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/physmem.h>
43 #include <machine/md_var.h>
46 * These structures are used internally to keep track of regions of physical
47 * ram, and regions within the physical ram that need to be excluded. An
48 * exclusion region can be excluded from crash dumps, from the vm pool of pages
49 * that can be allocated, or both, depending on the exclusion flags associated
53 #define MAX_HWCNT 32 /* ACPI needs more regions */
61 #define MAX_PHYS_ADDR 0xFFFFFFFFull
62 #elif defined(__aarch64__) || defined(__riscv)
63 #define MAX_PHYS_ADDR 0xFFFFFFFFFFFFFFFFull
72 static struct region hwregions[MAX_HWCNT];
73 static struct region exregions[MAX_EXCNT];
79 * These "avail lists" are globals used to communicate physical memory layout to
80 * other parts of the kernel. Within the arrays, each value is the starting
81 * address of a contiguous area of physical address space. The values at even
82 * indexes are areas that contain usable memory and the values at odd indexes
83 * are areas that aren't usable. Each list is terminated by a pair of zero
86 * dump_avail tells the dump code what regions to include in a crash dump, and
87 * phys_avail is the way we hand all the remaining physical ram we haven't used
88 * in early kernel init over to the vm system for allocation management.
90 * We size these arrays to hold twice as many available regions as we allow for
91 * hardware memory regions, to allow for the fact that exclusions can split a
92 * hardware region into two or more available regions. In the real world there
93 * will typically be one or two hardware regions and two or three exclusions.
95 * Each available region in this list occupies two array slots (the start of the
96 * available region and the start of the unavailable region that follows it).
98 #define MAX_AVAIL_REGIONS (MAX_HWCNT * 2)
99 #define MAX_AVAIL_ENTRIES (MAX_AVAIL_REGIONS * 2)
101 vm_paddr_t phys_avail[MAX_AVAIL_ENTRIES + 2]; /* +2 to allow for a pair */
102 vm_paddr_t dump_avail[MAX_AVAIL_ENTRIES + 2]; /* of zeroes to terminate. */
105 * realmem is the total number of hardware pages, excluded or not.
106 * Maxmem is one greater than the last physical page number.
112 * Print the contents of the physical and excluded region tables using the
113 * provided printf-like output function (which will be either printf or
117 physmem_dump_tables(int (*prfunc)(const char *, ...))
120 uintmax_t addr, size;
121 const unsigned int mbyte = 1024 * 1024;
123 prfunc("Physical memory chunk(s):\n");
124 for (i = 0; i < hwcnt; ++i) {
125 addr = hwregions[i].addr;
126 size = hwregions[i].size;
127 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages)\n", addr,
128 addr + size - 1, size / mbyte, size / PAGE_SIZE);
131 prfunc("Excluded memory regions:\n");
132 for (i = 0; i < excnt; ++i) {
133 addr = exregions[i].addr;
134 size = exregions[i].size;
135 flags = exregions[i].flags;
136 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages) %s %s\n",
137 addr, addr + size - 1, size / mbyte, size / PAGE_SIZE,
138 (flags & EXFLAG_NOALLOC) ? "NoAlloc" : "",
139 (flags & EXFLAG_NODUMP) ? "NoDump" : "");
143 prfunc("Avail lists:\n");
144 for (i = 0; phys_avail[i] != 0; ++i) {
145 prfunc(" phys_avail[%d] 0x%08x\n", i, phys_avail[i]);
147 for (i = 0; dump_avail[i] != 0; ++i) {
148 prfunc(" dump_avail[%d] 0x%08x\n", i, dump_avail[i]);
154 * Print the contents of the static mapping table. Used for bootverbose.
157 physmem_print_tables(void)
160 physmem_dump_tables(printf);
164 * Walk the list of hardware regions, processing it against the list of
165 * exclusions that contain the given exflags, and generating an "avail list".
167 * Updates the value at *pavail with the sum of all pages in all hw regions.
169 * Returns the number of pages of non-excluded memory added to the avail list.
172 regions_to_avail(vm_paddr_t *avail, uint32_t exflags, size_t maxavail,
173 long *pavail, long *prealmem)
175 size_t acnt, exi, hwi;
176 uint64_t end, start, xend, xstart;
177 long availmem, totalmem;
178 const struct region *exp, *hwp;
183 for (hwi = 0, hwp = hwregions; hwi < hwcnt; ++hwi, ++hwp) {
185 end = hwp->size + start;
186 totalmem += atop((vm_offset_t)(end - start));
187 for (exi = 0, exp = exregions; exi < excnt; ++exi, ++exp) {
189 * If the excluded region does not match given flags,
190 * continue checking with the next excluded region.
192 if ((exp->flags & exflags) == 0)
195 xend = exp->size + xstart;
197 * If the excluded region ends before this hw region,
198 * continue checking with the next excluded region.
203 * If the excluded region begins after this hw region
204 * we're done because both lists are sorted.
209 * If the excluded region completely covers this hw
210 * region, shrink this hw region to zero size.
212 if ((start >= xstart) && (end <= xend)) {
218 * If the excluded region falls wholly within this hw
219 * region without abutting or overlapping the beginning
220 * or end, create an available entry from the leading
221 * fragment, then adjust the start of this hw region to
222 * the end of the excluded region, and continue checking
223 * the next excluded region because another exclusion
224 * could affect the remainder of this hw region.
226 if ((xstart > start) && (xend < end)) {
228 avail[acnt - 1] == (vm_paddr_t)start) {
229 avail[acnt - 1] = (vm_paddr_t)xstart;
231 avail[acnt++] = (vm_paddr_t)start;
232 avail[acnt++] = (vm_paddr_t)xstart;
234 availmem += atop((vm_offset_t)(xstart - start));
239 * We know the excluded region overlaps either the start
240 * or end of this hardware region (but not both), trim
241 * the excluded portion off the appropriate end.
249 * If the trimming actions above left a non-zero size, create an
250 * available entry for it.
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 availmem += atop((vm_offset_t)(end - start));
261 if (acnt >= maxavail)
262 panic("Not enough space in the dump/phys_avail arrays");
267 if (prealmem != NULL)
268 *prealmem = totalmem;
273 * Insertion-sort a new entry into a regions list; sorted by start address.
276 insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr,
277 vm_size_t size, uint32_t flags)
280 struct region *ep, *rp;
283 for (i = 0, rp = regions; i < rcnt; ++i, ++rp) {
284 if (rp->addr == addr && rp->size == size) /* Pure dup. */
286 if (flags == rp->flags) {
287 if (addr + size == rp->addr) {
291 } else if (rp->addr + rp->size == addr) {
296 if (addr < rp->addr) {
297 bcopy(rp, rp + 1, (ep - rp) * sizeof(*rp));
310 * Add a hardware memory region.
313 physmem_hardware_region(uint64_t pa, uint64_t sz)
318 * Filter out the page at PA 0x00000000. The VM can't handle it, as
319 * pmap_extract() == 0 means failure.
326 } else if (pa > MAX_PHYS_ADDR) {
327 /* This range is past usable memory, ignore it */
332 * Also filter out the page at the end of the physical address space --
333 * if addr is non-zero and addr+size is zero we wrapped to the next byte
334 * beyond what vm_paddr_t can express. That leads to a NULL pointer
335 * deref early in startup; work around it by leaving the last page out.
337 * XXX This just in: subtract out a whole megabyte, not just 1 page.
338 * Reducing the size by anything less than 1MB results in the NULL
339 * pointer deref in _vm_map_lock_read(). Better to give up a megabyte
340 * than leave some folks with an unusable system while we investigate.
342 if ((pa + sz) > (MAX_PHYS_ADDR - 1024 * 1024)) {
343 sz = MAX_PHYS_ADDR - pa + 1;
344 if (sz <= 1024 * 1024)
350 * Round the starting address up to a page boundary, and truncate the
351 * ending page down to a page boundary.
353 adj = round_page(pa) - pa;
355 sz = trunc_page(sz - adj);
357 if (sz > 0 && hwcnt < nitems(hwregions))
358 hwcnt = insert_region(hwregions, hwcnt, pa, sz, 0);
362 * Add an exclusion region.
365 physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
370 * Truncate the starting address down to a page boundary, and round the
371 * ending page up to a page boundary.
373 adj = pa - trunc_page(pa);
375 sz = round_page(sz + adj);
377 if (excnt >= nitems(exregions))
378 panic("failed to exclude region %#jx-%#jx", (uintmax_t)pa,
379 (uintmax_t)(pa + sz));
380 excnt = insert_region(exregions, excnt, pa, sz, exflags);
384 physmem_avail(vm_paddr_t *avail, size_t maxavail)
387 return (regions_to_avail(avail, EXFLAG_NOALLOC, maxavail, NULL, NULL));
391 * Process all the regions added earlier into the global avail lists.
393 * Updates the kernel global 'physmem' with the number of physical pages
394 * available for use (all pages not in any exclusion region).
396 * Updates the kernel global 'Maxmem' with the page number one greater then the
397 * last page of physical memory in the system.
400 physmem_init_kernel_globals(void)
404 regions_to_avail(dump_avail, EXFLAG_NODUMP, MAX_AVAIL_ENTRIES, NULL,
406 nextidx = regions_to_avail(phys_avail, EXFLAG_NOALLOC,
407 MAX_AVAIL_ENTRIES, &physmem, &realmem);
409 panic("No memory entries in phys_avail");
410 Maxmem = atop(phys_avail[nextidx - 1]);
416 DB_SHOW_COMMAND(physmem, db_show_physmem)
419 physmem_dump_tables(db_printf);