) 381 movl %edx,PCB_KGSBASE+4(%rdi) 382 movl $MSR_EFER,%ecx 383 rdmsr 384 movl %eax,PCB_EFER(%rdi) 385 movl %edx,PCB_EFER+4(%rdi) 386 movl $MSR_STAR,%ecx 387 rdmsr 388 movl %eax,PCB_STAR(%rdi) 389 movl %edx,PCB_STAR+4(%rdi) 390 movl $MSR_LSTAR,%ecx 391 rdmsr 392 movl %eax,PCB_LSTAR(%rdi) 393 movl %edx,PCB_LSTAR+4(%rdi) 394 movl $MSR_CSTAR,%ecx 395 rdmsr 396 movl %eax,PCB_CSTAR(%rdi) 397 movl %edx,PCB_CSTAR+4(%rdi) 398 movl $MSR_SF_MASK,%ecx 399 rdmsr 400 movl %eax,PCB_SFMASK(%rdi) 401 movl %edx,PCB_SFMASK+4(%rdi) 402 movl xsave_mask,%eax 403 movl %eax,PCB_XSMASK(%rdi) 404 movl xsave_mask+4,%eax 405 movl %eax,PCB_XSMASK+4(%rdi) 406 407 sgdt PCB_GDT(%rdi) 408 sidt PCB_IDT(%rdi) 409 sldt PCB_LDT(%rdi) 410 str PCB_TR(%rdi) 411 4122: movq %rsi,%cr0 /* The previous %cr0 is saved in %rsi. */ 413 414 movl $1,%eax 415 ret 416END(savectx) 417 418/* 419 * resumectx(pcb) 420 * Resuming processor state from pcb. 421 */ 422ENTRY(resumectx) 423 /* Switch to KPML4phys. */ 424 movq KPML4phys,%rax 425 movq %rax,%cr3 426 427 /* Force kernel segment registers. */ 428 movl $KDSEL,%eax 429 movw %ax,%ds 430 movw %ax,%es 431 movw %ax,%ss 432 movl $KUF32SEL,%eax 433 movw %ax,%fs 434 movl $KUG32SEL,%eax 435 movw %ax,%gs 436 437 movl $MSR_FSBASE,%ecx 438 movl PCB_FSBASE(%rdi),%eax 439 movl 4 + PCB_FSBASE(%rdi),%edx 440 wrmsr 441 movl $MSR_GSBASE,%ecx 442 movl PCB_GSBASE(%rdi),%eax 443 movl 4 + PCB_GSBASE(%rdi),%edx 444 wrmsr 445 movl $MSR_KGSBASE,%ecx 446 movl PCB_KGSBASE(%rdi),%eax 447 movl 4 + PCB_KGSBASE(%rdi),%edx 448 wrmsr 449 450 /* Restore EFER. */ 451 movl $MSR_EFER,%ecx 452 movl PCB_EFER(%rdi),%eax 453 wrmsr 454 455 /* Restore fast syscall stuff. */ 456 movl $MSR_STAR,%ecx 457 movl PCB_STAR(%rdi),%eax 458 movl 4 + PCB_STAR(%rdi),%edx 459 wrmsr 460 movl $MSR_LSTAR,%ecx 461 movl PCB_LSTAR(%rdi),%eax 462 movl 4 + PCB_LSTAR(%rdi),%edx 463 wrmsr 464 movl $MSR_CSTAR,%ecx 465 movl PCB_CSTAR(%rdi),%eax 466 movl 4 + PCB_CSTAR(%rdi),%edx 467 wrmsr 468 movl $MSR_SF_MASK,%ecx 469 movl PCB_SFMASK(%rdi),%eax 470 wrmsr 471 472 /* Restore CR0 except for FPU mode. */ 473 movq PCB_CR0(%rdi),%rax 474 andq $~(CR0_EM | CR0_TS),%rax 475 movq %rax,%cr0 476 477 /* Restore CR2, CR4 and CR3. */ 478 movq PCB_CR2(%rdi),%rax 479 movq %rax,%cr2 480 movq PCB_CR4(%rdi),%rax 481 movq %rax,%cr4 482 movq PCB_CR3(%rdi),%rax 483 movq %rax,%cr3 484 485 /* Restore descriptor tables. */ 486 lidt PCB_IDT(%rdi) 487 lldt PCB_LDT(%rdi) 488 489#define SDT_SYSTSS 9 490#define SDT_SYSBSY 11 491 492 /* Clear "task busy" bit and reload TR. */ 493 movq PCPU(TSS),%rax 494 andb $(~SDT_SYSBSY | SDT_SYSTSS),5(%rax) 495 movw PCB_TR(%rdi),%ax 496 ltr %ax 497 498#undef SDT_SYSTSS 499#undef SDT_SYSBSY 500 501 /* Restore debug registers. */ 502 movq PCB_DR0(%rdi),%rax 503 movq %rax,%dr0 504 movq PCB_DR1(%rdi),%rax 505 movq %rax,%dr1 506 movq PCB_DR2(%rdi),%rax 507 movq %rax,%dr2 508 movq PCB_DR3(%rdi),%rax 509 movq %rax,%dr3 510 movq PCB_DR6(%rdi),%rax 511 movq %rax,%dr6 512 movq PCB_DR7(%rdi),%rax 513 movq %rax,%dr7 514 515 /* Restore FPU state. */ 516 fninit 517 movq PCB_FPUSUSPEND(%rdi),%rbx 518 movq PCB_XSMASK(%rdi),%rax 519 testq %rax,%rax 520 jz 1f 521 movq %rax,%rdx 522 shrq $32,%rdx 523 movl $XCR0,%ecx 524 xsetbv 525 xrstor (%rbx) 526 jmp 2f 5271: 528 fxrstor (%rbx) 5292: 530 531 /* Reload CR0. */ 532 movq PCB_CR0(%rdi),%rax 533 movq %rax,%cr0 534 535 /* Restore other callee saved registers. */ 536 movq PCB_R15(%rdi),%r15 537 movq PCB_R14(%rdi),%r14 538 movq PCB_R13(%rdi),%r13 539 movq PCB_R12(%rdi),%r12 540 movq PCB_RBP(%rdi),%rbp 541 movq PCB_RSP(%rdi),%rsp 542 movq PCB_RBX(%rdi),%rbx 543 544 /* Restore return address. */ 545 movq PCB_RIP(%rdi),%rax 546 movq %rax,(%rsp) 547 548 xorl %eax,%eax 549 ret 550END(resumectx) 551 552/* 553 * Wrapper around fpusave to care about TS0_CR. 554 */ 555ENTRY(ctx_fpusave) 556 movq %cr0,%rsi 557 clts 558 call fpusave 559 movq %rsi,%cr0 560 ret 561END(ctx_fpusave) 562