xref: /netbsd-src/sys/arch/amd64/include/pmap.h (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: pmap.h,v 1.22 2008/10/26 00:08:15 mrg Exp $	*/
2 
3 /*
4  *
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgment:
18  *      This product includes software developed by Charles D. Cranor and
19  *      Washington University.
20  * 4. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 2001 Wasabi Systems, Inc.
37  * All rights reserved.
38  *
39  * Written by Frank van der Linden for Wasabi Systems, Inc.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. All advertising materials mentioning features or use of this software
50  *    must display the following acknowledgement:
51  *      This product includes software developed for the NetBSD Project by
52  *      Wasabi Systems, Inc.
53  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
54  *    or promote products derived from this software without specific prior
55  *    written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
59  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
60  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
61  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
62  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
63  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
64  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
65  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
66  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
67  * POSSIBILITY OF SUCH DAMAGE.
68  */
69 
70 #ifndef	_AMD64_PMAP_H_
71 #define	_AMD64_PMAP_H_
72 
73 #ifdef __x86_64__
74 
75 #if defined(_KERNEL_OPT)
76 #include "opt_xen.h"
77 #endif
78 
79 #include <sys/atomic.h>
80 
81 #include <machine/pte.h>
82 #include <machine/segments.h>
83 #ifdef _KERNEL
84 #include <machine/cpufunc.h>
85 #endif
86 
87 #include <uvm/uvm_object.h>
88 #ifdef XEN
89 #include <xen/xenfunc.h>
90 #include <xen/xenpmap.h>
91 #endif /* XEN */
92 
93 /*
94  * The x86_64 pmap module closely resembles the i386 one. It uses
95  * the same recursive entry scheme, and the same alternate area
96  * trick for accessing non-current pmaps. See the i386 pmap.h
97  * for a description. The obvious difference is that 3 extra
98  * levels of page table need to be dealt with. The level 1 page
99  * table pages are at:
100  *
101  * l1: 0x00007f8000000000 - 0x00007fffffffffff     (39 bits, needs PML4 entry)
102  *
103  * The alternate space is at:
104  *
105  * l1: 0xffffff8000000000 - 0xffffffffffffffff     (39 bits, needs PML4 entry)
106  *
107  * The rest is kept as physical pages in 3 UVM objects, and is
108  * temporarily mapped for virtual access when needed.
109  *
110  * Note that address space is signed, so the layout for 48 bits is:
111  *
112  *  +---------------------------------+ 0xffffffffffffffff
113  *  |                                 |
114  *  |    alt.L1 table (PTE pages)     |
115  *  |                                 |
116  *  +---------------------------------+ 0xffffff8000000000
117  *  ~                                 ~
118  *  |                                 |
119  *  |         Kernel Space            |
120  *  |                                 |
121  *  |                                 |
122  *  +---------------------------------+ 0xffff800000000000 = 0x0000800000000000
123  *  |                                 |
124  *  |    alt.L1 table (PTE pages)     |
125  *  |                                 |
126  *  +---------------------------------+ 0x00007f8000000000
127  *  ~                                 ~
128  *  |                                 |
129  *  |         User Space              |
130  *  |                                 |
131  *  |                                 |
132  *  +---------------------------------+ 0x0000000000000000
133  *
134  * In other words, there is a 'VA hole' at 0x0000800000000000 -
135  * 0xffff800000000000 which will trap, just as on, for example,
136  * sparcv9.
137  *
138  * The unused space can be used if needed, but it adds a little more
139  * complexity to the calculations.
140  */
141 
142 /*
143  * The first generation of Hammer processors can use 48 bits of
144  * virtual memory, and 40 bits of physical memory. This will be
145  * more for later generations. These defines can be changed to
146  * variable names containing the # of bits, extracted from an
147  * extended cpuid instruction (variables are harder to use during
148  * bootstrap, though)
149  */
150 #define VIRT_BITS	48
151 #define PHYS_BITS	40
152 
153 /*
154  * Mask to get rid of the sign-extended part of addresses.
155  */
156 #define VA_SIGN_MASK		0xffff000000000000
157 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
158 /*
159  * XXXfvdl this one's not right.
160  */
161 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
162 
163 #define L4_SLOT_PTE		255
164 #ifndef XEN
165 #define L4_SLOT_KERN		256
166 #else
167 /* Xen use slots 256-272, let's move farther */
168 #define L4_SLOT_KERN		320
169 #endif
170 #define L4_SLOT_KERNBASE	511
171 #define L4_SLOT_APTE		510
172 
173 #define PDIR_SLOT_KERN	L4_SLOT_KERN
174 #define PDIR_SLOT_PTE	L4_SLOT_PTE
175 #define PDIR_SLOT_APTE	L4_SLOT_APTE
176 
177 /*
178  * the following defines give the virtual addresses of various MMU
179  * data structures:
180  * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
181  * PTD_BASE and APTD_BASE: the base VA of the recursive mapping of the PTD
182  * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
183  *
184  */
185 
186 #define PTE_BASE  ((pt_entry_t *) (L4_SLOT_PTE * NBPD_L4))
187 #define APTE_BASE ((pt_entry_t *) (VA_SIGN_NEG((L4_SLOT_APTE * NBPD_L4))))
188 
189 #define L1_BASE		PTE_BASE
190 #define AL1_BASE	APTE_BASE
191 
192 #define L2_BASE ((pd_entry_t *)((char *)L1_BASE + L4_SLOT_PTE * NBPD_L3))
193 #define L3_BASE ((pd_entry_t *)((char *)L2_BASE + L4_SLOT_PTE * NBPD_L2))
194 #define L4_BASE ((pd_entry_t *)((char *)L3_BASE + L4_SLOT_PTE * NBPD_L1))
195 
196 #define AL2_BASE ((pd_entry_t *)((char *)AL1_BASE + L4_SLOT_PTE * NBPD_L3))
197 #define AL3_BASE ((pd_entry_t *)((char *)AL2_BASE + L4_SLOT_PTE * NBPD_L2))
198 #define AL4_BASE ((pd_entry_t *)((char *)AL3_BASE + L4_SLOT_PTE * NBPD_L1))
199 
200 #define PDP_PDE		(L4_BASE + PDIR_SLOT_PTE)
201 #define APDP_PDE	(L4_BASE + PDIR_SLOT_APTE)
202 
203 #define PDP_BASE	L4_BASE
204 #define APDP_BASE	AL4_BASE
205 
206 #define NKL4_MAX_ENTRIES	(unsigned long)1
207 #define NKL3_MAX_ENTRIES	(unsigned long)(NKL4_MAX_ENTRIES * 512)
208 #define NKL2_MAX_ENTRIES	(unsigned long)(NKL3_MAX_ENTRIES * 512)
209 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * 512)
210 
211 #define NKL4_KIMG_ENTRIES	1
212 #define NKL3_KIMG_ENTRIES	1
213 #define NKL2_KIMG_ENTRIES	10
214 
215 /*
216  * Since kva space is below the kernel in its entirety, we start off
217  * with zero entries on each level.
218  */
219 #define NKL4_START_ENTRIES	0
220 #define NKL3_START_ENTRIES	0
221 #define NKL2_START_ENTRIES	0
222 #define NKL1_START_ENTRIES	0	/* XXX */
223 
224 #define NTOPLEVEL_PDES		(PAGE_SIZE / (sizeof (pd_entry_t)))
225 
226 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
227 
228 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME, L3_FRAME, L4_FRAME }
229 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT, L3_SHIFT, L4_SHIFT }
230 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES, \
231 				  NKL3_START_ENTRIES, NKL4_START_ENTRIES }
232 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES, \
233 				  NKL3_MAX_ENTRIES, NKL4_MAX_ENTRIES }
234 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2, NBPD_L3, NBPD_L4 }
235 #define PDES_INITIALIZER	{ L2_BASE, L3_BASE, L4_BASE }
236 #define APDES_INITIALIZER	{ AL2_BASE, AL3_BASE, AL4_BASE }
237 
238 #define PTP_LEVELS	4
239 
240 /*
241  * PG_AVAIL usage: we make use of the ignored bits of the PTE
242  */
243 
244 #define PG_W		PG_AVAIL1	/* "wired" mapping */
245 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
246 /* PG_AVAIL3 not used */
247 
248 #define	PG_X		0		/* XXX dummy */
249 
250 /*
251  * Number of PTE's per cache line.  8 byte pte, 64-byte cache line
252  * Used to avoid false sharing of cache lines.
253  */
254 #define NPTECL		8
255 
256 #include <x86/pmap.h>
257 
258 #ifndef XEN
259 #define pmap_pa2pte(a)			(a)
260 #define pmap_pte2pa(a)			((a) & PG_FRAME)
261 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
262 #define pmap_pte_cas(p, o, n)		atomic_cas_64((p), (o), (n))
263 #define pmap_pte_testset(p, n)		\
264     atomic_swap_ulong((volatile unsigned long *)p, n)
265 #define pmap_pte_setbits(p, b)		\
266     atomic_or_ulong((volatile unsigned long *)p, b)
267 #define pmap_pte_clearbits(p, b)	\
268     atomic_and_ulong((volatile unsigned long *)p, ~(b))
269 #define pmap_pte_flush()		/* nothing */
270 #else
271 static __inline pt_entry_t
272 pmap_pa2pte(paddr_t pa)
273 {
274 	return (pt_entry_t)xpmap_ptom_masked(pa);
275 }
276 
277 static __inline paddr_t
278 pmap_pte2pa(pt_entry_t pte)
279 {
280 	return xpmap_mtop_masked(pte & PG_FRAME);
281 }
282 static __inline void
283 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
284 {
285 	int s = splvm();
286 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
287 	splx(s);
288 }
289 
290 static __inline pt_entry_t
291 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
292 {
293 	int s = splvm();
294 	pt_entry_t opte = *ptep;
295 
296 	if (opte == o) {
297 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
298 		xpq_flush_queue();
299 	}
300 	splx(s);
301 	return opte;
302 }
303 
304 static __inline pt_entry_t
305 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
306 {
307 	int s = splvm();
308 	pt_entry_t opte = *pte;
309 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), npte);
310 	xpq_flush_queue();
311 	splx(s);
312 	return opte;
313 }
314 
315 static __inline void
316 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
317 {
318 	int s = splvm();
319 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
320 	xpq_flush_queue();
321 	splx(s);
322 }
323 
324 static __inline void
325 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
326 {
327 	int s = splvm();
328 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
329 	    (*pte) & ~bits);
330 	xpq_flush_queue();
331 	splx(s);
332 }
333 
334 static __inline void
335 pmap_pte_flush(void)
336 {
337 	int s = splvm();
338 	xpq_flush_queue();
339 	splx(s);
340 }
341 #endif
342 
343 void pmap_prealloc_lowmem_ptps(void);
344 void pmap_changeprot_local(vaddr_t, vm_prot_t);
345 
346 #else	/*	__x86_64__	*/
347 
348 #include <i386/pmap.h>
349 
350 #endif	/*	__x86_64__	*/
351 
352 #endif	/* _AMD64_PMAP_H_ */
353