xref: /netbsd-src/sys/arch/i386/include/pmap.h (revision 7f21db1c0118155e0dd40b75182e30c589d9f63e)
1 /*	$NetBSD: pmap.h,v 1.104 2010/02/09 22:51:13 jym 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  * When PAE is in use, the L3 page directory breaks up the address space in
185  * 4 1GB * regions, each of them broken in 512 2MB regions by the L2 PD
186  * (the size of the pages at the L1 level is still 4K).
187  * The kernel virtual space is mapped by the last entry in the L3 page,
188  * the first 3 entries mapping the user VA space.
189  * Because the L3 has only 4 entries of 1GB each, we can't use recursive
190  * mappings at this level for PDP_PDE and APDP_PDE (this would eat 2 of the
191  * 4GB virtual space). There's also restrictions imposed by Xen on the
192  * last entry of the L3 PD, which makes it hard to use one L3 page per pmap
193  * switch %cr3 to switch pmaps. So we use one static L3 page which is
194  * always loaded in %cr3, and we use it as 2 virtual PD pointers: one for
195  * kernel space (L3[3], always loaded), and one for user space (in fact the
196  * first 3 entries of the L3 PD), and we claim the VM has only a 2-level
197  * PTP (with the L2 index extended by 2 bytes).
198  * PTE_BASE and APTE_BASE will need 4 entries in the L2 page table.
199  * In addition, we can't recursively map L3[3] (Xen wants the ref count on
200  * this page to be exactly once), so we use a shadow PD page for the last
201  * L2 PD. The shadow page could be static too, but to make pm_pdir[]
202  * contigous we'll allocate/copy one page per pmap.
203  */
204 /* XXX MP should we allocate one APDP_PDE per processor?? */
205 
206 /*
207  * Mask to get rid of the sign-extended part of addresses.
208  */
209 #define VA_SIGN_MASK		0
210 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
211 /*
212  * XXXfvdl this one's not right.
213  */
214 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
215 
216 /*
217  * the following defines identify the slots used as described above.
218  */
219 #ifdef PAE
220 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-4) /* 1532: for recursive PDP map */
221 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 1536: start of kernel space */
222 #define	L2_SLOT_KERNBASE L2_SLOT_KERN
223 #define L2_SLOT_APTE	1960                 /* 1964-2047 reserved by Xen */
224 #else /* PAE */
225 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-1) /* 767: for recursive PDP map */
226 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 768: start of kernel space */
227 #define	L2_SLOT_KERNBASE L2_SLOT_KERN
228 #ifndef XEN
229 #define L2_SLOT_APTE	1023		 /* 1023: alternative recursive slot */
230 #else
231 #define L2_SLOT_APTE	1007		/* 1008-1023 reserved by Xen */
232 #endif
233 #endif /* PAE */
234 
235 
236 #define PDIR_SLOT_KERN	L2_SLOT_KERN
237 #define PDIR_SLOT_PTE	L2_SLOT_PTE
238 #define PDIR_SLOT_APTE	L2_SLOT_APTE
239 
240 /*
241  * the following defines give the virtual addresses of various MMU
242  * data structures:
243  * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
244  * PDP_BASE and APDP_BASE: the base VA of the recursive mapping of the PDP
245  * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
246  */
247 
248 #define PTE_BASE  ((pt_entry_t *) (PDIR_SLOT_PTE * NBPD_L2))
249 #define APTE_BASE ((pt_entry_t *) (VA_SIGN_NEG((PDIR_SLOT_APTE * NBPD_L2))))
250 
251 #define L1_BASE		PTE_BASE
252 #define AL1_BASE	APTE_BASE
253 
254 #define L2_BASE ((pd_entry_t *)((char *)L1_BASE + L2_SLOT_PTE * NBPD_L1))
255 #define AL2_BASE ((pd_entry_t *)((char *)AL1_BASE + L2_SLOT_PTE * NBPD_L1))
256 
257 #define PDP_PDE		(L2_BASE + PDIR_SLOT_PTE)
258 #ifdef PAE
259 /*
260  * when PAE is in use we can't write APDP_PDE though the recursive mapping,
261  * because it points to the shadow PD. Use the kernel PD instead, which is
262  * static
263  */
264 #define APDP_PDE	(&pmap_kl2pd[l2tol2(PDIR_SLOT_APTE)])
265 #define APDP_PDE_SHADOW	(L2_BASE + PDIR_SLOT_APTE)
266 #else /* PAE */
267 #define APDP_PDE	(L2_BASE + PDIR_SLOT_APTE)
268 #endif /* PAE */
269 
270 #define PDP_BASE	L2_BASE
271 #define APDP_BASE	AL2_BASE
272 
273 /* largest value (-1 for APTP space) */
274 #define NKL2_MAX_ENTRIES	(NTOPLEVEL_PDES - (KERNBASE/NBPD_L2) - 1)
275 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * NPDPG)
276 
277 #define NKL2_KIMG_ENTRIES	0	/* XXX unused */
278 
279 #define NKL2_START_ENTRIES	0	/* XXX computed on runtime */
280 #define NKL1_START_ENTRIES	0	/* XXX unused */
281 
282 #ifdef PAE
283 #define NTOPLEVEL_PDES		(PAGE_SIZE * 4 / (sizeof (pd_entry_t)))
284 #else
285 #define NTOPLEVEL_PDES		(PAGE_SIZE / (sizeof (pd_entry_t)))
286 #endif
287 
288 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
289 
290 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME }
291 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT }
292 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES }
293 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES }
294 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2 }
295 #define PDES_INITIALIZER	{ L2_BASE }
296 #define APDES_INITIALIZER	{ AL2_BASE }
297 
298 #define PTP_LEVELS	2
299 
300 /*
301  * PG_AVAIL usage: we make use of the ignored bits of the PTE
302  */
303 
304 #define PG_W		PG_AVAIL1	/* "wired" mapping */
305 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
306 #define PG_X		PG_AVAIL3	/* executable mapping */
307 
308 /*
309  * Number of PTE's per cache line.  4 byte pte, 32-byte cache line
310  * Used to avoid false sharing of cache lines.
311  */
312 #ifdef PAE
313 #define NPTECL		4
314 #else
315 #define NPTECL		8
316 #endif
317 
318 #include <x86/pmap.h>
319 
320 #ifndef XEN
321 #define pmap_pa2pte(a)			(a)
322 #define pmap_pte2pa(a)			((a) & PG_FRAME)
323 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
324 #define pmap_pte_cas(p, o, n)		atomic_cas_32((p), (o), (n))
325 #define pmap_pte_testset(p, n)		\
326     atomic_swap_ulong((volatile unsigned long *)p, n)
327 #define pmap_pte_setbits(p, b)		\
328     atomic_or_ulong((volatile unsigned long *)p, b)
329 #define pmap_pte_clearbits(p, b)	\
330     atomic_and_ulong((volatile unsigned long *)p, ~(b))
331 #define pmap_pte_flush()		/* nothing */
332 #else
333 static __inline pt_entry_t
334 pmap_pa2pte(paddr_t pa)
335 {
336 	return (pt_entry_t)xpmap_ptom_masked(pa);
337 }
338 
339 static __inline paddr_t
340 pmap_pte2pa(pt_entry_t pte)
341 {
342 	return xpmap_mtop_masked(pte & PG_FRAME);
343 }
344 static __inline void
345 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
346 {
347 	int s = splvm();
348 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
349 	splx(s);
350 }
351 
352 static __inline pt_entry_t
353 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
354 {
355 	int s = splvm();
356 	pt_entry_t opte = *ptep;
357 
358 	if (opte == o) {
359 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
360 		xpq_flush_queue();
361 	}
362 	splx(s);
363 	return opte;
364 }
365 
366 static __inline pt_entry_t
367 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
368 {
369 	int s = splvm();
370 	pt_entry_t opte = *pte;
371 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
372 	    npte);
373 	xpq_flush_queue();
374 	splx(s);
375 	return opte;
376 }
377 
378 static __inline void
379 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
380 {
381 	int s = splvm();
382 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
383 	xpq_flush_queue();
384 	splx(s);
385 }
386 
387 static __inline void
388 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
389 {
390 	int s = splvm();
391 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
392 	    (*pte) & ~bits);
393 	xpq_flush_queue();
394 	splx(s);
395 }
396 
397 static __inline void
398 pmap_pte_flush(void)
399 {
400 	int s = splvm();
401 	xpq_flush_queue();
402 	splx(s);
403 }
404 
405 #endif
406 
407 #ifdef PAE
408 /* addresses of static pages used for PAE pmap: */
409 /* the L3 page */
410 pd_entry_t *pmap_l3pd;
411 paddr_t pmap_l3paddr;
412 /* the kernel's L2 page */
413 pd_entry_t *pmap_kl2pd;
414 paddr_t pmap_kl2paddr;
415 #endif
416 
417 
418 struct trapframe;
419 
420 int	pmap_exec_fixup(struct vm_map *, struct trapframe *, struct pcb *);
421 void	pmap_ldt_cleanup(struct lwp *);
422 
423 #endif	/* _I386_PMAP_H_ */
424