1 /* $OpenBSD: pmap.h,v 1.33 2003/05/26 16:25:32 art Exp $ */
2 /* $NetBSD: pmap.h,v 1.44 2000/04/24 17:18:18 thorpej Exp $ */
3
4 /*
5 *
6 * Copyright (c) 1997 Charles D. Cranor and Washington University.
7 * All rights reserved.
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 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgment:
19 * This product includes software developed by Charles D. Cranor and
20 * Washington University.
21 * 4. The name of the author may not be used to endorse or promote products
22 * derived from this software without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
25 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
29 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
30 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
31 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
32 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
33 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 */
35
36 /*
37 * pmap.h: see pmap.c for the history of this pmap module.
38 */
39
40 #ifndef _I386_PMAP_H_
41 #define _I386_PMAP_H_
42
43 #include <machine/cpufunc.h>
44 #include <machine/pte.h>
45 #include <machine/segments.h>
46 #include <uvm/uvm_pglist.h>
47 #include <uvm/uvm_object.h>
48
49 /*
50 * See pte.h for a description of i386 MMU terminology and hardware
51 * interface.
52 *
53 * A pmap describes a process' 4GB virtual address space. This
54 * virtual address space can be broken up into 1024 4MB regions which
55 * are described by PDEs in the PDP. The PDEs are defined as follows:
56 *
57 * Ranges are inclusive -> exclusive, just like vm_map_entry start/end.
58 * The following assumes that KERNBASE is 0xd0000000.
59 *
60 * PDE#s VA range Usage
61 * 0->831 0x0 -> 0xcfc00000 user address space, note that the
62 * max user address is 0xcfbfe000
63 * the final two pages in the last 4MB
64 * used to be reserved for the UAREA
65 * but now are no longer used.
66 * 831 0xcfc00000-> recursive mapping of PDP (used for
67 * 0xd0000000 linear mapping of PTPs).
68 * 832->1023 0xd0000000-> kernel address space (constant
69 * 0xffc00000 across all pmaps/processes).
70 * 1023 0xffc00000-> "alternate" recursive PDP mapping
71 * <end> (for other pmaps).
72 *
73 *
74 * Note: A recursive PDP mapping provides a way to map all the PTEs for
75 * a 4GB address space into a linear chunk of virtual memory. In other
76 * words, the PTE for page 0 is the first int mapped into the 4MB recursive
77 * area. The PTE for page 1 is the second int. The very last int in the
78 * 4MB range is the PTE that maps VA 0xffffe000 (the last page in a 4GB
79 * address).
80 *
81 * All pmaps' PDs must have the same values in slots 832->1023 so that
82 * the kernel is always mapped in every process. These values are loaded
83 * into the PD at pmap creation time.
84 *
85 * At any one time only one pmap can be active on a processor. This is
86 * the pmap whose PDP is pointed to by processor register %cr3. This pmap
87 * will have all its PTEs mapped into memory at the recursive mapping
88 * point (slot #831 as show above). When the pmap code wants to find the
89 * PTE for a virtual address, all it has to do is the following:
90 *
91 * Address of PTE = (831 * 4MB) + (VA / NBPG) * sizeof(pt_entry_t)
92 * = 0xcfc00000 + (VA / 4096) * 4
93 *
94 * What happens if the pmap layer is asked to perform an operation
95 * on a pmap that is not the one which is currently active? In that
96 * case we take the PA of the PDP of non-active pmap and put it in
97 * slot 1023 of the active pmap. This causes the non-active pmap's
98 * PTEs to get mapped in the final 4MB of the 4GB address space
99 * (e.g. starting at 0xffc00000).
100 *
101 * The following figure shows the effects of the recursive PDP mapping:
102 *
103 * PDP (%cr3)
104 * +----+
105 * | 0| -> PTP#0 that maps VA 0x0 -> 0x400000
106 * | |
107 * | |
108 * | 831| -> points back to PDP (%cr3) mapping VA 0xcfc00000 -> 0xd0000000
109 * | 832| -> first kernel PTP (maps 0xd0000000 -> 0xe0400000)
110 * | |
111 * |1023| -> points to alternate pmap's PDP (maps 0xffc00000 -> end)
112 * +----+
113 *
114 * Note that the PDE#831 VA (0xcfc00000) is defined as "PTE_BASE".
115 * Note that the PDE#1023 VA (0xffc00000) is defined as "APTE_BASE".
116 *
117 * Starting at VA 0xdfc00000 the current active PDP (%cr3) acts as a
118 * PTP:
119 *
120 * PTP#831 == PDP(%cr3) => maps VA 0xcfc00000 -> 0xd0000000
121 * +----+
122 * | 0| -> maps the contents of PTP#0 at VA 0xcfc00000->0xcfc01000
123 * | |
124 * | |
125 * | 831| -> maps the contents of PTP#831 (the PDP) at VA 0xcff3f000
126 * | 832| -> maps the contents of first kernel PTP
127 * | |
128 * |1023|
129 * +----+
130 *
131 * Note that mapping of the PDP at PTP#831's VA (0xcff3f000) is
132 * defined as "PDP_BASE".... within that mapping there are two
133 * defines:
134 * "PDP_PDE" (0xcff3fcfc) is the VA of the PDE in the PDP
135 * which points back to itself.
136 * "APDP_PDE" (0xcff3fffc) is the VA of the PDE in the PDP which
137 * establishes the recursive mapping of the alternate pmap.
138 * To set the alternate PDP, one just has to put the correct
139 * PA info in *APDP_PDE.
140 *
141 * Note that in the APTE_BASE space, the APDP appears at VA
142 * "APDP_BASE" (0xfffff000).
143 */
144
145 /*
146 * The following defines identify the slots used as described above.
147 */
148
149 #define PDSLOT_PTE ((KERNBASE/NBPD)-1) /* 831: for recursive PDP map */
150 #define PDSLOT_KERN (KERNBASE/NBPD) /* 832: start of kernel space */
151 #define PDSLOT_APTE ((unsigned)1023) /* 1023: alternative recursive slot */
152
153 /*
154 * The following defines give the virtual addresses of various MMU
155 * data structures:
156 * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
157 * PTD_BASE and APTD_BASE: the base VA of the recursive mapping of the PTD
158 * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
159 */
160
161 #define PTE_BASE ((pt_entry_t *) (PDSLOT_PTE * NBPD) )
162 #define APTE_BASE ((pt_entry_t *) (PDSLOT_APTE * NBPD) )
163 #define PDP_BASE ((pd_entry_t *)(((char *)PTE_BASE) + (PDSLOT_PTE * NBPG)))
164 #define APDP_BASE ((pd_entry_t *)(((char *)APTE_BASE) + (PDSLOT_APTE * NBPG)))
165 #define PDP_PDE (PDP_BASE + PDSLOT_PTE)
166 #define APDP_PDE (PDP_BASE + PDSLOT_APTE)
167
168 /*
169 * XXXCDC: tmp xlate from old names:
170 * PTDPTDI -> PDSLOT_PTE
171 * KPTDI -> PDSLOT_KERN
172 * APTDPTDI -> PDSLOT_APTE
173 */
174
175 /*
176 * The following define determines how many PTPs should be set up for the
177 * kernel by locore.s at boot time. This should be large enough to
178 * get the VM system running. Once the VM system is running, the
179 * pmap module can add more PTPs to the kernel area on demand.
180 */
181
182 #ifndef NKPTP
183 #define NKPTP 4 /* 16MB to start */
184 #endif
185 #define NKPTP_MIN 4 /* smallest value we allow */
186 #define NKPTP_MAX (1024 - (KERNBASE/NBPD) - 1)
187 /* largest value (-1 for APTP space) */
188
189 /*
190 * various address macros
191 *
192 * vtopte: return a pointer to the PTE mapping a VA
193 * kvtopte: same as above (takes a KVA, but doesn't matter with this pmap)
194 * ptetov: given a pointer to a PTE, return the VA that it maps
195 * vtophys: translate a VA to the PA mapped to it
196 *
197 * plus alternative versions of the above
198 */
199
200 #define vtopte(VA) (PTE_BASE + i386_btop(VA))
201 #define kvtopte(VA) vtopte(VA)
202 #define ptetov(PT) (i386_ptob(PT - PTE_BASE))
203 #define vtophys(VA) ((*vtopte(VA) & PG_FRAME) | \
204 ((unsigned)(VA) & ~PG_FRAME))
205 #define avtopte(VA) (APTE_BASE + i386_btop(VA))
206 #define ptetoav(PT) (i386_ptob(PT - APTE_BASE))
207 #define avtophys(VA) ((*avtopte(VA) & PG_FRAME) | \
208 ((unsigned)(VA) & ~PG_FRAME))
209
210 /*
211 * pdei/ptei: generate index into PDP/PTP from a VA
212 */
213 #define pdei(VA) (((VA) & PD_MASK) >> PDSHIFT)
214 #define ptei(VA) (((VA) & PT_MASK) >> PGSHIFT)
215
216 /*
217 * PTP macros:
218 * A PTP's index is the PD index of the PDE that points to it.
219 * A PTP's offset is the byte-offset in the PTE space that this PTP is at.
220 * A PTP's VA is the first VA mapped by that PTP.
221 *
222 * Note that NBPG == number of bytes in a PTP (4096 bytes == 1024 entries)
223 * NBPD == number of bytes a PTP can map (4MB)
224 */
225
226 #define ptp_i2o(I) ((I) * NBPG) /* index => offset */
227 #define ptp_o2i(O) ((O) / NBPG) /* offset => index */
228 #define ptp_i2v(I) ((I) * NBPD) /* index => VA */
229 #define ptp_v2i(V) ((V) / NBPD) /* VA => index (same as pdei) */
230
231 /*
232 * PG_AVAIL usage: we make use of the ignored bits of the PTE
233 */
234
235 #define PG_W PG_AVAIL1 /* "wired" mapping */
236 #define PG_PVLIST PG_AVAIL2 /* mapping has entry on pvlist */
237 #define PG_X PG_AVAIL3 /* executable mapping */
238
239 #ifdef _KERNEL
240 /*
241 * pmap data structures: see pmap.c for details of locking.
242 */
243
244 struct pmap;
245 typedef struct pmap *pmap_t;
246
247 /*
248 * We maintain a list of all non-kernel pmaps.
249 */
250
251 LIST_HEAD(pmap_head, pmap); /* struct pmap_head: head of a pmap list */
252
253 /*
254 * The pmap structure
255 *
256 * Note that the pm_obj contains the simple_lock, the reference count,
257 * page list, and number of PTPs within the pmap.
258 */
259
260 struct pmap {
261 struct uvm_object pm_obj; /* object (lck by object lock) */
262 #define pm_lock pm_obj.vmobjlock
263 LIST_ENTRY(pmap) pm_list; /* list (lck by pm_list lock) */
264 pd_entry_t *pm_pdir; /* VA of PD (lck by object lock) */
265 u_int32_t pm_pdirpa; /* PA of PD (read-only after create) */
266 struct vm_page *pm_ptphint; /* pointer to a PTP in our pmap */
267 struct pmap_statistics pm_stats; /* pmap stats (lck by object lock) */
268
269 vaddr_t pm_hiexec; /* highest executable mapping */
270 int pm_flags; /* see below */
271
272 union descriptor *pm_ldt; /* user-set LDT */
273 int pm_ldt_len; /* number of LDT entries */
274 int pm_ldt_sel; /* LDT selector */
275 };
276
277 /* pm_flags */
278 #define PMF_USER_LDT 0x01 /* pmap has user-set LDT */
279
280 /*
281 * For each managed physical page we maintain a list of <PMAP,VA>s
282 * which it is mapped at. The list is headed by a pv_head structure.
283 * there is one pv_head per managed phys page (allocated at boot time).
284 * The pv_head structure points to a list of pv_entry structures (each
285 * describes one mapping).
286 */
287
288 struct pv_entry;
289
290 struct pv_head {
291 struct simplelock pvh_lock; /* locks every pv on this list */
292 struct pv_entry *pvh_list; /* head of list (locked by pvh_lock) */
293 };
294
295 struct pv_entry { /* locked by its list's pvh_lock */
296 struct pv_entry *pv_next; /* next entry */
297 struct pmap *pv_pmap; /* the pmap */
298 vaddr_t pv_va; /* the virtual address */
299 struct vm_page *pv_ptp; /* the vm_page of the PTP */
300 };
301
302 /*
303 * pv_entrys are dynamically allocated in chunks from a single page.
304 * we keep track of how many pv_entrys are in use for each page and
305 * we can free pv_entry pages if needed. There is one lock for the
306 * entire allocation system.
307 */
308
309 struct pv_page_info {
310 TAILQ_ENTRY(pv_page) pvpi_list;
311 struct pv_entry *pvpi_pvfree;
312 int pvpi_nfree;
313 };
314
315 /*
316 * number of pv_entries in a pv_page
317 * (note: won't work on systems where NPBG isn't a constant)
318 */
319
320 #define PVE_PER_PVPAGE ((NBPG - sizeof(struct pv_page_info)) / \
321 sizeof(struct pv_entry))
322
323 /*
324 * a pv_page: where pv_entrys are allocated from
325 */
326
327 struct pv_page {
328 struct pv_page_info pvinfo;
329 struct pv_entry pvents[PVE_PER_PVPAGE];
330 };
331
332 /*
333 * pmap_remove_record: a record of VAs that have been unmapped, used to
334 * flush TLB. If we have more than PMAP_RR_MAX then we stop recording.
335 */
336
337 #define PMAP_RR_MAX 16 /* max of 16 pages (64K) */
338
339 struct pmap_remove_record {
340 int prr_npages;
341 vaddr_t prr_vas[PMAP_RR_MAX];
342 };
343
344 /*
345 * Global kernel variables
346 */
347
348 extern pd_entry_t PTD[];
349
350 /* PTDpaddr: is the physical address of the kernel's PDP */
351 extern u_long PTDpaddr;
352
353 extern struct pmap kernel_pmap_store; /* kernel pmap */
354 extern int nkpde; /* current # of PDEs for kernel */
355 extern int pmap_pg_g; /* do we support PG_G? */
356
357 /*
358 * Macros
359 */
360
361 #define pmap_kernel() (&kernel_pmap_store)
362 #define pmap_resident_count(pmap) ((pmap)->pm_stats.resident_count)
363 #define pmap_update(pm) /* nada */
364
365 #define pmap_clear_modify(pg) pmap_change_attrs(pg, 0, PG_M)
366 #define pmap_clear_reference(pg) pmap_change_attrs(pg, 0, PG_U)
367 #define pmap_copy(DP,SP,D,L,S)
368 #define pmap_is_modified(pg) pmap_test_attrs(pg, PG_M)
369 #define pmap_is_referenced(pg) pmap_test_attrs(pg, PG_U)
370 #define pmap_phys_address(ppn) i386_ptob(ppn)
371 #define pmap_valid_entry(E) ((E) & PG_V) /* is PDE or PTE valid? */
372
373 #define pmap_proc_iflush(p,va,len) /* nothing */
374
375 /*
376 * Prototypes
377 */
378
379 void pmap_bootstrap(vaddr_t);
380 boolean_t pmap_change_attrs(struct vm_page *, int, int);
381 static void pmap_page_protect(struct vm_page *, vm_prot_t);
382 void pmap_page_remove(struct vm_page *);
383 static void pmap_protect(struct pmap *, vaddr_t,
384 vaddr_t, vm_prot_t);
385 void pmap_remove(struct pmap *, vaddr_t, vaddr_t);
386 boolean_t pmap_test_attrs(struct vm_page *, int);
387 static void pmap_update_pg(vaddr_t);
388 static void pmap_update_2pg(vaddr_t,vaddr_t);
389 void pmap_write_protect(struct pmap *, vaddr_t,
390 vaddr_t, vm_prot_t);
391 int pmap_exec_fixup(struct vm_map *, struct trapframe *,
392 struct pcb *);
393
394 vaddr_t reserve_dumppages(vaddr_t); /* XXX: not a pmap fn */
395
396 #define PMAP_GROWKERNEL /* turn on pmap_growkernel interface */
397
398 /*
399 * Do idle page zero'ing uncached to avoid polluting the cache.
400 */
401 boolean_t pmap_zero_page_uncached(paddr_t);
402 #define PMAP_PAGEIDLEZERO(pg) pmap_zero_page_uncached(VM_PAGE_TO_PHYS(pg))
403
404 /*
405 * Inline functions
406 */
407
408 /*
409 * pmap_update_pg: flush one page from the TLB (or flush the whole thing
410 * if hardware doesn't support one-page flushing)
411 */
412
413 __inline static void
pmap_update_pg(va)414 pmap_update_pg(va)
415 vaddr_t va;
416 {
417 #if defined(I386_CPU)
418 if (cpu_class == CPUCLASS_386)
419 tlbflush();
420 else
421 #endif
422 invlpg((u_int) va);
423 }
424
425 /*
426 * pmap_update_2pg: flush two pages from the TLB
427 */
428
429 __inline static void
pmap_update_2pg(va,vb)430 pmap_update_2pg(va, vb)
431 vaddr_t va, vb;
432 {
433 #if defined(I386_CPU)
434 if (cpu_class == CPUCLASS_386)
435 tlbflush();
436 else
437 #endif
438 {
439 invlpg((u_int) va);
440 invlpg((u_int) vb);
441 }
442 }
443
444 /*
445 * pmap_page_protect: change the protection of all recorded mappings
446 * of a managed page
447 *
448 * => This function is a front end for pmap_page_remove/pmap_change_attrs
449 * => We only have to worry about making the page more protected.
450 * Unprotecting a page is done on-demand at fault time.
451 */
452
453 __inline static void
pmap_page_protect(pg,prot)454 pmap_page_protect(pg, prot)
455 struct vm_page *pg;
456 vm_prot_t prot;
457 {
458 if ((prot & VM_PROT_WRITE) == 0) {
459 if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
460 (void) pmap_change_attrs(pg, PG_RO, PG_RW);
461 } else {
462 pmap_page_remove(pg);
463 }
464 }
465 }
466
467 /*
468 * pmap_protect: change the protection of pages in a pmap
469 *
470 * => This function is a front end for pmap_remove/pmap_write_protect.
471 * => We only have to worry about making the page more protected.
472 * Unprotecting a page is done on-demand at fault time.
473 */
474
475 __inline static void
pmap_protect(pmap,sva,eva,prot)476 pmap_protect(pmap, sva, eva, prot)
477 struct pmap *pmap;
478 vaddr_t sva, eva;
479 vm_prot_t prot;
480 {
481 if ((prot & VM_PROT_WRITE) == 0) {
482 if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
483 pmap_write_protect(pmap, sva, eva, prot);
484 } else {
485 pmap_remove(pmap, sva, eva);
486 }
487 }
488 }
489
490 #if defined(USER_LDT)
491 void pmap_ldt_cleanup(struct proc *);
492 #define PMAP_FORK
493 #endif /* USER_LDT */
494
495 #endif /* _KERNEL */
496 #endif /* _I386_PMAP_H_ */
497