xref: /freebsd/sys/kern/subr_physmem.c (revision c1d255d3)
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 
71 struct region {
72 	vm_paddr_t	addr;
73 	vm_size_t	size;
74 	uint32_t	flags;
75 };
76 
77 static struct region hwregions[MAX_HWCNT];
78 static struct region exregions[MAX_EXCNT];
79 
80 static size_t hwcnt;
81 static 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  */
87 long realmem;
88 long 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  */
95 static void
96 physmem_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  */
135 void
136 physmem_print_tables(void)
137 {
138 
139 	physmem_dump_tables(printf);
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  */
154 static size_t
155 regions_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  * Insertion-sort a new entry into a regions list; sorted by start address.
275  */
276 static size_t
277 insert_region(struct region *regions, size_t rcnt, vm_paddr_t addr,
278     vm_size_t size, uint32_t flags)
279 {
280 	size_t i;
281 	struct region *ep, *rp;
282 
283 	ep = regions + rcnt;
284 	for (i = 0, rp = regions; i < rcnt; ++i, ++rp) {
285 		if (rp->addr == addr && rp->size == size) /* Pure dup. */
286 			return (rcnt);
287 		if (flags == rp->flags) {
288 			if (addr + size == rp->addr) {
289 				rp->addr = addr;
290 				rp->size += size;
291 				return (rcnt);
292 			} else if (rp->addr + rp->size == addr) {
293 				rp->size += size;
294 				return (rcnt);
295 			}
296 		}
297 		if (addr < rp->addr) {
298 			bcopy(rp, rp + 1, (ep - rp) * sizeof(*rp));
299 			break;
300 		}
301 	}
302 	rp->addr  = addr;
303 	rp->size  = size;
304 	rp->flags = flags;
305 	rcnt++;
306 
307 	return (rcnt);
308 }
309 
310 /*
311  * Add a hardware memory region.
312  */
313 void
314 physmem_hardware_region(uint64_t pa, uint64_t sz)
315 {
316 	vm_offset_t adj;
317 
318 	/*
319 	 * Filter out the page at PA 0x00000000.  The VM can't handle it, as
320 	 * pmap_extract() == 0 means failure.
321 	 */
322 	if (pa == 0) {
323 		if (sz <= PAGE_SIZE)
324 			return;
325 		pa  = PAGE_SIZE;
326 		sz -= PAGE_SIZE;
327 	} else if (pa > MAX_PHYS_ADDR) {
328 		/* This range is past usable memory, ignore it */
329 		return;
330 	}
331 
332 	/*
333 	 * Also filter out the page at the end of the physical address space --
334 	 * if addr is non-zero and addr+size is zero we wrapped to the next byte
335 	 * beyond what vm_paddr_t can express.  That leads to a NULL pointer
336 	 * deref early in startup; work around it by leaving the last page out.
337 	 *
338 	 * XXX This just in:  subtract out a whole megabyte, not just 1 page.
339 	 * Reducing the size by anything less than 1MB results in the NULL
340 	 * pointer deref in _vm_map_lock_read().  Better to give up a megabyte
341 	 * than leave some folks with an unusable system while we investigate.
342 	 */
343 	if ((pa + sz) > (MAX_PHYS_ADDR - 1024 * 1024)) {
344 		sz = MAX_PHYS_ADDR - pa + 1;
345 		if (sz <= 1024 * 1024)
346 			return;
347 		sz -= 1024 * 1024;
348 	}
349 
350 	/*
351 	 * Round the starting address up to a page boundary, and truncate the
352 	 * ending page down to a page boundary.
353 	 */
354 	adj = round_page(pa) - pa;
355 	pa  = round_page(pa);
356 	sz  = trunc_page(sz - adj);
357 
358 	if (sz > 0 && hwcnt < nitems(hwregions))
359 		hwcnt = insert_region(hwregions, hwcnt, pa, sz, 0);
360 }
361 
362 /*
363  * Add an exclusion region.
364  */
365 void
366 physmem_exclude_region(vm_paddr_t pa, vm_size_t sz, uint32_t exflags)
367 {
368 	vm_offset_t adj;
369 
370 	/*
371 	 * Truncate the starting address down to a page boundary, and round the
372 	 * ending page up to a page boundary.
373 	 */
374 	adj = pa - trunc_page(pa);
375 	pa  = trunc_page(pa);
376 	sz  = round_page(sz + adj);
377 
378 	if (excnt >= nitems(exregions))
379 		panic("failed to exclude region %#jx-%#jx", (uintmax_t)pa,
380 		    (uintmax_t)(pa + sz));
381 	excnt = insert_region(exregions, excnt, pa, sz, exflags);
382 }
383 
384 size_t
385 physmem_avail(vm_paddr_t *avail, size_t maxavail)
386 {
387 
388 	return (regions_to_avail(avail, EXFLAG_NOALLOC, maxavail, 0, NULL, NULL));
389 }
390 
391 /*
392  * Process all the regions added earlier into the global avail lists.
393  *
394  * Updates the kernel global 'physmem' with the number of physical pages
395  * available for use (all pages not in any exclusion region).
396  *
397  * Updates the kernel global 'Maxmem' with the page number one greater then the
398  * last page of physical memory in the system.
399  */
400 void
401 physmem_init_kernel_globals(void)
402 {
403 	size_t nextidx;
404 	u_long hwphyssz;
405 
406 	hwphyssz = 0;
407 	TUNABLE_ULONG_FETCH("hw.physmem", &hwphyssz);
408 
409 	regions_to_avail(dump_avail, EXFLAG_NODUMP, PHYS_AVAIL_ENTRIES,
410 	    hwphyssz, NULL, NULL);
411 	nextidx = regions_to_avail(phys_avail, EXFLAG_NOALLOC,
412 	    PHYS_AVAIL_ENTRIES, hwphyssz, &physmem, &realmem);
413 	if (nextidx == 0)
414 		panic("No memory entries in phys_avail");
415 	Maxmem = atop(phys_avail[nextidx - 1]);
416 }
417 
418 #ifdef DDB
419 #include <ddb/ddb.h>
420 
421 DB_SHOW_COMMAND(physmem, db_show_physmem)
422 {
423 
424 	physmem_dump_tables(db_printf);
425 }
426 
427 #endif /* DDB */
428