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