vmm_mem.c revision 221940 
1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD$
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include <sys/param.h>
33#include <sys/lock.h>
34#include <sys/mutex.h>
35#include <sys/linker.h>
36#include <sys/systm.h>
37#include <sys/malloc.h>
38#include <sys/kernel.h>
39
40#include <vm/vm.h>
41#include <vm/pmap.h>
42
43#include <machine/md_var.h>
44#include <machine/metadata.h>
45#include <machine/pc/bios.h>
46#include <machine/vmparam.h>
47#include <machine/pmap.h>
48
49#include "vmm_util.h"
50#include "vmm_mem.h"
51
52static MALLOC_DEFINE(M_VMM_MEM, "vmm memory", "vmm memory");
53
54#define	MB		(1024 * 1024)
55#define	GB		(1024 * MB)
56
57#define	VMM_MEM_MAXSEGS	64
58
59/* protected by vmm_mem_mtx */
60static struct {
61	vm_paddr_t	base;
62	vm_size_t	length;
63} vmm_mem_avail[VMM_MEM_MAXSEGS];
64
65static int vmm_mem_nsegs;
66
67static vm_paddr_t maxaddr;
68
69static struct mtx vmm_mem_mtx;
70
71/*
72 * Steal any memory that was deliberately hidden from FreeBSD either by
73 * the use of MAXMEM kernel config option or the hw.physmem loader tunable.
74 */
75static int
76vmm_mem_steal_memory(void)
77{
78	int nsegs;
79	caddr_t kmdp;
80	uint32_t smapsize;
81	uint64_t base, length;
82	struct bios_smap *smapbase, *smap, *smapend;
83
84	/*
85	 * Borrowed from hammer_time() and getmemsize() in machdep.c
86	 */
87	kmdp = preload_search_by_type("elf kernel");
88	if (kmdp == NULL)
89		kmdp = preload_search_by_type("elf64 kernel");
90
91	smapbase = (struct bios_smap *)preload_search_info(kmdp,
92		MODINFO_METADATA | MODINFOMD_SMAP);
93	if (smapbase == NULL)
94		panic("No BIOS smap info from loader!");
95
96	smapsize = *((uint32_t *)smapbase - 1);
97	smapend = (struct bios_smap *)((uintptr_t)smapbase + smapsize);
98
99	nsegs = 0;
100	for (smap = smapbase; smap < smapend; smap++) {
101		/*
102		 * XXX
103		 * Assuming non-overlapping, monotonically increasing
104		 * memory segments.
105		 */
106		if (smap->type != SMAP_TYPE_MEMORY)
107			continue;
108		if (smap->length == 0)
109			break;
110
111		base = roundup(smap->base, NBPDR);
112		length = rounddown(smap->length, NBPDR);
113
114		/* Skip this segment if FreeBSD is using all of it. */
115		if (base + length <= ptoa(Maxmem))
116			continue;
117
118		/*
119		 * If FreeBSD is using part of this segment then adjust
120		 * 'base' and 'length' accordingly.
121		 */
122		if (base < ptoa(Maxmem)) {
123			uint64_t used;
124			used = roundup(ptoa(Maxmem), NBPDR) - base;
125			base += used;
126			length -= used;
127		}
128
129		if (length == 0)
130			continue;
131
132		vmm_mem_avail[nsegs].base = base;
133		vmm_mem_avail[nsegs].length = length;
134
135		if (base + length > maxaddr)
136			maxaddr = base + length;
137
138		if (0 && bootverbose) {
139			printf("vmm_mem_populate: index %d, base 0x%0lx, "
140			       "length %ld\n",
141			       nsegs, vmm_mem_avail[nsegs].base,
142			       vmm_mem_avail[nsegs].length);
143		}
144
145		nsegs++;
146		if (nsegs >= VMM_MEM_MAXSEGS) {
147			printf("vmm_mem_populate: maximum number of vmm memory "
148			       "segments reached!\n");
149			return (ENOSPC);
150		}
151	}
152
153	vmm_mem_nsegs = nsegs;
154
155	return (0);
156}
157
158static void
159vmm_mem_direct_map(vm_paddr_t start, vm_paddr_t end)
160{
161	vm_paddr_t addr, remaining;
162	int pdpi, pdi, superpage_size;
163	pml4_entry_t *pml4p;
164	pdp_entry_t *pdp;
165	pd_entry_t *pd;
166	uint64_t page_attr_bits;
167
168	if (end >= NBPML4)
169		panic("Cannot map memory beyond %ldGB", NBPML4 / GB);
170
171	if (vmm_supports_1G_pages())
172		superpage_size = NBPDP;
173	else
174		superpage_size = NBPDR;
175
176	/*
177	 * Get the page directory pointer page that contains the direct
178	 * map address mappings.
179	 */
180	pml4p = kernel_pmap->pm_pml4;
181	pdp = (pdp_entry_t *)PHYS_TO_DMAP(pml4p[DMPML4I] & ~PAGE_MASK);
182
183	page_attr_bits = PG_RW | PG_V | PG_PS | PG_G;
184	addr = start;
185	while (addr < end) {
186		remaining = end - addr;
187		pdpi = addr / NBPDP;
188		if (superpage_size == NBPDP &&
189		    remaining >= NBPDP &&
190		    addr % NBPDP == 0) {
191			/*
192			 * If there isn't a mapping for this address then
193			 * create one but if there is one already make sure
194			 * it matches what we expect it to be.
195			 */
196			if (pdp[pdpi] == 0) {
197				pdp[pdpi] = addr | page_attr_bits;
198				if (0 && bootverbose) {
199					printf("vmm_mem_populate: mapping "
200					       "0x%lx with 1GB page at "
201					       "pdpi %d\n", addr, pdpi);
202				}
203			} else {
204				pdp_entry_t pdpe = pdp[pdpi];
205				if ((pdpe & ~PAGE_MASK) != addr ||
206				    (pdpe & page_attr_bits) != page_attr_bits) {
207					panic("An invalid mapping 0x%016lx "
208					      "already exists for 0x%016lx\n",
209					      pdpe, addr);
210				}
211			}
212			addr += NBPDP;
213		} else {
214			if (remaining < NBPDR) {
215				panic("vmm_mem_populate: remaining (%ld) must "
216				      "be greater than NBPDR (%d)\n",
217				      remaining, NBPDR);
218			}
219			if (pdp[pdpi] == 0) {
220				/*
221				 * XXX we lose this memory forever because
222				 * we do not keep track of the virtual address
223				 * that would be required to free this page.
224				 */
225				pd = malloc(PAGE_SIZE, M_VMM_MEM,
226					    M_WAITOK | M_ZERO);
227				if ((uintptr_t)pd & PAGE_MASK) {
228					panic("vmm_mem_populate: page directory"
229					      "page not aligned on %d "
230					      "boundary\n", PAGE_SIZE);
231				}
232				pdp[pdpi] = vtophys(pd);
233				pdp[pdpi] |= PG_RW | PG_V | PG_U;
234				if (0 && bootverbose) {
235					printf("Creating page directory "
236					       "at pdp index %d for 0x%016lx\n",
237					       pdpi, addr);
238				}
239			}
240			pdi = (addr % NBPDP) / NBPDR;
241			pd = (pd_entry_t *)PHYS_TO_DMAP(pdp[pdpi] & ~PAGE_MASK);
242
243			/*
244			 * Create a new mapping if one doesn't already exist
245			 * or validate it if it does.
246			 */
247			if (pd[pdi] == 0) {
248				pd[pdi] = addr | page_attr_bits;
249				if (0 && bootverbose) {
250					printf("vmm_mem_populate: mapping "
251					       "0x%lx with 2MB page at "
252					       "pdpi %d, pdi %d\n",
253					       addr, pdpi, pdi);
254				}
255			} else {
256				pd_entry_t pde = pd[pdi];
257				if ((pde & ~PAGE_MASK) != addr ||
258				    (pde & page_attr_bits) != page_attr_bits) {
259					panic("An invalid mapping 0x%016lx "
260					      "already exists for 0x%016lx\n",
261					      pde, addr);
262				}
263			}
264			addr += NBPDR;
265		}
266	}
267}
268
269static int
270vmm_mem_populate(void)
271{
272	int seg, error;
273	vm_paddr_t start, end;
274
275	/* populate the vmm_mem_avail[] array */
276	error = vmm_mem_steal_memory();
277	if (error)
278		return (error);
279
280	/*
281	 * Now map the memory that was hidden from FreeBSD in
282	 * the direct map VA space.
283	 */
284	for (seg = 0; seg < vmm_mem_nsegs; seg++) {
285		start = vmm_mem_avail[seg].base;
286		end = start + vmm_mem_avail[seg].length;
287		if ((start & PDRMASK) != 0 || (end & PDRMASK) != 0) {
288			panic("start (0x%016lx) and end (0x%016lx) must be "
289			      "aligned on a %dMB boundary\n",
290			      start, end, NBPDR / MB);
291		}
292		vmm_mem_direct_map(start, end);
293	}
294
295	return (0);
296}
297
298int
299vmm_mem_init(void)
300{
301	int error;
302
303	mtx_init(&vmm_mem_mtx, "vmm_mem_mtx", NULL, MTX_DEF);
304
305	error = vmm_mem_populate();
306	if (error)
307		return (error);
308
309	return (0);
310}
311
312vm_paddr_t
313vmm_mem_alloc(size_t size)
314{
315	int i;
316	vm_paddr_t addr;
317
318	if ((size & PDRMASK) != 0) {
319		panic("vmm_mem_alloc: size 0x%0lx must be "
320		      "aligned on a 0x%0x boundary\n", size, NBPDR);
321	}
322
323	addr = 0;
324
325	mtx_lock(&vmm_mem_mtx);
326	for (i = 0; i < vmm_mem_nsegs; i++) {
327		if (vmm_mem_avail[i].length >= size) {
328			addr = vmm_mem_avail[i].base;
329			vmm_mem_avail[i].base += size;
330			vmm_mem_avail[i].length -= size;
331			/* remove a zero length segment */
332			if (vmm_mem_avail[i].length == 0) {
333				memmove(&vmm_mem_avail[i],
334					&vmm_mem_avail[i + 1],
335					(vmm_mem_nsegs - (i + 1)) *
336					 sizeof(vmm_mem_avail[0]));
337				vmm_mem_nsegs--;
338			}
339			break;
340		}
341	}
342	mtx_unlock(&vmm_mem_mtx);
343
344	return (addr);
345}
346
347void
348vmm_mem_free(vm_paddr_t base, size_t length)
349{
350	int i;
351
352	if ((base & PDRMASK) != 0 || (length & PDRMASK) != 0) {
353		panic("vmm_mem_free: base 0x%0lx and length 0x%0lx must be "
354		      "aligned on a 0x%0x boundary\n", base, length, NBPDR);
355	}
356
357	mtx_lock(&vmm_mem_mtx);
358
359	for (i = 0; i < vmm_mem_nsegs; i++) {
360		if (vmm_mem_avail[i].base > base)
361			break;
362	}
363
364	if (vmm_mem_nsegs >= VMM_MEM_MAXSEGS)
365		panic("vmm_mem_free: cannot free any more segments");
366
367	/* Create a new segment at index 'i' */
368	memmove(&vmm_mem_avail[i + 1], &vmm_mem_avail[i],
369		(vmm_mem_nsegs - i) * sizeof(vmm_mem_avail[0]));
370
371	vmm_mem_avail[i].base = base;
372	vmm_mem_avail[i].length = length;
373
374	vmm_mem_nsegs++;
375
376coalesce_some_more:
377	for (i = 0; i < vmm_mem_nsegs - 1; i++) {
378		if (vmm_mem_avail[i].base + vmm_mem_avail[i].length ==
379		    vmm_mem_avail[i + 1].base) {
380			vmm_mem_avail[i].length += vmm_mem_avail[i + 1].length;
381			memmove(&vmm_mem_avail[i + 1], &vmm_mem_avail[i + 2],
382			  (vmm_mem_nsegs - (i + 2)) * sizeof(vmm_mem_avail[0]));
383			vmm_mem_nsegs--;
384			goto coalesce_some_more;
385		}
386	}
387
388	mtx_unlock(&vmm_mem_mtx);
389}
390
391vm_paddr_t
392vmm_mem_maxaddr(void)
393{
394
395	return (maxaddr);
396}
397
398void
399vmm_mem_dump(void)
400{
401	int i;
402	vm_paddr_t base;
403	vm_size_t length;
404
405	mtx_lock(&vmm_mem_mtx);
406	for (i = 0; i < vmm_mem_nsegs; i++) {
407		base = vmm_mem_avail[i].base;
408		length = vmm_mem_avail[i].length;
409		printf("%-4d0x%016lx    0x%016lx\n", i, base, base + length);
410	}
411	mtx_unlock(&vmm_mem_mtx);
412}
413