vm_machdep.c revision 139345
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 139345 2004-12-27 06:42:25Z njl $"); 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/vnode.h> 66#include <sys/vmmeter.h> 67 68#include <machine/cpu.h> 69#include <machine/md_var.h> 70#include <machine/pcb.h> 71 72#include <vm/vm.h> 73#include <vm/vm_extern.h> 74#include <vm/vm_kern.h> 75#include <vm/vm_page.h> 76#include <vm/vm_map.h> 77#include <vm/vm_param.h> 78 79#include <amd64/isa/isa.h> 80 81static void cpu_reset_real(void); 82#ifdef SMP 83static void cpu_reset_proxy(void); 84static u_int cpu_reset_proxyid; 85static volatile u_int cpu_reset_proxy_active; 86#endif 87 88/* 89 * Finish a fork operation, with process p2 nearly set up. 90 * Copy and update the pcb, set up the stack so that the child 91 * ready to run and return to user mode. 92 */ 93void 94cpu_fork(td1, p2, td2, flags) 95 register struct thread *td1; 96 register struct proc *p2; 97 struct thread *td2; 98 int flags; 99{ 100 register struct proc *p1; 101 struct pcb *pcb2; 102 struct mdproc *mdp2; 103 104 p1 = td1->td_proc; 105 if ((flags & RFPROC) == 0) 106 return; 107 108 /* Ensure that p1's pcb is up to date. */ 109 fpuexit(td1); 110 111 /* Point the pcb to the top of the stack */ 112 pcb2 = (struct pcb *)(td2->td_kstack + 113 td2->td_kstack_pages * PAGE_SIZE) - 1; 114 td2->td_pcb = pcb2; 115 116 /* Copy p1's pcb */ 117 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 118 119 /* Point mdproc and then copy over td1's contents */ 120 mdp2 = &p2->p_md; 121 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 122 123 /* 124 * Create a new fresh stack for the new process. 125 * Copy the trap frame for the return to user mode as if from a 126 * syscall. This copies most of the user mode register values. 127 */ 128 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 129 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 130 131 td2->td_frame->tf_rax = 0; /* Child returns zero */ 132 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 133 td2->td_frame->tf_rdx = 1; 134 135 /* 136 * Set registers for trampoline to user mode. Leave space for the 137 * return address on stack. These are the kernel mode register values. 138 */ 139 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 140 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 141 pcb2->pcb_rbp = 0; 142 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 143 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 144 pcb2->pcb_rip = (register_t)fork_trampoline; 145 pcb2->pcb_rflags = td2->td_frame->tf_rflags & ~PSL_I; /* ints disabled */ 146 /*- 147 * pcb2->pcb_dr*: cloned above. 148 * pcb2->pcb_savefpu: cloned above. 149 * pcb2->pcb_flags: cloned above. 150 * pcb2->pcb_onfault: cloned above (always NULL here?). 151 * pcb2->pcb_[fg]sbase: cloned above 152 */ 153 154 /* 155 * Now, cpu_switch() can schedule the new process. 156 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 157 * containing the return address when exiting cpu_switch. 158 * This will normally be to fork_trampoline(), which will have 159 * %ebx loaded with the new proc's pointer. fork_trampoline() 160 * will set up a stack to call fork_return(p, frame); to complete 161 * the return to user-mode. 162 */ 163} 164 165/* 166 * Intercept the return address from a freshly forked process that has NOT 167 * been scheduled yet. 168 * 169 * This is needed to make kernel threads stay in kernel mode. 170 */ 171void 172cpu_set_fork_handler(td, func, arg) 173 struct thread *td; 174 void (*func)(void *); 175 void *arg; 176{ 177 /* 178 * Note that the trap frame follows the args, so the function 179 * is really called like this: func(arg, frame); 180 */ 181 td->td_pcb->pcb_r12 = (long) func; /* function */ 182 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 183} 184 185void 186cpu_exit(struct thread *td) 187{ 188 struct pcb *pcb = td->td_pcb; 189 190 if (pcb->pcb_flags & PCB_DBREGS) { 191 /* disable all hardware breakpoints */ 192 reset_dbregs(); 193 pcb->pcb_flags &= ~PCB_DBREGS; 194 } 195} 196 197void 198cpu_thread_exit(struct thread *td) 199{ 200 struct pcb *pcb = td->td_pcb; 201 202 if (td == PCPU_GET(fpcurthread)) 203 fpudrop(); 204 if (pcb->pcb_flags & PCB_DBREGS) { 205 /* disable all hardware breakpoints */ 206 reset_dbregs(); 207 pcb->pcb_flags &= ~PCB_DBREGS; 208 } 209} 210 211void 212cpu_thread_clean(struct thread *td) 213{ 214} 215 216void 217cpu_thread_swapin(struct thread *td) 218{ 219} 220 221void 222cpu_thread_swapout(struct thread *td) 223{ 224} 225 226void 227cpu_thread_setup(struct thread *td) 228{ 229 230 td->td_pcb = (struct pcb *)(td->td_kstack + 231 td->td_kstack_pages * PAGE_SIZE) - 1; 232 td->td_frame = (struct trapframe *)td->td_pcb - 1; 233} 234 235/* 236 * Initialize machine state (pcb and trap frame) for a new thread about to 237 * upcall. Pu t enough state in the new thread's PCB to get it to go back 238 * userret(), where we can intercept it again to set the return (upcall) 239 * Address and stack, along with those from upcals that are from other sources 240 * such as those generated in thread_userret() itself. 241 */ 242void 243cpu_set_upcall(struct thread *td, struct thread *td0) 244{ 245 struct pcb *pcb2; 246 247 /* Point the pcb to the top of the stack. */ 248 pcb2 = td->td_pcb; 249 250 /* 251 * Copy the upcall pcb. This loads kernel regs. 252 * Those not loaded individually below get their default 253 * values here. 254 * 255 * XXXKSE It might be a good idea to simply skip this as 256 * the values of the other registers may be unimportant. 257 * This would remove any requirement for knowing the KSE 258 * at this time (see the matching comment below for 259 * more analysis) (need a good safe default). 260 */ 261 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 262 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 263 264 /* 265 * Create a new fresh stack for the new thread. 266 * Don't forget to set this stack value into whatever supplies 267 * the address for the fault handlers. 268 * The contexts are filled in at the time we actually DO the 269 * upcall as only then do we know which KSE we got. 270 */ 271 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 272 273 /* 274 * Set registers for trampoline to user mode. Leave space for the 275 * return address on stack. These are the kernel mode register values. 276 */ 277 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 278 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 279 pcb2->pcb_rbp = 0; 280 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 281 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 282 pcb2->pcb_rip = (register_t)fork_trampoline; 283 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 284 /* 285 * If we didn't copy the pcb, we'd need to do the following registers: 286 * pcb2->pcb_dr*: cloned above. 287 * pcb2->pcb_savefpu: cloned above. 288 * pcb2->pcb_rflags: cloned above. 289 * pcb2->pcb_onfault: cloned above (always NULL here?). 290 * pcb2->pcb_[fg]sbase: cloned above 291 */ 292} 293 294/* 295 * Set that machine state for performing an upcall that has to 296 * be done in thread_userret() so that those upcalls generated 297 * in thread_userret() itself can be done as well. 298 */ 299void 300cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 301{ 302 303 /* 304 * Do any extra cleaning that needs to be done. 305 * The thread may have optional components 306 * that are not present in a fresh thread. 307 * This may be a recycled thread so make it look 308 * as though it's newly allocated. 309 */ 310 cpu_thread_clean(td); 311 312 /* 313 * Set the trap frame to point at the beginning of the uts 314 * function. 315 */ 316 td->td_frame->tf_rbp = 0; 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#ifdef SMP 331static void 332cpu_reset_proxy() 333{ 334 335 cpu_reset_proxy_active = 1; 336 while (cpu_reset_proxy_active == 1) 337 ; /* Wait for other cpu to see that we've started */ 338 stop_cpus((1<<cpu_reset_proxyid)); 339 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 340 DELAY(1000000); 341 cpu_reset_real(); 342} 343#endif 344 345void 346cpu_reset() 347{ 348#ifdef SMP 349 u_int cnt, map; 350 351 if (smp_active) { 352 map = PCPU_GET(other_cpus) & ~stopped_cpus; 353 if (map != 0) { 354 printf("cpu_reset: Stopping other CPUs\n"); 355 stop_cpus(map); 356 } 357 358 if (PCPU_GET(cpuid) != 0) { 359 cpu_reset_proxyid = PCPU_GET(cpuid); 360 cpustop_restartfunc = cpu_reset_proxy; 361 cpu_reset_proxy_active = 0; 362 printf("cpu_reset: Restarting BSP\n"); 363 started_cpus = (1<<0); /* Restart CPU #0 */ 364 365 cnt = 0; 366 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 367 cnt++; /* Wait for BSP to announce restart */ 368 if (cpu_reset_proxy_active == 0) 369 printf("cpu_reset: Failed to restart BSP\n"); 370 enable_intr(); 371 cpu_reset_proxy_active = 2; 372 373 while (1); 374 /* NOTREACHED */ 375 } 376 377 DELAY(1000000); 378 } 379#endif 380 cpu_reset_real(); 381 /* NOTREACHED */ 382} 383 384static void 385cpu_reset_real() 386{ 387 388 /* 389 * Attempt to do a CPU reset via the keyboard controller, 390 * do not turn off GateA20, as any machine that fails 391 * to do the reset here would then end up in no man's land. 392 */ 393 outb(IO_KBD + 4, 0xFE); 394 DELAY(500000); /* wait 0.5 sec to see if that did it */ 395 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 396 DELAY(1000000); /* wait 1 sec for printf to complete */ 397 398 /* Force a shutdown by unmapping entire address space. */ 399 bzero((caddr_t)PML4map, PAGE_SIZE); 400 401 /* "good night, sweet prince .... <THUNK!>" */ 402 invltlb(); 403 /* NOTREACHED */ 404 while(1); 405} 406 407/* 408 * Allocate an sf_buf for the given vm_page. On this machine, however, there 409 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 410 * returned. 411 */ 412struct sf_buf * 413sf_buf_alloc(struct vm_page *m, int pri) 414{ 415 416 return ((struct sf_buf *)m); 417} 418 419/* 420 * Free the sf_buf. In fact, do nothing because there are no resources 421 * associated with the sf_buf. 422 */ 423void 424sf_buf_free(struct sf_buf *sf) 425{ 426} 427 428/* 429 * Software interrupt handler for queued VM system processing. 430 */ 431void 432swi_vm(void *dummy) 433{ 434 if (busdma_swi_pending != 0) 435 busdma_swi(); 436} 437 438/* 439 * Tell whether this address is in some physical memory region. 440 * Currently used by the kernel coredump code in order to avoid 441 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 442 * or other unpredictable behaviour. 443 */ 444 445int 446is_physical_memory(vm_paddr_t addr) 447{ 448 449#ifdef DEV_ISA 450 /* The ISA ``memory hole''. */ 451 if (addr >= 0xa0000 && addr < 0x100000) 452 return 0; 453#endif 454 455 /* 456 * stuff other tests for known memory-mapped devices (PCI?) 457 * here 458 */ 459 460 return 1; 461} 462