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_npx.h" 50 #include "opt_reset.h" 51 #include "opt_cpu.h" 52 53 #include <sys/param.h> 54 #include <sys/systm.h> 55 #include <sys/bio.h> 56 #include <sys/buf.h> 57 #include <sys/kernel.h> 58 #include <sys/ktr.h> 59 #include <sys/lock.h> 60 #include <sys/malloc.h> 61 #include <sys/mbuf.h> 62 #include <sys/mutex.h> 63 #include <sys/proc.h> 64 #include <sys/sysent.h> 65 #include <sys/sf_buf.h> 66 #include <sys/smp.h> 67 #include <sys/sched.h> 68 #include <sys/sysctl.h> 69 #include <sys/unistd.h> 70 #include <sys/vnode.h> 71 #include <sys/vmmeter.h> 72 73 #include <machine/cpu.h> 74 #include <machine/cputypes.h> 75 #include <machine/md_var.h> 76 #include <machine/pcb.h> 77 #include <machine/pcb_ext.h> 78 #include <machine/smp.h> 79 #include <machine/vm86.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_extern.h> 83 #include <vm/vm_kern.h> 84 #include <vm/vm_page.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_param.h> 87 88 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf), 89 "__OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf."); 90 91 union savefpu * 92 get_pcb_user_save_td(struct thread *td) 93 { 94 vm_offset_t p; 95 96 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 97 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN); 98 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area")); 99 return ((union savefpu *)p); 100 } 101 102 union savefpu * 103 get_pcb_user_save_pcb(struct pcb *pcb) 104 { 105 vm_offset_t p; 106 107 p = (vm_offset_t)(pcb + 1); 108 return ((union savefpu *)p); 109 } 110 111 struct pcb * 112 get_pcb_td(struct thread *td) 113 { 114 vm_offset_t p; 115 116 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 117 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) - 118 sizeof(struct pcb); 119 return ((struct pcb *)p); 120 } 121 122 void * 123 alloc_fpusave(int flags) 124 { 125 void *res; 126 struct savefpu_ymm *sf; 127 128 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags); 129 if (use_xsave) { 130 sf = (struct savefpu_ymm *)res; 131 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd)); 132 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask; 133 } 134 return (res); 135 } 136 137 /* 138 * Common code shared between cpu_fork() and cpu_copy_thread() for 139 * initializing a thread. 140 */ 141 static void 142 copy_thread(struct thread *td1, struct thread *td2) 143 { 144 struct pcb *pcb2; 145 146 pcb2 = td2->td_pcb; 147 148 /* Ensure that td1's pcb is up to date for user threads. */ 149 if ((td2->td_pflags & TDP_KTHREAD) == 0) { 150 MPASS(td1 == curthread); 151 td1->td_pcb->pcb_gs = rgs(); 152 critical_enter(); 153 if (PCPU_GET(fpcurthread) == td1) 154 npxsave(td1->td_pcb->pcb_save); 155 critical_exit(); 156 } 157 158 /* Copy td1's pcb */ 159 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 160 161 /* Properly initialize pcb_save */ 162 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 163 164 /* Kernel threads start with clean NPX and segment bases. */ 165 if ((td2->td_pflags & TDP_KTHREAD) != 0) { 166 pcb2->pcb_gs = _udatasel; 167 set_fsbase(td2, 0); 168 set_gsbase(td2, 0); 169 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE | 170 PCB_KERNNPX | PCB_KERNNPX_THR); 171 } else { 172 MPASS((pcb2->pcb_flags & (PCB_KERNNPX | PCB_KERNNPX_THR)) == 0); 173 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2), 174 cpu_max_ext_state_size); 175 } 176 177 /* 178 * Set registers for trampoline to user mode. Leave space for the 179 * return address on stack. These are the kernel mode register values. 180 */ 181 pcb2->pcb_edi = 0; 182 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 183 pcb2->pcb_ebp = 0; 184 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); /* trampoline arg */ 185 pcb2->pcb_ebx = (int)td2; /* trampoline arg */ 186 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp; 187 /* 188 * If we didn't copy the pcb, we'd need to do the following registers: 189 * pcb2->pcb_cr3: cloned above. 190 * pcb2->pcb_dr*: cloned above. 191 * pcb2->pcb_savefpu: cloned above. 192 * pcb2->pcb_flags: cloned above. 193 * pcb2->pcb_onfault: cloned above (always NULL here?). 194 * pcb2->pcb_gs: cloned above. 195 * pcb2->pcb_ext: cleared below. 196 */ 197 pcb2->pcb_ext = NULL; 198 199 /* Setup to release spin count in fork_exit(). */ 200 td2->td_md.md_spinlock_count = 1; 201 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 202 } 203 204 /* 205 * Finish a fork operation, with process p2 nearly set up. 206 * Copy and update the pcb, set up the stack so that the child 207 * ready to run and return to user mode. 208 */ 209 void 210 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 211 { 212 struct proc *p1; 213 struct pcb *pcb2; 214 struct mdproc *mdp2; 215 216 p1 = td1->td_proc; 217 if ((flags & RFPROC) == 0) { 218 if ((flags & RFMEM) == 0) { 219 /* unshare user LDT */ 220 struct mdproc *mdp1 = &p1->p_md; 221 struct proc_ldt *pldt, *pldt1; 222 223 mtx_lock_spin(&dt_lock); 224 if ((pldt1 = mdp1->md_ldt) != NULL && 225 pldt1->ldt_refcnt > 1) { 226 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len); 227 if (pldt == NULL) 228 panic("could not copy LDT"); 229 mdp1->md_ldt = pldt; 230 set_user_ldt(mdp1); 231 user_ldt_deref(pldt1); 232 } else 233 mtx_unlock_spin(&dt_lock); 234 } 235 return; 236 } 237 238 /* Point the pcb to the top of the stack */ 239 pcb2 = get_pcb_td(td2); 240 td2->td_pcb = pcb2; 241 242 copy_thread(td1, td2); 243 244 /* Reset debug registers in the new process */ 245 x86_clear_dbregs(pcb2); 246 247 /* Point mdproc and then copy over td1's contents */ 248 mdp2 = &p2->p_md; 249 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 250 251 /* 252 * Copy the trap frame for the return to user mode as if from a 253 * syscall. This copies most of the user mode register values. 254 * The -VM86_STACK_SPACE (-16) is so we can expand the trapframe 255 * if we go to vm86. 256 */ 257 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 258 VM86_STACK_SPACE) - 1; 259 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 260 261 /* Set child return values. */ 262 p2->p_sysent->sv_set_fork_retval(td2); 263 264 /* 265 * If the parent process has the trap bit set (i.e. a debugger 266 * had single stepped the process to the system call), we need 267 * to clear the trap flag from the new frame. 268 */ 269 td2->td_frame->tf_eflags &= ~PSL_T; 270 271 /* Set cr3 for the new process. */ 272 pcb2->pcb_cr3 = pmap_get_cr3(vmspace_pmap(p2->p_vmspace)); 273 274 /* 275 * XXX don't copy the i/o pages. this should probably be fixed. 276 */ 277 pcb2->pcb_ext = NULL; 278 279 /* Copy the LDT, if necessary. */ 280 mtx_lock_spin(&dt_lock); 281 if (mdp2->md_ldt != NULL) { 282 if (flags & RFMEM) { 283 mdp2->md_ldt->ldt_refcnt++; 284 } else { 285 mdp2->md_ldt = user_ldt_alloc(mdp2, 286 mdp2->md_ldt->ldt_len); 287 if (mdp2->md_ldt == NULL) 288 panic("could not copy LDT"); 289 } 290 } 291 mtx_unlock_spin(&dt_lock); 292 293 /* 294 * Now, cpu_switch() can schedule the new process. 295 * pcb_esp is loaded pointing to the cpu_switch() stack frame 296 * containing the return address when exiting cpu_switch. 297 * This will normally be to fork_trampoline(), which will have 298 * %ebx loaded with the new proc's pointer. fork_trampoline() 299 * will set up a stack to call fork_return(p, frame); to complete 300 * the return to user-mode. 301 */ 302 } 303 304 void 305 x86_set_fork_retval(struct thread *td) 306 { 307 struct trapframe * frame = td->td_frame; 308 309 frame->tf_eax = 0; /* Child returns zero */ 310 frame->tf_eflags &= ~PSL_C; /* success */ 311 frame->tf_edx = 1; /* System V emulation */ 312 } 313 314 /* 315 * Intercept the return address from a freshly forked process that has NOT 316 * been scheduled yet. 317 * 318 * This is needed to make kernel threads stay in kernel mode. 319 */ 320 void 321 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg) 322 { 323 /* 324 * Note that the trap frame follows the args, so the function 325 * is really called like this: func(arg, frame); 326 */ 327 td->td_pcb->pcb_esi = (int) func; /* function */ 328 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 329 } 330 331 void 332 cpu_exit(struct thread *td) 333 { 334 335 /* 336 * If this process has a custom LDT, release it. Reset pc->pcb_gs 337 * and %gs before we free it in case they refer to an LDT entry. 338 */ 339 mtx_lock_spin(&dt_lock); 340 if (td->td_proc->p_md.md_ldt) { 341 td->td_pcb->pcb_gs = _udatasel; 342 load_gs(_udatasel); 343 user_ldt_free(td); 344 } else 345 mtx_unlock_spin(&dt_lock); 346 } 347 348 void 349 cpu_thread_exit(struct thread *td) 350 { 351 352 critical_enter(); 353 if (td == PCPU_GET(fpcurthread)) 354 npxdrop(); 355 critical_exit(); 356 357 /* Disable any hardware breakpoints. */ 358 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 359 reset_dbregs(); 360 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 361 } 362 } 363 364 void 365 cpu_thread_clean(struct thread *td) 366 { 367 struct pcb *pcb; 368 369 pcb = td->td_pcb; 370 if (pcb->pcb_ext != NULL) { 371 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 372 /* 373 * XXX do we need to move the TSS off the allocated pages 374 * before freeing them? (not done here) 375 */ 376 pmap_trm_free(pcb->pcb_ext, ctob(IOPAGES + 1)); 377 pcb->pcb_ext = NULL; 378 } 379 } 380 381 void 382 cpu_thread_swapin(struct thread *td) 383 { 384 } 385 386 void 387 cpu_thread_swapout(struct thread *td) 388 { 389 } 390 391 void 392 cpu_thread_alloc(struct thread *td) 393 { 394 struct pcb *pcb; 395 struct xstate_hdr *xhdr; 396 397 td->td_pcb = pcb = get_pcb_td(td); 398 td->td_frame = (struct trapframe *)((caddr_t)pcb - 399 VM86_STACK_SPACE) - 1; 400 pcb->pcb_ext = NULL; 401 pcb->pcb_save = get_pcb_user_save_pcb(pcb); 402 if (use_xsave) { 403 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1); 404 bzero(xhdr, sizeof(*xhdr)); 405 xhdr->xstate_bv = xsave_mask; 406 } 407 } 408 409 void 410 cpu_thread_free(struct thread *td) 411 { 412 413 cpu_thread_clean(td); 414 } 415 416 bool 417 cpu_exec_vmspace_reuse(struct proc *p __unused, vm_map_t map __unused) 418 { 419 420 return (true); 421 } 422 423 int 424 cpu_procctl(struct thread *td __unused, int idtype __unused, id_t id __unused, 425 int com __unused, void *data __unused) 426 { 427 428 return (EINVAL); 429 } 430 431 void 432 cpu_set_syscall_retval(struct thread *td, int error) 433 { 434 435 switch (error) { 436 case 0: 437 td->td_frame->tf_eax = td->td_retval[0]; 438 td->td_frame->tf_edx = td->td_retval[1]; 439 td->td_frame->tf_eflags &= ~PSL_C; 440 break; 441 442 case ERESTART: 443 /* 444 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int 445 * 0x80 is 2 bytes. We saved this in tf_err. 446 */ 447 td->td_frame->tf_eip -= td->td_frame->tf_err; 448 break; 449 450 case EJUSTRETURN: 451 break; 452 453 default: 454 td->td_frame->tf_eax = error; 455 td->td_frame->tf_eflags |= PSL_C; 456 break; 457 } 458 } 459 460 /* 461 * Initialize machine state, mostly pcb and trap frame for a new 462 * thread, about to return to userspace. Put enough state in the new 463 * thread's PCB to get it to go back to the fork_return(), which 464 * finalizes the thread state and handles peculiarities of the first 465 * return to userspace for the new thread. 466 */ 467 void 468 cpu_copy_thread(struct thread *td, struct thread *td0) 469 { 470 copy_thread(td0, td); 471 472 /* 473 * Copy user general-purpose registers. 474 * 475 * Some of these registers are rewritten by cpu_set_upcall() 476 * and linux_set_upcall(). 477 */ 478 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 479 480 /* If the current thread has the trap bit set (i.e. a debugger had 481 * single stepped the process to the system call), we need to clear 482 * the trap flag from the new frame. Otherwise, the new thread will 483 * receive a (likely unexpected) SIGTRAP when it executes the first 484 * instruction after returning to userland. 485 */ 486 td->td_frame->tf_eflags &= ~PSL_T; 487 } 488 489 /* 490 * Set that machine state for performing an upcall that starts 491 * the entry function with the given argument. 492 */ 493 void 494 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg, 495 stack_t *stack) 496 { 497 498 /* 499 * Do any extra cleaning that needs to be done. 500 * The thread may have optional components 501 * that are not present in a fresh thread. 502 * This may be a recycled thread so make it look 503 * as though it's newly allocated. 504 */ 505 cpu_thread_clean(td); 506 507 /* 508 * Set the trap frame to point at the beginning of the entry 509 * function. 510 */ 511 td->td_frame->tf_ebp = 0; 512 td->td_frame->tf_esp = 513 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 514 td->td_frame->tf_eip = (int)entry; 515 516 /* Return address sentinel value to stop stack unwinding. */ 517 suword((void *)td->td_frame->tf_esp, 0); 518 519 /* Pass the argument to the entry point. */ 520 suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 521 (int)arg); 522 } 523 524 int 525 cpu_set_user_tls(struct thread *td, void *tls_base) 526 { 527 struct segment_descriptor sd; 528 uint32_t base; 529 530 /* 531 * Construct a descriptor and store it in the pcb for 532 * the next context switch. Also store it in the gdt 533 * so that the load of tf_fs into %fs will activate it 534 * at return to userland. 535 */ 536 base = (uint32_t)tls_base; 537 sd.sd_lobase = base & 0xffffff; 538 sd.sd_hibase = (base >> 24) & 0xff; 539 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 540 sd.sd_hilimit = 0xf; 541 sd.sd_type = SDT_MEMRWA; 542 sd.sd_dpl = SEL_UPL; 543 sd.sd_p = 1; 544 sd.sd_xx = 0; 545 sd.sd_def32 = 1; 546 sd.sd_gran = 1; 547 critical_enter(); 548 /* set %gs */ 549 td->td_pcb->pcb_gsd = sd; 550 if (td == curthread) { 551 PCPU_GET(fsgs_gdt)[1] = sd; 552 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 553 } 554 critical_exit(); 555 return (0); 556 } 557 558 /* 559 * Convert kernel VA to physical address 560 */ 561 vm_paddr_t 562 kvtop(void *addr) 563 { 564 vm_paddr_t pa; 565 566 pa = pmap_kextract((vm_offset_t)addr); 567 if (pa == 0) 568 panic("kvtop: zero page frame"); 569 return (pa); 570 } 571 572 /* 573 * Get an sf_buf from the freelist. May block if none are available. 574 */ 575 void 576 sf_buf_map(struct sf_buf *sf, int flags) 577 { 578 579 pmap_sf_buf_map(sf); 580 #ifdef SMP 581 sf_buf_shootdown(sf, flags); 582 #endif 583 } 584 585 #ifdef SMP 586 static void 587 sf_buf_shootdown_curcpu_cb(pmap_t pmap __unused, 588 vm_offset_t addr1 __unused, vm_offset_t addr2 __unused) 589 { 590 } 591 592 void 593 sf_buf_shootdown(struct sf_buf *sf, int flags) 594 { 595 cpuset_t other_cpus; 596 u_int cpuid; 597 598 sched_pin(); 599 cpuid = PCPU_GET(cpuid); 600 if (!CPU_ISSET(cpuid, &sf->cpumask)) { 601 CPU_SET(cpuid, &sf->cpumask); 602 invlpg(sf->kva); 603 } 604 if ((flags & SFB_CPUPRIVATE) == 0) { 605 other_cpus = all_cpus; 606 CPU_CLR(cpuid, &other_cpus); 607 CPU_ANDNOT(&other_cpus, &sf->cpumask); 608 if (!CPU_EMPTY(&other_cpus)) { 609 CPU_OR(&sf->cpumask, &other_cpus); 610 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap, 611 sf_buf_shootdown_curcpu_cb); 612 } 613 } 614 sched_unpin(); 615 } 616 #endif 617 618 /* 619 * MD part of sf_buf_free(). 620 */ 621 int 622 sf_buf_unmap(struct sf_buf *sf) 623 { 624 625 return (0); 626 } 627 628 static void 629 sf_buf_invalidate(struct sf_buf *sf) 630 { 631 vm_page_t m = sf->m; 632 633 /* 634 * Use pmap_qenter to update the pte for 635 * existing mapping, in particular, the PAT 636 * settings are recalculated. 637 */ 638 pmap_qenter(sf->kva, &m, 1); 639 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE); 640 } 641 642 /* 643 * Invalidate the cache lines that may belong to the page, if 644 * (possibly old) mapping of the page by sf buffer exists. Returns 645 * TRUE when mapping was found and cache invalidated. 646 */ 647 boolean_t 648 sf_buf_invalidate_cache(vm_page_t m) 649 { 650 651 return (sf_buf_process_page(m, sf_buf_invalidate)); 652 } 653 654 /* 655 * Software interrupt handler for queued VM system processing. 656 */ 657 void 658 swi_vm(void *dummy) 659 { 660 if (busdma_swi_pending != 0) 661 busdma_swi(); 662 } 663 664 /* 665 * Tell whether this address is in some physical memory region. 666 * Currently used by the kernel coredump code in order to avoid 667 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 668 * or other unpredictable behaviour. 669 */ 670 671 int 672 is_physical_memory(vm_paddr_t addr) 673 { 674 675 #ifdef DEV_ISA 676 /* The ISA ``memory hole''. */ 677 if (addr >= 0xa0000 && addr < 0x100000) 678 return 0; 679 #endif 680 681 /* 682 * stuff other tests for known memory-mapped devices (PCI?) 683 * here 684 */ 685 686 return 1; 687 } 688