xref: /freebsd/sys/amd64/include/pmap.h (revision 28fdf718b30b43d1f9ecaa1d2de021d7b53f8a67)
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 
43 #ifdef __i386__
44 #include <i386/pmap.h>
45 #else /* !__i386__ */
46 
47 #ifndef _MACHINE_PMAP_H_
48 #define	_MACHINE_PMAP_H_
49 
50 #include <machine/pte.h>
51 
52 /*
53  * Define the PG_xx macros in terms of the bits on x86 PTEs.
54  */
55 #define	PG_V		X86_PG_V
56 #define	PG_RW		X86_PG_RW
57 #define	PG_U		X86_PG_U
58 #define	PG_NC_PWT	X86_PG_NC_PWT
59 #define	PG_NC_PCD	X86_PG_NC_PCD
60 #define	PG_A		X86_PG_A
61 #define	PG_M		X86_PG_M
62 #define	PG_PS		X86_PG_PS
63 #define	PG_PTE_PAT	X86_PG_PTE_PAT
64 #define	PG_G		X86_PG_G
65 #define	PG_AVAIL1	X86_PG_AVAIL1
66 #define	PG_AVAIL2	X86_PG_AVAIL2
67 #define	PG_AVAIL3	X86_PG_AVAIL3
68 #define	PG_PDE_PAT	X86_PG_PDE_PAT
69 #define	PG_NX		X86_PG_NX
70 #define	PG_PDE_CACHE	X86_PG_PDE_CACHE
71 #define	PG_PTE_CACHE	X86_PG_PTE_CACHE
72 
73 /* Our various interpretations of the above */
74 #define	PG_W		X86_PG_AVAIL3	/* "Wired" pseudoflag */
75 #define	PG_MANAGED	X86_PG_AVAIL2
76 #define	EPT_PG_EMUL_V	X86_PG_AVAIL(52)
77 #define	EPT_PG_EMUL_RW	X86_PG_AVAIL(53)
78 #define	PG_PROMOTED	X86_PG_AVAIL(54)	/* PDE only */
79 
80 /*
81  * Promotion to a 2MB (PDE) page mapping requires that the corresponding 4KB
82  * (PTE) page mappings have identical settings for the following fields:
83  */
84 #define	PG_PTE_PROMOTE	(PG_NX | PG_MANAGED | PG_W | PG_G | PG_PTE_CACHE | \
85 	    PG_M | PG_U | PG_RW | PG_V | PG_PKU_MASK)
86 
87 /*
88  * undef the PG_xx macros that define bits in the regular x86 PTEs that
89  * have a different position in nested PTEs. This is done when compiling
90  * code that needs to be aware of the differences between regular x86 and
91  * nested PTEs.
92  *
93  * The appropriate bitmask will be calculated at runtime based on the pmap
94  * type.
95  */
96 #ifdef AMD64_NPT_AWARE
97 #undef PG_AVAIL1		/* X86_PG_AVAIL1 aliases with EPT_PG_M */
98 #undef PG_G
99 #undef PG_A
100 #undef PG_M
101 #undef PG_PDE_PAT
102 #undef PG_PDE_CACHE
103 #undef PG_PTE_PAT
104 #undef PG_PTE_CACHE
105 #undef PG_RW
106 #undef PG_V
107 #endif
108 
109 /*
110  * Pte related macros.  This is complicated by having to deal with
111  * the sign extension of the 48th bit.
112  */
113 #define KV4ADDR(l4, l3, l2, l1) ( \
114 	((unsigned long)-1 << 47) | \
115 	((unsigned long)(l4) << PML4SHIFT) | \
116 	((unsigned long)(l3) << PDPSHIFT) | \
117 	((unsigned long)(l2) << PDRSHIFT) | \
118 	((unsigned long)(l1) << PAGE_SHIFT))
119 #define KV5ADDR(l5, l4, l3, l2, l1) (		\
120 	((unsigned long)-1 << 56) | \
121 	((unsigned long)(l5) << PML5SHIFT) | \
122 	((unsigned long)(l4) << PML4SHIFT) | \
123 	((unsigned long)(l3) << PDPSHIFT) | \
124 	((unsigned long)(l2) << PDRSHIFT) | \
125 	((unsigned long)(l1) << PAGE_SHIFT))
126 
127 #define UVADDR(l5, l4, l3, l2, l1) (	     \
128 	((unsigned long)(l5) << PML5SHIFT) | \
129 	((unsigned long)(l4) << PML4SHIFT) | \
130 	((unsigned long)(l3) << PDPSHIFT) | \
131 	((unsigned long)(l2) << PDRSHIFT) | \
132 	((unsigned long)(l1) << PAGE_SHIFT))
133 
134 /*
135  * Number of kernel PML4 slots.  Can be anywhere from 1 to 64 or so,
136  * but setting it larger than NDMPML4E makes no sense.
137  *
138  * Each slot provides .5 TB of kernel virtual space.
139  */
140 #define NKPML4E		4
141 
142 /*
143  * Number of PML4 slots for the KASAN shadow map.  It requires 1 byte of memory
144  * for every 8 bytes of the kernel address space.
145  */
146 #define	NKASANPML4E	((NKPML4E + 7) / 8)
147 
148 /*
149  * Number of PML4 slots for the KMSAN shadow and origin maps.  These are
150  * one-to-one with the kernel map.
151  */
152 #define	NKMSANSHADPML4E	NKPML4E
153 #define	NKMSANORIGPML4E	NKPML4E
154 
155 /*
156  * We use the same numbering of the page table pages for 5-level and
157  * 4-level paging structures.
158  */
159 #define	NUPML5E		(NPML5EPG / 2)		/* number of userland PML5
160 						   pages */
161 #define	NUPML4E		(NUPML5E * NPML4EPG)	/* number of userland PML4
162 						   pages */
163 #define	NUPDPE		(NUPML4E * NPDPEPG)	/* number of userland PDP
164 						   pages */
165 #define	NUPDE		(NUPDPE * NPDEPG)	/* number of userland PD
166 						   entries */
167 #define	NUP4ML4E	(NPML4EPG / 2)
168 
169 /*
170  * NDMPML4E is the maximum number of PML4 entries that will be
171  * used to implement the direct map.  It must be a power of two,
172  * and should generally exceed NKPML4E.  The maximum possible
173  * value is 64; using 128 will make the direct map intrude into
174  * the recursive page table map.
175  */
176 #define	NDMPML4E	8
177 
178 /*
179  * These values control the layout of virtual memory.  The starting address
180  * of the direct map, which is controlled by DMPML4I, must be a multiple of
181  * its size.  (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.)
182  *
183  * Note: KPML4I is the index of the (single) level 4 page that maps
184  * the KVA that holds KERNBASE, while KPML4BASE is the index of the
185  * first level 4 page that maps VM_MIN_KERNEL_ADDRESS.  If NKPML4E
186  * is 1, these are the same, otherwise KPML4BASE < KPML4I and extra
187  * level 4 PDEs are needed to map from VM_MIN_KERNEL_ADDRESS up to
188  * KERNBASE.
189  *
190  * (KPML4I combines with KPDPI to choose where KERNBASE starts.
191  * Or, in other words, KPML4I provides bits 39..47 of KERNBASE,
192  * and KPDPI provides bits 30..38.)
193  */
194 #define	PML4PML4I	(NPML4EPG / 2)	/* Index of recursive pml4 mapping */
195 #define	PML5PML5I	(NPML5EPG / 2)	/* Index of recursive pml5 mapping */
196 
197 #define	KPML4BASE	(NPML4EPG-NKPML4E) /* KVM at highest addresses */
198 #define	DMPML4I		rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */
199 
200 #define	KPML4I		(NPML4EPG-1)
201 #define	KPDPI		(NPDPEPG-2)	/* kernbase at -2GB */
202 
203 #define	KASANPML4I	(DMPML4I - NKASANPML4E) /* Below the direct map */
204 
205 #define	KMSANSHADPML4I	(KPML4BASE - NKMSANSHADPML4E)
206 #define	KMSANORIGPML4I	(DMPML4I - NKMSANORIGPML4E)
207 
208 /* Large map: index of the first and max last pml4 entry */
209 #define	LMSPML4I	(PML4PML4I + 1)
210 #define	LMEPML4I	(KASANPML4I - 1)
211 
212 /*
213  * XXX doesn't really belong here I guess...
214  */
215 #define ISA_HOLE_START    0xa0000
216 #define ISA_HOLE_LENGTH (0x100000-ISA_HOLE_START)
217 
218 #define	PMAP_PCID_NONE		0xffffffff
219 #define	PMAP_PCID_KERN		0
220 #define	PMAP_PCID_OVERMAX	0x1000
221 #define	PMAP_PCID_OVERMAX_KERN	0x800
222 #define	PMAP_PCID_USER_PT	0x800
223 
224 #define	PMAP_NO_CR3		0xffffffffffffffff
225 #define	PMAP_UCR3_NOMASK	0xffffffffffffffff
226 
227 #ifndef LOCORE
228 
229 #include <sys/kassert.h>
230 #include <sys/queue.h>
231 #include <sys/_cpuset.h>
232 #include <sys/_lock.h>
233 #include <sys/_mutex.h>
234 #include <sys/_pctrie.h>
235 #include <machine/_pmap.h>
236 #include <sys/_pv_entry.h>
237 #include <sys/_rangeset.h>
238 #include <sys/_smr.h>
239 
240 #include <vm/_vm_radix.h>
241 
242 typedef u_int64_t pd_entry_t;
243 typedef u_int64_t pt_entry_t;
244 typedef u_int64_t pdp_entry_t;
245 typedef u_int64_t pml4_entry_t;
246 typedef u_int64_t pml5_entry_t;
247 
248 /*
249  * Address of current address space page table maps and directories.
250  */
251 #ifdef _KERNEL
252 #define	addr_P4Tmap	(KV4ADDR(PML4PML4I, 0, 0, 0))
253 #define	addr_P4Dmap	(KV4ADDR(PML4PML4I, PML4PML4I, 0, 0))
254 #define	addr_P4DPmap	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, 0))
255 #define	addr_P4ML4map	(KV4ADDR(PML4PML4I, PML4PML4I, PML4PML4I, PML4PML4I))
256 #define	addr_P4ML4pml4e	(addr_PML4map + (PML4PML4I * sizeof(pml4_entry_t)))
257 #define	P4Tmap		((pt_entry_t *)(addr_P4Tmap))
258 #define	P4Dmap		((pd_entry_t *)(addr_P4Dmap))
259 
260 #define	addr_P5Tmap	(KV5ADDR(PML5PML5I, 0, 0, 0, 0))
261 #define	addr_P5Dmap	(KV5ADDR(PML5PML5I, PML5PML5I, 0, 0, 0))
262 #define	addr_P5DPmap	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, 0, 0))
263 #define	addr_P5ML4map	(KV5ADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, 0))
264 #define	addr_P5ML5map	\
265     (KVADDR(PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I, PML5PML5I))
266 #define	addr_P5ML5pml5e	(addr_P5ML5map + (PML5PML5I * sizeof(pml5_entry_t)))
267 #define	P5Tmap		((pt_entry_t *)(addr_P5Tmap))
268 #define	P5Dmap		((pd_entry_t *)(addr_P5Dmap))
269 
270 extern int nkpt;		/* Initial number of kernel page tables */
271 extern u_int64_t KPML4phys;	/* physical address of kernel level 4 */
272 extern u_int64_t KPML5phys;	/* physical address of kernel level 5 */
273 
274 /*
275  * virtual address to page table entry and
276  * to physical address.
277  * Note: these work recursively, thus vtopte of a pte will give
278  * the corresponding pde that in turn maps it.
279  */
280 pt_entry_t *vtopte(vm_offset_t);
281 #define	vtophys(va)	pmap_kextract(((vm_offset_t) (va)))
282 
283 #define	pte_load_store(ptep, pte)	atomic_swap_long(ptep, pte)
284 #define	pte_load_clear(ptep)		atomic_swap_long(ptep, 0)
285 #define	pte_store(ptep, pte) do { \
286 	*(u_long *)(ptep) = (u_long)(pte); \
287 } while (0)
288 #define	pte_clear(ptep)			pte_store(ptep, 0)
289 
290 #define	pde_store(pdep, pde)		pte_store(pdep, pde)
291 
292 extern pt_entry_t pg_nx;
293 
294 #endif /* _KERNEL */
295 
296 /*
297  * Pmap stuff
298  */
299 
300 /*
301  * Locks
302  * (p) PV list lock
303  */
304 struct md_page {
305 	TAILQ_HEAD(, pv_entry)	pv_list;  /* (p) */
306 	int			pv_gen;   /* (p) */
307 	int			pat_mode;
308 };
309 
310 enum pmap_type {
311 	PT_X86,			/* regular x86 page tables */
312 	PT_EPT,			/* Intel's nested page tables */
313 	PT_RVI,			/* AMD's nested page tables */
314 };
315 
316 /*
317  * The kernel virtual address (KVA) of the level 4 page table page is always
318  * within the direct map (DMAP) region.
319  */
320 struct pmap {
321 	struct mtx		pm_mtx;
322 	pml4_entry_t		*pm_pmltop;	/* KVA of top level page table */
323 	pml4_entry_t		*pm_pmltopu;	/* KVA of user top page table */
324 	uint64_t		pm_cr3;
325 	uint64_t		pm_ucr3;
326 	TAILQ_HEAD(,pv_chunk)	pm_pvchunk;	/* list of mappings in pmap */
327 	cpuset_t		pm_active;	/* active on cpus */
328 	enum pmap_type		pm_type;	/* regular or nested tables */
329 	struct pmap_statistics	pm_stats;	/* pmap statistics */
330 	struct vm_radix		pm_root;	/* spare page table pages */
331 	long			pm_eptgen;	/* EPT pmap generation id */
332 	smr_t			pm_eptsmr;
333 	int			pm_flags;
334 	struct pmap_pcid	*pm_pcidp;
335 	struct rangeset		pm_pkru;
336 };
337 
338 /* flags */
339 #define	PMAP_NESTED_IPIMASK	0xff
340 #define	PMAP_PDE_SUPERPAGE	(1 << 8)	/* supports 2MB superpages */
341 #define	PMAP_EMULATE_AD_BITS	(1 << 9)	/* needs A/D bits emulation */
342 #define	PMAP_SUPPORTS_EXEC_ONLY	(1 << 10)	/* execute only mappings ok */
343 
344 typedef struct pmap	*pmap_t;
345 
346 #ifdef _KERNEL
347 extern struct pmap	kernel_pmap_store;
348 #define kernel_pmap	(&kernel_pmap_store)
349 
350 #define	PMAP_LOCK(pmap)		mtx_lock(&(pmap)->pm_mtx)
351 #define	PMAP_LOCK_ASSERT(pmap, type) \
352 				mtx_assert(&(pmap)->pm_mtx, (type))
353 #define	PMAP_LOCK_DESTROY(pmap)	mtx_destroy(&(pmap)->pm_mtx)
354 #define	PMAP_LOCK_INIT(pmap)	mtx_init(&(pmap)->pm_mtx, "pmap", \
355 				    NULL, MTX_DEF | MTX_DUPOK)
356 #define	PMAP_LOCKED(pmap)	mtx_owned(&(pmap)->pm_mtx)
357 #define	PMAP_MTX(pmap)		(&(pmap)->pm_mtx)
358 #define	PMAP_TRYLOCK(pmap)	mtx_trylock(&(pmap)->pm_mtx)
359 #define	PMAP_UNLOCK(pmap)	mtx_unlock(&(pmap)->pm_mtx)
360 
361 int	pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags);
362 int	pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype);
363 
364 extern caddr_t	CADDR1;
365 extern pt_entry_t *CMAP1;
366 extern vm_offset_t virtual_avail;
367 extern vm_offset_t virtual_end;
368 extern vm_paddr_t dmaplimit;
369 extern int pmap_pcid_enabled;
370 extern int invpcid_works;
371 extern int invlpgb_works;
372 extern int invlpgb_maxcnt;
373 extern int pmap_pcid_invlpg_workaround;
374 extern int pmap_pcid_invlpg_workaround_uena;
375 
376 #define	pmap_page_get_memattr(m)	((vm_memattr_t)(m)->md.pat_mode)
377 #define	pmap_page_is_write_mapped(m)	(((m)->a.flags & PGA_WRITEABLE) != 0)
378 #define	pmap_unmapbios(va, sz)		pmap_unmapdev((va), (sz))
379 
380 #define	pmap_vm_page_alloc_check(m)					\
381 	KASSERT(m->phys_addr < kernphys ||				\
382 	    m->phys_addr >= kernphys + (vm_offset_t)&_end - KERNSTART,	\
383 	    ("allocating kernel page %p pa %#lx kernphys %#lx end %p", \
384 	    m, m->phys_addr, kernphys, &_end));
385 
386 struct thread;
387 
388 void	pmap_activate_boot(pmap_t pmap);
389 void	pmap_activate_sw(struct thread *);
390 void	pmap_allow_2m_x_ept_recalculate(void);
391 void	pmap_bootstrap(vm_paddr_t *);
392 int	pmap_cache_bits(pmap_t pmap, int mode, bool is_pde);
393 int	pmap_change_attr(vm_offset_t, vm_size_t, int);
394 int	pmap_change_prot(vm_offset_t, vm_size_t, vm_prot_t);
395 void	pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, bool invalidate);
396 void	pmap_flush_cache_range(vm_offset_t, vm_offset_t);
397 void	pmap_flush_cache_phys_range(vm_paddr_t, vm_paddr_t, vm_memattr_t);
398 void	pmap_init_pat(void);
399 void	pmap_kenter(vm_offset_t va, vm_paddr_t pa);
400 void	*pmap_kenter_temporary(vm_paddr_t pa, int i);
401 vm_paddr_t pmap_kextract(vm_offset_t);
402 void	pmap_kremove(vm_offset_t);
403 int	pmap_large_map(vm_paddr_t, vm_size_t, void **, vm_memattr_t);
404 void	pmap_large_map_wb(void *sva, vm_size_t len);
405 void	pmap_large_unmap(void *sva, vm_size_t len);
406 void	*pmap_mapbios(vm_paddr_t, vm_size_t);
407 void	*pmap_mapdev(vm_paddr_t, vm_size_t);
408 void	*pmap_mapdev_attr(vm_paddr_t, vm_size_t, int);
409 void	*pmap_mapdev_pciecfg(vm_paddr_t pa, vm_size_t size);
410 bool	pmap_not_in_di(void);
411 bool	pmap_page_is_mapped(vm_page_t m);
412 void	pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma);
413 void	pmap_page_set_memattr_noflush(vm_page_t m, vm_memattr_t ma);
414 void	pmap_pinit_pml4(vm_page_t);
415 void	pmap_pinit_pml5(vm_page_t);
416 bool	pmap_ps_enabled(pmap_t pmap);
417 void	pmap_unmapdev(void *, vm_size_t);
418 void	pmap_invalidate_page(pmap_t, vm_offset_t);
419 void	pmap_invalidate_range(pmap_t, vm_offset_t, vm_offset_t);
420 void	pmap_invalidate_all(pmap_t);
421 void	pmap_invalidate_cache(void);
422 void	pmap_invalidate_cache_pages(vm_page_t *pages, int count);
423 void	pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
424 void	pmap_force_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva);
425 void	pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num);
426 bool	pmap_map_io_transient(vm_page_t *, vm_offset_t *, int, bool);
427 void	pmap_unmap_io_transient(vm_page_t *, vm_offset_t *, int, bool);
428 void	pmap_map_delete(pmap_t, vm_offset_t, vm_offset_t);
429 void	pmap_pti_add_kva(vm_offset_t sva, vm_offset_t eva, bool exec);
430 void	pmap_pti_remove_kva(vm_offset_t sva, vm_offset_t eva);
431 void	pmap_pti_pcid_invalidate(uint64_t ucr3, uint64_t kcr3);
432 void	pmap_pti_pcid_invlpg(uint64_t ucr3, uint64_t kcr3, vm_offset_t va);
433 void	pmap_pti_pcid_invlrng(uint64_t ucr3, uint64_t kcr3, vm_offset_t sva,
434 	    vm_offset_t eva);
435 int	pmap_pkru_clear(pmap_t pmap, vm_offset_t sva, vm_offset_t eva);
436 int	pmap_pkru_set(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
437 	    u_int keyidx, int flags);
438 void	pmap_thread_init_invl_gen(struct thread *td);
439 int	pmap_vmspace_copy(pmap_t dst_pmap, pmap_t src_pmap);
440 void	pmap_page_array_startup(long count);
441 vm_page_t pmap_page_alloc_below_4g(bool zeroed);
442 
443 #if defined(KASAN) || defined(KMSAN)
444 void	pmap_san_enter(vm_offset_t);
445 #endif
446 
447 /*
448  * Returns a pointer to a set of CPUs on which the pmap is currently active.
449  * Note that the set can be modified without any mutual exclusion, so a copy
450  * must be made if a stable value is required.
451  */
452 static __inline volatile cpuset_t *
pmap_invalidate_cpu_mask(pmap_t pmap)453 pmap_invalidate_cpu_mask(pmap_t pmap)
454 {
455 	return (&pmap->pm_active);
456 }
457 
458 #if defined(_SYS_PCPU_H_) && defined(_MACHINE_CPUFUNC_H_)
459 /*
460  * It seems that AlderLake+ small cores have some microarchitectural
461  * bug, which results in the INVLPG instruction failing to flush all
462  * global TLB entries when PCID is enabled.  Work around it for now,
463  * by doing global invalidation on small cores instead of INVLPG.
464  */
465 static __inline void
pmap_invlpg(pmap_t pmap,vm_offset_t va)466 pmap_invlpg(pmap_t pmap, vm_offset_t va)
467 {
468 	if (pmap == kernel_pmap && PCPU_GET(pcid_invlpg_workaround)) {
469 		struct invpcid_descr d = { 0 };
470 
471 		invpcid(&d, INVPCID_CTXGLOB);
472 	} else {
473 		invlpg(va);
474 	}
475 }
476 #endif /* sys/pcpu.h && machine/cpufunc.h */
477 
478 #if defined(_SYS_PCPU_H_)
479 /* Return pcid for the pmap pmap on current cpu */
480 static __inline uint32_t
pmap_get_pcid(pmap_t pmap)481 pmap_get_pcid(pmap_t pmap)
482 {
483 	struct pmap_pcid *pcidp;
484 
485 	MPASS(pmap_pcid_enabled);
486 	pcidp = zpcpu_get(pmap->pm_pcidp);
487 	return (pcidp->pm_pcid);
488 }
489 #endif /* sys/pcpu.h */
490 
491 /*
492  * Invalidation request.  PCPU pc_smp_tlb_op uses u_int instead of the
493  * enum to avoid both namespace and ABI issues (with enums).
494  */
495 enum invl_op_codes {
496 	INVL_OP_TLB               = 1,
497 	INVL_OP_TLB_INVPCID       = 2,
498 	INVL_OP_TLB_INVPCID_PTI   = 3,
499 	INVL_OP_TLB_PCID          = 4,
500 	INVL_OP_PGRNG             = 5,
501 	INVL_OP_PGRNG_INVPCID     = 6,
502 	INVL_OP_PGRNG_PCID        = 7,
503 	INVL_OP_PG                = 8,
504 	INVL_OP_PG_INVPCID        = 9,
505 	INVL_OP_PG_PCID           = 10,
506 	INVL_OP_CACHE             = 11,
507 };
508 
509 typedef void (*smp_invl_local_cb_t)(struct pmap *, vm_offset_t addr1,
510     vm_offset_t addr2);
511 typedef void (*smp_targeted_tlb_shootdown_t)(pmap_t, vm_offset_t, vm_offset_t,
512     smp_invl_local_cb_t, enum invl_op_codes);
513 
514 void smp_targeted_tlb_shootdown_native(pmap_t, vm_offset_t, vm_offset_t,
515     smp_invl_local_cb_t, enum invl_op_codes);
516 extern smp_targeted_tlb_shootdown_t smp_targeted_tlb_shootdown;
517 
518 #endif /* _KERNEL */
519 
520 /* Return various clipped indexes for a given VA */
521 static __inline vm_pindex_t
pmap_pte_index(vm_offset_t va)522 pmap_pte_index(vm_offset_t va)
523 {
524 
525 	return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
526 }
527 
528 static __inline vm_pindex_t
pmap_pde_index(vm_offset_t va)529 pmap_pde_index(vm_offset_t va)
530 {
531 
532 	return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
533 }
534 
535 static __inline vm_pindex_t
pmap_pdpe_index(vm_offset_t va)536 pmap_pdpe_index(vm_offset_t va)
537 {
538 
539 	return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
540 }
541 
542 static __inline vm_pindex_t
pmap_pml4e_index(vm_offset_t va)543 pmap_pml4e_index(vm_offset_t va)
544 {
545 
546 	return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
547 }
548 
549 static __inline vm_pindex_t
pmap_pml5e_index(vm_offset_t va)550 pmap_pml5e_index(vm_offset_t va)
551 {
552 
553 	return ((va >> PML5SHIFT) & ((1ul << NPML5EPGSHIFT) - 1));
554 }
555 
556 #endif /* !LOCORE */
557 
558 #endif /* !_MACHINE_PMAP_H_ */
559 
560 #endif /* __i386__ */
561