.. | .. |
---|
9 | 9 | |
---|
10 | 10 | #include <trace/events/tlb.h> |
---|
11 | 11 | |
---|
12 | | -#include <asm/pgalloc.h> |
---|
13 | 12 | #include <asm/tlbflush.h> |
---|
14 | 13 | #include <asm/paravirt.h> |
---|
15 | | -#include <asm/mpx.h> |
---|
| 14 | +#include <asm/debugreg.h> |
---|
16 | 15 | |
---|
17 | 16 | extern atomic64_t last_mm_ctx_id; |
---|
18 | 17 | |
---|
19 | | -#ifndef CONFIG_PARAVIRT |
---|
| 18 | +#ifndef CONFIG_PARAVIRT_XXL |
---|
20 | 19 | static inline void paravirt_activate_mm(struct mm_struct *prev, |
---|
21 | 20 | struct mm_struct *next) |
---|
22 | 21 | { |
---|
23 | 22 | } |
---|
24 | | -#endif /* !CONFIG_PARAVIRT */ |
---|
| 23 | +#endif /* !CONFIG_PARAVIRT_XXL */ |
---|
25 | 24 | |
---|
26 | 25 | #ifdef CONFIG_PERF_EVENTS |
---|
27 | | - |
---|
| 26 | +DECLARE_STATIC_KEY_FALSE(rdpmc_never_available_key); |
---|
28 | 27 | DECLARE_STATIC_KEY_FALSE(rdpmc_always_available_key); |
---|
29 | | - |
---|
30 | | -static inline void load_mm_cr4(struct mm_struct *mm) |
---|
31 | | -{ |
---|
32 | | - if (static_branch_unlikely(&rdpmc_always_available_key) || |
---|
33 | | - atomic_read(&mm->context.perf_rdpmc_allowed)) |
---|
34 | | - cr4_set_bits(X86_CR4_PCE); |
---|
35 | | - else |
---|
36 | | - cr4_clear_bits(X86_CR4_PCE); |
---|
37 | | -} |
---|
38 | | -#else |
---|
39 | | -static inline void load_mm_cr4(struct mm_struct *mm) {} |
---|
| 28 | +void cr4_update_pce(void *ignored); |
---|
40 | 29 | #endif |
---|
41 | 30 | |
---|
42 | 31 | #ifdef CONFIG_MODIFY_LDT_SYSCALL |
---|
.. | .. |
---|
66 | 55 | int slot; |
---|
67 | 56 | }; |
---|
68 | 57 | |
---|
69 | | -/* This is a multiple of PAGE_SIZE. */ |
---|
70 | | -#define LDT_SLOT_STRIDE (LDT_ENTRIES * LDT_ENTRY_SIZE) |
---|
71 | | - |
---|
72 | | -static inline void *ldt_slot_va(int slot) |
---|
73 | | -{ |
---|
74 | | - return (void *)(LDT_BASE_ADDR + LDT_SLOT_STRIDE * slot); |
---|
75 | | -} |
---|
76 | | - |
---|
77 | 58 | /* |
---|
78 | 59 | * Used for LDT copy/destruction. |
---|
79 | 60 | */ |
---|
.. | .. |
---|
96 | 77 | static inline void ldt_arch_exit_mmap(struct mm_struct *mm) { } |
---|
97 | 78 | #endif |
---|
98 | 79 | |
---|
| 80 | +#ifdef CONFIG_MODIFY_LDT_SYSCALL |
---|
| 81 | +extern void load_mm_ldt(struct mm_struct *mm); |
---|
| 82 | +extern void switch_ldt(struct mm_struct *prev, struct mm_struct *next); |
---|
| 83 | +#else |
---|
99 | 84 | static inline void load_mm_ldt(struct mm_struct *mm) |
---|
100 | 85 | { |
---|
101 | | -#ifdef CONFIG_MODIFY_LDT_SYSCALL |
---|
102 | | - struct ldt_struct *ldt; |
---|
103 | | - |
---|
104 | | - /* READ_ONCE synchronizes with smp_store_release */ |
---|
105 | | - ldt = READ_ONCE(mm->context.ldt); |
---|
106 | | - |
---|
107 | | - /* |
---|
108 | | - * Any change to mm->context.ldt is followed by an IPI to all |
---|
109 | | - * CPUs with the mm active. The LDT will not be freed until |
---|
110 | | - * after the IPI is handled by all such CPUs. This means that, |
---|
111 | | - * if the ldt_struct changes before we return, the values we see |
---|
112 | | - * will be safe, and the new values will be loaded before we run |
---|
113 | | - * any user code. |
---|
114 | | - * |
---|
115 | | - * NB: don't try to convert this to use RCU without extreme care. |
---|
116 | | - * We would still need IRQs off, because we don't want to change |
---|
117 | | - * the local LDT after an IPI loaded a newer value than the one |
---|
118 | | - * that we can see. |
---|
119 | | - */ |
---|
120 | | - |
---|
121 | | - if (unlikely(ldt)) { |
---|
122 | | - if (static_cpu_has(X86_FEATURE_PTI)) { |
---|
123 | | - if (WARN_ON_ONCE((unsigned long)ldt->slot > 1)) { |
---|
124 | | - /* |
---|
125 | | - * Whoops -- either the new LDT isn't mapped |
---|
126 | | - * (if slot == -1) or is mapped into a bogus |
---|
127 | | - * slot (if slot > 1). |
---|
128 | | - */ |
---|
129 | | - clear_LDT(); |
---|
130 | | - return; |
---|
131 | | - } |
---|
132 | | - |
---|
133 | | - /* |
---|
134 | | - * If page table isolation is enabled, ldt->entries |
---|
135 | | - * will not be mapped in the userspace pagetables. |
---|
136 | | - * Tell the CPU to access the LDT through the alias |
---|
137 | | - * at ldt_slot_va(ldt->slot). |
---|
138 | | - */ |
---|
139 | | - set_ldt(ldt_slot_va(ldt->slot), ldt->nr_entries); |
---|
140 | | - } else { |
---|
141 | | - set_ldt(ldt->entries, ldt->nr_entries); |
---|
142 | | - } |
---|
143 | | - } else { |
---|
144 | | - clear_LDT(); |
---|
145 | | - } |
---|
146 | | -#else |
---|
147 | 86 | clear_LDT(); |
---|
148 | | -#endif |
---|
149 | 87 | } |
---|
150 | | - |
---|
151 | 88 | static inline void switch_ldt(struct mm_struct *prev, struct mm_struct *next) |
---|
152 | 89 | { |
---|
153 | | -#ifdef CONFIG_MODIFY_LDT_SYSCALL |
---|
154 | | - /* |
---|
155 | | - * Load the LDT if either the old or new mm had an LDT. |
---|
156 | | - * |
---|
157 | | - * An mm will never go from having an LDT to not having an LDT. Two |
---|
158 | | - * mms never share an LDT, so we don't gain anything by checking to |
---|
159 | | - * see whether the LDT changed. There's also no guarantee that |
---|
160 | | - * prev->context.ldt actually matches LDTR, but, if LDTR is non-NULL, |
---|
161 | | - * then prev->context.ldt will also be non-NULL. |
---|
162 | | - * |
---|
163 | | - * If we really cared, we could optimize the case where prev == next |
---|
164 | | - * and we're exiting lazy mode. Most of the time, if this happens, |
---|
165 | | - * we don't actually need to reload LDTR, but modify_ldt() is mostly |
---|
166 | | - * used by legacy code and emulators where we don't need this level of |
---|
167 | | - * performance. |
---|
168 | | - * |
---|
169 | | - * This uses | instead of || because it generates better code. |
---|
170 | | - */ |
---|
171 | | - if (unlikely((unsigned long)prev->context.ldt | |
---|
172 | | - (unsigned long)next->context.ldt)) |
---|
173 | | - load_mm_ldt(next); |
---|
174 | | -#endif |
---|
175 | | - |
---|
176 | 90 | DEBUG_LOCKS_WARN_ON(preemptible()); |
---|
177 | 91 | } |
---|
| 92 | +#endif |
---|
178 | 93 | |
---|
179 | | -void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk); |
---|
| 94 | +extern void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk); |
---|
180 | 95 | |
---|
181 | 96 | /* |
---|
182 | 97 | * Init a new mm. Used on mm copies, like at fork() |
---|
.. | .. |
---|
271 | 186 | } |
---|
272 | 187 | #endif |
---|
273 | 188 | |
---|
274 | | -static inline void arch_bprm_mm_init(struct mm_struct *mm, |
---|
275 | | - struct vm_area_struct *vma) |
---|
| 189 | +static inline void arch_unmap(struct mm_struct *mm, unsigned long start, |
---|
| 190 | + unsigned long end) |
---|
276 | 191 | { |
---|
277 | | - mpx_mm_init(mm); |
---|
278 | | -} |
---|
279 | | - |
---|
280 | | -static inline void arch_unmap(struct mm_struct *mm, struct vm_area_struct *vma, |
---|
281 | | - unsigned long start, unsigned long end) |
---|
282 | | -{ |
---|
283 | | - /* |
---|
284 | | - * mpx_notify_unmap() goes and reads a rarely-hot |
---|
285 | | - * cacheline in the mm_struct. That can be expensive |
---|
286 | | - * enough to be seen in profiles. |
---|
287 | | - * |
---|
288 | | - * The mpx_notify_unmap() call and its contents have been |
---|
289 | | - * observed to affect munmap() performance on hardware |
---|
290 | | - * where MPX is not present. |
---|
291 | | - * |
---|
292 | | - * The unlikely() optimizes for the fast case: no MPX |
---|
293 | | - * in the CPU, or no MPX use in the process. Even if |
---|
294 | | - * we get this wrong (in the unlikely event that MPX |
---|
295 | | - * is widely enabled on some system) the overhead of |
---|
296 | | - * MPX itself (reading bounds tables) is expected to |
---|
297 | | - * overwhelm the overhead of getting this unlikely() |
---|
298 | | - * consistently wrong. |
---|
299 | | - */ |
---|
300 | | - if (unlikely(cpu_feature_enabled(X86_FEATURE_MPX))) |
---|
301 | | - mpx_notify_unmap(mm, vma, start, end); |
---|
302 | 192 | } |
---|
303 | 193 | |
---|
304 | 194 | /* |
---|
.. | .. |
---|
310 | 200 | * So do not enforce things if the VMA is not from the current |
---|
311 | 201 | * mm, or if we are in a kernel thread. |
---|
312 | 202 | */ |
---|
313 | | -static inline bool vma_is_foreign(struct vm_area_struct *vma) |
---|
314 | | -{ |
---|
315 | | - if (!current->mm) |
---|
316 | | - return true; |
---|
317 | | - /* |
---|
318 | | - * Should PKRU be enforced on the access to this VMA? If |
---|
319 | | - * the VMA is from another process, then PKRU has no |
---|
320 | | - * relevance and should not be enforced. |
---|
321 | | - */ |
---|
322 | | - if (current->mm != vma->vm_mm) |
---|
323 | | - return true; |
---|
324 | | - |
---|
325 | | - return false; |
---|
326 | | -} |
---|
327 | | - |
---|
328 | 203 | static inline bool arch_vma_access_permitted(struct vm_area_struct *vma, |
---|
329 | 204 | bool write, bool execute, bool foreign) |
---|
330 | 205 | { |
---|
.. | .. |
---|
337 | 212 | return __pkru_allows_pkey(vma_pkey(vma), write); |
---|
338 | 213 | } |
---|
339 | 214 | |
---|
340 | | -/* |
---|
341 | | - * This can be used from process context to figure out what the value of |
---|
342 | | - * CR3 is without needing to do a (slow) __read_cr3(). |
---|
343 | | - * |
---|
344 | | - * It's intended to be used for code like KVM that sneakily changes CR3 |
---|
345 | | - * and needs to restore it. It needs to be used very carefully. |
---|
346 | | - */ |
---|
347 | | -static inline unsigned long __get_current_cr3_fast(void) |
---|
348 | | -{ |
---|
349 | | - unsigned long cr3 = build_cr3(this_cpu_read(cpu_tlbstate.loaded_mm)->pgd, |
---|
350 | | - this_cpu_read(cpu_tlbstate.loaded_mm_asid)); |
---|
351 | | - |
---|
352 | | - /* For now, be very restrictive about when this can be called. */ |
---|
353 | | - VM_WARN_ON(in_nmi() || preemptible()); |
---|
354 | | - |
---|
355 | | - VM_BUG_ON(cr3 != __read_cr3()); |
---|
356 | | - return cr3; |
---|
357 | | -} |
---|
| 215 | +unsigned long __get_current_cr3_fast(void); |
---|
358 | 216 | |
---|
359 | 217 | #endif /* _ASM_X86_MMU_CONTEXT_H */ |
---|