xref: /netbsd-src/sys/arch/i386/include/pmap.h (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*	$NetBSD: pmap.h,v 1.109 2011/02/01 20:09:08 chuck 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 #ifndef	_I386_PMAP_H_
64 #define	_I386_PMAP_H_
65 
66 #if defined(_KERNEL_OPT)
67 #include "opt_user_ldt.h"
68 #include "opt_xen.h"
69 #endif
70 
71 #include <sys/atomic.h>
72 
73 #include <i386/pte.h>
74 #include <machine/segments.h>
75 #if defined(_KERNEL)
76 #include <machine/cpufunc.h>
77 #endif
78 
79 #include <uvm/uvm_object.h>
80 #ifdef XEN
81 #include <xen/xenfunc.h>
82 #include <xen/xenpmap.h>
83 #endif /* XEN */
84 
85 /*
86  * see pte.h for a description of i386 MMU terminology and hardware
87  * interface.
88  *
89  * a pmap describes a processes' 4GB virtual address space.  when PAE
90  * is not in use, this virtual address space can be broken up into 1024 4MB
91  * regions which are described by PDEs in the PDP.  the PDEs are defined as
92  * follows:
93  *
94  * (ranges are inclusive -> exclusive, just like vm_map_entry start/end)
95  * (the following assumes that KERNBASE is 0xc0000000)
96  *
97  * PDE#s	VA range		usage
98  * 0->766	0x0 -> 0xbfc00000	user address space
99  * 767		0xbfc00000->		recursive mapping of PDP (used for
100  *			0xc0000000	linear mapping of PTPs)
101  * 768->1023	0xc0000000->		kernel address space (constant
102  *			0xffc00000	across all pmap's/processes)
103  * 1023		0xffc00000->		"alternate" recursive PDP mapping
104  *			<end>		(for other pmaps)
105  *
106  *
107  * note: a recursive PDP mapping provides a way to map all the PTEs for
108  * a 4GB address space into a linear chunk of virtual memory.  in other
109  * words, the PTE for page 0 is the first int mapped into the 4MB recursive
110  * area.  the PTE for page 1 is the second int.  the very last int in the
111  * 4MB range is the PTE that maps VA 0xfffff000 (the last page in a 4GB
112  * address).
113  *
114  * all pmap's PD's must have the same values in slots 768->1023 so that
115  * the kernel is always mapped in every process.  these values are loaded
116  * into the PD at pmap creation time.
117  *
118  * at any one time only one pmap can be active on a processor.  this is
119  * the pmap whose PDP is pointed to by processor register %cr3.  this pmap
120  * will have all its PTEs mapped into memory at the recursive mapping
121  * point (slot #767 as show above).  when the pmap code wants to find the
122  * PTE for a virtual address, all it has to do is the following:
123  *
124  * address of PTE = (767 * 4MB) + (VA / PAGE_SIZE) * sizeof(pt_entry_t)
125  *                = 0xbfc00000 + (VA / 4096) * 4
126  *
127  * what happens if the pmap layer is asked to perform an operation
128  * on a pmap that is not the one which is currently active?  in that
129  * case we take the PA of the PDP of non-active pmap and put it in
130  * slot 1023 of the active pmap.  this causes the non-active pmap's
131  * PTEs to get mapped in the final 4MB of the 4GB address space
132  * (e.g. starting at 0xffc00000).
133  *
134  * the following figure shows the effects of the recursive PDP mapping:
135  *
136  *   PDP (%cr3)
137  *   +----+
138  *   |   0| -> PTP#0 that maps VA 0x0 -> 0x400000
139  *   |    |
140  *   |    |
141  *   | 767| -> points back to PDP (%cr3) mapping VA 0xbfc00000 -> 0xc0000000
142  *   | 768| -> first kernel PTP (maps 0xc0000000 -> 0xc0400000)
143  *   |    |
144  *   |1023| -> points to alternate pmap's PDP (maps 0xffc00000 -> end)
145  *   +----+
146  *
147  * note that the PDE#767 VA (0xbfc00000) is defined as "PTE_BASE"
148  * note that the PDE#1023 VA (0xffc00000) is defined as "APTE_BASE"
149  *
150  * starting at VA 0xbfc00000 the current active PDP (%cr3) acts as a
151  * PTP:
152  *
153  * PTP#767 == PDP(%cr3) => maps VA 0xbfc00000 -> 0xc0000000
154  *   +----+
155  *   |   0| -> maps the contents of PTP#0 at VA 0xbfc00000->0xbfc01000
156  *   |    |
157  *   |    |
158  *   | 767| -> maps contents of PTP#767 (the PDP) at VA 0xbfeff000
159  *   | 768| -> maps contents of first kernel PTP
160  *   |    |
161  *   |1023|
162  *   +----+
163  *
164  * note that mapping of the PDP at PTP#767's VA (0xbfeff000) is
165  * defined as "PDP_BASE".... within that mapping there are two
166  * defines:
167  *   "PDP_PDE" (0xbfeffbfc) is the VA of the PDE in the PDP
168  *      which points back to itself.
169  *   "APDP_PDE" (0xbfeffffc) is the VA of the PDE in the PDP which
170  *      establishes the recursive mapping of the alternate pmap.
171  *      to set the alternate PDP, one just has to put the correct
172  *	PA info in *APDP_PDE.
173  *
174  * note that in the APTE_BASE space, the APDP appears at VA
175  * "APDP_BASE" (0xfffff000).
176  *
177  * - PAE support -
178  * ---------------
179  *
180  * PAE adds another layer of indirection during address translation, breaking
181  * up the translation process in 3 different levels:
182  * - L3 page directory, containing 4 * 64-bits addresses (index determined by
183  * bits [31:30] from the virtual address). This breaks up the address space
184  * in 4 1GB regions.
185  * - the PD (L2), containing 512 64-bits addresses, breaking each L3 region
186  * in 512 * 2MB regions.
187  * - the PT (L1), also containing 512 64-bits addresses (at L1, the size of
188  * the pages is still 4K).
189  *
190  * The kernel virtual space is mapped by the last entry in the L3 page,
191  * the first 3 entries mapping the user VA space.
192  *
193  * Because the L3 has only 4 entries of 1GB each, we can't use recursive
194  * mappings at this level for PDP_PDE and APDP_PDE (this would eat up 2 of
195  * the 4GB virtual space). There are also restrictions imposed by Xen on the
196  * last entry of the L3 PD (reference count to this page cannot be bigger
197  * than 1), which makes it hard to use one L3 page per pmap to switch
198  * between pmaps using %cr3.
199  *
200  * As such, each CPU gets its own L3 page that is always loaded into its %cr3
201  * (ci_pae_l3_pd in the associated cpu_info struct). We claim that the VM has
202  * only a 2-level PTP (similar to the non-PAE case). L2 PD is now 4 contiguous
203  * pages long (corresponding to the 4 entries of the L3), and the different
204  * index/slots (like PDP_PDE) are adapted accordingly.
205  *
206  * Kernel space remains in L3[3], L3[0-2] maps the user VA space. Switching
207  * between pmaps consists in modifying the first 3 entries of the CPU's L3 page.
208  *
209  * PTE_BASE and APTE_BASE will need 4 entries in the L2 PD pages to map the
210  * L2 pages recursively.
211  *
212  * In addition, for Xen, we can't recursively map L3[3] (Xen wants the ref
213  * count on this page to be exactly one), so we use a shadow PD page for
214  * the last L2 PD. The shadow page could be static too, but to make pm_pdir[]
215  * contiguous we'll allocate/copy one page per pmap.
216  */
217 /* XXX MP should we allocate one APDP_PDE per processor?? */
218 
219 /*
220  * Mask to get rid of the sign-extended part of addresses.
221  */
222 #define VA_SIGN_MASK		0
223 #define VA_SIGN_NEG(va)		((va) | VA_SIGN_MASK)
224 /*
225  * XXXfvdl this one's not right.
226  */
227 #define VA_SIGN_POS(va)		((va) & ~VA_SIGN_MASK)
228 
229 /*
230  * the following defines identify the slots used as described above.
231  */
232 #ifdef PAE
233 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-4) /* 1532: for recursive PDP map */
234 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 1536: start of kernel space */
235 #ifndef XEN
236 #define L2_SLOT_APTE	2044		/* 2044: alternative recursive slot */
237 #else
238 #define L2_SLOT_APTE	1960		/* 1964-2047 reserved by Xen */
239 #endif
240 #else /* PAE */
241 #define L2_SLOT_PTE	(KERNBASE/NBPD_L2-1) /* 767: for recursive PDP map */
242 #define L2_SLOT_KERN	(KERNBASE/NBPD_L2)   /* 768: start of kernel space */
243 #ifndef XEN
244 #define L2_SLOT_APTE	1023		/* 1023: alternative recursive slot */
245 #else
246 #define L2_SLOT_APTE	1007		/* 1008-1023 reserved by Xen */
247 #endif
248 #endif /* PAE */
249 
250 #define	L2_SLOT_KERNBASE L2_SLOT_KERN
251 
252 #define PDIR_SLOT_KERN	L2_SLOT_KERN
253 #define PDIR_SLOT_PTE	L2_SLOT_PTE
254 #define PDIR_SLOT_APTE	L2_SLOT_APTE
255 
256 /*
257  * the following defines give the virtual addresses of various MMU
258  * data structures:
259  * PTE_BASE and APTE_BASE: the base VA of the linear PTE mappings
260  * PDP_BASE and APDP_BASE: the base VA of the recursive mapping of the PDP
261  * PDP_PDE and APDP_PDE: the VA of the PDE that points back to the PDP/APDP
262  */
263 
264 #define PTE_BASE  ((pt_entry_t *) (PDIR_SLOT_PTE * NBPD_L2))
265 #define APTE_BASE ((pt_entry_t *) (VA_SIGN_NEG((PDIR_SLOT_APTE * NBPD_L2))))
266 
267 #define L1_BASE		PTE_BASE
268 #define AL1_BASE	APTE_BASE
269 
270 #define L2_BASE ((pd_entry_t *)((char *)L1_BASE + L2_SLOT_PTE * NBPD_L1))
271 #define AL2_BASE ((pd_entry_t *)((char *)AL1_BASE + L2_SLOT_PTE * NBPD_L1))
272 
273 #define PDP_PDE		(L2_BASE + PDIR_SLOT_PTE)
274 #if defined(PAE) && defined(XEN)
275 /*
276  * when PAE is in use under Xen, we can't write APDP_PDE through the recursive
277  * mapping, because it points to the shadow PD. Use the kernel PD instead,
278  * which is static
279  */
280 #define APDP_PDE	(&pmap_kl2pd[l2tol2(PDIR_SLOT_APTE)])
281 #define APDP_PDE_SHADOW	(L2_BASE + PDIR_SLOT_APTE)
282 #else /* PAE && XEN */
283 #define APDP_PDE	(L2_BASE + PDIR_SLOT_APTE)
284 #endif /* PAE && XEN */
285 
286 #define PDP_BASE	L2_BASE
287 #define APDP_BASE	AL2_BASE
288 
289 /* largest value (-1 for APTP space) */
290 #define NKL2_MAX_ENTRIES	(NTOPLEVEL_PDES - (KERNBASE/NBPD_L2) - 1)
291 #define NKL1_MAX_ENTRIES	(unsigned long)(NKL2_MAX_ENTRIES * NPDPG)
292 
293 #define NKL2_KIMG_ENTRIES	0	/* XXX unused */
294 
295 #define NKL2_START_ENTRIES	0	/* XXX computed on runtime */
296 #define NKL1_START_ENTRIES	0	/* XXX unused */
297 
298 #define NTOPLEVEL_PDES		(PAGE_SIZE * PDP_SIZE / (sizeof (pd_entry_t)))
299 
300 #define NPDPG			(PAGE_SIZE / sizeof (pd_entry_t))
301 
302 #define PTP_MASK_INITIALIZER	{ L1_FRAME, L2_FRAME }
303 #define PTP_SHIFT_INITIALIZER	{ L1_SHIFT, L2_SHIFT }
304 #define NKPTP_INITIALIZER	{ NKL1_START_ENTRIES, NKL2_START_ENTRIES }
305 #define NKPTPMAX_INITIALIZER	{ NKL1_MAX_ENTRIES, NKL2_MAX_ENTRIES }
306 #define NBPD_INITIALIZER	{ NBPD_L1, NBPD_L2 }
307 #define PDES_INITIALIZER	{ L2_BASE }
308 #define APDES_INITIALIZER	{ AL2_BASE }
309 
310 #define PTP_LEVELS	2
311 
312 /*
313  * PG_AVAIL usage: we make use of the ignored bits of the PTE
314  */
315 
316 #define PG_W		PG_AVAIL1	/* "wired" mapping */
317 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
318 #define PG_X		PG_AVAIL3	/* executable mapping */
319 
320 /*
321  * Number of PTE's per cache line.  4 byte pte, 32-byte cache line
322  * Used to avoid false sharing of cache lines.
323  */
324 #ifdef PAE
325 #define NPTECL		4
326 #else
327 #define NPTECL		8
328 #endif
329 
330 #include <x86/pmap.h>
331 
332 #ifndef XEN
333 #define pmap_pa2pte(a)			(a)
334 #define pmap_pte2pa(a)			((a) & PG_FRAME)
335 #define pmap_pte_set(p, n)		do { *(p) = (n); } while (0)
336 #define pmap_pte_flush()		/* nothing */
337 
338 #ifdef PAE
339 #define pmap_pte_cas(p, o, n)		atomic_cas_64((p), (o), (n))
340 #define pmap_pte_testset(p, n)		\
341     atomic_swap_64((volatile uint64_t *)p, n)
342 #define pmap_pte_setbits(p, b)		\
343     atomic_or_64((volatile uint64_t *)p, b)
344 #define pmap_pte_clearbits(p, b)	\
345     atomic_and_64((volatile uint64_t *)p, ~(b))
346 #else /* PAE */
347 #define pmap_pte_cas(p, o, n)		atomic_cas_32((p), (o), (n))
348 #define pmap_pte_testset(p, n)		\
349     atomic_swap_ulong((volatile unsigned long *)p, n)
350 #define pmap_pte_setbits(p, b)		\
351     atomic_or_ulong((volatile unsigned long *)p, b)
352 #define pmap_pte_clearbits(p, b)	\
353     atomic_and_ulong((volatile unsigned long *)p, ~(b))
354 #endif /* PAE */
355 
356 #else /* XEN */
357 static __inline pt_entry_t
358 pmap_pa2pte(paddr_t pa)
359 {
360 	return (pt_entry_t)xpmap_ptom_masked(pa);
361 }
362 
363 static __inline paddr_t
364 pmap_pte2pa(pt_entry_t pte)
365 {
366 	return xpmap_mtop_masked(pte & PG_FRAME);
367 }
368 static __inline void
369 pmap_pte_set(pt_entry_t *pte, pt_entry_t npte)
370 {
371 	int s = splvm();
372 	xpq_queue_pte_update(xpmap_ptetomach(pte), npte);
373 	splx(s);
374 }
375 
376 static __inline pt_entry_t
377 pmap_pte_cas(volatile pt_entry_t *ptep, pt_entry_t o, pt_entry_t n)
378 {
379 	int s = splvm();
380 	pt_entry_t opte = *ptep;
381 
382 	if (opte == o) {
383 		xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(ptep)), n);
384 		xpq_flush_queue();
385 	}
386 	splx(s);
387 	return opte;
388 }
389 
390 static __inline pt_entry_t
391 pmap_pte_testset(volatile pt_entry_t *pte, pt_entry_t npte)
392 {
393 	int s = splvm();
394 	pt_entry_t opte = *pte;
395 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
396 	    npte);
397 	xpq_flush_queue();
398 	splx(s);
399 	return opte;
400 }
401 
402 static __inline void
403 pmap_pte_setbits(volatile pt_entry_t *pte, pt_entry_t bits)
404 {
405 	int s = splvm();
406 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)), (*pte) | bits);
407 	xpq_flush_queue();
408 	splx(s);
409 }
410 
411 static __inline void
412 pmap_pte_clearbits(volatile pt_entry_t *pte, pt_entry_t bits)
413 {
414 	int s = splvm();
415 	xpq_queue_pte_update(xpmap_ptetomach(__UNVOLATILE(pte)),
416 	    (*pte) & ~bits);
417 	xpq_flush_queue();
418 	splx(s);
419 }
420 
421 static __inline void
422 pmap_pte_flush(void)
423 {
424 	int s = splvm();
425 	xpq_flush_queue();
426 	splx(s);
427 }
428 
429 #endif
430 
431 #ifdef PAE
432 /* Address of the static kernel's L2 page */
433 pd_entry_t *pmap_kl2pd;
434 paddr_t pmap_kl2paddr;
435 #endif
436 
437 
438 struct trapframe;
439 
440 int	pmap_exec_fixup(struct vm_map *, struct trapframe *, struct pcb *);
441 void	pmap_ldt_cleanup(struct lwp *);
442 
443 #include <x86/pmap_pv.h>
444 
445 #define	__HAVE_VM_PAGE_MD
446 #define	VM_MDPAGE_INIT(pg) \
447 	memset(&(pg)->mdpage, 0, sizeof((pg)->mdpage)); \
448 	PMAP_PAGE_INIT(&(pg)->mdpage.mp_pp)
449 
450 struct vm_page_md {
451 	struct pmap_page mp_pp;
452 };
453 
454 #endif	/* _I386_PMAP_H_ */
455