FreeBSD kernel kern code
subr_physmem.c
Go to the documentation of this file.
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2014 Ian Lepore <ian@freebsd.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include "opt_acpi.h"
33#include "opt_ddb.h"
34
35/*
36 * Routines for describing and initializing anything related to physical memory.
37 */
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/physmem.h>
43#include <vm/vm.h>
44#include <vm/vm_param.h>
45#include <vm/vm_page.h>
46#include <vm/vm_phys.h>
47#include <vm/vm_dumpset.h>
48#include <machine/md_var.h>
49
50/*
51 * These structures are used internally to keep track of regions of physical
52 * ram, and regions within the physical ram that need to be excluded. An
53 * exclusion region can be excluded from crash dumps, from the vm pool of pages
54 * that can be allocated, or both, depending on the exclusion flags associated
55 * with the region.
56 */
57#ifdef DEV_ACPI
58#define MAX_HWCNT 32 /* ACPI needs more regions */
59#define MAX_EXCNT 32
60#else
61#define MAX_HWCNT 16
62#define MAX_EXCNT 16
63#endif
64
65#if defined(__arm__)
66#define MAX_PHYS_ADDR 0xFFFFFFFFull
67#elif defined(__aarch64__) || defined(__riscv)
68#define MAX_PHYS_ADDR 0xFFFFFFFFFFFFFFFFull
69#endif
70
71struct region {
72 vm_paddr_t addr;
73 vm_size_t size;
74 uint32_t flags;
75};
76
77static struct region hwregions[MAX_HWCNT];
78static struct region exregions[MAX_EXCNT];
79
80static size_t hwcnt;
81static size_t excnt;
82
83/*
84 * realmem is the total number of hardware pages, excluded or not.
85 * Maxmem is one greater than the last physical page number.
86 */
88long Maxmem;
89
90/*
91 * Print the contents of the physical and excluded region tables using the
92 * provided printf-like output function (which will be either printf or
93 * db_printf).
94 */
95static void
96physmem_dump_tables(int (*prfunc)(const char *, ...))
97{
98 int flags, i;
99 uintmax_t addr, size;
100 const unsigned int mbyte = 1024 * 1024;
101
102 prfunc("Physical memory chunk(s):\n");
103 for (i = 0; i < hwcnt; ++i) {
104 addr = hwregions[i].addr;
105 size = hwregions[i].size;
106 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages)\n", addr,
107 addr + size - 1, size / mbyte, size / PAGE_SIZE);
108 }
109
110 prfunc("Excluded memory regions:\n");
111 for (i = 0; i < excnt; ++i) {
112 addr = exregions[i].addr;
113 size = exregions[i].size;
114 flags = exregions[i].flags;
115 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages) %s %s\n",
116 addr, addr + size - 1, size / mbyte, size / PAGE_SIZE,
117 (flags & EXFLAG_NOALLOC) ? "NoAlloc" : "",
118 (flags & EXFLAG_NODUMP) ? "NoDump" : "");
119 }
120
121#ifdef DEBUG
122 prfunc("Avail lists:\n");
123 for (i = 0; phys_avail[i] != 0; ++i) {
124 prfunc(" phys_avail[%d] 0x%08x\n", i, phys_avail[i]);
125 }
126 for (i = 0; dump_avail[i] != 0; ++i) {
127 prfunc(" dump_avail[%d] 0x%08x\n", i, dump_avail[i]);
128 }
129#endif
130}
131
132/*
133 * Print the contents of the static mapping table. Used for bootverbose.
134 */
135void
137{
138
140}
141
142/*
143 * Walk the list of hardware regions, processing it against the list of
144 * exclusions that contain the given exflags, and generating an "avail list".
145 *
146 * If maxphyssz is not zero it sets upper limit, in bytes, for the total
147 * "avail list" size. Walk stops once the limit is reached and the last region
148 * is cut short if necessary.
149 *
150 * Updates the value at *pavail with the sum of all pages in all hw regions.
151 *
152 * Returns the number of pages of non-excluded memory added to the avail list.
153 */
154static size_t
155regions_to_avail(vm_paddr_t *avail, uint32_t exflags, size_t maxavail,
156 uint64_t maxphyssz, long *pavail, long *prealmem)
157{
158 size_t acnt, exi, hwi;
159 uint64_t end, start, xend, xstart;
160 long availmem, totalmem;
161 const struct region *exp, *hwp;
162 uint64_t availsz;
163
164 totalmem = 0;
165 availmem = 0;
166 availsz = 0;
167 acnt = 0;
168 for (hwi = 0, hwp = hwregions; hwi < hwcnt; ++hwi, ++hwp) {
169 start = hwp->addr;
170 end = hwp->size + start;
171 totalmem += atop((vm_offset_t)(end - start));
172 for (exi = 0, exp = exregions; exi < excnt; ++exi, ++exp) {
173 /*
174 * If the excluded region does not match given flags,
175 * continue checking with the next excluded region.
176 */
177 if ((exp->flags & exflags) == 0)
178 continue;
179 xstart = exp->addr;
180 xend = exp->size + xstart;
181 /*
182 * If the excluded region ends before this hw region,
183 * continue checking with the next excluded region.
184 */
185 if (xend <= start)
186 continue;
187 /*
188 * If the excluded region begins after this hw region
189 * we're done because both lists are sorted.
190 */
191 if (xstart >= end)
192 break;
193 /*
194 * If the excluded region completely covers this hw
195 * region, shrink this hw region to zero size.
196 */
197 if ((start >= xstart) && (end <= xend)) {
198 start = xend;
199 end = xend;
200 break;
201 }
202 /*
203 * If the excluded region falls wholly within this hw
204 * region without abutting or overlapping the beginning
205 * or end, create an available entry from the leading
206 * fragment, then adjust the start of this hw region to
207 * the end of the excluded region, and continue checking
208 * the next excluded region because another exclusion
209 * could affect the remainder of this hw region.
210 */
211 if ((xstart > start) && (xend < end)) {
212
213 if ((maxphyssz != 0) &&
214 (availsz + xstart - start > maxphyssz)) {
215 xstart = maxphyssz + start - availsz;
216 }
217 if (xstart <= start)
218 continue;
219 if (acnt > 0 &&
220 avail[acnt - 1] == (vm_paddr_t)start) {
221 avail[acnt - 1] = (vm_paddr_t)xstart;
222 } else {
223 avail[acnt++] = (vm_paddr_t)start;
224 avail[acnt++] = (vm_paddr_t)xstart;
225 }
226 availsz += (xstart - start);
227 availmem += atop((vm_offset_t)(xstart - start));
228 start = xend;
229 continue;
230 }
231 /*
232 * We know the excluded region overlaps either the start
233 * or end of this hardware region (but not both), trim
234 * the excluded portion off the appropriate end.
235 */
236 if (xstart <= start)
237 start = xend;
238 else
239 end = xstart;
240 }
241 /*
242 * If the trimming actions above left a non-zero size, create an
243 * available entry for it.
244 */
245 if (end > start) {
246 if ((maxphyssz != 0) &&
247 (availsz + end - start > maxphyssz)) {
248 end = maxphyssz + start - availsz;
249 }
250 if (end <= start)
251 break;
252
253 if (acnt > 0 && avail[acnt - 1] == (vm_paddr_t)start) {
254 avail[acnt - 1] = (vm_paddr_t)end;
255 } else {
256 avail[acnt++] = (vm_paddr_t)start;
257 avail[acnt++] = (vm_paddr_t)end;
258 }
259 availsz += end - start;
260 availmem += atop((vm_offset_t)(end - start));
261 }
262 if (acnt >= maxavail)
263 panic("Not enough space in the dump/phys_avail arrays");
264 }
265
266 if (pavail != NULL)
267 *pavail = availmem;
268 if (prealmem != NULL)
269 *prealmem = totalmem;
270 return (acnt);
271}
272
273/*
274 * Check if the region at idx can be merged with the region above it.
275 */
276static size_t
277merge_upper_regions(struct region *regions, size_t rcnt, size_t idx)
278{
279 struct region *lower, *upper;
280 vm_paddr_t lend, uend;
281 size_t i, mergecnt, movecnt;
282
283 lower = &regions[idx];
284 lend = lower->addr + lower->size;
285
286 /*
287 * Continue merging in upper entries as long as we have entries to
288 * merge; the new block could have spanned more than one, although one
289 * is likely the common case.
290 */
291 for (i = idx + 1; i < rcnt; i++) {
292 upper = &regions[i];
293 if (lend < upper->addr || lower->flags != upper->flags)
294 break;
295
296 uend = upper->addr + upper->size;
297 if (uend > lend) {
298 lower->size += uend - lend;
299 lend = lower->addr + lower->size;
300 }
301
302 if (uend >= lend) {
303 /*
304 * If we didn't move past the end of the upper region,
305 * then we don't need to bother checking for another
306 * merge because it would have been done already. Just
307 * increment i once more to maintain the invariant that
308 * i is one past the last entry merged.
309 */
310 i++;
311 break;
312 }
313 }
314
315 /*
316 * We merged in the entries from [idx + 1, i); physically move the tail
317 * end at [i, rcnt) if we need to.
318 */
319 mergecnt = i - (idx + 1);
320 if (mergecnt > 0) {
321 movecnt = rcnt - i;
322 if (movecnt == 0) {
323 /* Merged all the way to the end, just decrease rcnt. */
324 rcnt = idx + 1;
325 } else {
326 memmove(&regions[idx + 1], &regions[idx + mergecnt + 1],
327 movecnt * sizeof(*regions));
328 rcnt -= mergecnt;
329 }
330 }
331 return (rcnt);
332}
333
334/*
335 * Insertion-sort a new entry into a regions list; sorted by start address.
336 */
337static size_t
338insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr,
339 vm_size_t size, uint32_t flags)
340{
341 size_t i;
342 vm_paddr_t nend, rend;
343 struct region *ep, *rp;
344
345 nend = addr + size;
346 ep = regions + rcnt;
347 for (i = 0, rp = regions; i < rcnt; ++i, ++rp) {
348 if (flags == rp->flags) {
349 rend = rp->addr + rp->size;
350 if (addr <= rp->addr && nend >= rp->addr) {
351 /*
352 * New mapping overlaps at the beginning, shift
353 * for any difference in the beginning then
354 * shift if the new mapping extends past.
355 */
356 rp->size += rp->addr - addr;
357 rp->addr = addr;
358 if (nend > rend) {
359 rp->size += nend - rend;
360 rcnt = merge_upper_regions(regions,
361 rcnt, i);
362 }
363 return (rcnt);
364 } else if (addr <= rend && nend > rp->addr) {
365 /*
366 * New mapping is either entirely contained
367 * within or it's overlapping at the end.
368 */
369 if (nend > rend) {
370 rp->size += nend - rend;
371 rcnt = merge_upper_regions(regions,
372 rcnt, i);
373 }
374 return (rcnt);
375 }
376 }
377 if (addr < rp->addr) {
378 bcopy(rp, rp + 1, (ep - rp) * sizeof(*rp));
379 break;
380 }
381 }
382 rp->addr = addr;
383 rp->size = size;
384 rp->flags = flags;
385 rcnt++;
386
387 return (rcnt);
388}
389
390/*
391 * Add a hardware memory region.
392 */
393void
394physmem_hardware_region(uint64_t pa, uint64_t sz)
395{
396 vm_offset_t adj;
397
398 /*
399 * Filter out the page at PA 0x00000000. The VM can't handle it, as
400 * pmap_extract() == 0 means failure.
401 */
402 if (pa == 0) {
403 if (sz <= PAGE_SIZE)
404 return;
405 pa = PAGE_SIZE;
406 sz -= PAGE_SIZE;
407 } else if (pa > MAX_PHYS_ADDR) {
408 /* This range is past usable memory, ignore it */
409 return;
410 }
411
412 /*
413 * Also filter out the page at the end of the physical address space --
414 * if addr is non-zero and addr+size is zero we wrapped to the next byte
415 * beyond what vm_paddr_t can express. That leads to a NULL pointer
416 * deref early in startup; work around it by leaving the last page out.
417 *
418 * XXX This just in: subtract out a whole megabyte, not just 1 page.
419 * Reducing the size by anything less than 1MB results in the NULL
420 * pointer deref in _vm_map_lock_read(). Better to give up a megabyte
421 * than leave some folks with an unusable system while we investigate.
422 */
423 if ((pa + sz) > (MAX_PHYS_ADDR - 1024 * 1024)) {
424 sz = MAX_PHYS_ADDR - pa + 1;
425 if (sz <= 1024 * 1024)
426 return;
427 sz -= 1024 * 1024;
428 }
429
430 /*
431 * Round the starting address up to a page boundary, and truncate the
432 * ending page down to a page boundary.
433 */
434 adj = round_page(pa) - pa;
435 pa = round_page(pa);
436 sz = trunc_page(sz - adj);
437
438 if (sz > 0 && hwcnt < nitems(hwregions))
439 hwcnt = insert_region(hwregions, hwcnt, pa, sz, 0);
440}
441
442/*
443 * Add an exclusion region.
444 */
445void
446physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
447{
448 vm_offset_t adj;
449
450 /*
451 * Truncate the starting address down to a page boundary, and round the
452 * ending page up to a page boundary.
453 */
454 adj = pa - trunc_page(pa);
455 pa = trunc_page(pa);
456 sz = round_page(sz + adj);
457
458 if (excnt >= nitems(exregions))
459 panic("failed to exclude region %#jx-%#jx", (uintmax_t)pa,
460 (uintmax_t)(pa + sz));
461 excnt = insert_region(exregions, excnt, pa, sz, exflags);
462}
463
464size_t
465physmem_avail(vm_paddr_t *avail, size_t maxavail)
466{
467
468 return (regions_to_avail(avail, EXFLAG_NOALLOC, maxavail, 0, NULL, NULL));
469}
470
471/*
472 * Process all the regions added earlier into the global avail lists.
473 *
474 * Updates the kernel global 'physmem' with the number of physical pages
475 * available for use (all pages not in any exclusion region).
476 *
477 * Updates the kernel global 'Maxmem' with the page number one greater then the
478 * last page of physical memory in the system.
479 */
480void
482{
483 size_t nextidx;
484 u_long hwphyssz;
485
486 hwphyssz = 0;
487 TUNABLE_ULONG_FETCH("hw.physmem", &hwphyssz);
488
489 regions_to_avail(dump_avail, EXFLAG_NODUMP, PHYS_AVAIL_ENTRIES,
490 hwphyssz, NULL, NULL);
491 nextidx = regions_to_avail(phys_avail, EXFLAG_NOALLOC,
492 PHYS_AVAIL_ENTRIES, hwphyssz, &physmem, &realmem);
493 if (nextidx == 0)
494 panic("No memory entries in phys_avail");
495 Maxmem = atop(phys_avail[nextidx - 1]);
496}
497
498#ifdef DDB
499#include <ddb/ddb.h>
500
501DB_SHOW_COMMAND(physmem, db_show_physmem)
502{
503
504 physmem_dump_tables(db_printf);
505}
506
507#endif /* DDB */
void panic(const char *fmt,...)
void *** start
Definition: linker_if.m:98
uint64_t * addr
Definition: msi_if.m:89
vm_size_t size
Definition: subr_physmem.c:73
uint32_t flags
Definition: subr_physmem.c:74
vm_paddr_t addr
Definition: subr_physmem.c:72
void physmem_hardware_region(uint64_t pa, uint64_t sz)
Definition: subr_physmem.c:394
void physmem_init_kernel_globals(void)
Definition: subr_physmem.c:481
size_t physmem_avail(vm_paddr_t *avail, size_t maxavail)
Definition: subr_physmem.c:465
static struct region exregions[MAX_EXCNT]
Definition: subr_physmem.c:78
static size_t excnt
Definition: subr_physmem.c:81
void physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
Definition: subr_physmem.c:446
static size_t hwcnt
Definition: subr_physmem.c:80
#define MAX_HWCNT
Definition: subr_physmem.c:61
static size_t merge_upper_regions(struct region *regions, size_t rcnt, size_t idx)
Definition: subr_physmem.c:277
static struct region hwregions[MAX_HWCNT]
Definition: subr_physmem.c:77
static void physmem_dump_tables(int(*prfunc)(const char *,...))
Definition: subr_physmem.c:96
long realmem
Definition: subr_physmem.c:87
__FBSDID("$FreeBSD$")
void physmem_print_tables(void)
Definition: subr_physmem.c:136
long Maxmem
Definition: subr_physmem.c:88
#define MAX_EXCNT
Definition: subr_physmem.c:62
static size_t insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr, vm_size_t size, uint32_t flags)
Definition: subr_physmem.c:338
static size_t regions_to_avail(vm_paddr_t *avail, uint32_t exflags, size_t maxavail, uint64_t maxphyssz, long *pavail, long *prealmem)
Definition: subr_physmem.c:155
int printf(const char *fmt,...)
Definition: subr_prf.c:397
uint16_t flags
Definition: subr_stats.c:2