Merge tag 'arm64-perf' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
[linux-2.6-block.git] / arch / arm64 / include / asm / kvm_mmu.h
CommitLineData
37c43753
MZ
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>
20475f78 23#include <asm/cpufeature.h>
37c43753
MZ
24
25/*
cedbb8b7 26 * As ARMv8.0 only has the TTBR0_EL2 register, we cannot express
37c43753
MZ
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).
cedbb8b7
MZ
33 *
34 * ARMv8.1 (using VHE) does have a TTBR1_EL2, and doesn't use these
35 * macros (the entire kernel runs at EL2).
37c43753
MZ
36 */
37#define HYP_PAGE_OFFSET_SHIFT VA_BITS
38#define HYP_PAGE_OFFSET_MASK ((UL(1) << HYP_PAGE_OFFSET_SHIFT) - 1)
39#define HYP_PAGE_OFFSET (PAGE_OFFSET & HYP_PAGE_OFFSET_MASK)
40
41/*
42 * Our virtual mapping for the idmap-ed MMU-enable code. Must be
43 * shared across all the page-tables. Conveniently, we use the last
44 * possible page, where no kernel mapping will ever exist.
45 */
46#define TRAMPOLINE_VA (HYP_PAGE_OFFSET_MASK & PAGE_MASK)
47
38f791a4
CD
48/*
49 * KVM_MMU_CACHE_MIN_PAGES is the number of stage2 page table translation
50 * levels in addition to the PGD and potentially the PUD which are
51 * pre-allocated (we pre-allocate the fake PGD and the PUD when the Stage-2
52 * tables use one level of tables less than the kernel.
53 */
54#ifdef CONFIG_ARM64_64K_PAGES
55#define KVM_MMU_CACHE_MIN_PAGES 1
56#else
57#define KVM_MMU_CACHE_MIN_PAGES 2
58#endif
59
37c43753
MZ
60#ifdef __ASSEMBLY__
61
cedbb8b7
MZ
62#include <asm/alternative.h>
63#include <asm/cpufeature.h>
64
37c43753
MZ
65/*
66 * Convert a kernel VA into a HYP VA.
67 * reg: VA to be converted.
68 */
69.macro kern_hyp_va reg
cedbb8b7 70alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
37c43753 71 and \reg, \reg, #HYP_PAGE_OFFSET_MASK
cedbb8b7
MZ
72alternative_else
73 nop
74alternative_endif
37c43753
MZ
75.endm
76
77#else
78
38f791a4 79#include <asm/pgalloc.h>
37c43753
MZ
80#include <asm/cachetype.h>
81#include <asm/cacheflush.h>
e4c5a685
AB
82#include <asm/mmu_context.h>
83#include <asm/pgtable.h>
37c43753
MZ
84
85#define KERN_TO_HYP(kva) ((unsigned long)kva - PAGE_OFFSET + HYP_PAGE_OFFSET)
86
87/*
dbff124e 88 * We currently only support a 40bit IPA.
37c43753 89 */
dbff124e 90#define KVM_PHYS_SHIFT (40)
37c43753
MZ
91#define KVM_PHYS_SIZE (1UL << KVM_PHYS_SHIFT)
92#define KVM_PHYS_MASK (KVM_PHYS_SIZE - 1UL)
93
37c43753
MZ
94int create_hyp_mappings(void *from, void *to);
95int create_hyp_io_mappings(void *from, void *to, phys_addr_t);
96void free_boot_hyp_pgd(void);
97void free_hyp_pgds(void);
98
957db105 99void stage2_unmap_vm(struct kvm *kvm);
37c43753
MZ
100int kvm_alloc_stage2_pgd(struct kvm *kvm);
101void kvm_free_stage2_pgd(struct kvm *kvm);
102int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
c40f2f8f 103 phys_addr_t pa, unsigned long size, bool writable);
37c43753
MZ
104
105int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run);
106
107void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
108
109phys_addr_t kvm_mmu_get_httbr(void);
110phys_addr_t kvm_mmu_get_boot_httbr(void);
111phys_addr_t kvm_get_idmap_vector(void);
67f69197 112phys_addr_t kvm_get_idmap_start(void);
37c43753
MZ
113int kvm_mmu_init(void);
114void kvm_clear_hyp_idmap(void);
115
116#define kvm_set_pte(ptep, pte) set_pte(ptep, pte)
ad361f09 117#define kvm_set_pmd(pmdp, pmd) set_pmd(pmdp, pmd)
37c43753 118
37c43753 119static inline void kvm_clean_pgd(pgd_t *pgd) {}
38f791a4 120static inline void kvm_clean_pmd(pmd_t *pmd) {}
37c43753
MZ
121static inline void kvm_clean_pmd_entry(pmd_t *pmd) {}
122static inline void kvm_clean_pte(pte_t *pte) {}
123static inline void kvm_clean_pte_entry(pte_t *pte) {}
124
125static inline void kvm_set_s2pte_writable(pte_t *pte)
126{
127 pte_val(*pte) |= PTE_S2_RDWR;
128}
129
ad361f09
CD
130static inline void kvm_set_s2pmd_writable(pmd_t *pmd)
131{
132 pmd_val(*pmd) |= PMD_S2_RDWR;
133}
134
8199ed0e
MS
135static inline void kvm_set_s2pte_readonly(pte_t *pte)
136{
137 pte_val(*pte) = (pte_val(*pte) & ~PTE_S2_RDWR) | PTE_S2_RDONLY;
138}
139
140static inline bool kvm_s2pte_readonly(pte_t *pte)
141{
142 return (pte_val(*pte) & PTE_S2_RDWR) == PTE_S2_RDONLY;
143}
144
145static inline void kvm_set_s2pmd_readonly(pmd_t *pmd)
146{
147 pmd_val(*pmd) = (pmd_val(*pmd) & ~PMD_S2_RDWR) | PMD_S2_RDONLY;
148}
149
150static inline bool kvm_s2pmd_readonly(pmd_t *pmd)
151{
152 return (pmd_val(*pmd) & PMD_S2_RDWR) == PMD_S2_RDONLY;
153}
154
155
a3c8bd31
MZ
156#define kvm_pgd_addr_end(addr, end) pgd_addr_end(addr, end)
157#define kvm_pud_addr_end(addr, end) pud_addr_end(addr, end)
158#define kvm_pmd_addr_end(addr, end) pmd_addr_end(addr, end)
159
38f791a4
CD
160/*
161 * In the case where PGDIR_SHIFT is larger than KVM_PHYS_SHIFT, we can address
162 * the entire IPA input range with a single pgd entry, and we would only need
163 * one pgd entry. Note that in this case, the pgd is actually not used by
164 * the MMU for Stage-2 translations, but is merely a fake pgd used as a data
165 * structure for the kernel pgtable macros to work.
166 */
167#if PGDIR_SHIFT > KVM_PHYS_SHIFT
168#define PTRS_PER_S2_PGD_SHIFT 0
169#else
170#define PTRS_PER_S2_PGD_SHIFT (KVM_PHYS_SHIFT - PGDIR_SHIFT)
171#endif
172#define PTRS_PER_S2_PGD (1 << PTRS_PER_S2_PGD_SHIFT)
38f791a4 173
04b8dc85
MZ
174#define kvm_pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_S2_PGD - 1))
175
38f791a4
CD
176/*
177 * If we are concatenating first level stage-2 page tables, we would have less
178 * than or equal to 16 pointers in the fake PGD, because that's what the
9f25e6ad 179 * architecture allows. In this case, (4 - CONFIG_PGTABLE_LEVELS)
38f791a4
CD
180 * represents the first level for the host, and we add 1 to go to the next
181 * level (which uses contatenation) for the stage-2 tables.
182 */
183#if PTRS_PER_S2_PGD <= 16
9f25e6ad 184#define KVM_PREALLOC_LEVEL (4 - CONFIG_PGTABLE_LEVELS + 1)
38f791a4
CD
185#else
186#define KVM_PREALLOC_LEVEL (0)
187#endif
188
38f791a4
CD
189static inline void *kvm_get_hwpgd(struct kvm *kvm)
190{
191 pgd_t *pgd = kvm->arch.pgd;
192 pud_t *pud;
193
194 if (KVM_PREALLOC_LEVEL == 0)
195 return pgd;
196
197 pud = pud_offset(pgd, 0);
198 if (KVM_PREALLOC_LEVEL == 1)
199 return pud;
200
201 BUG_ON(KVM_PREALLOC_LEVEL != 2);
202 return pmd_offset(pud, 0);
203}
204
a987370f 205static inline unsigned int kvm_get_hwpgd_size(void)
38f791a4 206{
a987370f
MZ
207 if (KVM_PREALLOC_LEVEL > 0)
208 return PTRS_PER_S2_PGD * PAGE_SIZE;
209 return PTRS_PER_S2_PGD * sizeof(pgd_t);
38f791a4
CD
210}
211
4f853a71
CD
212static inline bool kvm_page_empty(void *ptr)
213{
214 struct page *ptr_page = virt_to_page(ptr);
215 return page_count(ptr_page) == 1;
216}
217
38f791a4
CD
218#define kvm_pte_table_empty(kvm, ptep) kvm_page_empty(ptep)
219
220#ifdef __PAGETABLE_PMD_FOLDED
221#define kvm_pmd_table_empty(kvm, pmdp) (0)
222#else
223#define kvm_pmd_table_empty(kvm, pmdp) \
224 (kvm_page_empty(pmdp) && (!(kvm) || KVM_PREALLOC_LEVEL < 2))
225#endif
226
227#ifdef __PAGETABLE_PUD_FOLDED
228#define kvm_pud_table_empty(kvm, pudp) (0)
4f853a71 229#else
38f791a4
CD
230#define kvm_pud_table_empty(kvm, pudp) \
231 (kvm_page_empty(pudp) && (!(kvm) || KVM_PREALLOC_LEVEL < 1))
4f853a71 232#endif
4f853a71
CD
233
234
37c43753
MZ
235struct kvm;
236
2d58b733
MZ
237#define kvm_flush_dcache_to_poc(a,l) __flush_dcache_area((a), (l))
238
239static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
37c43753 240{
2d58b733
MZ
241 return (vcpu_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
242}
243
ba049e93
DW
244static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu,
245 kvm_pfn_t pfn,
0d3e4d4f
MZ
246 unsigned long size,
247 bool ipa_uncached)
2d58b733 248{
0d3e4d4f
MZ
249 void *va = page_address(pfn_to_page(pfn));
250
840f4bfb 251 if (!vcpu_has_cache_enabled(vcpu) || ipa_uncached)
0d3e4d4f 252 kvm_flush_dcache_to_poc(va, size);
2d58b733 253
37c43753 254 if (!icache_is_aliasing()) { /* PIPT */
0d3e4d4f
MZ
255 flush_icache_range((unsigned long)va,
256 (unsigned long)va + size);
37c43753
MZ
257 } else if (!icache_is_aivivt()) { /* non ASID-tagged VIVT */
258 /* any kind of VIPT cache */
259 __flush_icache_all();
260 }
261}
262
363ef89f
MZ
263static inline void __kvm_flush_dcache_pte(pte_t pte)
264{
265 struct page *page = pte_page(pte);
266 kvm_flush_dcache_to_poc(page_address(page), PAGE_SIZE);
267}
268
269static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
270{
271 struct page *page = pmd_page(pmd);
272 kvm_flush_dcache_to_poc(page_address(page), PMD_SIZE);
273}
274
275static inline void __kvm_flush_dcache_pud(pud_t pud)
276{
277 struct page *page = pud_page(pud);
278 kvm_flush_dcache_to_poc(page_address(page), PUD_SIZE);
279}
280
4fda342c 281#define kvm_virt_to_phys(x) __virt_to_phys((unsigned long)(x))
37c43753 282
3c1e7165
MZ
283void kvm_set_way_flush(struct kvm_vcpu *vcpu);
284void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
9d218a1f 285
e4c5a685
AB
286static inline bool __kvm_cpu_uses_extended_idmap(void)
287{
288 return __cpu_uses_extended_idmap();
289}
290
291static inline void __kvm_extend_hypmap(pgd_t *boot_hyp_pgd,
292 pgd_t *hyp_pgd,
293 pgd_t *merged_hyp_pgd,
294 unsigned long hyp_idmap_start)
295{
296 int idmap_idx;
297
298 /*
299 * Use the first entry to access the HYP mappings. It is
300 * guaranteed to be free, otherwise we wouldn't use an
301 * extended idmap.
302 */
303 VM_BUG_ON(pgd_val(merged_hyp_pgd[0]));
304 merged_hyp_pgd[0] = __pgd(__pa(hyp_pgd) | PMD_TYPE_TABLE);
305
306 /*
307 * Create another extended level entry that points to the boot HYP map,
308 * which contains an ID mapping of the HYP init code. We essentially
309 * merge the boot and runtime HYP maps by doing so, but they don't
310 * overlap anyway, so this is fine.
311 */
312 idmap_idx = hyp_idmap_start >> VA_BITS;
313 VM_BUG_ON(pgd_val(merged_hyp_pgd[idmap_idx]));
314 merged_hyp_pgd[idmap_idx] = __pgd(__pa(boot_hyp_pgd) | PMD_TYPE_TABLE);
315}
316
20475f78
VM
317static inline unsigned int kvm_get_vmid_bits(void)
318{
319 int reg = read_system_reg(SYS_ID_AA64MMFR1_EL1);
320
28c5dcb2 321 return (cpuid_feature_extract_unsigned_field(reg, ID_AA64MMFR1_VMIDBITS_SHIFT) == 2) ? 16 : 8;
20475f78
VM
322}
323
37c43753
MZ
324#endif /* __ASSEMBLY__ */
325#endif /* __ARM64_KVM_MMU_H__ */