1 /*- 2 * SPDX-License-Identifier: BSD-4-Clause 3 * 4 * Copyright (c) 1982, 1986 The Regents of the University of California. 5 * Copyright (c) 1989, 1990 William Jolitz 6 * Copyright (c) 1994 John Dyson 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * the Systems Programming Group of the University of Utah Computer 11 * Science Department, and William Jolitz. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 42 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_isa.h" 49 #include "opt_cpu.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/bio.h> 54 #include <sys/buf.h> 55 #include <sys/kernel.h> 56 #include <sys/ktr.h> 57 #include <sys/lock.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/mutex.h> 61 #include <sys/priv.h> 62 #include <sys/proc.h> 63 #include <sys/procctl.h> 64 #include <sys/smp.h> 65 #include <sys/sysctl.h> 66 #include <sys/sysent.h> 67 #include <sys/unistd.h> 68 #include <sys/vnode.h> 69 #include <sys/vmmeter.h> 70 #include <sys/wait.h> 71 72 #include <machine/cpu.h> 73 #include <machine/md_var.h> 74 #include <machine/pcb.h> 75 #include <machine/smp.h> 76 #include <machine/specialreg.h> 77 #include <machine/tss.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_extern.h> 81 #include <vm/vm_kern.h> 82 #include <vm/vm_page.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_param.h> 85 86 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf), 87 "OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf."); 88 89 void 90 set_top_of_stack_td(struct thread *td) 91 { 92 td->td_md.md_stack_base = td->td_kstack + 93 td->td_kstack_pages * PAGE_SIZE - 94 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN); 95 } 96 97 struct savefpu * 98 get_pcb_user_save_td(struct thread *td) 99 { 100 vm_offset_t p; 101 102 p = td->td_md.md_stack_base; 103 KASSERT((p % XSAVE_AREA_ALIGN) == 0, 104 ("Unaligned pcb_user_save area ptr %#lx td %p", p, td)); 105 return ((struct savefpu *)p); 106 } 107 108 struct pcb * 109 get_pcb_td(struct thread *td) 110 { 111 112 return (&td->td_md.md_pcb); 113 } 114 115 struct savefpu * 116 get_pcb_user_save_pcb(struct pcb *pcb) 117 { 118 struct thread *td; 119 120 td = __containerof(pcb, struct thread, td_md.md_pcb); 121 return (get_pcb_user_save_td(td)); 122 } 123 124 void * 125 alloc_fpusave(int flags) 126 { 127 void *res; 128 struct savefpu_ymm *sf; 129 130 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags); 131 if (use_xsave) { 132 sf = (struct savefpu_ymm *)res; 133 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd)); 134 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask; 135 } 136 return (res); 137 } 138 139 /* 140 * Finish a fork operation, with process p2 nearly set up. 141 * Copy and update the pcb, set up the stack so that the child 142 * ready to run and return to user mode. 143 */ 144 void 145 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 146 { 147 struct proc *p1; 148 struct pcb *pcb2; 149 struct mdproc *mdp1, *mdp2; 150 struct proc_ldt *pldt; 151 152 p1 = td1->td_proc; 153 if ((flags & RFPROC) == 0) { 154 if ((flags & RFMEM) == 0) { 155 /* unshare user LDT */ 156 mdp1 = &p1->p_md; 157 mtx_lock(&dt_lock); 158 if ((pldt = mdp1->md_ldt) != NULL && 159 pldt->ldt_refcnt > 1 && 160 user_ldt_alloc(p1, 1) == NULL) 161 panic("could not copy LDT"); 162 mtx_unlock(&dt_lock); 163 } 164 return; 165 } 166 167 /* Ensure that td1's pcb is up to date for user processes. */ 168 if ((td2->td_pflags & TDP_KTHREAD) == 0) { 169 MPASS(td1 == curthread); 170 fpuexit(td1); 171 update_pcb_bases(td1->td_pcb); 172 } 173 174 /* Point the stack and pcb to the actual location */ 175 set_top_of_stack_td(td2); 176 td2->td_pcb = pcb2 = get_pcb_td(td2); 177 178 /* Copy td1's pcb */ 179 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 180 181 /* Properly initialize pcb_save */ 182 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 183 184 /* Kernel processes start with clean FPU and segment bases. */ 185 if ((td2->td_pflags & TDP_KTHREAD) != 0) { 186 pcb2->pcb_fsbase = 0; 187 pcb2->pcb_gsbase = 0; 188 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE | 189 PCB_KERNFPU | PCB_KERNFPU_THR); 190 } else { 191 MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0); 192 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2), 193 cpu_max_ext_state_size); 194 } 195 196 /* Point mdproc and then copy over td1's contents */ 197 mdp2 = &p2->p_md; 198 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 199 200 /* 201 * Copy the trap frame for the return to user mode as if from a 202 * syscall. This copies most of the user mode register values. 203 */ 204 td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1; 205 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 206 207 td2->td_frame->tf_rax = 0; /* Child returns zero */ 208 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 209 td2->td_frame->tf_rdx = 1; 210 211 /* 212 * If the parent process has the trap bit set (i.e. a debugger 213 * had single stepped the process to the system call), we need 214 * to clear the trap flag from the new frame. 215 */ 216 td2->td_frame->tf_rflags &= ~PSL_T; 217 218 /* 219 * Set registers for trampoline to user mode. Leave space for the 220 * return address on stack. These are the kernel mode register values. 221 */ 222 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 223 pcb2->pcb_rbp = 0; 224 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 225 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 226 pcb2->pcb_rip = (register_t)fork_trampoline; 227 /*- 228 * pcb2->pcb_dr*: cloned above. 229 * pcb2->pcb_savefpu: cloned above. 230 * pcb2->pcb_flags: cloned above. 231 * pcb2->pcb_onfault: cloned above (always NULL here?). 232 * pcb2->pcb_[fg]sbase: cloned above 233 */ 234 235 /* Setup to release spin count in fork_exit(). */ 236 td2->td_md.md_spinlock_count = 1; 237 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 238 pmap_thread_init_invl_gen(td2); 239 240 /* As an i386, do not copy io permission bitmap. */ 241 pcb2->pcb_tssp = NULL; 242 243 /* New segment registers. */ 244 set_pcb_flags_raw(pcb2, PCB_FULL_IRET); 245 246 /* Copy the LDT, if necessary. */ 247 mdp1 = &td1->td_proc->p_md; 248 mdp2 = &p2->p_md; 249 if (mdp1->md_ldt == NULL) { 250 mdp2->md_ldt = NULL; 251 return; 252 } 253 mtx_lock(&dt_lock); 254 if (mdp1->md_ldt != NULL) { 255 if (flags & RFMEM) { 256 mdp1->md_ldt->ldt_refcnt++; 257 mdp2->md_ldt = mdp1->md_ldt; 258 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct 259 system_segment_descriptor)); 260 } else { 261 mdp2->md_ldt = NULL; 262 mdp2->md_ldt = user_ldt_alloc(p2, 0); 263 if (mdp2->md_ldt == NULL) 264 panic("could not copy LDT"); 265 amd64_set_ldt_data(td2, 0, max_ldt_segment, 266 (struct user_segment_descriptor *) 267 mdp1->md_ldt->ldt_base); 268 } 269 } else 270 mdp2->md_ldt = NULL; 271 mtx_unlock(&dt_lock); 272 273 /* 274 * Now, cpu_switch() can schedule the new process. 275 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 276 * containing the return address when exiting cpu_switch. 277 * This will normally be to fork_trampoline(), which will have 278 * %ebx loaded with the new proc's pointer. fork_trampoline() 279 * will set up a stack to call fork_return(p, frame); to complete 280 * the return to user-mode. 281 */ 282 } 283 284 /* 285 * Intercept the return address from a freshly forked process that has NOT 286 * been scheduled yet. 287 * 288 * This is needed to make kernel threads stay in kernel mode. 289 */ 290 void 291 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg) 292 { 293 /* 294 * Note that the trap frame follows the args, so the function 295 * is really called like this: func(arg, frame); 296 */ 297 td->td_pcb->pcb_r12 = (long) func; /* function */ 298 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 299 } 300 301 void 302 cpu_exit(struct thread *td) 303 { 304 305 /* 306 * If this process has a custom LDT, release it. 307 */ 308 if (td->td_proc->p_md.md_ldt != NULL) 309 user_ldt_free(td); 310 } 311 312 void 313 cpu_thread_exit(struct thread *td) 314 { 315 struct pcb *pcb; 316 317 critical_enter(); 318 if (td == PCPU_GET(fpcurthread)) 319 fpudrop(); 320 critical_exit(); 321 322 pcb = td->td_pcb; 323 324 /* Disable any hardware breakpoints. */ 325 if (pcb->pcb_flags & PCB_DBREGS) { 326 reset_dbregs(); 327 clear_pcb_flags(pcb, PCB_DBREGS); 328 } 329 } 330 331 void 332 cpu_thread_clean(struct thread *td) 333 { 334 struct pcb *pcb; 335 336 pcb = td->td_pcb; 337 338 /* 339 * Clean TSS/iomap 340 */ 341 if (pcb->pcb_tssp != NULL) { 342 pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp, 343 (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1)); 344 kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1)); 345 pcb->pcb_tssp = NULL; 346 } 347 } 348 349 void 350 cpu_thread_swapin(struct thread *td) 351 { 352 } 353 354 void 355 cpu_thread_swapout(struct thread *td) 356 { 357 } 358 359 void 360 cpu_thread_alloc(struct thread *td) 361 { 362 struct pcb *pcb; 363 struct xstate_hdr *xhdr; 364 365 set_top_of_stack_td(td); 366 td->td_pcb = pcb = get_pcb_td(td); 367 td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1; 368 pcb->pcb_save = get_pcb_user_save_pcb(pcb); 369 if (use_xsave) { 370 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1); 371 bzero(xhdr, sizeof(*xhdr)); 372 xhdr->xstate_bv = xsave_mask; 373 } 374 } 375 376 void 377 cpu_thread_free(struct thread *td) 378 { 379 380 cpu_thread_clean(td); 381 } 382 383 bool 384 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map) 385 { 386 387 return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) == 388 (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3)); 389 } 390 391 static void 392 cpu_procctl_kpti_ctl(struct proc *p, int val) 393 { 394 395 if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC) 396 p->p_md.md_flags |= P_MD_KPTI; 397 if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC) 398 p->p_md.md_flags &= ~P_MD_KPTI; 399 } 400 401 static void 402 cpu_procctl_kpti_status(struct proc *p, int *val) 403 { 404 *val = (p->p_md.md_flags & P_MD_KPTI) != 0 ? 405 PROC_KPTI_CTL_ENABLE_ON_EXEC: 406 PROC_KPTI_CTL_DISABLE_ON_EXEC; 407 if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3) 408 *val |= PROC_KPTI_STATUS_ACTIVE; 409 } 410 411 static int 412 cpu_procctl_la_ctl(struct proc *p, int val) 413 { 414 int error; 415 416 error = 0; 417 switch (val) { 418 case PROC_LA_CTL_LA48_ON_EXEC: 419 p->p_md.md_flags |= P_MD_LA48; 420 p->p_md.md_flags &= ~P_MD_LA57; 421 break; 422 case PROC_LA_CTL_LA57_ON_EXEC: 423 if (la57) { 424 p->p_md.md_flags &= ~P_MD_LA48; 425 p->p_md.md_flags |= P_MD_LA57; 426 } else { 427 error = ENOTSUP; 428 } 429 break; 430 case PROC_LA_CTL_DEFAULT_ON_EXEC: 431 p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57); 432 break; 433 } 434 return (error); 435 } 436 437 static void 438 cpu_procctl_la_status(struct proc *p, int *val) 439 { 440 int res; 441 442 if ((p->p_md.md_flags & P_MD_LA48) != 0) 443 res = PROC_LA_CTL_LA48_ON_EXEC; 444 else if ((p->p_md.md_flags & P_MD_LA57) != 0) 445 res = PROC_LA_CTL_LA57_ON_EXEC; 446 else 447 res = PROC_LA_CTL_DEFAULT_ON_EXEC; 448 if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48) 449 res |= PROC_LA_STATUS_LA48; 450 else 451 res |= PROC_LA_STATUS_LA57; 452 *val = res; 453 } 454 455 int 456 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data) 457 { 458 struct proc *p; 459 int error, val; 460 461 switch (com) { 462 case PROC_KPTI_CTL: 463 case PROC_KPTI_STATUS: 464 case PROC_LA_CTL: 465 case PROC_LA_STATUS: 466 if (idtype != P_PID) { 467 error = EINVAL; 468 break; 469 } 470 if (com == PROC_KPTI_CTL) { 471 /* sad but true and not a joke */ 472 error = priv_check(td, PRIV_IO); 473 if (error != 0) 474 break; 475 } 476 if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) { 477 error = copyin(data, &val, sizeof(val)); 478 if (error != 0) 479 break; 480 } 481 if (com == PROC_KPTI_CTL && 482 val != PROC_KPTI_CTL_ENABLE_ON_EXEC && 483 val != PROC_KPTI_CTL_DISABLE_ON_EXEC) { 484 error = EINVAL; 485 break; 486 } 487 if (com == PROC_LA_CTL && 488 val != PROC_LA_CTL_LA48_ON_EXEC && 489 val != PROC_LA_CTL_LA57_ON_EXEC && 490 val != PROC_LA_CTL_DEFAULT_ON_EXEC) { 491 error = EINVAL; 492 break; 493 } 494 error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p); 495 if (error != 0) 496 break; 497 switch (com) { 498 case PROC_KPTI_CTL: 499 cpu_procctl_kpti_ctl(p, val); 500 break; 501 case PROC_KPTI_STATUS: 502 cpu_procctl_kpti_status(p, &val); 503 break; 504 case PROC_LA_CTL: 505 error = cpu_procctl_la_ctl(p, val); 506 break; 507 case PROC_LA_STATUS: 508 cpu_procctl_la_status(p, &val); 509 break; 510 } 511 PROC_UNLOCK(p); 512 if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS) 513 error = copyout(&val, data, sizeof(val)); 514 break; 515 default: 516 error = EINVAL; 517 break; 518 } 519 return (error); 520 } 521 522 void 523 cpu_set_syscall_retval(struct thread *td, int error) 524 { 525 struct trapframe *frame; 526 527 frame = td->td_frame; 528 if (__predict_true(error == 0)) { 529 frame->tf_rax = td->td_retval[0]; 530 frame->tf_rdx = td->td_retval[1]; 531 frame->tf_rflags &= ~PSL_C; 532 return; 533 } 534 535 switch (error) { 536 case ERESTART: 537 /* 538 * Reconstruct pc, we know that 'syscall' is 2 bytes, 539 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 540 * We saved this in tf_err. 541 * %r10 (which was holding the value of %rcx) is restored 542 * for the next iteration. 543 * %r10 restore is only required for freebsd/amd64 processes, 544 * but shall be innocent for any ia32 ABI. 545 * 546 * Require full context restore to get the arguments 547 * in the registers reloaded at return to usermode. 548 */ 549 frame->tf_rip -= frame->tf_err; 550 frame->tf_r10 = frame->tf_rcx; 551 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 552 break; 553 554 case EJUSTRETURN: 555 break; 556 557 default: 558 frame->tf_rax = error; 559 frame->tf_rflags |= PSL_C; 560 break; 561 } 562 } 563 564 /* 565 * Initialize machine state, mostly pcb and trap frame for a new 566 * thread, about to return to userspace. Put enough state in the new 567 * thread's PCB to get it to go back to the fork_return(), which 568 * finalizes the thread state and handles peculiarities of the first 569 * return to userspace for the new thread. 570 */ 571 void 572 cpu_copy_thread(struct thread *td, struct thread *td0) 573 { 574 struct pcb *pcb2; 575 576 pcb2 = td->td_pcb; 577 578 /* Ensure that td0's pcb is up to date for user threads. */ 579 if ((td->td_pflags & TDP_KTHREAD) == 0) { 580 MPASS(td0 == curthread); 581 fpuexit(td0); 582 update_pcb_bases(td0->td_pcb); 583 } 584 585 /* 586 * Copy the upcall pcb. This loads kernel regs. 587 * Those not loaded individually below get their default 588 * values here. 589 */ 590 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 591 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 592 593 /* Kernel threads start with clean FPU and segment bases. */ 594 if ((td->td_pflags & TDP_KTHREAD) != 0) { 595 pcb2->pcb_fsbase = 0; 596 pcb2->pcb_gsbase = 0; 597 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE | 598 PCB_KERNFPU | PCB_KERNFPU_THR); 599 } else { 600 MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0); 601 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save, 602 cpu_max_ext_state_size); 603 } 604 set_pcb_flags_raw(pcb2, PCB_FULL_IRET); 605 606 607 /* 608 * Copy user general-purpose registers. 609 * 610 * Some of these registers are rewritten by cpu_set_upcall() 611 * and linux_set_upcall_kse(). 612 */ 613 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 614 615 /* If the current thread has the trap bit set (i.e. a debugger had 616 * single stepped the process to the system call), we need to clear 617 * the trap flag from the new frame. Otherwise, the new thread will 618 * receive a (likely unexpected) SIGTRAP when it executes the first 619 * instruction after returning to userland. 620 */ 621 td->td_frame->tf_rflags &= ~PSL_T; 622 623 /* 624 * Set registers for trampoline to user mode. Leave space for the 625 * return address on stack. These are the kernel mode register values. 626 */ 627 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 628 pcb2->pcb_rbp = 0; 629 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 630 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 631 pcb2->pcb_rip = (register_t)fork_trampoline; 632 /* 633 * If we didn't copy the pcb, we'd need to do the following registers: 634 * pcb2->pcb_dr*: cloned above. 635 * pcb2->pcb_savefpu: cloned above. 636 * pcb2->pcb_onfault: cloned above (always NULL here?). 637 * pcb2->pcb_[fg]sbase: cloned above 638 */ 639 640 /* Setup to release spin count in fork_exit(). */ 641 td->td_md.md_spinlock_count = 1; 642 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 643 pmap_thread_init_invl_gen(td); 644 } 645 646 /* 647 * Set that machine state for performing an upcall that starts 648 * the entry function with the given argument. 649 */ 650 void 651 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg, 652 stack_t *stack) 653 { 654 655 /* 656 * Do any extra cleaning that needs to be done. 657 * The thread may have optional components 658 * that are not present in a fresh thread. 659 * This may be a recycled thread so make it look 660 * as though it's newly allocated. 661 */ 662 cpu_thread_clean(td); 663 664 #ifdef COMPAT_FREEBSD32 665 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 666 /* 667 * Set the trap frame to point at the beginning of the entry 668 * function. 669 */ 670 td->td_frame->tf_rbp = 0; 671 td->td_frame->tf_rsp = 672 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 673 td->td_frame->tf_rip = (uintptr_t)entry; 674 675 /* Return address sentinel value to stop stack unwinding. */ 676 suword32((void *)td->td_frame->tf_rsp, 0); 677 678 /* Pass the argument to the entry point. */ 679 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 680 (uint32_t)(uintptr_t)arg); 681 682 return; 683 } 684 #endif 685 686 /* 687 * Set the trap frame to point at the beginning of the uts 688 * function. 689 */ 690 td->td_frame->tf_rbp = 0; 691 td->td_frame->tf_rsp = 692 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 693 td->td_frame->tf_rsp -= 8; 694 td->td_frame->tf_rip = (register_t)entry; 695 td->td_frame->tf_ds = _udatasel; 696 td->td_frame->tf_es = _udatasel; 697 td->td_frame->tf_fs = _ufssel; 698 td->td_frame->tf_gs = _ugssel; 699 td->td_frame->tf_flags = TF_HASSEGS; 700 701 /* Return address sentinel value to stop stack unwinding. */ 702 suword((void *)td->td_frame->tf_rsp, 0); 703 704 /* Pass the argument to the entry point. */ 705 td->td_frame->tf_rdi = (register_t)arg; 706 } 707 708 int 709 cpu_set_user_tls(struct thread *td, void *tls_base) 710 { 711 struct pcb *pcb; 712 713 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 714 return (EINVAL); 715 716 pcb = td->td_pcb; 717 set_pcb_flags(pcb, PCB_FULL_IRET); 718 #ifdef COMPAT_FREEBSD32 719 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 720 pcb->pcb_gsbase = (register_t)tls_base; 721 return (0); 722 } 723 #endif 724 pcb->pcb_fsbase = (register_t)tls_base; 725 return (0); 726 } 727 728 /* 729 * Software interrupt handler for queued VM system processing. 730 */ 731 void 732 swi_vm(void *dummy) 733 { 734 if (busdma_swi_pending != 0) 735 busdma_swi(); 736 } 737 738 /* 739 * Tell whether this address is in some physical memory region. 740 * Currently used by the kernel coredump code in order to avoid 741 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 742 * or other unpredictable behaviour. 743 */ 744 745 int 746 is_physical_memory(vm_paddr_t addr) 747 { 748 749 #ifdef DEV_ISA 750 /* The ISA ``memory hole''. */ 751 if (addr >= 0xa0000 && addr < 0x100000) 752 return 0; 753 #endif 754 755 /* 756 * stuff other tests for known memory-mapped devices (PCI?) 757 * here 758 */ 759 760 return 1; 761 } 762