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 * 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. 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.inc" 40#include "opt_sched.h" 41 42/*****************************************************************************/ 43/* Scheduling */ 44/*****************************************************************************/ 45 46 .text 47 48/* 49 * cpu_throw() 50 * 51 * This is the second half of cpu_switch(). It is used when the current 52 * thread is either a dummy or slated to die, and we no longer care 53 * about its state. This is only a slight optimization and is probably 54 * not worth it anymore. Note that we need to clear the pm_active bits so 55 * we do need the old proc if it still exists. 56 * %rdi = oldtd 57 * %rsi = newtd 58 */ 59ENTRY(cpu_throw) 60 movq %rsi,%r12 61 movq %rsi,%rdi 62 call pmap_activate_sw 63 jmp sw1 64END(cpu_throw) 65 66/* 67 * cpu_switch(old, new, mtx) 68 * 69 * Save the current thread state, then select the next thread to run 70 * and load its state. 71 * %rdi = oldtd 72 * %rsi = newtd 73 * %rdx = mtx 74 */ 75ENTRY(cpu_switch) 76 /* Switch to new thread. First, save context. */ 77 leaq TD_MD_PCB(%rdi),%r8 78 79 movq (%rsp),%rax /* Hardware registers */ 80 movq %r15,PCB_R15(%r8) 81 movq %r14,PCB_R14(%r8) 82 movq %r13,PCB_R13(%r8) 83 movq %r12,PCB_R12(%r8) 84 movq %rbp,PCB_RBP(%r8) 85 movq %rsp,PCB_RSP(%r8) 86 movq %rbx,PCB_RBX(%r8) 87 movq %rax,PCB_RIP(%r8) 88 89 testl $PCB_FULL_IRET,PCB_FLAGS(%r8) 90 jnz 2f 91 orl $PCB_FULL_IRET,PCB_FLAGS(%r8) 92 testl $TDP_KTHREAD,TD_PFLAGS(%rdi) 93 jnz 2f 94 testb $CPUID_STDEXT_FSGSBASE,cpu_stdext_feature(%rip) 95 jz 2f 96 movl %fs,%eax 97 cmpl $KUF32SEL,%eax 98 jne 1f 99 rdfsbase %rax 100 movq %rax,PCB_FSBASE(%r8) 1011: movl %gs,%eax 102 cmpl $KUG32SEL,%eax 103 jne 2f 104 movq %rdx,%r12 105 movl $MSR_KGSBASE,%ecx /* Read user gs base */ 106 rdmsr 107 shlq $32,%rdx 108 orq %rdx,%rax 109 movq %rax,PCB_GSBASE(%r8) 110 movq %r12,%rdx 111 1122: 113 testl $PCB_DBREGS,PCB_FLAGS(%r8) 114 jnz store_dr /* static predict not taken */ 115done_store_dr: 116 117 /* have we used fp, and need a save? */ 118 cmpq %rdi,PCPU(FPCURTHREAD) 119 jne ctx_switch_fpusave_done 120 movq PCB_SAVEFPU(%r8),%r9 121 clts 122 cmpl $0,use_xsave(%rip) 123 jne 1f 124 fxsave (%r9) 125 jmp ctx_switch_fpusave_done 1261: movq %rdx,%rcx 127 movl xsave_mask,%eax 128 movl xsave_mask+4,%edx 129 testl $PCB_32BIT,PCB_FLAGS(%r8) 130 jne ctx_switch_xsave32 131 .globl ctx_switch_xsave 132ctx_switch_xsave: 133 /* This is patched to xsaveopt if supported, see fpuinit_bsp1() */ 134 xsave64 (%r9) 135ctx_switch_xsave_done: 136 movq %rcx,%rdx 137ctx_switch_fpusave_done: 138 /* Save is done. Now fire up new thread. Leave old vmspace. */ 139 movq %rsi,%r12 140 movq %rdi,%r13 141 movq %rdx,%r15 142 movq %rsi,%rdi 143 callq pmap_activate_sw 144 movq %r15,TD_LOCK(%r13) /* Release the old thread */ 145sw1: 146 leaq TD_MD_PCB(%r12),%r8 147#if defined(SCHED_ULE) && defined(SMP) 148 movq $blocked_lock, %rdx 149 movq TD_LOCK(%r12),%rcx 150 cmpq %rcx, %rdx 151 je sw1wait 152sw1cont: 153#endif 154 /* 155 * At this point, we've switched address spaces and are ready 156 * to load up the rest of the next context. 157 */ 158 159 /* Skip loading LDT and user fsbase/gsbase for kthreads */ 160 testl $TDP_KTHREAD,TD_PFLAGS(%r12) 161 jnz do_kthread 162 163 /* 164 * Load ldt register 165 */ 166 movq TD_PROC(%r12),%rcx 167 cmpq $0, P_MD+MD_LDT(%rcx) 168 jne do_ldt 169 xorl %eax,%eax 170ld_ldt: lldt %ax 171 172 /* Restore fs base in GDT */ 173 movl PCB_FSBASE(%r8),%eax 174 movq PCPU(FS32P),%rdx 175 movw %ax,2(%rdx) 176 shrl $16,%eax 177 movb %al,4(%rdx) 178 shrl $8,%eax 179 movb %al,7(%rdx) 180 181 /* Restore gs base in GDT */ 182 movl PCB_GSBASE(%r8),%eax 183 movq PCPU(GS32P),%rdx 184 movw %ax,2(%rdx) 185 shrl $16,%eax 186 movb %al,4(%rdx) 187 shrl $8,%eax 188 movb %al,7(%rdx) 189 190do_kthread: 191 /* Do we need to reload tss ? */ 192 movq PCPU(TSSP),%rax 193 movq PCB_TSSP(%r8),%rdx 194 movq PCPU(PRVSPACE),%r13 195 addq $PC_COMMONTSS,%r13 196 testq %rdx,%rdx 197 cmovzq %r13,%rdx 198 cmpq %rax,%rdx 199 jne do_tss 200done_tss: 201 movq TD_MD_STACK_BASE(%r12),%r9 202 movq %r9,PCPU(RSP0) 203 movq %r8,PCPU(CURPCB) 204 movq PCPU(PTI_RSP0),%rax 205 cmpq $~0,PCPU(UCR3) 206 cmove %r9,%rax 207 movq %rax,TSS_RSP0(%rdx) 208 movq %r12,PCPU(CURTHREAD) /* into next thread */ 209 210 /* Test if debug registers should be restored. */ 211 testl $PCB_DBREGS,PCB_FLAGS(%r8) 212 jnz load_dr /* static predict not taken */ 213done_load_dr: 214 215 /* Restore context. */ 216 movq PCB_R15(%r8),%r15 217 movq PCB_R14(%r8),%r14 218 movq PCB_R13(%r8),%r13 219 movq PCB_R12(%r8),%r12 220 movq PCB_RBP(%r8),%rbp 221 movq PCB_RSP(%r8),%rsp 222 movq PCB_RBX(%r8),%rbx 223 movq PCB_RIP(%r8),%rax 224 movq %rax,(%rsp) 225 movq PCPU(CURTHREAD),%rdi 226 call fpu_activate_sw 227 cmpb $0,cpu_flush_rsb_ctxsw(%rip) 228 jne rsb_flush 229 ret 230 231 /* 232 * We order these strangely for several reasons. 233 * 1: I wanted to use static branch prediction hints 234 * 2: Most athlon64/opteron cpus don't have them. They define 235 * a forward branch as 'predict not taken'. Intel cores have 236 * the 'rep' prefix to invert this. 237 * So, to make it work on both forms of cpu we do the detour. 238 * We use jumps rather than call in order to avoid the stack. 239 */ 240 241store_dr: 242 movq %dr7,%rax /* yes, do the save */ 243 movq %dr0,%r15 244 movq %dr1,%r14 245 movq %dr2,%r13 246 movq %dr3,%r12 247 movq %dr6,%r11 248 movq %r15,PCB_DR0(%r8) 249 movq %r14,PCB_DR1(%r8) 250 movq %r13,PCB_DR2(%r8) 251 movq %r12,PCB_DR3(%r8) 252 movq %r11,PCB_DR6(%r8) 253 movq %rax,PCB_DR7(%r8) 254 andq $0x0000fc00, %rax /* disable all watchpoints */ 255 movq %rax,%dr7 256 jmp done_store_dr 257 258load_dr: 259 movq %dr7,%rax 260 movq PCB_DR0(%r8),%r15 261 movq PCB_DR1(%r8),%r14 262 movq PCB_DR2(%r8),%r13 263 movq PCB_DR3(%r8),%r12 264 movq PCB_DR6(%r8),%r11 265 movq PCB_DR7(%r8),%rcx 266 movq %r15,%dr0 267 movq %r14,%dr1 268 /* Preserve reserved bits in %dr7 */ 269 andq $0x0000fc00,%rax 270 andq $~0x0000fc00,%rcx 271 movq %r13,%dr2 272 movq %r12,%dr3 273 orq %rcx,%rax 274 movq %r11,%dr6 275 movq %rax,%dr7 276 jmp done_load_dr 277 278do_tss: movq %rdx,PCPU(TSSP) 279 movq %rdx,%rcx 280 movq PCPU(TSS),%rax 281 movw %cx,2(%rax) 282 shrq $16,%rcx 283 movb %cl,4(%rax) 284 shrq $8,%rcx 285 movb %cl,7(%rax) 286 shrq $8,%rcx 287 movl %ecx,8(%rax) 288 movb $0x89,5(%rax) /* unset busy */ 289 movl $TSSSEL,%eax 290 ltr %ax 291 jmp done_tss 292 293do_ldt: movq PCPU(LDT),%rax 294 movq P_MD+MD_LDT_SD(%rcx),%rdx 295 movq %rdx,(%rax) 296 movq P_MD+MD_LDT_SD+8(%rcx),%rdx 297 movq %rdx,8(%rax) 298 movl $LDTSEL,%eax 299 jmp ld_ldt 300 301 .globl ctx_switch_xsave32 302ctx_switch_xsave32: 303 xsave (%r9) 304 jmp ctx_switch_xsave_done 305END(cpu_switch) 306 307/* 308 * savectx(pcb) 309 * Update pcb, saving current processor state. 310 */ 311ENTRY(savectx) 312 /* Save caller's return address. */ 313 movq (%rsp),%rax 314 movq %rax,PCB_RIP(%rdi) 315 316 movq %rbx,PCB_RBX(%rdi) 317 movq %rsp,PCB_RSP(%rdi) 318 movq %rbp,PCB_RBP(%rdi) 319 movq %r12,PCB_R12(%rdi) 320 movq %r13,PCB_R13(%rdi) 321 movq %r14,PCB_R14(%rdi) 322 movq %r15,PCB_R15(%rdi) 323 324 movq %cr0,%rax 325 movq %rax,PCB_CR0(%rdi) 326 movq %cr2,%rax 327 movq %rax,PCB_CR2(%rdi) 328 movq %cr3,%rax 329 movq %rax,PCB_CR3(%rdi) 330 movq %cr4,%rax 331 movq %rax,PCB_CR4(%rdi) 332 333 movq %dr0,%rax 334 movq %rax,PCB_DR0(%rdi) 335 movq %dr1,%rax 336 movq %rax,PCB_DR1(%rdi) 337 movq %dr2,%rax 338 movq %rax,PCB_DR2(%rdi) 339 movq %dr3,%rax 340 movq %rax,PCB_DR3(%rdi) 341 movq %dr6,%rax 342 movq %rax,PCB_DR6(%rdi) 343 movq %dr7,%rax 344 movq %rax,PCB_DR7(%rdi) 345 346 movl $MSR_FSBASE,%ecx 347 rdmsr 348 movl %eax,PCB_FSBASE(%rdi) 349 movl %edx,PCB_FSBASE+4(%rdi) 350 movl $MSR_GSBASE,%ecx 351 rdmsr 352 movl %eax,PCB_GSBASE(%rdi) 353 movl %edx,PCB_GSBASE+4(%rdi) 354 movl $MSR_KGSBASE,%ecx 355 rdmsr 356 movl %eax,PCB_KGSBASE(%rdi) 357 movl %edx,PCB_KGSBASE+4(%rdi) 358 movl $MSR_EFER,%ecx 359 rdmsr 360 movl %eax,PCB_EFER(%rdi) 361 movl %edx,PCB_EFER+4(%rdi) 362 movl $MSR_STAR,%ecx 363 rdmsr 364 movl %eax,PCB_STAR(%rdi) 365 movl %edx,PCB_STAR+4(%rdi) 366 movl $MSR_LSTAR,%ecx 367 rdmsr 368 movl %eax,PCB_LSTAR(%rdi) 369 movl %edx,PCB_LSTAR+4(%rdi) 370 movl $MSR_CSTAR,%ecx 371 rdmsr 372 movl %eax,PCB_CSTAR(%rdi) 373 movl %edx,PCB_CSTAR+4(%rdi) 374 movl $MSR_SF_MASK,%ecx 375 rdmsr 376 movl %eax,PCB_SFMASK(%rdi) 377 movl %edx,PCB_SFMASK+4(%rdi) 378 379 sgdt PCB_GDT(%rdi) 380 sidt PCB_IDT(%rdi) 381 sldt PCB_LDT(%rdi) 382 str PCB_TR(%rdi) 383 384 movl $1,%eax 385 ret 386END(savectx) 387 388/* 389 * resumectx(pcb) 390 * Resuming processor state from pcb. 391 */ 392ENTRY(resumectx) 393 /* Switch to KPML5/4phys. */ 394 movq KPML4phys,%rax 395 movq KPML5phys,%rcx 396 cmpl $0, la57 397 cmovne %rcx, %rax 398 movq %rax,%cr3 399 400 /* Force kernel segment registers. */ 401 movl $KDSEL,%eax 402 movw %ax,%ds 403 movw %ax,%es 404 movw %ax,%ss 405 movl $KUF32SEL,%eax 406 movw %ax,%fs 407 movl $KUG32SEL,%eax 408 movw %ax,%gs 409 410 movl $MSR_FSBASE,%ecx 411 movl PCB_FSBASE(%rdi),%eax 412 movl 4 + PCB_FSBASE(%rdi),%edx 413 wrmsr 414 movl $MSR_GSBASE,%ecx 415 movl PCB_GSBASE(%rdi),%eax 416 movl 4 + PCB_GSBASE(%rdi),%edx 417 wrmsr 418 movl $MSR_KGSBASE,%ecx 419 movl PCB_KGSBASE(%rdi),%eax 420 movl 4 + PCB_KGSBASE(%rdi),%edx 421 wrmsr 422 423 /* Restore EFER one more time. */ 424 movl $MSR_EFER,%ecx 425 movl PCB_EFER(%rdi),%eax 426 wrmsr 427 428 /* Restore fast syscall stuff. */ 429 movl $MSR_STAR,%ecx 430 movl PCB_STAR(%rdi),%eax 431 movl 4 + PCB_STAR(%rdi),%edx 432 wrmsr 433 movl $MSR_LSTAR,%ecx 434 movl PCB_LSTAR(%rdi),%eax 435 movl 4 + PCB_LSTAR(%rdi),%edx 436 wrmsr 437 movl $MSR_CSTAR,%ecx 438 movl PCB_CSTAR(%rdi),%eax 439 movl 4 + PCB_CSTAR(%rdi),%edx 440 wrmsr 441 movl $MSR_SF_MASK,%ecx 442 movl PCB_SFMASK(%rdi),%eax 443 wrmsr 444 445 /* Restore CR0, CR2, CR4 and CR3. */ 446 movq PCB_CR0(%rdi),%rax 447 movq %rax,%cr0 448 movq PCB_CR2(%rdi),%rax 449 movq %rax,%cr2 450 movq PCB_CR4(%rdi),%rax 451 movq %rax,%cr4 452 movq PCB_CR3(%rdi),%rax 453 movq %rax,%cr3 454 455 /* Restore descriptor tables. */ 456 lidt PCB_IDT(%rdi) 457 lldt PCB_LDT(%rdi) 458 459#define SDT_SYSTSS 9 460#define SDT_SYSBSY 11 461 462 /* Clear "task busy" bit and reload TR. */ 463 movq PCPU(TSS),%rax 464 andb $(~SDT_SYSBSY | SDT_SYSTSS),5(%rax) 465 movw PCB_TR(%rdi),%ax 466 ltr %ax 467 468#undef SDT_SYSTSS 469#undef SDT_SYSBSY 470 471 /* Restore debug registers. */ 472 movq PCB_DR0(%rdi),%rax 473 movq %rax,%dr0 474 movq PCB_DR1(%rdi),%rax 475 movq %rax,%dr1 476 movq PCB_DR2(%rdi),%rax 477 movq %rax,%dr2 478 movq PCB_DR3(%rdi),%rax 479 movq %rax,%dr3 480 movq PCB_DR6(%rdi),%rax 481 movq %rax,%dr6 482 movq PCB_DR7(%rdi),%rax 483 movq %rax,%dr7 484 485 /* Restore other callee saved registers. */ 486 movq PCB_R15(%rdi),%r15 487 movq PCB_R14(%rdi),%r14 488 movq PCB_R13(%rdi),%r13 489 movq PCB_R12(%rdi),%r12 490 movq PCB_RBP(%rdi),%rbp 491 movq PCB_RSP(%rdi),%rsp 492 movq PCB_RBX(%rdi),%rbx 493 494 /* Restore return address. */ 495 movq PCB_RIP(%rdi),%rax 496 movq %rax,(%rsp) 497 498 xorl %eax,%eax 499 ret 500END(resumectx) 501 502/* Wait for the new thread to become unblocked */ 503#if defined(SCHED_ULE) && defined(SMP) 504sw1wait: 5051: 506 pause 507 movq TD_LOCK(%r12),%rcx 508 cmpq %rcx, %rdx 509 je 1b 510 jmp sw1cont 511#endif 512