vm_machdep.c revision 133529
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 133529 2004-08-11 23:23:05Z 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/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_thread_setup(struct thread *td) 229{ 230 231 td->td_pcb = (struct pcb *)(td->td_kstack + 232 td->td_kstack_pages * PAGE_SIZE) - 1; 233 td->td_frame = (struct trapframe *)td->td_pcb - 1; 234} 235 236/* 237 * Initialize machine state (pcb and trap frame) for a new thread about to 238 * upcall. Pu t enough state in the new thread's PCB to get it to go back 239 * userret(), where we can intercept it again to set the return (upcall) 240 * Address and stack, along with those from upcals that are from other sources 241 * such as those generated in thread_userret() itself. 242 */ 243void 244cpu_set_upcall(struct thread *td, struct thread *td0) 245{ 246 struct pcb *pcb2; 247 248 /* Point the pcb to the top of the stack. */ 249 pcb2 = td->td_pcb; 250 251 /* 252 * Copy the upcall pcb. This loads kernel regs. 253 * Those not loaded individually below get their default 254 * values here. 255 * 256 * XXXKSE It might be a good idea to simply skip this as 257 * the values of the other registers may be unimportant. 258 * This would remove any requirement for knowing the KSE 259 * at this time (see the matching comment below for 260 * more analysis) (need a good safe default). 261 */ 262 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 263 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 264 265 /* 266 * Create a new fresh stack for the new thread. 267 * Don't forget to set this stack value into whatever supplies 268 * the address for the fault handlers. 269 * The contexts are filled in at the time we actually DO the 270 * upcall as only then do we know which KSE we got. 271 */ 272 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 273 274 /* 275 * Set registers for trampoline to user mode. Leave space for the 276 * return address on stack. These are the kernel mode register values. 277 */ 278 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 279 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 280 pcb2->pcb_rbp = 0; 281 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 282 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 283 pcb2->pcb_rip = (register_t)fork_trampoline; 284 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 285 /* 286 * If we didn't copy the pcb, we'd need to do the following registers: 287 * pcb2->pcb_dr*: cloned above. 288 * pcb2->pcb_savefpu: cloned above. 289 * pcb2->pcb_rflags: cloned above. 290 * pcb2->pcb_onfault: cloned above (always NULL here?). 291 * pcb2->pcb_[fg]sbase: cloned above 292 */ 293} 294 295/* 296 * Set that machine state for performing an upcall that has to 297 * be done in thread_userret() so that those upcalls generated 298 * in thread_userret() itself can be done as well. 299 */ 300void 301cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 302{ 303 304 /* 305 * Do any extra cleaning that needs to be done. 306 * The thread may have optional components 307 * that are not present in a fresh thread. 308 * This may be a recycled thread so make it look 309 * as though it's newly allocated. 310 */ 311 cpu_thread_clean(td); 312 313 /* 314 * Set the trap frame to point at the beginning of the uts 315 * function. 316 */ 317 td->td_frame->tf_rsp = 318 ((register_t)ku->ku_stack.ss_sp + ku->ku_stack.ss_size) & ~0x0f; 319 td->td_frame->tf_rsp -= 8; 320 td->td_frame->tf_rbp = 0; 321 td->td_frame->tf_rip = (register_t)ku->ku_func; 322 323 /* 324 * Pass the address of the mailbox for this kse to the uts 325 * function as a parameter on the stack. 326 */ 327 td->td_frame->tf_rdi = (register_t)ku->ku_mailbox; 328} 329 330 331/* 332 * Force reset the processor by invalidating the entire address space! 333 */ 334 335#ifdef SMP 336static void 337cpu_reset_proxy() 338{ 339 340 cpu_reset_proxy_active = 1; 341 while (cpu_reset_proxy_active == 1) 342 ; /* Wait for other cpu to see that we've started */ 343 stop_cpus((1<<cpu_reset_proxyid)); 344 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 345 DELAY(1000000); 346 cpu_reset_real(); 347} 348#endif 349 350void 351cpu_reset() 352{ 353#ifdef SMP 354 if (smp_active == 0) { 355 cpu_reset_real(); 356 /* NOTREACHED */ 357 } else { 358 359 u_int map; 360 int cnt; 361 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 362 363 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 364 365 if (map != 0) { 366 printf("cpu_reset: Stopping other CPUs\n"); 367 stop_cpus(map); /* Stop all other CPUs */ 368 } 369 370 if (PCPU_GET(cpuid) == 0) { 371 DELAY(1000000); 372 cpu_reset_real(); 373 /* NOTREACHED */ 374 } else { 375 /* We are not BSP (CPU #0) */ 376 377 cpu_reset_proxyid = PCPU_GET(cpuid); 378 cpustop_restartfunc = cpu_reset_proxy; 379 cpu_reset_proxy_active = 0; 380 printf("cpu_reset: Restarting BSP\n"); 381 started_cpus = (1<<0); /* Restart CPU #0 */ 382 383 cnt = 0; 384 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 385 cnt++; /* Wait for BSP to announce restart */ 386 if (cpu_reset_proxy_active == 0) 387 printf("cpu_reset: Failed to restart BSP\n"); 388 enable_intr(); 389 cpu_reset_proxy_active = 2; 390 391 while (1); 392 /* NOTREACHED */ 393 } 394 } 395#else 396 cpu_reset_real(); 397#endif 398} 399 400static void 401cpu_reset_real() 402{ 403 404 /* 405 * Attempt to do a CPU reset via the keyboard controller, 406 * do not turn of the GateA20, as any machine that fails 407 * to do the reset here would then end up in no man's land. 408 */ 409 410 outb(IO_KBD + 4, 0xFE); 411 DELAY(500000); /* wait 0.5 sec to see if that did it */ 412 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 413 DELAY(1000000); /* wait 1 sec for printf to complete */ 414 /* force a shutdown by unmapping entire address space ! */ 415 bzero((caddr_t)PML4map, PAGE_SIZE); 416 417 /* "good night, sweet prince .... <THUNK!>" */ 418 invltlb(); 419 /* NOTREACHED */ 420 while(1); 421} 422 423/* 424 * Allocate an sf_buf for the given vm_page. On this machine, however, there 425 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 426 * returned. 427 */ 428struct sf_buf * 429sf_buf_alloc(struct vm_page *m, int pri) 430{ 431 432 return ((struct sf_buf *)m); 433} 434 435/* 436 * Free the sf_buf. In fact, do nothing because there are no resources 437 * associated with the sf_buf. 438 */ 439void 440sf_buf_free(struct sf_buf *sf) 441{ 442} 443 444/* 445 * Software interrupt handler for queued VM system processing. 446 */ 447void 448swi_vm(void *dummy) 449{ 450 if (busdma_swi_pending != 0) 451 busdma_swi(); 452} 453 454/* 455 * Tell whether this address is in some physical memory region. 456 * Currently used by the kernel coredump code in order to avoid 457 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 458 * or other unpredictable behaviour. 459 */ 460 461int 462is_physical_memory(vm_paddr_t addr) 463{ 464 465#ifdef DEV_ISA 466 /* The ISA ``memory hole''. */ 467 if (addr >= 0xa0000 && addr < 0x100000) 468 return 0; 469#endif 470 471 /* 472 * stuff other tests for known memory-mapped devices (PCI?) 473 * here 474 */ 475 476 return 1; 477} 478