sys/uvm/uvm_glue.c

   1 /*      $NetBSD: uvm_glue.c,v 1.142 2009/11/21 17:45:02 rmind Exp $     */
   2
   3 /*
   4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
   5  * Copyright (c) 1991, 1993, The Regents of the University of California.
   6  *
   7  * All rights reserved.
   8  *
   9  * This code is derived from software contributed to Berkeley by
  10  * The Mach Operating System project at Carnegie-Mellon University.
  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 Charles D. Cranor,
  23  *      Washington University, the University of California, Berkeley and
  24  *      its contributors.
  25  * 4. Neither the name of the University nor the names of its contributors
  26  *    may be used to endorse or promote products derived from this software
  27  *    without specific prior written permission.
  28  *
  29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  39  * SUCH DAMAGE.
  40  *
  41  *      @(#)vm_glue.c   8.6 (Berkeley) 1/5/94
  42  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
  43  *
  44  *
  45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
  46  * All rights reserved.
  47  *
  48  * Permission to use, copy, modify and distribute this software and
  49  * its documentation is hereby granted, provided that both the copyright
  50  * notice and this permission notice appear in all copies of the
  51  * software, derivative works or modified versions, and any portions
  52  * thereof, and that both notices appear in supporting documentation.
  53  *
  54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
  56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  57  *
  58  * Carnegie Mellon requests users of this software to return to
  59  *
  60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  61  *  School of Computer Science
  62  *  Carnegie Mellon University
  63  *  Pittsburgh PA 15213-3890
  64  *
  65  * any improvements or extensions that they make and grant Carnegie the
  66  * rights to redistribute these changes.
  67  */
  68
  69 #include <sys/cdefs.h>
  70 __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.142 2009/11/21 17:45:02 rmind Exp $");
  71
  72 #include "opt_kgdb.h"
  73 #include "opt_kstack.h"
  74 #include "opt_uvmhist.h"
  75
  76 /*
  77  * uvm_glue.c: glue functions
  78  */
  79
  80 #include <sys/param.h>
  81 #include <sys/systm.h>
  82 #include <sys/proc.h>
  83 #include <sys/resourcevar.h>
  84 #include <sys/buf.h>
  85 #include <sys/user.h>
  86 #include <sys/syncobj.h>
  87 #include <sys/cpu.h>
  88 #include <sys/atomic.h>
  89
  90 #include <uvm/uvm.h>
  91
  92 /*
  93  * XXXCDC: do these really belong here?
  94  */
  95
  96 /*
  97  * uvm_kernacc: can the kernel access a region of memory
  98  *
  99  * - used only by /dev/kmem driver (mem.c)
 100  */
 101
 102 bool
 103 uvm_kernacc(void *addr, size_t len, int rw)
 104 {
 105         bool rv;
 106         vaddr_t saddr, eaddr;
 107         vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
 108
 109         saddr = trunc_page((vaddr_t)addr);
 110         eaddr = round_page((vaddr_t)addr + len);
 111         vm_map_lock_read(kernel_map);
 112         rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
 113         vm_map_unlock_read(kernel_map);
 114
 115         return(rv);
 116 }
 117
 118 #ifdef KGDB
 119 /*
 120  * Change protections on kernel pages from addr to addr+len
 121  * (presumably so debugger can plant a breakpoint).
 122  *
 123  * We force the protection change at the pmap level.  If we were
 124  * to use vm_map_protect a change to allow writing would be lazily-
 125  * applied meaning we would still take a protection fault, something
 126  * we really don't want to do.  It would also fragment the kernel
 127  * map unnecessarily.  We cannot use pmap_protect since it also won't
 128  * enforce a write-enable request.  Using pmap_enter is the only way
 129  * we can ensure the change takes place properly.
 130  */
 131 void
 132 uvm_chgkprot(void *addr, size_t len, int rw)
 133 {
 134         vm_prot_t prot;
 135         paddr_t pa;
 136         vaddr_t sva, eva;
 137
 138         prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
 139         eva = round_page((vaddr_t)addr + len);
 140         for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
 141                 /*
 142                  * Extract physical address for the page.
 143                  */
 144                 if (pmap_extract(pmap_kernel(), sva, &pa) == false)
 145                         panic("%s: invalid page", __func__);
 146                 pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
 147         }
 148         pmap_update(pmap_kernel());
 149 }
 150 #endif
 151
 152 /*
 153  * uvm_vslock: wire user memory for I/O
 154  *
 155  * - called from physio and sys___sysctl
 156  * - XXXCDC: consider nuking this (or making it a macro?)
 157  */
 158
 159 int
 160 uvm_vslock(struct vmspace *vs, void *addr, size_t len, vm_prot_t access_type)
 161 {
 162         struct vm_map *map;
 163         vaddr_t start, end;
 164         int error;
 165
 166         map = &vs->vm_map;
 167         start = trunc_page((vaddr_t)addr);
 168         end = round_page((vaddr_t)addr + len);
 169         error = uvm_fault_wire(map, start, end, access_type, 0);
 170         return error;
 171 }
 172
 173 /*
 174  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
 175  *
 176  * - called from physio and sys___sysctl
 177  * - XXXCDC: consider nuking this (or making it a macro?)
 178  */
 179
 180 void
 181 uvm_vsunlock(struct vmspace *vs, void *addr, size_t len)
 182 {
 183         uvm_fault_unwire(&vs->vm_map, trunc_page((vaddr_t)addr),
 184                 round_page((vaddr_t)addr + len));
 185 }
 186
 187 /*
 188  * uvm_proc_fork: fork a virtual address space
 189  *
 190  * - the address space is copied as per parent map's inherit values
 191  */
 192 void
 193 uvm_proc_fork(struct proc *p1, struct proc *p2, bool shared)
 194 {
 195
 196         if (shared == true) {
 197                 p2->p_vmspace = NULL;
 198                 uvmspace_share(p1, p2);
 199         } else {
 200                 p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
 201         }
 202
 203         cpu_proc_fork(p1, p2);
 204 }
 205
 206 /*
 207  * uvm_lwp_fork: fork a thread
 208  *
 209  * - a new "user" structure is allocated for the child process
 210  *      [filled in by MD layer...]
 211  * - if specified, the child gets a new user stack described by
 212  *      stack and stacksize
 213  * - NOTE: the kernel stack may be at a different location in the child
 214  *      process, and thus addresses of automatic variables may be invalid
 215  *      after cpu_lwp_fork returns in the child process.  We do nothing here
 216  *      after cpu_lwp_fork returns.
 217  */
 218 void
 219 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
 220     void (*func)(void *), void *arg)
 221 {
 222
 223         /* Fill stack with magic number. */
 224         kstack_setup_magic(l2);
 225
 226         /*
 227          * cpu_lwp_fork() copy and update the pcb, and make the child ready
 228          * to run.  If this is a normal user fork, the child will exit
 229          * directly to user mode via child_return() on its first time
 230          * slice and will not return here.  If this is a kernel thread,
 231          * the specified entry point will be executed.
 232          */
 233         cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
 234
 235         /* Inactive emap for new LWP. */
 236         l2->l_emap_gen = UVM_EMAP_INACTIVE;
 237 }
 238
 239 #ifndef USPACE_ALIGN
 240 #define USPACE_ALIGN    0
 241 #endif
 242
 243 static pool_cache_t uvm_uarea_cache;
 244
 245 static void *
 246 uarea_poolpage_alloc(struct pool *pp, int flags)
 247 {
 248 #if defined(PMAP_MAP_POOLPAGE)
 249         if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
 250                 struct vm_page *pg;
 251                 vaddr_t va;
 252
 253                 pg = uvm_pagealloc(NULL, 0, NULL,
 254                    ((flags & PR_WAITOK) == 0 ? UVM_KMF_NOWAIT : 0));
 255                 if (pg == NULL)
 256                         return NULL;
 257                 va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
 258                 if (va == 0)
 259                         uvm_pagefree(pg);
 260                 return (void *)va;
 261         }
 262 #endif
 263         return (void *)uvm_km_alloc(kernel_map, pp->pr_alloc->pa_pagesz,
 264             USPACE_ALIGN, UVM_KMF_WIRED |
 265             ((flags & PR_WAITOK) ? UVM_KMF_WAITVA :
 266             (UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)));
 267 }
 268
 269 static void
 270 uarea_poolpage_free(struct pool *pp, void *addr)
 271 {
 272 #if defined(PMAP_MAP_POOLPAGE)
 273         if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
 274                 paddr_t pa;
 275
 276                 pa = PMAP_UNMAP_POOLPAGE((vaddr_t) addr);
 277                 KASSERT(pa != 0);
 278                 uvm_pagefree(PHYS_TO_VM_PAGE(pa));
 279                 return;
 280         }
 281 #endif
 282         uvm_km_free(kernel_map, (vaddr_t)addr, pp->pr_alloc->pa_pagesz,
 283             UVM_KMF_WIRED);
 284 }
 285
 286 static struct pool_allocator uvm_uarea_allocator = {
 287         .pa_alloc = uarea_poolpage_alloc,
 288         .pa_free = uarea_poolpage_free,
 289         .pa_pagesz = USPACE,
 290 };
 291
 292 void
 293 uvm_uarea_init(void)
 294 {
 295         int flags = PR_NOTOUCH;
 296
 297         /*
 298          * specify PR_NOALIGN unless the alignment provided by
 299          * the backend (USPACE_ALIGN) is sufficient to provide
 300          * pool page size (UPSACE) alignment.
 301          */
 302
 303         if ((USPACE_ALIGN == 0 && USPACE != PAGE_SIZE) ||
 304             (USPACE_ALIGN % USPACE) != 0) {
 305                 flags |= PR_NOALIGN;
 306         }
 307
 308         uvm_uarea_cache = pool_cache_init(USPACE, USPACE_ALIGN, 0, flags,
 309             "uarea", &uvm_uarea_allocator, IPL_NONE, NULL, NULL, NULL);
 310 }
 311
 312 /*
 313  * uvm_uarea_alloc: allocate a u-area
 314  */
 315
 316 vaddr_t
 317 uvm_uarea_alloc(void)
 318 {
 319
 320         return (vaddr_t)pool_cache_get(uvm_uarea_cache, PR_WAITOK);
 321 }
 322
 323 /*
 324  * uvm_uarea_free: free a u-area
 325  */
 326
 327 void
 328 uvm_uarea_free(vaddr_t uaddr)
 329 {
 330
 331         pool_cache_put(uvm_uarea_cache, (void *)uaddr);
 332 }
 333
 334 vaddr_t
 335 uvm_lwp_getuarea(lwp_t *l)
 336 {
 337
 338         return (vaddr_t)l->l_addr - UAREA_USER_OFFSET;
 339 }
 340
 341 void
 342 uvm_lwp_setuarea(lwp_t *l, vaddr_t addr)
 343 {
 344
 345         l->l_addr = (void *)(addr + UAREA_USER_OFFSET);
 346 }
 347
 348 /*
 349  * uvm_proc_exit: exit a virtual address space
 350  *
 351  * - borrow proc0's address space because freeing the vmspace
 352  *   of the dead process may block.
 353  */
 354
 355 void
 356 uvm_proc_exit(struct proc *p)
 357 {
 358         struct lwp *l = curlwp; /* XXX */
 359         struct vmspace *ovm;
 360
 361         KASSERT(p == l->l_proc);
 362         ovm = p->p_vmspace;
 363
 364         /*
 365          * borrow proc0's address space.
 366          */
 367         KPREEMPT_DISABLE(l);
 368         pmap_deactivate(l);
 369         p->p_vmspace = proc0.p_vmspace;
 370         pmap_activate(l);
 371         KPREEMPT_ENABLE(l);
 372
 373         uvmspace_free(ovm);
 374 }
 375
 376 void
 377 uvm_lwp_exit(struct lwp *l)
 378 {
 379         vaddr_t va = uvm_lwp_getuarea(l);
 380
 381         uvm_uarea_free(va);
 382 #ifdef DIAGNOSTIC
 383         uvm_lwp_setuarea(l, (vaddr_t)NULL);
 384 #endif
 385 }
 386
 387 /*
 388  * uvm_init_limit: init per-process VM limits
 389  *
 390  * - called for process 0 and then inherited by all others.
 391  */
 392
 393 void
 394 uvm_init_limits(struct proc *p)
 395 {
 396
 397         /*
 398          * Set up the initial limits on process VM.  Set the maximum
 399          * resident set size to be all of (reasonably) available memory.
 400          * This causes any single, large process to start random page
 401          * replacement once it fills memory.
 402          */
 403
 404         p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
 405         p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
 406         p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
 407         p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
 408         p->p_rlimit[RLIMIT_AS].rlim_cur = RLIM_INFINITY;
 409         p->p_rlimit[RLIMIT_AS].rlim_max = RLIM_INFINITY;
 410         p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
 411 }
 412
 413 /*
 414  * uvm_scheduler: process zero main loop.
 415  */
 416 void
 417 uvm_scheduler(void)
 418 {
 419         lwp_t *l = curlwp;
 420
 421         lwp_lock(l);
 422         l->l_priority = PRI_VM;
 423         l->l_class = SCHED_FIFO;
 424         lwp_unlock(l);
 425
 426         for (;;) {
 427                 /* XXX/TODO: move some workload to this LWP? */
 428                 (void)kpause("uvm", false, 0, NULL);
 429         }
 430 }