vm_machdep.c revision 140554
1274016Sngie/*- 2274016Sngie * Copyright (c) 1982, 1986 The Regents of the University of California. 3274016Sngie * Copyright (c) 1989, 1990 William Jolitz 4274016Sngie * Copyright (c) 1994 John Dyson 5274016Sngie * All rights reserved. 6274016Sngie * 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 140554 2005-01-21 05:57:45Z peter $"); 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 /* 167 * Now, cpu_switch() can schedule the new process. 168 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 169 * containing the return address when exiting cpu_switch. 170 * This will normally be to fork_trampoline(), which will have 171 * %ebx loaded with the new proc's pointer. fork_trampoline() 172 * will set up a stack to call fork_return(p, frame); to complete 173 * the return to user-mode. 174 */ 175} 176 177/* 178 * Intercept the return address from a freshly forked process that has NOT 179 * been scheduled yet. 180 * 181 * This is needed to make kernel threads stay in kernel mode. 182 */ 183void 184cpu_set_fork_handler(td, func, arg) 185 struct thread *td; 186 void (*func)(void *); 187 void *arg; 188{ 189 /* 190 * Note that the trap frame follows the args, so the function 191 * is really called like this: func(arg, frame); 192 */ 193 td->td_pcb->pcb_r12 = (long) func; /* function */ 194 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 195} 196 197void 198cpu_exit(struct thread *td) 199{ 200} 201 202void 203cpu_thread_exit(struct thread *td) 204{ 205 struct pcb *pcb = td->td_pcb; 206 207 if (td == PCPU_GET(fpcurthread)) 208 fpudrop(); 209 if (pcb->pcb_flags & PCB_DBREGS) { 210 /* disable all hardware breakpoints */ 211 reset_dbregs(); 212 pcb->pcb_flags &= ~PCB_DBREGS; 213 } 214} 215 216void 217cpu_thread_clean(struct thread *td) 218{ 219} 220 221void 222cpu_thread_swapin(struct thread *td) 223{ 224} 225 226void 227cpu_thread_swapout(struct thread *td) 228{ 229} 230 231void 232cpu_thread_setup(struct thread *td) 233{ 234 235 td->td_pcb = (struct pcb *)(td->td_kstack + 236 td->td_kstack_pages * PAGE_SIZE) - 1; 237 td->td_frame = (struct trapframe *)td->td_pcb - 1; 238} 239 240/* 241 * Initialize machine state (pcb and trap frame) for a new thread about to 242 * upcall. Put enough state in the new thread's PCB to get it to go back 243 * userret(), where we can intercept it again to set the return (upcall) 244 * Address and stack, along with those from upcals that are from other sources 245 * such as those generated in thread_userret() itself. 246 */ 247void 248cpu_set_upcall(struct thread *td, struct thread *td0) 249{ 250 struct pcb *pcb2; 251 252 /* Point the pcb to the top of the stack. */ 253 pcb2 = td->td_pcb; 254 255 /* 256 * Copy the upcall pcb. This loads kernel regs. 257 * Those not loaded individually below get their default 258 * values here. 259 * 260 * XXXKSE It might be a good idea to simply skip this as 261 * the values of the other registers may be unimportant. 262 * This would remove any requirement for knowing the KSE 263 * at this time (see the matching comment below for 264 * more analysis) (need a good safe default). 265 */ 266 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 267 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 268 269 /* 270 * Create a new fresh stack for the new thread. 271 * Don't forget to set this stack value into whatever supplies 272 * the address for the fault handlers. 273 * The contexts are filled in at the time we actually DO the 274 * upcall as only then do we know which KSE we got. 275 */ 276 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 277 278 /* 279 * Set registers for trampoline to user mode. Leave space for the 280 * return address on stack. These are the kernel mode register values. 281 */ 282 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 283 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 284 pcb2->pcb_rbp = 0; 285 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 286 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 287 pcb2->pcb_rip = (register_t)fork_trampoline; 288 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 289 /* 290 * If we didn't copy the pcb, we'd need to do the following registers: 291 * pcb2->pcb_dr*: cloned above. 292 * pcb2->pcb_savefpu: cloned above. 293 * pcb2->pcb_rflags: cloned above. 294 * pcb2->pcb_onfault: cloned above (always NULL here?). 295 * pcb2->pcb_[fg]sbase: cloned above 296 */ 297} 298 299/* 300 * Set that machine state for performing an upcall that has to 301 * be done in thread_userret() so that those upcalls generated 302 * in thread_userret() itself can be done as well. 303 */ 304void 305cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 306{ 307 308 /* 309 * Do any extra cleaning that needs to be done. 310 * The thread may have optional components 311 * that are not present in a fresh thread. 312 * This may be a recycled thread so make it look 313 * as though it's newly allocated. 314 */ 315 cpu_thread_clean(td); 316 317 /* 318 * Set the trap frame to point at the beginning of the uts 319 * function. 320 */ 321 td->td_frame->tf_rbp = 0; 322 td->td_frame->tf_rsp = 323 ((register_t)ku->ku_stack.ss_sp + ku->ku_stack.ss_size) & ~0x0f; 324 td->td_frame->tf_rsp -= 8; 325 td->td_frame->tf_rbp = 0; 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#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 u_int cnt, map; 355 356 if (smp_active) { 357 map = PCPU_GET(other_cpus) & ~stopped_cpus; 358 if (map != 0) { 359 printf("cpu_reset: Stopping other CPUs\n"); 360 stop_cpus(map); 361 } 362 363 if (PCPU_GET(cpuid) != 0) { 364 cpu_reset_proxyid = PCPU_GET(cpuid); 365 cpustop_restartfunc = cpu_reset_proxy; 366 cpu_reset_proxy_active = 0; 367 printf("cpu_reset: Restarting BSP\n"); 368 started_cpus = (1<<0); /* Restart CPU #0 */ 369 370 cnt = 0; 371 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 372 cnt++; /* Wait for BSP to announce restart */ 373 if (cpu_reset_proxy_active == 0) 374 printf("cpu_reset: Failed to restart BSP\n"); 375 enable_intr(); 376 cpu_reset_proxy_active = 2; 377 378 while (1); 379 /* NOTREACHED */ 380 } 381 382 DELAY(1000000); 383 } 384#endif 385 cpu_reset_real(); 386 /* NOTREACHED */ 387} 388 389static void 390cpu_reset_real() 391{ 392 393 /* 394 * Attempt to do a CPU reset via the keyboard controller, 395 * do not turn off GateA20, as any machine that fails 396 * to do the reset here would then end up in no man's land. 397 */ 398 outb(IO_KBD + 4, 0xFE); 399 DELAY(500000); /* wait 0.5 sec to see if that did it */ 400 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 401 DELAY(1000000); /* wait 1 sec for printf to complete */ 402 403 /* Force a shutdown by unmapping entire address space. */ 404 bzero((caddr_t)PML4map, PAGE_SIZE); 405 406 /* "good night, sweet prince .... <THUNK!>" */ 407 invltlb(); 408 /* NOTREACHED */ 409 while(1); 410} 411 412/* 413 * Allocate an sf_buf for the given vm_page. On this machine, however, there 414 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 415 * returned. 416 */ 417struct sf_buf * 418sf_buf_alloc(struct vm_page *m, int pri) 419{ 420 421 return ((struct sf_buf *)m); 422} 423 424/* 425 * Free the sf_buf. In fact, do nothing because there are no resources 426 * associated with the sf_buf. 427 */ 428void 429sf_buf_free(struct sf_buf *sf) 430{ 431} 432 433/* 434 * Software interrupt handler for queued VM system processing. 435 */ 436void 437swi_vm(void *dummy) 438{ 439 if (busdma_swi_pending != 0) 440 busdma_swi(); 441} 442 443/* 444 * Tell whether this address is in some physical memory region. 445 * Currently used by the kernel coredump code in order to avoid 446 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 447 * or other unpredictable behaviour. 448 */ 449 450int 451is_physical_memory(vm_paddr_t addr) 452{ 453 454#ifdef DEV_ISA 455 /* The ISA ``memory hole''. */ 456 if (addr >= 0xa0000 && addr < 0x100000) 457 return 0; 458#endif 459 460 /* 461 * stuff other tests for known memory-mapped devices (PCI?) 462 * here 463 */ 464 465 return 1; 466} 467