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 * Copyright (c) 2008 The DragonFly Project. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the Systems Programming Group of the University of Utah Computer 10 * Science Department, and William Jolitz. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by the University of 23 * California, Berkeley and its contributors. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 * 40 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 41 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 42 * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $ 43 */ 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/malloc.h> 48 #include <sys/proc.h> 49 #include <sys/buf.h> 50 #include <sys/interrupt.h> 51 #include <sys/vnode.h> 52 #include <sys/vmmeter.h> 53 #include <sys/kernel.h> 54 #include <sys/sysctl.h> 55 #include <sys/unistd.h> 56 57 #include <machine/clock.h> 58 #include <machine/cpu.h> 59 #include <machine/md_var.h> 60 #include <machine/smp.h> 61 #include <machine/pcb.h> 62 #include <machine/pcb_ext.h> 63 #include <machine/segments.h> 64 #include <machine/globaldata.h> /* npxthread */ 65 66 #include <vm/vm.h> 67 #include <vm/vm_param.h> 68 #include <sys/lock.h> 69 #include <vm/vm_kern.h> 70 #include <vm/vm_page.h> 71 #include <vm/vm_map.h> 72 #include <vm/vm_extern.h> 73 74 #include <sys/thread2.h> 75 #include <sys/mplock2.h> 76 77 #include <bus/isa/isa.h> 78 79 static void cpu_reset_real (void); 80 /* 81 * Finish a fork operation, with lwp lp2 nearly set up. 82 * Copy and update the pcb, set up the stack so that the child 83 * ready to run and return to user mode. 84 */ 85 void 86 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags) 87 { 88 struct pcb *pcb2; 89 90 if ((flags & RFPROC) == 0) { 91 if ((flags & RFMEM) == 0) { 92 /* unshare user LDT */ 93 struct pcb *pcb1 = lp1->lwp_thread->td_pcb; 94 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 95 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 96 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 97 user_ldt_free(pcb1); 98 pcb1->pcb_ldt = pcb_ldt; 99 set_user_ldt(pcb1); 100 } 101 } 102 return; 103 } 104 105 /* Ensure that lp1's pcb is up to date. */ 106 if (mdcpu->gd_npxthread == lp1->lwp_thread) 107 npxsave(lp1->lwp_thread->td_savefpu); 108 109 /* 110 * Copy lp1's PCB. This really only applies to the 111 * debug registers and FP state, but its faster to just copy the 112 * whole thing. Because we only save the PCB at switchout time, 113 * the register state may not be current. 114 */ 115 pcb2 = lp2->lwp_thread->td_pcb; 116 *pcb2 = *lp1->lwp_thread->td_pcb; 117 118 /* 119 * Create a new fresh stack for the new process. 120 * Copy the trap frame for the return to user mode as if from a 121 * syscall. This copies the user mode register values. 122 * 123 * pcb_rsp must allocate an additional call-return pointer below 124 * the trap frame which will be restored by cpu_heavy_restore from 125 * PCB_RIP, and the thread's td_sp pointer must allocate an 126 * additonal two quadwords below the pcb_rsp call-return pointer to 127 * hold the LWKT restore function pointer and rflags. 128 * 129 * The LWKT restore function pointer must be set to cpu_heavy_restore, 130 * which is our standard heavy-weight process switch-in function. 131 * YYY eventually we should shortcut fork_return and fork_trampoline 132 * to use the LWKT restore function directly so we can get rid of 133 * all the extra crap we are setting up. 134 */ 135 lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1; 136 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs)); 137 138 /* 139 * Set registers for trampoline to user mode. Leave space for the 140 * return address on stack. These are the kernel mode register values. 141 */ 142 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pml4); 143 pcb2->pcb_cr3 |= PG_RW | PG_U | PG_V; 144 pcb2->pcb_rbx = (unsigned long)fork_return; /* fork_trampoline argument */ 145 pcb2->pcb_rbp = 0; 146 pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *); 147 pcb2->pcb_r12 = (unsigned long)lp2; /* fork_trampoline argument */ 148 pcb2->pcb_r13 = 0; 149 pcb2->pcb_r14 = 0; 150 pcb2->pcb_r15 = 0; 151 pcb2->pcb_rip = (unsigned long)fork_trampoline; 152 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *)); 153 *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER; 154 lp2->lwp_thread->td_sp -= sizeof(void *); 155 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore; 156 157 /* 158 * pcb2->pcb_ldt: duplicated below, if necessary. 159 * pcb2->pcb_savefpu: cloned above. 160 * pcb2->pcb_flags: cloned above (always 0 here?). 161 * pcb2->pcb_onfault: cloned above (always NULL here?). 162 */ 163 164 /* 165 * XXX don't copy the i/o pages. this should probably be fixed. 166 */ 167 pcb2->pcb_ext = 0; 168 169 /* Copy the LDT, if necessary. */ 170 if (pcb2->pcb_ldt != 0) { 171 if (flags & RFMEM) { 172 pcb2->pcb_ldt->ldt_refcnt++; 173 } else { 174 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 175 pcb2->pcb_ldt->ldt_len); 176 } 177 } 178 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls, 179 sizeof(lp2->lwp_thread->td_tls)); 180 /* 181 * Now, cpu_switch() can schedule the new lwp. 182 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 183 * containing the return address when exiting cpu_switch. 184 * This will normally be to fork_trampoline(), which will have 185 * %rbx loaded with the new lwp's pointer. fork_trampoline() 186 * will set up a stack to call fork_return(lp, frame); to complete 187 * the return to user-mode. 188 */ 189 } 190 191 /* 192 * Prepare new lwp to return to the address specified in params. 193 */ 194 int 195 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params) 196 { 197 struct trapframe *regs = lp->lwp_md.md_regs; 198 void *bad_return = NULL; 199 int error; 200 201 regs->tf_rip = (long)params->func; 202 regs->tf_rsp = (long)params->stack; 203 /* Set up argument for function call */ 204 regs->tf_rdi = (long)params->arg; /* JG Can this be in userspace addresses? */ 205 /* 206 * Set up fake return address. As the lwp function may never return, 207 * we simply copy out a NULL pointer and force the lwp to receive 208 * a SIGSEGV if it returns anyways. 209 */ 210 regs->tf_rsp -= sizeof(void *); 211 error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return)); 212 if (error) 213 return (error); 214 215 cpu_set_fork_handler(lp, 216 (void (*)(void *, struct trapframe *))generic_lwp_return, lp); 217 return (0); 218 } 219 220 /* 221 * Intercept the return address from a freshly forked process that has NOT 222 * been scheduled yet. 223 * 224 * This is needed to make kernel threads stay in kernel mode. 225 */ 226 void 227 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *), 228 void *arg) 229 { 230 /* 231 * Note that the trap frame follows the args, so the function 232 * is really called like this: func(arg, frame); 233 */ 234 lp->lwp_thread->td_pcb->pcb_rbx = (long)func; /* function */ 235 lp->lwp_thread->td_pcb->pcb_r12 = (long)arg; /* first arg */ 236 } 237 238 void 239 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg) 240 { 241 td->td_pcb->pcb_rbx = (long)func; 242 td->td_pcb->pcb_r12 = (long)arg; 243 td->td_switch = cpu_lwkt_switch; 244 td->td_sp -= sizeof(void *); 245 *(void **)td->td_sp = rfunc; /* exit function on return */ 246 td->td_sp -= sizeof(void *); 247 *(void **)td->td_sp = cpu_kthread_restore; 248 } 249 250 void 251 cpu_lwp_exit(void) 252 { 253 struct thread *td = curthread; 254 struct pcb *pcb; 255 256 npxexit(); 257 pcb = td->td_pcb; 258 259 /* Some i386 functionality was dropped */ 260 KKASSERT(pcb->pcb_ext == NULL); 261 262 /* 263 * disable all hardware breakpoints 264 */ 265 if (pcb->pcb_flags & PCB_DBREGS) { 266 reset_dbregs(); 267 pcb->pcb_flags &= ~PCB_DBREGS; 268 } 269 td->td_gd->gd_cnt.v_swtch++; 270 271 crit_enter_quick(td); 272 if (td->td_flags & TDF_TSLEEPQ) 273 tsleep_remove(td); 274 lwkt_deschedule_self(td); 275 lwkt_remove_tdallq(td); 276 cpu_thread_exit(); 277 } 278 279 /* 280 * Terminate the current thread. The caller must have already acquired 281 * the thread's rwlock and placed it on a reap list or otherwise notified 282 * a reaper of its existance. We set a special assembly switch function which 283 * releases td_rwlock after it has cleaned up the MMU state and switched 284 * out the stack. 285 * 286 * Must be caller from a critical section and with the thread descheduled. 287 */ 288 void 289 cpu_thread_exit(void) 290 { 291 curthread->td_switch = cpu_exit_switch; 292 curthread->td_flags |= TDF_EXITING; 293 lwkt_switch(); 294 panic("cpu_thread_exit: lwkt_switch() unexpectedly returned"); 295 } 296 297 void 298 cpu_reset(void) 299 { 300 cpu_reset_real(); 301 } 302 303 static void 304 cpu_reset_real(void) 305 { 306 /* 307 * Attempt to do a CPU reset via the keyboard controller, 308 * do not turn of the GateA20, as any machine that fails 309 * to do the reset here would then end up in no man's land. 310 */ 311 312 #if !defined(BROKEN_KEYBOARD_RESET) 313 outb(IO_KBD + 4, 0xFE); 314 DELAY(500000); /* wait 0.5 sec to see if that did it */ 315 kprintf("Keyboard reset did not work, attempting CPU shutdown\n"); 316 DELAY(1000000); /* wait 1 sec for kprintf to complete */ 317 #endif 318 #if JG 319 /* force a shutdown by unmapping entire address space ! */ 320 bzero((caddr_t) PTD, PAGE_SIZE); 321 #endif 322 323 /* "good night, sweet prince .... <THUNK!>" */ 324 cpu_invltlb(); 325 /* NOTREACHED */ 326 while(1); 327 } 328 329 /* 330 * Convert kernel VA to physical address 331 */ 332 vm_paddr_t 333 kvtop(void *addr) 334 { 335 vm_paddr_t pa; 336 337 pa = pmap_kextract((vm_offset_t)addr); 338 if (pa == 0) 339 panic("kvtop: zero page frame"); 340 return (pa); 341 } 342 343 static void 344 swi_vm(void *arg, void *frame) 345 { 346 if (busdma_swi_pending != 0) 347 busdma_swi(); 348 } 349 350 static void 351 swi_vm_setup(void *arg) 352 { 353 register_swi(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0); 354 } 355 356 SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL); 357 358 359 /* 360 * Tell whether this address is in some physical memory region. 361 * Currently used by the kernel coredump code in order to avoid 362 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 363 * or other unpredictable behaviour. 364 */ 365 366 int 367 is_physical_memory(vm_offset_t addr) 368 { 369 #if NISA > 0 370 /* The ISA ``memory hole''. */ 371 if (addr >= 0xa0000 && addr < 0x100000) 372 return 0; 373 #endif 374 /* 375 * stuff other tests for known memory-mapped devices (PCI?) 376 * here 377 */ 378 379 return 1; 380 } 381 382 /* 383 * platform-specific vmspace initialization (nothing for x86_64) 384 */ 385 void 386 cpu_vmspace_alloc(struct vmspace *vm __unused) 387 { 388 } 389 390 void 391 cpu_vmspace_free(struct vmspace *vm __unused) 392 { 393 } 394 395 int 396 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot) 397 { 398 vm_offset_t addr; 399 400 if (saddr < KvaStart) 401 return EFAULT; 402 if (eaddr >= KvaEnd) 403 return EFAULT; 404 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) { 405 if (pmap_extract(&kernel_pmap, addr) == 0) 406 return EFAULT; 407 } 408 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot)) 409 return EFAULT; 410 return 0; 411 } 412 413