xref: /openbmc/linux/arch/arm64/include/asm/kvm_mmu.h (revision bc5aa3a0)
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 #include <asm/cpufeature.h>
24 
25 /*
26  * As ARMv8.0 only has the TTBR0_EL2 register, we cannot express
27  * "negative" addresses. This makes it impossible to directly share
28  * mappings with the kernel.
29  *
30  * Instead, give the HYP mode its own VA region at a fixed offset from
31  * the kernel by just masking the top bits (which are all ones for a
32  * kernel address). We need to find out how many bits to mask.
33  *
34  * We want to build a set of page tables that cover both parts of the
35  * idmap (the trampoline page used to initialize EL2), and our normal
36  * runtime VA space, at the same time.
37  *
38  * Given that the kernel uses VA_BITS for its entire address space,
39  * and that half of that space (VA_BITS - 1) is used for the linear
40  * mapping, we can also limit the EL2 space to (VA_BITS - 1).
41  *
42  * The main question is "Within the VA_BITS space, does EL2 use the
43  * top or the bottom half of that space to shadow the kernel's linear
44  * mapping?". As we need to idmap the trampoline page, this is
45  * determined by the range in which this page lives.
46  *
47  * If the page is in the bottom half, we have to use the top half. If
48  * the page is in the top half, we have to use the bottom half:
49  *
50  * T = __virt_to_phys(__hyp_idmap_text_start)
51  * if (T & BIT(VA_BITS - 1))
52  *	HYP_VA_MIN = 0  //idmap in upper half
53  * else
54  *	HYP_VA_MIN = 1 << (VA_BITS - 1)
55  * HYP_VA_MAX = HYP_VA_MIN + (1 << (VA_BITS - 1)) - 1
56  *
57  * This of course assumes that the trampoline page exists within the
58  * VA_BITS range. If it doesn't, then it means we're in the odd case
59  * where the kernel idmap (as well as HYP) uses more levels than the
60  * kernel runtime page tables (as seen when the kernel is configured
61  * for 4k pages, 39bits VA, and yet memory lives just above that
62  * limit, forcing the idmap to use 4 levels of page tables while the
63  * kernel itself only uses 3). In this particular case, it doesn't
64  * matter which side of VA_BITS we use, as we're guaranteed not to
65  * conflict with anything.
66  *
67  * When using VHE, there are no separate hyp mappings and all KVM
68  * functionality is already mapped as part of the main kernel
69  * mappings, and none of this applies in that case.
70  */
71 
72 #define HYP_PAGE_OFFSET_HIGH_MASK	((UL(1) << VA_BITS) - 1)
73 #define HYP_PAGE_OFFSET_LOW_MASK	((UL(1) << (VA_BITS - 1)) - 1)
74 
75 #ifdef __ASSEMBLY__
76 
77 #include <asm/alternative.h>
78 #include <asm/cpufeature.h>
79 
80 /*
81  * Convert a kernel VA into a HYP VA.
82  * reg: VA to be converted.
83  *
84  * This generates the following sequences:
85  * - High mask:
86  *		and x0, x0, #HYP_PAGE_OFFSET_HIGH_MASK
87  *		nop
88  * - Low mask:
89  *		and x0, x0, #HYP_PAGE_OFFSET_HIGH_MASK
90  *		and x0, x0, #HYP_PAGE_OFFSET_LOW_MASK
91  * - VHE:
92  *		nop
93  *		nop
94  *
95  * The "low mask" version works because the mask is a strict subset of
96  * the "high mask", hence performing the first mask for nothing.
97  * Should be completely invisible on any viable CPU.
98  */
99 .macro kern_hyp_va	reg
100 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
101 	and     \reg, \reg, #HYP_PAGE_OFFSET_HIGH_MASK
102 alternative_else
103 	nop
104 alternative_endif
105 alternative_if_not ARM64_HYP_OFFSET_LOW
106 	nop
107 alternative_else
108 	and     \reg, \reg, #HYP_PAGE_OFFSET_LOW_MASK
109 alternative_endif
110 .endm
111 
112 #else
113 
114 #include <asm/pgalloc.h>
115 #include <asm/cachetype.h>
116 #include <asm/cacheflush.h>
117 #include <asm/mmu_context.h>
118 #include <asm/pgtable.h>
119 
120 static inline unsigned long __kern_hyp_va(unsigned long v)
121 {
122 	asm volatile(ALTERNATIVE("and %0, %0, %1",
123 				 "nop",
124 				 ARM64_HAS_VIRT_HOST_EXTN)
125 		     : "+r" (v)
126 		     : "i" (HYP_PAGE_OFFSET_HIGH_MASK));
127 	asm volatile(ALTERNATIVE("nop",
128 				 "and %0, %0, %1",
129 				 ARM64_HYP_OFFSET_LOW)
130 		     : "+r" (v)
131 		     : "i" (HYP_PAGE_OFFSET_LOW_MASK));
132 	return v;
133 }
134 
135 #define kern_hyp_va(v) 	(typeof(v))(__kern_hyp_va((unsigned long)(v)))
136 
137 /*
138  * We currently only support a 40bit IPA.
139  */
140 #define KVM_PHYS_SHIFT	(40)
141 #define KVM_PHYS_SIZE	(1UL << KVM_PHYS_SHIFT)
142 #define KVM_PHYS_MASK	(KVM_PHYS_SIZE - 1UL)
143 
144 #include <asm/stage2_pgtable.h>
145 
146 int create_hyp_mappings(void *from, void *to, pgprot_t prot);
147 int create_hyp_io_mappings(void *from, void *to, phys_addr_t);
148 void free_hyp_pgds(void);
149 
150 void stage2_unmap_vm(struct kvm *kvm);
151 int kvm_alloc_stage2_pgd(struct kvm *kvm);
152 void kvm_free_stage2_pgd(struct kvm *kvm);
153 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
154 			  phys_addr_t pa, unsigned long size, bool writable);
155 
156 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run);
157 
158 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
159 
160 phys_addr_t kvm_mmu_get_httbr(void);
161 phys_addr_t kvm_get_idmap_vector(void);
162 phys_addr_t kvm_get_idmap_start(void);
163 int kvm_mmu_init(void);
164 void kvm_clear_hyp_idmap(void);
165 
166 #define	kvm_set_pte(ptep, pte)		set_pte(ptep, pte)
167 #define	kvm_set_pmd(pmdp, pmd)		set_pmd(pmdp, pmd)
168 
169 static inline void kvm_clean_pgd(pgd_t *pgd) {}
170 static inline void kvm_clean_pmd(pmd_t *pmd) {}
171 static inline void kvm_clean_pmd_entry(pmd_t *pmd) {}
172 static inline void kvm_clean_pte(pte_t *pte) {}
173 static inline void kvm_clean_pte_entry(pte_t *pte) {}
174 
175 static inline pte_t kvm_s2pte_mkwrite(pte_t pte)
176 {
177 	pte_val(pte) |= PTE_S2_RDWR;
178 	return pte;
179 }
180 
181 static inline pmd_t kvm_s2pmd_mkwrite(pmd_t pmd)
182 {
183 	pmd_val(pmd) |= PMD_S2_RDWR;
184 	return pmd;
185 }
186 
187 static inline void kvm_set_s2pte_readonly(pte_t *pte)
188 {
189 	pteval_t pteval;
190 	unsigned long tmp;
191 
192 	asm volatile("//	kvm_set_s2pte_readonly\n"
193 	"	prfm	pstl1strm, %2\n"
194 	"1:	ldxr	%0, %2\n"
195 	"	and	%0, %0, %3		// clear PTE_S2_RDWR\n"
196 	"	orr	%0, %0, %4		// set PTE_S2_RDONLY\n"
197 	"	stxr	%w1, %0, %2\n"
198 	"	cbnz	%w1, 1b\n"
199 	: "=&r" (pteval), "=&r" (tmp), "+Q" (pte_val(*pte))
200 	: "L" (~PTE_S2_RDWR), "L" (PTE_S2_RDONLY));
201 }
202 
203 static inline bool kvm_s2pte_readonly(pte_t *pte)
204 {
205 	return (pte_val(*pte) & PTE_S2_RDWR) == PTE_S2_RDONLY;
206 }
207 
208 static inline void kvm_set_s2pmd_readonly(pmd_t *pmd)
209 {
210 	kvm_set_s2pte_readonly((pte_t *)pmd);
211 }
212 
213 static inline bool kvm_s2pmd_readonly(pmd_t *pmd)
214 {
215 	return kvm_s2pte_readonly((pte_t *)pmd);
216 }
217 
218 static inline bool kvm_page_empty(void *ptr)
219 {
220 	struct page *ptr_page = virt_to_page(ptr);
221 	return page_count(ptr_page) == 1;
222 }
223 
224 #define hyp_pte_table_empty(ptep) kvm_page_empty(ptep)
225 
226 #ifdef __PAGETABLE_PMD_FOLDED
227 #define hyp_pmd_table_empty(pmdp) (0)
228 #else
229 #define hyp_pmd_table_empty(pmdp) kvm_page_empty(pmdp)
230 #endif
231 
232 #ifdef __PAGETABLE_PUD_FOLDED
233 #define hyp_pud_table_empty(pudp) (0)
234 #else
235 #define hyp_pud_table_empty(pudp) kvm_page_empty(pudp)
236 #endif
237 
238 struct kvm;
239 
240 #define kvm_flush_dcache_to_poc(a,l)	__flush_dcache_area((a), (l))
241 
242 static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
243 {
244 	return (vcpu_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
245 }
246 
247 static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu,
248 					       kvm_pfn_t pfn,
249 					       unsigned long size,
250 					       bool ipa_uncached)
251 {
252 	void *va = page_address(pfn_to_page(pfn));
253 
254 	if (!vcpu_has_cache_enabled(vcpu) || ipa_uncached)
255 		kvm_flush_dcache_to_poc(va, size);
256 
257 	if (!icache_is_aliasing()) {		/* PIPT */
258 		flush_icache_range((unsigned long)va,
259 				   (unsigned long)va + size);
260 	} else if (!icache_is_aivivt()) {	/* non ASID-tagged VIVT */
261 		/* any kind of VIPT cache */
262 		__flush_icache_all();
263 	}
264 }
265 
266 static inline void __kvm_flush_dcache_pte(pte_t pte)
267 {
268 	struct page *page = pte_page(pte);
269 	kvm_flush_dcache_to_poc(page_address(page), PAGE_SIZE);
270 }
271 
272 static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
273 {
274 	struct page *page = pmd_page(pmd);
275 	kvm_flush_dcache_to_poc(page_address(page), PMD_SIZE);
276 }
277 
278 static inline void __kvm_flush_dcache_pud(pud_t pud)
279 {
280 	struct page *page = pud_page(pud);
281 	kvm_flush_dcache_to_poc(page_address(page), PUD_SIZE);
282 }
283 
284 #define kvm_virt_to_phys(x)		__virt_to_phys((unsigned long)(x))
285 
286 void kvm_set_way_flush(struct kvm_vcpu *vcpu);
287 void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
288 
289 static inline bool __kvm_cpu_uses_extended_idmap(void)
290 {
291 	return __cpu_uses_extended_idmap();
292 }
293 
294 static inline void __kvm_extend_hypmap(pgd_t *boot_hyp_pgd,
295 				       pgd_t *hyp_pgd,
296 				       pgd_t *merged_hyp_pgd,
297 				       unsigned long hyp_idmap_start)
298 {
299 	int idmap_idx;
300 
301 	/*
302 	 * Use the first entry to access the HYP mappings. It is
303 	 * guaranteed to be free, otherwise we wouldn't use an
304 	 * extended idmap.
305 	 */
306 	VM_BUG_ON(pgd_val(merged_hyp_pgd[0]));
307 	merged_hyp_pgd[0] = __pgd(__pa(hyp_pgd) | PMD_TYPE_TABLE);
308 
309 	/*
310 	 * Create another extended level entry that points to the boot HYP map,
311 	 * which contains an ID mapping of the HYP init code. We essentially
312 	 * merge the boot and runtime HYP maps by doing so, but they don't
313 	 * overlap anyway, so this is fine.
314 	 */
315 	idmap_idx = hyp_idmap_start >> VA_BITS;
316 	VM_BUG_ON(pgd_val(merged_hyp_pgd[idmap_idx]));
317 	merged_hyp_pgd[idmap_idx] = __pgd(__pa(boot_hyp_pgd) | PMD_TYPE_TABLE);
318 }
319 
320 static inline unsigned int kvm_get_vmid_bits(void)
321 {
322 	int reg = read_system_reg(SYS_ID_AA64MMFR1_EL1);
323 
324 	return (cpuid_feature_extract_unsigned_field(reg, ID_AA64MMFR1_VMIDBITS_SHIFT) == 2) ? 16 : 8;
325 }
326 
327 #endif /* __ASSEMBLY__ */
328 #endif /* __ARM64_KVM_MMU_H__ */
329