]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/include/asm/tlbflush.h
x86/mm: Remove the UP asm/tlbflush.h code, always use the (formerly) SMP code
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / include / asm / tlbflush.h
1 #ifndef _ASM_X86_TLBFLUSH_H
2 #define _ASM_X86_TLBFLUSH_H
3
4 #include <linux/mm.h>
5 #include <linux/sched.h>
6
7 #include <asm/processor.h>
8 #include <asm/cpufeature.h>
9 #include <asm/special_insns.h>
10 #include <asm/smp.h>
11
12 static inline void __invpcid(unsigned long pcid, unsigned long addr,
13 unsigned long type)
14 {
15 struct { u64 d[2]; } desc = { { pcid, addr } };
16
17 /*
18 * The memory clobber is because the whole point is to invalidate
19 * stale TLB entries and, especially if we're flushing global
20 * mappings, we don't want the compiler to reorder any subsequent
21 * memory accesses before the TLB flush.
22 *
23 * The hex opcode is invpcid (%ecx), %eax in 32-bit mode and
24 * invpcid (%rcx), %rax in long mode.
25 */
26 asm volatile (".byte 0x66, 0x0f, 0x38, 0x82, 0x01"
27 : : "m" (desc), "a" (type), "c" (&desc) : "memory");
28 }
29
30 #define INVPCID_TYPE_INDIV_ADDR 0
31 #define INVPCID_TYPE_SINGLE_CTXT 1
32 #define INVPCID_TYPE_ALL_INCL_GLOBAL 2
33 #define INVPCID_TYPE_ALL_NON_GLOBAL 3
34
35 /* Flush all mappings for a given pcid and addr, not including globals. */
36 static inline void invpcid_flush_one(unsigned long pcid,
37 unsigned long addr)
38 {
39 __invpcid(pcid, addr, INVPCID_TYPE_INDIV_ADDR);
40 }
41
42 /* Flush all mappings for a given PCID, not including globals. */
43 static inline void invpcid_flush_single_context(unsigned long pcid)
44 {
45 __invpcid(pcid, 0, INVPCID_TYPE_SINGLE_CTXT);
46 }
47
48 /* Flush all mappings, including globals, for all PCIDs. */
49 static inline void invpcid_flush_all(void)
50 {
51 __invpcid(0, 0, INVPCID_TYPE_ALL_INCL_GLOBAL);
52 }
53
54 /* Flush all mappings for all PCIDs except globals. */
55 static inline void invpcid_flush_all_nonglobals(void)
56 {
57 __invpcid(0, 0, INVPCID_TYPE_ALL_NON_GLOBAL);
58 }
59
60 #ifdef CONFIG_PARAVIRT
61 #include <asm/paravirt.h>
62 #else
63 #define __flush_tlb() __native_flush_tlb()
64 #define __flush_tlb_global() __native_flush_tlb_global()
65 #define __flush_tlb_single(addr) __native_flush_tlb_single(addr)
66 #endif
67
68 struct tlb_state {
69 struct mm_struct *active_mm;
70 int state;
71
72 /*
73 * Access to this CR4 shadow and to H/W CR4 is protected by
74 * disabling interrupts when modifying either one.
75 */
76 unsigned long cr4;
77 };
78 DECLARE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate);
79
80 /* Initialize cr4 shadow for this CPU. */
81 static inline void cr4_init_shadow(void)
82 {
83 this_cpu_write(cpu_tlbstate.cr4, __read_cr4());
84 }
85
86 /* Set in this cpu's CR4. */
87 static inline void cr4_set_bits(unsigned long mask)
88 {
89 unsigned long cr4;
90
91 cr4 = this_cpu_read(cpu_tlbstate.cr4);
92 if ((cr4 | mask) != cr4) {
93 cr4 |= mask;
94 this_cpu_write(cpu_tlbstate.cr4, cr4);
95 __write_cr4(cr4);
96 }
97 }
98
99 /* Clear in this cpu's CR4. */
100 static inline void cr4_clear_bits(unsigned long mask)
101 {
102 unsigned long cr4;
103
104 cr4 = this_cpu_read(cpu_tlbstate.cr4);
105 if ((cr4 & ~mask) != cr4) {
106 cr4 &= ~mask;
107 this_cpu_write(cpu_tlbstate.cr4, cr4);
108 __write_cr4(cr4);
109 }
110 }
111
112 static inline void cr4_toggle_bits(unsigned long mask)
113 {
114 unsigned long cr4;
115
116 cr4 = this_cpu_read(cpu_tlbstate.cr4);
117 cr4 ^= mask;
118 this_cpu_write(cpu_tlbstate.cr4, cr4);
119 __write_cr4(cr4);
120 }
121
122 /* Read the CR4 shadow. */
123 static inline unsigned long cr4_read_shadow(void)
124 {
125 return this_cpu_read(cpu_tlbstate.cr4);
126 }
127
128 /*
129 * Save some of cr4 feature set we're using (e.g. Pentium 4MB
130 * enable and PPro Global page enable), so that any CPU's that boot
131 * up after us can get the correct flags. This should only be used
132 * during boot on the boot cpu.
133 */
134 extern unsigned long mmu_cr4_features;
135 extern u32 *trampoline_cr4_features;
136
137 static inline void cr4_set_bits_and_update_boot(unsigned long mask)
138 {
139 mmu_cr4_features |= mask;
140 if (trampoline_cr4_features)
141 *trampoline_cr4_features = mmu_cr4_features;
142 cr4_set_bits(mask);
143 }
144
145 static inline void __native_flush_tlb(void)
146 {
147 /*
148 * If current->mm == NULL then we borrow a mm which may change during a
149 * task switch and therefore we must not be preempted while we write CR3
150 * back:
151 */
152 preempt_disable();
153 native_write_cr3(native_read_cr3());
154 preempt_enable();
155 }
156
157 static inline void __native_flush_tlb_global_irq_disabled(void)
158 {
159 unsigned long cr4;
160
161 cr4 = this_cpu_read(cpu_tlbstate.cr4);
162 /* clear PGE */
163 native_write_cr4(cr4 & ~X86_CR4_PGE);
164 /* write old PGE again and flush TLBs */
165 native_write_cr4(cr4);
166 }
167
168 static inline void __native_flush_tlb_global(void)
169 {
170 unsigned long flags;
171
172 if (static_cpu_has(X86_FEATURE_INVPCID)) {
173 /*
174 * Using INVPCID is considerably faster than a pair of writes
175 * to CR4 sandwiched inside an IRQ flag save/restore.
176 */
177 invpcid_flush_all();
178 return;
179 }
180
181 /*
182 * Read-modify-write to CR4 - protect it from preemption and
183 * from interrupts. (Use the raw variant because this code can
184 * be called from deep inside debugging code.)
185 */
186 raw_local_irq_save(flags);
187
188 __native_flush_tlb_global_irq_disabled();
189
190 raw_local_irq_restore(flags);
191 }
192
193 static inline void __native_flush_tlb_single(unsigned long addr)
194 {
195 asm volatile("invlpg (%0)" ::"r" (addr) : "memory");
196 }
197
198 static inline void __flush_tlb_all(void)
199 {
200 if (boot_cpu_has(X86_FEATURE_PGE))
201 __flush_tlb_global();
202 else
203 __flush_tlb();
204 }
205
206 static inline void __flush_tlb_one(unsigned long addr)
207 {
208 count_vm_tlb_event(NR_TLB_LOCAL_FLUSH_ONE);
209 __flush_tlb_single(addr);
210 }
211
212 #define TLB_FLUSH_ALL -1UL
213
214 /*
215 * TLB flushing:
216 *
217 * - flush_tlb_all() flushes all processes TLBs
218 * - flush_tlb_mm(mm) flushes the specified mm context TLB's
219 * - flush_tlb_page(vma, vmaddr) flushes one page
220 * - flush_tlb_range(vma, start, end) flushes a range of pages
221 * - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
222 * - flush_tlb_others(cpumask, info) flushes TLBs on other cpus
223 *
224 * ..but the i386 has somewhat limited tlb flushing capabilities,
225 * and page-granular flushes are available only on i486 and up.
226 */
227 struct flush_tlb_info {
228 struct mm_struct *mm;
229 unsigned long start;
230 unsigned long end;
231 };
232
233 #define local_flush_tlb() __flush_tlb()
234
235 #define flush_tlb_mm(mm) flush_tlb_mm_range(mm, 0UL, TLB_FLUSH_ALL, 0UL)
236
237 #define flush_tlb_range(vma, start, end) \
238 flush_tlb_mm_range(vma->vm_mm, start, end, vma->vm_flags)
239
240 extern void flush_tlb_all(void);
241 extern void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
242 unsigned long end, unsigned long vmflag);
243 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
244
245 static inline void flush_tlb_page(struct vm_area_struct *vma, unsigned long a)
246 {
247 flush_tlb_mm_range(vma->vm_mm, a, a + PAGE_SIZE, VM_NONE);
248 }
249
250 void native_flush_tlb_others(const struct cpumask *cpumask,
251 const struct flush_tlb_info *info);
252
253 #define TLBSTATE_OK 1
254 #define TLBSTATE_LAZY 2
255
256 static inline void reset_lazy_tlbstate(void)
257 {
258 this_cpu_write(cpu_tlbstate.state, 0);
259 this_cpu_write(cpu_tlbstate.active_mm, &init_mm);
260 }
261
262 static inline void arch_tlbbatch_add_mm(struct arch_tlbflush_unmap_batch *batch,
263 struct mm_struct *mm)
264 {
265 cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
266 }
267
268 extern void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch);
269
270 #ifndef CONFIG_PARAVIRT
271 #define flush_tlb_others(mask, info) \
272 native_flush_tlb_others(mask, info)
273 #endif
274
275 #endif /* _ASM_X86_TLBFLUSH_H */