vm_machdep.c revision 68900
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 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41 * $FreeBSD: head/sys/powerpc/aim/vm_machdep.c 68900 2000-11-19 12:28:42Z dfr $ 42 */ 43/* 44 * Copyright (c) 1994, 1995, 1996 Carnegie-Mellon University. 45 * All rights reserved. 46 * 47 * Author: Chris G. Demetriou 48 * 49 * Permission to use, copy, modify and distribute this software and 50 * its documentation is hereby granted, provided that both the copyright 51 * notice and this permission notice appear in all copies of the 52 * software, derivative works or modified versions, and any portions 53 * thereof, and that both notices appear in supporting documentation. 54 * 55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58 * 59 * Carnegie Mellon requests users of this software to return to 60 * 61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 62 * School of Computer Science 63 * Carnegie Mellon University 64 * Pittsburgh PA 15213-3890 65 * 66 * any improvements or extensions that they make and grant Carnegie the 67 * rights to redistribute these changes. 68 */ 69 70#include <sys/param.h> 71#include <sys/systm.h> 72#include <sys/proc.h> 73#include <sys/malloc.h> 74#include <sys/bio.h> 75#include <sys/buf.h> 76#include <sys/mutex.h> 77#include <sys/vnode.h> 78#include <sys/vmmeter.h> 79#include <sys/kernel.h> 80#include <sys/sysctl.h> 81#include <sys/unistd.h> 82 83#include <machine/clock.h> 84#include <machine/cpu.h> 85#include <machine/fpu.h> 86#include <machine/md_var.h> 87#include <machine/prom.h> 88 89#include <vm/vm.h> 90#include <vm/vm_param.h> 91#include <sys/lock.h> 92#include <vm/vm_kern.h> 93#include <vm/vm_page.h> 94#include <vm/vm_map.h> 95#include <vm/vm_extern.h> 96 97#include <sys/user.h> 98 99/* 100 * quick version of vm_fault 101 */ 102int 103vm_fault_quick(v, prot) 104 caddr_t v; 105 int prot; 106{ 107 int r; 108 if (prot & VM_PROT_WRITE) 109 r = subyte(v, fubyte(v)); 110 else 111 r = fubyte(v); 112 return(r); 113} 114 115/* 116 * Finish a fork operation, with process p2 nearly set up. 117 * Copy and update the pcb, set up the stack so that the child 118 * ready to run and return to user mode. 119 */ 120void 121cpu_fork(p1, p2, flags) 122 register struct proc *p1, *p2; 123 int flags; 124{ 125 if ((flags & RFPROC) == 0) 126 return; 127 128 p2->p_md.md_tf = p1->p_md.md_tf; 129 p2->p_md.md_flags = p1->p_md.md_flags & (MDP_FPUSED | MDP_UAC_MASK); 130 131 /* 132 * Cache the physical address of the pcb, so we can 133 * swap to it easily. 134 */ 135 p2->p_md.md_pcbpaddr = (void*)vtophys((vm_offset_t)&p2->p_addr->u_pcb); 136 137 /* 138 * Copy floating point state from the FP chip to the PCB 139 * if this process has state stored there. 140 */ 141 alpha_fpstate_save(p1, 0); 142 143 /* 144 * Copy pcb and stack from proc p1 to p2. We do this as 145 * cheaply as possible, copying only the active part of the 146 * stack. The stack and pcb need to agree. Make sure that the 147 * new process has FEN disabled. 148 */ 149 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 150 p2->p_addr->u_pcb.pcb_hw.apcb_usp = alpha_pal_rdusp(); 151 p2->p_addr->u_pcb.pcb_hw.apcb_flags &= ~ALPHA_PCB_FLAGS_FEN; 152 153 /* 154 * Set the floating point state. 155 */ 156 if ((p2->p_addr->u_pcb.pcb_fp_control & IEEE_INHERIT) == 0) { 157 p2->p_addr->u_pcb.pcb_fp_control = 0; 158 p2->p_addr->u_pcb.pcb_fp.fpr_cr = (FPCR_DYN_NORMAL 159 | FPCR_INVD | FPCR_DZED 160 | FPCR_OVFD | FPCR_INED 161 | FPCR_UNFD); 162 } 163 164 /* 165 * Arrange for a non-local goto when the new process 166 * is started, to resume here, returning nonzero from setjmp. 167 */ 168#ifdef DIAGNOSTIC 169 if (p1 != curproc) 170 panic("cpu_fork: curproc"); 171 alpha_fpstate_check(p1); 172#endif 173 174 /* 175 * create the child's kernel stack, from scratch. 176 */ 177 { 178 struct user *up = p2->p_addr; 179 struct trapframe *p2tf; 180 181 /* 182 * Pick a stack pointer, leaving room for a trapframe; 183 * copy trapframe from parent so return to user mode 184 * will be to right address, with correct registers. 185 */ 186 p2tf = p2->p_md.md_tf = (struct trapframe *) 187 ((char *)p2->p_addr + USPACE - sizeof(struct trapframe)); 188 bcopy(p1->p_md.md_tf, p2->p_md.md_tf, 189 sizeof(struct trapframe)); 190 191 /* 192 * Set up return-value registers as fork() libc stub expects. 193 */ 194 p2tf->tf_regs[FRAME_V0] = 0; /* child's pid (linux) */ 195 p2tf->tf_regs[FRAME_A3] = 0; /* no error */ 196 p2tf->tf_regs[FRAME_A4] = 1; /* is child (FreeBSD) */ 197 198 /* 199 * Arrange for continuation at child_return(), which 200 * will return to exception_return(). Note that the child 201 * process doesn't stay in the kernel for long! 202 * 203 * This is an inlined version of cpu_set_kpc. 204 */ 205 up->u_pcb.pcb_hw.apcb_ksp = (u_int64_t)p2tf; 206 up->u_pcb.pcb_context[0] = 207 (u_int64_t)child_return; /* s0: pc */ 208 up->u_pcb.pcb_context[1] = 209 (u_int64_t)exception_return; /* s1: ra */ 210 up->u_pcb.pcb_context[2] = (u_long) p2; /* s2: a0 */ 211 up->u_pcb.pcb_context[7] = 212 (u_int64_t)switch_trampoline; /* ra: assembly magic */ 213 214 /* 215 * Clear the saved recursion count for sched_lock 216 * since the child needs only one count which is 217 * released in switch_trampoline. 218 */ 219 up->u_pcb.pcb_schednest = 0; 220 } 221} 222 223/* 224 * Intercept the return address from a freshly forked process that has NOT 225 * been scheduled yet. 226 * 227 * This is needed to make kernel threads stay in kernel mode. 228 */ 229void 230cpu_set_fork_handler(p, func, arg) 231 struct proc *p; 232 void (*func) __P((void *)); 233 void *arg; 234{ 235 /* 236 * Note that the trap frame follows the args, so the function 237 * is really called like this: func(arg, frame); 238 */ 239 p->p_addr->u_pcb.pcb_context[0] = (u_long) func; 240 p->p_addr->u_pcb.pcb_context[2] = (u_long) arg; 241} 242 243/* 244 * cpu_exit is called as the last action during exit. 245 * We release the address space of the process, block interrupts, 246 * and call switch_exit. switch_exit switches to proc0's PCB and stack, 247 * then jumps into the middle of cpu_switch, as if it were switching 248 * from proc0. 249 */ 250void 251cpu_exit(p) 252 register struct proc *p; 253{ 254 alpha_fpstate_drop(p); 255 256 mtx_enter(&sched_lock, MTX_SPIN); 257 mtx_exit(&Giant, MTX_DEF | MTX_NOSWITCH); 258 mtx_assert(&Giant, MA_NOTOWNED); 259 cnt.v_swtch++; 260 cpu_switch(); 261 panic("cpu_exit"); 262} 263 264void 265cpu_wait(p) 266 struct proc *p; 267{ 268 /* drop per-process resources */ 269 pmap_dispose_proc(p); 270 271 /* and clean-out the vmspace */ 272 vmspace_free(p->p_vmspace); 273} 274 275/* 276 * Dump the machine specific header information at the start of a core dump. 277 */ 278int 279cpu_coredump(p, vp, cred) 280 struct proc *p; 281 struct vnode *vp; 282 struct ucred *cred; 283{ 284 285 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 286 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 287 p)); 288} 289 290#ifdef notyet 291static void 292setredzone(pte, vaddr) 293 u_short *pte; 294 caddr_t vaddr; 295{ 296/* eventually do this by setting up an expand-down stack segment 297 for ss0: selector, allowing stack access down to top of u. 298 this means though that protection violations need to be handled 299 thru a double fault exception that must do an integral task 300 switch to a known good context, within which a dump can be 301 taken. a sensible scheme might be to save the initial context 302 used by sched (that has physical memory mapped 1:1 at bottom) 303 and take the dump while still in mapped mode */ 304} 305#endif 306 307/* 308 * Map an IO request into kernel virtual address space. 309 * 310 * All requests are (re)mapped into kernel VA space. 311 * Notice that we use b_bufsize for the size of the buffer 312 * to be mapped. b_bcount might be modified by the driver. 313 */ 314void 315vmapbuf(bp) 316 register struct buf *bp; 317{ 318 register caddr_t addr, v, kva; 319 vm_offset_t pa; 320 321 if ((bp->b_flags & B_PHYS) == 0) 322 panic("vmapbuf"); 323 324 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 325 addr < bp->b_data + bp->b_bufsize; 326 addr += PAGE_SIZE, v += PAGE_SIZE) { 327 /* 328 * Do the vm_fault if needed; do the copy-on-write thing 329 * when reading stuff off device into memory. 330 */ 331 vm_fault_quick(addr, 332 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 333 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 334 if (pa == 0) 335 panic("vmapbuf: page not present"); 336 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 337 pmap_kenter((vm_offset_t) v, pa); 338 } 339 340 kva = bp->b_saveaddr; 341 bp->b_saveaddr = bp->b_data; 342 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 343} 344 345/* 346 * Free the io map PTEs associated with this IO operation. 347 * We also invalidate the TLB entries and restore the original b_addr. 348 */ 349void 350vunmapbuf(bp) 351 register struct buf *bp; 352{ 353 register caddr_t addr; 354 vm_offset_t pa; 355 356 if ((bp->b_flags & B_PHYS) == 0) 357 panic("vunmapbuf"); 358 359 for (addr = (caddr_t)trunc_page(bp->b_data); 360 addr < bp->b_data + bp->b_bufsize; 361 addr += PAGE_SIZE) { 362 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 363 pmap_kremove((vm_offset_t) addr); 364 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 365 } 366 367 bp->b_data = bp->b_saveaddr; 368} 369 370/* 371 * Reset back to firmware. 372 */ 373void 374cpu_reset() 375{ 376 prom_halt(0); 377} 378 379int 380grow_stack(p, sp) 381 struct proc *p; 382 size_t sp; 383{ 384 int rv; 385 386 rv = vm_map_growstack (p, sp); 387 if (rv != KERN_SUCCESS) 388 return (0); 389 390 return (1); 391} 392 393 394static int cnt_prezero; 395 396SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 397 398/* 399 * Implement the pre-zeroed page mechanism. 400 * This routine is called from the idle loop. 401 */ 402 403#define ZIDLE_LO(v) ((v) * 2 / 3) 404#define ZIDLE_HI(v) ((v) * 4 / 5) 405 406int 407vm_page_zero_idle() 408{ 409 static int free_rover; 410 static int zero_state; 411 vm_page_t m; 412 int s; 413 414 /* 415 * Attempt to maintain approximately 1/2 of our free pages in a 416 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 417 * generally zeroing a page when the system is near steady-state. 418 * Otherwise we might get 'flutter' during disk I/O / IPC or 419 * fast sleeps. We also do not want to be continuously zeroing 420 * pages because doing so may flush our L1 and L2 caches too much. 421 */ 422 423 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 424 return(0); 425 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 426 return(0); 427 428#ifdef SMP 429 if (KLOCK_ENTER(M_TRY)) { 430#endif 431 s = splvm(); 432 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 433 zero_state = 0; 434 if (m != NULL && (m->flags & PG_ZERO) == 0) { 435 vm_page_queues[m->queue].lcnt--; 436 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 437 m->queue = PQ_NONE; 438 splx(s); 439#if 0 440 rel_mplock(); 441#endif 442 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 443#if 0 444 get_mplock(); 445#endif 446 (void)splvm(); 447 vm_page_flag_set(m, PG_ZERO); 448 m->queue = PQ_FREE + m->pc; 449 vm_page_queues[m->queue].lcnt++; 450 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 451 pageq); 452 ++vm_page_zero_count; 453 ++cnt_prezero; 454 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 455 zero_state = 1; 456 } 457 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 458 splx(s); 459#ifdef SMP 460 KLOCK_EXIT; 461#endif 462 return (1); 463#ifdef SMP 464 } 465#endif 466 return (0); 467} 468 469/* 470 * Software interrupt handler for queued VM system processing. 471 */ 472void 473swi_vm(void *dummy) 474{ 475 if (busdma_swi_pending != 0) 476 busdma_swi(); 477} 478 479/* 480 * Tell whether this address is in some physical memory region. 481 * Currently used by the kernel coredump code in order to avoid 482 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 483 * or other unpredictable behaviour. 484 */ 485 486 487int 488is_physical_memory(addr) 489 vm_offset_t addr; 490{ 491 /* 492 * stuff other tests for known memory-mapped devices (PCI?) 493 * here 494 */ 495 496 return 1; 497} 498