arch/x86/include/asm/mmu_context.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _ASM_X86_MMU_CONTEXT_H
   3 #define _ASM_X86_MMU_CONTEXT_H
   4
   5 #include <asm/desc.h>
   6 #include <linux/atomic.h>
   7 #include <linux/mm_types.h>
   8 #include <linux/pkeys.h>
   9
  10 #include <trace/events/tlb.h>
  11
  12 #include <asm/pgalloc.h>
  13 #include <asm/tlbflush.h>
  14 #include <asm/paravirt.h>
  15 #include <asm/mpx.h>
  16
  17 extern atomic64_t last_mm_ctx_id;
  18
  19 #ifndef CONFIG_PARAVIRT
  20 static inline void paravirt_activate_mm(struct mm_struct *prev,
  21                                         struct mm_struct *next)
  22 {
  23 }
  24 #endif  /* !CONFIG_PARAVIRT */
  25
  26 #ifdef CONFIG_PERF_EVENTS
  27 extern struct static_key rdpmc_always_available;
  28
  29 static inline void load_mm_cr4(struct mm_struct *mm)
  30 {
  31         if (static_key_false(&rdpmc_always_available) ||
  32             atomic_read(&mm->context.perf_rdpmc_allowed))
  33                 cr4_set_bits(X86_CR4_PCE);
  34         else
  35                 cr4_clear_bits(X86_CR4_PCE);
  36 }
  37 #else
  38 static inline void load_mm_cr4(struct mm_struct *mm) {}
  39 #endif
  40
  41 #ifdef CONFIG_MODIFY_LDT_SYSCALL
  42 /*
  43  * ldt_structs can be allocated, used, and freed, but they are never
  44  * modified while live.
  45  */
  46 struct ldt_struct {
  47         /*
  48          * Xen requires page-aligned LDTs with special permissions.  This is
  49          * needed to prevent us from installing evil descriptors such as
  50          * call gates.  On native, we could merge the ldt_struct and LDT
  51          * allocations, but it's not worth trying to optimize.
  52          */
  53         struct desc_struct      *entries;
  54         unsigned int            nr_entries;
  55
  56         /*
  57          * If PTI is in use, then the entries array is not mapped while we're
  58          * in user mode.  The whole array will be aliased at the addressed
  59          * given by ldt_slot_va(slot).  We use two slots so that we can allocate
  60          * and map, and enable a new LDT without invalidating the mapping
  61          * of an older, still-in-use LDT.
  62          *
  63          * slot will be -1 if this LDT doesn't have an alias mapping.
  64          */
  65         int                     slot;
  66 };
  67
  68 /* This is a multiple of PAGE_SIZE. */
  69 #define LDT_SLOT_STRIDE (LDT_ENTRIES * LDT_ENTRY_SIZE)
  70
  71 static inline void *ldt_slot_va(int slot)
  72 {
  73         return (void *)(LDT_BASE_ADDR + LDT_SLOT_STRIDE * slot);
  74 }
  75
  76 /*
  77  * Used for LDT copy/destruction.
  78  */
  79 static inline void init_new_context_ldt(struct mm_struct *mm)
  80 {
  81         mm->context.ldt = NULL;
  82         init_rwsem(&mm->context.ldt_usr_sem);
  83 }
  84 int ldt_dup_context(struct mm_struct *oldmm, struct mm_struct *mm);
  85 void destroy_context_ldt(struct mm_struct *mm);
  86 void ldt_arch_exit_mmap(struct mm_struct *mm);
  87 #else   /* CONFIG_MODIFY_LDT_SYSCALL */
  88 static inline void init_new_context_ldt(struct mm_struct *mm) { }
  89 static inline int ldt_dup_context(struct mm_struct *oldmm,
  90                                   struct mm_struct *mm)
  91 {
  92         return 0;
  93 }
  94 static inline void destroy_context_ldt(struct mm_struct *mm) { }
  95 static inline void ldt_arch_exit_mmap(struct mm_struct *mm) { }
  96 #endif
  97
  98 static inline void load_mm_ldt(struct mm_struct *mm)
  99 {
 100 #ifdef CONFIG_MODIFY_LDT_SYSCALL
 101         struct ldt_struct *ldt;
 102
 103         /* READ_ONCE synchronizes with smp_store_release */
 104         ldt = READ_ONCE(mm->context.ldt);
 105
 106         /*
 107          * Any change to mm->context.ldt is followed by an IPI to all
 108          * CPUs with the mm active.  The LDT will not be freed until
 109          * after the IPI is handled by all such CPUs.  This means that,
 110          * if the ldt_struct changes before we return, the values we see
 111          * will be safe, and the new values will be loaded before we run
 112          * any user code.
 113          *
 114          * NB: don't try to convert this to use RCU without extreme care.
 115          * We would still need IRQs off, because we don't want to change
 116          * the local LDT after an IPI loaded a newer value than the one
 117          * that we can see.
 118          */
 119
 120         if (unlikely(ldt)) {
 121                 if (static_cpu_has(X86_FEATURE_PTI)) {
 122                         if (WARN_ON_ONCE((unsigned long)ldt->slot > 1)) {
 123                                 /*
 124                                  * Whoops -- either the new LDT isn't mapped
 125                                  * (if slot == -1) or is mapped into a bogus
 126                                  * slot (if slot > 1).
 127                                  */
 128                                 clear_LDT();
 129                                 return;
 130                         }
 131
 132                         /*
 133                          * If page table isolation is enabled, ldt->entries
 134                          * will not be mapped in the userspace pagetables.
 135                          * Tell the CPU to access the LDT through the alias
 136                          * at ldt_slot_va(ldt->slot).
 137                          */
 138                         set_ldt(ldt_slot_va(ldt->slot), ldt->nr_entries);
 139                 } else {
 140                         set_ldt(ldt->entries, ldt->nr_entries);
 141                 }
 142         } else {
 143                 clear_LDT();
 144         }
 145 #else
 146         clear_LDT();
 147 #endif
 148 }
 149
 150 static inline void switch_ldt(struct mm_struct *prev, struct mm_struct *next)
 151 {
 152 #ifdef CONFIG_MODIFY_LDT_SYSCALL
 153         /*
 154          * Load the LDT if either the old or new mm had an LDT.
 155          *
 156          * An mm will never go from having an LDT to not having an LDT.  Two
 157          * mms never share an LDT, so we don't gain anything by checking to
 158          * see whether the LDT changed.  There's also no guarantee that
 159          * prev->context.ldt actually matches LDTR, but, if LDTR is non-NULL,
 160          * then prev->context.ldt will also be non-NULL.
 161          *
 162          * If we really cared, we could optimize the case where prev == next
 163          * and we're exiting lazy mode.  Most of the time, if this happens,
 164          * we don't actually need to reload LDTR, but modify_ldt() is mostly
 165          * used by legacy code and emulators where we don't need this level of
 166          * performance.
 167          *
 168          * This uses | instead of || because it generates better code.
 169          */
 170         if (unlikely((unsigned long)prev->context.ldt |
 171                      (unsigned long)next->context.ldt))
 172                 load_mm_ldt(next);
 173 #endif
 174
 175         DEBUG_LOCKS_WARN_ON(preemptible());
 176 }
 177
 178 void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk);
 179
 180 static inline int init_new_context(struct task_struct *tsk,
 181                                    struct mm_struct *mm)
 182 {
 183         mutex_init(&mm->context.lock);
 184
 185         mm->context.ctx_id = atomic64_inc_return(&last_mm_ctx_id);
 186         atomic64_set(&mm->context.tlb_gen, 0);
 187
 188 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 189         if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
 190                 /* pkey 0 is the default and allocated implicitly */
 191                 mm->context.pkey_allocation_map = 0x1;
 192                 /* -1 means unallocated or invalid */
 193                 mm->context.execute_only_pkey = -1;
 194         }
 195 #endif
 196         init_new_context_ldt(mm);
 197         return 0;
 198 }
 199 static inline void destroy_context(struct mm_struct *mm)
 200 {
 201         destroy_context_ldt(mm);
 202 }
 203
 204 extern void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 205                       struct task_struct *tsk);
 206
 207 extern void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
 208                                struct task_struct *tsk);
 209 #define switch_mm_irqs_off switch_mm_irqs_off
 210
 211 #define activate_mm(prev, next)                 \
 212 do {                                            \
 213         paravirt_activate_mm((prev), (next));   \
 214         switch_mm((prev), (next), NULL);        \
 215 } while (0);
 216
 217 #ifdef CONFIG_X86_32
 218 #define deactivate_mm(tsk, mm)                  \
 219 do {                                            \
 220         lazy_load_gs(0);                        \
 221 } while (0)
 222 #else
 223 #define deactivate_mm(tsk, mm)                  \
 224 do {                                            \
 225         load_gs_index(0);                       \
 226         loadsegment(fs, 0);                     \
 227 } while (0)
 228 #endif
 229
 230 static inline int arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
 231 {
 232         paravirt_arch_dup_mmap(oldmm, mm);
 233         return ldt_dup_context(oldmm, mm);
 234 }
 235
 236 static inline void arch_exit_mmap(struct mm_struct *mm)
 237 {
 238         paravirt_arch_exit_mmap(mm);
 239         ldt_arch_exit_mmap(mm);
 240 }
 241
 242 #ifdef CONFIG_X86_64
 243 static inline bool is_64bit_mm(struct mm_struct *mm)
 244 {
 245         return  !IS_ENABLED(CONFIG_IA32_EMULATION) ||
 246                 !(mm->context.ia32_compat == TIF_IA32);
 247 }
 248 #else
 249 static inline bool is_64bit_mm(struct mm_struct *mm)
 250 {
 251         return false;
 252 }
 253 #endif
 254
 255 static inline void arch_bprm_mm_init(struct mm_struct *mm,
 256                 struct vm_area_struct *vma)
 257 {
 258         mpx_mm_init(mm);
 259 }
 260
 261 static inline void arch_unmap(struct mm_struct *mm, struct vm_area_struct *vma,
 262                               unsigned long start, unsigned long end)
 263 {
 264         /*
 265          * mpx_notify_unmap() goes and reads a rarely-hot
 266          * cacheline in the mm_struct.  That can be expensive
 267          * enough to be seen in profiles.
 268          *
 269          * The mpx_notify_unmap() call and its contents have been
 270          * observed to affect munmap() performance on hardware
 271          * where MPX is not present.
 272          *
 273          * The unlikely() optimizes for the fast case: no MPX
 274          * in the CPU, or no MPX use in the process.  Even if
 275          * we get this wrong (in the unlikely event that MPX
 276          * is widely enabled on some system) the overhead of
 277          * MPX itself (reading bounds tables) is expected to
 278          * overwhelm the overhead of getting this unlikely()
 279          * consistently wrong.
 280          */
 281         if (unlikely(cpu_feature_enabled(X86_FEATURE_MPX)))
 282                 mpx_notify_unmap(mm, vma, start, end);
 283 }
 284
 285 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 286 static inline int vma_pkey(struct vm_area_struct *vma)
 287 {
 288         unsigned long vma_pkey_mask = VM_PKEY_BIT0 | VM_PKEY_BIT1 |
 289                                       VM_PKEY_BIT2 | VM_PKEY_BIT3;
 290
 291         return (vma->vm_flags & vma_pkey_mask) >> VM_PKEY_SHIFT;
 292 }
 293 #else
 294 static inline int vma_pkey(struct vm_area_struct *vma)
 295 {
 296         return 0;
 297 }
 298 #endif
 299
 300 /*
 301  * We only want to enforce protection keys on the current process
 302  * because we effectively have no access to PKRU for other
 303  * processes or any way to tell *which * PKRU in a threaded
 304  * process we could use.
 305  *
 306  * So do not enforce things if the VMA is not from the current
 307  * mm, or if we are in a kernel thread.
 308  */
 309 static inline bool vma_is_foreign(struct vm_area_struct *vma)
 310 {
 311         if (!current->mm)
 312                 return true;
 313         /*
 314          * Should PKRU be enforced on the access to this VMA?  If
 315          * the VMA is from another process, then PKRU has no
 316          * relevance and should not be enforced.
 317          */
 318         if (current->mm != vma->vm_mm)
 319                 return true;
 320
 321         return false;
 322 }
 323
 324 static inline bool arch_vma_access_permitted(struct vm_area_struct *vma,
 325                 bool write, bool execute, bool foreign)
 326 {
 327         /* pkeys never affect instruction fetches */
 328         if (execute)
 329                 return true;
 330         /* allow access if the VMA is not one from this process */
 331         if (foreign || vma_is_foreign(vma))
 332                 return true;
 333         return __pkru_allows_pkey(vma_pkey(vma), write);
 334 }
 335
 336 /*
 337  * This can be used from process context to figure out what the value of
 338  * CR3 is without needing to do a (slow) __read_cr3().
 339  *
 340  * It's intended to be used for code like KVM that sneakily changes CR3
 341  * and needs to restore it.  It needs to be used very carefully.
 342  */
 343 static inline unsigned long __get_current_cr3_fast(void)
 344 {
 345         unsigned long cr3 = build_cr3(this_cpu_read(cpu_tlbstate.loaded_mm)->pgd,
 346                 this_cpu_read(cpu_tlbstate.loaded_mm_asid));
 347
 348         /* For now, be very restrictive about when this can be called. */
 349         VM_WARN_ON(in_nmi() || preemptible());
 350
 351         VM_BUG_ON(cr3 != __read_cr3());
 352         return cr3;
 353 }
 354
 355 #endif /* _ASM_X86_MMU_CONTEXT_H */