xref: /freebsd/sys/i386/include/pmap.h (revision 282a3889ebf826db9839be296ff1dd903f6d6d6e)
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  */
99 #ifndef KVA_PAGES
100 #ifdef PAE
101 #define KVA_PAGES	512
102 #else
103 #define KVA_PAGES	256
104 #endif
105 #endif
106 
107 /*
108  * Pte related macros
109  */
110 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
111 
112 /* Initial number of kernel page tables. */
113 #ifndef NKPT
114 #ifdef PAE
115 /* 152 page tables needed to map 16G (76B "struct vm_page", 2M page tables). */
116 #define	NKPT		240
117 #else
118 /* 18 page tables needed to map 4G (72B "struct vm_page", 4M page tables). */
119 #define	NKPT		30
120 #endif
121 #endif
122 
123 #ifndef NKPDE
124 #ifdef SMP
125 #define NKPDE	(KVA_PAGES - 1) /* number of page tables/pde's */
126 #else
127 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
128 #endif
129 #endif
130 
131 /*
132  * The *PTDI values control the layout of virtual memory
133  *
134  * XXX This works for now, but I am not real happy with it, I'll fix it
135  * right after I fix locore.s and the magic 28K hole
136  *
137  * SMP_PRIVPAGES: The per-cpu address space is 0xff80000 -> 0xffbfffff
138  */
139 #ifdef SMP
140 #define MPPTDI		(NPDEPTD-1)	/* per cpu ptd entry */
141 #define	KPTDI		(MPPTDI-NKPDE)	/* start of kernel virtual pde's */
142 #else
143 #define	KPTDI		(NPDEPTD-NKPDE)/* start of kernel virtual pde's */
144 #endif	/* SMP */
145 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
146 
147 /*
148  * XXX doesn't really belong here I guess...
149  */
150 #define ISA_HOLE_START    0xa0000
151 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
152 
153 #ifndef LOCORE
154 
155 #include <sys/queue.h>
156 #include <sys/_lock.h>
157 #include <sys/_mutex.h>
158 
159 #ifdef PAE
160 
161 typedef uint64_t pdpt_entry_t;
162 typedef uint64_t pd_entry_t;
163 typedef uint64_t pt_entry_t;
164 
165 #define	PTESHIFT	(3)
166 #define	PDESHIFT	(3)
167 
168 #else
169 
170 typedef uint32_t pd_entry_t;
171 typedef uint32_t pt_entry_t;
172 
173 #define	PTESHIFT	(2)
174 #define	PDESHIFT	(2)
175 
176 #endif
177 
178 /*
179  * Address of current and alternate address space page table maps
180  * and directories.
181  */
182 #ifdef _KERNEL
183 extern pt_entry_t PTmap[];
184 extern pd_entry_t PTD[];
185 extern pd_entry_t PTDpde[];
186 
187 #ifdef PAE
188 extern pdpt_entry_t *IdlePDPT;
189 #endif
190 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
191 #endif
192 
193 #ifdef _KERNEL
194 /*
195  * virtual address to page table entry and
196  * to physical address.
197  * Note: these work recursively, thus vtopte of a pte will give
198  * the corresponding pde that in turn maps it.
199  */
200 #define	vtopte(va)	(PTmap + i386_btop(va))
201 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
202 
203 /*
204  *	Routine:	pmap_kextract
205  *	Function:
206  *		Extract the physical page address associated
207  *		kernel virtual address.
208  */
209 static __inline vm_paddr_t
210 pmap_kextract(vm_offset_t va)
211 {
212 	vm_paddr_t pa;
213 
214 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
215 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
216 	} else {
217 		pa = *vtopte(va);
218 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
219 	}
220 	return pa;
221 }
222 
223 #ifdef PAE
224 
225 static __inline pt_entry_t
226 pte_load(pt_entry_t *ptep)
227 {
228 	pt_entry_t r;
229 
230 	__asm __volatile(
231 	    "lock; cmpxchg8b %1"
232 	    : "=A" (r)
233 	    : "m" (*ptep), "a" (0), "d" (0), "b" (0), "c" (0));
234 	return (r);
235 }
236 
237 static __inline pt_entry_t
238 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
239 {
240 	pt_entry_t r;
241 
242 	r = *ptep;
243 	__asm __volatile(
244 	    "1:\n"
245 	    "\tlock; cmpxchg8b %1\n"
246 	    "\tjnz 1b"
247 	    : "+A" (r)
248 	    : "m" (*ptep), "b" ((uint32_t)v), "c" ((uint32_t)(v >> 32)));
249 	return (r);
250 }
251 
252 /* XXXRU move to atomic.h? */
253 static __inline int
254 atomic_cmpset_64(volatile uint64_t *dst, uint64_t exp, uint64_t src)
255 {
256 	int64_t res = exp;
257 
258 	__asm __volatile (
259 	"	lock ;			"
260 	"	cmpxchg8b %2 ;		"
261 	"	setz	%%al ;		"
262 	"	movzbl	%%al,%0 ;	"
263 	"# atomic_cmpset_64"
264 	: "+A" (res),			/* 0 (result) */
265 	  "=m" (*dst)			/* 1 */
266 	: "m" (*dst),			/* 2 */
267 	  "b" ((uint32_t)src),
268 	  "c" ((uint32_t)(src >> 32)));
269 
270 	return (res);
271 }
272 
273 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
274 
275 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
276 
277 extern pt_entry_t pg_nx;
278 
279 #else /* PAE */
280 
281 static __inline pt_entry_t
282 pte_load(pt_entry_t *ptep)
283 {
284 	pt_entry_t r;
285 
286 	r = *ptep;
287 	return (r);
288 }
289 
290 static __inline pt_entry_t
291 pte_load_store(pt_entry_t *ptep, pt_entry_t pte)
292 {
293 	pt_entry_t r;
294 
295 	__asm __volatile(
296 	    "xchgl %0,%1"
297 	    : "=m" (*ptep),
298 	      "=r" (r)
299 	    : "1" (pte),
300 	      "m" (*ptep));
301 	return (r);
302 }
303 
304 #define	pte_load_clear(pte)	atomic_readandclear_int(pte)
305 
306 static __inline void
307 pte_store(pt_entry_t *ptep, pt_entry_t pte)
308 {
309 
310 	*ptep = pte;
311 }
312 
313 #endif /* PAE */
314 
315 #define	pte_clear(ptep)		pte_store((ptep), (pt_entry_t)0ULL)
316 
317 #define	pde_store(pdep, pde)	pte_store((pdep), (pde))
318 
319 #endif /* _KERNEL */
320 
321 /*
322  * Pmap stuff
323  */
324 struct	pv_entry;
325 struct	pv_chunk;
326 
327 struct md_page {
328 	int pv_list_count;
329 	TAILQ_HEAD(,pv_entry)	pv_list;
330 };
331 
332 struct pmap {
333 	struct mtx		pm_mtx;
334 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
335 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
336 	u_int			pm_active;	/* active on cpus */
337 	struct pmap_statistics	pm_stats;	/* pmap statistics */
338 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
339 #ifdef PAE
340 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
341 						   table */
342 #endif
343 };
344 
345 typedef struct pmap	*pmap_t;
346 
347 #ifdef _KERNEL
348 extern struct pmap	kernel_pmap_store;
349 #define kernel_pmap	(&kernel_pmap_store)
350 
351 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
352 #define	PMAP_LOCK_ASSERT(pmap, type) \
353 				mtx_assert(&(pmap)->pm_mtx, (type))
354 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
355 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
356 				    NULL, MTX_DEF | MTX_DUPOK)
357 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
358 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
359 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
360 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
361 #endif
362 
363 /*
364  * For each vm_page_t, there is a list of all currently valid virtual
365  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
366  */
367 typedef struct pv_entry {
368 	vm_offset_t	pv_va;		/* virtual address for mapping */
369 	TAILQ_ENTRY(pv_entry)	pv_list;
370 } *pv_entry_t;
371 
372 /*
373  * pv_entries are allocated in chunks per-process.  This avoids the
374  * need to track per-pmap assignments.
375  */
376 #define	_NPCM	11
377 #define	_NPCPV	336
378 struct pv_chunk {
379 	pmap_t			pc_pmap;
380 	TAILQ_ENTRY(pv_chunk)	pc_list;
381 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
382 	uint32_t		pc_spare[2];
383 	struct pv_entry		pc_pventry[_NPCPV];
384 };
385 
386 #ifdef	_KERNEL
387 
388 #define NPPROVMTRR		8
389 #define PPRO_VMTRRphysBase0	0x200
390 #define PPRO_VMTRRphysMask0	0x201
391 struct ppro_vmtrr {
392 	u_int64_t base, mask;
393 };
394 extern struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR];
395 
396 extern caddr_t	CADDR1;
397 extern pt_entry_t *CMAP1;
398 extern vm_paddr_t phys_avail[];
399 extern vm_paddr_t dump_avail[];
400 extern int pseflag;
401 extern int pgeflag;
402 extern char *ptvmmap;		/* poor name! */
403 extern vm_offset_t virtual_avail;
404 extern vm_offset_t virtual_end;
405 
406 #define	pmap_page_is_mapped(m)	(!TAILQ_EMPTY(&(m)->md.pv_list))
407 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
408 
409 void	pmap_bootstrap(vm_paddr_t);
410 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
411 void	pmap_init_pat(void);
412 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
413 void	pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode);
414 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
415 void	pmap_kremove(vm_offset_t);
416 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
417 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
418 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
419 void	pmap_unmapdev(vm_offset_t, vm_size_t);
420 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
421 void	pmap_set_pg(void);
422 void	pmap_invalidate_page(pmap_t, vm_offset_t);
423 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
424 void	pmap_invalidate_all(pmap_t);
425 void	pmap_invalidate_cache(void);
426 
427 #endif /* _KERNEL */
428 
429 #endif /* !LOCORE */
430 
431 #endif /* !_MACHINE_PMAP_H_ */
432