xref: /netbsd-src/sys/uvm/uvm_glue.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: uvm_glue.c,v 1.168 2019/05/08 16:00:01 chs 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. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vm_glue.c	8.6 (Berkeley) 1/5/94
37  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
38  *
39  *
40  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
41  * All rights reserved.
42  *
43  * Permission to use, copy, modify and distribute this software and
44  * its documentation is hereby granted, provided that both the copyright
45  * notice and this permission notice appear in all copies of the
46  * software, derivative works or modified versions, and any portions
47  * thereof, and that both notices appear in supporting documentation.
48  *
49  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
50  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
51  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52  *
53  * Carnegie Mellon requests users of this software to return to
54  *
55  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
56  *  School of Computer Science
57  *  Carnegie Mellon University
58  *  Pittsburgh PA 15213-3890
59  *
60  * any improvements or extensions that they make and grant Carnegie the
61  * rights to redistribute these changes.
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.168 2019/05/08 16:00:01 chs Exp $");
66 
67 #include "opt_kgdb.h"
68 #include "opt_kstack.h"
69 #include "opt_uvmhist.h"
70 #include "opt_kasan.h"
71 
72 /*
73  * uvm_glue.c: glue functions
74  */
75 
76 #include <sys/param.h>
77 #include <sys/kernel.h>
78 
79 #include <sys/systm.h>
80 #include <sys/proc.h>
81 #include <sys/resourcevar.h>
82 #include <sys/buf.h>
83 #include <sys/syncobj.h>
84 #include <sys/cpu.h>
85 #include <sys/atomic.h>
86 #include <sys/lwp.h>
87 #include <sys/asan.h>
88 
89 #include <uvm/uvm.h>
90 
91 /*
92  * uvm_kernacc: test if kernel can access a memory region.
93  *
94  * => Currently used only by /dev/kmem driver (dev/mm.c).
95  */
96 bool
97 uvm_kernacc(void *addr, size_t len, vm_prot_t prot)
98 {
99 	vaddr_t saddr = trunc_page((vaddr_t)addr);
100 	vaddr_t eaddr = round_page(saddr + len);
101 	bool rv;
102 
103 	vm_map_lock_read(kernel_map);
104 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
105 	vm_map_unlock_read(kernel_map);
106 
107 	return rv;
108 }
109 
110 #ifdef KGDB
111 /*
112  * Change protections on kernel pages from addr to addr+len
113  * (presumably so debugger can plant a breakpoint).
114  *
115  * We force the protection change at the pmap level.  If we were
116  * to use vm_map_protect a change to allow writing would be lazily-
117  * applied meaning we would still take a protection fault, something
118  * we really don't want to do.  It would also fragment the kernel
119  * map unnecessarily.  We cannot use pmap_protect since it also won't
120  * enforce a write-enable request.  Using pmap_enter is the only way
121  * we can ensure the change takes place properly.
122  */
123 void
124 uvm_chgkprot(void *addr, size_t len, int rw)
125 {
126 	vm_prot_t prot;
127 	paddr_t pa;
128 	vaddr_t sva, eva;
129 
130 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
131 	eva = round_page((vaddr_t)addr + len);
132 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
133 		/*
134 		 * Extract physical address for the page.
135 		 */
136 		if (pmap_extract(pmap_kernel(), sva, &pa) == false)
137 			panic("%s: invalid page", __func__);
138 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
139 	}
140 	pmap_update(pmap_kernel());
141 }
142 #endif
143 
144 /*
145  * uvm_vslock: wire user memory for I/O
146  *
147  * - called from physio and sys___sysctl
148  * - XXXCDC: consider nuking this (or making it a macro?)
149  */
150 
151 int
152 uvm_vslock(struct vmspace *vs, void *addr, size_t len, vm_prot_t access_type)
153 {
154 	struct vm_map *map;
155 	vaddr_t start, end;
156 	int error;
157 
158 	map = &vs->vm_map;
159 	start = trunc_page((vaddr_t)addr);
160 	end = round_page((vaddr_t)addr + len);
161 	error = uvm_fault_wire(map, start, end, access_type, 0);
162 	return error;
163 }
164 
165 /*
166  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
167  *
168  * - called from physio and sys___sysctl
169  * - XXXCDC: consider nuking this (or making it a macro?)
170  */
171 
172 void
173 uvm_vsunlock(struct vmspace *vs, void *addr, size_t len)
174 {
175 	uvm_fault_unwire(&vs->vm_map, trunc_page((vaddr_t)addr),
176 		round_page((vaddr_t)addr + len));
177 }
178 
179 /*
180  * uvm_proc_fork: fork a virtual address space
181  *
182  * - the address space is copied as per parent map's inherit values
183  */
184 void
185 uvm_proc_fork(struct proc *p1, struct proc *p2, bool shared)
186 {
187 
188 	if (shared == true) {
189 		p2->p_vmspace = NULL;
190 		uvmspace_share(p1, p2);
191 	} else {
192 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
193 	}
194 
195 	cpu_proc_fork(p1, p2);
196 }
197 
198 /*
199  * uvm_lwp_fork: fork a thread
200  *
201  * - a new PCB structure is allocated for the child process,
202  *	and filled in by MD layer
203  * - if specified, the child gets a new user stack described by
204  *	stack and stacksize
205  * - NOTE: the kernel stack may be at a different location in the child
206  *	process, and thus addresses of automatic variables may be invalid
207  *	after cpu_lwp_fork returns in the child process.  We do nothing here
208  *	after cpu_lwp_fork returns.
209  */
210 void
211 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
212     void (*func)(void *), void *arg)
213 {
214 
215 	/* Fill stack with magic number. */
216 	kstack_setup_magic(l2);
217 
218 	/*
219 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
220  	 * to run.  If this is a normal user fork, the child will exit
221 	 * directly to user mode via child_return() on its first time
222 	 * slice and will not return here.  If this is a kernel thread,
223 	 * the specified entry point will be executed.
224 	 */
225 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
226 
227 	/* Inactive emap for new LWP. */
228 	l2->l_emap_gen = UVM_EMAP_INACTIVE;
229 }
230 
231 #ifndef USPACE_ALIGN
232 #define	USPACE_ALIGN	0
233 #endif
234 
235 static pool_cache_t uvm_uarea_cache;
236 #if defined(__HAVE_CPU_UAREA_ROUTINES)
237 static pool_cache_t uvm_uarea_system_cache;
238 #else
239 #define uvm_uarea_system_cache uvm_uarea_cache
240 #endif
241 
242 static void *
243 uarea_poolpage_alloc(struct pool *pp, int flags)
244 {
245 
246 	KASSERT((flags & PR_WAITOK) != 0);
247 
248 #if defined(PMAP_MAP_POOLPAGE)
249 	while (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
250 		struct vm_page *pg;
251 		vaddr_t va;
252 #if defined(PMAP_ALLOC_POOLPAGE)
253 		pg = PMAP_ALLOC_POOLPAGE(0);
254 #else
255 		pg = uvm_pagealloc(NULL, 0, NULL, 0);
256 #endif
257 		if (pg == NULL) {
258 			uvm_wait("uarea");
259 			continue;
260 		}
261 		va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
262 		KASSERT(va != 0);
263 		return (void *)va;
264 	}
265 #endif
266 #if defined(__HAVE_CPU_UAREA_ROUTINES)
267 	void *va = cpu_uarea_alloc(false);
268 	if (va)
269 		return (void *)va;
270 #endif
271 	return (void *)uvm_km_alloc(kernel_map, pp->pr_alloc->pa_pagesz,
272 	    USPACE_ALIGN, UVM_KMF_WIRED | UVM_KMF_WAITVA);
273 }
274 
275 static void
276 uarea_poolpage_free(struct pool *pp, void *addr)
277 {
278 #if defined(PMAP_MAP_POOLPAGE)
279 	if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
280 		paddr_t pa;
281 
282 		pa = PMAP_UNMAP_POOLPAGE((vaddr_t) addr);
283 		KASSERT(pa != 0);
284 		uvm_pagefree(PHYS_TO_VM_PAGE(pa));
285 		return;
286 	}
287 #endif
288 #if defined(__HAVE_CPU_UAREA_ROUTINES)
289 	if (cpu_uarea_free(addr))
290 		return;
291 #endif
292 	uvm_km_free(kernel_map, (vaddr_t)addr, pp->pr_alloc->pa_pagesz,
293 	    UVM_KMF_WIRED);
294 }
295 
296 static struct pool_allocator uvm_uarea_allocator = {
297 	.pa_alloc = uarea_poolpage_alloc,
298 	.pa_free = uarea_poolpage_free,
299 	.pa_pagesz = USPACE,
300 };
301 
302 #if defined(__HAVE_CPU_UAREA_ROUTINES)
303 static void *
304 uarea_system_poolpage_alloc(struct pool *pp, int flags)
305 {
306 	void * const va = cpu_uarea_alloc(true);
307 	if (va != NULL)
308 		return va;
309 
310 	return (void *)uvm_km_alloc(kernel_map, pp->pr_alloc->pa_pagesz,
311 	    USPACE_ALIGN, UVM_KMF_WIRED |
312 	    ((flags & PR_WAITOK) ? UVM_KMF_WAITVA :
313 	    (UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)));
314 }
315 
316 static void
317 uarea_system_poolpage_free(struct pool *pp, void *addr)
318 {
319 	if (cpu_uarea_free(addr))
320 		return;
321 
322 	uvm_km_free(kernel_map, (vaddr_t)addr, pp->pr_alloc->pa_pagesz,
323 	    UVM_KMF_WIRED);
324 }
325 
326 static struct pool_allocator uvm_uarea_system_allocator = {
327 	.pa_alloc = uarea_system_poolpage_alloc,
328 	.pa_free = uarea_system_poolpage_free,
329 	.pa_pagesz = USPACE,
330 };
331 #endif /* __HAVE_CPU_UAREA_ROUTINES */
332 
333 void
334 uvm_uarea_init(void)
335 {
336 	int flags = PR_NOTOUCH;
337 
338 	/*
339 	 * specify PR_NOALIGN unless the alignment provided by
340 	 * the backend (USPACE_ALIGN) is sufficient to provide
341 	 * pool page size (UPSACE) alignment.
342 	 */
343 
344 	if ((USPACE_ALIGN == 0 && USPACE != PAGE_SIZE) ||
345 	    (USPACE_ALIGN % USPACE) != 0) {
346 		flags |= PR_NOALIGN;
347 	}
348 
349 	uvm_uarea_cache = pool_cache_init(USPACE, USPACE_ALIGN, 0, flags,
350 	    "uarea", &uvm_uarea_allocator, IPL_NONE, NULL, NULL, NULL);
351 #if defined(__HAVE_CPU_UAREA_ROUTINES)
352 	uvm_uarea_system_cache = pool_cache_init(USPACE, USPACE_ALIGN,
353 	    0, flags, "uareasys", &uvm_uarea_system_allocator,
354 	    IPL_NONE, NULL, NULL, NULL);
355 #endif
356 }
357 
358 /*
359  * uvm_uarea_alloc: allocate a u-area
360  */
361 
362 vaddr_t
363 uvm_uarea_alloc(void)
364 {
365 
366 	return (vaddr_t)pool_cache_get(uvm_uarea_cache, PR_WAITOK);
367 }
368 
369 vaddr_t
370 uvm_uarea_system_alloc(struct cpu_info *ci)
371 {
372 #ifdef __HAVE_CPU_UAREA_ALLOC_IDLELWP
373 	if (__predict_false(ci != NULL))
374 		return cpu_uarea_alloc_idlelwp(ci);
375 #endif
376 
377 	return (vaddr_t)pool_cache_get(uvm_uarea_system_cache, PR_WAITOK);
378 }
379 
380 /*
381  * uvm_uarea_free: free a u-area
382  */
383 
384 void
385 uvm_uarea_free(vaddr_t uaddr)
386 {
387 
388 	kasan_mark((void *)uaddr, USPACE, USPACE, 0);
389 	pool_cache_put(uvm_uarea_cache, (void *)uaddr);
390 }
391 
392 void
393 uvm_uarea_system_free(vaddr_t uaddr)
394 {
395 
396 	kasan_mark((void *)uaddr, USPACE, USPACE, 0);
397 	pool_cache_put(uvm_uarea_system_cache, (void *)uaddr);
398 }
399 
400 vaddr_t
401 uvm_lwp_getuarea(lwp_t *l)
402 {
403 
404 	return (vaddr_t)l->l_addr - UAREA_PCB_OFFSET;
405 }
406 
407 void
408 uvm_lwp_setuarea(lwp_t *l, vaddr_t addr)
409 {
410 
411 	l->l_addr = (void *)(addr + UAREA_PCB_OFFSET);
412 }
413 
414 /*
415  * uvm_proc_exit: exit a virtual address space
416  *
417  * - borrow proc0's address space because freeing the vmspace
418  *   of the dead process may block.
419  */
420 
421 void
422 uvm_proc_exit(struct proc *p)
423 {
424 	struct lwp *l = curlwp; /* XXX */
425 	struct vmspace *ovm;
426 
427 	KASSERT(p == l->l_proc);
428 	ovm = p->p_vmspace;
429 	KASSERT(ovm != NULL);
430 
431 	if (__predict_false(ovm == proc0.p_vmspace))
432 		return;
433 
434 	/*
435 	 * borrow proc0's address space.
436 	 */
437 	kpreempt_disable();
438 	pmap_deactivate(l);
439 	p->p_vmspace = proc0.p_vmspace;
440 	pmap_activate(l);
441 	kpreempt_enable();
442 
443 	uvmspace_free(ovm);
444 }
445 
446 void
447 uvm_lwp_exit(struct lwp *l)
448 {
449 	vaddr_t va = uvm_lwp_getuarea(l);
450 	bool system = (l->l_flag & LW_SYSTEM) != 0;
451 
452 	if (system)
453 		uvm_uarea_system_free(va);
454 	else
455 		uvm_uarea_free(va);
456 #ifdef DIAGNOSTIC
457 	uvm_lwp_setuarea(l, (vaddr_t)NULL);
458 #endif
459 }
460 
461 /*
462  * uvm_init_limit: init per-process VM limits
463  *
464  * - called for process 0 and then inherited by all others.
465  */
466 
467 void
468 uvm_init_limits(struct proc *p)
469 {
470 
471 	/*
472 	 * Set up the initial limits on process VM.  Set the maximum
473 	 * resident set size to be all of (reasonably) available memory.
474 	 * This causes any single, large process to start random page
475 	 * replacement once it fills memory.
476 	 */
477 
478 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
479 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
480 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
481 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
482 	p->p_rlimit[RLIMIT_AS].rlim_cur = RLIM_INFINITY;
483 	p->p_rlimit[RLIMIT_AS].rlim_max = RLIM_INFINITY;
484 	p->p_rlimit[RLIMIT_RSS].rlim_cur = MIN(
485 	    VM_MAXUSER_ADDRESS, ctob((rlim_t)uvmexp.free));
486 }
487 
488 /*
489  * uvm_scheduler: process zero main loop.
490  */
491 
492 extern struct loadavg averunnable;
493 
494 void
495 uvm_scheduler(void)
496 {
497 	lwp_t *l = curlwp;
498 
499 	lwp_lock(l);
500 	l->l_priority = PRI_VM;
501 	l->l_class = SCHED_FIFO;
502 	lwp_unlock(l);
503 
504 	for (;;) {
505 		sched_pstats();
506 		(void)kpause("uvm", false, hz, NULL);
507 	}
508 }
509