2 * SPDX-License-Identifier: BSD-2-Clause
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("$FreeBSD$");
38 * Routines for describing and initializing anything related to physical memory.
41 #include <sys/param.h>
42 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/module.h>
46 #include <sys/physmem.h>
50 #include <vm/vm_param.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_phys.h>
53 #include <vm/vm_dumpset.h>
55 #include <machine/md_var.h>
56 #include <machine/resource.h>
64 * These structures are used internally to keep track of regions of physical
65 * ram, and regions within the physical ram that need to be excluded. An
66 * exclusion region can be excluded from crash dumps, from the vm pool of pages
67 * that can be allocated, or both, depending on the exclusion flags associated
71 #define MAX_HWCNT 32 /* ACPI needs more regions */
79 #define MAX_PHYS_ADDR 0xFFFFFFFFull
80 #elif defined(__aarch64__) || defined(__amd64__) || defined(__riscv)
81 #define MAX_PHYS_ADDR 0xFFFFFFFFFFFFFFFFull
90 static struct region hwregions[MAX_HWCNT];
91 static struct region exregions[MAX_EXCNT];
97 * realmem is the total number of hardware pages, excluded or not.
98 * Maxmem is one greater than the last physical page number.
105 panic(const char *fmt, ...)
110 vfprintf(stderr, fmt, va);
111 fprintf(stderr, "\n");
118 * Print the contents of the physical and excluded region tables using the
119 * provided printf-like output function (which will be either printf or
123 physmem_dump_tables(int (*prfunc)(const char *, ...) __printflike(1, 2))
127 uintmax_t addr, size;
128 const unsigned int mbyte = 1024 * 1024;
130 prfunc("Physical memory chunk(s):\n");
131 for (i = 0; i < hwcnt; ++i) {
132 addr = hwregions[i].addr;
133 size = hwregions[i].size;
134 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages)\n", addr,
135 addr + size - 1, size / mbyte, size / PAGE_SIZE);
138 prfunc("Excluded memory regions:\n");
139 for (i = 0; i < excnt; ++i) {
140 addr = exregions[i].addr;
141 size = exregions[i].size;
142 flags = exregions[i].flags;
143 prfunc(" 0x%08jx - 0x%08jx, %5ju MB (%7ju pages) %s %s\n",
144 addr, addr + size - 1, size / mbyte, size / PAGE_SIZE,
145 (flags & EXFLAG_NOALLOC) ? "NoAlloc" : "",
146 (flags & EXFLAG_NODUMP) ? "NoDump" : "");
150 prfunc("Avail lists:\n");
151 for (i = 0; phys_avail[i] != 0; ++i) {
152 prfunc(" phys_avail[%zu] 0x%08jx\n", i,
153 (uintmax_t)phys_avail[i]);
155 for (i = 0; dump_avail[i] != 0; ++i) {
156 prfunc(" dump_avail[%zu] 0x%08jx\n", i,
157 (uintmax_t)dump_avail[i]);
163 * Print the contents of the static mapping table. Used for bootverbose.
166 physmem_print_tables(void)
169 physmem_dump_tables(printf);
173 * Walk the list of hardware regions, processing it against the list of
174 * exclusions that contain the given exflags, and generating an "avail list".
176 * If maxphyssz is not zero it sets upper limit, in bytes, for the total
177 * "avail list" size. Walk stops once the limit is reached and the last region
178 * is cut short if necessary.
180 * Updates the value at *pavail with the sum of all pages in all hw regions.
182 * Returns the number of pages of non-excluded memory added to the avail list.
185 regions_to_avail(vm_paddr_t *avail, uint32_t exflags, size_t maxavail,
186 uint64_t maxphyssz, long *pavail, long *prealmem)
188 size_t acnt, exi, hwi;
189 uint64_t adj, end, start, xend, xstart;
190 long availmem, totalmem;
191 const struct region *exp, *hwp;
198 for (hwi = 0, hwp = hwregions; hwi < hwcnt; ++hwi, ++hwp) {
199 adj = round_page(hwp->addr) - hwp->addr;
200 start = round_page(hwp->addr);
201 end = trunc_page(hwp->size + adj) + start;
202 totalmem += atop((vm_offset_t)(end - start));
203 for (exi = 0, exp = exregions; exi < excnt; ++exi, ++exp) {
205 * If the excluded region does not match given flags,
206 * continue checking with the next excluded region.
208 if ((exp->flags & exflags) == 0)
211 xend = exp->size + xstart;
213 * If the excluded region ends before this hw region,
214 * continue checking with the next excluded region.
219 * If the excluded region begins after this hw region
220 * we're done because both lists are sorted.
225 * If the excluded region completely covers this hw
226 * region, shrink this hw region to zero size.
228 if ((start >= xstart) && (end <= xend)) {
234 * If the excluded region falls wholly within this hw
235 * region without abutting or overlapping the beginning
236 * or end, create an available entry from the leading
237 * fragment, then adjust the start of this hw region to
238 * the end of the excluded region, and continue checking
239 * the next excluded region because another exclusion
240 * could affect the remainder of this hw region.
242 if ((xstart > start) && (xend < end)) {
244 if ((maxphyssz != 0) &&
245 (availsz + xstart - start > maxphyssz)) {
246 xstart = maxphyssz + start - availsz;
251 avail[acnt - 1] == (vm_paddr_t)start) {
252 avail[acnt - 1] = (vm_paddr_t)xstart;
254 avail[acnt++] = (vm_paddr_t)start;
255 avail[acnt++] = (vm_paddr_t)xstart;
257 availsz += (xstart - start);
258 availmem += atop((vm_offset_t)(xstart - start));
263 * We know the excluded region overlaps either the start
264 * or end of this hardware region (but not both), trim
265 * the excluded portion off the appropriate end.
273 * If the trimming actions above left a non-zero size, create an
274 * available entry for it.
277 if ((maxphyssz != 0) &&
278 (availsz + end - start > maxphyssz)) {
279 end = maxphyssz + start - availsz;
284 if (acnt > 0 && avail[acnt - 1] == (vm_paddr_t)start) {
285 avail[acnt - 1] = (vm_paddr_t)end;
287 avail[acnt++] = (vm_paddr_t)start;
288 avail[acnt++] = (vm_paddr_t)end;
290 availsz += end - start;
291 availmem += atop((vm_offset_t)(end - start));
293 if (acnt >= maxavail)
294 panic("Not enough space in the dump/phys_avail arrays");
299 if (prealmem != NULL)
300 *prealmem = totalmem;
305 * Check if the region at idx can be merged with the region above it.
308 merge_upper_regions(struct region *regions, size_t rcnt, size_t idx)
310 struct region *lower, *upper;
311 vm_paddr_t lend, uend;
312 size_t i, mergecnt, movecnt;
314 lower = ®ions[idx];
315 lend = lower->addr + lower->size;
318 * Continue merging in upper entries as long as we have entries to
319 * merge; the new block could have spanned more than one, although one
320 * is likely the common case.
322 for (i = idx + 1; i < rcnt; i++) {
324 if (lend < upper->addr || lower->flags != upper->flags)
327 uend = upper->addr + upper->size;
329 lower->size += uend - lend;
330 lend = lower->addr + lower->size;
335 * If we didn't move past the end of the upper region,
336 * then we don't need to bother checking for another
337 * merge because it would have been done already. Just
338 * increment i once more to maintain the invariant that
339 * i is one past the last entry merged.
347 * We merged in the entries from [idx + 1, i); physically move the tail
348 * end at [i, rcnt) if we need to.
350 mergecnt = i - (idx + 1);
354 /* Merged all the way to the end, just decrease rcnt. */
357 memmove(®ions[idx + 1], ®ions[idx + mergecnt + 1],
358 movecnt * sizeof(*regions));
366 * Insertion-sort a new entry into a regions list; sorted by start address.
369 insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr,
370 vm_size_t size, uint32_t flags)
373 vm_paddr_t nend, rend;
374 struct region *ep, *rp;
378 for (i = 0, rp = regions; i < rcnt; ++i, ++rp) {
379 rend = rp->addr + rp->size;
380 if (flags == rp->flags) {
381 if (addr <= rp->addr && nend >= rp->addr) {
383 * New mapping overlaps at the beginning, shift
384 * for any difference in the beginning then
385 * shift if the new mapping extends past.
387 rp->size += rp->addr - addr;
390 rp->size += nend - rend;
391 rcnt = merge_upper_regions(regions,
395 } else if (addr <= rend && nend > rp->addr) {
397 * New mapping is either entirely contained
398 * within or it's overlapping at the end.
401 rp->size += nend - rend;
402 rcnt = merge_upper_regions(regions,
407 } else if ((flags != 0) && (rp->flags != 0)) {
409 * If we're duplicating an entry that already exists
410 * exactly, just upgrade its flags as needed. We could
411 * do more if we find that we have differently specified
412 * flags clipping existing excluding regions, but that's
415 if (addr == rp->addr && nend == rend) {
421 if (addr < rp->addr) {
422 bcopy(rp, rp + 1, (ep - rp) * sizeof(*rp));
435 * Add a hardware memory region.
438 physmem_hardware_region(uint64_t pa, uint64_t sz)
441 * Filter out the page at PA 0x00000000. The VM can't handle it, as
442 * pmap_extract() == 0 means failure.
449 } else if (pa > MAX_PHYS_ADDR) {
450 /* This range is past usable memory, ignore it */
455 * Also filter out the page at the end of the physical address space --
456 * if addr is non-zero and addr+size is zero we wrapped to the next byte
457 * beyond what vm_paddr_t can express. That leads to a NULL pointer
458 * deref early in startup; work around it by leaving the last page out.
460 * XXX This just in: subtract out a whole megabyte, not just 1 page.
461 * Reducing the size by anything less than 1MB results in the NULL
462 * pointer deref in _vm_map_lock_read(). Better to give up a megabyte
463 * than leave some folks with an unusable system while we investigate.
465 if ((pa + sz) > (MAX_PHYS_ADDR - 1024 * 1024)) {
466 sz = MAX_PHYS_ADDR - pa + 1;
467 if (sz <= 1024 * 1024)
472 if (sz > 0 && hwcnt < nitems(hwregions))
473 hwcnt = insert_region(hwregions, hwcnt, pa, sz, 0);
477 * Add an exclusion region.
480 physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
485 * Truncate the starting address down to a page boundary, and round the
486 * ending page up to a page boundary.
488 adj = pa - trunc_page(pa);
490 sz = round_page(sz + adj);
492 if (excnt >= nitems(exregions))
493 panic("failed to exclude region %#jx-%#jx", (uintmax_t)pa,
494 (uintmax_t)(pa + sz));
495 excnt = insert_region(exregions, excnt, pa, sz, exflags);
499 physmem_avail(vm_paddr_t *avail, size_t maxavail)
502 return (regions_to_avail(avail, EXFLAG_NOALLOC, maxavail, 0, NULL, NULL));
506 physmem_excluded(vm_paddr_t pa, vm_size_t sz)
508 const struct region *exp;
511 for (exi = 0, exp = exregions; exi < excnt; ++exi, ++exp) {
512 if (pa < exp->addr || pa + sz > exp->addr + exp->size)
521 * Process all the regions added earlier into the global avail lists.
523 * Updates the kernel global 'physmem' with the number of physical pages
524 * available for use (all pages not in any exclusion region).
526 * Updates the kernel global 'Maxmem' with the page number one greater then the
527 * last page of physical memory in the system.
530 physmem_init_kernel_globals(void)
536 TUNABLE_ULONG_FETCH("hw.physmem", &hwphyssz);
538 regions_to_avail(dump_avail, EXFLAG_NODUMP, PHYS_AVAIL_ENTRIES,
539 hwphyssz, NULL, NULL);
540 nextidx = regions_to_avail(phys_avail, EXFLAG_NOALLOC,
541 PHYS_AVAIL_ENTRIES, hwphyssz, &physmem, &realmem);
543 panic("No memory entries in phys_avail");
544 Maxmem = atop(phys_avail[nextidx - 1]);
550 DB_SHOW_COMMAND_FLAGS(physmem, db_show_physmem, DB_CMD_MEMSAFE)
553 physmem_dump_tables(db_printf);
559 * ram pseudo driver - this reserves I/O space resources corresponding to physical
564 ram_identify(driver_t *driver, device_t parent)
567 if (resource_disabled("ram", 0))
569 if (BUS_ADD_CHILD(parent, 0, "ram", 0) == NULL)
570 panic("ram_identify");
574 ram_probe(device_t dev)
578 device_set_desc(dev, "System RAM");
579 return (BUS_PROBE_SPECIFIC);
583 ram_attach(device_t dev)
585 vm_paddr_t avail_list[PHYS_AVAIL_COUNT];
586 rman_res_t start, end;
592 /* Get the avail list. */
593 bzero(avail_list, sizeof(avail_list));
594 regions_to_avail(avail_list, EXFLAG_NOALLOC | EXFLAG_NODUMP,
595 PHYS_AVAIL_COUNT, 0, NULL, NULL);
597 /* Reserve all memory regions. */
598 for (i = 0; avail_list[i + 1] != 0; i += 2) {
599 start = avail_list[i];
600 end = avail_list[i + 1];
604 "reserving memory region: %jx-%jx\n",
605 (uintmax_t)start, (uintmax_t)end);
607 if (bus_alloc_resource(dev, SYS_RES_MEMORY, &rid, start, end,
608 end - start, 0) == NULL)
609 panic("ram_attach: resource %d failed to attach", rid);
613 /* Now, reserve the excluded memory regions. */
614 for (i = 0, hwp = exregions; i < excnt; i++, hwp++) {
616 end = hwp->addr + hwp->size;
620 "reserving excluded region: %jx-%jx\n",
621 (uintmax_t)start, (uintmax_t)(end - 1));
624 * Best-effort attempt to reserve the range. This may fail, as
625 * sometimes the excluded ranges provided by the device tree
626 * will cover or overlap some I/O range.
628 if (bus_alloc_resource(dev, SYS_RES_MEMORY, &rid, start, end,
629 end - start, 0) == NULL) {
631 device_printf(dev, "failed to reserve region\n");
640 static device_method_t ram_methods[] = {
641 /* Device interface */
642 DEVMETHOD(device_identify, ram_identify),
643 DEVMETHOD(device_probe, ram_probe),
644 DEVMETHOD(device_attach, ram_attach),
649 DEFINE_CLASS_0(ram, ram_driver, ram_methods, /* no softc */ 1);
650 DRIVER_MODULE(ram, nexus, ram_driver, 0, 0);