xref: /linux/arch/powerpc/mm/cacheflush.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 
3 #include <linux/highmem.h>
4 #include <linux/kprobes.h>
5 
6 /**
7  * flush_coherent_icache() - if a CPU has a coherent icache, flush it
8  * Return true if the cache was flushed, false otherwise
9  */
10 static inline bool flush_coherent_icache(void)
11 {
12 	/*
13 	 * For a snooping icache, we still need a dummy icbi to purge all the
14 	 * prefetched instructions from the ifetch buffers. We also need a sync
15 	 * before the icbi to order the the actual stores to memory that might
16 	 * have modified instructions with the icbi.
17 	 */
18 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
19 		mb(); /* sync */
20 		icbi((void *)PAGE_OFFSET);
21 		mb(); /* sync */
22 		isync();
23 		return true;
24 	}
25 
26 	return false;
27 }
28 
29 /**
30  * invalidate_icache_range() - Flush the icache by issuing icbi across an address range
31  * @start: the start address
32  * @stop: the stop address (exclusive)
33  */
34 static void invalidate_icache_range(unsigned long start, unsigned long stop)
35 {
36 	unsigned long shift = l1_icache_shift();
37 	unsigned long bytes = l1_icache_bytes();
38 	char *addr = (char *)(start & ~(bytes - 1));
39 	unsigned long size = stop - (unsigned long)addr + (bytes - 1);
40 	unsigned long i;
41 
42 	for (i = 0; i < size >> shift; i++, addr += bytes)
43 		icbi(addr);
44 
45 	mb(); /* sync */
46 	isync();
47 }
48 
49 /**
50  * flush_icache_range: Write any modified data cache blocks out to memory
51  * and invalidate the corresponding blocks in the instruction cache
52  *
53  * Generic code will call this after writing memory, before executing from it.
54  *
55  * @start: the start address
56  * @stop: the stop address (exclusive)
57  */
58 void flush_icache_range(unsigned long start, unsigned long stop)
59 {
60 	if (flush_coherent_icache())
61 		return;
62 
63 	clean_dcache_range(start, stop);
64 
65 	if (IS_ENABLED(CONFIG_44x)) {
66 		/*
67 		 * Flash invalidate on 44x because we are passed kmapped
68 		 * addresses and this doesn't work for userspace pages due to
69 		 * the virtually tagged icache.
70 		 */
71 		iccci((void *)start);
72 		mb(); /* sync */
73 		isync();
74 	} else
75 		invalidate_icache_range(start, stop);
76 }
77 EXPORT_SYMBOL(flush_icache_range);
78 
79 #ifdef CONFIG_HIGHMEM
80 /**
81  * flush_dcache_icache_phys() - Flush a page by it's physical address
82  * @physaddr: the physical address of the page
83  */
84 static void flush_dcache_icache_phys(unsigned long physaddr)
85 {
86 	unsigned long bytes = l1_dcache_bytes();
87 	unsigned long nb = PAGE_SIZE / bytes;
88 	unsigned long addr = physaddr & PAGE_MASK;
89 	unsigned long msr, msr0;
90 	unsigned long loop1 = addr, loop2 = addr;
91 
92 	msr0 = mfmsr();
93 	msr = msr0 & ~MSR_DR;
94 	/*
95 	 * This must remain as ASM to prevent potential memory accesses
96 	 * while the data MMU is disabled
97 	 */
98 	asm volatile(
99 		"   mtctr %2;\n"
100 		"   mtmsr %3;\n"
101 		"   isync;\n"
102 		"0: dcbst   0, %0;\n"
103 		"   addi    %0, %0, %4;\n"
104 		"   bdnz    0b;\n"
105 		"   sync;\n"
106 		"   mtctr %2;\n"
107 		"1: icbi    0, %1;\n"
108 		"   addi    %1, %1, %4;\n"
109 		"   bdnz    1b;\n"
110 		"   sync;\n"
111 		"   mtmsr %5;\n"
112 		"   isync;\n"
113 		: "+&r" (loop1), "+&r" (loop2)
114 		: "r" (nb), "r" (msr), "i" (bytes), "r" (msr0)
115 		: "ctr", "memory");
116 }
117 NOKPROBE_SYMBOL(flush_dcache_icache_phys)
118 #else
119 static void flush_dcache_icache_phys(unsigned long physaddr)
120 {
121 }
122 #endif
123 
124 /**
125  * __flush_dcache_icache(): Flush a particular page from the data cache to RAM.
126  * Note: this is necessary because the instruction cache does *not*
127  * snoop from the data cache.
128  *
129  * @p: the address of the page to flush
130  */
131 static void __flush_dcache_icache(void *p)
132 {
133 	unsigned long addr = (unsigned long)p & PAGE_MASK;
134 
135 	clean_dcache_range(addr, addr + PAGE_SIZE);
136 
137 	/*
138 	 * We don't flush the icache on 44x. Those have a virtual icache and we
139 	 * don't have access to the virtual address here (it's not the page
140 	 * vaddr but where it's mapped in user space). The flushing of the
141 	 * icache on these is handled elsewhere, when a change in the address
142 	 * space occurs, before returning to user space.
143 	 */
144 
145 	if (mmu_has_feature(MMU_FTR_TYPE_44x))
146 		return;
147 
148 	invalidate_icache_range(addr, addr + PAGE_SIZE);
149 }
150 
151 static void flush_dcache_icache_hugepage(struct page *page)
152 {
153 	int i;
154 	int nr = compound_nr(page);
155 
156 	if (!PageHighMem(page)) {
157 		for (i = 0; i < nr; i++)
158 			__flush_dcache_icache(lowmem_page_address(page + i));
159 	} else {
160 		for (i = 0; i < nr; i++) {
161 			void *start = kmap_local_page(page + i);
162 
163 			__flush_dcache_icache(start);
164 			kunmap_local(start);
165 		}
166 	}
167 }
168 
169 void flush_dcache_icache_page(struct page *page)
170 {
171 	if (flush_coherent_icache())
172 		return;
173 
174 	if (PageCompound(page))
175 		return flush_dcache_icache_hugepage(page);
176 
177 	if (!PageHighMem(page)) {
178 		__flush_dcache_icache(lowmem_page_address(page));
179 	} else if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
180 		void *start = kmap_local_page(page);
181 
182 		__flush_dcache_icache(start);
183 		kunmap_local(start);
184 	} else {
185 		flush_dcache_icache_phys(page_to_phys(page));
186 	}
187 }
188 EXPORT_SYMBOL(flush_dcache_icache_page);
189 
190 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
191 {
192 	clear_page(page);
193 
194 	/*
195 	 * We shouldn't have to do this, but some versions of glibc
196 	 * require it (ld.so assumes zero filled pages are icache clean)
197 	 * - Anton
198 	 */
199 	flush_dcache_page(pg);
200 }
201 EXPORT_SYMBOL(clear_user_page);
202 
203 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
204 		    struct page *pg)
205 {
206 	copy_page(vto, vfrom);
207 
208 	/*
209 	 * We should be able to use the following optimisation, however
210 	 * there are two problems.
211 	 * Firstly a bug in some versions of binutils meant PLT sections
212 	 * were not marked executable.
213 	 * Secondly the first word in the GOT section is blrl, used
214 	 * to establish the GOT address. Until recently the GOT was
215 	 * not marked executable.
216 	 * - Anton
217 	 */
218 #if 0
219 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
220 		return;
221 #endif
222 
223 	flush_dcache_page(pg);
224 }
225 
226 void flush_icache_user_page(struct vm_area_struct *vma, struct page *page,
227 			     unsigned long addr, int len)
228 {
229 	void *maddr;
230 
231 	maddr = kmap_local_page(page) + (addr & ~PAGE_MASK);
232 	flush_icache_range((unsigned long)maddr, (unsigned long)maddr + len);
233 	kunmap_local(maddr);
234 }
235