xref: /netbsd-src/sys/arch/i386/include/pmap.h (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: pmap.h,v 1.108 2010/11/14 13:33:21 uebayasi 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  * Copyright (c) 2001 Wasabi Systems, Inc.
37  * All rights reserved.
38  *
39  * Written by Frank van der Linden for Wasabi Systems, Inc.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. All advertising materials mentioning features or use of this software
50  *    must display the following acknowledgement:
51  *      This product includes software developed for the NetBSD Project by
52  *      Wasabi Systems, Inc.
53  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
54  *    or promote products derived from this software without specific prior
55  *    written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
59  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
60  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
61  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
62  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
63  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
64  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
65  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
66  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
67  * POSSIBILITY OF SUCH DAMAGE.
68  */
69 
70 #ifndef	_I386_PMAP_H_
71 #define	_I386_PMAP_H_
72 
73 #if defined(_KERNEL_OPT)
74 #include "opt_user_ldt.h"
75 #include "opt_xen.h"
76 #endif
77 
78 #include <sys/atomic.h>
79 
80 #include <i386/pte.h>
81 #include <machine/segments.h>
82 #if defined(_KERNEL)
83 #include <machine/cpufunc.h>
84 #endif
85 
86 #include <uvm/uvm_object.h>
87 #ifdef XEN
88 #include <xen/xenfunc.h>
89 #include <xen/xenpmap.h>
90 #endif /* XEN */
91 
92 /*
93  * see pte.h for a description of i386 MMU terminology and hardware
94  * interface.
95  *
96  * a pmap describes a processes' 4GB virtual address space.  when PAE
97  * is not in use, this virtual address space can be broken up into 1024 4MB
98  * regions which are described by PDEs in the PDP.  the PDEs are defined as
99  * follows:
100  *
101  * (ranges are inclusive -> exclusive, just like vm_map_entry start/end)
102  * (the following assumes that KERNBASE is 0xc0000000)
103  *
104  * PDE#s	VA range		usage
105  * 0->766	0x0 -> 0xbfc00000	user address space
106  * 767		0xbfc00000->		recursive mapping of PDP (used for
107  *			0xc0000000	linear mapping of PTPs)
108  * 768->1023	0xc0000000->		kernel address space (constant
109  *			0xffc00000	across all pmap's/processes)
110  * 1023		0xffc00000->		"alternate" recursive PDP mapping
111  *			<end>		(for other pmaps)
112  *
113  *
114  * note: a recursive PDP mapping provides a way to map all the PTEs for
115  * a 4GB address space into a linear chunk of virtual memory.  in other
116  * words, the PTE for page 0 is the first int mapped into the 4MB recursive
117  * area.  the PTE for page 1 is the second int.  the very last int in the
118  * 4MB range is the PTE that maps VA 0xfffff000 (the last page in a 4GB
119  * address).
120  *
121  * all pmap's PD's must have the same values in slots 768->1023 so that
122  * the kernel is always mapped in every process.  these values are loaded
123  * into the PD at pmap creation time.
124  *
125  * at any one time only one pmap can be active on a processor.  this is
126  * the pmap whose PDP is pointed to by processor register %cr3.  this pmap
127  * will have all its PTEs mapped into memory at the recursive mapping
128  * point (slot #767 as show above).  when the pmap code wants to find the
129  * PTE for a virtual address, all it has to do is the following:
130  *
131  * address of PTE = (767 * 4MB) + (VA / PAGE_SIZE) * sizeof(pt_entry_t)
132  *                = 0xbfc00000 + (VA / 4096) * 4
133  *
134  * what happens if the pmap layer is asked to perform an operation
135  * on a pmap that is not the one which is currently active?  in that
136  * case we take the PA of the PDP of non-active pmap and put it in
137  * slot 1023 of the active pmap.  this causes the non-active pmap's
138  * PTEs to get mapped in the final 4MB of the 4GB address space
139  * (e.g. starting at 0xffc00000).
140  *
141  * the following figure shows the effects of the recursive PDP mapping:
142  *
143  *   PDP (%cr3)
144  *   +----+
145  *   |   0| -> PTP#0 that maps VA 0x0 -> 0x400000
146  *   |    |
147  *   |    |
148  *   | 767| -> points back to PDP (%cr3) mapping VA 0xbfc00000 -> 0xc0000000
149  *   | 768| -> first kernel PTP (maps 0xc0000000 -> 0xc0400000)
150  *   |    |
151  *   |1023| -> points to alternate pmap's PDP (maps 0xffc00000 -> end)
152  *   +----+
153  *
154  * note that the PDE#767 VA (0xbfc00000) is defined as "PTE_BASE"
155  * note that the PDE#1023 VA (0xffc00000) is defined as "APTE_BASE"
156  *
157  * starting at VA 0xbfc00000 the current active PDP (%cr3) acts as a
158  * PTP:
159  *
160  * PTP#767 == PDP(%cr3) => maps VA 0xbfc00000 -> 0xc0000000
161  *   +----+
162  *   |   0| -> maps the contents of PTP#0 at VA 0xbfc00000->0xbfc01000
163  *   |    |
164  *   |    |
165  *   | 767| -> maps contents of PTP#767 (the PDP) at VA 0xbfeff000
166  *   | 768| -> maps contents of first kernel PTP
167  *   |    |
168  *   |1023|
169  *   +----+
170  *
171  * note that mapping of the PDP at PTP#767's VA (0xbfeff000) is
172  * defined as "PDP_BASE".... within that mapping there are two
173  * defines:
174  *   "PDP_PDE" (0xbfeffbfc) is the VA of the PDE in the PDP
175  *      which points back to itself.
176  *   "APDP_PDE" (0xbfeffffc) is the VA of the PDE in the PDP which
177  *      establishes the recursive mapping of the alternate pmap.
178  *      to set the alternate PDP, one just has to put the correct
179  *	PA info in *APDP_PDE.
180  *
181  * note that in the APTE_BASE space, the APDP appears at VA
182  * "APDP_BASE" (0xfffff000).
183  *
184  * - PAE support -
185  * ---------------
186  *
187  * PAE adds another layer of indirection during address translation, breaking
188  * up the translation process in 3 different levels:
189  * - L3 page directory, containing 4 * 64-bits addresses (index determined by
190  * bits [31:30] from the virtual address). This breaks up the address space
191  * in 4 1GB regions.
192  * - the PD (L2), containing 512 64-bits addresses, breaking each L3 region
193  * in 512 * 2MB regions.
194  * - the PT (L1), also containing 512 64-bits addresses (at L1, the size of
195  * the pages is still 4K).
196  *
197  * The kernel virtual space is mapped by the last entry in the L3 page,
198  * the first 3 entries mapping the user VA space.
199  *
200  * Because the L3 has only 4 entries of 1GB each, we can't use recursive
201  * mappings at this level for PDP_PDE and APDP_PDE (this would eat up 2 of
202  * the 4GB virtual space). There are also restrictions imposed by Xen on the
203  * last entry of the L3 PD (reference count to this page cannot be bigger
204  * than 1), which makes it hard to use one L3 page per pmap to switch
205  * between pmaps using %cr3.
206  *
207  * As such, each CPU gets its own L3 page that is always loaded into its %cr3
208  * (ci_pae_l3_pd in the associated cpu_info struct). We claim that the VM has
209  * only a 2-level PTP (similar to the non-PAE case). L2 PD is now 4 contiguous
210  * pages long (corresponding to the 4 entries of the L3), and the different
211  * index/slots (like PDP_PDE) are adapted accordingly.
212  *
213  * Kernel space remains in L3[3], L3[0-2] maps the user VA space. Switching
214  * between pmaps consists in modifying the first 3 entries of the CPU's L3 page.
215  *
216  * PTE_BASE and APTE_BASE will need 4 entries in the L2 PD pages to map the
217  * L2 pages recursively.
218  *
219  * In addition, for Xen, we can't recursively map L3[3] (Xen wants the ref
220  * count on this page to be exactly one), so we use a shadow PD page for
221  * the last L2 PD. The shadow page could be static too, but to make pm_pdir[]
222  * contiguous we'll allocate/copy one page per pmap.
223  */
224 /* XXX MP should we allocate one APDP_PDE per processor?? */
225 
226 /*
227  * Mask to get rid of the sign-extended part of addresses.
228  */
229 #define VA_SIGN_MASK		0
230 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
231 /*
232  * XXXfvdl this one's not right.
233  */
234 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
235 
236 /*
237  * the following defines identify the slots used as described above.
238  */
239 #ifdef PAE
240 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-4) /* 1532: for recursive PDP map */
241 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 1536: start of kernel space */
242 #ifndef XEN
243 #define L2_SLOT_APTE	2044		/* 2044: alternative recursive slot */
244 #else
245 #define L2_SLOT_APTE	1960		/* 1964-2047 reserved by Xen */
246 #endif
247 #else /* PAE */
248 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-1) /* 767: for recursive PDP map */
249 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 768: start of kernel space */
250 #ifndef XEN
251 #define L2_SLOT_APTE	1023		/* 1023: alternative recursive slot */
252 #else
253 #define L2_SLOT_APTE	1007		/* 1008-1023 reserved by Xen */
254 #endif
255 #endif /* PAE */
256 
257 #define	L2_SLOT_KERNBASE L2_SLOT_KERN
258 
259 #define PDIR_SLOT_KERN	L2_SLOT_KERN
260 #define PDIR_SLOT_PTE	L2_SLOT_PTE
261 #define PDIR_SLOT_APTE	L2_SLOT_APTE
262 
263 /*
264  * the following defines give the virtual addresses of various MMU
265  * data structures:
266  * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
267  * PDP_BASE and APDP_BASE: the base VA of the recursive mapping of the PDP
268  * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
269  */
270 
271 #define PTE_BASE  ((pt_entry_t *) (PDIR_SLOT_PTE * NBPD_L2))
272 #define APTE_BASE ((pt_entry_t *) (VA_SIGN_NEG((PDIR_SLOT_APTE * NBPD_L2))))
273 
274 #define L1_BASE		PTE_BASE
275 #define AL1_BASE	APTE_BASE
276 
277 #define L2_BASE ((pd_entry_t *)((char *)L1_BASE + L2_SLOT_PTE * NBPD_L1))
278 #define AL2_BASE ((pd_entry_t *)((char *)AL1_BASE + L2_SLOT_PTE * NBPD_L1))
279 
280 #define PDP_PDE		(L2_BASE + PDIR_SLOT_PTE)
281 #if defined(PAE) && defined(XEN)
282 /*
283  * when PAE is in use under Xen, we can't write APDP_PDE through the recursive
284  * mapping, because it points to the shadow PD. Use the kernel PD instead,
285  * which is static
286  */
287 #define APDP_PDE	(&pmap_kl2pd[l2tol2(PDIR_SLOT_APTE)])
288 #define APDP_PDE_SHADOW	(L2_BASE + PDIR_SLOT_APTE)
289 #else /* PAE && XEN */
290 #define APDP_PDE	(L2_BASE + PDIR_SLOT_APTE)
291 #endif /* PAE && XEN */
292 
293 #define PDP_BASE	L2_BASE
294 #define APDP_BASE	AL2_BASE
295 
296 /* largest value (-1 for APTP space) */
297 #define NKL2_MAX_ENTRIES	(NTOPLEVEL_PDES - (KERNBASE/NBPD_L2) - 1)
298 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * NPDPG)
299 
300 #define NKL2_KIMG_ENTRIES	0	/* XXX unused */
301 
302 #define NKL2_START_ENTRIES	0	/* XXX computed on runtime */
303 #define NKL1_START_ENTRIES	0	/* XXX unused */
304 
305 #define NTOPLEVEL_PDES		(PAGE_SIZE * PDP_SIZE / (sizeof (pd_entry_t)))
306 
307 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
308 
309 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME }
310 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT }
311 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES }
312 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES }
313 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2 }
314 #define PDES_INITIALIZER	{ L2_BASE }
315 #define APDES_INITIALIZER	{ AL2_BASE }
316 
317 #define PTP_LEVELS	2
318 
319 /*
320  * PG_AVAIL usage: we make use of the ignored bits of the PTE
321  */
322 
323 #define PG_W		PG_AVAIL1	/* "wired" mapping */
324 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
325 #define PG_X		PG_AVAIL3	/* executable mapping */
326 
327 /*
328  * Number of PTE's per cache line.  4 byte pte, 32-byte cache line
329  * Used to avoid false sharing of cache lines.
330  */
331 #ifdef PAE
332 #define NPTECL		4
333 #else
334 #define NPTECL		8
335 #endif
336 
337 #include <x86/pmap.h>
338 
339 #ifndef XEN
340 #define pmap_pa2pte(a)			(a)
341 #define pmap_pte2pa(a)			((a) & PG_FRAME)
342 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
343 #define pmap_pte_flush()		/* nothing */
344 
345 #ifdef PAE
346 #define pmap_pte_cas(p, o, n)		atomic_cas_64((p), (o), (n))
347 #define pmap_pte_testset(p, n)		\
348     atomic_swap_64((volatile uint64_t *)p, n)
349 #define pmap_pte_setbits(p, b)		\
350     atomic_or_64((volatile uint64_t *)p, b)
351 #define pmap_pte_clearbits(p, b)	\
352     atomic_and_64((volatile uint64_t *)p, ~(b))
353 #else /* PAE */
354 #define pmap_pte_cas(p, o, n)		atomic_cas_32((p), (o), (n))
355 #define pmap_pte_testset(p, n)		\
356     atomic_swap_ulong((volatile unsigned long *)p, n)
357 #define pmap_pte_setbits(p, b)		\
358     atomic_or_ulong((volatile unsigned long *)p, b)
359 #define pmap_pte_clearbits(p, b)	\
360     atomic_and_ulong((volatile unsigned long *)p, ~(b))
361 #endif /* PAE */
362 
363 #else /* XEN */
364 static __inline pt_entry_t
365 pmap_pa2pte(paddr_t pa)
366 {
367 	return (pt_entry_t)xpmap_ptom_masked(pa);
368 }
369 
370 static __inline paddr_t
371 pmap_pte2pa(pt_entry_t pte)
372 {
373 	return xpmap_mtop_masked(pte & PG_FRAME);
374 }
375 static __inline void
376 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
377 {
378 	int s = splvm();
379 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
380 	splx(s);
381 }
382 
383 static __inline pt_entry_t
384 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
385 {
386 	int s = splvm();
387 	pt_entry_t opte = *ptep;
388 
389 	if (opte == o) {
390 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
391 		xpq_flush_queue();
392 	}
393 	splx(s);
394 	return opte;
395 }
396 
397 static __inline pt_entry_t
398 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
399 {
400 	int s = splvm();
401 	pt_entry_t opte = *pte;
402 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
403 	    npte);
404 	xpq_flush_queue();
405 	splx(s);
406 	return opte;
407 }
408 
409 static __inline void
410 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
411 {
412 	int s = splvm();
413 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
414 	xpq_flush_queue();
415 	splx(s);
416 }
417 
418 static __inline void
419 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
420 {
421 	int s = splvm();
422 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
423 	    (*pte) & ~bits);
424 	xpq_flush_queue();
425 	splx(s);
426 }
427 
428 static __inline void
429 pmap_pte_flush(void)
430 {
431 	int s = splvm();
432 	xpq_flush_queue();
433 	splx(s);
434 }
435 
436 #endif
437 
438 #ifdef PAE
439 /* Address of the static kernel's L2 page */
440 pd_entry_t *pmap_kl2pd;
441 paddr_t pmap_kl2paddr;
442 #endif
443 
444 
445 struct trapframe;
446 
447 int	pmap_exec_fixup(struct vm_map *, struct trapframe *, struct pcb *);
448 void	pmap_ldt_cleanup(struct lwp *);
449 
450 #include <x86/pmap_pv.h>
451 
452 #define	__HAVE_VM_PAGE_MD
453 #define	VM_MDPAGE_INIT(pg) \
454 	memset(&(pg)->mdpage, 0, sizeof((pg)->mdpage)); \
455 	PMAP_PAGE_INIT(&(pg)->mdpage.mp_pp)
456 
457 struct vm_page_md {
458 	struct pmap_page mp_pp;
459 };
460 
461 #endif	/* _I386_PMAP_H_ */
462