vm_machdep.c revision 255273
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 255273 2013-09-05 23:00:24Z nwhitehorn $ 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/ktr.h> 77#include <sys/lock.h> 78#include <sys/mutex.h> 79#include <sys/vnode.h> 80#include <sys/vmmeter.h> 81#include <sys/kernel.h> 82#include <sys/mbuf.h> 83#include <sys/sf_buf.h> 84#include <sys/sysctl.h> 85#include <sys/sysent.h> 86#include <sys/unistd.h> 87 88#include <machine/cpu.h> 89#include <machine/fpu.h> 90#include <machine/frame.h> 91#include <machine/md_var.h> 92#include <machine/pcb.h> 93 94#include <dev/ofw/openfirm.h> 95 96#include <vm/vm.h> 97#include <vm/vm_param.h> 98#include <vm/vm_kern.h> 99#include <vm/vm_page.h> 100#include <vm/vm_map.h> 101#include <vm/vm_extern.h> 102 103/* 104 * On systems without a direct mapped region (e.g. PPC64), 105 * we use the same code as the Book E implementation. Since 106 * we need to have runtime detection of this, define some machinery 107 * for sf_bufs in this case, and ignore it on systems with direct maps. 108 */ 109 110#ifndef NSFBUFS 111#define NSFBUFS (512 + maxusers * 16) 112#endif 113 114static void sf_buf_init(void *arg); 115SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL); 116 117LIST_HEAD(sf_head, sf_buf); 118 119/* A hash table of active sendfile(2) buffers */ 120static struct sf_head *sf_buf_active; 121static u_long sf_buf_hashmask; 122 123#define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask) 124 125static TAILQ_HEAD(, sf_buf) sf_buf_freelist; 126static u_int sf_buf_alloc_want; 127 128/* 129 * A lock used to synchronize access to the hash table and free list 130 */ 131static struct mtx sf_buf_lock; 132 133#ifdef __powerpc64__ 134extern uintptr_t tocbase; 135#endif 136 137 138/* 139 * Finish a fork operation, with process p2 nearly set up. 140 * Copy and update the pcb, set up the stack so that the child 141 * ready to run and return to user mode. 142 */ 143void 144cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 145{ 146 struct trapframe *tf; 147 struct callframe *cf; 148 struct pcb *pcb; 149 150 KASSERT(td1 == curthread || td1 == &thread0, 151 ("cpu_fork: p1 not curproc and not proc0")); 152 CTR3(KTR_PROC, "cpu_fork: called td1=%p p2=%p flags=%x", 153 td1, p2, flags); 154 155 if ((flags & RFPROC) == 0) 156 return; 157 158 pcb = (struct pcb *)((td2->td_kstack + 159 td2->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb)) & ~0x2fUL); 160 td2->td_pcb = pcb; 161 162 /* Copy the pcb */ 163 bcopy(td1->td_pcb, pcb, sizeof(struct pcb)); 164 165 /* 166 * Create a fresh stack for the new process. 167 * Copy the trap frame for the return to user mode as if from a 168 * syscall. This copies most of the user mode register values. 169 */ 170 tf = (struct trapframe *)pcb - 1; 171 bcopy(td1->td_frame, tf, sizeof(*tf)); 172 173 /* Set up trap frame. */ 174 tf->fixreg[FIRSTARG] = 0; 175 tf->fixreg[FIRSTARG + 1] = 0; 176 tf->cr &= ~0x10000000; 177 178 td2->td_frame = tf; 179 180 cf = (struct callframe *)tf - 1; 181 memset(cf, 0, sizeof(struct callframe)); 182 #ifdef __powerpc64__ 183 cf->cf_toc = tocbase; 184 #endif 185 cf->cf_func = (register_t)fork_return; 186 cf->cf_arg0 = (register_t)td2; 187 cf->cf_arg1 = (register_t)tf; 188 189 pcb->pcb_sp = (register_t)cf; 190 KASSERT(pcb->pcb_sp % 16 == 0, ("stack misaligned")); 191 #ifdef __powerpc64__ 192 pcb->pcb_lr = ((register_t *)fork_trampoline)[0]; 193 pcb->pcb_toc = ((register_t *)fork_trampoline)[1]; 194 #else 195 pcb->pcb_lr = (register_t)fork_trampoline; 196 #endif 197 pcb->pcb_cpu.aim.usr_vsid = 0; 198 199 /* Setup to release spin count in fork_exit(). */ 200 td2->td_md.md_spinlock_count = 1; 201 td2->td_md.md_saved_msr = PSL_KERNSET; 202 203 /* 204 * Now cpu_switch() can schedule the new process. 205 */ 206} 207 208/* 209 * Intercept the return address from a freshly forked process that has NOT 210 * been scheduled yet. 211 * 212 * This is needed to make kernel threads stay in kernel mode. 213 */ 214void 215cpu_set_fork_handler(td, func, arg) 216 struct thread *td; 217 void (*func)(void *); 218 void *arg; 219{ 220 struct callframe *cf; 221 222 CTR4(KTR_PROC, "%s called with td=%p func=%p arg=%p", 223 __func__, td, func, arg); 224 225 cf = (struct callframe *)td->td_pcb->pcb_sp; 226 227 cf->cf_func = (register_t)func; 228 cf->cf_arg0 = (register_t)arg; 229} 230 231void 232cpu_exit(td) 233 register struct thread *td; 234{ 235} 236 237/* 238 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 239 */ 240static void 241sf_buf_init(void *arg) 242{ 243 struct sf_buf *sf_bufs; 244 vm_offset_t sf_base; 245 int i; 246 247 /* Don't bother on systems with a direct map */ 248 249 if (hw_direct_map) 250 return; 251 252 nsfbufs = NSFBUFS; 253 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs); 254 255 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask); 256 TAILQ_INIT(&sf_buf_freelist); 257 sf_base = kva_alloc(nsfbufs * PAGE_SIZE); 258 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT | M_ZERO); 259 260 for (i = 0; i < nsfbufs; i++) { 261 sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 262 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry); 263 } 264 sf_buf_alloc_want = 0; 265 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF); 266} 267 268/* 269 * Get an sf_buf from the freelist. Will block if none are available. 270 */ 271struct sf_buf * 272sf_buf_alloc(struct vm_page *m, int flags) 273{ 274 struct sf_head *hash_list; 275 struct sf_buf *sf; 276 int error; 277 278 if (hw_direct_map) { 279 /* Shortcut the direct mapped case */ 280 281 return ((struct sf_buf *)m); 282 } 283 284 hash_list = &sf_buf_active[SF_BUF_HASH(m)]; 285 mtx_lock(&sf_buf_lock); 286 LIST_FOREACH(sf, hash_list, list_entry) { 287 if (sf->m == m) { 288 sf->ref_count++; 289 if (sf->ref_count == 1) { 290 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 291 nsfbufsused++; 292 nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 293 } 294 goto done; 295 } 296 } 297 298 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) { 299 if (flags & SFB_NOWAIT) 300 goto done; 301 302 sf_buf_alloc_want++; 303 SFSTAT_INC(sf_allocwait); 304 error = msleep(&sf_buf_freelist, &sf_buf_lock, 305 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0); 306 sf_buf_alloc_want--; 307 308 /* 309 * If we got a signal, don't risk going back to sleep. 310 */ 311 if (error) 312 goto done; 313 } 314 315 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 316 if (sf->m != NULL) 317 LIST_REMOVE(sf, list_entry); 318 319 LIST_INSERT_HEAD(hash_list, sf, list_entry); 320 sf->ref_count = 1; 321 sf->m = m; 322 nsfbufsused++; 323 nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 324 pmap_qenter(sf->kva, &sf->m, 1); 325done: 326 mtx_unlock(&sf_buf_lock); 327 return (sf); 328} 329 330/* 331 * Detatch mapped page and release resources back to the system. 332 * 333 * Remove a reference from the given sf_buf, adding it to the free 334 * list when its reference count reaches zero. A freed sf_buf still, 335 * however, retains its virtual-to-physical mapping until it is 336 * recycled or reactivated by sf_buf_alloc(9). 337 */ 338void 339sf_buf_free(struct sf_buf *sf) 340{ 341 if (hw_direct_map) 342 return; 343 344 mtx_lock(&sf_buf_lock); 345 sf->ref_count--; 346 if (sf->ref_count == 0) { 347 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry); 348 nsfbufsused--; 349 350 if (sf_buf_alloc_want > 0) 351 wakeup(&sf_buf_freelist); 352 } 353 mtx_unlock(&sf_buf_lock); 354} 355 356/* 357 * Software interrupt handler for queued VM system processing. 358 */ 359void 360swi_vm(void *dummy) 361{ 362 363 if (busdma_swi_pending != 0) 364 busdma_swi(); 365} 366 367/* 368 * Tell whether this address is in some physical memory region. 369 * Currently used by the kernel coredump code in order to avoid 370 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 371 * or other unpredictable behaviour. 372 */ 373 374 375int 376is_physical_memory(addr) 377 vm_offset_t addr; 378{ 379 /* 380 * stuff other tests for known memory-mapped devices (PCI?) 381 * here 382 */ 383 384 return 1; 385} 386 387/* 388 * CPU threading functions related to the VM layer. These could be used 389 * to map the SLB bits required for the kernel stack instead of forcing a 390 * fixed-size KVA. 391 */ 392 393void 394cpu_thread_swapin(struct thread *td) 395{ 396} 397 398void 399cpu_thread_swapout(struct thread *td) 400{ 401} 402 403