xref: /freebsd/sys/i386/include/pmap.h (revision bfe691b2f75de2224c7ceb304ebcdef2b42d4179)
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_I		0x10	/* during an instruction fetch */
92 
93 /*
94  * Size of Kernel address space.  This is the number of page table pages
95  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
96  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
97  */
98 #ifndef KVA_PAGES
99 #ifdef PAE
100 #define KVA_PAGES	512
101 #else
102 #define KVA_PAGES	256
103 #endif
104 #endif
105 
106 /*
107  * Pte related macros
108  */
109 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
110 
111 /* Initial number of kernel page tables. */
112 #ifndef NKPT
113 #ifdef PAE
114 /* 152 page tables needed to map 16G (76B "struct vm_page", 2M page tables). */
115 #define	NKPT		240
116 #else
117 /* 18 page tables needed to map 4G (72B "struct vm_page", 4M page tables). */
118 #define	NKPT		30
119 #endif
120 #endif
121 
122 #ifndef NKPDE
123 #ifdef SMP
124 #define NKPDE	(KVA_PAGES - 1) /* number of page tables/pde's */
125 #else
126 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
127 #endif
128 #endif
129 
130 /*
131  * The *PTDI values control the layout of virtual memory
132  *
133  * XXX This works for now, but I am not real happy with it, I'll fix it
134  * right after I fix locore.s and the magic 28K hole
135  *
136  * SMP_PRIVPAGES: The per-cpu address space is 0xff80000 -> 0xffbfffff
137  */
138 #ifdef SMP
139 #define MPPTDI		(NPDEPTD-1)	/* per cpu ptd entry */
140 #define	KPTDI		(MPPTDI-NKPDE)	/* start of kernel virtual pde's */
141 #else
142 #define	KPTDI		(NPDEPTD-NKPDE)/* start of kernel virtual pde's */
143 #endif	/* SMP */
144 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
145 
146 /*
147  * XXX doesn't really belong here I guess...
148  */
149 #define ISA_HOLE_START    0xa0000
150 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
151 
152 #ifndef LOCORE
153 
154 #include <sys/queue.h>
155 #include <sys/_lock.h>
156 #include <sys/_mutex.h>
157 
158 #ifdef PAE
159 
160 typedef uint64_t pdpt_entry_t;
161 typedef uint64_t pd_entry_t;
162 typedef uint64_t pt_entry_t;
163 
164 #define	PTESHIFT	(3)
165 #define	PDESHIFT	(3)
166 
167 #else
168 
169 typedef uint32_t pd_entry_t;
170 typedef uint32_t pt_entry_t;
171 
172 #define	PTESHIFT	(2)
173 #define	PDESHIFT	(2)
174 
175 #endif
176 
177 /*
178  * Address of current and alternate address space page table maps
179  * and directories.
180  */
181 #ifdef _KERNEL
182 extern pt_entry_t PTmap[];
183 extern pd_entry_t PTD[];
184 extern pd_entry_t PTDpde[];
185 
186 #ifdef PAE
187 extern pdpt_entry_t *IdlePDPT;
188 #endif
189 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
190 #endif
191 
192 #ifdef _KERNEL
193 /*
194  * virtual address to page table entry and
195  * to physical address.
196  * Note: these work recursively, thus vtopte of a pte will give
197  * the corresponding pde that in turn maps it.
198  */
199 #define	vtopte(va)	(PTmap + i386_btop(va))
200 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
201 
202 /*
203  *	Routine:	pmap_kextract
204  *	Function:
205  *		Extract the physical page address associated
206  *		kernel virtual address.
207  */
208 static __inline vm_paddr_t
209 pmap_kextract(vm_offset_t va)
210 {
211 	vm_paddr_t pa;
212 
213 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
214 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
215 	} else {
216 		pa = *vtopte(va);
217 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
218 	}
219 	return pa;
220 }
221 
222 #ifdef PAE
223 
224 static __inline pt_entry_t
225 pte_load(pt_entry_t *ptep)
226 {
227 	pt_entry_t r;
228 
229 	__asm __volatile(
230 	    "lock; cmpxchg8b %1"
231 	    : "=A" (r)
232 	    : "m" (*ptep), "a" (0), "d" (0), "b" (0), "c" (0));
233 	return (r);
234 }
235 
236 static __inline pt_entry_t
237 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
238 {
239 	pt_entry_t r;
240 
241 	r = *ptep;
242 	__asm __volatile(
243 	    "1:\n"
244 	    "\tlock; cmpxchg8b %1\n"
245 	    "\tjnz 1b"
246 	    : "+A" (r)
247 	    : "m" (*ptep), "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)));
248 	return (r);
249 }
250 
251 /* XXXRU move to atomic.h? */
252 static __inline int
253 atomic_cmpset_64(volatile uint64_t *dst, uint64_t exp, uint64_t src)
254 {
255 	int64_t res = exp;
256 
257 	__asm __volatile (
258 	"	lock ;			"
259 	"	cmpxchg8b %2 ;		"
260 	"	setz	%%al ;		"
261 	"	movzbl	%%al,%0 ;	"
262 	"# atomic_cmpset_64"
263 	: "+A" (res),			/* 0 (result) */
264 	  "=m" (*dst)			/* 1 */
265 	: "m" (*dst),			/* 2 */
266 	  "b" ((uint32_t)src),
267 	  "c" ((uint32_t)(src >> 32)));
268 
269 	return (res);
270 }
271 
272 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
273 
274 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
275 
276 extern pt_entry_t pg_nx;
277 
278 #else /* PAE */
279 
280 static __inline pt_entry_t
281 pte_load(pt_entry_t *ptep)
282 {
283 	pt_entry_t r;
284 
285 	r = *ptep;
286 	return (r);
287 }
288 
289 static __inline pt_entry_t
290 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
291 {
292 	pt_entry_t r;
293 
294 	__asm __volatile(
295 	    "xchgl %0,%1"
296 	    : "=m" (*ptep),
297 	      "=r" (r)
298 	    : "1" (pte),
299 	      "m" (*ptep));
300 	return (r);
301 }
302 
303 #define	pte_load_clear(pte)	atomic_readandclear_int(pte)
304 
305 static __inline void
306 pte_store(pt_entry_t *ptep, pt_entry_t pte)
307 {
308 
309 	*ptep = pte;
310 }
311 
312 #endif /* PAE */
313 
314 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
315 
316 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
317 
318 #endif /* _KERNEL */
319 
320 /*
321  * Pmap stuff
322  */
323 struct	pv_entry;
324 struct	pv_chunk;
325 
326 struct md_page {
327 	int pv_list_count;
328 	TAILQ_HEAD(,pv_entry)	pv_list;
329 };
330 
331 struct pmap {
332 	struct mtx		pm_mtx;
333 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
334 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
335 	u_int			pm_active;	/* active on cpus */
336 	struct pmap_statistics	pm_stats;	/* pmap statistics */
337 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
338 #ifdef PAE
339 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
340 						   table */
341 #endif
342 };
343 
344 typedef struct pmap	*pmap_t;
345 
346 #ifdef _KERNEL
347 extern struct pmap	kernel_pmap_store;
348 #define kernel_pmap	(&kernel_pmap_store)
349 
350 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
351 #define	PMAP_LOCK_ASSERT(pmap, type) \
352 				mtx_assert(&(pmap)->pm_mtx, (type))
353 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
354 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
355 				    NULL, MTX_DEF | MTX_DUPOK)
356 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
357 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
358 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
359 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
360 #endif
361 
362 /*
363  * For each vm_page_t, there is a list of all currently valid virtual
364  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
365  */
366 typedef struct pv_entry {
367 	vm_offset_t	pv_va;		/* virtual address for mapping */
368 	TAILQ_ENTRY(pv_entry)	pv_list;
369 } *pv_entry_t;
370 
371 /*
372  * pv_entries are allocated in chunks per-process.  This avoids the
373  * need to track per-pmap assignments.
374  */
375 #define	_NPCM	11
376 #define	_NPCPV	336
377 struct pv_chunk {
378 	pmap_t			pc_pmap;
379 	TAILQ_ENTRY(pv_chunk)	pc_list;
380 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
381 	uint32_t		pc_spare[2];
382 	struct pv_entry		pc_pventry[_NPCPV];
383 };
384 
385 #ifdef	_KERNEL
386 
387 #define NPPROVMTRR		8
388 #define PPRO_VMTRRphysBase0	0x200
389 #define PPRO_VMTRRphysMask0	0x201
390 struct ppro_vmtrr {
391 	u_int64_t base, mask;
392 };
393 extern struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR];
394 
395 extern caddr_t	CADDR1;
396 extern pt_entry_t *CMAP1;
397 extern vm_paddr_t phys_avail[];
398 extern vm_paddr_t dump_avail[];
399 extern int pseflag;
400 extern int pgeflag;
401 extern char *ptvmmap;		/* poor name! */
402 extern vm_offset_t virtual_avail;
403 extern vm_offset_t virtual_end;
404 
405 #define	pmap_page_is_mapped(m)	(!TAILQ_EMPTY(&(m)->md.pv_list))
406 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
407 
408 void	pmap_bootstrap(vm_paddr_t);
409 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
410 void	pmap_init_pat(void);
411 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
412 void	pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode);
413 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
414 void	pmap_kremove(vm_offset_t);
415 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
416 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
417 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
418 void	pmap_unmapdev(vm_offset_t, vm_size_t);
419 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
420 void	pmap_set_pg(void);
421 void	pmap_invalidate_page(pmap_t, vm_offset_t);
422 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
423 void	pmap_invalidate_all(pmap_t);
424 void	pmap_invalidate_cache(void);
425 
426 #endif /* _KERNEL */
427 
428 #endif /* !LOCORE */
429 
430 #endif /* !_MACHINE_PMAP_H_ */
431