vm_machdep.c revision 72746
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 72746 2001-02-20 05:26:15Z jhb $ 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 fork_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)fork_return; /* s0: a0 */ 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: a1 */ 211 up->u_pcb.pcb_context[7] = 212 (u_int64_t)switch_trampoline; /* ra: assembly magic */ 213 } 214} 215 216/* 217 * Intercept the return address from a freshly forked process that has NOT 218 * been scheduled yet. 219 * 220 * This is needed to make kernel threads stay in kernel mode. 221 */ 222void 223cpu_set_fork_handler(p, func, arg) 224 struct proc *p; 225 void (*func) __P((void *)); 226 void *arg; 227{ 228 /* 229 * Note that the trap frame follows the args, so the function 230 * is really called like this: func(arg, frame); 231 */ 232 p->p_addr->u_pcb.pcb_context[0] = (u_long) func; 233 p->p_addr->u_pcb.pcb_context[2] = (u_long) arg; 234} 235 236/* 237 * cpu_exit is called as the last action during exit. 238 * We release the address space of the process, block interrupts, 239 * and call switch_exit. switch_exit switches to proc0's PCB and stack, 240 * then jumps into the middle of cpu_switch, as if it were switching 241 * from proc0. 242 */ 243void 244cpu_exit(p) 245 register struct proc *p; 246{ 247 alpha_fpstate_drop(p); 248 249 mtx_lock_spin(&sched_lock); 250 mtx_unlock_flags(&Giant, MTX_NOSWITCH); 251 mtx_assert(&Giant, MA_NOTOWNED); 252 253 /* 254 * We have to wait until after releasing all locks before 255 * changing p_stat. If we block on a mutex then we will be 256 * back at SRUN when we resume and our parent will never 257 * harvest us. 258 */ 259 p->p_stat = SZOMB; 260 261 mp_fixme("assumption: p_pptr won't change at this time"); 262 wakeup(p->p_pptr); 263 264 cnt.v_swtch++; 265 cpu_switch(); 266 panic("cpu_exit"); 267} 268 269void 270cpu_wait(p) 271 struct proc *p; 272{ 273 /* drop per-process resources */ 274 pmap_dispose_proc(p); 275 276 /* and clean-out the vmspace */ 277 vmspace_free(p->p_vmspace); 278} 279 280/* 281 * Dump the machine specific header information at the start of a core dump. 282 */ 283int 284cpu_coredump(p, vp, cred) 285 struct proc *p; 286 struct vnode *vp; 287 struct ucred *cred; 288{ 289 290 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 291 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 292 p)); 293} 294 295#ifdef notyet 296static void 297setredzone(pte, vaddr) 298 u_short *pte; 299 caddr_t vaddr; 300{ 301/* eventually do this by setting up an expand-down stack segment 302 for ss0: selector, allowing stack access down to top of u. 303 this means though that protection violations need to be handled 304 thru a double fault exception that must do an integral task 305 switch to a known good context, within which a dump can be 306 taken. a sensible scheme might be to save the initial context 307 used by sched (that has physical memory mapped 1:1 at bottom) 308 and take the dump while still in mapped mode */ 309} 310#endif 311 312/* 313 * Map an IO request into kernel virtual address space. 314 * 315 * All requests are (re)mapped into kernel VA space. 316 * Notice that we use b_bufsize for the size of the buffer 317 * to be mapped. b_bcount might be modified by the driver. 318 */ 319void 320vmapbuf(bp) 321 register struct buf *bp; 322{ 323 register caddr_t addr, v, kva; 324 vm_offset_t pa; 325 326 if ((bp->b_flags & B_PHYS) == 0) 327 panic("vmapbuf"); 328 329 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 330 addr < bp->b_data + bp->b_bufsize; 331 addr += PAGE_SIZE, v += PAGE_SIZE) { 332 /* 333 * Do the vm_fault if needed; do the copy-on-write thing 334 * when reading stuff off device into memory. 335 */ 336 vm_fault_quick(addr, 337 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 338 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 339 if (pa == 0) 340 panic("vmapbuf: page not present"); 341 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 342 pmap_kenter((vm_offset_t) v, pa); 343 } 344 345 kva = bp->b_saveaddr; 346 bp->b_saveaddr = bp->b_data; 347 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 348} 349 350/* 351 * Free the io map PTEs associated with this IO operation. 352 * We also invalidate the TLB entries and restore the original b_addr. 353 */ 354void 355vunmapbuf(bp) 356 register struct buf *bp; 357{ 358 register caddr_t addr; 359 vm_offset_t pa; 360 361 if ((bp->b_flags & B_PHYS) == 0) 362 panic("vunmapbuf"); 363 364 for (addr = (caddr_t)trunc_page(bp->b_data); 365 addr < bp->b_data + bp->b_bufsize; 366 addr += PAGE_SIZE) { 367 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 368 pmap_kremove((vm_offset_t) addr); 369 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 370 } 371 372 bp->b_data = bp->b_saveaddr; 373} 374 375/* 376 * Reset back to firmware. 377 */ 378void 379cpu_reset() 380{ 381 prom_halt(0); 382} 383 384int 385grow_stack(p, sp) 386 struct proc *p; 387 size_t sp; 388{ 389 int rv; 390 391 rv = vm_map_growstack (p, sp); 392 if (rv != KERN_SUCCESS) 393 return (0); 394 395 return (1); 396} 397 398 399static int cnt_prezero; 400 401SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 402 403/* 404 * Implement the pre-zeroed page mechanism. 405 * This routine is called from the idle loop. 406 */ 407 408#define ZIDLE_LO(v) ((v) * 2 / 3) 409#define ZIDLE_HI(v) ((v) * 4 / 5) 410 411int 412vm_page_zero_idle() 413{ 414 static int free_rover; 415 static int zero_state; 416 vm_page_t m; 417 int s; 418 419 /* 420 * Attempt to maintain approximately 1/2 of our free pages in a 421 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 422 * generally zeroing a page when the system is near steady-state. 423 * Otherwise we might get 'flutter' during disk I/O / IPC or 424 * fast sleeps. We also do not want to be continuously zeroing 425 * pages because doing so may flush our L1 and L2 caches too much. 426 */ 427 428 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 429 return(0); 430 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 431 return(0); 432 433 if (mtx_trylock(&Giant)) { 434 s = splvm(); 435 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 436 zero_state = 0; 437 if (m != NULL && (m->flags & PG_ZERO) == 0) { 438 vm_page_queues[m->queue].lcnt--; 439 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 440 m->queue = PQ_NONE; 441 splx(s); 442#if 0 443 rel_mplock(); 444#endif 445 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 446#if 0 447 get_mplock(); 448#endif 449 (void)splvm(); 450 vm_page_flag_set(m, PG_ZERO); 451 m->queue = PQ_FREE + m->pc; 452 vm_page_queues[m->queue].lcnt++; 453 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 454 pageq); 455 ++vm_page_zero_count; 456 ++cnt_prezero; 457 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 458 zero_state = 1; 459 } 460 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 461 splx(s); 462 mtx_unlock(&Giant); 463 return (1); 464 } 465 return (0); 466} 467 468/* 469 * Software interrupt handler for queued VM system processing. 470 */ 471void 472swi_vm(void *dummy) 473{ 474 if (busdma_swi_pending != 0) 475 busdma_swi(); 476} 477 478/* 479 * Tell whether this address is in some physical memory region. 480 * Currently used by the kernel coredump code in order to avoid 481 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 482 * or other unpredictable behaviour. 483 */ 484 485 486int 487is_physical_memory(addr) 488 vm_offset_t addr; 489{ 490 /* 491 * stuff other tests for known memory-mapped devices (PCI?) 492 * here 493 */ 494 495 return 1; 496} 497