xref: /freebsd/sys/i386/include/pmap.h (revision 1669d8afc64812c8d2d1d147ae1fd42ff441e1b1)
1 /*-
2  * Copyright (c) 1991 Regents of the University of California.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * the Systems Programming Group of the University of Utah Computer
7  * Science Department and William Jolitz of UUNET Technologies Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * Derived from hp300 version by Mike Hibler, this version by William
34  * Jolitz uses a recursive map [a pde points to the page directory] to
35  * map the page tables using the pagetables themselves. This is done to
36  * reduce the impact on kernel virtual memory for lots of sparse address
37  * space, and to reduce the cost of memory to each process.
38  *
39  *	from: hp300: @(#)pmap.h	7.2 (Berkeley) 12/16/90
40  *	from: @(#)pmap.h	7.4 (Berkeley) 5/12/91
41  * $FreeBSD$
42  */
43 
44 #ifndef _MACHINE_PMAP_H_
45 #define	_MACHINE_PMAP_H_
46 
47 /*
48  * Page-directory and page-table entries follow this format, with a few
49  * of the fields not present here and there, depending on a lot of things.
50  */
51 				/* ---- Intel Nomenclature ---- */
52 #define	PG_V		0x001	/* P	Valid			*/
53 #define PG_RW		0x002	/* R/W	Read/Write		*/
54 #define PG_U		0x004	/* U/S  User/Supervisor		*/
55 #define	PG_NC_PWT	0x008	/* PWT	Write through		*/
56 #define	PG_NC_PCD	0x010	/* PCD	Cache disable		*/
57 #define PG_A		0x020	/* A	Accessed		*/
58 #define	PG_M		0x040	/* D	Dirty			*/
59 #define	PG_PS		0x080	/* PS	Page size (0=4k,1=4M)	*/
60 #define	PG_PTE_PAT	0x080	/* PAT	PAT index		*/
61 #define	PG_G		0x100	/* G	Global			*/
62 #define	PG_AVAIL1	0x200	/*    /	Available for system	*/
63 #define	PG_AVAIL2	0x400	/*   <	programmers use		*/
64 #define	PG_AVAIL3	0x800	/*    \				*/
65 #define	PG_PDE_PAT	0x1000	/* PAT	PAT index		*/
66 #ifdef PAE
67 #define	PG_NX		(1ull<<63) /* No-execute */
68 #endif
69 
70 
71 /* Our various interpretations of the above */
72 #define PG_W		PG_AVAIL1	/* "Wired" pseudoflag */
73 #define	PG_MANAGED	PG_AVAIL2
74 #ifdef PAE
75 #define	PG_FRAME	(0x000ffffffffff000ull)
76 #define	PG_PS_FRAME	(0x000fffffffe00000ull)
77 #else
78 #define	PG_FRAME	(~PAGE_MASK)
79 #define	PG_PS_FRAME	(0xffc00000)
80 #endif
81 #define	PG_PROT		(PG_RW|PG_U)	/* all protection bits . */
82 #define PG_N		(PG_NC_PWT|PG_NC_PCD)	/* Non-cacheable */
83 
84 /*
85  * Page Protection Exception bits
86  */
87 
88 #define PGEX_P		0x01	/* Protection violation vs. not present */
89 #define PGEX_W		0x02	/* during a Write cycle */
90 #define PGEX_U		0x04	/* access from User mode (UPL) */
91 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
92 #define PGEX_I		0x10	/* during an instruction fetch */
93 
94 /*
95  * Size of Kernel address space.  This is the number of page table pages
96  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
97  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
98  * For PAE, the page table page unit size is 2MB.  This means that 512 pages
99  * is 1 Gigabyte.  Double everything.  It must be a multiple of 8 for PAE.
100  */
101 #ifndef KVA_PAGES
102 #ifdef PAE
103 #define KVA_PAGES	512
104 #else
105 #define KVA_PAGES	256
106 #endif
107 #endif
108 
109 /*
110  * Pte related macros
111  */
112 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
113 
114 /* Initial number of kernel page tables. */
115 #ifndef NKPT
116 #ifdef PAE
117 /* 152 page tables needed to map 16G (76B "struct vm_page", 2M page tables). */
118 #define	NKPT		240
119 #else
120 /* 18 page tables needed to map 4G (72B "struct vm_page", 4M page tables). */
121 #define	NKPT		30
122 #endif
123 #endif
124 
125 #ifndef NKPDE
126 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
127 #endif
128 
129 /*
130  * The *PTDI values control the layout of virtual memory
131  *
132  * XXX This works for now, but I am not real happy with it, I'll fix it
133  * right after I fix locore.s and the magic 28K hole
134  */
135 #define	KPTDI		(NPDEPTD-NKPDE)	/* start of kernel virtual pde's */
136 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
137 
138 /*
139  * XXX doesn't really belong here I guess...
140  */
141 #define ISA_HOLE_START    0xa0000
142 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
143 
144 #ifndef LOCORE
145 
146 #include <sys/queue.h>
147 #include <sys/_lock.h>
148 #include <sys/_mutex.h>
149 
150 #ifdef PAE
151 
152 typedef uint64_t pdpt_entry_t;
153 typedef uint64_t pd_entry_t;
154 typedef uint64_t pt_entry_t;
155 
156 #define	PTESHIFT	(3)
157 #define	PDESHIFT	(3)
158 
159 #else
160 
161 typedef uint32_t pd_entry_t;
162 typedef uint32_t pt_entry_t;
163 
164 #define	PTESHIFT	(2)
165 #define	PDESHIFT	(2)
166 
167 #endif
168 
169 /*
170  * Address of current and alternate address space page table maps
171  * and directories.
172  */
173 #ifdef _KERNEL
174 extern pt_entry_t PTmap[];
175 extern pd_entry_t PTD[];
176 extern pd_entry_t PTDpde[];
177 
178 #ifdef PAE
179 extern pdpt_entry_t *IdlePDPT;
180 #endif
181 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
182 #endif
183 
184 #ifdef _KERNEL
185 /*
186  * virtual address to page table entry and
187  * to physical address.
188  * Note: these work recursively, thus vtopte of a pte will give
189  * the corresponding pde that in turn maps it.
190  */
191 #define	vtopte(va)	(PTmap + i386_btop(va))
192 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
193 
194 /*
195  *	Routine:	pmap_kextract
196  *	Function:
197  *		Extract the physical page address associated
198  *		kernel virtual address.
199  */
200 static __inline vm_paddr_t
201 pmap_kextract(vm_offset_t va)
202 {
203 	vm_paddr_t pa;
204 
205 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
206 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
207 	} else {
208 		pa = *vtopte(va);
209 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
210 	}
211 	return pa;
212 }
213 
214 #ifdef PAE
215 
216 static __inline pt_entry_t
217 pte_load(pt_entry_t *ptep)
218 {
219 	pt_entry_t r;
220 
221 	__asm __volatile(
222 	    "lock; cmpxchg8b %1"
223 	    : "=A" (r)
224 	    : "m" (*ptep), "a" (0), "d" (0), "b" (0), "c" (0));
225 	return (r);
226 }
227 
228 static __inline pt_entry_t
229 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
230 {
231 	pt_entry_t r;
232 
233 	r = *ptep;
234 	__asm __volatile(
235 	    "1:\n"
236 	    "\tlock; cmpxchg8b %1\n"
237 	    "\tjnz 1b"
238 	    : "+A" (r)
239 	    : "m" (*ptep), "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)));
240 	return (r);
241 }
242 
243 /* XXXRU move to atomic.h? */
244 static __inline int
245 atomic_cmpset_64(volatile uint64_t *dst, uint64_t exp, uint64_t src)
246 {
247 	int64_t res = exp;
248 
249 	__asm __volatile (
250 	"	lock ;			"
251 	"	cmpxchg8b %2 ;		"
252 	"	setz	%%al ;		"
253 	"	movzbl	%%al,%0 ;	"
254 	"# atomic_cmpset_64"
255 	: "+A" (res),			/* 0 (result) */
256 	  "=m" (*dst)			/* 1 */
257 	: "m" (*dst),			/* 2 */
258 	  "b" ((uint32_t)src),
259 	  "c" ((uint32_t)(src >> 32)));
260 
261 	return (res);
262 }
263 
264 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
265 
266 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
267 
268 extern pt_entry_t pg_nx;
269 
270 #else /* PAE */
271 
272 static __inline pt_entry_t
273 pte_load(pt_entry_t *ptep)
274 {
275 	pt_entry_t r;
276 
277 	r = *ptep;
278 	return (r);
279 }
280 
281 static __inline pt_entry_t
282 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
283 {
284 	pt_entry_t r;
285 
286 	__asm __volatile(
287 	    "xchgl %0,%1"
288 	    : "=m" (*ptep),
289 	      "=r" (r)
290 	    : "1" (pte),
291 	      "m" (*ptep));
292 	return (r);
293 }
294 
295 #define	pte_load_clear(pte)	atomic_readandclear_int(pte)
296 
297 static __inline void
298 pte_store(pt_entry_t *ptep, pt_entry_t pte)
299 {
300 
301 	*ptep = pte;
302 }
303 
304 #endif /* PAE */
305 
306 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
307 
308 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
309 
310 #endif /* _KERNEL */
311 
312 /*
313  * Pmap stuff
314  */
315 struct	pv_entry;
316 struct	pv_chunk;
317 
318 struct md_page {
319 	TAILQ_HEAD(,pv_entry)	pv_list;
320 };
321 
322 struct pmap {
323 	struct mtx		pm_mtx;
324 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
325 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
326 	u_int			pm_active;	/* active on cpus */
327 	struct pmap_statistics	pm_stats;	/* pmap statistics */
328 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
329 #ifdef PAE
330 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
331 						   table */
332 #endif
333 };
334 
335 typedef struct pmap	*pmap_t;
336 
337 #ifdef _KERNEL
338 extern struct pmap	kernel_pmap_store;
339 #define kernel_pmap	(&kernel_pmap_store)
340 
341 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
342 #define	PMAP_LOCK_ASSERT(pmap, type) \
343 				mtx_assert(&(pmap)->pm_mtx, (type))
344 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
345 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
346 				    NULL, MTX_DEF | MTX_DUPOK)
347 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
348 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
349 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
350 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
351 #endif
352 
353 /*
354  * For each vm_page_t, there is a list of all currently valid virtual
355  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
356  */
357 typedef struct pv_entry {
358 	vm_offset_t	pv_va;		/* virtual address for mapping */
359 	TAILQ_ENTRY(pv_entry)	pv_list;
360 } *pv_entry_t;
361 
362 /*
363  * pv_entries are allocated in chunks per-process.  This avoids the
364  * need to track per-pmap assignments.
365  */
366 #define	_NPCM	11
367 #define	_NPCPV	336
368 struct pv_chunk {
369 	pmap_t			pc_pmap;
370 	TAILQ_ENTRY(pv_chunk)	pc_list;
371 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
372 	uint32_t		pc_spare[2];
373 	struct pv_entry		pc_pventry[_NPCPV];
374 };
375 
376 #ifdef	_KERNEL
377 
378 #define NPPROVMTRR		8
379 #define PPRO_VMTRRphysBase0	0x200
380 #define PPRO_VMTRRphysMask0	0x201
381 struct ppro_vmtrr {
382 	u_int64_t base, mask;
383 };
384 extern struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR];
385 
386 extern caddr_t	CADDR1;
387 extern pt_entry_t *CMAP1;
388 extern vm_paddr_t phys_avail[];
389 extern vm_paddr_t dump_avail[];
390 extern int pseflag;
391 extern int pgeflag;
392 extern char *ptvmmap;		/* poor name! */
393 extern vm_offset_t virtual_avail;
394 extern vm_offset_t virtual_end;
395 
396 #define	pmap_page_is_mapped(m)	(!TAILQ_EMPTY(&(m)->md.pv_list))
397 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
398 
399 void	pmap_bootstrap(vm_paddr_t);
400 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
401 void	pmap_init_pat(void);
402 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
403 void	pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode);
404 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
405 void	pmap_kremove(vm_offset_t);
406 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
407 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
408 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
409 void	pmap_unmapdev(vm_offset_t, vm_size_t);
410 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
411 void	pmap_set_pg(void);
412 void	pmap_invalidate_page(pmap_t, vm_offset_t);
413 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
414 void	pmap_invalidate_all(pmap_t);
415 void	pmap_invalidate_cache(void);
416 
417 #endif /* _KERNEL */
418 
419 #endif /* !LOCORE */
420 
421 #endif /* !_MACHINE_PMAP_H_ */
422