xref: /netbsd-src/sys/arch/amd64/include/pmap.h (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*	$NetBSD: pmap.h,v 1.24 2011/02/01 20:09:08 chuck Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
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 ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*
29  * Copyright (c) 2001 Wasabi Systems, Inc.
30  * All rights reserved.
31  *
32  * Written by Frank van der Linden for Wasabi Systems, Inc.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  * 3. All advertising materials mentioning features or use of this software
43  *    must display the following acknowledgement:
44  *      This product includes software developed for the NetBSD Project by
45  *      Wasabi Systems, Inc.
46  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
47  *    or promote products derived from this software without specific prior
48  *    written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
52  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
53  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
54  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
55  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
56  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
57  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
58  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
59  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
60  * POSSIBILITY OF SUCH DAMAGE.
61  */
62 
63 #ifndef	_AMD64_PMAP_H_
64 #define	_AMD64_PMAP_H_
65 
66 #ifdef __x86_64__
67 
68 #if defined(_KERNEL_OPT)
69 #include "opt_xen.h"
70 #endif
71 
72 #include <sys/atomic.h>
73 
74 #include <machine/pte.h>
75 #include <machine/segments.h>
76 #ifdef _KERNEL
77 #include <machine/cpufunc.h>
78 #endif
79 
80 #include <uvm/uvm_object.h>
81 #ifdef XEN
82 #include <xen/xenfunc.h>
83 #include <xen/xenpmap.h>
84 #endif /* XEN */
85 
86 /*
87  * The x86_64 pmap module closely resembles the i386 one. It uses
88  * the same recursive entry scheme, and the same alternate area
89  * trick for accessing non-current pmaps. See the i386 pmap.h
90  * for a description. The obvious difference is that 3 extra
91  * levels of page table need to be dealt with. The level 1 page
92  * table pages are at:
93  *
94  * l1: 0x00007f8000000000 - 0x00007fffffffffff     (39 bits, needs PML4 entry)
95  *
96  * The alternate space is at:
97  *
98  * l1: 0xffffff8000000000 - 0xffffffffffffffff     (39 bits, needs PML4 entry)
99  *
100  * The rest is kept as physical pages in 3 UVM objects, and is
101  * temporarily mapped for virtual access when needed.
102  *
103  * Note that address space is signed, so the layout for 48 bits is:
104  *
105  *  +---------------------------------+ 0xffffffffffffffff
106  *  |                                 |
107  *  |    alt.L1 table (PTE pages)     |
108  *  |                                 |
109  *  +---------------------------------+ 0xffffff8000000000
110  *  ~                                 ~
111  *  |                                 |
112  *  |         Kernel Space            |
113  *  |                                 |
114  *  |                                 |
115  *  +---------------------------------+ 0xffff800000000000 = 0x0000800000000000
116  *  |                                 |
117  *  |    alt.L1 table (PTE pages)     |
118  *  |                                 |
119  *  +---------------------------------+ 0x00007f8000000000
120  *  ~                                 ~
121  *  |                                 |
122  *  |         User Space              |
123  *  |                                 |
124  *  |                                 |
125  *  +---------------------------------+ 0x0000000000000000
126  *
127  * In other words, there is a 'VA hole' at 0x0000800000000000 -
128  * 0xffff800000000000 which will trap, just as on, for example,
129  * sparcv9.
130  *
131  * The unused space can be used if needed, but it adds a little more
132  * complexity to the calculations.
133  */
134 
135 /*
136  * The first generation of Hammer processors can use 48 bits of
137  * virtual memory, and 40 bits of physical memory. This will be
138  * more for later generations. These defines can be changed to
139  * variable names containing the # of bits, extracted from an
140  * extended cpuid instruction (variables are harder to use during
141  * bootstrap, though)
142  */
143 #define VIRT_BITS	48
144 #define PHYS_BITS	40
145 
146 /*
147  * Mask to get rid of the sign-extended part of addresses.
148  */
149 #define VA_SIGN_MASK		0xffff000000000000
150 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
151 /*
152  * XXXfvdl this one's not right.
153  */
154 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
155 
156 #define L4_SLOT_PTE		255
157 #ifndef XEN
158 #define L4_SLOT_KERN		256
159 #else
160 /* Xen use slots 256-272, let's move farther */
161 #define L4_SLOT_KERN		320
162 #endif
163 #define L4_SLOT_KERNBASE	511
164 #define L4_SLOT_APTE		510
165 
166 #define PDIR_SLOT_KERN	L4_SLOT_KERN
167 #define PDIR_SLOT_PTE	L4_SLOT_PTE
168 #define PDIR_SLOT_APTE	L4_SLOT_APTE
169 
170 /*
171  * the following defines give the virtual addresses of various MMU
172  * data structures:
173  * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
174  * PTD_BASE and APTD_BASE: the base VA of the recursive mapping of the PTD
175  * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
176  *
177  */
178 
179 #define PTE_BASE  ((pt_entry_t *) (L4_SLOT_PTE * NBPD_L4))
180 #define APTE_BASE ((pt_entry_t *) (VA_SIGN_NEG((L4_SLOT_APTE * NBPD_L4))))
181 
182 #define L1_BASE		PTE_BASE
183 #define AL1_BASE	APTE_BASE
184 
185 #define L2_BASE ((pd_entry_t *)((char *)L1_BASE + L4_SLOT_PTE * NBPD_L3))
186 #define L3_BASE ((pd_entry_t *)((char *)L2_BASE + L4_SLOT_PTE * NBPD_L2))
187 #define L4_BASE ((pd_entry_t *)((char *)L3_BASE + L4_SLOT_PTE * NBPD_L1))
188 
189 #define AL2_BASE ((pd_entry_t *)((char *)AL1_BASE + L4_SLOT_PTE * NBPD_L3))
190 #define AL3_BASE ((pd_entry_t *)((char *)AL2_BASE + L4_SLOT_PTE * NBPD_L2))
191 #define AL4_BASE ((pd_entry_t *)((char *)AL3_BASE + L4_SLOT_PTE * NBPD_L1))
192 
193 #define PDP_PDE		(L4_BASE + PDIR_SLOT_PTE)
194 #define APDP_PDE	(L4_BASE + PDIR_SLOT_APTE)
195 
196 #define PDP_BASE	L4_BASE
197 #define APDP_BASE	AL4_BASE
198 
199 #define NKL4_MAX_ENTRIES	(unsigned long)1
200 #define NKL3_MAX_ENTRIES	(unsigned long)(NKL4_MAX_ENTRIES * 512)
201 #define NKL2_MAX_ENTRIES	(unsigned long)(NKL3_MAX_ENTRIES * 512)
202 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * 512)
203 
204 #define NKL4_KIMG_ENTRIES	1
205 #define NKL3_KIMG_ENTRIES	1
206 #define NKL2_KIMG_ENTRIES	10
207 
208 /*
209  * Since kva space is below the kernel in its entirety, we start off
210  * with zero entries on each level.
211  */
212 #define NKL4_START_ENTRIES	0
213 #define NKL3_START_ENTRIES	0
214 #define NKL2_START_ENTRIES	0
215 #define NKL1_START_ENTRIES	0	/* XXX */
216 
217 #define NTOPLEVEL_PDES		(PAGE_SIZE / (sizeof (pd_entry_t)))
218 
219 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
220 
221 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME, L3_FRAME, L4_FRAME }
222 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT, L3_SHIFT, L4_SHIFT }
223 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES, \
224 				  NKL3_START_ENTRIES, NKL4_START_ENTRIES }
225 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES, \
226 				  NKL3_MAX_ENTRIES, NKL4_MAX_ENTRIES }
227 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2, NBPD_L3, NBPD_L4 }
228 #define PDES_INITIALIZER	{ L2_BASE, L3_BASE, L4_BASE }
229 #define APDES_INITIALIZER	{ AL2_BASE, AL3_BASE, AL4_BASE }
230 
231 #define PTP_LEVELS	4
232 
233 /*
234  * PG_AVAIL usage: we make use of the ignored bits of the PTE
235  */
236 
237 #define PG_W		PG_AVAIL1	/* "wired" mapping */
238 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
239 /* PG_AVAIL3 not used */
240 
241 #define	PG_X		0		/* XXX dummy */
242 
243 /*
244  * Number of PTE's per cache line.  8 byte pte, 64-byte cache line
245  * Used to avoid false sharing of cache lines.
246  */
247 #define NPTECL		8
248 
249 #include <x86/pmap.h>
250 
251 #ifndef XEN
252 #define pmap_pa2pte(a)			(a)
253 #define pmap_pte2pa(a)			((a) & PG_FRAME)
254 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
255 #define pmap_pte_cas(p, o, n)		atomic_cas_64((p), (o), (n))
256 #define pmap_pte_testset(p, n)		\
257     atomic_swap_ulong((volatile unsigned long *)p, n)
258 #define pmap_pte_setbits(p, b)		\
259     atomic_or_ulong((volatile unsigned long *)p, b)
260 #define pmap_pte_clearbits(p, b)	\
261     atomic_and_ulong((volatile unsigned long *)p, ~(b))
262 #define pmap_pte_flush()		/* nothing */
263 #else
264 static __inline pt_entry_t
265 pmap_pa2pte(paddr_t pa)
266 {
267 	return (pt_entry_t)xpmap_ptom_masked(pa);
268 }
269 
270 static __inline paddr_t
271 pmap_pte2pa(pt_entry_t pte)
272 {
273 	return xpmap_mtop_masked(pte & PG_FRAME);
274 }
275 static __inline void
276 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
277 {
278 	int s = splvm();
279 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
280 	splx(s);
281 }
282 
283 static __inline pt_entry_t
284 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
285 {
286 	int s = splvm();
287 	pt_entry_t opte = *ptep;
288 
289 	if (opte == o) {
290 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
291 		xpq_flush_queue();
292 	}
293 	splx(s);
294 	return opte;
295 }
296 
297 static __inline pt_entry_t
298 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
299 {
300 	int s = splvm();
301 	pt_entry_t opte = *pte;
302 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), npte);
303 	xpq_flush_queue();
304 	splx(s);
305 	return opte;
306 }
307 
308 static __inline void
309 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
310 {
311 	int s = splvm();
312 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
313 	xpq_flush_queue();
314 	splx(s);
315 }
316 
317 static __inline void
318 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
319 {
320 	int s = splvm();
321 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
322 	    (*pte) & ~bits);
323 	xpq_flush_queue();
324 	splx(s);
325 }
326 
327 static __inline void
328 pmap_pte_flush(void)
329 {
330 	int s = splvm();
331 	xpq_flush_queue();
332 	splx(s);
333 }
334 #endif
335 
336 void pmap_prealloc_lowmem_ptps(void);
337 void pmap_changeprot_local(vaddr_t, vm_prot_t);
338 
339 #include <x86/pmap_pv.h>
340 
341 #define	__HAVE_VM_PAGE_MD
342 #define	VM_MDPAGE_INIT(pg) \
343 	memset(&(pg)->mdpage, 0, sizeof((pg)->mdpage)); \
344 	PMAP_PAGE_INIT(&(pg)->mdpage.mp_pp)
345 
346 struct vm_page_md {
347 	struct pmap_page mp_pp;
348 };
349 
350 #else	/*	!__x86_64__	*/
351 
352 #include <i386/pmap.h>
353 
354 #endif	/*	__x86_64__	*/
355 
356 #endif	/* _AMD64_PMAP_H_ */
357