arch/x86/include/asm/mmu_context.h

   1 #ifndef _ASM_X86_MMU_CONTEXT_H
   2 #define _ASM_X86_MMU_CONTEXT_H
   3
   4 #include <asm/desc.h>
   5 #include <linux/atomic.h>
   6 #include <linux/mm_types.h>
   7 #include <linux/pkeys.h>
   8
   9 #include <trace/events/tlb.h>
  10
  11 #include <asm/pgalloc.h>
  12 #include <asm/tlbflush.h>
  13 #include <asm/paravirt.h>
  14 #include <asm/mpx.h>
  15
  16 extern atomic64_t last_mm_ctx_id;
  17
  18 #ifndef CONFIG_PARAVIRT
  19 static inline void paravirt_activate_mm(struct mm_struct *prev,
  20                                         struct mm_struct *next)
  21 {
  22 }
  23 #endif  /* !CONFIG_PARAVIRT */
  24
  25 #ifdef CONFIG_PERF_EVENTS
  26 extern struct static_key rdpmc_always_available;
  27
  28 static inline void load_mm_cr4(struct mm_struct *mm)
  29 {
  30         if (static_key_false(&rdpmc_always_available) ||
  31             atomic_read(&mm->context.perf_rdpmc_allowed))
  32                 cr4_set_bits(X86_CR4_PCE);
  33         else
  34                 cr4_clear_bits(X86_CR4_PCE);
  35 }
  36 #else
  37 static inline void load_mm_cr4(struct mm_struct *mm) {}
  38 #endif
  39
  40 #ifdef CONFIG_MODIFY_LDT_SYSCALL
  41 /*
  42  * ldt_structs can be allocated, used, and freed, but they are never
  43  * modified while live.
  44  */
  45 struct ldt_struct {
  46         /*
  47          * Xen requires page-aligned LDTs with special permissions.  This is
  48          * needed to prevent us from installing evil descriptors such as
  49          * call gates.  On native, we could merge the ldt_struct and LDT
  50          * allocations, but it's not worth trying to optimize.
  51          */
  52         struct desc_struct *entries;
  53         unsigned int nr_entries;
  54 };
  55
  56 /*
  57  * Used for LDT copy/destruction.
  58  */
  59 int init_new_context_ldt(struct task_struct *tsk, struct mm_struct *mm);
  60 void destroy_context_ldt(struct mm_struct *mm);
  61 #else   /* CONFIG_MODIFY_LDT_SYSCALL */
  62 static inline int init_new_context_ldt(struct task_struct *tsk,
  63                                        struct mm_struct *mm)
  64 {
  65         return 0;
  66 }
  67 static inline void destroy_context_ldt(struct mm_struct *mm) {}
  68 #endif
  69
  70 static inline void load_mm_ldt(struct mm_struct *mm)
  71 {
  72 #ifdef CONFIG_MODIFY_LDT_SYSCALL
  73         struct ldt_struct *ldt;
  74
  75         /* lockless_dereference synchronizes with smp_store_release */
  76         ldt = lockless_dereference(mm->context.ldt);
  77
  78         /*
  79          * Any change to mm->context.ldt is followed by an IPI to all
  80          * CPUs with the mm active.  The LDT will not be freed until
  81          * after the IPI is handled by all such CPUs.  This means that,
  82          * if the ldt_struct changes before we return, the values we see
  83          * will be safe, and the new values will be loaded before we run
  84          * any user code.
  85          *
  86          * NB: don't try to convert this to use RCU without extreme care.
  87          * We would still need IRQs off, because we don't want to change
  88          * the local LDT after an IPI loaded a newer value than the one
  89          * that we can see.
  90          */
  91
  92         if (unlikely(ldt))
  93                 set_ldt(ldt->entries, ldt->nr_entries);
  94         else
  95                 clear_LDT();
  96 #else
  97         clear_LDT();
  98 #endif
  99 }
 100
 101 static inline void switch_ldt(struct mm_struct *prev, struct mm_struct *next)
 102 {
 103 #ifdef CONFIG_MODIFY_LDT_SYSCALL
 104         /*
 105          * Load the LDT if either the old or new mm had an LDT.
 106          *
 107          * An mm will never go from having an LDT to not having an LDT.  Two
 108          * mms never share an LDT, so we don't gain anything by checking to
 109          * see whether the LDT changed.  There's also no guarantee that
 110          * prev->context.ldt actually matches LDTR, but, if LDTR is non-NULL,
 111          * then prev->context.ldt will also be non-NULL.
 112          *
 113          * If we really cared, we could optimize the case where prev == next
 114          * and we're exiting lazy mode.  Most of the time, if this happens,
 115          * we don't actually need to reload LDTR, but modify_ldt() is mostly
 116          * used by legacy code and emulators where we don't need this level of
 117          * performance.
 118          *
 119          * This uses | instead of || because it generates better code.
 120          */
 121         if (unlikely((unsigned long)prev->context.ldt |
 122                      (unsigned long)next->context.ldt))
 123                 load_mm_ldt(next);
 124 #endif
 125
 126         DEBUG_LOCKS_WARN_ON(preemptible());
 127 }
 128
 129 static inline void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
 130 {
 131         int cpu = smp_processor_id();
 132
 133         if (cpumask_test_cpu(cpu, mm_cpumask(mm)))
 134                 cpumask_clear_cpu(cpu, mm_cpumask(mm));
 135 }
 136
 137 static inline int init_new_context(struct task_struct *tsk,
 138                                    struct mm_struct *mm)
 139 {
 140         mm->context.ctx_id = atomic64_inc_return(&last_mm_ctx_id);
 141         atomic64_set(&mm->context.tlb_gen, 0);
 142
 143         #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 144         if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
 145                 /* pkey 0 is the default and always allocated */
 146                 mm->context.pkey_allocation_map = 0x1;
 147                 /* -1 means unallocated or invalid */
 148                 mm->context.execute_only_pkey = -1;
 149         }
 150         #endif
 151         return init_new_context_ldt(tsk, mm);
 152 }
 153 static inline void destroy_context(struct mm_struct *mm)
 154 {
 155         destroy_context_ldt(mm);
 156 }
 157
 158 extern void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 159                       struct task_struct *tsk);
 160
 161 extern void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
 162                                struct task_struct *tsk);
 163 #define switch_mm_irqs_off switch_mm_irqs_off
 164
 165 #define activate_mm(prev, next)                 \
 166 do {                                            \
 167         paravirt_activate_mm((prev), (next));   \
 168         switch_mm((prev), (next), NULL);        \
 169 } while (0);
 170
 171 #ifdef CONFIG_X86_32
 172 #define deactivate_mm(tsk, mm)                  \
 173 do {                                            \
 174         lazy_load_gs(0);                        \
 175 } while (0)
 176 #else
 177 #define deactivate_mm(tsk, mm)                  \
 178 do {                                            \
 179         load_gs_index(0);                       \
 180         loadsegment(fs, 0);                     \
 181 } while (0)
 182 #endif
 183
 184 static inline void arch_dup_mmap(struct mm_struct *oldmm,
 185                                  struct mm_struct *mm)
 186 {
 187         paravirt_arch_dup_mmap(oldmm, mm);
 188 }
 189
 190 static inline void arch_exit_mmap(struct mm_struct *mm)
 191 {
 192         paravirt_arch_exit_mmap(mm);
 193 }
 194
 195 #ifdef CONFIG_X86_64
 196 static inline bool is_64bit_mm(struct mm_struct *mm)
 197 {
 198         return  !IS_ENABLED(CONFIG_IA32_EMULATION) ||
 199                 !(mm->context.ia32_compat == TIF_IA32);
 200 }
 201 #else
 202 static inline bool is_64bit_mm(struct mm_struct *mm)
 203 {
 204         return false;
 205 }
 206 #endif
 207
 208 static inline void arch_bprm_mm_init(struct mm_struct *mm,
 209                 struct vm_area_struct *vma)
 210 {
 211         mpx_mm_init(mm);
 212 }
 213
 214 static inline void arch_unmap(struct mm_struct *mm, struct vm_area_struct *vma,
 215                               unsigned long start, unsigned long end)
 216 {
 217         /*
 218          * mpx_notify_unmap() goes and reads a rarely-hot
 219          * cacheline in the mm_struct.  That can be expensive
 220          * enough to be seen in profiles.
 221          *
 222          * The mpx_notify_unmap() call and its contents have been
 223          * observed to affect munmap() performance on hardware
 224          * where MPX is not present.
 225          *
 226          * The unlikely() optimizes for the fast case: no MPX
 227          * in the CPU, or no MPX use in the process.  Even if
 228          * we get this wrong (in the unlikely event that MPX
 229          * is widely enabled on some system) the overhead of
 230          * MPX itself (reading bounds tables) is expected to
 231          * overwhelm the overhead of getting this unlikely()
 232          * consistently wrong.
 233          */
 234         if (unlikely(cpu_feature_enabled(X86_FEATURE_MPX)))
 235                 mpx_notify_unmap(mm, vma, start, end);
 236 }
 237
 238 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 239 static inline int vma_pkey(struct vm_area_struct *vma)
 240 {
 241         unsigned long vma_pkey_mask = VM_PKEY_BIT0 | VM_PKEY_BIT1 |
 242                                       VM_PKEY_BIT2 | VM_PKEY_BIT3;
 243
 244         return (vma->vm_flags & vma_pkey_mask) >> VM_PKEY_SHIFT;
 245 }
 246 #else
 247 static inline int vma_pkey(struct vm_area_struct *vma)
 248 {
 249         return 0;
 250 }
 251 #endif
 252
 253 /*
 254  * We only want to enforce protection keys on the current process
 255  * because we effectively have no access to PKRU for other
 256  * processes or any way to tell *which * PKRU in a threaded
 257  * process we could use.
 258  *
 259  * So do not enforce things if the VMA is not from the current
 260  * mm, or if we are in a kernel thread.
 261  */
 262 static inline bool vma_is_foreign(struct vm_area_struct *vma)
 263 {
 264         if (!current->mm)
 265                 return true;
 266         /*
 267          * Should PKRU be enforced on the access to this VMA?  If
 268          * the VMA is from another process, then PKRU has no
 269          * relevance and should not be enforced.
 270          */
 271         if (current->mm != vma->vm_mm)
 272                 return true;
 273
 274         return false;
 275 }
 276
 277 static inline bool arch_vma_access_permitted(struct vm_area_struct *vma,
 278                 bool write, bool execute, bool foreign)
 279 {
 280         /* pkeys never affect instruction fetches */
 281         if (execute)
 282                 return true;
 283         /* allow access if the VMA is not one from this process */
 284         if (foreign || vma_is_foreign(vma))
 285                 return true;
 286         return __pkru_allows_pkey(vma_pkey(vma), write);
 287 }
 288
 289 /*
 290  * If PCID is on, ASID-aware code paths put the ASID+1 into the PCID
 291  * bits.  This serves two purposes.  It prevents a nasty situation in
 292  * which PCID-unaware code saves CR3, loads some other value (with PCID
 293  * == 0), and then restores CR3, thus corrupting the TLB for ASID 0 if
 294  * the saved ASID was nonzero.  It also means that any bugs involving
 295  * loading a PCID-enabled CR3 with CR4.PCIDE off will trigger
 296  * deterministically.
 297  */
 298
 299 static inline unsigned long build_cr3(struct mm_struct *mm, u16 asid)
 300 {
 301         if (static_cpu_has(X86_FEATURE_PCID)) {
 302                 VM_WARN_ON_ONCE(asid > 4094);
 303                 return __sme_pa(mm->pgd) | (asid + 1);
 304         } else {
 305                 VM_WARN_ON_ONCE(asid != 0);
 306                 return __sme_pa(mm->pgd);
 307         }
 308 }
 309
 310 static inline unsigned long build_cr3_noflush(struct mm_struct *mm, u16 asid)
 311 {
 312         VM_WARN_ON_ONCE(asid > 4094);
 313         return __sme_pa(mm->pgd) | (asid + 1) | CR3_NOFLUSH;
 314 }
 315
 316 /*
 317  * This can be used from process context to figure out what the value of
 318  * CR3 is without needing to do a (slow) __read_cr3().
 319  *
 320  * It's intended to be used for code like KVM that sneakily changes CR3
 321  * and needs to restore it.  It needs to be used very carefully.
 322  */
 323 static inline unsigned long __get_current_cr3_fast(void)
 324 {
 325         unsigned long cr3 = build_cr3(this_cpu_read(cpu_tlbstate.loaded_mm),
 326                 this_cpu_read(cpu_tlbstate.loaded_mm_asid));
 327
 328         /* For now, be very restrictive about when this can be called. */
 329         VM_WARN_ON(in_nmi() || preemptible());
 330
 331         VM_BUG_ON(cr3 != __read_cr3());
 332         return cr3;
 333 }
 334
 335 #endif /* _ASM_X86_MMU_CONTEXT_H */