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 /* 95 * Unshare user LDT. > 1 test is MPSAFE. While 96 * it can potentially race a 2->1 transition, the 97 * worst that happens is that we do an unnecessary 98 * ldt replacement. 99 */ 100 struct pcb *pcb1 = lp1->lwp_thread->td_pcb; 101 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 102 103 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 104 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 105 user_ldt_free(pcb1); 106 pcb1->pcb_ldt = pcb_ldt; 107 set_user_ldt(pcb1); 108 } 109 } 110 return; 111 } 112 113 /* Ensure that lp1's pcb is up to date. */ 114 if (mdcpu->gd_npxthread == lp1->lwp_thread) 115 npxsave(lp1->lwp_thread->td_savefpu); 116 117 /* 118 * Copy lp1's PCB. This really only applies to the 119 * debug registers and FP state, but its faster to just copy the 120 * whole thing. Because we only save the PCB at switchout time, 121 * the register state may not be current. 122 */ 123 pcb2 = lp2->lwp_thread->td_pcb; 124 *pcb2 = *lp1->lwp_thread->td_pcb; 125 126 /* 127 * Create a new fresh stack for the new process. 128 * Copy the trap frame for the return to user mode as if from a 129 * syscall. This copies the user mode register values. 130 * 131 * pcb_rsp must allocate an additional call-return pointer below 132 * the trap frame which will be restored by cpu_heavy_restore from 133 * PCB_RIP, and the thread's td_sp pointer must allocate an 134 * additonal two quadwords below the pcb_rsp call-return pointer to 135 * hold the LWKT restore function pointer and rflags. 136 * 137 * The LWKT restore function pointer must be set to cpu_heavy_restore, 138 * which is our standard heavy-weight process switch-in function. 139 * YYY eventually we should shortcut fork_return and fork_trampoline 140 * to use the LWKT restore function directly so we can get rid of 141 * all the extra crap we are setting up. 142 */ 143 lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1; 144 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs)); 145 146 /* 147 * Set registers for trampoline to user mode. Leave space for the 148 * return address on stack. These are the kernel mode register values. 149 */ 150 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pml4); 151 pcb2->pcb_rbx = (unsigned long)fork_return; /* fork_trampoline argument */ 152 pcb2->pcb_rbp = 0; 153 pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *); 154 pcb2->pcb_r12 = (unsigned long)lp2; /* fork_trampoline argument */ 155 pcb2->pcb_r13 = 0; 156 pcb2->pcb_r14 = 0; 157 pcb2->pcb_r15 = 0; 158 pcb2->pcb_rip = (unsigned long)fork_trampoline; 159 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *)); 160 *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER; 161 lp2->lwp_thread->td_sp -= sizeof(void *); 162 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore; 163 164 /* 165 * pcb2->pcb_ldt: duplicated below, if necessary. 166 * pcb2->pcb_savefpu: cloned above. 167 * pcb2->pcb_flags: cloned above (always 0 here?). 168 * pcb2->pcb_onfault: cloned above (always NULL here). 169 * pcb2->pcb_onfault_sp:cloned above (dont care) 170 */ 171 172 /* 173 * XXX don't copy the i/o pages. this should probably be fixed. 174 */ 175 pcb2->pcb_ext = NULL; 176 177 /* Copy the LDT, if necessary. */ 178 if (pcb2->pcb_ldt != NULL) { 179 if (flags & RFMEM) { 180 atomic_add_int(&pcb2->pcb_ldt->ldt_refcnt, 1); 181 } else { 182 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 183 pcb2->pcb_ldt->ldt_len); 184 } 185 } 186 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls, 187 sizeof(lp2->lwp_thread->td_tls)); 188 /* 189 * Now, cpu_switch() can schedule the new lwp. 190 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 191 * containing the return address when exiting cpu_switch. 192 * This will normally be to fork_trampoline(), which will have 193 * %rbx loaded with the new lwp's pointer. fork_trampoline() 194 * will set up a stack to call fork_return(lp, frame); to complete 195 * the return to user-mode. 196 */ 197 } 198 199 /* 200 * Prepare new lwp to return to the address specified in params. 201 */ 202 int 203 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params) 204 { 205 struct trapframe *regs = lp->lwp_md.md_regs; 206 void *bad_return = NULL; 207 int error; 208 209 regs->tf_rip = (long)params->lwp_func; 210 regs->tf_rsp = (long)params->lwp_stack; 211 /* Set up argument for function call */ 212 regs->tf_rdi = (long)params->lwp_arg; 213 214 /* 215 * Set up fake return address. As the lwp function may never return, 216 * we simply copy out a NULL pointer and force the lwp to receive 217 * a SIGSEGV if it returns anyways. 218 */ 219 regs->tf_rsp -= sizeof(void *); 220 error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return)); 221 if (error) 222 return (error); 223 224 if (lp->lwp_proc->p_vmm) { 225 lp->lwp_thread->td_pcb->pcb_cr3 = KPML4phys; 226 cpu_set_fork_handler(lp, 227 (void (*)(void *, struct trapframe *))vmm_lwp_return, lp); 228 } else { 229 cpu_set_fork_handler(lp, 230 (void (*)(void *, struct trapframe *))generic_lwp_return, lp); 231 } 232 return (0); 233 } 234 235 /* 236 * Intercept the return address from a freshly forked process that has NOT 237 * been scheduled yet. 238 * 239 * This is needed to make kernel threads stay in kernel mode. 240 */ 241 void 242 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *), 243 void *arg) 244 { 245 /* 246 * Note that the trap frame follows the args, so the function 247 * is really called like this: func(arg, frame); 248 */ 249 lp->lwp_thread->td_pcb->pcb_rbx = (long)func; /* function */ 250 lp->lwp_thread->td_pcb->pcb_r12 = (long)arg; /* first arg */ 251 } 252 253 void 254 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg) 255 { 256 td->td_pcb->pcb_rbx = (long)func; 257 td->td_pcb->pcb_r12 = (long)arg; 258 td->td_switch = cpu_lwkt_switch; 259 td->td_sp -= sizeof(void *); 260 *(void **)td->td_sp = rfunc; /* exit function on return */ 261 td->td_sp -= sizeof(void *); 262 *(void **)td->td_sp = cpu_kthread_restore; 263 } 264 265 void 266 cpu_lwp_exit(void) 267 { 268 struct thread *td = curthread; 269 struct pcb *pcb; 270 271 pcb = td->td_pcb; 272 273 /* Some i386 functionality was dropped */ 274 KKASSERT(pcb->pcb_ext == NULL); 275 276 /* 277 * disable all hardware breakpoints 278 */ 279 if (pcb->pcb_flags & PCB_DBREGS) { 280 reset_dbregs(); 281 pcb->pcb_flags &= ~PCB_DBREGS; 282 } 283 td->td_gd->gd_cnt.v_swtch++; 284 285 crit_enter_quick(td); 286 if (td->td_flags & TDF_TSLEEPQ) 287 tsleep_remove(td); 288 lwkt_deschedule_self(td); 289 lwkt_remove_tdallq(td); 290 cpu_thread_exit(); 291 } 292 293 /* 294 * Terminate the current thread. The caller must have already acquired 295 * the thread's rwlock and placed it on a reap list or otherwise notified 296 * a reaper of its existance. We set a special assembly switch function which 297 * releases td_rwlock after it has cleaned up the MMU state and switched 298 * out the stack. 299 * 300 * Must be caller from a critical section and with the thread descheduled. 301 */ 302 void 303 cpu_thread_exit(void) 304 { 305 npxexit(); 306 curthread->td_switch = cpu_exit_switch; 307 curthread->td_flags |= TDF_EXITING; 308 lwkt_switch(); 309 panic("cpu_thread_exit: lwkt_switch() unexpectedly returned"); 310 } 311 312 void 313 cpu_reset(void) 314 { 315 cpu_reset_real(); 316 } 317 318 static void 319 cpu_reset_real(void) 320 { 321 /* 322 * Attempt to do a CPU reset via the keyboard controller, 323 * do not turn off the GateA20, as any machine that fails 324 * to do the reset here would then end up in no man's land. 325 */ 326 327 #if !defined(BROKEN_KEYBOARD_RESET) 328 outb(IO_KBD + 4, 0xFE); 329 DELAY(500000); /* wait 0.5 sec to see if that did it */ 330 kprintf("Keyboard reset did not work, attempting CPU shutdown\n"); 331 DELAY(1000000); /* wait 1 sec for kprintf to complete */ 332 #endif 333 #if 0 /* JG */ 334 /* force a shutdown by unmapping entire address space ! */ 335 bzero((caddr_t) PTD, PAGE_SIZE); 336 #endif 337 338 /* "good night, sweet prince .... <THUNK!>" */ 339 cpu_invltlb(); 340 /* NOTREACHED */ 341 while(1); 342 } 343 344 /* 345 * Convert kernel VA to physical address 346 */ 347 vm_paddr_t 348 kvtop(void *addr) 349 { 350 vm_paddr_t pa; 351 352 pa = pmap_kextract((vm_offset_t)addr); 353 if (pa == 0) 354 panic("kvtop: zero page frame"); 355 return (pa); 356 } 357 358 static void 359 swi_vm(void *arg, void *frame) 360 { 361 if (busdma_swi_pending != 0) 362 busdma_swi(); 363 } 364 365 static void 366 swi_vm_setup(void *arg) 367 { 368 register_swi_mp(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0); 369 } 370 371 SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL); 372 373 /* 374 * platform-specific vmspace initialization (nothing for x86_64) 375 */ 376 void 377 cpu_vmspace_alloc(struct vmspace *vm __unused) 378 { 379 } 380 381 void 382 cpu_vmspace_free(struct vmspace *vm __unused) 383 { 384 } 385 386 int 387 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot) 388 { 389 vm_offset_t addr; 390 391 if (saddr < KvaStart) 392 return EFAULT; 393 if (eaddr >= KvaEnd) 394 return EFAULT; 395 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) { 396 if (pmap_kextract(addr) == 0) 397 return EFAULT; 398 } 399 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot)) 400 return EFAULT; 401 return 0; 402 } 403 404 #if 0 405 406 void _test_frame_enter(struct trapframe *frame); 407 void _test_frame_exit(struct trapframe *frame); 408 409 void 410 _test_frame_enter(struct trapframe *frame) 411 { 412 thread_t td = curthread; 413 414 if (ISPL(frame->tf_cs) == SEL_UPL) { 415 KKASSERT(td->td_lwp); 416 KASSERT(td->td_lwp->lwp_md.md_regs == frame, 417 ("_test_frame_exit: Frame mismatch %p %p", 418 td->td_lwp->lwp_md.md_regs, frame)); 419 td->td_lwp->lwp_saveusp = (void *)frame->tf_rsp; 420 td->td_lwp->lwp_saveupc = (void *)frame->tf_rip; 421 } 422 if ((char *)frame < td->td_kstack || 423 (char *)frame > td->td_kstack + td->td_kstack_size) { 424 panic("_test_frame_exit: frame not on kstack %p kstack=%p", 425 frame, td->td_kstack); 426 } 427 } 428 429 void 430 _test_frame_exit(struct trapframe *frame) 431 { 432 thread_t td = curthread; 433 434 if (ISPL(frame->tf_cs) == SEL_UPL) { 435 KKASSERT(td->td_lwp); 436 KASSERT(td->td_lwp->lwp_md.md_regs == frame, 437 ("_test_frame_exit: Frame mismatch %p %p", 438 td->td_lwp->lwp_md.md_regs, frame)); 439 if (td->td_lwp->lwp_saveusp != (void *)frame->tf_rsp) { 440 kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n", 441 td->td_comm, td->td_proc->p_pid, 442 td->td_lwp->lwp_saveusp, 443 (void *)frame->tf_rsp); 444 } 445 if (td->td_lwp->lwp_saveupc != (void *)frame->tf_rip) { 446 kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n", 447 td->td_comm, td->td_proc->p_pid, 448 td->td_lwp->lwp_saveupc, 449 (void *)frame->tf_rip); 450 } 451 452 /* 453 * adulterate the fields to catch entries that 454 * don't run through test_frame_enter 455 */ 456 td->td_lwp->lwp_saveusp = 457 (void *)~(intptr_t)td->td_lwp->lwp_saveusp; 458 td->td_lwp->lwp_saveupc = 459 (void *)~(intptr_t)td->td_lwp->lwp_saveupc; 460 } 461 if ((char *)frame < td->td_kstack || 462 (char *)frame > td->td_kstack + td->td_kstack_size) { 463 panic("_test_frame_exit: frame not on kstack %p kstack=%p", 464 frame, td->td_kstack); 465 } 466 } 467 468 #endif 469