xref: /kvm-unit-tests/lib/arm/mmu.c (revision d5b60621b0e6863fd62e4657b7a2ec70bf7eee34)
1 /*
2  * MMU enable and page table manipulation functions
3  *
4  * Copyright (C) 2014, Red Hat Inc, Andrew Jones <drjones@redhat.com>
5  *
6  * This work is licensed under the terms of the GNU LGPL, version 2.
7  */
8 #include <asm/setup.h>
9 #include <asm/thread_info.h>
10 #include <asm/cpumask.h>
11 #include <asm/mmu.h>
12 #include <asm/setup.h>
13 #include <asm/page.h>
14 
15 #include "alloc_page.h"
16 #include "vmalloc.h"
17 #include <asm/pgtable-hwdef.h>
18 #include <asm/pgtable.h>
19 
20 extern unsigned long etext;
21 
22 pgd_t *mmu_idmap;
23 
24 /* CPU 0 starts with disabled MMU */
25 static cpumask_t mmu_disabled_cpumask = { {1} };
26 unsigned int mmu_disabled_cpu_count = 1;
27 
28 bool __mmu_enabled(void)
29 {
30 	int cpu = current_thread_info()->cpu;
31 
32 	/*
33 	 * mmu_enabled is called from places that are guarding the
34 	 * use of exclusive ops (which require the mmu to be enabled).
35 	 * That means we CANNOT call anything from here that may use a
36 	 * spinlock, atomic bitop, etc., otherwise we'll recurse.
37 	 * [cpumask_]test_bit is safe though.
38 	 */
39 	return !cpumask_test_cpu(cpu, &mmu_disabled_cpumask);
40 }
41 
42 void mmu_mark_enabled(int cpu)
43 {
44 	if (cpumask_test_and_clear_cpu(cpu, &mmu_disabled_cpumask))
45 		--mmu_disabled_cpu_count;
46 }
47 
48 void mmu_mark_disabled(int cpu)
49 {
50 	if (!cpumask_test_and_set_cpu(cpu, &mmu_disabled_cpumask))
51 		++mmu_disabled_cpu_count;
52 }
53 
54 extern void asm_mmu_enable(phys_addr_t pgtable);
55 void mmu_enable(pgd_t *pgtable)
56 {
57 	struct thread_info *info = current_thread_info();
58 
59 	asm_mmu_enable(__pa(pgtable));
60 	flush_tlb_all();
61 
62 	info->pgtable = pgtable;
63 	mmu_mark_enabled(info->cpu);
64 }
65 
66 extern void asm_mmu_disable(void);
67 void mmu_disable(void)
68 {
69 	int cpu = current_thread_info()->cpu;
70 
71 	mmu_mark_disabled(cpu);
72 
73 	asm_mmu_disable();
74 }
75 
76 static pteval_t *get_pte(pgd_t *pgtable, uintptr_t vaddr)
77 {
78 	pgd_t *pgd = pgd_offset(pgtable, vaddr);
79 	pmd_t *pmd = pmd_alloc(pgd, vaddr);
80 	pte_t *pte = pte_alloc(pmd, vaddr);
81 
82 	return &pte_val(*pte);
83 }
84 
85 static pteval_t *install_pte(pgd_t *pgtable, uintptr_t vaddr, pteval_t pte)
86 {
87 	pteval_t *p_pte = get_pte(pgtable, vaddr);
88 
89 	*p_pte = pte;
90 	flush_tlb_page(vaddr);
91 	return p_pte;
92 }
93 
94 static pteval_t *install_page_prot(pgd_t *pgtable, phys_addr_t phys,
95 				   uintptr_t vaddr, pgprot_t prot)
96 {
97 	pteval_t pte = phys;
98 	pte |= PTE_TYPE_PAGE | PTE_AF | PTE_SHARED;
99 	pte |= pgprot_val(prot);
100 	return install_pte(pgtable, vaddr, pte);
101 }
102 
103 pteval_t *install_page(pgd_t *pgtable, phys_addr_t phys, void *virt)
104 {
105 	return install_page_prot(pgtable, phys, (uintptr_t)virt,
106 				 __pgprot(PTE_WBWA | PTE_USER));
107 }
108 
109 phys_addr_t virt_to_pte_phys(pgd_t *pgtable, void *mem)
110 {
111 	return (*get_pte(pgtable, (uintptr_t)mem) & PHYS_MASK & -PAGE_SIZE)
112 		+ ((ulong)mem & (PAGE_SIZE - 1));
113 }
114 
115 void mmu_set_range_ptes(pgd_t *pgtable, uintptr_t virt_offset,
116 			phys_addr_t phys_start, phys_addr_t phys_end,
117 			pgprot_t prot)
118 {
119 	phys_addr_t paddr = phys_start & PAGE_MASK;
120 	uintptr_t vaddr = virt_offset & PAGE_MASK;
121 	uintptr_t virt_end = phys_end - paddr + vaddr;
122 
123 	for (; vaddr < virt_end; vaddr += PAGE_SIZE, paddr += PAGE_SIZE)
124 		install_page_prot(pgtable, paddr, vaddr, prot);
125 }
126 
127 void mmu_set_range_sect(pgd_t *pgtable, uintptr_t virt_offset,
128 			phys_addr_t phys_start, phys_addr_t phys_end,
129 			pgprot_t prot)
130 {
131 	phys_addr_t paddr = phys_start & PGDIR_MASK;
132 	uintptr_t vaddr = virt_offset & PGDIR_MASK;
133 	uintptr_t virt_end = phys_end - paddr + vaddr;
134 
135 	for (; vaddr < virt_end; vaddr += PGDIR_SIZE, paddr += PGDIR_SIZE) {
136 		pgd_t *pgd = pgd_offset(pgtable, vaddr);
137 		pgd_val(*pgd) = paddr;
138 		pgd_val(*pgd) |= PMD_TYPE_SECT | PMD_SECT_AF | PMD_SECT_S;
139 		pgd_val(*pgd) |= pgprot_val(prot);
140 		flush_tlb_page(vaddr);
141 	}
142 }
143 
144 void *setup_mmu(phys_addr_t phys_end)
145 {
146 	uintptr_t code_end = (uintptr_t)&etext;
147 
148 	/* 0G-1G = I/O, 1G-3G = identity, 3G-4G = vmalloc */
149 	if (phys_end > (3ul << 30))
150 		phys_end = 3ul << 30;
151 
152 #ifdef __aarch64__
153 	init_alloc_vpage((void*)(4ul << 30));
154 #endif
155 
156 	mmu_idmap = alloc_page();
157 
158 	/*
159 	 * mach-virt I/O regions:
160 	 *   - The first 1G (arm/arm64)
161 	 *   - 512M at 256G (arm64, arm uses highmem=off)
162 	 *   - 512G at 512G (arm64, arm uses highmem=off)
163 	 */
164 	mmu_set_range_sect(mmu_idmap,
165 		0, 0, (1ul << 30),
166 		__pgprot(PMD_SECT_UNCACHED | PMD_SECT_USER));
167 #ifdef __aarch64__
168 	mmu_set_range_sect(mmu_idmap,
169 		(1ul << 38), (1ul << 38), (1ul << 38) | (1ul << 29),
170 		__pgprot(PMD_SECT_UNCACHED | PMD_SECT_USER));
171 	mmu_set_range_sect(mmu_idmap,
172 		(1ul << 39), (1ul << 39), (1ul << 40),
173 		__pgprot(PMD_SECT_UNCACHED | PMD_SECT_USER));
174 #endif
175 
176 	/* armv8 requires code shared between EL1 and EL0 to be read-only */
177 	mmu_set_range_ptes(mmu_idmap, PHYS_OFFSET,
178 		PHYS_OFFSET, code_end,
179 		__pgprot(PTE_WBWA | PTE_RDONLY | PTE_USER));
180 
181 	mmu_set_range_ptes(mmu_idmap, code_end,
182 		code_end, phys_end,
183 		__pgprot(PTE_WBWA | PTE_USER));
184 
185 	mmu_enable(mmu_idmap);
186 	return mmu_idmap;
187 }
188 
189 phys_addr_t __virt_to_phys(unsigned long addr)
190 {
191 	if (mmu_enabled()) {
192 		pgd_t *pgtable = current_thread_info()->pgtable;
193 		return virt_to_pte_phys(pgtable, (void *)addr);
194 	}
195 	return addr;
196 }
197 
198 unsigned long __phys_to_virt(phys_addr_t addr)
199 {
200 	/*
201 	 * We don't guarantee that phys_to_virt(virt_to_phys(vaddr)) == vaddr, but
202 	 * the default page tables do identity map all physical addresses, which
203 	 * means phys_to_virt(virt_to_phys((void *)paddr)) == paddr.
204 	 */
205 	assert(!mmu_enabled() || __virt_to_phys(addr) == addr);
206 	return addr;
207 }
208 
209 void mmu_clear_user(unsigned long vaddr)
210 {
211 	pgd_t *pgtable;
212 	pteval_t *pte;
213 
214 	if (!mmu_enabled())
215 		return;
216 
217 	pgtable = current_thread_info()->pgtable;
218 	pte = get_pte(pgtable, vaddr);
219 
220 	*pte &= ~PTE_USER;
221 	flush_tlb_page(vaddr);
222 }
223