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/pioctl.h> 64 #include <sys/proc.h> 65 #include <sys/sysent.h> 66 #include <sys/sf_buf.h> 67 #include <sys/smp.h> 68 #include <sys/sched.h> 69 #include <sys/sysctl.h> 70 #include <sys/unistd.h> 71 #include <sys/vnode.h> 72 #include <sys/vmmeter.h> 73 74 #include <machine/cpu.h> 75 #include <machine/cputypes.h> 76 #include <machine/md_var.h> 77 #include <machine/pcb.h> 78 #include <machine/pcb_ext.h> 79 #include <machine/smp.h> 80 #include <machine/vm86.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_kern.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_param.h> 88 89 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf), 90 "__OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf."); 91 92 union savefpu * 93 get_pcb_user_save_td(struct thread *td) 94 { 95 vm_offset_t p; 96 97 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 98 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN); 99 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area")); 100 return ((union savefpu *)p); 101 } 102 103 union savefpu * 104 get_pcb_user_save_pcb(struct pcb *pcb) 105 { 106 vm_offset_t p; 107 108 p = (vm_offset_t)(pcb + 1); 109 return ((union savefpu *)p); 110 } 111 112 struct pcb * 113 get_pcb_td(struct thread *td) 114 { 115 vm_offset_t p; 116 117 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 118 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) - 119 sizeof(struct pcb); 120 return ((struct pcb *)p); 121 } 122 123 void * 124 alloc_fpusave(int flags) 125 { 126 void *res; 127 struct savefpu_ymm *sf; 128 129 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags); 130 if (use_xsave) { 131 sf = (struct savefpu_ymm *)res; 132 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd)); 133 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask; 134 } 135 return (res); 136 } 137 /* 138 * Finish a fork operation, with process p2 nearly set up. 139 * Copy and update the pcb, set up the stack so that the child 140 * ready to run and return to user mode. 141 */ 142 void 143 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 144 { 145 struct proc *p1; 146 struct pcb *pcb2; 147 struct mdproc *mdp2; 148 149 p1 = td1->td_proc; 150 if ((flags & RFPROC) == 0) { 151 if ((flags & RFMEM) == 0) { 152 /* unshare user LDT */ 153 struct mdproc *mdp1 = &p1->p_md; 154 struct proc_ldt *pldt, *pldt1; 155 156 mtx_lock_spin(&dt_lock); 157 if ((pldt1 = mdp1->md_ldt) != NULL && 158 pldt1->ldt_refcnt > 1) { 159 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len); 160 if (pldt == NULL) 161 panic("could not copy LDT"); 162 mdp1->md_ldt = pldt; 163 set_user_ldt(mdp1); 164 user_ldt_deref(pldt1); 165 } else 166 mtx_unlock_spin(&dt_lock); 167 } 168 return; 169 } 170 171 /* Ensure that td1's pcb is up to date. */ 172 if (td1 == curthread) 173 td1->td_pcb->pcb_gs = rgs(); 174 critical_enter(); 175 if (PCPU_GET(fpcurthread) == td1) 176 npxsave(td1->td_pcb->pcb_save); 177 critical_exit(); 178 179 /* Point the pcb to the top of the stack */ 180 pcb2 = get_pcb_td(td2); 181 td2->td_pcb = pcb2; 182 183 /* Copy td1's pcb */ 184 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 185 186 /* Properly initialize pcb_save */ 187 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 188 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2), 189 cpu_max_ext_state_size); 190 191 /* Point mdproc and then copy over td1's contents */ 192 mdp2 = &p2->p_md; 193 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 194 195 /* 196 * Create a new fresh stack for the new process. 197 * Copy the trap frame for the return to user mode as if from a 198 * syscall. This copies most of the user mode register values. 199 * The -VM86_STACK_SPACE (-16) is so we can expand the trapframe 200 * if we go to vm86. 201 */ 202 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 203 VM86_STACK_SPACE) - 1; 204 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 205 206 td2->td_frame->tf_eax = 0; /* Child returns zero */ 207 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 208 td2->td_frame->tf_edx = 1; 209 210 /* 211 * If the parent process has the trap bit set (i.e. a debugger had 212 * single stepped the process to the system call), we need to clear 213 * the trap flag from the new frame unless the debugger had set PF_FORK 214 * on the parent. Otherwise, the child will receive a (likely 215 * unexpected) SIGTRAP when it executes the first instruction after 216 * returning to userland. 217 */ 218 if ((p1->p_pfsflags & PF_FORK) == 0) 219 td2->td_frame->tf_eflags &= ~PSL_T; 220 221 /* 222 * Set registers for trampoline to user mode. Leave space for the 223 * return address on stack. These are the kernel mode register values. 224 */ 225 pcb2->pcb_cr3 = pmap_get_cr3(vmspace_pmap(p2->p_vmspace)); 226 pcb2->pcb_edi = 0; 227 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 228 pcb2->pcb_ebp = 0; 229 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 230 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 231 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp; 232 /*- 233 * pcb2->pcb_dr*: cloned above. 234 * pcb2->pcb_savefpu: cloned above. 235 * pcb2->pcb_flags: cloned above. 236 * pcb2->pcb_onfault: cloned above (always NULL here?). 237 * pcb2->pcb_gs: cloned above. 238 * pcb2->pcb_ext: cleared below. 239 */ 240 241 /* 242 * XXX don't copy the i/o pages. this should probably be fixed. 243 */ 244 pcb2->pcb_ext = 0; 245 246 /* Copy the LDT, if necessary. */ 247 mtx_lock_spin(&dt_lock); 248 if (mdp2->md_ldt != NULL) { 249 if (flags & RFMEM) { 250 mdp2->md_ldt->ldt_refcnt++; 251 } else { 252 mdp2->md_ldt = user_ldt_alloc(mdp2, 253 mdp2->md_ldt->ldt_len); 254 if (mdp2->md_ldt == NULL) 255 panic("could not copy LDT"); 256 } 257 } 258 mtx_unlock_spin(&dt_lock); 259 260 /* Setup to release spin count in fork_exit(). */ 261 td2->td_md.md_spinlock_count = 1; 262 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 263 264 /* 265 * Now, cpu_switch() can schedule the new process. 266 * pcb_esp is loaded pointing to the cpu_switch() stack frame 267 * containing the return address when exiting cpu_switch. 268 * This will normally be to fork_trampoline(), which will have 269 * %ebx loaded with the new proc's pointer. fork_trampoline() 270 * will set up a stack to call fork_return(p, frame); to complete 271 * the return to user-mode. 272 */ 273 } 274 275 /* 276 * Intercept the return address from a freshly forked process that has NOT 277 * been scheduled yet. 278 * 279 * This is needed to make kernel threads stay in kernel mode. 280 */ 281 void 282 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg) 283 { 284 /* 285 * Note that the trap frame follows the args, so the function 286 * is really called like this: func(arg, frame); 287 */ 288 td->td_pcb->pcb_esi = (int) func; /* function */ 289 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 290 } 291 292 void 293 cpu_exit(struct thread *td) 294 { 295 296 /* 297 * If this process has a custom LDT, release it. Reset pc->pcb_gs 298 * and %gs before we free it in case they refer to an LDT entry. 299 */ 300 mtx_lock_spin(&dt_lock); 301 if (td->td_proc->p_md.md_ldt) { 302 td->td_pcb->pcb_gs = _udatasel; 303 load_gs(_udatasel); 304 user_ldt_free(td); 305 } else 306 mtx_unlock_spin(&dt_lock); 307 } 308 309 void 310 cpu_thread_exit(struct thread *td) 311 { 312 313 critical_enter(); 314 if (td == PCPU_GET(fpcurthread)) 315 npxdrop(); 316 critical_exit(); 317 318 /* Disable any hardware breakpoints. */ 319 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 320 reset_dbregs(); 321 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 322 } 323 } 324 325 void 326 cpu_thread_clean(struct thread *td) 327 { 328 struct pcb *pcb; 329 330 pcb = td->td_pcb; 331 if (pcb->pcb_ext != NULL) { 332 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 333 /* 334 * XXX do we need to move the TSS off the allocated pages 335 * before freeing them? (not done here) 336 */ 337 pmap_trm_free(pcb->pcb_ext, ctob(IOPAGES + 1)); 338 pcb->pcb_ext = NULL; 339 } 340 } 341 342 void 343 cpu_thread_swapin(struct thread *td) 344 { 345 } 346 347 void 348 cpu_thread_swapout(struct thread *td) 349 { 350 } 351 352 void 353 cpu_thread_alloc(struct thread *td) 354 { 355 struct pcb *pcb; 356 struct xstate_hdr *xhdr; 357 358 td->td_pcb = pcb = get_pcb_td(td); 359 td->td_frame = (struct trapframe *)((caddr_t)pcb - 360 VM86_STACK_SPACE) - 1; 361 pcb->pcb_ext = NULL; 362 pcb->pcb_save = get_pcb_user_save_pcb(pcb); 363 if (use_xsave) { 364 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1); 365 bzero(xhdr, sizeof(*xhdr)); 366 xhdr->xstate_bv = xsave_mask; 367 } 368 } 369 370 void 371 cpu_thread_free(struct thread *td) 372 { 373 374 cpu_thread_clean(td); 375 } 376 377 bool 378 cpu_exec_vmspace_reuse(struct proc *p __unused, vm_map_t map __unused) 379 { 380 381 return (true); 382 } 383 384 int 385 cpu_procctl(struct thread *td __unused, int idtype __unused, id_t id __unused, 386 int com __unused, void *data __unused) 387 { 388 389 return (EINVAL); 390 } 391 392 void 393 cpu_set_syscall_retval(struct thread *td, int error) 394 { 395 396 switch (error) { 397 case 0: 398 td->td_frame->tf_eax = td->td_retval[0]; 399 td->td_frame->tf_edx = td->td_retval[1]; 400 td->td_frame->tf_eflags &= ~PSL_C; 401 break; 402 403 case ERESTART: 404 /* 405 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int 406 * 0x80 is 2 bytes. We saved this in tf_err. 407 */ 408 td->td_frame->tf_eip -= td->td_frame->tf_err; 409 break; 410 411 case EJUSTRETURN: 412 break; 413 414 default: 415 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error); 416 td->td_frame->tf_eflags |= PSL_C; 417 break; 418 } 419 } 420 421 /* 422 * Initialize machine state, mostly pcb and trap frame for a new 423 * thread, about to return to userspace. Put enough state in the new 424 * thread's PCB to get it to go back to the fork_return(), which 425 * finalizes the thread state and handles peculiarities of the first 426 * return to userspace for the new thread. 427 */ 428 void 429 cpu_copy_thread(struct thread *td, struct thread *td0) 430 { 431 struct pcb *pcb2; 432 433 /* Point the pcb to the top of the stack. */ 434 pcb2 = td->td_pcb; 435 436 /* 437 * Copy the upcall pcb. This loads kernel regs. 438 * Those not loaded individually below get their default 439 * values here. 440 */ 441 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 442 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE | 443 PCB_KERNNPX); 444 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 445 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save, 446 cpu_max_ext_state_size); 447 448 /* 449 * Create a new fresh stack for the new thread. 450 */ 451 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 452 453 /* If the current thread has the trap bit set (i.e. a debugger had 454 * single stepped the process to the system call), we need to clear 455 * the trap flag from the new frame. Otherwise, the new thread will 456 * receive a (likely unexpected) SIGTRAP when it executes the first 457 * instruction after returning to userland. 458 */ 459 td->td_frame->tf_eflags &= ~PSL_T; 460 461 /* 462 * Set registers for trampoline to user mode. Leave space for the 463 * return address on stack. These are the kernel mode register values. 464 */ 465 pcb2->pcb_edi = 0; 466 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 467 pcb2->pcb_ebp = 0; 468 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 469 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 470 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp; 471 pcb2->pcb_gs = rgs(); 472 /* 473 * If we didn't copy the pcb, we'd need to do the following registers: 474 * pcb2->pcb_cr3: cloned above. 475 * pcb2->pcb_dr*: cloned above. 476 * pcb2->pcb_savefpu: cloned above. 477 * pcb2->pcb_flags: cloned above. 478 * pcb2->pcb_onfault: cloned above (always NULL here?). 479 * pcb2->pcb_gs: cloned above. 480 * pcb2->pcb_ext: cleared below. 481 */ 482 pcb2->pcb_ext = NULL; 483 484 /* Setup to release spin count in fork_exit(). */ 485 td->td_md.md_spinlock_count = 1; 486 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 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 void 587 sf_buf_shootdown(struct sf_buf *sf, int flags) 588 { 589 cpuset_t other_cpus; 590 u_int cpuid; 591 592 sched_pin(); 593 cpuid = PCPU_GET(cpuid); 594 if (!CPU_ISSET(cpuid, &sf->cpumask)) { 595 CPU_SET(cpuid, &sf->cpumask); 596 invlpg(sf->kva); 597 } 598 if ((flags & SFB_CPUPRIVATE) == 0) { 599 other_cpus = all_cpus; 600 CPU_CLR(cpuid, &other_cpus); 601 CPU_NAND(&other_cpus, &sf->cpumask); 602 if (!CPU_EMPTY(&other_cpus)) { 603 CPU_OR(&sf->cpumask, &other_cpus); 604 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap); 605 } 606 } 607 sched_unpin(); 608 } 609 #endif 610 611 /* 612 * MD part of sf_buf_free(). 613 */ 614 int 615 sf_buf_unmap(struct sf_buf *sf) 616 { 617 618 return (0); 619 } 620 621 static void 622 sf_buf_invalidate(struct sf_buf *sf) 623 { 624 vm_page_t m = sf->m; 625 626 /* 627 * Use pmap_qenter to update the pte for 628 * existing mapping, in particular, the PAT 629 * settings are recalculated. 630 */ 631 pmap_qenter(sf->kva, &m, 1); 632 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE); 633 } 634 635 /* 636 * Invalidate the cache lines that may belong to the page, if 637 * (possibly old) mapping of the page by sf buffer exists. Returns 638 * TRUE when mapping was found and cache invalidated. 639 */ 640 boolean_t 641 sf_buf_invalidate_cache(vm_page_t m) 642 { 643 644 return (sf_buf_process_page(m, sf_buf_invalidate)); 645 } 646 647 /* 648 * Software interrupt handler for queued VM system processing. 649 */ 650 void 651 swi_vm(void *dummy) 652 { 653 if (busdma_swi_pending != 0) 654 busdma_swi(); 655 } 656 657 /* 658 * Tell whether this address is in some physical memory region. 659 * Currently used by the kernel coredump code in order to avoid 660 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 661 * or other unpredictable behaviour. 662 */ 663 664 int 665 is_physical_memory(vm_paddr_t addr) 666 { 667 668 #ifdef DEV_ISA 669 /* The ISA ``memory hole''. */ 670 if (addr >= 0xa0000 && addr < 0x100000) 671 return 0; 672 #endif 673 674 /* 675 * stuff other tests for known memory-mapped devices (PCI?) 676 * here 677 */ 678 679 return 1; 680 } 681