vm_machdep.c revision 147889
1/*- 2 * Copyright (c) 1982, 1986 The Regents of the University of California. 3 * Copyright (c) 1989, 1990 William Jolitz 4 * Copyright (c) 1994 John Dyson 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * the Systems Programming Group of the University of Utah Computer 9 * Science Department, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41 */ 42 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: head/sys/amd64/amd64/vm_machdep.c 147889 2005-07-10 23:31:11Z davidxu $"); 45 46#include "opt_isa.h" 47#include "opt_cpu.h" 48 49#include <sys/param.h> 50#include <sys/systm.h> 51#include <sys/bio.h> 52#include <sys/buf.h> 53#include <sys/kse.h> 54#include <sys/kernel.h> 55#include <sys/ktr.h> 56#include <sys/lock.h> 57#include <sys/malloc.h> 58#include <sys/mbuf.h> 59#include <sys/mutex.h> 60#include <sys/pioctl.h> 61#include <sys/proc.h> 62#include <sys/sf_buf.h> 63#include <sys/smp.h> 64#include <sys/sysctl.h> 65#include <sys/unistd.h> 66#include <sys/vnode.h> 67#include <sys/vmmeter.h> 68 69#include <machine/cpu.h> 70#include <machine/md_var.h> 71#include <machine/pcb.h> 72 73#include <vm/vm.h> 74#include <vm/vm_extern.h> 75#include <vm/vm_kern.h> 76#include <vm/vm_page.h> 77#include <vm/vm_map.h> 78#include <vm/vm_param.h> 79 80#include <amd64/isa/isa.h> 81 82static void cpu_reset_real(void); 83#ifdef SMP 84static void cpu_reset_proxy(void); 85static u_int cpu_reset_proxyid; 86static volatile u_int cpu_reset_proxy_active; 87#endif 88 89/* 90 * Finish a fork operation, with process p2 nearly set up. 91 * Copy and update the pcb, set up the stack so that the child 92 * ready to run and return to user mode. 93 */ 94void 95cpu_fork(td1, p2, td2, flags) 96 register struct thread *td1; 97 register struct proc *p2; 98 struct thread *td2; 99 int flags; 100{ 101 register struct proc *p1; 102 struct pcb *pcb2; 103 struct mdproc *mdp2; 104 105 p1 = td1->td_proc; 106 if ((flags & RFPROC) == 0) 107 return; 108 109 /* Ensure that p1's pcb is up to date. */ 110 fpuexit(td1); 111 112 /* Point the pcb to the top of the stack */ 113 pcb2 = (struct pcb *)(td2->td_kstack + 114 td2->td_kstack_pages * PAGE_SIZE) - 1; 115 td2->td_pcb = pcb2; 116 117 /* Copy p1's pcb */ 118 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 119 120 /* Point mdproc and then copy over td1's contents */ 121 mdp2 = &p2->p_md; 122 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 123 124 /* 125 * Create a new fresh stack for the new process. 126 * Copy the trap frame for the return to user mode as if from a 127 * syscall. This copies most of the user mode register values. 128 */ 129 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 130 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 131 132 td2->td_frame->tf_rax = 0; /* Child returns zero */ 133 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 134 td2->td_frame->tf_rdx = 1; 135 136 /* 137 * If the parent process has the trap bit set (i.e. a debugger had 138 * single stepped the process to the system call), we need to clear 139 * the trap flag from the new frame unless the debugger had set PF_FORK 140 * on the parent. Otherwise, the child will receive a (likely 141 * unexpected) SIGTRAP when it executes the first instruction after 142 * returning to userland. 143 */ 144 if ((p1->p_pfsflags & PF_FORK) == 0) 145 td2->td_frame->tf_rflags &= ~PSL_T; 146 147 /* 148 * Set registers for trampoline to user mode. Leave space for the 149 * return address on stack. These are the kernel mode register values. 150 */ 151 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 152 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 153 pcb2->pcb_rbp = 0; 154 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 155 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 156 pcb2->pcb_rip = (register_t)fork_trampoline; 157 pcb2->pcb_rflags = td2->td_frame->tf_rflags & ~PSL_I; /* ints disabled */ 158 /*- 159 * pcb2->pcb_dr*: cloned above. 160 * pcb2->pcb_savefpu: cloned above. 161 * pcb2->pcb_flags: cloned above. 162 * pcb2->pcb_onfault: cloned above (always NULL here?). 163 * pcb2->pcb_[fg]sbase: cloned above 164 */ 165 166 /* Setup to release sched_lock in fork_exit(). */ 167 td2->td_md.md_spinlock_count = 1; 168 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 169 170 /* 171 * Now, cpu_switch() can schedule the new process. 172 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 173 * containing the return address when exiting cpu_switch. 174 * This will normally be to fork_trampoline(), which will have 175 * %ebx loaded with the new proc's pointer. fork_trampoline() 176 * will set up a stack to call fork_return(p, frame); to complete 177 * the return to user-mode. 178 */ 179} 180 181/* 182 * Intercept the return address from a freshly forked process that has NOT 183 * been scheduled yet. 184 * 185 * This is needed to make kernel threads stay in kernel mode. 186 */ 187void 188cpu_set_fork_handler(td, func, arg) 189 struct thread *td; 190 void (*func)(void *); 191 void *arg; 192{ 193 /* 194 * Note that the trap frame follows the args, so the function 195 * is really called like this: func(arg, frame); 196 */ 197 td->td_pcb->pcb_r12 = (long) func; /* function */ 198 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 199} 200 201void 202cpu_exit(struct thread *td) 203{ 204} 205 206void 207cpu_thread_exit(struct thread *td) 208{ 209 210 if (td == PCPU_GET(fpcurthread)) 211 fpudrop(); 212 213 /* Disable any hardware breakpoints. */ 214 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 215 reset_dbregs(); 216 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 217 } 218} 219 220void 221cpu_thread_clean(struct thread *td) 222{ 223} 224 225void 226cpu_thread_swapin(struct thread *td) 227{ 228} 229 230void 231cpu_thread_swapout(struct thread *td) 232{ 233} 234 235void 236cpu_thread_setup(struct thread *td) 237{ 238 239 td->td_pcb = (struct pcb *)(td->td_kstack + 240 td->td_kstack_pages * PAGE_SIZE) - 1; 241 td->td_frame = (struct trapframe *)td->td_pcb - 1; 242} 243 244/* 245 * Initialize machine state (pcb and trap frame) for a new thread about to 246 * upcall. Put enough state in the new thread's PCB to get it to go back 247 * userret(), where we can intercept it again to set the return (upcall) 248 * Address and stack, along with those from upcals that are from other sources 249 * such as those generated in thread_userret() itself. 250 */ 251void 252cpu_set_upcall(struct thread *td, struct thread *td0) 253{ 254 struct pcb *pcb2; 255 256 /* Point the pcb to the top of the stack. */ 257 pcb2 = td->td_pcb; 258 259 /* 260 * Copy the upcall pcb. This loads kernel regs. 261 * Those not loaded individually below get their default 262 * values here. 263 * 264 * XXXKSE It might be a good idea to simply skip this as 265 * the values of the other registers may be unimportant. 266 * This would remove any requirement for knowing the KSE 267 * at this time (see the matching comment below for 268 * more analysis) (need a good safe default). 269 */ 270 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 271 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 272 273 /* 274 * Create a new fresh stack for the new thread. 275 * Don't forget to set this stack value into whatever supplies 276 * the address for the fault handlers. 277 * The contexts are filled in at the time we actually DO the 278 * upcall as only then do we know which KSE we got. 279 */ 280 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 281 282 /* 283 * Set registers for trampoline to user mode. Leave space for the 284 * return address on stack. These are the kernel mode register values. 285 */ 286 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 287 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 288 pcb2->pcb_rbp = 0; 289 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 290 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 291 pcb2->pcb_rip = (register_t)fork_trampoline; 292 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 293 /* 294 * If we didn't copy the pcb, we'd need to do the following registers: 295 * pcb2->pcb_dr*: cloned above. 296 * pcb2->pcb_savefpu: cloned above. 297 * pcb2->pcb_rflags: cloned above. 298 * pcb2->pcb_onfault: cloned above (always NULL here?). 299 * pcb2->pcb_[fg]sbase: cloned above 300 */ 301 302 /* Setup to release sched_lock in fork_exit(). */ 303 td->td_md.md_spinlock_count = 1; 304 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 305} 306 307/* 308 * Set that machine state for performing an upcall that has to 309 * be done in thread_userret() so that those upcalls generated 310 * in thread_userret() itself can be done as well. 311 */ 312void 313cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 314 stack_t *stack) 315{ 316 317 /* 318 * Do any extra cleaning that needs to be done. 319 * The thread may have optional components 320 * that are not present in a fresh thread. 321 * This may be a recycled thread so make it look 322 * as though it's newly allocated. 323 */ 324 cpu_thread_clean(td); 325 326 /* 327 * Set the trap frame to point at the beginning of the uts 328 * function. 329 */ 330 td->td_frame->tf_rbp = 0; 331 td->td_frame->tf_rsp = 332 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 333 td->td_frame->tf_rsp -= 8; 334 td->td_frame->tf_rbp = 0; 335 td->td_frame->tf_rip = (register_t)entry; 336 337 /* 338 * Pass the address of the mailbox for this kse to the uts 339 * function as a parameter on the stack. 340 */ 341 td->td_frame->tf_rdi = (register_t)arg; 342} 343 344int 345cpu_set_user_tls(struct thread *td, void *tls_base) 346{ 347 348 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 349 return (EINVAL); 350 351 if (td == curthread) { 352 critical_enter(); 353 td->td_pcb->pcb_fsbase = (register_t)tls_base; 354 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase); 355 critical_exit(); 356 } else { 357 td->td_pcb->pcb_fsbase = (register_t)tls_base; 358 } 359 return (0); 360} 361 362#ifdef SMP 363static void 364cpu_reset_proxy() 365{ 366 367 cpu_reset_proxy_active = 1; 368 while (cpu_reset_proxy_active == 1) 369 ; /* Wait for other cpu to see that we've started */ 370 stop_cpus((1<<cpu_reset_proxyid)); 371 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 372 DELAY(1000000); 373 cpu_reset_real(); 374} 375#endif 376 377void 378cpu_reset() 379{ 380#ifdef SMP 381 u_int cnt, map; 382 383 if (smp_active) { 384 map = PCPU_GET(other_cpus) & ~stopped_cpus; 385 if (map != 0) { 386 printf("cpu_reset: Stopping other CPUs\n"); 387 stop_cpus(map); 388 } 389 390 if (PCPU_GET(cpuid) != 0) { 391 cpu_reset_proxyid = PCPU_GET(cpuid); 392 cpustop_restartfunc = cpu_reset_proxy; 393 cpu_reset_proxy_active = 0; 394 printf("cpu_reset: Restarting BSP\n"); 395 started_cpus = (1<<0); /* Restart CPU #0 */ 396 397 cnt = 0; 398 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 399 cnt++; /* Wait for BSP to announce restart */ 400 if (cpu_reset_proxy_active == 0) 401 printf("cpu_reset: Failed to restart BSP\n"); 402 enable_intr(); 403 cpu_reset_proxy_active = 2; 404 405 while (1); 406 /* NOTREACHED */ 407 } 408 409 DELAY(1000000); 410 } 411#endif 412 cpu_reset_real(); 413 /* NOTREACHED */ 414} 415 416static void 417cpu_reset_real() 418{ 419 420 /* 421 * Attempt to do a CPU reset via the keyboard controller, 422 * do not turn off GateA20, as any machine that fails 423 * to do the reset here would then end up in no man's land. 424 */ 425 outb(IO_KBD + 4, 0xFE); 426 DELAY(500000); /* wait 0.5 sec to see if that did it */ 427 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 428 DELAY(1000000); /* wait 1 sec for printf to complete */ 429 430 /* Force a shutdown by unmapping entire address space. */ 431 bzero((caddr_t)PML4map, PAGE_SIZE); 432 433 /* "good night, sweet prince .... <THUNK!>" */ 434 invltlb(); 435 /* NOTREACHED */ 436 while(1); 437} 438 439/* 440 * Allocate an sf_buf for the given vm_page. On this machine, however, there 441 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 442 * returned. 443 */ 444struct sf_buf * 445sf_buf_alloc(struct vm_page *m, int pri) 446{ 447 448 return ((struct sf_buf *)m); 449} 450 451/* 452 * Free the sf_buf. In fact, do nothing because there are no resources 453 * associated with the sf_buf. 454 */ 455void 456sf_buf_free(struct sf_buf *sf) 457{ 458} 459 460/* 461 * Software interrupt handler for queued VM system processing. 462 */ 463void 464swi_vm(void *dummy) 465{ 466 if (busdma_swi_pending != 0) 467 busdma_swi(); 468} 469 470/* 471 * Tell whether this address is in some physical memory region. 472 * Currently used by the kernel coredump code in order to avoid 473 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 474 * or other unpredictable behaviour. 475 */ 476 477int 478is_physical_memory(vm_paddr_t addr) 479{ 480 481#ifdef DEV_ISA 482 /* The ISA ``memory hole''. */ 483 if (addr >= 0xa0000 && addr < 0x100000) 484 return 0; 485#endif 486 487 /* 488 * stuff other tests for known memory-mapped devices (PCI?) 489 * here 490 */ 491 492 return 1; 493} 494