2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2014 Ian Lepore <ian@freebsd.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
36 * Routines for describing and initializing anything related to physical memory.
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/physmem.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>
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
58 #define MAX_HWCNT 32 /* ACPI needs more regions */
66 #define MAX_PHYS_ADDR 0xFFFFFFFFull
67 #elif defined(__aarch64__) || defined(__riscv)
68 #define MAX_PHYS_ADDR 0xFFFFFFFFFFFFFFFFull
77 static struct region hwregions[MAX_HWCNT];
78 static struct region exregions[MAX_EXCNT];
84 * realmem is the total number of hardware pages, excluded or not.
85 * Maxmem is one greater than the last physical page number.
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
96 physmem_dump_tables(int (*prfunc)(const char *, ...))
100 const unsigned int mbyte = 1024 * 1024;
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);
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" : "");
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]);
126 for (i = 0; dump_avail[i] != 0; ++i) {
127 prfunc(" dump_avail[%d] 0x%08x\n", i, dump_avail[i]);
133 * Print the contents of the static mapping table. Used for bootverbose.
136 physmem_print_tables(void)
139 physmem_dump_tables(printf);
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".
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.
150 * Updates the value at *pavail with the sum of all pages in all hw regions.
152 * Returns the number of pages of non-excluded memory added to the avail list.
155 regions_to_avail(vm_paddr_t *avail, uint32_t exflags, size_t maxavail,
156 uint64_t maxphyssz, long *pavail, long *prealmem)
158 size_t acnt, exi, hwi;
159 uint64_t end, start, xend, xstart;
160 long availmem, totalmem;
161 const struct region *exp, *hwp;
168 for (hwi = 0, hwp = hwregions; hwi < hwcnt; ++hwi, ++hwp) {
170 end = hwp->size + start;
171 totalmem += atop((vm_offset_t)(end - start));
172 for (exi = 0, exp = exregions; exi < excnt; ++exi, ++exp) {
174 * If the excluded region does not match given flags,
175 * continue checking with the next excluded region.
177 if ((exp->flags & exflags) == 0)
180 xend = exp->size + xstart;
182 * If the excluded region ends before this hw region,
183 * continue checking with the next excluded region.
188 * If the excluded region begins after this hw region
189 * we're done because both lists are sorted.
194 * If the excluded region completely covers this hw
195 * region, shrink this hw region to zero size.
197 if ((start >= xstart) && (end <= xend)) {
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.
211 if ((xstart > start) && (xend < end)) {
213 if ((maxphyssz != 0) &&
214 (availsz + xstart - start > maxphyssz)) {
215 xstart = maxphyssz + start - availsz;
220 avail[acnt - 1] == (vm_paddr_t)start) {
221 avail[acnt - 1] = (vm_paddr_t)xstart;
223 avail[acnt++] = (vm_paddr_t)start;
224 avail[acnt++] = (vm_paddr_t)xstart;
226 availsz += (xstart - start);
227 availmem += atop((vm_offset_t)(xstart - start));
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.
242 * If the trimming actions above left a non-zero size, create an
243 * available entry for it.
246 if ((maxphyssz != 0) &&
247 (availsz + end - start > maxphyssz)) {
248 end = maxphyssz + start - availsz;
253 if (acnt > 0 && avail[acnt - 1] == (vm_paddr_t)start) {
254 avail[acnt - 1] = (vm_paddr_t)end;
256 avail[acnt++] = (vm_paddr_t)start;
257 avail[acnt++] = (vm_paddr_t)end;
259 availsz += end - start;
260 availmem += atop((vm_offset_t)(end - start));
262 if (acnt >= maxavail)
263 panic("Not enough space in the dump/phys_avail arrays");
268 if (prealmem != NULL)
269 *prealmem = totalmem;
274 * Check if the region at idx can be merged with the region above it.
277 merge_upper_regions(struct region *regions, size_t rcnt, size_t idx)
279 struct region *lower, *upper;
280 vm_paddr_t lend, uend;
281 size_t i, mergecnt, movecnt;
283 lower = ®ions[idx];
284 lend = lower->addr + lower->size;
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.
291 for (i = idx + 1; i < rcnt; i++) {
293 if (lend < upper->addr || lower->flags != upper->flags)
296 uend = upper->addr + upper->size;
298 lower->size += uend - lend;
299 lend = lower->addr + lower->size;
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.
316 * We merged in the entries from [idx + 1, i); physically move the tail
317 * end at [i, rcnt) if we need to.
319 mergecnt = i - (idx + 1);
323 /* Merged all the way to the end, just decrease rcnt. */
326 memmove(®ions[idx + 1], ®ions[idx + mergecnt + 1],
327 movecnt * sizeof(*regions));
335 * Insertion-sort a new entry into a regions list; sorted by start address.
338 insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr,
339 vm_size_t size, uint32_t flags)
342 vm_paddr_t nend, rend;
343 struct region *ep, *rp;
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) {
352 * New mapping overlaps at the beginning, shift
353 * for any difference in the beginning then
354 * shift if the new mapping extends past.
356 rp->size += rp->addr - addr;
359 rp->size += nend - rend;
360 rcnt = merge_upper_regions(regions,
364 } else if (addr <= rend && nend > rp->addr) {
366 * New mapping is either entirely contained
367 * within or it's overlapping at the end.
370 rp->size += nend - rend;
371 rcnt = merge_upper_regions(regions,
377 if (addr < rp->addr) {
378 bcopy(rp, rp + 1, (ep - rp) * sizeof(*rp));
391 * Add a hardware memory region.
394 physmem_hardware_region(uint64_t pa, uint64_t sz)
399 * Filter out the page at PA 0x00000000. The VM can't handle it, as
400 * pmap_extract() == 0 means failure.
407 } else if (pa > MAX_PHYS_ADDR) {
408 /* This range is past usable memory, ignore it */
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.
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.
423 if ((pa + sz) > (MAX_PHYS_ADDR - 1024 * 1024)) {
424 sz = MAX_PHYS_ADDR - pa + 1;
425 if (sz <= 1024 * 1024)
431 * Round the starting address up to a page boundary, and truncate the
432 * ending page down to a page boundary.
434 adj = round_page(pa) - pa;
436 sz = trunc_page(sz - adj);
438 if (sz > 0 && hwcnt < nitems(hwregions))
439 hwcnt = insert_region(hwregions, hwcnt, pa, sz, 0);
443 * Add an exclusion region.
446 physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
451 * Truncate the starting address down to a page boundary, and round the
452 * ending page up to a page boundary.
454 adj = pa - trunc_page(pa);
456 sz = round_page(sz + adj);
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);
465 physmem_avail(vm_paddr_t *avail, size_t maxavail)
468 return (regions_to_avail(avail, EXFLAG_NOALLOC, maxavail, 0, NULL, NULL));
472 * Process all the regions added earlier into the global avail lists.
474 * Updates the kernel global 'physmem' with the number of physical pages
475 * available for use (all pages not in any exclusion region).
477 * Updates the kernel global 'Maxmem' with the page number one greater then the
478 * last page of physical memory in the system.
481 physmem_init_kernel_globals(void)
487 TUNABLE_ULONG_FETCH("hw.physmem", &hwphyssz);
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);
494 panic("No memory entries in phys_avail");
495 Maxmem = atop(phys_avail[nextidx - 1]);
501 DB_SHOW_COMMAND(physmem, db_show_physmem)
504 physmem_dump_tables(db_printf);