vm_machdep.c revision 118156
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 * @(#)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 118156 2003-07-29 12:44:16Z davidxu $"); 45 46#include "opt_isa.h" 47#include "opt_kstack_pages.h" 48 49#include <sys/param.h> 50#include <sys/systm.h> 51#include <sys/malloc.h> 52#include <sys/proc.h> 53#include <sys/kse.h> 54#include <sys/bio.h> 55#include <sys/buf.h> 56#include <sys/vnode.h> 57#include <sys/vmmeter.h> 58#include <sys/kernel.h> 59#include <sys/ktr.h> 60#include <sys/mutex.h> 61#include <sys/sysctl.h> 62#include <sys/unistd.h> 63 64#include <machine/cpu.h> 65#include <machine/md_var.h> 66#include <machine/pcb.h> 67 68#include <vm/vm.h> 69#include <vm/vm_param.h> 70#include <sys/lock.h> 71#include <vm/vm_kern.h> 72#include <vm/vm_page.h> 73#include <vm/vm_map.h> 74#include <vm/vm_extern.h> 75 76#include <sys/user.h> 77 78#include <amd64/isa/isa.h> 79 80static void cpu_reset_real(void); 81 82/* 83 * Finish a fork operation, with process p2 nearly set up. 84 * Copy and update the pcb, set up the stack so that the child 85 * ready to run and return to user mode. 86 */ 87void 88cpu_fork(td1, p2, td2, flags) 89 register struct thread *td1; 90 register struct proc *p2; 91 struct thread *td2; 92 int flags; 93{ 94 register struct proc *p1; 95 struct pcb *pcb2; 96 struct mdproc *mdp2; 97 register_t savecrit; 98 99 p1 = td1->td_proc; 100 if ((flags & RFPROC) == 0) 101 return; 102 103 /* Ensure that p1's pcb is up to date. */ 104 savecrit = intr_disable(); 105 if (PCPU_GET(fpcurthread) == td1) 106 npxsave(&td1->td_pcb->pcb_save); 107 intr_restore(savecrit); 108 109 /* Point the pcb to the top of the stack */ 110 pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 111 td2->td_pcb = pcb2; 112 113 /* Copy p1's pcb */ 114 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 115 116 /* Point mdproc and then copy over td1's contents */ 117 mdp2 = &p2->p_md; 118 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 119 120 /* 121 * Create a new fresh stack for the new process. 122 * Copy the trap frame for the return to user mode as if from a 123 * syscall. This copies most of the user mode register values. 124 */ 125 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 126 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 127 128 td2->td_frame->tf_rax = 0; /* Child returns zero */ 129 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 130 td2->td_frame->tf_rdx = 1; 131 132 /* 133 * Set registers for trampoline to user mode. Leave space for the 134 * return address on stack. These are the kernel mode register values. 135 */ 136 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 137 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 138 pcb2->pcb_rbp = 0; 139 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 140 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 141 pcb2->pcb_rip = (register_t)fork_trampoline; 142 pcb2->pcb_rflags = td2->td_frame->tf_rflags & ~PSL_I; /* ints disabled */ 143 /*- 144 * pcb2->pcb_savefpu: cloned above. 145 * pcb2->pcb_flags: cloned above. 146 * pcb2->pcb_onfault: cloned above (always NULL here?). 147 * pcb2->pcb_[fg]sbase: cloned above 148 */ 149 150 /* 151 * Now, cpu_switch() can schedule the new process. 152 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 153 * containing the return address when exiting cpu_switch. 154 * This will normally be to fork_trampoline(), which will have 155 * %ebx loaded with the new proc's pointer. fork_trampoline() 156 * will set up a stack to call fork_return(p, frame); to complete 157 * the return to user-mode. 158 */ 159} 160 161/* 162 * Intercept the return address from a freshly forked process that has NOT 163 * been scheduled yet. 164 * 165 * This is needed to make kernel threads stay in kernel mode. 166 */ 167void 168cpu_set_fork_handler(td, func, arg) 169 struct thread *td; 170 void (*func)(void *); 171 void *arg; 172{ 173 /* 174 * Note that the trap frame follows the args, so the function 175 * is really called like this: func(arg, frame); 176 */ 177 td->td_pcb->pcb_r12 = (long) func; /* function */ 178 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 179} 180 181void 182cpu_exit(struct thread *td) 183{ 184 struct mdproc *mdp; 185 186 mdp = &td->td_proc->p_md; 187} 188 189void 190cpu_thread_exit(struct thread *td) 191{ 192 193 npxexit(td); 194} 195 196void 197cpu_thread_clean(struct thread *td) 198{ 199} 200 201void 202cpu_sched_exit(td) 203 register struct thread *td; 204{ 205} 206 207void 208cpu_thread_setup(struct thread *td) 209{ 210 211 td->td_pcb = 212 (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 213 td->td_frame = (struct trapframe *)td->td_pcb - 1; 214} 215 216/* 217 * Initialize machine state (pcb and trap frame) for a new thread about to 218 * upcall. Pu t enough state in the new thread's PCB to get it to go back 219 * userret(), where we can intercept it again to set the return (upcall) 220 * Address and stack, along with those from upcals that are from other sources 221 * such as those generated in thread_userret() itself. 222 */ 223void 224cpu_set_upcall(struct thread *td, struct thread *td0) 225{ 226 struct pcb *pcb2; 227 228 /* Point the pcb to the top of the stack. */ 229 pcb2 = td->td_pcb; 230 231 /* 232 * Copy the upcall pcb. This loads kernel regs. 233 * Those not loaded individually below get their default 234 * values here. 235 * 236 * XXXKSE It might be a good idea to simply skip this as 237 * the values of the other registers may be unimportant. 238 * This would remove any requirement for knowing the KSE 239 * at this time (see the matching comment below for 240 * more analysis) (need a good safe default). 241 */ 242 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 243 244 /* 245 * Create a new fresh stack for the new thread. 246 * Don't forget to set this stack value into whatever supplies 247 * the address for the fault handlers. 248 * The contexts are filled in at the time we actually DO the 249 * upcall as only then do we know which KSE we got. 250 */ 251 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 252 253 /* 254 * Set registers for trampoline to user mode. Leave space for the 255 * return address on stack. These are the kernel mode register values. 256 */ 257 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 258 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 259 pcb2->pcb_rbp = 0; 260 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 261 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 262 pcb2->pcb_rip = (register_t)fork_trampoline; 263 pcb2->pcb_rflags = PSL_KERNEL; /* ints disabled */ 264 /* 265 * If we didn't copy the pcb, we'd need to do the following registers: 266 * pcb2->pcb_savefpu: cloned above. 267 * pcb2->pcb_rflags: cloned above. 268 * pcb2->pcb_onfault: cloned above (always NULL here?). 269 * pcb2->pcb_[fg]sbase: cloned above 270 */ 271} 272 273/* 274 * Set that machine state for performing an upcall that has to 275 * be done in thread_userret() so that those upcalls generated 276 * in thread_userret() itself can be done as well. 277 */ 278void 279cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 280{ 281 282 /* 283 * Do any extra cleaning that needs to be done. 284 * The thread may have optional components 285 * that are not present in a fresh thread. 286 * This may be a recycled thread so make it look 287 * as though it's newly allocated. 288 */ 289 cpu_thread_clean(td); 290 291 /* 292 * Set the trap frame to point at the beginning of the uts 293 * function. 294 */ 295 td->td_frame->tf_rsp = 296 ((register_t)ku->ku_stack.ss_sp + ku->ku_stack.ss_size) & ~0x0f; 297 td->td_frame->tf_rsp -= 8; 298 td->td_frame->tf_rip = (register_t)ku->ku_func; 299 300 /* 301 * Pass the address of the mailbox for this kse to the uts 302 * function as a parameter on the stack. 303 */ 304 td->td_frame->tf_rdi = (register_t)ku->ku_mailbox; 305} 306 307 308/* 309 * Force reset the processor by invalidating the entire address space! 310 */ 311 312void 313cpu_reset() 314{ 315 cpu_reset_real(); 316} 317 318static void 319cpu_reset_real() 320{ 321 322 /* 323 * Attempt to do a CPU reset via the keyboard controller, 324 * do not turn of the GateA20, as any machine that fails 325 * to do the reset here would then end up in no man's land. 326 */ 327 328 outb(IO_KBD + 4, 0xFE); 329 DELAY(500000); /* wait 0.5 sec to see if that did it */ 330 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 331 DELAY(1000000); /* wait 1 sec for printf to complete */ 332 /* force a shutdown by unmapping entire address space ! */ 333 bzero((caddr_t)PML4map, PAGE_SIZE); 334 335 /* "good night, sweet prince .... <THUNK!>" */ 336 invltlb(); 337 /* NOTREACHED */ 338 while(1); 339} 340 341/* 342 * Software interrupt handler for queued VM system processing. 343 */ 344void 345swi_vm(void *dummy) 346{ 347 if (busdma_swi_pending != 0) 348 busdma_swi(); 349} 350 351/* 352 * Tell whether this address is in some physical memory region. 353 * Currently used by the kernel coredump code in order to avoid 354 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 355 * or other unpredictable behaviour. 356 */ 357 358int 359is_physical_memory(addr) 360 vm_offset_t addr; 361{ 362 363#ifdef DEV_ISA 364 /* The ISA ``memory hole''. */ 365 if (addr >= 0xa0000 && addr < 0x100000) 366 return 0; 367#endif 368 369 /* 370 * stuff other tests for known memory-mapped devices (PCI?) 371 * here 372 */ 373 374 return 1; 375} 376