xref: /netbsd-src/sys/arch/i386/include/pmap.h (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: pmap.h,v 1.73 2003/04/09 18:22:15 thorpej Exp $	*/
2 
3 /*
4  *
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgment:
18  *      This product includes software developed by Charles D. Cranor and
19  *      Washington University.
20  * 4. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * pmap.h: see pmap.c for the history of this pmap module.
37  */
38 
39 #ifndef	_I386_PMAP_H_
40 #define	_I386_PMAP_H_
41 
42 #if defined(_KERNEL_OPT)
43 #include "opt_user_ldt.h"
44 #include "opt_largepages.h"
45 #endif
46 
47 #include <machine/cpufunc.h>
48 #include <machine/pte.h>
49 #include <machine/segments.h>
50 #include <uvm/uvm_object.h>
51 
52 /*
53  * see pte.h for a description of i386 MMU terminology and hardware
54  * interface.
55  *
56  * a pmap describes a processes' 4GB virtual address space.  this
57  * virtual address space can be broken up into 1024 4MB regions which
58  * are described by PDEs in the PDP.  the PDEs are defined as follows:
59  *
60  * (ranges are inclusive -> exclusive, just like vm_map_entry start/end)
61  * (the following assumes that KERNBASE is 0xc0000000)
62  *
63  * PDE#s	VA range		usage
64  * 0->766	0x0 -> 0xbfc00000	user address space
65  * 767		0xbfc00000->		recursive mapping of PDP (used for
66  *			0xc0000000	linear mapping of PTPs)
67  * 768->1023	0xc0000000->		kernel address space (constant
68  *			0xffc00000	across all pmap's/processes)
69  * 1023		0xffc00000->		"alternate" recursive PDP mapping
70  *			<end>		(for other pmaps)
71  *
72  *
73  * note: a recursive PDP mapping provides a way to map all the PTEs for
74  * a 4GB address space into a linear chunk of virtual memory.  in other
75  * words, the PTE for page 0 is the first int mapped into the 4MB recursive
76  * area.  the PTE for page 1 is the second int.  the very last int in the
77  * 4MB range is the PTE that maps VA 0xffffe000 (the last page in a 4GB
78  * address).
79  *
80  * all pmap's PD's must have the same values in slots 768->1023 so that
81  * the kernel is always mapped in every process.  these values are loaded
82  * into the PD at pmap creation time.
83  *
84  * at any one time only one pmap can be active on a processor.  this is
85  * the pmap whose PDP is pointed to by processor register %cr3.  this pmap
86  * will have all its PTEs mapped into memory at the recursive mapping
87  * point (slot #767 as show above).  when the pmap code wants to find the
88  * PTE for a virtual address, all it has to do is the following:
89  *
90  * address of PTE = (767 * 4MB) + (VA / PAGE_SIZE) * sizeof(pt_entry_t)
91  *                = 0xbfc00000 + (VA / 4096) * 4
92  *
93  * what happens if the pmap layer is asked to perform an operation
94  * on a pmap that is not the one which is currently active?  in that
95  * case we take the PA of the PDP of non-active pmap and put it in
96  * slot 1023 of the active pmap.  this causes the non-active pmap's
97  * PTEs to get mapped in the final 4MB of the 4GB address space
98  * (e.g. starting at 0xffc00000).
99  *
100  * the following figure shows the effects of the recursive PDP mapping:
101  *
102  *   PDP (%cr3)
103  *   +----+
104  *   |   0| -> PTP#0 that maps VA 0x0 -> 0x400000
105  *   |    |
106  *   |    |
107  *   | 767| -> points back to PDP (%cr3) mapping VA 0xbfc00000 -> 0xc0000000
108  *   | 768| -> first kernel PTP (maps 0xc0000000 -> 0xf0400000)
109  *   |    |
110  *   |1023| -> points to alternate pmap's PDP (maps 0xffc00000 -> end)
111  *   +----+
112  *
113  * note that the PDE#767 VA (0xbfc00000) is defined as "PTE_BASE"
114  * note that the PDE#1023 VA (0xffc00000) is defined as "APTE_BASE"
115  *
116  * starting at VA 0xbfc00000 the current active PDP (%cr3) acts as a
117  * PTP:
118  *
119  * PTP#767 == PDP(%cr3) => maps VA 0xbfc00000 -> 0xc0000000
120  *   +----+
121  *   |   0| -> maps the contents of PTP#0 at VA 0xbfc00000->0xbfc01000
122  *   |    |
123  *   |    |
124  *   | 767| -> maps contents of PTP#767 (the PDP) at VA 0xbffbf000
125  *   | 768| -> maps contents of first kernel PTP
126  *   |    |
127  *   |1023|
128  *   +----+
129  *
130  * note that mapping of the PDP at PTP#767's VA (0xbffbf000) is
131  * defined as "PDP_BASE".... within that mapping there are two
132  * defines:
133  *   "PDP_PDE" (0xbfeffbfc) is the VA of the PDE in the PDP
134  *      which points back to itself.
135  *   "APDP_PDE" (0xbfeffffc) is the VA of the PDE in the PDP which
136  *      establishes the recursive mapping of the alternate pmap.
137  *      to set the alternate PDP, one just has to put the correct
138  *	PA info in *APDP_PDE.
139  *
140  * note that in the APTE_BASE space, the APDP appears at VA
141  * "APDP_BASE" (0xfffff000).
142  */
143 /* XXX MP should we allocate one APDP_PDE per processor?? */
144 
145 /*
146  * the following defines identify the slots used as described above.
147  */
148 
149 #define PDSLOT_PTE	((KERNBASE/NBPD)-1) /* 767: for recursive PDP map */
150 #define PDSLOT_KERN	(KERNBASE/NBPD)	    /* 768: 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 * PAGE_SIZE)))
164 #define APDP_BASE ((pd_entry_t *)(((char *)APTE_BASE) + (PDSLOT_APTE * PAGE_SIZE)))
165 #define PDP_PDE		(PDP_BASE + PDSLOT_PTE)
166 #define APDP_PDE	(PDP_BASE + PDSLOT_APTE)
167 
168 /*
169  * the follow define determines how many PTPs should be set up for the
170  * kernel by locore.s at boot time.  this should be large enough to
171  * get the VM system running.  once the VM system is running, the
172  * pmap module can add more PTPs to the kernel area on demand.
173  */
174 
175 #ifndef NKPTP
176 #define NKPTP		4	/* 16MB to start */
177 #endif
178 #define NKPTP_MIN	4	/* smallest value we allow */
179 #define NKPTP_MAX	(1024 - (KERNBASE/NBPD) - 1)
180 				/* largest value (-1 for APTP space) */
181 
182 /*
183  * pdei/ptei: generate index into PDP/PTP from a VA
184  */
185 #define	pdei(VA)	(((VA) & PD_MASK) >> PDSHIFT)
186 #define	ptei(VA)	(((VA) & PT_MASK) >> PGSHIFT)
187 
188 /*
189  * PTP macros:
190  *   a PTP's index is the PD index of the PDE that points to it
191  *   a PTP's offset is the byte-offset in the PTE space that this PTP is at
192  *   a PTP's VA is the first VA mapped by that PTP
193  *
194  * note that PAGE_SIZE == number of bytes in a PTP (4096 bytes == 1024 entries)
195  *           NBPD == number of bytes a PTP can map (4MB)
196  */
197 
198 #define ptp_i2o(I)	((I) * PAGE_SIZE)	/* index => offset */
199 #define ptp_o2i(O)	((O) / PAGE_SIZE)	/* offset => index */
200 #define ptp_i2v(I)	((I) * NBPD)	/* index => VA */
201 #define ptp_v2i(V)	((V) / NBPD)	/* VA => index (same as pdei) */
202 
203 /*
204  * PG_AVAIL usage: we make use of the ignored bits of the PTE
205  */
206 
207 #define PG_W		PG_AVAIL1	/* "wired" mapping */
208 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
209 /* PG_AVAIL3 not used */
210 
211 /*
212  * Number of PTE's per cache line.  4 byte pte, 32-byte cache line
213  * Used to avoid false sharing of cache lines.
214  */
215 #define NPTECL			8
216 
217 #ifdef _KERNEL
218 /*
219  * pmap data structures: see pmap.c for details of locking.
220  */
221 
222 struct pmap;
223 typedef struct pmap *pmap_t;
224 
225 /*
226  * we maintain a list of all non-kernel pmaps
227  */
228 
229 LIST_HEAD(pmap_head, pmap); /* struct pmap_head: head of a pmap list */
230 
231 /*
232  * the pmap structure
233  *
234  * note that the pm_obj contains the simple_lock, the reference count,
235  * page list, and number of PTPs within the pmap.
236  *
237  * XXX If we ever support processor numbers higher than 31, we'll have
238  * XXX to rethink the CPU mask.
239  */
240 
241 struct pmap {
242 	struct uvm_object pm_obj;	/* object (lck by object lock) */
243 #define	pm_lock	pm_obj.vmobjlock
244 	LIST_ENTRY(pmap) pm_list;	/* list (lck by pm_list lock) */
245 	pd_entry_t *pm_pdir;		/* VA of PD (lck by object lock) */
246 	u_int32_t pm_pdirpa;		/* PA of PD (read-only after create) */
247 	struct vm_page *pm_ptphint;	/* pointer to a PTP in our pmap */
248 	struct pmap_statistics pm_stats;  /* pmap stats (lck by object lock) */
249 
250 	int pm_flags;			/* see below */
251 
252 	union descriptor *pm_ldt;	/* user-set LDT */
253 	int pm_ldt_len;			/* number of LDT entries */
254 	int pm_ldt_sel;			/* LDT selector */
255 	u_int32_t pm_cpus;		/* mask of CPUs using pmap */
256 };
257 
258 /* pm_flags */
259 #define	PMF_USER_LDT	0x01	/* pmap has user-set LDT */
260 
261 /*
262  * for each managed physical page we maintain a list of <PMAP,VA>'s
263  * which it is mapped at.  the list is headed by a pv_head structure.
264  * there is one pv_head per managed phys page (allocated at boot time).
265  * the pv_head structure points to a list of pv_entry structures (each
266  * describes one mapping).
267  */
268 
269 struct pv_entry;
270 
271 struct pv_head {
272 	struct simplelock pvh_lock;	/* locks every pv on this list */
273 	struct pv_entry *pvh_list;	/* head of list (locked by pvh_lock) */
274 };
275 
276 struct pv_entry {			/* locked by its list's pvh_lock */
277 	struct pv_entry *pv_next;	/* next entry */
278 	struct pmap *pv_pmap;		/* the pmap */
279 	vaddr_t pv_va;			/* the virtual address */
280 	struct vm_page *pv_ptp;		/* the vm_page of the PTP */
281 };
282 
283 /*
284  * pv_entrys are dynamically allocated in chunks from a single page.
285  * we keep track of how many pv_entrys are in use for each page and
286  * we can free pv_entry pages if needed.  there is one lock for the
287  * entire allocation system.
288  */
289 
290 struct pv_page_info {
291 	TAILQ_ENTRY(pv_page) pvpi_list;
292 	struct pv_entry *pvpi_pvfree;
293 	int pvpi_nfree;
294 };
295 
296 /*
297  * number of pv_entry's in a pv_page
298  * (note: won't work on systems where NPBG isn't a constant)
299  */
300 
301 #define PVE_PER_PVPAGE ((PAGE_SIZE - sizeof(struct pv_page_info)) / \
302 			sizeof(struct pv_entry))
303 
304 /*
305  * a pv_page: where pv_entrys are allocated from
306  */
307 
308 struct pv_page {
309 	struct pv_page_info pvinfo;
310 	struct pv_entry pvents[PVE_PER_PVPAGE];
311 };
312 
313 /*
314  * global kernel variables
315  */
316 
317 /* PTDpaddr: is the physical address of the kernel's PDP */
318 extern u_long PTDpaddr;
319 
320 extern struct pmap kernel_pmap_store;	/* kernel pmap */
321 extern int nkpde;			/* current # of PDEs for kernel */
322 extern int pmap_pg_g;			/* do we support PG_G? */
323 
324 /*
325  * macros
326  */
327 
328 #define	pmap_kernel()			(&kernel_pmap_store)
329 #define	pmap_resident_count(pmap)	((pmap)->pm_stats.resident_count)
330 #define	pmap_wired_count(pmap)		((pmap)->pm_stats.wired_count)
331 #define	pmap_update(pmap)		/* nothing (yet) */
332 
333 #define pmap_clear_modify(pg)		pmap_clear_attrs(pg, PG_M)
334 #define pmap_clear_reference(pg)	pmap_clear_attrs(pg, PG_U)
335 #define pmap_copy(DP,SP,D,L,S)
336 #define pmap_is_modified(pg)		pmap_test_attrs(pg, PG_M)
337 #define pmap_is_referenced(pg)		pmap_test_attrs(pg, PG_U)
338 #define pmap_move(DP,SP,D,L,S)
339 #define pmap_phys_address(ppn)		x86_ptob(ppn)
340 #define pmap_valid_entry(E) 		((E) & PG_V) /* is PDE or PTE valid? */
341 
342 
343 /*
344  * prototypes
345  */
346 
347 void		pmap_activate __P((struct lwp *));
348 void		pmap_bootstrap __P((vaddr_t));
349 boolean_t	pmap_clear_attrs __P((struct vm_page *, int));
350 void		pmap_deactivate __P((struct lwp *));
351 void		pmap_page_remove  __P((struct vm_page *));
352 void		pmap_remove __P((struct pmap *, vaddr_t, vaddr_t));
353 boolean_t	pmap_test_attrs __P((struct vm_page *, int));
354 void		pmap_write_protect __P((struct pmap *, vaddr_t,
355 				vaddr_t, vm_prot_t));
356 
357 vaddr_t reserve_dumppages __P((vaddr_t)); /* XXX: not a pmap fn */
358 
359 void	pmap_tlb_shootdown __P((pmap_t, vaddr_t, pt_entry_t, int32_t *));
360 void	pmap_tlb_shootnow __P((int32_t));
361 void	pmap_do_tlb_shootdown __P((struct cpu_info *));
362 
363 #define PMAP_GROWKERNEL		/* turn on pmap_growkernel interface */
364 
365 /*
366  * Do idle page zero'ing uncached to avoid polluting the cache.
367  */
368 boolean_t			pmap_pageidlezero __P((paddr_t));
369 #define	PMAP_PAGEIDLEZERO(pa)	pmap_pageidlezero((pa))
370 
371 /*
372  * inline functions
373  */
374 
375 /*ARGSUSED*/
376 static __inline void
377 pmap_remove_all(struct pmap *pmap)
378 {
379 	/* Nothing. */
380 }
381 
382 /*
383  * pmap_update_pg: flush one page from the TLB (or flush the whole thing
384  *	if hardware doesn't support one-page flushing)
385  */
386 
387 __inline static void __attribute__((__unused__))
388 pmap_update_pg(vaddr_t va)
389 {
390 #if defined(I386_CPU)
391 	if (cpu_class == CPUCLASS_386)
392 		tlbflush();
393 	else
394 #endif
395 		invlpg((u_int) va);
396 }
397 
398 /*
399  * pmap_update_2pg: flush two pages from the TLB
400  */
401 
402 __inline static void __attribute__((__unused__))
403 pmap_update_2pg(vaddr_t va, vaddr_t vb)
404 {
405 #if defined(I386_CPU)
406 	if (cpu_class == CPUCLASS_386)
407 		tlbflush();
408 	else
409 #endif
410 	{
411 		invlpg((u_int) va);
412 		invlpg((u_int) vb);
413 	}
414 }
415 
416 /*
417  * pmap_page_protect: change the protection of all recorded mappings
418  *	of a managed page
419  *
420  * => this function is a frontend for pmap_page_remove/pmap_clear_attrs
421  * => we only have to worry about making the page more protected.
422  *	unprotecting a page is done on-demand at fault time.
423  */
424 
425 __inline static void __attribute__((__unused__))
426 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
427 {
428 	if ((prot & VM_PROT_WRITE) == 0) {
429 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
430 			(void) pmap_clear_attrs(pg, PG_RW);
431 		} else {
432 			pmap_page_remove(pg);
433 		}
434 	}
435 }
436 
437 /*
438  * pmap_protect: change the protection of pages in a pmap
439  *
440  * => this function is a frontend for pmap_remove/pmap_write_protect
441  * => we only have to worry about making the page more protected.
442  *	unprotecting a page is done on-demand at fault time.
443  */
444 
445 __inline static void __attribute__((__unused__))
446 pmap_protect(struct pmap *pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
447 {
448 	if ((prot & VM_PROT_WRITE) == 0) {
449 		if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
450 			pmap_write_protect(pmap, sva, eva, prot);
451 		} else {
452 			pmap_remove(pmap, sva, eva);
453 		}
454 	}
455 }
456 
457 /*
458  * various address inlines
459  *
460  *  vtopte: return a pointer to the PTE mapping a VA, works only for
461  *  user and PT addresses
462  *
463  *  kvtopte: return a pointer to the PTE mapping a kernel VA
464  */
465 
466 #include <lib/libkern/libkern.h>
467 
468 static __inline pt_entry_t * __attribute__((__unused__))
469 vtopte(vaddr_t va)
470 {
471 
472 	KASSERT(va < (PDSLOT_KERN << PDSHIFT));
473 
474 	return (PTE_BASE + x86_btop(va));
475 }
476 
477 static __inline pt_entry_t * __attribute__((__unused__))
478 kvtopte(vaddr_t va)
479 {
480 
481 	KASSERT(va >= (PDSLOT_KERN << PDSHIFT));
482 
483 #ifdef LARGEPAGES
484 	{
485 		pd_entry_t *pde;
486 
487 		pde = PDP_BASE + pdei(va);
488 		if (*pde & PG_PS)
489 			return ((pt_entry_t *)pde);
490 	}
491 #endif
492 
493 	return (PTE_BASE + x86_btop(va));
494 }
495 
496 #define pmap_cpu_has_pg_n()		(cpu_class != CPUCLASS_386)
497 #define pmap_cpu_has_invlpg()		(cpu_class != CPUCLASS_386)
498 
499 paddr_t vtophys __P((vaddr_t));
500 vaddr_t	pmap_map __P((vaddr_t, paddr_t, paddr_t, vm_prot_t));
501 
502 #if defined(USER_LDT)
503 void	pmap_ldt_cleanup __P((struct lwp *));
504 #define	PMAP_FORK
505 #endif /* USER_LDT */
506 
507 /*
508  * Hooks for the pool allocator.
509  */
510 #define	POOL_VTOPHYS(va)	vtophys((vaddr_t) (va))
511 
512 #endif /* _KERNEL */
513 #endif	/* _I386_PMAP_H_ */
514