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 #include <sys/lwp.h> 57 58 #include <machine/clock.h> 59 #include <machine/cpu.h> 60 #include <machine/md_var.h> 61 #include <machine/smp.h> 62 #include <machine/pcb.h> 63 #include <machine/pcb_ext.h> 64 #include <machine/segments.h> 65 #include <machine/globaldata.h> /* npxthread */ 66 #include <machine/vmm.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/thread2.h> 77 #include <sys/mplock2.h> 78 79 #include <bus/isa/isa.h> 80 81 static void cpu_reset_real (void); 82 /* 83 * Finish a fork operation, with lwp lp2 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 */ 87 void 88 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags) 89 { 90 struct pcb *pcb2; 91 92 if ((flags & RFPROC) == 0) { 93 if ((flags & RFMEM) == 0) { 94 /* unshare user LDT */ 95 struct pcb *pcb1 = lp1->lwp_thread->td_pcb; 96 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 97 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 98 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 99 user_ldt_free(pcb1); 100 pcb1->pcb_ldt = pcb_ldt; 101 set_user_ldt(pcb1); 102 } 103 } 104 return; 105 } 106 107 /* Ensure that lp1's pcb is up to date. */ 108 if (mdcpu->gd_npxthread == lp1->lwp_thread) 109 npxsave(lp1->lwp_thread->td_savefpu); 110 111 /* 112 * Copy lp1's PCB. This really only applies to the 113 * debug registers and FP state, but its faster to just copy the 114 * whole thing. Because we only save the PCB at switchout time, 115 * the register state may not be current. 116 */ 117 pcb2 = lp2->lwp_thread->td_pcb; 118 *pcb2 = *lp1->lwp_thread->td_pcb; 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 the user mode register values. 124 * 125 * pcb_rsp must allocate an additional call-return pointer below 126 * the trap frame which will be restored by cpu_heavy_restore from 127 * PCB_RIP, and the thread's td_sp pointer must allocate an 128 * additonal two quadwords below the pcb_rsp call-return pointer to 129 * hold the LWKT restore function pointer and rflags. 130 * 131 * The LWKT restore function pointer must be set to cpu_heavy_restore, 132 * which is our standard heavy-weight process switch-in function. 133 * YYY eventually we should shortcut fork_return and fork_trampoline 134 * to use the LWKT restore function directly so we can get rid of 135 * all the extra crap we are setting up. 136 */ 137 lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1; 138 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs)); 139 140 /* 141 * Set registers for trampoline to user mode. Leave space for the 142 * return address on stack. These are the kernel mode register values. 143 */ 144 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pml4); 145 pcb2->pcb_rbx = (unsigned long)fork_return; /* fork_trampoline argument */ 146 pcb2->pcb_rbp = 0; 147 pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *); 148 pcb2->pcb_r12 = (unsigned long)lp2; /* fork_trampoline argument */ 149 pcb2->pcb_r13 = 0; 150 pcb2->pcb_r14 = 0; 151 pcb2->pcb_r15 = 0; 152 pcb2->pcb_rip = (unsigned long)fork_trampoline; 153 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *)); 154 *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER; 155 lp2->lwp_thread->td_sp -= sizeof(void *); 156 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore; 157 158 /* 159 * pcb2->pcb_ldt: duplicated below, if necessary. 160 * pcb2->pcb_savefpu: cloned above. 161 * pcb2->pcb_flags: cloned above (always 0 here?). 162 * pcb2->pcb_onfault: cloned above (always NULL here). 163 * pcb2->pcb_onfault_sp:cloned above (dont care) 164 */ 165 166 /* 167 * XXX don't copy the i/o pages. this should probably be fixed. 168 */ 169 pcb2->pcb_ext = NULL; 170 171 /* Copy the LDT, if necessary. */ 172 if (pcb2->pcb_ldt != NULL) { 173 if (flags & RFMEM) { 174 pcb2->pcb_ldt->ldt_refcnt++; 175 } else { 176 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 177 pcb2->pcb_ldt->ldt_len); 178 } 179 } 180 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls, 181 sizeof(lp2->lwp_thread->td_tls)); 182 /* 183 * Now, cpu_switch() can schedule the new lwp. 184 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 185 * containing the return address when exiting cpu_switch. 186 * This will normally be to fork_trampoline(), which will have 187 * %rbx loaded with the new lwp's pointer. fork_trampoline() 188 * will set up a stack to call fork_return(lp, frame); to complete 189 * the return to user-mode. 190 */ 191 } 192 193 /* 194 * Prepare new lwp to return to the address specified in params. 195 */ 196 int 197 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params) 198 { 199 struct trapframe *regs = lp->lwp_md.md_regs; 200 void *bad_return = NULL; 201 int error; 202 203 regs->tf_rip = (long)params->lwp_func; 204 regs->tf_rsp = (long)params->lwp_stack; 205 /* Set up argument for function call */ 206 regs->tf_rdi = (long)params->lwp_arg; 207 208 /* 209 * Set up fake return address. As the lwp function may never return, 210 * we simply copy out a NULL pointer and force the lwp to receive 211 * a SIGSEGV if it returns anyways. 212 */ 213 regs->tf_rsp -= sizeof(void *); 214 error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return)); 215 if (error) 216 return (error); 217 218 if (lp->lwp_proc->p_vmm) { 219 lp->lwp_thread->td_pcb->pcb_cr3 = KPML4phys; 220 cpu_set_fork_handler(lp, 221 (void (*)(void *, struct trapframe *))vmm_lwp_return, lp); 222 } else { 223 cpu_set_fork_handler(lp, 224 (void (*)(void *, struct trapframe *))generic_lwp_return, lp); 225 } 226 return (0); 227 } 228 229 /* 230 * Intercept the return address from a freshly forked process that has NOT 231 * been scheduled yet. 232 * 233 * This is needed to make kernel threads stay in kernel mode. 234 */ 235 void 236 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *), 237 void *arg) 238 { 239 /* 240 * Note that the trap frame follows the args, so the function 241 * is really called like this: func(arg, frame); 242 */ 243 lp->lwp_thread->td_pcb->pcb_rbx = (long)func; /* function */ 244 lp->lwp_thread->td_pcb->pcb_r12 = (long)arg; /* first arg */ 245 } 246 247 void 248 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg) 249 { 250 td->td_pcb->pcb_rbx = (long)func; 251 td->td_pcb->pcb_r12 = (long)arg; 252 td->td_switch = cpu_lwkt_switch; 253 td->td_sp -= sizeof(void *); 254 *(void **)td->td_sp = rfunc; /* exit function on return */ 255 td->td_sp -= sizeof(void *); 256 *(void **)td->td_sp = cpu_kthread_restore; 257 } 258 259 void 260 cpu_lwp_exit(void) 261 { 262 struct thread *td = curthread; 263 struct pcb *pcb; 264 265 pcb = td->td_pcb; 266 267 /* Some i386 functionality was dropped */ 268 KKASSERT(pcb->pcb_ext == NULL); 269 270 /* 271 * disable all hardware breakpoints 272 */ 273 if (pcb->pcb_flags & PCB_DBREGS) { 274 reset_dbregs(); 275 pcb->pcb_flags &= ~PCB_DBREGS; 276 } 277 td->td_gd->gd_cnt.v_swtch++; 278 279 crit_enter_quick(td); 280 if (td->td_flags & TDF_TSLEEPQ) 281 tsleep_remove(td); 282 lwkt_deschedule_self(td); 283 lwkt_remove_tdallq(td); 284 cpu_thread_exit(); 285 } 286 287 /* 288 * Terminate the current thread. The caller must have already acquired 289 * the thread's rwlock and placed it on a reap list or otherwise notified 290 * a reaper of its existance. We set a special assembly switch function which 291 * releases td_rwlock after it has cleaned up the MMU state and switched 292 * out the stack. 293 * 294 * Must be caller from a critical section and with the thread descheduled. 295 */ 296 void 297 cpu_thread_exit(void) 298 { 299 npxexit(); 300 curthread->td_switch = cpu_exit_switch; 301 curthread->td_flags |= TDF_EXITING; 302 lwkt_switch(); 303 panic("cpu_thread_exit: lwkt_switch() unexpectedly returned"); 304 } 305 306 void 307 cpu_reset(void) 308 { 309 cpu_reset_real(); 310 } 311 312 static void 313 cpu_reset_real(void) 314 { 315 /* 316 * Attempt to do a CPU reset via the keyboard controller, 317 * do not turn off the GateA20, as any machine that fails 318 * to do the reset here would then end up in no man's land. 319 */ 320 321 #if !defined(BROKEN_KEYBOARD_RESET) 322 outb(IO_KBD + 4, 0xFE); 323 DELAY(500000); /* wait 0.5 sec to see if that did it */ 324 kprintf("Keyboard reset did not work, attempting CPU shutdown\n"); 325 DELAY(1000000); /* wait 1 sec for kprintf to complete */ 326 #endif 327 #if 0 /* JG */ 328 /* force a shutdown by unmapping entire address space ! */ 329 bzero((caddr_t) PTD, PAGE_SIZE); 330 #endif 331 332 /* "good night, sweet prince .... <THUNK!>" */ 333 cpu_invltlb(); 334 /* NOTREACHED */ 335 while(1); 336 } 337 338 /* 339 * Convert kernel VA to physical address 340 */ 341 vm_paddr_t 342 kvtop(void *addr) 343 { 344 vm_paddr_t pa; 345 346 pa = pmap_kextract((vm_offset_t)addr); 347 if (pa == 0) 348 panic("kvtop: zero page frame"); 349 return (pa); 350 } 351 352 static void 353 swi_vm(void *arg, void *frame) 354 { 355 if (busdma_swi_pending != 0) 356 busdma_swi(); 357 } 358 359 static void 360 swi_vm_setup(void *arg) 361 { 362 register_swi_mp(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0); 363 } 364 365 SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL); 366 367 /* 368 * platform-specific vmspace initialization (nothing for x86_64) 369 */ 370 void 371 cpu_vmspace_alloc(struct vmspace *vm __unused) 372 { 373 } 374 375 void 376 cpu_vmspace_free(struct vmspace *vm __unused) 377 { 378 } 379 380 int 381 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot) 382 { 383 vm_offset_t addr; 384 385 if (saddr < KvaStart) 386 return EFAULT; 387 if (eaddr >= KvaEnd) 388 return EFAULT; 389 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) { 390 if (pmap_kextract(addr) == 0) 391 return EFAULT; 392 } 393 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot)) 394 return EFAULT; 395 return 0; 396 } 397 398 #if 0 399 400 void _test_frame_enter(struct trapframe *frame); 401 void _test_frame_exit(struct trapframe *frame); 402 403 void 404 _test_frame_enter(struct trapframe *frame) 405 { 406 thread_t td = curthread; 407 408 if (ISPL(frame->tf_cs) == SEL_UPL) { 409 KKASSERT(td->td_lwp); 410 KASSERT(td->td_lwp->lwp_md.md_regs == frame, 411 ("_test_frame_exit: Frame mismatch %p %p", 412 td->td_lwp->lwp_md.md_regs, frame)); 413 td->td_lwp->lwp_saveusp = (void *)frame->tf_rsp; 414 td->td_lwp->lwp_saveupc = (void *)frame->tf_rip; 415 } 416 if ((char *)frame < td->td_kstack || 417 (char *)frame > td->td_kstack + td->td_kstack_size) { 418 panic("_test_frame_exit: frame not on kstack %p kstack=%p", 419 frame, td->td_kstack); 420 } 421 } 422 423 void 424 _test_frame_exit(struct trapframe *frame) 425 { 426 thread_t td = curthread; 427 428 if (ISPL(frame->tf_cs) == SEL_UPL) { 429 KKASSERT(td->td_lwp); 430 KASSERT(td->td_lwp->lwp_md.md_regs == frame, 431 ("_test_frame_exit: Frame mismatch %p %p", 432 td->td_lwp->lwp_md.md_regs, frame)); 433 if (td->td_lwp->lwp_saveusp != (void *)frame->tf_rsp) { 434 kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n", 435 td->td_comm, td->td_proc->p_pid, 436 td->td_lwp->lwp_saveusp, 437 (void *)frame->tf_rsp); 438 } 439 if (td->td_lwp->lwp_saveupc != (void *)frame->tf_rip) { 440 kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n", 441 td->td_comm, td->td_proc->p_pid, 442 td->td_lwp->lwp_saveupc, 443 (void *)frame->tf_rip); 444 } 445 446 /* 447 * adulterate the fields to catch entries that 448 * don't run through test_frame_enter 449 */ 450 td->td_lwp->lwp_saveusp = 451 (void *)~(intptr_t)td->td_lwp->lwp_saveusp; 452 td->td_lwp->lwp_saveupc = 453 (void *)~(intptr_t)td->td_lwp->lwp_saveupc; 454 } 455 if ((char *)frame < td->td_kstack || 456 (char *)frame > td->td_kstack + td->td_kstack_size) { 457 panic("_test_frame_exit: frame not on kstack %p kstack=%p", 458 frame, td->td_kstack); 459 } 460 } 461 462 #endif 463