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