xref: /linux/arch/arm64/include/asm/kvm_mmu.h (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 /*
2  * Copyright (C) 2012,2013 - ARM Ltd
3  * Author: Marc Zyngier <marc.zyngier@arm.com>
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License version 2 as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16  */
17 
18 #ifndef __ARM64_KVM_MMU_H__
19 #define __ARM64_KVM_MMU_H__
20 
21 #include <asm/page.h>
22 #include <asm/memory.h>
23 
24 /*
25  * As we only have the TTBR0_EL2 register, we cannot express
26  * "negative" addresses. This makes it impossible to directly share
27  * mappings with the kernel.
28  *
29  * Instead, give the HYP mode its own VA region at a fixed offset from
30  * the kernel by just masking the top bits (which are all ones for a
31  * kernel address).
32  */
33 #define HYP_PAGE_OFFSET_SHIFT	VA_BITS
34 #define HYP_PAGE_OFFSET_MASK	((UL(1) << HYP_PAGE_OFFSET_SHIFT) - 1)
35 #define HYP_PAGE_OFFSET		(PAGE_OFFSET & HYP_PAGE_OFFSET_MASK)
36 
37 /*
38  * Our virtual mapping for the idmap-ed MMU-enable code. Must be
39  * shared across all the page-tables. Conveniently, we use the last
40  * possible page, where no kernel mapping will ever exist.
41  */
42 #define TRAMPOLINE_VA		(HYP_PAGE_OFFSET_MASK & PAGE_MASK)
43 
44 /*
45  * KVM_MMU_CACHE_MIN_PAGES is the number of stage2 page table translation
46  * levels in addition to the PGD and potentially the PUD which are
47  * pre-allocated (we pre-allocate the fake PGD and the PUD when the Stage-2
48  * tables use one level of tables less than the kernel.
49  */
50 #ifdef CONFIG_ARM64_64K_PAGES
51 #define KVM_MMU_CACHE_MIN_PAGES	1
52 #else
53 #define KVM_MMU_CACHE_MIN_PAGES	2
54 #endif
55 
56 #ifdef __ASSEMBLY__
57 
58 /*
59  * Convert a kernel VA into a HYP VA.
60  * reg: VA to be converted.
61  */
62 .macro kern_hyp_va	reg
63 	and	\reg, \reg, #HYP_PAGE_OFFSET_MASK
64 .endm
65 
66 #else
67 
68 #include <asm/pgalloc.h>
69 #include <asm/cachetype.h>
70 #include <asm/cacheflush.h>
71 
72 #define KERN_TO_HYP(kva)	((unsigned long)kva - PAGE_OFFSET + HYP_PAGE_OFFSET)
73 
74 /*
75  * We currently only support a 40bit IPA.
76  */
77 #define KVM_PHYS_SHIFT	(40)
78 #define KVM_PHYS_SIZE	(1UL << KVM_PHYS_SHIFT)
79 #define KVM_PHYS_MASK	(KVM_PHYS_SIZE - 1UL)
80 
81 int create_hyp_mappings(void *from, void *to);
82 int create_hyp_io_mappings(void *from, void *to, phys_addr_t);
83 void free_boot_hyp_pgd(void);
84 void free_hyp_pgds(void);
85 
86 void stage2_unmap_vm(struct kvm *kvm);
87 int kvm_alloc_stage2_pgd(struct kvm *kvm);
88 void kvm_free_stage2_pgd(struct kvm *kvm);
89 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
90 			  phys_addr_t pa, unsigned long size, bool writable);
91 
92 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run);
93 
94 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
95 
96 phys_addr_t kvm_mmu_get_httbr(void);
97 phys_addr_t kvm_mmu_get_boot_httbr(void);
98 phys_addr_t kvm_get_idmap_vector(void);
99 int kvm_mmu_init(void);
100 void kvm_clear_hyp_idmap(void);
101 
102 #define	kvm_set_pte(ptep, pte)		set_pte(ptep, pte)
103 #define	kvm_set_pmd(pmdp, pmd)		set_pmd(pmdp, pmd)
104 
105 static inline void kvm_clean_pgd(pgd_t *pgd) {}
106 static inline void kvm_clean_pmd(pmd_t *pmd) {}
107 static inline void kvm_clean_pmd_entry(pmd_t *pmd) {}
108 static inline void kvm_clean_pte(pte_t *pte) {}
109 static inline void kvm_clean_pte_entry(pte_t *pte) {}
110 
111 static inline void kvm_set_s2pte_writable(pte_t *pte)
112 {
113 	pte_val(*pte) |= PTE_S2_RDWR;
114 }
115 
116 static inline void kvm_set_s2pmd_writable(pmd_t *pmd)
117 {
118 	pmd_val(*pmd) |= PMD_S2_RDWR;
119 }
120 
121 #define kvm_pgd_addr_end(addr, end)	pgd_addr_end(addr, end)
122 #define kvm_pud_addr_end(addr, end)	pud_addr_end(addr, end)
123 #define kvm_pmd_addr_end(addr, end)	pmd_addr_end(addr, end)
124 
125 /*
126  * In the case where PGDIR_SHIFT is larger than KVM_PHYS_SHIFT, we can address
127  * the entire IPA input range with a single pgd entry, and we would only need
128  * one pgd entry.  Note that in this case, the pgd is actually not used by
129  * the MMU for Stage-2 translations, but is merely a fake pgd used as a data
130  * structure for the kernel pgtable macros to work.
131  */
132 #if PGDIR_SHIFT > KVM_PHYS_SHIFT
133 #define PTRS_PER_S2_PGD_SHIFT	0
134 #else
135 #define PTRS_PER_S2_PGD_SHIFT	(KVM_PHYS_SHIFT - PGDIR_SHIFT)
136 #endif
137 #define PTRS_PER_S2_PGD		(1 << PTRS_PER_S2_PGD_SHIFT)
138 #define S2_PGD_ORDER		get_order(PTRS_PER_S2_PGD * sizeof(pgd_t))
139 
140 /*
141  * If we are concatenating first level stage-2 page tables, we would have less
142  * than or equal to 16 pointers in the fake PGD, because that's what the
143  * architecture allows.  In this case, (4 - CONFIG_ARM64_PGTABLE_LEVELS)
144  * represents the first level for the host, and we add 1 to go to the next
145  * level (which uses contatenation) for the stage-2 tables.
146  */
147 #if PTRS_PER_S2_PGD <= 16
148 #define KVM_PREALLOC_LEVEL	(4 - CONFIG_ARM64_PGTABLE_LEVELS + 1)
149 #else
150 #define KVM_PREALLOC_LEVEL	(0)
151 #endif
152 
153 /**
154  * kvm_prealloc_hwpgd - allocate inital table for VTTBR
155  * @kvm:	The KVM struct pointer for the VM.
156  * @pgd:	The kernel pseudo pgd
157  *
158  * When the kernel uses more levels of page tables than the guest, we allocate
159  * a fake PGD and pre-populate it to point to the next-level page table, which
160  * will be the real initial page table pointed to by the VTTBR.
161  *
162  * When KVM_PREALLOC_LEVEL==2, we allocate a single page for the PMD and
163  * the kernel will use folded pud.  When KVM_PREALLOC_LEVEL==1, we
164  * allocate 2 consecutive PUD pages.
165  */
166 static inline int kvm_prealloc_hwpgd(struct kvm *kvm, pgd_t *pgd)
167 {
168 	unsigned int i;
169 	unsigned long hwpgd;
170 
171 	if (KVM_PREALLOC_LEVEL == 0)
172 		return 0;
173 
174 	hwpgd = __get_free_pages(GFP_KERNEL | __GFP_ZERO, PTRS_PER_S2_PGD_SHIFT);
175 	if (!hwpgd)
176 		return -ENOMEM;
177 
178 	for (i = 0; i < PTRS_PER_S2_PGD; i++) {
179 		if (KVM_PREALLOC_LEVEL == 1)
180 			pgd_populate(NULL, pgd + i,
181 				     (pud_t *)hwpgd + i * PTRS_PER_PUD);
182 		else if (KVM_PREALLOC_LEVEL == 2)
183 			pud_populate(NULL, pud_offset(pgd, 0) + i,
184 				     (pmd_t *)hwpgd + i * PTRS_PER_PMD);
185 	}
186 
187 	return 0;
188 }
189 
190 static inline void *kvm_get_hwpgd(struct kvm *kvm)
191 {
192 	pgd_t *pgd = kvm->arch.pgd;
193 	pud_t *pud;
194 
195 	if (KVM_PREALLOC_LEVEL == 0)
196 		return pgd;
197 
198 	pud = pud_offset(pgd, 0);
199 	if (KVM_PREALLOC_LEVEL == 1)
200 		return pud;
201 
202 	BUG_ON(KVM_PREALLOC_LEVEL != 2);
203 	return pmd_offset(pud, 0);
204 }
205 
206 static inline void kvm_free_hwpgd(struct kvm *kvm)
207 {
208 	if (KVM_PREALLOC_LEVEL > 0) {
209 		unsigned long hwpgd = (unsigned long)kvm_get_hwpgd(kvm);
210 		free_pages(hwpgd, PTRS_PER_S2_PGD_SHIFT);
211 	}
212 }
213 
214 static inline bool kvm_page_empty(void *ptr)
215 {
216 	struct page *ptr_page = virt_to_page(ptr);
217 	return page_count(ptr_page) == 1;
218 }
219 
220 #define kvm_pte_table_empty(kvm, ptep) kvm_page_empty(ptep)
221 
222 #ifdef __PAGETABLE_PMD_FOLDED
223 #define kvm_pmd_table_empty(kvm, pmdp) (0)
224 #else
225 #define kvm_pmd_table_empty(kvm, pmdp) \
226 	(kvm_page_empty(pmdp) && (!(kvm) || KVM_PREALLOC_LEVEL < 2))
227 #endif
228 
229 #ifdef __PAGETABLE_PUD_FOLDED
230 #define kvm_pud_table_empty(kvm, pudp) (0)
231 #else
232 #define kvm_pud_table_empty(kvm, pudp) \
233 	(kvm_page_empty(pudp) && (!(kvm) || KVM_PREALLOC_LEVEL < 1))
234 #endif
235 
236 
237 struct kvm;
238 
239 #define kvm_flush_dcache_to_poc(a,l)	__flush_dcache_area((a), (l))
240 
241 static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
242 {
243 	return (vcpu_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
244 }
245 
246 static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
247 					       unsigned long size,
248 					       bool ipa_uncached)
249 {
250 	void *va = page_address(pfn_to_page(pfn));
251 
252 	if (!vcpu_has_cache_enabled(vcpu) || ipa_uncached)
253 		kvm_flush_dcache_to_poc(va, size);
254 
255 	if (!icache_is_aliasing()) {		/* PIPT */
256 		flush_icache_range((unsigned long)va,
257 				   (unsigned long)va + size);
258 	} else if (!icache_is_aivivt()) {	/* non ASID-tagged VIVT */
259 		/* any kind of VIPT cache */
260 		__flush_icache_all();
261 	}
262 }
263 
264 static inline void __kvm_flush_dcache_pte(pte_t pte)
265 {
266 	struct page *page = pte_page(pte);
267 	kvm_flush_dcache_to_poc(page_address(page), PAGE_SIZE);
268 }
269 
270 static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
271 {
272 	struct page *page = pmd_page(pmd);
273 	kvm_flush_dcache_to_poc(page_address(page), PMD_SIZE);
274 }
275 
276 static inline void __kvm_flush_dcache_pud(pud_t pud)
277 {
278 	struct page *page = pud_page(pud);
279 	kvm_flush_dcache_to_poc(page_address(page), PUD_SIZE);
280 }
281 
282 #define kvm_virt_to_phys(x)		__virt_to_phys((unsigned long)(x))
283 
284 void kvm_set_way_flush(struct kvm_vcpu *vcpu);
285 void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
286 
287 #endif /* __ASSEMBLY__ */
288 #endif /* __ARM64_KVM_MMU_H__ */
289