xref: /freebsd/sys/i386/include/pmap.h (revision a7496c776d77fa07a5fcf3b73e16cb649982521d)
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 /* Page level cache control fields used to determine the PAT type */
85 #define PG_PDE_CACHE	(PG_PDE_PAT | PG_NC_PWT | PG_NC_PCD)
86 #define PG_PTE_CACHE	(PG_PTE_PAT | PG_NC_PWT | PG_NC_PCD)
87 
88 /*
89  * Promotion to a 2 or 4MB (PDE) page mapping requires that the corresponding
90  * 4KB (PTE) page mappings have identical settings for the following fields:
91  */
92 #define PG_PTE_PROMOTE	(PG_MANAGED | PG_W | PG_G | PG_PTE_PAT | \
93 	    PG_M | PG_A | PG_NC_PCD | PG_NC_PWT | PG_U | PG_RW | PG_V)
94 
95 /*
96  * Page Protection Exception bits
97  */
98 
99 #define PGEX_P		0x01	/* Protection violation vs. not present */
100 #define PGEX_W		0x02	/* during a Write cycle */
101 #define PGEX_U		0x04	/* access from User mode (UPL) */
102 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
103 #define PGEX_I		0x10	/* during an instruction fetch */
104 
105 /*
106  * Size of Kernel address space.  This is the number of page table pages
107  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
108  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
109  * For PAE, the page table page unit size is 2MB.  This means that 512 pages
110  * is 1 Gigabyte.  Double everything.  It must be a multiple of 8 for PAE.
111  */
112 #ifndef KVA_PAGES
113 #ifdef PAE
114 #define KVA_PAGES	512
115 #else
116 #define KVA_PAGES	256
117 #endif
118 #endif
119 
120 /*
121  * Pte related macros
122  */
123 #define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDRSHIFT)|((pti)<<PAGE_SHIFT)))
124 
125 /*
126  * The initial number of kernel page table pages that are constructed
127  * by locore must be sufficient to map vm_page_array.  That number can
128  * be calculated as follows:
129  *     max_phys / PAGE_SIZE * sizeof(struct vm_page) / NBPDR
130  * PAE:      max_phys 16G, sizeof(vm_page) 76, NBPDR 2M, 152 page table pages.
131  * Non-PAE:  max_phys 4G,  sizeof(vm_page) 68, NBPDR 4M, 18 page table pages.
132  */
133 #ifndef NKPT
134 #ifdef PAE
135 #define	NKPT		240
136 #else
137 #define	NKPT		30
138 #endif
139 #endif
140 
141 #ifndef NKPDE
142 #define NKPDE	(KVA_PAGES)	/* number of page tables/pde's */
143 #endif
144 
145 /*
146  * The *PTDI values control the layout of virtual memory
147  *
148  * XXX This works for now, but I am not real happy with it, I'll fix it
149  * right after I fix locore.s and the magic 28K hole
150  */
151 #define	KPTDI		(NPDEPTD-NKPDE)	/* start of kernel virtual pde's */
152 #define	PTDPTDI		(KPTDI-NPGPTD)	/* ptd entry that points to ptd! */
153 
154 /*
155  * XXX doesn't really belong here I guess...
156  */
157 #define ISA_HOLE_START    0xa0000
158 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
159 
160 #ifndef LOCORE
161 
162 #include <sys/queue.h>
163 #include <sys/_cpuset.h>
164 #include <sys/_lock.h>
165 #include <sys/_mutex.h>
166 
167 #include <vm/_vm_radix.h>
168 
169 #ifdef PAE
170 
171 typedef uint64_t pdpt_entry_t;
172 typedef uint64_t pd_entry_t;
173 typedef uint64_t pt_entry_t;
174 
175 #define	PTESHIFT	(3)
176 #define	PDESHIFT	(3)
177 
178 #else
179 
180 typedef uint32_t pd_entry_t;
181 typedef uint32_t pt_entry_t;
182 
183 #define	PTESHIFT	(2)
184 #define	PDESHIFT	(2)
185 
186 #endif
187 
188 /*
189  * Address of current address space page table maps and directories.
190  */
191 #ifdef _KERNEL
192 extern pt_entry_t PTmap[];
193 extern pd_entry_t PTD[];
194 extern pd_entry_t PTDpde[];
195 
196 #ifdef PAE
197 extern pdpt_entry_t *IdlePDPT;
198 #endif
199 extern pd_entry_t *IdlePTD;	/* physical address of "Idle" state directory */
200 
201 /*
202  * Translate a virtual address to the kernel virtual address of its page table
203  * entry (PTE).  This can be used recursively.  If the address of a PTE as
204  * previously returned by this macro is itself given as the argument, then the
205  * address of the page directory entry (PDE) that maps the PTE will be
206  * returned.
207  *
208  * This macro may be used before pmap_bootstrap() is called.
209  */
210 #define	vtopte(va)	(PTmap + i386_btop(va))
211 
212 /*
213  * Translate a virtual address to its physical address.
214  *
215  * This macro may be used before pmap_bootstrap() is called.
216  */
217 #define	vtophys(va)	pmap_kextract((vm_offset_t)(va))
218 
219 #if defined(XEN)
220 #include <sys/param.h>
221 
222 #include <xen/xen-os.h>
223 
224 #include <machine/xen/xenvar.h>
225 #include <machine/xen/xenpmap.h>
226 
227 extern pt_entry_t pg_nx;
228 
229 #define PG_KERNEL  (PG_V | PG_A | PG_RW | PG_M)
230 
231 #define MACH_TO_VM_PAGE(ma) PHYS_TO_VM_PAGE(xpmap_mtop((ma)))
232 #define VM_PAGE_TO_MACH(m) xpmap_ptom(VM_PAGE_TO_PHYS((m)))
233 
234 #define VTOM(va) xpmap_ptom(VTOP(va))
235 
236 static __inline vm_paddr_t
237 pmap_kextract_ma(vm_offset_t va)
238 {
239         vm_paddr_t ma;
240         if ((ma = PTD[va >> PDRSHIFT]) & PG_PS) {
241                 ma = (ma & ~(NBPDR - 1)) | (va & (NBPDR - 1));
242         } else {
243                 ma = (*vtopte(va) & PG_FRAME) | (va & PAGE_MASK);
244         }
245         return ma;
246 }
247 
248 static __inline vm_paddr_t
249 pmap_kextract(vm_offset_t va)
250 {
251         return xpmap_mtop(pmap_kextract_ma(va));
252 }
253 #define vtomach(va)     pmap_kextract_ma(((vm_offset_t) (va)))
254 
255 vm_paddr_t pmap_extract_ma(struct pmap *pmap, vm_offset_t va);
256 
257 void    pmap_kenter_ma(vm_offset_t va, vm_paddr_t pa);
258 void    pmap_map_readonly(struct pmap *pmap, vm_offset_t va, int len);
259 void    pmap_map_readwrite(struct pmap *pmap, vm_offset_t va, int len);
260 
261 static __inline pt_entry_t
262 pte_load_store(pt_entry_t *ptep, pt_entry_t v)
263 {
264 	pt_entry_t r;
265 
266 	r = *ptep;
267 	PT_SET_VA(ptep, v, TRUE);
268 	return (r);
269 }
270 
271 static __inline pt_entry_t
272 pte_load_store_ma(pt_entry_t *ptep, pt_entry_t v)
273 {
274 	pt_entry_t r;
275 
276 	r = *ptep;
277 	PT_SET_VA_MA(ptep, v, TRUE);
278 	return (r);
279 }
280 
281 #define	pte_load_clear(ptep)	pte_load_store((ptep), (pt_entry_t)0ULL)
282 
283 #define	pte_store(ptep, pte)	pte_load_store((ptep), (pt_entry_t)pte)
284 #define	pte_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
285 #define	pde_store_ma(ptep, pte)	pte_load_store_ma((ptep), (pt_entry_t)pte)
286 
287 #elif !defined(XEN)
288 
289 /*
290  * KPTmap is a linear mapping of the kernel page table.  It differs from the
291  * recursive mapping in two ways: (1) it only provides access to kernel page
292  * table pages, and not user page table pages, and (2) it provides access to
293  * a kernel page table page after the corresponding virtual addresses have
294  * been promoted to a 2/4MB page mapping.
295  *
296  * KPTmap is first initialized by locore to support just NPKT page table
297  * pages.  Later, it is reinitialized by pmap_bootstrap() to allow for
298  * expansion of the kernel page table.
299  */
300 extern pt_entry_t *KPTmap;
301 
302 /*
303  * Extract from the kernel page table the physical address that is mapped by
304  * the given virtual address "va".
305  *
306  * This function may be used before pmap_bootstrap() is called.
307  */
308 static __inline vm_paddr_t
309 pmap_kextract(vm_offset_t va)
310 {
311 	vm_paddr_t pa;
312 
313 	if ((pa = PTD[va >> PDRSHIFT]) & PG_PS) {
314 		pa = (pa & PG_PS_FRAME) | (va & PDRMASK);
315 	} else {
316 		/*
317 		 * Beware of a concurrent promotion that changes the PDE at
318 		 * this point!  For example, vtopte() must not be used to
319 		 * access the PTE because it would use the new PDE.  It is,
320 		 * however, safe to use the old PDE because the page table
321 		 * page is preserved by the promotion.
322 		 */
323 		pa = KPTmap[i386_btop(va)];
324 		pa = (pa & PG_FRAME) | (va & PAGE_MASK);
325 	}
326 	return (pa);
327 }
328 #endif
329 
330 #if !defined(XEN)
331 #define PT_UPDATES_FLUSH()
332 #endif
333 
334 #if defined(PAE) && !defined(XEN)
335 
336 #define	pde_cmpset(pdep, old, new)	atomic_cmpset_64_i586(pdep, old, new)
337 #define	pte_load_store(ptep, pte)	atomic_swap_64_i586(ptep, pte)
338 #define	pte_load_clear(ptep)		atomic_swap_64_i586(ptep, 0)
339 #define	pte_store(ptep, pte)		atomic_store_rel_64_i586(ptep, pte)
340 
341 extern pt_entry_t pg_nx;
342 
343 #elif !defined(PAE) && !defined(XEN)
344 
345 #define	pde_cmpset(pdep, old, new)	atomic_cmpset_int(pdep, old, new)
346 #define	pte_load_store(ptep, pte)	atomic_swap_int(ptep, pte)
347 #define	pte_load_clear(ptep)		atomic_swap_int(ptep, 0)
348 #define	pte_store(ptep, pte) do { \
349 	*(u_int *)(ptep) = (u_int)(pte); \
350 } while (0)
351 
352 #endif /* PAE */
353 
354 #define	pte_clear(ptep)			pte_store(ptep, 0)
355 
356 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
357 
358 #endif /* _KERNEL */
359 
360 /*
361  * Pmap stuff
362  */
363 struct	pv_entry;
364 struct	pv_chunk;
365 
366 struct md_page {
367 	TAILQ_HEAD(,pv_entry)	pv_list;
368 	int			pat_mode;
369 };
370 
371 struct pmap {
372 	struct mtx		pm_mtx;
373 	pd_entry_t		*pm_pdir;	/* KVA of page directory */
374 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
375 	cpuset_t		pm_active;	/* active on cpus */
376 	struct pmap_statistics	pm_stats;	/* pmap statistics */
377 	LIST_ENTRY(pmap) 	pm_list;	/* List of all pmaps */
378 #ifdef PAE
379 	pdpt_entry_t		*pm_pdpt;	/* KVA of page director pointer
380 						   table */
381 #endif
382 	struct vm_radix		pm_root;	/* spare page table pages */
383 };
384 
385 typedef struct pmap	*pmap_t;
386 
387 #ifdef _KERNEL
388 extern struct pmap	kernel_pmap_store;
389 #define kernel_pmap	(&kernel_pmap_store)
390 
391 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
392 #define	PMAP_LOCK_ASSERT(pmap, type) \
393 				mtx_assert(&(pmap)->pm_mtx, (type))
394 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
395 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
396 				    NULL, MTX_DEF | MTX_DUPOK)
397 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
398 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
399 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
400 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
401 #endif
402 
403 /*
404  * For each vm_page_t, there is a list of all currently valid virtual
405  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
406  */
407 typedef struct pv_entry {
408 	vm_offset_t	pv_va;		/* virtual address for mapping */
409 	TAILQ_ENTRY(pv_entry)	pv_next;
410 } *pv_entry_t;
411 
412 /*
413  * pv_entries are allocated in chunks per-process.  This avoids the
414  * need to track per-pmap assignments.
415  */
416 #define	_NPCM	11
417 #define	_NPCPV	336
418 struct pv_chunk {
419 	pmap_t			pc_pmap;
420 	TAILQ_ENTRY(pv_chunk)	pc_list;
421 	uint32_t		pc_map[_NPCM];	/* bitmap; 1 = free */
422 	TAILQ_ENTRY(pv_chunk)	pc_lru;
423 	struct pv_entry		pc_pventry[_NPCPV];
424 };
425 
426 #ifdef	_KERNEL
427 
428 extern caddr_t	CADDR3;
429 extern pt_entry_t *CMAP3;
430 extern vm_paddr_t phys_avail[];
431 extern vm_paddr_t dump_avail[];
432 extern int pseflag;
433 extern int pgeflag;
434 extern char *ptvmmap;		/* poor name! */
435 extern vm_offset_t virtual_avail;
436 extern vm_offset_t virtual_end;
437 
438 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
439 #define	pmap_page_is_write_mapped(m)	(((m)->aflags & PGA_WRITEABLE) != 0)
440 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
441 
442 /*
443  * Only the following functions or macros may be used before pmap_bootstrap()
444  * is called: pmap_kenter(), pmap_kextract(), pmap_kremove(), vtophys(), and
445  * vtopte().
446  */
447 void	pmap_bootstrap(vm_paddr_t);
448 int	pmap_cache_bits(int mode, boolean_t is_pde);
449 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
450 void	pmap_init_pat(void);
451 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
452 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
453 void	pmap_kremove(vm_offset_t);
454 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
455 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
456 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
457 boolean_t pmap_page_is_mapped(vm_page_t m);
458 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
459 void	pmap_unmapdev(vm_offset_t, vm_size_t);
460 pt_entry_t *pmap_pte(pmap_t, vm_offset_t) __pure2;
461 void	pmap_invalidate_page(pmap_t, vm_offset_t);
462 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
463 void	pmap_invalidate_all(pmap_t);
464 void	pmap_invalidate_cache(void);
465 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
466 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva,
467 	    boolean_t force);
468 
469 #endif /* _KERNEL */
470 
471 #endif /* !LOCORE */
472 
473 #endif /* !_MACHINE_PMAP_H_ */
474