xref: /dragonfly/sys/vm/vm_zone.c (revision 86fe9e07) 
1 /*
2  * Copyright (c) 1997, 1998 John S. Dyson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *	notice immediately at the beginning of the file, without modification,
10  *	this list of conditions, and the following disclaimer.
11  * 2. Absolutely no warranty of function or purpose is made by the author
12  *	John S. Dyson.
13  *
14  * $FreeBSD: src/sys/vm/vm_zone.c,v 1.30.2.6 2002/10/10 19:50:16 dillon Exp $
15  * $DragonFly: src/sys/vm/vm_zone.c,v 1.15 2004/05/11 18:05:08 dillon Exp $
16  */
17 
18 #include <sys/param.h>
19 #include <sys/systm.h>
20 #include <sys/kernel.h>
21 #include <sys/lock.h>
22 #include <sys/malloc.h>
23 #include <sys/sysctl.h>
24 #include <sys/vmmeter.h>
25 
26 #include <vm/vm.h>
27 #include <vm/vm_object.h>
28 #include <vm/vm_page.h>
29 #include <vm/vm_map.h>
30 #include <vm/vm_kern.h>
31 #include <vm/vm_extern.h>
32 #include <vm/vm_zone.h>
33 
34 static MALLOC_DEFINE(M_ZONE, "ZONE", "Zone header");
35 
36 #define	ZONE_ERROR_INVALID 0
37 #define	ZONE_ERROR_NOTFREE 1
38 #define	ZONE_ERROR_ALREADYFREE 2
39 
40 #define	ZONE_ROUNDING	32
41 
42 #define	ZENTRY_FREE	0x12342378
43 
44 static void *zget(vm_zone_t z);
45 
46 /*
47  * Return an item from the specified zone.   This function is interrupt/MP
48  * thread safe, but might block.
49  */
50 void *
51 zalloc(vm_zone_t z)
52 {
53 	void *item;
54 	lwkt_tokref ilock;
55 
56 #ifdef INVARIANTS
57 	if (z == NULL)
58 		zerror(ZONE_ERROR_INVALID);
59 #endif
60 	lwkt_gettoken(&ilock, &z->zlock);
61 	if (z->zfreecnt <= z->zfreemin) {
62 		item = zget(z);
63 		/*
64 		 * PANICFAIL allows the caller to assume that the zalloc()
65 		 * will always succeed.  If it doesn't, we panic here.
66 		 */
67 		if (item == NULL && (z->zflags & ZONE_PANICFAIL))
68 			panic("zalloc(%s) failed", z->zname);
69 	} else {
70 		item = z->zitems;
71 #ifdef INVARIANTS
72 		KASSERT(item != NULL, ("zitems unexpectedly NULL"));
73 		if (((void **) item)[1] != (void *) ZENTRY_FREE)
74 			zerror(ZONE_ERROR_NOTFREE);
75 		((void **) item)[1] = 0;
76 #endif
77 		z->zitems = ((void **) item)[0];
78 		z->zfreecnt--;
79 		z->znalloc++;
80 	}
81 	lwkt_reltoken(&ilock);
82 	return item;
83 }
84 
85 /*
86  * Free an item to the specified zone.   This function is interrupt/MP
87  * thread safe, but might block.
88  */
89 void
90 zfree(vm_zone_t z, void *item)
91 {
92 	lwkt_tokref ilock;
93 
94 	lwkt_gettoken(&ilock, &z->zlock);
95 	((void **) item)[0] = z->zitems;
96 #ifdef INVARIANTS
97 	if (((void **) item)[1] == (void *) ZENTRY_FREE)
98 		zerror(ZONE_ERROR_ALREADYFREE);
99 	((void **) item)[1] = (void *) ZENTRY_FREE;
100 #endif
101 	z->zitems = item;
102 	z->zfreecnt++;
103 	lwkt_reltoken(&ilock);
104 }
105 
106 /*
107  * This file comprises a very simple zone allocator.  This is used
108  * in lieu of the malloc allocator, where needed or more optimal.
109  *
110  * Note that the initial implementation of this had coloring, and
111  * absolutely no improvement (actually perf degradation) occurred.
112  *
113  * Note also that the zones are type stable.  The only restriction is
114  * that the first two longwords of a data structure can be changed
115  * between allocations.  Any data that must be stable between allocations
116  * must reside in areas after the first two longwords.
117  *
118  * zinitna, zinit, zbootinit are the initialization routines.
119  * zalloc, zfree, are the allocation/free routines.
120  */
121 
122 static struct vm_zone *zlist;
123 static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS);
124 static int zone_kmem_pages, zone_kern_pages, zone_kmem_kvaspace;
125 
126 /*
127  * Create a zone, but don't allocate the zone structure.  If the
128  * zone had been previously created by the zone boot code, initialize
129  * various parts of the zone code.
130  *
131  * If waits are not allowed during allocation (e.g. during interrupt
132  * code), a-priori allocate the kernel virtual space, and allocate
133  * only pages when needed.
134  *
135  * Arguments:
136  * z		pointer to zone structure.
137  * obj		pointer to VM object (opt).
138  * name		name of zone.
139  * size		size of zone entries.
140  * nentries	number of zone entries allocated (only ZONE_INTERRUPT.)
141  * flags	ZONE_INTERRUPT -- items can be allocated at interrupt time.
142  * zalloc	number of pages allocated when memory is needed.
143  *
144  * Note that when using ZONE_INTERRUPT, the size of the zone is limited
145  * by the nentries argument.  The size of the memory allocatable is
146  * unlimited if ZONE_INTERRUPT is not set.
147  *
148  */
149 int
150 zinitna(vm_zone_t z, vm_object_t obj, char *name, int size,
151 	int nentries, int flags, int zalloc)
152 {
153 	int totsize;
154 
155 	if ((z->zflags & ZONE_BOOT) == 0) {
156 		z->zsize = (size + ZONE_ROUNDING - 1) & ~(ZONE_ROUNDING - 1);
157 		lwkt_token_init(&z->zlock);
158 		z->zfreecnt = 0;
159 		z->ztotal = 0;
160 		z->zmax = 0;
161 		z->zname = name;
162 		z->znalloc = 0;
163 		z->zitems = NULL;
164 
165 		z->znext = zlist;
166 		zlist = z;
167 	}
168 
169 	z->zflags |= flags;
170 
171 	/*
172 	 * If we cannot wait, allocate KVA space up front, and we will fill
173 	 * in pages as needed.  This is particularly required when creating
174 	 * an allocation space for map entries in kernel_map, because we
175 	 * do not want to go into a recursion deadlock with
176 	 * vm_map_entry_reserve().
177 	 */
178 	if (z->zflags & ZONE_INTERRUPT) {
179 
180 		totsize = round_page(z->zsize * nentries);
181 		zone_kmem_kvaspace += totsize;
182 
183 		z->zkva = kmem_alloc_pageable(kernel_map, totsize);
184 		if (z->zkva == 0) {
185 			zlist = z->znext;
186 			return 0;
187 		}
188 
189 		z->zpagemax = totsize / PAGE_SIZE;
190 		if (obj == NULL) {
191 			z->zobj = vm_object_allocate(OBJT_DEFAULT, z->zpagemax);
192 		} else {
193 			z->zobj = obj;
194 			_vm_object_allocate(OBJT_DEFAULT, z->zpagemax, obj);
195 		}
196 		z->zallocflag = VM_ALLOC_SYSTEM | VM_ALLOC_INTERRUPT;
197 		z->zmax += nentries;
198 	} else {
199 		z->zallocflag = VM_ALLOC_NORMAL | VM_ALLOC_SYSTEM;
200 		z->zmax = 0;
201 	}
202 
203 
204 	if (z->zsize > PAGE_SIZE)
205 		z->zfreemin = 1;
206 	else
207 		z->zfreemin = PAGE_SIZE / z->zsize;
208 
209 	z->zpagecount = 0;
210 	if (zalloc)
211 		z->zalloc = zalloc;
212 	else
213 		z->zalloc = 1;
214 
215 	return 1;
216 }
217 
218 /*
219  * Subroutine same as zinitna, except zone data structure is allocated
220  * automatically by malloc.  This routine should normally be used, except
221  * in certain tricky startup conditions in the VM system -- then
222  * zbootinit and zinitna can be used.  Zinit is the standard zone
223  * initialization call.
224  */
225 vm_zone_t
226 zinit(char *name, int size, int nentries, int flags, int zalloc)
227 {
228 	vm_zone_t z;
229 
230 	z = (vm_zone_t) malloc(sizeof (struct vm_zone), M_ZONE, M_NOWAIT);
231 	if (z == NULL)
232 		return NULL;
233 
234 	z->zflags = 0;
235 	if (zinitna(z, NULL, name, size, nentries, flags, zalloc) == 0) {
236 		free(z, M_ZONE);
237 		return NULL;
238 	}
239 
240 	return z;
241 }
242 
243 /*
244  * Initialize a zone before the system is fully up.  This routine should
245  * only be called before full VM startup.
246  */
247 void
248 zbootinit(vm_zone_t z, char *name, int size, void *item, int nitems)
249 {
250 	int i;
251 
252 	z->zname = name;
253 	z->zsize = size;
254 	z->zpagemax = 0;
255 	z->zobj = NULL;
256 	z->zflags = ZONE_BOOT;
257 	z->zfreemin = 0;
258 	z->zallocflag = 0;
259 	z->zpagecount = 0;
260 	z->zalloc = 0;
261 	z->znalloc = 0;
262 	lwkt_token_init(&z->zlock);
263 
264 	bzero(item, nitems * z->zsize);
265 	z->zitems = NULL;
266 	for (i = 0; i < nitems; i++) {
267 		((void **) item)[0] = z->zitems;
268 #ifdef INVARIANTS
269 		((void **) item)[1] = (void *) ZENTRY_FREE;
270 #endif
271 		z->zitems = item;
272 		(char *) item += z->zsize;
273 	}
274 	z->zfreecnt = nitems;
275 	z->zmax = nitems;
276 	z->ztotal = nitems;
277 
278 	if (zlist == 0) {
279 		zlist = z;
280 	} else {
281 		z->znext = zlist;
282 		zlist = z;
283 	}
284 }
285 
286 /*
287  * void *zalloc(vm_zone_t zone) --
288  *	Returns an item from a specified zone.  May not be called from a
289  *	FAST interrupt or IPI function.
290  *
291  * void zfree(vm_zone_t zone, void *item) --
292  *	Frees an item back to a specified zone.  May not be called from a
293  *	FAST interrupt or IPI function.
294  */
295 
296 /*
297  * Internal zone routine.  Not to be called from external (non vm_zone) code.
298  */
299 static void *
300 zget(vm_zone_t z)
301 {
302 	int i;
303 	vm_page_t m;
304 	int nitems, nbytes;
305 	void *item;
306 
307 	if (z == NULL)
308 		panic("zget: null zone");
309 
310 	if (z->zflags & ZONE_INTERRUPT) {
311 		nbytes = z->zpagecount * PAGE_SIZE;
312 		nbytes -= nbytes % z->zsize;
313 		item = (char *) z->zkva + nbytes;
314 		for (i = 0; ((i < z->zalloc) && (z->zpagecount < z->zpagemax));
315 		     i++) {
316 			vm_offset_t zkva;
317 
318 			m = vm_page_alloc(z->zobj, z->zpagecount,
319 					  z->zallocflag);
320 			/* note: z might be modified due to blocking */
321 			if (m == NULL)
322 				break;
323 
324 			zkva = z->zkva + z->zpagecount * PAGE_SIZE;
325 			pmap_kenter(zkva, VM_PAGE_TO_PHYS(m)); /* YYY */
326 			bzero((caddr_t) zkva, PAGE_SIZE);
327 			z->zpagecount++;
328 			zone_kmem_pages++;
329 			vmstats.v_wire_count++;
330 		}
331 		nitems = ((z->zpagecount * PAGE_SIZE) - nbytes) / z->zsize;
332 	} else {
333 		nbytes = z->zalloc * PAGE_SIZE;
334 
335 		{
336 			item = (void *)kmem_alloc3(kernel_map, nbytes, KM_KRESERVE);
337 			/* note: z might be modified due to blocking */
338 			if (item != NULL)
339 				zone_kern_pages += z->zalloc;
340 		}
341 		if (item != NULL) {
342 			bzero(item, nbytes);
343 		} else {
344 			nbytes = 0;
345 		}
346 		nitems = nbytes / z->zsize;
347 	}
348 	z->ztotal += nitems;
349 
350 	/*
351 	 * Save one for immediate allocation
352 	 */
353 	if (nitems != 0) {
354 		nitems -= 1;
355 		for (i = 0; i < nitems; i++) {
356 			((void **) item)[0] = z->zitems;
357 #ifdef INVARIANTS
358 			((void **) item)[1] = (void *) ZENTRY_FREE;
359 #endif
360 			z->zitems = item;
361 			(char *) item += z->zsize;
362 		}
363 		z->zfreecnt += nitems;
364 		z->znalloc++;
365 	} else if (z->zfreecnt > 0) {
366 		item = z->zitems;
367 		z->zitems = ((void **) item)[0];
368 #ifdef INVARIANTS
369 		if (((void **) item)[1] != (void *) ZENTRY_FREE)
370 			zerror(ZONE_ERROR_NOTFREE);
371 		((void **) item)[1] = 0;
372 #endif
373 		z->zfreecnt--;
374 		z->znalloc++;
375 	} else {
376 		item = NULL;
377 	}
378 
379 	/*
380 	 * Recover any reserve missing due to a zalloc/kreserve/krelease
381 	 * recursion.
382 	 */
383 	vm_map_entry_reserve(0);
384 
385 	return item;
386 }
387 
388 static int
389 sysctl_vm_zone(SYSCTL_HANDLER_ARGS)
390 {
391 	int error=0;
392 	vm_zone_t curzone, nextzone;
393 	char tmpbuf[128];
394 	char tmpname[14];
395 
396 	snprintf(tmpbuf, sizeof(tmpbuf),
397 	    "\nITEM            SIZE     LIMIT    USED    FREE  REQUESTS\n");
398 	error = SYSCTL_OUT(req, tmpbuf, strlen(tmpbuf));
399 	if (error)
400 		return (error);
401 
402 	for (curzone = zlist; curzone; curzone = nextzone) {
403 		int i;
404 		int len;
405 		int offset;
406 
407 		nextzone = curzone->znext;
408 		len = strlen(curzone->zname);
409 		if (len >= (sizeof(tmpname) - 1))
410 			len = (sizeof(tmpname) - 1);
411 		for(i = 0; i < sizeof(tmpname) - 1; i++)
412 			tmpname[i] = ' ';
413 		tmpname[i] = 0;
414 		memcpy(tmpname, curzone->zname, len);
415 		tmpname[len] = ':';
416 		offset = 0;
417 		if (curzone == zlist) {
418 			offset = 1;
419 			tmpbuf[0] = '\n';
420 		}
421 
422 		snprintf(tmpbuf + offset, sizeof(tmpbuf) - offset,
423 			"%s %6.6u, %8.8u, %6.6u, %6.6u, %8.8u\n",
424 			tmpname, curzone->zsize, curzone->zmax,
425 			(curzone->ztotal - curzone->zfreecnt),
426 			curzone->zfreecnt, curzone->znalloc);
427 
428 		len = strlen((char *)tmpbuf);
429 		if (nextzone == NULL)
430 			tmpbuf[len - 1] = 0;
431 
432 		error = SYSCTL_OUT(req, tmpbuf, len);
433 
434 		if (error)
435 			return (error);
436 	}
437 	return (0);
438 }
439 
440 #if defined(INVARIANTS)
441 void
442 zerror(int error)
443 {
444 	char *msg;
445 
446 	switch (error) {
447 	case ZONE_ERROR_INVALID:
448 		msg = "zone: invalid zone";
449 		break;
450 	case ZONE_ERROR_NOTFREE:
451 		msg = "zone: entry not free";
452 		break;
453 	case ZONE_ERROR_ALREADYFREE:
454 		msg = "zone: freeing free entry";
455 		break;
456 	default:
457 		msg = "zone: invalid error";
458 		break;
459 	}
460 	panic(msg);
461 }
462 #endif
463 
464 SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD, \
465 	NULL, 0, sysctl_vm_zone, "A", "Zone Info");
466 
467 SYSCTL_INT(_vm, OID_AUTO, zone_kmem_pages,
468 	CTLFLAG_RD, &zone_kmem_pages, 0, "Number of interrupt safe pages allocated by zone");
469 SYSCTL_INT(_vm, OID_AUTO, zone_kmem_kvaspace,
470 	CTLFLAG_RD, &zone_kmem_kvaspace, 0, "KVA space allocated by zone");
471 SYSCTL_INT(_vm, OID_AUTO, zone_kern_pages,
472 	CTLFLAG_RD, &zone_kern_pages, 0, "Number of non-interrupt safe pages allocated by zone");
473