vm_machdep.c revision 209975
136865Sdfr/*- 236865Sdfr * Copyright (c) 1982, 1986 The Regents of the University of California. 336865Sdfr * Copyright (c) 1989, 1990 William Jolitz 436865Sdfr * Copyright (c) 1994 John Dyson 536865Sdfr * All rights reserved. 636865Sdfr * 736865Sdfr * This code is derived from software contributed to Berkeley by 836865Sdfr * the Systems Programming Group of the University of Utah Computer 936865Sdfr * Science Department, and William Jolitz. 1036865Sdfr * 1136865Sdfr * Redistribution and use in source and binary forms, with or without 1236865Sdfr * modification, are permitted provided that the following conditions 1336865Sdfr * are met: 1436865Sdfr * 1. Redistributions of source code must retain the above copyright 1536865Sdfr * notice, this list of conditions and the following disclaimer. 1636865Sdfr * 2. Redistributions in binary form must reproduce the above copyright 1736865Sdfr * notice, this list of conditions and the following disclaimer in the 1836865Sdfr * documentation and/or other materials provided with the distribution. 1936865Sdfr * 3. All advertising materials mentioning features or use of this software 2036865Sdfr * must display the following acknowledgement: 2136865Sdfr * This product includes software developed by the University of 2236865Sdfr * California, Berkeley and its contributors. 2336865Sdfr * 4. Neither the name of the University nor the names of its contributors 2436865Sdfr * may be used to endorse or promote products derived from this software 2536865Sdfr * without specific prior written permission. 2636865Sdfr * 2736865Sdfr * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2836865Sdfr * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2936865Sdfr * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 3036865Sdfr * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 3136865Sdfr * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 3236865Sdfr * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 3336865Sdfr * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 3436865Sdfr * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3536865Sdfr * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3636865Sdfr * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3736865Sdfr * SUCH DAMAGE. 3836865Sdfr * 3936865Sdfr * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 4036865Sdfr * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 4150477Speter * $FreeBSD: head/sys/powerpc/aim/vm_machdep.c 209975 2010-07-13 05:32:19Z nwhitehorn $ 4236865Sdfr */ 43139825Simp/*- 4436865Sdfr * Copyright (c) 1994, 1995, 1996 Carnegie-Mellon University. 4536865Sdfr * All rights reserved. 4636865Sdfr * 4736865Sdfr * Author: Chris G. Demetriou 48126474Sgrehan * 4936865Sdfr * Permission to use, copy, modify and distribute this software and 5036865Sdfr * its documentation is hereby granted, provided that both the copyright 5136865Sdfr * notice and this permission notice appear in all copies of the 5236865Sdfr * software, derivative works or modified versions, and any portions 5336865Sdfr * thereof, and that both notices appear in supporting documentation. 54126474Sgrehan * 55126474Sgrehan * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56126474Sgrehan * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 5736865Sdfr * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58126474Sgrehan * 5936865Sdfr * Carnegie Mellon requests users of this software to return to 6036865Sdfr * 6136865Sdfr * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 6236865Sdfr * School of Computer Science 6336865Sdfr * Carnegie Mellon University 6436865Sdfr * Pittsburgh PA 15213-3890 6536865Sdfr * 6636865Sdfr * any improvements or extensions that they make and grant Carnegie the 6736865Sdfr * rights to redistribute these changes. 6836865Sdfr */ 6936865Sdfr 7036865Sdfr#include <sys/param.h> 7136865Sdfr#include <sys/systm.h> 7236865Sdfr#include <sys/proc.h> 7336865Sdfr#include <sys/malloc.h> 7460041Sphk#include <sys/bio.h> 7536865Sdfr#include <sys/buf.h> 7685201Smp#include <sys/ktr.h> 7785201Smp#include <sys/lock.h> 7867365Sjhb#include <sys/mutex.h> 7936865Sdfr#include <sys/vnode.h> 8036865Sdfr#include <sys/vmmeter.h> 8136865Sdfr#include <sys/kernel.h> 82119563Salc#include <sys/mbuf.h> 83122780Salc#include <sys/sf_buf.h> 8436865Sdfr#include <sys/sysctl.h> 85199135Skib#include <sys/sysent.h> 8654207Speter#include <sys/unistd.h> 8736865Sdfr 8836865Sdfr#include <machine/cpu.h> 8941499Sdfr#include <machine/fpu.h> 9097397Sbenno#include <machine/frame.h> 9136865Sdfr#include <machine/md_var.h> 92138129Sdas#include <machine/pcb.h> 9336865Sdfr 9478342Sbenno#include <dev/ofw/openfirm.h> 9578342Sbenno 9636865Sdfr#include <vm/vm.h> 9736865Sdfr#include <vm/vm_param.h> 9836865Sdfr#include <vm/vm_kern.h> 9936865Sdfr#include <vm/vm_page.h> 10036865Sdfr#include <vm/vm_map.h> 10136865Sdfr#include <vm/vm_extern.h> 10236865Sdfr 10336865Sdfr/* 104190681Snwhitehorn * On systems without a direct mapped region (e.g. PPC64), 105190681Snwhitehorn * we use the same code as the Book E implementation. Since 106190681Snwhitehorn * we need to have runtime detection of this, define some machinery 107190681Snwhitehorn * for sf_bufs in this case, and ignore it on systems with direct maps. 108190681Snwhitehorn */ 109190681Snwhitehorn 110190681Snwhitehorn#ifndef NSFBUFS 111190681Snwhitehorn#define NSFBUFS (512 + maxusers * 16) 112190681Snwhitehorn#endif 113190681Snwhitehorn 114190681Snwhitehornstatic void sf_buf_init(void *arg); 115190681SnwhitehornSYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL); 116190681Snwhitehorn 117190681SnwhitehornLIST_HEAD(sf_head, sf_buf); 118190681Snwhitehorn 119190681Snwhitehorn/* A hash table of active sendfile(2) buffers */ 120190681Snwhitehornstatic struct sf_head *sf_buf_active; 121190681Snwhitehornstatic u_long sf_buf_hashmask; 122190681Snwhitehorn 123190681Snwhitehorn#define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask) 124190681Snwhitehorn 125190681Snwhitehornstatic TAILQ_HEAD(, sf_buf) sf_buf_freelist; 126190681Snwhitehornstatic u_int sf_buf_alloc_want; 127190681Snwhitehorn 128190681Snwhitehorn/* 129190681Snwhitehorn * A lock used to synchronize access to the hash table and free list 130190681Snwhitehorn */ 131190681Snwhitehornstatic struct mtx sf_buf_lock; 132190681Snwhitehorn 133209975Snwhitehorn#ifdef __powerpc64__ 134209975Snwhitehornextern uintptr_t tocbase; 135209975Snwhitehorn#endif 136190681Snwhitehorn 137209975Snwhitehorn 138190681Snwhitehorn/* 13936865Sdfr * Finish a fork operation, with process p2 nearly set up. 14036865Sdfr * Copy and update the pcb, set up the stack so that the child 14136865Sdfr * ready to run and return to user mode. 14236865Sdfr */ 14336865Sdfrvoid 14490361Sjuliancpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 14536865Sdfr{ 14684977Sbenno struct proc *p1; 14784977Sbenno struct trapframe *tf; 14884977Sbenno struct callframe *cf; 14990643Sbenno struct pcb *pcb; 15084977Sbenno 15191467Sbenno KASSERT(td1 == curthread || td1 == &thread0, 15290643Sbenno ("cpu_fork: p1 not curproc and not proc0")); 153209975Snwhitehorn CTR3(KTR_PROC, "cpu_fork: called td1=%p p2=%p flags=%x", 154209975Snwhitehorn td1, p2, flags); 15584977Sbenno 15684977Sbenno if ((flags & RFPROC) == 0) 15784977Sbenno return; 15884977Sbenno 15984977Sbenno p1 = td1->td_proc; 16084977Sbenno 161127582Speter pcb = (struct pcb *)((td2->td_kstack + 162209975Snwhitehorn td2->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb)) & ~0x2fUL); 16390643Sbenno td2->td_pcb = pcb; 16484977Sbenno 16590643Sbenno /* Copy the pcb */ 16690643Sbenno bcopy(td1->td_pcb, pcb, sizeof(struct pcb)); 16784977Sbenno 16884977Sbenno /* 16990643Sbenno * Create a fresh stack for the new process. 17084977Sbenno * Copy the trap frame for the return to user mode as if from a 17184977Sbenno * syscall. This copies most of the user mode register values. 17284977Sbenno */ 17390643Sbenno tf = (struct trapframe *)pcb - 1; 17490643Sbenno bcopy(td1->td_frame, tf, sizeof(*tf)); 17584977Sbenno 17697397Sbenno /* Set up trap frame. */ 17797397Sbenno tf->fixreg[FIRSTARG] = 0; 17897397Sbenno tf->fixreg[FIRSTARG + 1] = 0; 17997397Sbenno tf->cr &= ~0x10000000; 18090643Sbenno 18190643Sbenno td2->td_frame = tf; 18290643Sbenno 18390643Sbenno cf = (struct callframe *)tf - 1; 184132520Sgrehan memset(cf, 0, sizeof(struct callframe)); 185209975Snwhitehorn #ifdef __powerpc64__ 186209975Snwhitehorn cf->cf_toc = tocbase; 187209975Snwhitehorn #endif 18891467Sbenno cf->cf_func = (register_t)fork_return; 18991467Sbenno cf->cf_arg0 = (register_t)td2; 19091467Sbenno cf->cf_arg1 = (register_t)tf; 19184977Sbenno 19291467Sbenno pcb->pcb_sp = (register_t)cf; 193209975Snwhitehorn #ifdef __powerpc64__ 194209975Snwhitehorn pcb->pcb_lr = ((register_t *)fork_trampoline)[0]; 195209975Snwhitehorn pcb->pcb_toc = ((register_t *)fork_trampoline)[1]; 196209975Snwhitehorn #else 19791467Sbenno pcb->pcb_lr = (register_t)fork_trampoline; 198209975Snwhitehorn #endif 199209975Snwhitehorn pcb->pcb_cpu.aim.usr_vsid = 0; 200209975Snwhitehorn pcb->pcb_cpu.aim.usr_esid = 0; 20184977Sbenno 202170305Sjeff /* Setup to release spin count in fork_exit(). */ 203144637Sjhb td2->td_md.md_spinlock_count = 1; 204144637Sjhb td2->td_md.md_saved_msr = PSL_KERNSET; 205144637Sjhb 20690643Sbenno /* 20790643Sbenno * Now cpu_switch() can schedule the new process. 20890643Sbenno */ 20936865Sdfr} 21036865Sdfr 21136865Sdfr/* 21236865Sdfr * Intercept the return address from a freshly forked process that has NOT 21336865Sdfr * been scheduled yet. 21436865Sdfr * 21536865Sdfr * This is needed to make kernel threads stay in kernel mode. 21636865Sdfr */ 21736865Sdfrvoid 21883682Smpcpu_set_fork_handler(td, func, arg) 21983682Smp struct thread *td; 22092842Salfred void (*func)(void *); 22148391Speter void *arg; 22236865Sdfr{ 22384977Sbenno struct callframe *cf; 22484977Sbenno 225209975Snwhitehorn CTR4(KTR_PROC, "%s called with td=%p func=%p arg=%p", 226209975Snwhitehorn __func__, td, func, arg); 22784977Sbenno 22891467Sbenno cf = (struct callframe *)td->td_pcb->pcb_sp; 22984977Sbenno 23091467Sbenno cf->cf_func = (register_t)func; 23191467Sbenno cf->cf_arg0 = (register_t)arg; 23236865Sdfr} 23336865Sdfr 23436865Sdfrvoid 23583682Smpcpu_exit(td) 23683682Smp register struct thread *td; 23736865Sdfr{ 23836865Sdfr} 23936865Sdfr 24036865Sdfr/* 24165557Sjasone * Reset back to firmware. 24236865Sdfr */ 24336865Sdfrvoid 24436865Sdfrcpu_reset() 24536865Sdfr{ 246165608Smarcel OF_reboot(); 24736865Sdfr} 24836865Sdfr 24936865Sdfr/* 250190681Snwhitehorn * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 251119563Salc */ 252190681Snwhitehornstatic void 253190681Snwhitehornsf_buf_init(void *arg) 254190681Snwhitehorn{ 255190681Snwhitehorn struct sf_buf *sf_bufs; 256190681Snwhitehorn vm_offset_t sf_base; 257190681Snwhitehorn int i; 258190681Snwhitehorn 259190681Snwhitehorn /* Don't bother on systems with a direct map */ 260190681Snwhitehorn 261190681Snwhitehorn if (hw_direct_map) 262190681Snwhitehorn return; 263190681Snwhitehorn 264190681Snwhitehorn nsfbufs = NSFBUFS; 265190681Snwhitehorn TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs); 266190681Snwhitehorn 267190681Snwhitehorn sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask); 268190681Snwhitehorn TAILQ_INIT(&sf_buf_freelist); 269190681Snwhitehorn sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE); 270190681Snwhitehorn sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, M_NOWAIT | M_ZERO); 271190681Snwhitehorn 272190681Snwhitehorn for (i = 0; i < nsfbufs; i++) { 273190681Snwhitehorn sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 274190681Snwhitehorn TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry); 275190681Snwhitehorn } 276190681Snwhitehorn sf_buf_alloc_want = 0; 277190681Snwhitehorn mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF); 278190681Snwhitehorn} 279190681Snwhitehorn 280190681Snwhitehorn/* 281190681Snwhitehorn * Get an sf_buf from the freelist. Will block if none are available. 282190681Snwhitehorn */ 283119563Salcstruct sf_buf * 284190681Snwhitehornsf_buf_alloc(struct vm_page *m, int flags) 285119563Salc{ 286190681Snwhitehorn struct sf_head *hash_list; 287190681Snwhitehorn struct sf_buf *sf; 288190681Snwhitehorn int error; 289119563Salc 290190681Snwhitehorn if (hw_direct_map) { 291190681Snwhitehorn /* Shortcut the direct mapped case */ 292190681Snwhitehorn 293190681Snwhitehorn return ((struct sf_buf *)m); 294190681Snwhitehorn } 295190681Snwhitehorn 296190681Snwhitehorn hash_list = &sf_buf_active[SF_BUF_HASH(m)]; 297190681Snwhitehorn mtx_lock(&sf_buf_lock); 298190681Snwhitehorn LIST_FOREACH(sf, hash_list, list_entry) { 299190681Snwhitehorn if (sf->m == m) { 300190681Snwhitehorn sf->ref_count++; 301190681Snwhitehorn if (sf->ref_count == 1) { 302190681Snwhitehorn TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 303190681Snwhitehorn nsfbufsused++; 304190681Snwhitehorn nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 305190681Snwhitehorn } 306190681Snwhitehorn goto done; 307190681Snwhitehorn } 308190681Snwhitehorn } 309190681Snwhitehorn 310190681Snwhitehorn while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) { 311190681Snwhitehorn if (flags & SFB_NOWAIT) 312190681Snwhitehorn goto done; 313190681Snwhitehorn 314190681Snwhitehorn sf_buf_alloc_want++; 315190681Snwhitehorn mbstat.sf_allocwait++; 316190681Snwhitehorn error = msleep(&sf_buf_freelist, &sf_buf_lock, 317190681Snwhitehorn (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0); 318190681Snwhitehorn sf_buf_alloc_want--; 319190681Snwhitehorn 320190681Snwhitehorn /* 321190681Snwhitehorn * If we got a signal, don't risk going back to sleep. 322190681Snwhitehorn */ 323190681Snwhitehorn if (error) 324190681Snwhitehorn goto done; 325190681Snwhitehorn } 326190681Snwhitehorn 327190681Snwhitehorn TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 328190681Snwhitehorn if (sf->m != NULL) 329190681Snwhitehorn LIST_REMOVE(sf, list_entry); 330190681Snwhitehorn 331190681Snwhitehorn LIST_INSERT_HEAD(hash_list, sf, list_entry); 332190681Snwhitehorn sf->ref_count = 1; 333190681Snwhitehorn sf->m = m; 334190681Snwhitehorn nsfbufsused++; 335190681Snwhitehorn nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 336190681Snwhitehorn pmap_qenter(sf->kva, &sf->m, 1); 337190681Snwhitehorndone: 338190681Snwhitehorn mtx_unlock(&sf_buf_lock); 339190681Snwhitehorn return (sf); 340119563Salc} 341119563Salc 342119563Salc/* 343190681Snwhitehorn * Detatch mapped page and release resources back to the system. 344190681Snwhitehorn * 345190681Snwhitehorn * Remove a reference from the given sf_buf, adding it to the free 346190681Snwhitehorn * list when its reference count reaches zero. A freed sf_buf still, 347190681Snwhitehorn * however, retains its virtual-to-physical mapping until it is 348190681Snwhitehorn * recycled or reactivated by sf_buf_alloc(9). 349119563Salc */ 350119563Salcvoid 351127086Salcsf_buf_free(struct sf_buf *sf) 352119563Salc{ 353190681Snwhitehorn if (hw_direct_map) 354190681Snwhitehorn return; 355190681Snwhitehorn 356190681Snwhitehorn mtx_lock(&sf_buf_lock); 357190681Snwhitehorn sf->ref_count--; 358190681Snwhitehorn if (sf->ref_count == 0) { 359190681Snwhitehorn TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry); 360190681Snwhitehorn nsfbufsused--; 361190681Snwhitehorn 362190681Snwhitehorn if (sf_buf_alloc_want > 0) 363190681Snwhitehorn wakeup_one(&sf_buf_freelist); 364190681Snwhitehorn } 365190681Snwhitehorn mtx_unlock(&sf_buf_lock); 366119563Salc} 367119563Salc 368119563Salc/* 36936865Sdfr * Software interrupt handler for queued VM system processing. 370126474Sgrehan */ 371126474Sgrehanvoid 372126474Sgrehanswi_vm(void *dummy) 373126474Sgrehan{ 374209950Snwhitehorn 37536865Sdfr if (busdma_swi_pending != 0) 37636865Sdfr busdma_swi(); 37736865Sdfr} 37836865Sdfr 37936865Sdfr/* 38036865Sdfr * Tell whether this address is in some physical memory region. 38136865Sdfr * Currently used by the kernel coredump code in order to avoid 38236865Sdfr * dumping the ``ISA memory hole'' which could cause indefinite hangs, 38336865Sdfr * or other unpredictable behaviour. 38436865Sdfr */ 38536865Sdfr 38636865Sdfr 38736865Sdfrint 38836865Sdfris_physical_memory(addr) 38936865Sdfr vm_offset_t addr; 39036865Sdfr{ 39136865Sdfr /* 39236865Sdfr * stuff other tests for known memory-mapped devices (PCI?) 39336865Sdfr * here 39436865Sdfr */ 39536865Sdfr 39636865Sdfr return 1; 39736865Sdfr} 39899659Sbenno 39999659Sbenno/* 400209975Snwhitehorn * CPU threading functions related to the VM layer. These could be used 401209975Snwhitehorn * to map the SLB bits required for the kernel stack instead of forcing a 402209975Snwhitehorn * fixed-size KVA. 40399659Sbenno */ 40499659Sbenno 40599659Sbennovoid 406119004Smarcelcpu_thread_swapin(struct thread *td) 407119004Smarcel{ 40899659Sbenno} 40999659Sbenno 41099659Sbennovoid 411119004Smarcelcpu_thread_swapout(struct thread *td) 412119004Smarcel{ 413119004Smarcel} 414119004Smarcel 415