xref: /dflybsd-src/sys/vm/vm_zone.c (revision ac2e3f5effc58aa364c7e5c199f35ebbae7cda81)
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 1.7 2003/07/29 21:24:33 hmp 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 
55 #ifdef INVARIANTS
56 	if (z == NULL)
57 		zerror(ZONE_ERROR_INVALID);
58 #endif
59 	lwkt_gettoken(&z->zlock);
60 	if (z->zfreecnt <= z->zfreemin) {
61 		item = zget(z);
62 		/*
63 		 * PANICFAIL allows the caller to assume that the zalloc()
64 		 * will always succeed.  If it doesn't, we panic here.
65 		 */
66 		if (item == NULL && (z->zflags & ZONE_PANICFAIL))
67 			panic("zalloc(%s) failed", z->zname);
68 	} else {
69 		item = z->zitems;
70 		z->zitems = ((void **) item)[0];
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->zfreecnt--;
78 		z->znalloc++;
79 	}
80 	lwkt_reltoken(&z->zlock);
81 	return item;
82 }
83 
84 /*
85  * Free an item to the specified zone.   This function is interrupt/MP
86  * thread safe, but might block.
87  */
88 void
89 zfree(vm_zone_t z, void *item)
90 {
91 	lwkt_gettoken(&z->zlock);
92 	((void **) item)[0] = z->zitems;
93 #ifdef INVARIANTS
94 	if (((void **) item)[1] == (void *) ZENTRY_FREE)
95 		zerror(ZONE_ERROR_ALREADYFREE);
96 	((void **) item)[1] = (void *) ZENTRY_FREE;
97 #endif
98 	z->zitems = item;
99 	z->zfreecnt++;
100 	lwkt_reltoken(&z->zlock);
101 }
102 
103 /*
104  * This file comprises a very simple zone allocator.  This is used
105  * in lieu of the malloc allocator, where needed or more optimal.
106  *
107  * Note that the initial implementation of this had coloring, and
108  * absolutely no improvement (actually perf degradation) occurred.
109  *
110  * Note also that the zones are type stable.  The only restriction is
111  * that the first two longwords of a data structure can be changed
112  * between allocations.  Any data that must be stable between allocations
113  * must reside in areas after the first two longwords.
114  *
115  * zinitna, zinit, zbootinit are the initialization routines.
116  * zalloc, zfree, are the interrupt/lock unsafe allocation/free routines.
117  * zalloci, zfreei, are the interrupt/lock safe allocation/free routines.
118  */
119 
120 static struct vm_zone *zlist;
121 static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS);
122 static int zone_kmem_pages, zone_kern_pages, zone_kmem_kvaspace;
123 
124 /*
125  * Create a zone, but don't allocate the zone structure.  If the
126  * zone had been previously created by the zone boot code, initialize
127  * various parts of the zone code.
128  *
129  * If waits are not allowed during allocation (e.g. during interrupt
130  * code), a-priori allocate the kernel virtual space, and allocate
131  * only pages when needed.
132  *
133  * Arguments:
134  * z		pointer to zone structure.
135  * obj		pointer to VM object (opt).
136  * name		name of zone.
137  * size		size of zone entries.
138  * nentries	number of zone entries allocated (only ZONE_INTERRUPT.)
139  * flags	ZONE_INTERRUPT -- items can be allocated at interrupt time.
140  * zalloc	number of pages allocated when memory is needed.
141  *
142  * Note that when using ZONE_INTERRUPT, the size of the zone is limited
143  * by the nentries argument.  The size of the memory allocatable is
144  * unlimited if ZONE_INTERRUPT is not set.
145  *
146  */
147 int
148 zinitna(vm_zone_t z, vm_object_t obj, char *name, int size,
149 	int nentries, int flags, int zalloc)
150 {
151 	int totsize;
152 
153 	if ((z->zflags & ZONE_BOOT) == 0) {
154 		z->zsize = (size + ZONE_ROUNDING - 1) & ~(ZONE_ROUNDING - 1);
155 		lwkt_inittoken(&z->zlock);
156 		z->zfreecnt = 0;
157 		z->ztotal = 0;
158 		z->zmax = 0;
159 		z->zname = name;
160 		z->znalloc = 0;
161 		z->zitems = NULL;
162 
163 		z->znext = zlist;
164 		zlist = z;
165 	}
166 
167 	z->zflags |= flags;
168 
169 	/*
170 	 * If we cannot wait, allocate KVA space up front, and we will fill
171 	 * in pages as needed.
172 	 */
173 	if (z->zflags & ZONE_INTERRUPT) {
174 
175 		totsize = round_page(z->zsize * nentries);
176 		zone_kmem_kvaspace += totsize;
177 
178 		z->zkva = kmem_alloc_pageable(kernel_map, totsize);
179 		if (z->zkva == 0) {
180 			zlist = z->znext;
181 			return 0;
182 		}
183 
184 		z->zpagemax = totsize / PAGE_SIZE;
185 		if (obj == NULL) {
186 			z->zobj = vm_object_allocate(OBJT_DEFAULT, z->zpagemax);
187 		} else {
188 			z->zobj = obj;
189 			_vm_object_allocate(OBJT_DEFAULT, z->zpagemax, obj);
190 		}
191 		z->zallocflag = VM_ALLOC_INTERRUPT;
192 		z->zmax += nentries;
193 	} else {
194 		z->zallocflag = VM_ALLOC_SYSTEM;
195 		z->zmax = 0;
196 	}
197 
198 
199 	if (z->zsize > PAGE_SIZE)
200 		z->zfreemin = 1;
201 	else
202 		z->zfreemin = PAGE_SIZE / z->zsize;
203 
204 	z->zpagecount = 0;
205 	if (zalloc)
206 		z->zalloc = zalloc;
207 	else
208 		z->zalloc = 1;
209 
210 	return 1;
211 }
212 
213 /*
214  * Subroutine same as zinitna, except zone data structure is allocated
215  * automatically by malloc.  This routine should normally be used, except
216  * in certain tricky startup conditions in the VM system -- then
217  * zbootinit and zinitna can be used.  Zinit is the standard zone
218  * initialization call.
219  */
220 vm_zone_t
221 zinit(char *name, int size, int nentries, int flags, int zalloc)
222 {
223 	vm_zone_t z;
224 
225 	z = (vm_zone_t) malloc(sizeof (struct vm_zone), M_ZONE, M_NOWAIT);
226 	if (z == NULL)
227 		return NULL;
228 
229 	z->zflags = 0;
230 	if (zinitna(z, NULL, name, size, nentries, flags, zalloc) == 0) {
231 		free(z, M_ZONE);
232 		return NULL;
233 	}
234 
235 	return z;
236 }
237 
238 /*
239  * Initialize a zone before the system is fully up.  This routine should
240  * only be called before full VM startup.
241  */
242 void
243 zbootinit(vm_zone_t z, char *name, int size, void *item, int nitems)
244 {
245 	int i;
246 
247 	z->zname = name;
248 	z->zsize = size;
249 	z->zpagemax = 0;
250 	z->zobj = NULL;
251 	z->zflags = ZONE_BOOT;
252 	z->zfreemin = 0;
253 	z->zallocflag = 0;
254 	z->zpagecount = 0;
255 	z->zalloc = 0;
256 	z->znalloc = 0;
257 	lwkt_inittoken(&z->zlock);
258 
259 	bzero(item, nitems * z->zsize);
260 	z->zitems = NULL;
261 	for (i = 0; i < nitems; i++) {
262 		((void **) item)[0] = z->zitems;
263 #ifdef INVARIANTS
264 		((void **) item)[1] = (void *) ZENTRY_FREE;
265 #endif
266 		z->zitems = item;
267 		(char *) item += z->zsize;
268 	}
269 	z->zfreecnt = nitems;
270 	z->zmax = nitems;
271 	z->ztotal = nitems;
272 
273 	if (zlist == 0) {
274 		zlist = z;
275 	} else {
276 		z->znext = zlist;
277 		zlist = z;
278 	}
279 }
280 
281 /*
282  * void *zalloc(vm_zone_t zone) --
283  *	Returns an item from a specified zone.
284  *
285  * void zfree(vm_zone_t zone, void *item) --
286  *  Frees an item back to a specified zone.
287  *
288  * void *zalloci(vm_zone_t zone) --
289  *	Returns an item from a specified zone, interrupt safe.
290  *
291  * void zfreei(vm_zone_t zone, void *item) --
292  *  Frees an item back to a specified zone, interrupt safe.
293  *
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 			if (m == NULL)
321 				break;
322 			lwkt_regettoken(&z->zlock);
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 		 * Check to see if the kernel map is already locked.
337 		 * We could allow for recursive locks, but that eliminates
338 		 * a valuable debugging mechanism, and opens up the kernel
339 		 * map for potential corruption by inconsistent data structure
340 		 * manipulation.  We could also use the interrupt allocation
341 		 * mechanism, but that has size limitations.   Luckily, we
342 		 * have kmem_map that is a submap of kernel map available
343 		 * for memory allocation, and manipulation of that map doesn't
344 		 * affect the kernel map structures themselves.
345 		 *
346 		 * We can wait, so just do normal map allocation in the
347 		 * appropriate map.
348 		 */
349 		if (lockstatus(&kernel_map->lock, NULL)) {
350 			int s;
351 			s = splvm();
352 			item = (void *) kmem_malloc(kmem_map, nbytes, M_WAITOK);
353 			lwkt_regettoken(&z->zlock);
354 			if (item != NULL)
355 				zone_kmem_pages += z->zalloc;
356 			splx(s);
357 		} else {
358 			item = (void *) kmem_alloc(kernel_map, nbytes);
359 			lwkt_regettoken(&z->zlock);
360 			if (item != NULL)
361 				zone_kern_pages += z->zalloc;
362 		}
363 		if (item != NULL) {
364 			bzero(item, nbytes);
365 		} else {
366 			nbytes = 0;
367 		}
368 		nitems = nbytes / z->zsize;
369 	}
370 	z->ztotal += nitems;
371 
372 	/*
373 	 * Save one for immediate allocation
374 	 */
375 	if (nitems != 0) {
376 		nitems -= 1;
377 		for (i = 0; i < nitems; i++) {
378 			((void **) item)[0] = z->zitems;
379 #ifdef INVARIANTS
380 			((void **) item)[1] = (void *) ZENTRY_FREE;
381 #endif
382 			z->zitems = item;
383 			(char *) item += z->zsize;
384 		}
385 		z->zfreecnt += nitems;
386 		z->znalloc++;
387 	} else if (z->zfreecnt > 0) {
388 		item = z->zitems;
389 		z->zitems = ((void **) item)[0];
390 #ifdef INVARIANTS
391 		if (((void **) item)[1] != (void *) ZENTRY_FREE)
392 			zerror(ZONE_ERROR_NOTFREE);
393 		((void **) item)[1] = 0;
394 #endif
395 		z->zfreecnt--;
396 		z->znalloc++;
397 	} else {
398 		item = NULL;
399 	}
400 
401 	return item;
402 }
403 
404 static int
405 sysctl_vm_zone(SYSCTL_HANDLER_ARGS)
406 {
407 	int error=0;
408 	vm_zone_t curzone, nextzone;
409 	char tmpbuf[128];
410 	char tmpname[14];
411 
412 	snprintf(tmpbuf, sizeof(tmpbuf),
413 	    "\nITEM            SIZE     LIMIT    USED    FREE  REQUESTS\n");
414 	error = SYSCTL_OUT(req, tmpbuf, strlen(tmpbuf));
415 	if (error)
416 		return (error);
417 
418 	for (curzone = zlist; curzone; curzone = nextzone) {
419 		int i;
420 		int len;
421 		int offset;
422 
423 		nextzone = curzone->znext;
424 		len = strlen(curzone->zname);
425 		if (len >= (sizeof(tmpname) - 1))
426 			len = (sizeof(tmpname) - 1);
427 		for(i = 0; i < sizeof(tmpname) - 1; i++)
428 			tmpname[i] = ' ';
429 		tmpname[i] = 0;
430 		memcpy(tmpname, curzone->zname, len);
431 		tmpname[len] = ':';
432 		offset = 0;
433 		if (curzone == zlist) {
434 			offset = 1;
435 			tmpbuf[0] = '\n';
436 		}
437 
438 		snprintf(tmpbuf + offset, sizeof(tmpbuf) - offset,
439 			"%s %6.6u, %8.8u, %6.6u, %6.6u, %8.8u\n",
440 			tmpname, curzone->zsize, curzone->zmax,
441 			(curzone->ztotal - curzone->zfreecnt),
442 			curzone->zfreecnt, curzone->znalloc);
443 
444 		len = strlen((char *)tmpbuf);
445 		if (nextzone == NULL)
446 			tmpbuf[len - 1] = 0;
447 
448 		error = SYSCTL_OUT(req, tmpbuf, len);
449 
450 		if (error)
451 			return (error);
452 	}
453 	return (0);
454 }
455 
456 #if defined(INVARIANTS)
457 void
458 zerror(int error)
459 {
460 	char *msg;
461 
462 	switch (error) {
463 	case ZONE_ERROR_INVALID:
464 		msg = "zone: invalid zone";
465 		break;
466 	case ZONE_ERROR_NOTFREE:
467 		msg = "zone: entry not free";
468 		break;
469 	case ZONE_ERROR_ALREADYFREE:
470 		msg = "zone: freeing free entry";
471 		break;
472 	default:
473 		msg = "zone: invalid error";
474 		break;
475 	}
476 	panic(msg);
477 }
478 #endif
479 
480 SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD, \
481 	NULL, 0, sysctl_vm_zone, "A", "Zone Info");
482 
483 SYSCTL_INT(_vm, OID_AUTO, zone_kmem_pages,
484 	CTLFLAG_RD, &zone_kmem_pages, 0, "Number of interrupt safe pages allocated by zone");
485 SYSCTL_INT(_vm, OID_AUTO, zone_kmem_kvaspace,
486 	CTLFLAG_RD, &zone_kmem_kvaspace, 0, "KVA space allocated by zone");
487 SYSCTL_INT(_vm, OID_AUTO, zone_kern_pages,
488 	CTLFLAG_RD, &zone_kern_pages, 0, "Number of non-interrupt safe pages allocated by zone");
489