xref: /freebsd-14-stable/sys/amd64/include/pmap.h (revision edb7173b8c99fd4177a0e6eca28f2e4496335df3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2003 Peter Wemm.
5  * Copyright (c) 1991 Regents of the University of California.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department and William Jolitz of UUNET Technologies Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * Derived from hp300 version by Mike Hibler, this version by William
37  * Jolitz uses a recursive map [a pde points to the page directory] to
38  * map the page tables using the pagetables themselves. This is done to
39  * reduce the impact on kernel virtual memory for lots of sparse address
40  * space, and to reduce the cost of memory to each process.
41  *
42  *	from: hp300: @(#)pmap.h	7.2 (Berkeley) 12/16/90
43  *	from: @(#)pmap.h	7.4 (Berkeley) 5/12/91
44  */
45 
46 #ifdef __i386__
47 #include <i386/pmap.h>
48 #else /* !__i386__ */
49 
50 #ifndef _MACHINE_PMAP_H_
51 #define	_MACHINE_PMAP_H_
52 
53 #include <machine/pte.h>
54 
55 /*
56  * Define the PG_xx macros in terms of the bits on x86 PTEs.
57  */
58 #define	PG_V		X86_PG_V
59 #define	PG_RW		X86_PG_RW
60 #define	PG_U		X86_PG_U
61 #define	PG_NC_PWT	X86_PG_NC_PWT
62 #define	PG_NC_PCD	X86_PG_NC_PCD
63 #define	PG_A		X86_PG_A
64 #define	PG_M		X86_PG_M
65 #define	PG_PS		X86_PG_PS
66 #define	PG_PTE_PAT	X86_PG_PTE_PAT
67 #define	PG_G		X86_PG_G
68 #define	PG_AVAIL1	X86_PG_AVAIL1
69 #define	PG_AVAIL2	X86_PG_AVAIL2
70 #define	PG_AVAIL3	X86_PG_AVAIL3
71 #define	PG_PDE_PAT	X86_PG_PDE_PAT
72 #define	PG_NX		X86_PG_NX
73 #define	PG_PDE_CACHE	X86_PG_PDE_CACHE
74 #define	PG_PTE_CACHE	X86_PG_PTE_CACHE
75 
76 /* Our various interpretations of the above */
77 #define	PG_W		X86_PG_AVAIL3	/* "Wired" pseudoflag */
78 #define	PG_MANAGED	X86_PG_AVAIL2
79 #define	EPT_PG_EMUL_V	X86_PG_AVAIL(52)
80 #define	EPT_PG_EMUL_RW	X86_PG_AVAIL(53)
81 #define	PG_PROMOTED	X86_PG_AVAIL(54)	/* PDE only */
82 
83 /*
84  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
85  * (PTE) page mappings have identical settings for the following fields:
86  */
87 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_CACHE | \
88 	    PG_M | PG_U | PG_RW | PG_V | PG_PKU_MASK)
89 
90 /*
91  * undef the PG_xx macros that define bits in the regular x86 PTEs that
92  * have a different position in nested PTEs. This is done when compiling
93  * code that needs to be aware of the differences between regular x86 and
94  * nested PTEs.
95  *
96  * The appropriate bitmask will be calculated at runtime based on the pmap
97  * type.
98  */
99 #ifdef AMD64_NPT_AWARE
100 #undef PG_AVAIL1		/* X86_PG_AVAIL1 aliases with EPT_PG_M */
101 #undef PG_G
102 #undef PG_A
103 #undef PG_M
104 #undef PG_PDE_PAT
105 #undef PG_PDE_CACHE
106 #undef PG_PTE_PAT
107 #undef PG_PTE_CACHE
108 #undef PG_RW
109 #undef PG_V
110 #endif
111 
112 /*
113  * Pte related macros.  This is complicated by having to deal with
114  * the sign extension of the 48th bit.
115  */
116 #define KV4ADDR(l4, l3, l2, l1) ( \
117 	((unsigned long)-1 << 47) | \
118 	((unsigned long)(l4) << PML4SHIFT) | \
119 	((unsigned long)(l3) << PDPSHIFT) | \
120 	((unsigned long)(l2) << PDRSHIFT) | \
121 	((unsigned long)(l1) << PAGE_SHIFT))
122 #define KV5ADDR(l5, l4, l3, l2, l1) (		\
123 	((unsigned long)-1 << 56) | \
124 	((unsigned long)(l5) << PML5SHIFT) | \
125 	((unsigned long)(l4) << PML4SHIFT) | \
126 	((unsigned long)(l3) << PDPSHIFT) | \
127 	((unsigned long)(l2) << PDRSHIFT) | \
128 	((unsigned long)(l1) << PAGE_SHIFT))
129 
130 #define UVADDR(l5, l4, l3, l2, l1) (	     \
131 	((unsigned long)(l5) << PML5SHIFT) | \
132 	((unsigned long)(l4) << PML4SHIFT) | \
133 	((unsigned long)(l3) << PDPSHIFT) | \
134 	((unsigned long)(l2) << PDRSHIFT) | \
135 	((unsigned long)(l1) << PAGE_SHIFT))
136 
137 /*
138  * Number of kernel PML4 slots.  Can be anywhere from 1 to 64 or so,
139  * but setting it larger than NDMPML4E makes no sense.
140  *
141  * Each slot provides .5 TB of kernel virtual space.
142  */
143 #define NKPML4E		4
144 
145 /*
146  * Number of PML4 slots for the KASAN shadow map.  It requires 1 byte of memory
147  * for every 8 bytes of the kernel address space.
148  */
149 #define	NKASANPML4E	((NKPML4E + 7) / 8)
150 
151 /*
152  * Number of PML4 slots for the KMSAN shadow and origin maps.  These are
153  * one-to-one with the kernel map.
154  */
155 #define	NKMSANSHADPML4E	NKPML4E
156 #define	NKMSANORIGPML4E	NKPML4E
157 
158 /*
159  * We use the same numbering of the page table pages for 5-level and
160  * 4-level paging structures.
161  */
162 #define	NUPML5E		(NPML5EPG / 2)		/* number of userland PML5
163 						   pages */
164 #define	NUPML4E		(NUPML5E * NPML4EPG)	/* number of userland PML4
165 						   pages */
166 #define	NUPDPE		(NUPML4E * NPDPEPG)	/* number of userland PDP
167 						   pages */
168 #define	NUPDE		(NUPDPE * NPDEPG)	/* number of userland PD
169 						   entries */
170 #define	NUP4ML4E	(NPML4EPG / 2)
171 
172 /*
173  * NDMPML4E is the maximum number of PML4 entries that will be
174  * used to implement the direct map.  It must be a power of two,
175  * and should generally exceed NKPML4E.  The maximum possible
176  * value is 64; using 128 will make the direct map intrude into
177  * the recursive page table map.
178  */
179 #define	NDMPML4E	8
180 
181 /*
182  * These values control the layout of virtual memory.  The starting address
183  * of the direct map, which is controlled by DMPML4I, must be a multiple of
184  * its size.  (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.)
185  *
186  * Note: KPML4I is the index of the (single) level 4 page that maps
187  * the KVA that holds KERNBASE, while KPML4BASE is the index of the
188  * first level 4 page that maps VM_MIN_KERNEL_ADDRESS.  If NKPML4E
189  * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra
190  * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to
191  * KERNBASE.
192  *
193  * (KPML4I combines with KPDPI to choose where KERNBASE starts.
194  * Or, in other words, KPML4I provides bits 39..47 of KERNBASE,
195  * and KPDPI provides bits 30..38.)
196  */
197 #define	PML4PML4I	(NPML4EPG / 2)	/* Index of recursive pml4 mapping */
198 #define	PML5PML5I	(NPML5EPG / 2)	/* Index of recursive pml5 mapping */
199 
200 #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */
201 #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */
202 
203 #define	KPML4I		(NPML4EPG-1)
204 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
205 
206 #define	KASANPML4I	(DMPML4I - NKASANPML4E) /* Below the direct map */
207 
208 #define	KMSANSHADPML4I	(KPML4BASE - NKMSANSHADPML4E)
209 #define	KMSANORIGPML4I	(DMPML4I - NKMSANORIGPML4E)
210 
211 /* Large map: index of the first and max last pml4 entry */
212 #define	LMSPML4I	(PML4PML4I + 1)
213 #define	LMEPML4I	(KASANPML4I - 1)
214 
215 /*
216  * XXX doesn't really belong here I guess...
217  */
218 #define ISA_HOLE_START    0xa0000
219 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
220 
221 #define	PMAP_PCID_NONE		0xffffffff
222 #define	PMAP_PCID_KERN		0
223 #define	PMAP_PCID_OVERMAX	0x1000
224 #define	PMAP_PCID_OVERMAX_KERN	0x800
225 #define	PMAP_PCID_USER_PT	0x800
226 
227 #define	PMAP_NO_CR3		0xffffffffffffffff
228 #define	PMAP_UCR3_NOMASK	0xffffffffffffffff
229 
230 #ifndef LOCORE
231 
232 #include <sys/kassert.h>
233 #include <sys/queue.h>
234 #include <sys/_cpuset.h>
235 #include <sys/_lock.h>
236 #include <sys/_mutex.h>
237 #include <sys/_pctrie.h>
238 #include <machine/_pmap.h>
239 #include <sys/_pv_entry.h>
240 #include <sys/_rangeset.h>
241 #include <sys/_smr.h>
242 
243 #include <vm/_vm_radix.h>
244 
245 typedef u_int64_t pd_entry_t;
246 typedef u_int64_t pt_entry_t;
247 typedef u_int64_t pdp_entry_t;
248 typedef u_int64_t pml4_entry_t;
249 typedef u_int64_t pml5_entry_t;
250 
251 /*
252  * Address of current address space page table maps and directories.
253  */
254 #ifdef _KERNEL
255 #define	addr_P4Tmap	(KV4ADDR(PML4PML4I, 0, 0, 0))
256 #define	addr_P4Dmap	(KV4ADDR(PML4PML4I, PML4PML4I, 0, 0))
257 #define	addr_P4DPmap	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
258 #define	addr_P4ML4map	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
259 #define	addr_P4ML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
260 #define	P4Tmap		((pt_entry_t *)(addr_P4Tmap))
261 #define	P4Dmap		((pd_entry_t *)(addr_P4Dmap))
262 
263 #define	addr_P5Tmap	(KV5ADDR(PML5PML5I, 0, 0, 0, 0))
264 #define	addr_P5Dmap	(KV5ADDR(PML5PML5I, PML5PML5I, 0, 0, 0))
265 #define	addr_P5DPmap	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, 0, 0))
266 #define	addr_P5ML4map	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, 0))
267 #define	addr_P5ML5map	\
268     (KVADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I))
269 #define	addr_P5ML5pml5e	(addr_P5ML5map + (PML5PML5I * sizeof(pml5_entry_t)))
270 #define	P5Tmap		((pt_entry_t *)(addr_P5Tmap))
271 #define	P5Dmap		((pd_entry_t *)(addr_P5Dmap))
272 
273 extern int nkpt;		/* Initial number of kernel page tables */
274 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
275 extern u_int64_t KPML5phys;	/* physical address of kernel level 5 */
276 
277 /*
278  * virtual address to page table entry and
279  * to physical address.
280  * Note: these work recursively, thus vtopte of a pte will give
281  * the corresponding pde that in turn maps it.
282  */
283 pt_entry_t *vtopte(vm_offset_t);
284 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
285 
286 #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte)
287 #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0)
288 #define	pte_store(ptep, pte) do { \
289 	*(u_long *)(ptep) = (u_long)(pte); \
290 } while (0)
291 #define	pte_clear(ptep)			pte_store(ptep, 0)
292 
293 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
294 
295 extern pt_entry_t pg_nx;
296 
297 #endif /* _KERNEL */
298 
299 /*
300  * Pmap stuff
301  */
302 
303 /*
304  * Locks
305  * (p) PV list lock
306  */
307 struct md_page {
308 	TAILQ_HEAD(, pv_entry)	pv_list;  /* (p) */
309 	int			pv_gen;   /* (p) */
310 	int			pat_mode;
311 };
312 
313 enum pmap_type {
314 	PT_X86,			/* regular x86 page tables */
315 	PT_EPT,			/* Intel's nested page tables */
316 	PT_RVI,			/* AMD's nested page tables */
317 };
318 
319 /*
320  * The kernel virtual address (KVA) of the level 4 page table page is always
321  * within the direct map (DMAP) region.
322  */
323 struct pmap {
324 	struct mtx		pm_mtx;
325 	pml4_entry_t		*pm_pmltop;	/* KVA of top level page table */
326 	pml4_entry_t		*pm_pmltopu;	/* KVA of user top page table */
327 	uint64_t		pm_cr3;
328 	uint64_t		pm_ucr3;
329 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
330 	cpuset_t		pm_active;	/* active on cpus */
331 	enum pmap_type		pm_type;	/* regular or nested tables */
332 	struct pmap_statistics	pm_stats;	/* pmap statistics */
333 	struct vm_radix		pm_root;	/* spare page table pages */
334 	long			pm_eptgen;	/* EPT pmap generation id */
335 	smr_t			pm_eptsmr;
336 	int			pm_flags;
337 	struct pmap_pcid	*pm_pcidp;
338 	struct rangeset		pm_pkru;
339 };
340 
341 /* flags */
342 #define	PMAP_NESTED_IPIMASK	0xff
343 #define	PMAP_PDE_SUPERPAGE	(1 << 8)	/* supports 2MB superpages */
344 #define	PMAP_EMULATE_AD_BITS	(1 << 9)	/* needs A/D bits emulation */
345 #define	PMAP_SUPPORTS_EXEC_ONLY	(1 << 10)	/* execute only mappings ok */
346 
347 typedef struct pmap	*pmap_t;
348 
349 #ifdef _KERNEL
350 extern struct pmap	kernel_pmap_store;
351 #define kernel_pmap	(&kernel_pmap_store)
352 
353 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
354 #define	PMAP_LOCK_ASSERT(pmap, type) \
355 				mtx_assert(&(pmap)->pm_mtx, (type))
356 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
357 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
358 				    NULL, MTX_DEF | MTX_DUPOK)
359 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
360 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
361 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
362 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
363 
364 int	pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags);
365 int	pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype);
366 
367 extern caddr_t	CADDR1;
368 extern pt_entry_t *CMAP1;
369 extern vm_offset_t virtual_avail;
370 extern vm_offset_t virtual_end;
371 extern vm_paddr_t dmaplimit;
372 extern int pmap_pcid_enabled;
373 extern int invpcid_works;
374 extern int invlpgb_works;
375 extern int invlpgb_maxcnt;
376 extern int pmap_pcid_invlpg_workaround;
377 extern int pmap_pcid_invlpg_workaround_uena;
378 
379 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
380 #define	pmap_page_is_write_mapped(m)	(((m)->a.flags & PGA_WRITEABLE) != 0)
381 #define	pmap_unmapbios(va, sz)		pmap_unmapdev((va), (sz))
382 
383 #define	pmap_vm_page_alloc_check(m)					\
384 	KASSERT(m->phys_addr < kernphys ||				\
385 	    m->phys_addr >= kernphys + (vm_offset_t)&_end - KERNSTART,	\
386 	    ("allocating kernel page %p pa %#lx kernphys %#lx end %p", \
387 	    m, m->phys_addr, kernphys, &_end));
388 
389 struct thread;
390 
391 void	pmap_activate_boot(pmap_t pmap);
392 void	pmap_activate_sw(struct thread *);
393 void	pmap_allow_2m_x_ept_recalculate(void);
394 void	pmap_bootstrap(vm_paddr_t *);
395 int	pmap_cache_bits(pmap_t pmap, int mode, boolean_t is_pde);
396 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
397 int	pmap_change_prot(vm_offset_t, vm_size_t, vm_prot_t);
398 void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, boolean_t invalidate);
399 void	pmap_flush_cache_range(vm_offset_t, vm_offset_t);
400 void	pmap_flush_cache_phys_range(vm_paddr_t, vm_paddr_t, vm_memattr_t);
401 void	pmap_init_pat(void);
402 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
403 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
404 vm_paddr_t pmap_kextract(vm_offset_t);
405 void	pmap_kremove(vm_offset_t);
406 int	pmap_large_map(vm_paddr_t, vm_size_t, void **, vm_memattr_t);
407 void	pmap_large_map_wb(void *sva, vm_size_t len);
408 void	pmap_large_unmap(void *sva, vm_size_t len);
409 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
410 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
411 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
412 void	*pmap_mapdev_pciecfg(vm_paddr_t pa, vm_size_t size);
413 bool	pmap_not_in_di(void);
414 boolean_t pmap_page_is_mapped(vm_page_t m);
415 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
416 void	pmap_page_set_memattr_noflush(vm_page_t m, vm_memattr_t ma);
417 void	pmap_pinit_pml4(vm_page_t);
418 void	pmap_pinit_pml5(vm_page_t);
419 bool	pmap_ps_enabled(pmap_t pmap);
420 void	pmap_unmapdev(void *, vm_size_t);
421 void	pmap_invalidate_page(pmap_t, vm_offset_t);
422 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
423 void	pmap_invalidate_all(pmap_t);
424 void	pmap_invalidate_cache(void);
425 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
426 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
427 void	pmap_force_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
428 void	pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num);
429 bool	pmap_map_io_transient(vm_page_t *, vm_offset_t *, int, bool);
430 void	pmap_unmap_io_transient(vm_page_t *, vm_offset_t *, int, bool);
431 void	pmap_map_delete(pmap_t, vm_offset_t, vm_offset_t);
432 void	pmap_pti_add_kva(vm_offset_t sva, vm_offset_t eva, bool exec);
433 void	pmap_pti_remove_kva(vm_offset_t sva, vm_offset_t eva);
434 void	pmap_pti_pcid_invalidate(uint64_t ucr3, uint64_t kcr3);
435 void	pmap_pti_pcid_invlpg(uint64_t ucr3, uint64_t kcr3, vm_offset_t va);
436 void	pmap_pti_pcid_invlrng(uint64_t ucr3, uint64_t kcr3, vm_offset_t sva,
437 	    vm_offset_t eva);
438 int	pmap_pkru_clear(pmap_t pmap, vm_offset_t sva, vm_offset_t eva);
439 int	pmap_pkru_set(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
440 	    u_int keyidx, int flags);
441 void	pmap_thread_init_invl_gen(struct thread *td);
442 int	pmap_vmspace_copy(pmap_t dst_pmap, pmap_t src_pmap);
443 void	pmap_page_array_startup(long count);
444 vm_page_t pmap_page_alloc_below_4g(bool zeroed);
445 
446 #if defined(KASAN) || defined(KMSAN)
447 void	pmap_san_enter(vm_offset_t);
448 #endif
449 
450 /*
451  * Returns a pointer to a set of CPUs on which the pmap is currently active.
452  * Note that the set can be modified without any mutual exclusion, so a copy
453  * must be made if a stable value is required.
454  */
455 static __inline volatile cpuset_t *
pmap_invalidate_cpu_mask(pmap_t pmap)456 pmap_invalidate_cpu_mask(pmap_t pmap)
457 {
458 	return (&pmap->pm_active);
459 }
460 
461 #if defined(_SYS_PCPU_H_) && defined(_MACHINE_CPUFUNC_H_)
462 /*
463  * It seems that AlderLake+ small cores have some microarchitectural
464  * bug, which results in the INVLPG instruction failing to flush all
465  * global TLB entries when PCID is enabled.  Work around it for now,
466  * by doing global invalidation on small cores instead of INVLPG.
467  */
468 static __inline void
pmap_invlpg(pmap_t pmap,vm_offset_t va)469 pmap_invlpg(pmap_t pmap, vm_offset_t va)
470 {
471 	if (pmap == kernel_pmap && PCPU_GET(pcid_invlpg_workaround)) {
472 		struct invpcid_descr d = { 0 };
473 
474 		invpcid(&d, INVPCID_CTXGLOB);
475 	} else {
476 		invlpg(va);
477 	}
478 }
479 #endif /* sys/pcpu.h && machine/cpufunc.h */
480 
481 #if defined(_SYS_PCPU_H_)
482 /* Return pcid for the pmap pmap on current cpu */
483 static __inline uint32_t
pmap_get_pcid(pmap_t pmap)484 pmap_get_pcid(pmap_t pmap)
485 {
486 	struct pmap_pcid *pcidp;
487 
488 	MPASS(pmap_pcid_enabled);
489 	pcidp = zpcpu_get(pmap->pm_pcidp);
490 	return (pcidp->pm_pcid);
491 }
492 #endif /* sys/pcpu.h */
493 
494 /*
495  * Invalidation request.  PCPU pc_smp_tlb_op uses u_int instead of the
496  * enum to avoid both namespace and ABI issues (with enums).
497  */
498 enum invl_op_codes {
499 	INVL_OP_TLB               = 1,
500 	INVL_OP_TLB_INVPCID       = 2,
501 	INVL_OP_TLB_INVPCID_PTI   = 3,
502 	INVL_OP_TLB_PCID          = 4,
503 	INVL_OP_PGRNG             = 5,
504 	INVL_OP_PGRNG_INVPCID     = 6,
505 	INVL_OP_PGRNG_PCID        = 7,
506 	INVL_OP_PG                = 8,
507 	INVL_OP_PG_INVPCID        = 9,
508 	INVL_OP_PG_PCID           = 10,
509 	INVL_OP_CACHE             = 11,
510 };
511 
512 typedef void (*smp_invl_local_cb_t)(struct pmap *, vm_offset_t addr1,
513     vm_offset_t addr2);
514 typedef void (*smp_targeted_tlb_shootdown_t)(pmap_t, vm_offset_t, vm_offset_t,
515     smp_invl_local_cb_t, enum invl_op_codes);
516 
517 void smp_targeted_tlb_shootdown_native(pmap_t, vm_offset_t, vm_offset_t,
518     smp_invl_local_cb_t, enum invl_op_codes);
519 extern smp_targeted_tlb_shootdown_t smp_targeted_tlb_shootdown;
520 
521 #endif /* _KERNEL */
522 
523 /* Return various clipped indexes for a given VA */
524 static __inline vm_pindex_t
pmap_pte_index(vm_offset_t va)525 pmap_pte_index(vm_offset_t va)
526 {
527 
528 	return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
529 }
530 
531 static __inline vm_pindex_t
pmap_pde_index(vm_offset_t va)532 pmap_pde_index(vm_offset_t va)
533 {
534 
535 	return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
536 }
537 
538 static __inline vm_pindex_t
pmap_pdpe_index(vm_offset_t va)539 pmap_pdpe_index(vm_offset_t va)
540 {
541 
542 	return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
543 }
544 
545 static __inline vm_pindex_t
pmap_pml4e_index(vm_offset_t va)546 pmap_pml4e_index(vm_offset_t va)
547 {
548 
549 	return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
550 }
551 
552 static __inline vm_pindex_t
pmap_pml5e_index(vm_offset_t va)553 pmap_pml5e_index(vm_offset_t va)
554 {
555 
556 	return ((va >> PML5SHIFT) & ((1ul << NPML5EPGSHIFT) - 1));
557 }
558 
559 #endif /* !LOCORE */
560 
561 #endif /* !_MACHINE_PMAP_H_ */
562 
563 #endif /* __i386__ */
564