vm_machdep.c revision 128384
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 128384 2004-04-18 05:36:37Z alc $"); 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/proc.h> 61#include <sys/sf_buf.h> 62#include <sys/smp.h> 63#include <sys/sysctl.h> 64#include <sys/unistd.h> 65#include <sys/user.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 * Set registers for trampoline to user mode. Leave space for the 138 * return address on stack. These are the kernel mode register values. 139 */ 140 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 141 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 142 pcb2->pcb_rbp = 0; 143 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 144 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 145 pcb2->pcb_rip = (register_t)fork_trampoline; 146 pcb2->pcb_rflags = td2->td_frame->tf_rflags & ~PSL_I; /* ints disabled */ 147 /*- 148 * pcb2->pcb_dr*: cloned above. 149 * pcb2->pcb_savefpu: cloned above. 150 * pcb2->pcb_flags: cloned above. 151 * pcb2->pcb_onfault: cloned above (always NULL here?). 152 * pcb2->pcb_[fg]sbase: cloned above 153 */ 154 155 /* 156 * Now, cpu_switch() can schedule the new process. 157 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 158 * containing the return address when exiting cpu_switch. 159 * This will normally be to fork_trampoline(), which will have 160 * %ebx loaded with the new proc's pointer. fork_trampoline() 161 * will set up a stack to call fork_return(p, frame); to complete 162 * the return to user-mode. 163 */ 164} 165 166/* 167 * Intercept the return address from a freshly forked process that has NOT 168 * been scheduled yet. 169 * 170 * This is needed to make kernel threads stay in kernel mode. 171 */ 172void 173cpu_set_fork_handler(td, func, arg) 174 struct thread *td; 175 void (*func)(void *); 176 void *arg; 177{ 178 /* 179 * Note that the trap frame follows the args, so the function 180 * is really called like this: func(arg, frame); 181 */ 182 td->td_pcb->pcb_r12 = (long) func; /* function */ 183 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 184} 185 186void 187cpu_exit(struct thread *td) 188{ 189 struct pcb *pcb = td->td_pcb; 190 191 if (pcb->pcb_flags & PCB_DBREGS) { 192 /* disable all hardware breakpoints */ 193 reset_dbregs(); 194 pcb->pcb_flags &= ~PCB_DBREGS; 195 } 196} 197 198void 199cpu_thread_exit(struct thread *td) 200{ 201 struct pcb *pcb = td->td_pcb; 202 203 if (td == PCPU_GET(fpcurthread)) 204 fpudrop(); 205 if (pcb->pcb_flags & PCB_DBREGS) { 206 /* disable all hardware breakpoints */ 207 reset_dbregs(); 208 pcb->pcb_flags &= ~PCB_DBREGS; 209 } 210} 211 212void 213cpu_thread_clean(struct thread *td) 214{ 215} 216 217void 218cpu_thread_swapin(struct thread *td) 219{ 220} 221 222void 223cpu_thread_swapout(struct thread *td) 224{ 225} 226 227void 228cpu_sched_exit(td) 229 register struct thread *td; 230{ 231} 232 233void 234cpu_thread_setup(struct thread *td) 235{ 236 237 td->td_pcb = (struct pcb *)(td->td_kstack + 238 td->td_kstack_pages * PAGE_SIZE) - 1; 239 td->td_frame = (struct trapframe *)td->td_pcb - 1; 240} 241 242/* 243 * Initialize machine state (pcb and trap frame) for a new thread about to 244 * upcall. Pu t enough state in the new thread's PCB to get it to go back 245 * userret(), where we can intercept it again to set the return (upcall) 246 * Address and stack, along with those from upcals that are from other sources 247 * such as those generated in thread_userret() itself. 248 */ 249void 250cpu_set_upcall(struct thread *td, struct thread *td0) 251{ 252 struct pcb *pcb2; 253 254 /* Point the pcb to the top of the stack. */ 255 pcb2 = td->td_pcb; 256 257 /* 258 * Copy the upcall pcb. This loads kernel regs. 259 * Those not loaded individually below get their default 260 * values here. 261 * 262 * XXXKSE It might be a good idea to simply skip this as 263 * the values of the other registers may be unimportant. 264 * This would remove any requirement for knowing the KSE 265 * at this time (see the matching comment below for 266 * more analysis) (need a good safe default). 267 */ 268 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 269 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 270 271 /* 272 * Create a new fresh stack for the new thread. 273 * Don't forget to set this stack value into whatever supplies 274 * the address for the fault handlers. 275 * The contexts are filled in at the time we actually DO the 276 * upcall as only then do we know which KSE we got. 277 */ 278 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 279 280 /* 281 * Set registers for trampoline to user mode. Leave space for the 282 * return address on stack. These are the kernel mode register values. 283 */ 284 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 285 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 286 pcb2->pcb_rbp = 0; 287 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 288 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 289 pcb2->pcb_rip = (register_t)fork_trampoline; 290 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 291 /* 292 * If we didn't copy the pcb, we'd need to do the following registers: 293 * pcb2->pcb_dr*: cloned above. 294 * pcb2->pcb_savefpu: cloned above. 295 * pcb2->pcb_rflags: cloned above. 296 * pcb2->pcb_onfault: cloned above (always NULL here?). 297 * pcb2->pcb_[fg]sbase: cloned above 298 */ 299} 300 301/* 302 * Set that machine state for performing an upcall that has to 303 * be done in thread_userret() so that those upcalls generated 304 * in thread_userret() itself can be done as well. 305 */ 306void 307cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 308{ 309 310 /* 311 * Do any extra cleaning that needs to be done. 312 * The thread may have optional components 313 * that are not present in a fresh thread. 314 * This may be a recycled thread so make it look 315 * as though it's newly allocated. 316 */ 317 cpu_thread_clean(td); 318 319 /* 320 * Set the trap frame to point at the beginning of the uts 321 * function. 322 */ 323 td->td_frame->tf_rsp = 324 ((register_t)ku->ku_stack.ss_sp + ku->ku_stack.ss_size) & ~0x0f; 325 td->td_frame->tf_rsp -= 8; 326 td->td_frame->tf_rip = (register_t)ku->ku_func; 327 328 /* 329 * Pass the address of the mailbox for this kse to the uts 330 * function as a parameter on the stack. 331 */ 332 td->td_frame->tf_rdi = (register_t)ku->ku_mailbox; 333} 334 335 336/* 337 * Force reset the processor by invalidating the entire address space! 338 */ 339 340#ifdef SMP 341static void 342cpu_reset_proxy() 343{ 344 345 cpu_reset_proxy_active = 1; 346 while (cpu_reset_proxy_active == 1) 347 ; /* Wait for other cpu to see that we've started */ 348 stop_cpus((1<<cpu_reset_proxyid)); 349 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 350 DELAY(1000000); 351 cpu_reset_real(); 352} 353#endif 354 355void 356cpu_reset() 357{ 358#ifdef SMP 359 if (smp_active == 0) { 360 cpu_reset_real(); 361 /* NOTREACHED */ 362 } else { 363 364 u_int map; 365 int cnt; 366 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 367 368 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 369 370 if (map != 0) { 371 printf("cpu_reset: Stopping other CPUs\n"); 372 stop_cpus(map); /* Stop all other CPUs */ 373 } 374 375 if (PCPU_GET(cpuid) == 0) { 376 DELAY(1000000); 377 cpu_reset_real(); 378 /* NOTREACHED */ 379 } else { 380 /* We are not BSP (CPU #0) */ 381 382 cpu_reset_proxyid = PCPU_GET(cpuid); 383 cpustop_restartfunc = cpu_reset_proxy; 384 cpu_reset_proxy_active = 0; 385 printf("cpu_reset: Restarting BSP\n"); 386 started_cpus = (1<<0); /* Restart CPU #0 */ 387 388 cnt = 0; 389 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 390 cnt++; /* Wait for BSP to announce restart */ 391 if (cpu_reset_proxy_active == 0) 392 printf("cpu_reset: Failed to restart BSP\n"); 393 enable_intr(); 394 cpu_reset_proxy_active = 2; 395 396 while (1); 397 /* NOTREACHED */ 398 } 399 } 400#else 401 cpu_reset_real(); 402#endif 403} 404 405static void 406cpu_reset_real() 407{ 408 409 /* 410 * Attempt to do a CPU reset via the keyboard controller, 411 * do not turn of the GateA20, as any machine that fails 412 * to do the reset here would then end up in no man's land. 413 */ 414 415 outb(IO_KBD + 4, 0xFE); 416 DELAY(500000); /* wait 0.5 sec to see if that did it */ 417 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 418 DELAY(1000000); /* wait 1 sec for printf to complete */ 419 /* force a shutdown by unmapping entire address space ! */ 420 bzero((caddr_t)PML4map, PAGE_SIZE); 421 422 /* "good night, sweet prince .... <THUNK!>" */ 423 invltlb(); 424 /* NOTREACHED */ 425 while(1); 426} 427 428/* 429 * Allocate an sf_buf for the given vm_page. On this machine, however, there 430 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 431 * returned. 432 */ 433struct sf_buf * 434sf_buf_alloc(struct vm_page *m, int pri) 435{ 436 437 return ((struct sf_buf *)m); 438} 439 440/* 441 * Free the sf_buf. In fact, do nothing because there are no resources 442 * associated with the sf_buf. 443 */ 444void 445sf_buf_free(struct sf_buf *sf) 446{ 447} 448 449/* 450 * Software interrupt handler for queued VM system processing. 451 */ 452void 453swi_vm(void *dummy) 454{ 455 if (busdma_swi_pending != 0) 456 busdma_swi(); 457} 458 459/* 460 * Tell whether this address is in some physical memory region. 461 * Currently used by the kernel coredump code in order to avoid 462 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 463 * or other unpredictable behaviour. 464 */ 465 466int 467is_physical_memory(vm_paddr_t addr) 468{ 469 470#ifdef DEV_ISA 471 /* The ISA ``memory hole''. */ 472 if (addr >= 0xa0000 && addr < 0x100000) 473 return 0; 474#endif 475 476 /* 477 * stuff other tests for known memory-mapped devices (PCI?) 478 * here 479 */ 480 481 return 1; 482} 483