xref: /illumos-gate/usr/src/contrib/bhyve/amd64/machine/pmap.h (revision d0b3c59ba652f183eeec1414dd9fbdc56bf05cc8)
1 /*-
2  * Copyright (c) 2003 Peter Wemm.
3  * Copyright (c) 1991 Regents of the University of California.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department and William Jolitz of UUNET Technologies Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Derived from hp300 version by Mike Hibler, this version by William
35  * Jolitz uses a recursive map [a pde points to the page directory] to
36  * map the page tables using the pagetables themselves. This is done to
37  * reduce the impact on kernel virtual memory for lots of sparse address
38  * space, and to reduce the cost of memory to each process.
39  *
40  *	from: hp300: @(#)pmap.h	7.2 (Berkeley) 12/16/90
41  *	from: @(#)pmap.h	7.4 (Berkeley) 5/12/91
42  * $FreeBSD$
43  */
44 
45 #ifndef _MACHINE_PMAP_H_
46 #define	_MACHINE_PMAP_H_
47 
48 /*
49  * Page-directory and page-table entries follow this format, with a few
50  * of the fields not present here and there, depending on a lot of things.
51  */
52 				/* ---- Intel Nomenclature ---- */
53 #define	X86_PG_V	0x001	/* P	Valid			*/
54 #define	X86_PG_RW	0x002	/* R/W	Read/Write		*/
55 #define	X86_PG_U	0x004	/* U/S  User/Supervisor		*/
56 #define	X86_PG_NC_PWT	0x008	/* PWT	Write through		*/
57 #define	X86_PG_NC_PCD	0x010	/* PCD	Cache disable		*/
58 #define	X86_PG_A	0x020	/* A	Accessed		*/
59 #define	X86_PG_M	0x040	/* D	Dirty			*/
60 #define	X86_PG_PS	0x080	/* PS	Page size (0=4k,1=2M)	*/
61 #define	X86_PG_PTE_PAT	0x080	/* PAT	PAT index		*/
62 #define	X86_PG_G	0x100	/* G	Global			*/
63 #define	X86_PG_AVAIL1	0x200	/*    /	Available for system	*/
64 #define	X86_PG_AVAIL2	0x400	/*   <	programmers use		*/
65 #define	X86_PG_AVAIL3	0x800	/*    \				*/
66 #define	X86_PG_PDE_PAT	0x1000	/* PAT	PAT index		*/
67 #define	X86_PG_NX	(1ul<<63) /* No-execute */
68 #define	X86_PG_AVAIL(x)	(1ul << (x))
69 
70 /* Page level cache control fields used to determine the PAT type */
71 #define	X86_PG_PDE_CACHE (X86_PG_PDE_PAT | X86_PG_NC_PWT | X86_PG_NC_PCD)
72 #define	X86_PG_PTE_CACHE (X86_PG_PTE_PAT | X86_PG_NC_PWT | X86_PG_NC_PCD)
73 
74 /*
75  * Intel extended page table (EPT) bit definitions.
76  */
77 #define	EPT_PG_READ		0x001	/* R	Read		*/
78 #define	EPT_PG_WRITE		0x002	/* W	Write		*/
79 #define	EPT_PG_EXECUTE		0x004	/* X	Execute		*/
80 #define	EPT_PG_IGNORE_PAT	0x040	/* IPAT	Ignore PAT	*/
81 #define	EPT_PG_PS		0x080	/* PS	Page size	*/
82 #define	EPT_PG_A		0x100	/* A	Accessed	*/
83 #define	EPT_PG_M		0x200	/* D	Dirty		*/
84 #define	EPT_PG_MEMORY_TYPE(x)	((x) << 3) /* MT Memory Type	*/
85 
86 /*
87  * Define the PG_xx macros in terms of the bits on x86 PTEs.
88  */
89 #define	PG_V		X86_PG_V
90 #define	PG_RW		X86_PG_RW
91 #define	PG_U		X86_PG_U
92 #define	PG_NC_PWT	X86_PG_NC_PWT
93 #define	PG_NC_PCD	X86_PG_NC_PCD
94 #define	PG_A		X86_PG_A
95 #define	PG_M		X86_PG_M
96 #define	PG_PS		X86_PG_PS
97 #define	PG_PTE_PAT	X86_PG_PTE_PAT
98 #define	PG_G		X86_PG_G
99 #define	PG_AVAIL1	X86_PG_AVAIL1
100 #define	PG_AVAIL2	X86_PG_AVAIL2
101 #define	PG_AVAIL3	X86_PG_AVAIL3
102 #define	PG_PDE_PAT	X86_PG_PDE_PAT
103 #define	PG_NX		X86_PG_NX
104 #define	PG_PDE_CACHE	X86_PG_PDE_CACHE
105 #define	PG_PTE_CACHE	X86_PG_PTE_CACHE
106 
107 /* Our various interpretations of the above */
108 #define	PG_W		X86_PG_AVAIL3	/* "Wired" pseudoflag */
109 #define	PG_MANAGED	X86_PG_AVAIL2
110 #define	EPT_PG_EMUL_V	X86_PG_AVAIL(52)
111 #define	EPT_PG_EMUL_RW	X86_PG_AVAIL(53)
112 #define	PG_PROMOTED	X86_PG_AVAIL(54)	/* PDE only */
113 #define	PG_FRAME	(0x000ffffffffff000ul)
114 #define	PG_PS_FRAME	(0x000fffffffe00000ul)
115 
116 /*
117  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
118  * (PTE) page mappings have identical settings for the following fields:
119  */
120 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_CACHE | \
121 	    PG_M | PG_A | PG_U | PG_RW | PG_V)
122 
123 /*
124  * Page Protection Exception bits
125  */
126 
127 #define PGEX_P		0x01	/* Protection violation vs. not present */
128 #define PGEX_W		0x02	/* during a Write cycle */
129 #define PGEX_U		0x04	/* access from User mode (UPL) */
130 #define PGEX_RSV	0x08	/* reserved PTE field is non-zero */
131 #define PGEX_I		0x10	/* during an instruction fetch */
132 
133 /*
134  * undef the PG_xx macros that define bits in the regular x86 PTEs that
135  * have a different position in nested PTEs. This is done when compiling
136  * code that needs to be aware of the differences between regular x86 and
137  * nested PTEs.
138  *
139  * The appropriate bitmask will be calculated at runtime based on the pmap
140  * type.
141  */
142 #ifdef AMD64_NPT_AWARE
143 #undef PG_AVAIL1		/* X86_PG_AVAIL1 aliases with EPT_PG_M */
144 #undef PG_G
145 #undef PG_A
146 #undef PG_M
147 #undef PG_PDE_PAT
148 #undef PG_PDE_CACHE
149 #undef PG_PTE_PAT
150 #undef PG_PTE_CACHE
151 #undef PG_RW
152 #undef PG_V
153 #endif
154 
155 /*
156  * Pte related macros.  This is complicated by having to deal with
157  * the sign extension of the 48th bit.
158  */
159 #define KVADDR(l4, l3, l2, l1) ( \
160 	((unsigned long)-1 << 47) | \
161 	((unsigned long)(l4) << PML4SHIFT) | \
162 	((unsigned long)(l3) << PDPSHIFT) | \
163 	((unsigned long)(l2) << PDRSHIFT) | \
164 	((unsigned long)(l1) << PAGE_SHIFT))
165 
166 #define UVADDR(l4, l3, l2, l1) ( \
167 	((unsigned long)(l4) << PML4SHIFT) | \
168 	((unsigned long)(l3) << PDPSHIFT) | \
169 	((unsigned long)(l2) << PDRSHIFT) | \
170 	((unsigned long)(l1) << PAGE_SHIFT))
171 
172 /*
173  * Number of kernel PML4 slots.  Can be anywhere from 1 to 64 or so,
174  * but setting it larger than NDMPML4E makes no sense.
175  *
176  * Each slot provides .5 TB of kernel virtual space.
177  */
178 #define NKPML4E		4
179 
180 #define	NUPML4E		(NPML4EPG/2)	/* number of userland PML4 pages */
181 #define	NUPDPE		(NUPML4E*NPDPEPG)/* number of userland PDP pages */
182 #define	NUPDE		(NUPDPE*NPDEPG)	/* number of userland PD entries */
183 
184 /*
185  * NDMPML4E is the maximum number of PML4 entries that will be
186  * used to implement the direct map.  It must be a power of two,
187  * and should generally exceed NKPML4E.  The maximum possible
188  * value is 64; using 128 will make the direct map intrude into
189  * the recursive page table map.
190  */
191 #define	NDMPML4E	8
192 
193 /*
194  * These values control the layout of virtual memory.  The starting address
195  * of the direct map, which is controlled by DMPML4I, must be a multiple of
196  * its size.  (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.)
197  *
198  * Note: KPML4I is the index of the (single) level 4 page that maps
199  * the KVA that holds KERNBASE, while KPML4BASE is the index of the
200  * first level 4 page that maps VM_MIN_KERNEL_ADDRESS.  If NKPML4E
201  * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra
202  * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to
203  * KERNBASE.
204  *
205  * (KPML4I combines with KPDPI to choose where KERNBASE starts.
206  * Or, in other words, KPML4I provides bits 39..47 of KERNBASE,
207  * and KPDPI provides bits 30..38.)
208  */
209 #define	PML4PML4I	(NPML4EPG/2)	/* Index of recursive pml4 mapping */
210 
211 #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */
212 #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */
213 
214 #define	KPML4I		(NPML4EPG-1)
215 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
216 
217 /*
218  * XXX doesn't really belong here I guess...
219  */
220 #define ISA_HOLE_START    0xa0000
221 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
222 
223 #define	PMAP_PCID_NONE		0xffffffff
224 #define	PMAP_PCID_KERN		0
225 #define	PMAP_PCID_OVERMAX	0x1000
226 
227 #ifndef LOCORE
228 
229 #include <sys/queue.h>
230 #include <sys/_cpuset.h>
231 #include <sys/_lock.h>
232 #include <sys/_mutex.h>
233 
234 #include <vm/_vm_radix.h>
235 
236 typedef u_int64_t pd_entry_t;
237 typedef u_int64_t pt_entry_t;
238 typedef u_int64_t pdp_entry_t;
239 typedef u_int64_t pml4_entry_t;
240 
241 /*
242  * Address of current address space page table maps and directories.
243  */
244 #ifdef _KERNEL
245 #define	addr_PTmap	(KVADDR(PML4PML4I, 0, 0, 0))
246 #define	addr_PDmap	(KVADDR(PML4PML4I, PML4PML4I, 0, 0))
247 #define	addr_PDPmap	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
248 #define	addr_PML4map	(KVADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
249 #define	addr_PML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
250 #define	PTmap		((pt_entry_t *)(addr_PTmap))
251 #define	PDmap		((pd_entry_t *)(addr_PDmap))
252 #define	PDPmap		((pd_entry_t *)(addr_PDPmap))
253 #define	PML4map		((pd_entry_t *)(addr_PML4map))
254 #define	PML4pml4e	((pd_entry_t *)(addr_PML4pml4e))
255 
256 extern int nkpt;		/* Initial number of kernel page tables */
257 extern u_int64_t KPDPphys;	/* physical address of kernel level 3 */
258 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
259 
260 /*
261  * virtual address to page table entry and
262  * to physical address.
263  * Note: these work recursively, thus vtopte of a pte will give
264  * the corresponding pde that in turn maps it.
265  */
266 pt_entry_t *vtopte(vm_offset_t);
267 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
268 
269 #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte)
270 #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0)
271 #define	pte_store(ptep, pte) do { \
272 	*(u_long *)(ptep) = (u_long)(pte); \
273 } while (0)
274 #define	pte_clear(ptep)			pte_store(ptep, 0)
275 
276 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
277 
278 extern pt_entry_t pg_nx;
279 
280 #endif /* _KERNEL */
281 
282 /*
283  * Pmap stuff
284  */
285 struct	pv_entry;
286 struct	pv_chunk;
287 
288 /*
289  * Locks
290  * (p) PV list lock
291  */
292 struct md_page {
293 	TAILQ_HEAD(, pv_entry)	pv_list;  /* (p) */
294 	int			pv_gen;   /* (p) */
295 	int			pat_mode;
296 };
297 
298 enum pmap_type {
299 	PT_X86,			/* regular x86 page tables */
300 	PT_EPT,			/* Intel's nested page tables */
301 	PT_RVI,			/* AMD's nested page tables */
302 };
303 
304 struct pmap_pcids {
305 	uint32_t	pm_pcid;
306 	uint32_t	pm_gen;
307 };
308 
309 /*
310  * The kernel virtual address (KVA) of the level 4 page table page is always
311  * within the direct map (DMAP) region.
312  */
313 struct pmap {
314 	struct mtx		pm_mtx;
315 	pml4_entry_t		*pm_pml4;	/* KVA of level 4 page table */
316 	uint64_t		pm_cr3;
317 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
318 	cpuset_t		pm_active;	/* active on cpus */
319 	enum pmap_type		pm_type;	/* regular or nested tables */
320 	struct pmap_statistics	pm_stats;	/* pmap statistics */
321 	struct vm_radix		pm_root;	/* spare page table pages */
322 	long			pm_eptgen;	/* EPT pmap generation id */
323 	int			pm_flags;
324 	struct pmap_pcids	pm_pcids[MAXCPU];
325 };
326 
327 /* flags */
328 #define	PMAP_NESTED_IPIMASK	0xff
329 #define	PMAP_PDE_SUPERPAGE	(1 << 8)	/* supports 2MB superpages */
330 #define	PMAP_EMULATE_AD_BITS	(1 << 9)	/* needs A/D bits emulation */
331 #define	PMAP_SUPPORTS_EXEC_ONLY	(1 << 10)	/* execute only mappings ok */
332 
333 typedef struct pmap	*pmap_t;
334 
335 #ifdef _KERNEL
336 extern struct pmap	kernel_pmap_store;
337 #define kernel_pmap	(&kernel_pmap_store)
338 
339 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
340 #define	PMAP_LOCK_ASSERT(pmap, type) \
341 				mtx_assert(&(pmap)->pm_mtx, (type))
342 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
343 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
344 				    NULL, MTX_DEF | MTX_DUPOK)
345 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
346 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
347 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
348 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
349 
350 int	pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags);
351 int	pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype);
352 #endif
353 
354 /*
355  * For each vm_page_t, there is a list of all currently valid virtual
356  * mappings of that page.  An entry is a pv_entry_t, the list is pv_list.
357  */
358 typedef struct pv_entry {
359 	vm_offset_t	pv_va;		/* virtual address for mapping */
360 	TAILQ_ENTRY(pv_entry)	pv_next;
361 } *pv_entry_t;
362 
363 /*
364  * pv_entries are allocated in chunks per-process.  This avoids the
365  * need to track per-pmap assignments.
366  */
367 #define	_NPCM	3
368 #define	_NPCPV	168
369 struct pv_chunk {
370 	pmap_t			pc_pmap;
371 	TAILQ_ENTRY(pv_chunk)	pc_list;
372 	uint64_t		pc_map[_NPCM];	/* bitmap; 1 = free */
373 	TAILQ_ENTRY(pv_chunk)	pc_lru;
374 	struct pv_entry		pc_pventry[_NPCPV];
375 };
376 
377 #ifdef	_KERNEL
378 
379 extern caddr_t	CADDR1;
380 extern pt_entry_t *CMAP1;
381 extern vm_paddr_t phys_avail[];
382 extern vm_paddr_t dump_avail[];
383 extern vm_offset_t virtual_avail;
384 extern vm_offset_t virtual_end;
385 extern vm_paddr_t dmaplimit;
386 extern int pmap_pcid_enabled;
387 extern int invpcid_works;
388 
389 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
390 #define	pmap_page_is_write_mapped(m)	(((m)->aflags & PGA_WRITEABLE) != 0)
391 #define	pmap_unmapbios(va, sz)	pmap_unmapdev((va), (sz))
392 
393 struct thread;
394 
395 void	pmap_activate_sw(struct thread *);
396 void	pmap_bootstrap(vm_paddr_t *);
397 int	pmap_cache_bits(pmap_t pmap, int mode, boolean_t is_pde);
398 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
399 void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, boolean_t invalidate);
400 void	pmap_init_pat(void);
401 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
402 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
403 vm_paddr_t pmap_kextract(vm_offset_t);
404 void	pmap_kremove(vm_offset_t);
405 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
406 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
407 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
408 boolean_t pmap_page_is_mapped(vm_page_t m);
409 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
410 void	pmap_pinit_pml4(vm_page_t);
411 void	pmap_unmapdev(vm_offset_t, vm_size_t);
412 void	pmap_invalidate_page(pmap_t, vm_offset_t);
413 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
414 void	pmap_invalidate_all(pmap_t);
415 void	pmap_invalidate_cache(void);
416 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
417 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva,
418 	    boolean_t force);
419 void	pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num);
420 boolean_t pmap_map_io_transient(vm_page_t *, vm_offset_t *, int, boolean_t);
421 void	pmap_unmap_io_transient(vm_page_t *, vm_offset_t *, int, boolean_t);
422 #endif /* _KERNEL */
423 
424 /* Return various clipped indexes for a given VA */
425 static __inline vm_pindex_t
pmap_pte_index(vm_offset_t va)426 pmap_pte_index(vm_offset_t va)
427 {
428 
429 	return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
430 }
431 
432 static __inline vm_pindex_t
pmap_pde_index(vm_offset_t va)433 pmap_pde_index(vm_offset_t va)
434 {
435 
436 	return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
437 }
438 
439 static __inline vm_pindex_t
pmap_pdpe_index(vm_offset_t va)440 pmap_pdpe_index(vm_offset_t va)
441 {
442 
443 	return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
444 }
445 
446 static __inline vm_pindex_t
pmap_pml4e_index(vm_offset_t va)447 pmap_pml4e_index(vm_offset_t va)
448 {
449 
450 	return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
451 }
452 
453 #endif /* !LOCORE */
454 
455 #endif /* !_MACHINE_PMAP_H_ */
456