xref: /netbsd-src/sys/arch/x86/include/pmap.h (revision 1b9578b8c2c1f848eeb16dabbfd7d1f0d9fdefbd)
1 /*	$NetBSD: pmap.h,v 1.40 2011/06/13 04:30:40 tls 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 /*
64  * pmap.h: see pmap.c for the history of this pmap module.
65  */
66 
67 #ifndef _X86_PMAP_H_
68 #define	_X86_PMAP_H_
69 
70 /*
71  * pl*_pi: index in the ptp page for a pde mapping a VA.
72  * (pl*_i below is the index in the virtual array of all pdes per level)
73  */
74 #define pl1_pi(VA)	(((VA_SIGN_POS(VA)) & L1_MASK) >> L1_SHIFT)
75 #define pl2_pi(VA)	(((VA_SIGN_POS(VA)) & L2_MASK) >> L2_SHIFT)
76 #define pl3_pi(VA)	(((VA_SIGN_POS(VA)) & L3_MASK) >> L3_SHIFT)
77 #define pl4_pi(VA)	(((VA_SIGN_POS(VA)) & L4_MASK) >> L4_SHIFT)
78 
79 /*
80  * pl*_i: generate index into pde/pte arrays in virtual space
81  *
82  * pl_i(va, X) == plX_i(va) <= pl_i_roundup(va, X)
83  */
84 #define pl1_i(VA)	(((VA_SIGN_POS(VA)) & L1_FRAME) >> L1_SHIFT)
85 #define pl2_i(VA)	(((VA_SIGN_POS(VA)) & L2_FRAME) >> L2_SHIFT)
86 #define pl3_i(VA)	(((VA_SIGN_POS(VA)) & L3_FRAME) >> L3_SHIFT)
87 #define pl4_i(VA)	(((VA_SIGN_POS(VA)) & L4_FRAME) >> L4_SHIFT)
88 #define pl_i(va, lvl) \
89         (((VA_SIGN_POS(va)) & ptp_masks[(lvl)-1]) >> ptp_shifts[(lvl)-1])
90 
91 #define	pl_i_roundup(va, lvl)	pl_i((va)+ ~ptp_masks[(lvl)-1], (lvl))
92 
93 /*
94  * PTP macros:
95  *   a PTP's index is the PD index of the PDE that points to it
96  *   a PTP's offset is the byte-offset in the PTE space that this PTP is at
97  *   a PTP's VA is the first VA mapped by that PTP
98  */
99 
100 #define ptp_va2o(va, lvl)	(pl_i(va, (lvl)+1) * PAGE_SIZE)
101 
102 /* size of a PDP: usually one page, except for PAE */
103 #ifdef PAE
104 #define PDP_SIZE 4
105 #else
106 #define PDP_SIZE 1
107 #endif
108 
109 
110 #if defined(_KERNEL)
111 /*
112  * pmap data structures: see pmap.c for details of locking.
113  */
114 
115 /*
116  * we maintain a list of all non-kernel pmaps
117  */
118 
119 LIST_HEAD(pmap_head, pmap); /* struct pmap_head: head of a pmap list */
120 
121 /*
122  * the pmap structure
123  *
124  * note that the pm_obj contains the lock pointer, the reference count,
125  * page list, and number of PTPs within the pmap.
126  *
127  * pm_lock is the same as the lock for vm object 0.  Changes to
128  * the other objects may only be made if that lock has been taken
129  * (the other object locks are only used when uvm_pagealloc is called)
130  *
131  * XXX If we ever support processor numbers higher than 31, we'll have
132  * XXX to rethink the CPU mask.
133  */
134 
135 struct pmap {
136 	struct uvm_object pm_obj[PTP_LEVELS-1]; /* objects for lvl >= 1) */
137 #define	pm_lock	pm_obj[0].vmobjlock
138 	kmutex_t pm_obj_lock[PTP_LEVELS-1];	/* locks for pm_objs */
139 	LIST_ENTRY(pmap) pm_list;	/* list (lck by pm_list lock) */
140 	pd_entry_t *pm_pdir;		/* VA of PD (lck by object lock) */
141 	paddr_t pm_pdirpa[PDP_SIZE];	/* PA of PDs (read-only after create) */
142 	struct vm_page *pm_ptphint[PTP_LEVELS-1];
143 					/* pointer to a PTP in our pmap */
144 	struct pmap_statistics pm_stats;  /* pmap stats (lck by object lock) */
145 
146 #if !defined(__x86_64__)
147 	vaddr_t pm_hiexec;		/* highest executable mapping */
148 #endif /* !defined(__x86_64__) */
149 	int pm_flags;			/* see below */
150 
151 	union descriptor *pm_ldt;	/* user-set LDT */
152 	size_t pm_ldt_len;		/* size of LDT in bytes */
153 	int pm_ldt_sel;			/* LDT selector */
154 	uint32_t pm_cpus;		/* mask of CPUs using pmap */
155 	uint32_t pm_kernel_cpus;	/* mask of CPUs using kernel part
156 					 of pmap */
157 	uint64_t pm_ncsw;		/* for assertions */
158 	struct vm_page *pm_gc_ptp;	/* pages from pmap g/c */
159 };
160 
161 /* macro to access pm_pdirpa slots */
162 #ifdef PAE
163 #define pmap_pdirpa(pmap, index) \
164 	((pmap)->pm_pdirpa[l2tol3(index)] + l2tol2(index) * sizeof(pd_entry_t))
165 #else
166 #define pmap_pdirpa(pmap, index) \
167 	((pmap)->pm_pdirpa[0] + (index) * sizeof(pd_entry_t))
168 #endif
169 
170 /*
171  * MD flags that we use for pmap_enter and pmap_kenter_pa:
172  */
173 
174 /*
175  * global kernel variables
176  */
177 
178 /*
179  * PDPpaddr is the physical address of the kernel's PDP.
180  * - i386 non-PAE and amd64: PDPpaddr corresponds directly to the %cr3
181  * value associated to the kernel process, proc0.
182  * - i386 PAE: it still represents the PA of the kernel's PDP (L2). Due to
183  * the L3 PD, it cannot be considered as the equivalent of a %cr3 any more.
184  * - Xen: it corresponds to the PFN of the kernel's PDP.
185  */
186 extern u_long PDPpaddr;
187 
188 extern int pmap_pg_g;			/* do we support PG_G? */
189 extern long nkptp[PTP_LEVELS];
190 
191 /*
192  * macros
193  */
194 
195 #define	pmap_resident_count(pmap)	((pmap)->pm_stats.resident_count)
196 #define	pmap_wired_count(pmap)		((pmap)->pm_stats.wired_count)
197 
198 #define pmap_clear_modify(pg)		pmap_clear_attrs(pg, PG_M)
199 #define pmap_clear_reference(pg)	pmap_clear_attrs(pg, PG_U)
200 #define pmap_copy(DP,SP,D,L,S)
201 #define pmap_is_modified(pg)		pmap_test_attrs(pg, PG_M)
202 #define pmap_is_referenced(pg)		pmap_test_attrs(pg, PG_U)
203 #define pmap_move(DP,SP,D,L,S)
204 #define pmap_phys_address(ppn)		(x86_ptob(ppn) & ~X86_MMAP_FLAG_MASK)
205 #define pmap_mmap_flags(ppn)		x86_mmap_flags(ppn)
206 #define pmap_valid_entry(E) 		((E) & PG_V) /* is PDE or PTE valid? */
207 
208 #if defined(__x86_64__) || defined(PAE)
209 #define X86_MMAP_FLAG_SHIFT	(64 - PGSHIFT)
210 #else
211 #define X86_MMAP_FLAG_SHIFT	(32 - PGSHIFT)
212 #endif
213 
214 #define X86_MMAP_FLAG_MASK	0xf
215 #define X86_MMAP_FLAG_PREFETCH	0x1
216 
217 /*
218  * prototypes
219  */
220 
221 void		pmap_activate(struct lwp *);
222 void		pmap_bootstrap(vaddr_t);
223 bool		pmap_clear_attrs(struct vm_page *, unsigned);
224 void		pmap_deactivate(struct lwp *);
225 void		pmap_page_remove (struct vm_page *);
226 void		pmap_remove(struct pmap *, vaddr_t, vaddr_t);
227 bool		pmap_test_attrs(struct vm_page *, unsigned);
228 void		pmap_write_protect(struct pmap *, vaddr_t, vaddr_t, vm_prot_t);
229 void		pmap_load(void);
230 paddr_t		pmap_init_tmp_pgtbl(paddr_t);
231 void		pmap_remove_all(struct pmap *);
232 void		pmap_ldt_sync(struct pmap *);
233 
234 void		pmap_emap_enter(vaddr_t, paddr_t, vm_prot_t);
235 void		pmap_emap_remove(vaddr_t, vsize_t);
236 void		pmap_emap_sync(bool);
237 
238 void		pmap_map_ptes(struct pmap *, struct pmap **, pd_entry_t **,
239 		    pd_entry_t * const **);
240 void		pmap_unmap_ptes(struct pmap *, struct pmap *);
241 
242 int		pmap_pdes_invalid(vaddr_t, pd_entry_t * const *, pd_entry_t *);
243 
244 u_int		x86_mmap_flags(paddr_t);
245 
246 bool		pmap_is_curpmap(struct pmap *);
247 
248 vaddr_t reserve_dumppages(vaddr_t); /* XXX: not a pmap fn */
249 
250 typedef enum tlbwhy {
251 	TLBSHOOT_APTE,
252 	TLBSHOOT_KENTER,
253 	TLBSHOOT_KREMOVE,
254 	TLBSHOOT_FREE_PTP1,
255 	TLBSHOOT_FREE_PTP2,
256 	TLBSHOOT_REMOVE_PTE,
257 	TLBSHOOT_REMOVE_PTES,
258 	TLBSHOOT_SYNC_PV1,
259 	TLBSHOOT_SYNC_PV2,
260 	TLBSHOOT_WRITE_PROTECT,
261 	TLBSHOOT_ENTER,
262 	TLBSHOOT_UPDATE,
263 	TLBSHOOT_BUS_DMA,
264 	TLBSHOOT_BUS_SPACE,
265 	TLBSHOOT__MAX,
266 } tlbwhy_t;
267 
268 void		pmap_tlb_init(void);
269 void		pmap_tlb_shootdown(pmap_t, vaddr_t, pt_entry_t, tlbwhy_t);
270 void		pmap_tlb_shootnow(void);
271 void		pmap_tlb_intr(void);
272 
273 #define	__HAVE_PMAP_EMAP
274 
275 #define PMAP_GROWKERNEL		/* turn on pmap_growkernel interface */
276 #define PMAP_FORK		/* turn on pmap_fork interface */
277 
278 /*
279  * Do idle page zero'ing uncached to avoid polluting the cache.
280  */
281 bool	pmap_pageidlezero(paddr_t);
282 #define	PMAP_PAGEIDLEZERO(pa)	pmap_pageidlezero((pa))
283 
284 /*
285  * inline functions
286  */
287 
288 __inline static bool __unused
289 pmap_pdes_valid(vaddr_t va, pd_entry_t * const *pdes, pd_entry_t *lastpde)
290 {
291 	return pmap_pdes_invalid(va, pdes, lastpde) == 0;
292 }
293 
294 /*
295  * pmap_update_pg: flush one page from the TLB (or flush the whole thing
296  *	if hardware doesn't support one-page flushing)
297  */
298 
299 __inline static void __unused
300 pmap_update_pg(vaddr_t va)
301 {
302 	invlpg(va);
303 }
304 
305 /*
306  * pmap_update_2pg: flush two pages from the TLB
307  */
308 
309 __inline static void __unused
310 pmap_update_2pg(vaddr_t va, vaddr_t vb)
311 {
312 	invlpg(va);
313 	invlpg(vb);
314 }
315 
316 /*
317  * pmap_page_protect: change the protection of all recorded mappings
318  *	of a managed page
319  *
320  * => this function is a frontend for pmap_page_remove/pmap_clear_attrs
321  * => we only have to worry about making the page more protected.
322  *	unprotecting a page is done on-demand at fault time.
323  */
324 
325 __inline static void __unused
326 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
327 {
328 	if ((prot & VM_PROT_WRITE) == 0) {
329 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
330 			(void) pmap_clear_attrs(pg, PG_RW);
331 		} else {
332 			pmap_page_remove(pg);
333 		}
334 	}
335 }
336 
337 /*
338  * pmap_protect: change the protection of pages in a pmap
339  *
340  * => this function is a frontend for pmap_remove/pmap_write_protect
341  * => we only have to worry about making the page more protected.
342  *	unprotecting a page is done on-demand at fault time.
343  */
344 
345 __inline static void __unused
346 pmap_protect(struct pmap *pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
347 {
348 	if ((prot & VM_PROT_WRITE) == 0) {
349 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
350 			pmap_write_protect(pmap, sva, eva, prot);
351 		} else {
352 			pmap_remove(pmap, sva, eva);
353 		}
354 	}
355 }
356 
357 /*
358  * various address inlines
359  *
360  *  vtopte: return a pointer to the PTE mapping a VA, works only for
361  *  user and PT addresses
362  *
363  *  kvtopte: return a pointer to the PTE mapping a kernel VA
364  */
365 
366 #include <lib/libkern/libkern.h>
367 
368 static __inline pt_entry_t * __unused
369 vtopte(vaddr_t va)
370 {
371 
372 	KASSERT(va < VM_MIN_KERNEL_ADDRESS);
373 
374 	return (PTE_BASE + pl1_i(va));
375 }
376 
377 static __inline pt_entry_t * __unused
378 kvtopte(vaddr_t va)
379 {
380 	pd_entry_t *pde;
381 
382 	KASSERT(va >= VM_MIN_KERNEL_ADDRESS);
383 
384 	pde = L2_BASE + pl2_i(va);
385 	if (*pde & PG_PS)
386 		return ((pt_entry_t *)pde);
387 
388 	return (PTE_BASE + pl1_i(va));
389 }
390 
391 paddr_t vtophys(vaddr_t);
392 vaddr_t	pmap_map(vaddr_t, paddr_t, paddr_t, vm_prot_t);
393 void	pmap_cpu_init_late(struct cpu_info *);
394 bool	sse2_idlezero_page(void *);
395 
396 
397 #ifdef XEN
398 
399 #include <sys/bitops.h>
400 
401 #define XPTE_MASK	L1_FRAME
402 /* Selects the index of a PTE in (A)PTE_BASE */
403 #define XPTE_SHIFT	(L1_SHIFT - ilog2(sizeof(pt_entry_t)))
404 
405 /* PTE access inline fuctions */
406 
407 /*
408  * Get the machine address of the pointed pte
409  * We use hardware MMU to get value so works only for levels 1-3
410  */
411 
412 static __inline paddr_t
413 xpmap_ptetomach(pt_entry_t *pte)
414 {
415 	pt_entry_t *up_pte;
416 	vaddr_t va = (vaddr_t) pte;
417 
418 	va = ((va & XPTE_MASK) >> XPTE_SHIFT) | (vaddr_t) PTE_BASE;
419 	up_pte = (pt_entry_t *) va;
420 
421 	return (paddr_t) (((*up_pte) & PG_FRAME) + (((vaddr_t) pte) & (~PG_FRAME & ~VA_SIGN_MASK)));
422 }
423 
424 /*
425  * xpmap_update()
426  * Update an active pt entry with Xen
427  * Equivalent to *pte = npte
428  */
429 
430 static __inline void
431 xpmap_update (pt_entry_t *pte, pt_entry_t npte)
432 {
433         int s = splvm();
434 
435         xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
436         xpq_flush_queue();
437         splx(s);
438 }
439 
440 
441 /* Xen helpers to change bits of a pte */
442 #define XPMAP_UPDATE_DIRECT	1	/* Update direct map entry flags too */
443 
444 paddr_t	vtomach(vaddr_t);
445 #define vtomfn(va) (vtomach(va) >> PAGE_SHIFT)
446 
447 void	pmap_apte_flush(struct pmap *);
448 void	pmap_unmap_apdp(void);
449 
450 #endif	/* XEN */
451 
452 /* pmap functions with machine addresses */
453 void	pmap_kenter_ma(vaddr_t, paddr_t, vm_prot_t, u_int);
454 int	pmap_enter_ma(struct pmap *, vaddr_t, paddr_t, paddr_t,
455 	    vm_prot_t, u_int, int);
456 bool	pmap_extract_ma(pmap_t, vaddr_t, paddr_t *);
457 
458 /*
459  * Hooks for the pool allocator.
460  */
461 #define	POOL_VTOPHYS(va)	vtophys((vaddr_t) (va))
462 
463 #endif /* _KERNEL */
464 
465 #endif /* _X86_PMAP_H_ */
466