xref: /netbsd-src/sys/arch/amd64/include/pmap.h (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: pmap.h,v 1.46 2018/05/19 15:03:26 jdolecek 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 and it
88  * uses the same recursive entry scheme. See the i386 pmap.h
89  * for a description. The obvious difference is that 3 extra
90  * levels of page table need to be dealt with. The level 1 page
91  * table pages are at:
92  *
93  * l1: 0x00007f8000000000 - 0x00007fffffffffff     (39 bits, needs PML4 entry)
94  *
95  * The rest is kept as physical pages in 3 UVM objects, and is
96  * temporarily mapped for virtual access when needed.
97  *
98  * Note that address space is signed, so the layout for 48 bits is:
99  *
100  *  +---------------------------------+ 0xffffffffffffffff
101  *  |                                 |
102  *  |         Unused                  |
103  *  |                                 |
104  *  +---------------------------------+ 0xffffff8000000000
105  *  ~                                 ~
106  *  |                                 |
107  *  |         Kernel Space            |
108  *  |                                 |
109  *  |                                 |
110  *  +---------------------------------+ 0xffff800000000000 = 0x0000800000000000
111  *  |                                 |
112  *  |    alt.L1 table (PTE pages)     |
113  *  |                                 |
114  *  +---------------------------------+ 0x00007f8000000000
115  *  ~                                 ~
116  *  |                                 |
117  *  |         User Space              |
118  *  |                                 |
119  *  |                                 |
120  *  +---------------------------------+ 0x0000000000000000
121  *
122  * In other words, there is a 'VA hole' at 0x0000800000000000 -
123  * 0xffff800000000000 which will trap, just as on, for example,
124  * sparcv9.
125  *
126  * The unused space can be used if needed, but it adds a little more
127  * complexity to the calculations.
128  */
129 
130 /*
131  * Mask to get rid of the sign-extended part of addresses.
132  */
133 #define VA_SIGN_MASK		0xffff000000000000
134 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
135 /*
136  * XXXfvdl this one's not right.
137  */
138 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
139 
140 #define L4_SLOT_PTE		255
141 #ifndef XEN
142 #define L4_SLOT_KERN		256 /* pl4_i(VM_MIN_KERNEL_ADDRESS) */
143 #else
144 /* Xen use slots 256-272, let's move farther */
145 #define L4_SLOT_KERN		320 /* pl4_i(VM_MIN_KERNEL_ADDRESS) */
146 #endif
147 #define L4_SLOT_KERNBASE	511 /* pl4_i(KERNBASE) */
148 
149 #define PDIR_SLOT_KERN	L4_SLOT_KERN
150 #define PDIR_SLOT_PTE	L4_SLOT_PTE
151 
152 /*
153  * The following defines give the virtual addresses of various MMU
154  * data structures:
155  * PTE_BASE: the base VA of the linear PTE mappings
156  * PDP_BASE: the base VA of the recursive mapping of the PTD
157  * PDP_PDE: the VA of the PDE that points back to the PDP
158  */
159 
160 #define PTE_BASE	((pt_entry_t *)(L4_SLOT_PTE * NBPD_L4))
161 #define KERN_BASE	((pt_entry_t *)(L4_SLOT_KERN * NBPD_L4))
162 
163 #define L1_BASE	PTE_BASE
164 #define L2_BASE	((pd_entry_t *)((char *)L1_BASE + L4_SLOT_PTE * NBPD_L3))
165 #define L3_BASE	((pd_entry_t *)((char *)L2_BASE + L4_SLOT_PTE * NBPD_L2))
166 #define L4_BASE	((pd_entry_t *)((char *)L3_BASE + L4_SLOT_PTE * NBPD_L1))
167 
168 #define PDP_PDE		(L4_BASE + PDIR_SLOT_PTE)
169 
170 #define PDP_BASE	L4_BASE
171 
172 #define NKL4_MAX_ENTRIES	(unsigned long)64
173 #define NKL3_MAX_ENTRIES	(unsigned long)(NKL4_MAX_ENTRIES * 512)
174 #define NKL2_MAX_ENTRIES	(unsigned long)(NKL3_MAX_ENTRIES * 512)
175 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * 512)
176 
177 #define NKL4_KIMG_ENTRIES	1
178 #define NKL3_KIMG_ENTRIES	1
179 #define NKL2_KIMG_ENTRIES	48
180 
181 /*
182  * Since kva space is below the kernel in its entirety, we start off
183  * with zero entries on each level.
184  */
185 #define NKL4_START_ENTRIES	0
186 #define NKL3_START_ENTRIES	0
187 #define NKL2_START_ENTRIES	0
188 #define NKL1_START_ENTRIES	0	/* XXX */
189 
190 #define NTOPLEVEL_PDES		(PAGE_SIZE / (sizeof (pd_entry_t)))
191 
192 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
193 
194 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME, L3_FRAME, L4_FRAME }
195 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT, L3_SHIFT, L4_SHIFT }
196 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES, \
197 				  NKL3_START_ENTRIES, NKL4_START_ENTRIES }
198 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES, \
199 				  NKL3_MAX_ENTRIES, NKL4_MAX_ENTRIES }
200 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2, NBPD_L3, NBPD_L4 }
201 #define PDES_INITIALIZER	{ L2_BASE, L3_BASE, L4_BASE }
202 
203 #define PTP_LEVELS	4
204 
205 /*
206  * PG_AVAIL usage: we make use of the ignored bits of the PTE
207  */
208 
209 #define PG_W		PG_AVAIL1	/* "wired" mapping */
210 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
211 /* PG_AVAIL3 not used */
212 
213 #define	PG_X		0		/* XXX dummy */
214 
215 /*
216  * Number of PTE's per cache line.  8 byte pte, 64-byte cache line
217  * Used to avoid false sharing of cache lines.
218  */
219 #define NPTECL		8
220 
221 void svs_pmap_sync(struct pmap *, int);
222 void svs_lwp_switch(struct lwp *, struct lwp *);
223 void svs_pdir_switch(struct pmap *);
224 void svs_init(void);
225 extern bool svs_enabled;
226 
227 #include <x86/pmap.h>
228 
229 #ifndef XEN
230 #define pmap_pa2pte(a)			(a)
231 #define pmap_pte2pa(a)			((a) & PG_FRAME)
232 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
233 #define pmap_pte_cas(p, o, n)		atomic_cas_64((p), (o), (n))
234 #define pmap_pte_testset(p, n)		\
235     atomic_swap_ulong((volatile unsigned long *)p, n)
236 #define pmap_pte_setbits(p, b)		\
237     atomic_or_ulong((volatile unsigned long *)p, b)
238 #define pmap_pte_clearbits(p, b)	\
239     atomic_and_ulong((volatile unsigned long *)p, ~(b))
240 #define pmap_pte_flush()		/* nothing */
241 #else
242 extern kmutex_t pte_lock;
243 
244 static __inline pt_entry_t
245 pmap_pa2pte(paddr_t pa)
246 {
247 	return (pt_entry_t)xpmap_ptom_masked(pa);
248 }
249 
250 static __inline paddr_t
251 pmap_pte2pa(pt_entry_t pte)
252 {
253 	return xpmap_mtop_masked(pte & PG_FRAME);
254 }
255 
256 static __inline void
257 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
258 {
259 	int s = splvm();
260 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
261 	splx(s);
262 }
263 
264 static __inline pt_entry_t
265 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
266 {
267 	pt_entry_t opte;
268 
269 	mutex_enter(&pte_lock);
270 	opte = *ptep;
271 	if (opte == o) {
272 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
273 		xpq_flush_queue();
274 	}
275 
276 	mutex_exit(&pte_lock);
277 	return opte;
278 }
279 
280 static __inline pt_entry_t
281 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
282 {
283 	pt_entry_t opte;
284 
285 	mutex_enter(&pte_lock);
286 	opte = *pte;
287 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), npte);
288 	xpq_flush_queue();
289 	mutex_exit(&pte_lock);
290 	return opte;
291 }
292 
293 static __inline void
294 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
295 {
296 	mutex_enter(&pte_lock);
297 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
298 	xpq_flush_queue();
299 	mutex_exit(&pte_lock);
300 }
301 
302 static __inline void
303 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
304 {
305 	mutex_enter(&pte_lock);
306 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
307 	    (*pte) & ~bits);
308 	xpq_flush_queue();
309 	mutex_exit(&pte_lock);
310 }
311 
312 static __inline void
313 pmap_pte_flush(void)
314 {
315 	int s = splvm();
316 	xpq_flush_queue();
317 	splx(s);
318 }
319 #endif
320 
321 #ifdef __HAVE_DIRECT_MAP
322 #define PMAP_DIRECT
323 
324 static __inline int
325 pmap_direct_process(paddr_t pa, voff_t pgoff, size_t len,
326     int (*process)(void *, size_t, void *), void *arg)
327 {
328 	vaddr_t va = PMAP_DIRECT_MAP(pa);
329 
330 	return process((void *)(va + pgoff), len, arg);
331 }
332 
333 #endif /* __HAVE_DIRECT_MAP */
334 
335 void pmap_changeprot_local(vaddr_t, vm_prot_t);
336 
337 #include <x86/pmap_pv.h>
338 
339 #define	__HAVE_VM_PAGE_MD
340 #define	VM_MDPAGE_INIT(pg) \
341 	memset(&(pg)->mdpage, 0, sizeof((pg)->mdpage)); \
342 	PMAP_PAGE_INIT(&(pg)->mdpage.mp_pp)
343 
344 struct vm_page_md {
345 	struct pmap_page mp_pp;
346 };
347 
348 #else	/*	!__x86_64__	*/
349 
350 #include <i386/pmap.h>
351 
352 #endif	/*	__x86_64__	*/
353 
354 #endif	/* _AMD64_PMAP_H_ */
355