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/debugreg.h>
  16
  17 extern atomic64_t last_mm_ctx_id;
  18
  19 #ifndef CONFIG_PARAVIRT_XXL
  20 static inline void paravirt_activate_mm(struct mm_struct *prev,
  21                                         struct mm_struct *next)
  22 {
  23 }
  24 #endif  /* !CONFIG_PARAVIRT_XXL */
  25
  26 #ifdef CONFIG_PERF_EVENTS
  27
  28 DECLARE_STATIC_KEY_FALSE(rdpmc_never_available_key);
  29 DECLARE_STATIC_KEY_FALSE(rdpmc_always_available_key);
  30
  31 static inline void load_mm_cr4_irqsoff(struct mm_struct *mm)
  32 {
  33         if (static_branch_unlikely(&rdpmc_always_available_key) ||
  34             (!static_branch_unlikely(&rdpmc_never_available_key) &&
  35              atomic_read(&mm->context.perf_rdpmc_allowed)))
  36                 cr4_set_bits_irqsoff(X86_CR4_PCE);
  37         else
  38                 cr4_clear_bits_irqsoff(X86_CR4_PCE);
  39 }
  40 #else
  41 static inline void load_mm_cr4_irqsoff(struct mm_struct *mm) {}
  42 #endif
  43
  44 #ifdef CONFIG_MODIFY_LDT_SYSCALL
  45 /*
  46  * ldt_structs can be allocated, used, and freed, but they are never
  47  * modified while live.
  48  */
  49 struct ldt_struct {
  50         /*
  51          * Xen requires page-aligned LDTs with special permissions.  This is
  52          * needed to prevent us from installing evil descriptors such as
  53          * call gates.  On native, we could merge the ldt_struct and LDT
  54          * allocations, but it's not worth trying to optimize.
  55          */
  56         struct desc_struct      *entries;
  57         unsigned int            nr_entries;
  58
  59         /*
  60          * If PTI is in use, then the entries array is not mapped while we're
  61          * in user mode.  The whole array will be aliased at the addressed
  62          * given by ldt_slot_va(slot).  We use two slots so that we can allocate
  63          * and map, and enable a new LDT without invalidating the mapping
  64          * of an older, still-in-use LDT.
  65          *
  66          * slot will be -1 if this LDT doesn't have an alias mapping.
  67          */
  68         int                     slot;
  69 };
  70
  71 /*
  72  * Used for LDT copy/destruction.
  73  */
  74 static inline void init_new_context_ldt(struct mm_struct *mm)
  75 {
  76         mm->context.ldt = NULL;
  77         init_rwsem(&mm->context.ldt_usr_sem);
  78 }
  79 int ldt_dup_context(struct mm_struct *oldmm, struct mm_struct *mm);
  80 void destroy_context_ldt(struct mm_struct *mm);
  81 void ldt_arch_exit_mmap(struct mm_struct *mm);
  82 #else   /* CONFIG_MODIFY_LDT_SYSCALL */
  83 static inline void init_new_context_ldt(struct mm_struct *mm) { }
  84 static inline int ldt_dup_context(struct mm_struct *oldmm,
  85                                   struct mm_struct *mm)
  86 {
  87         return 0;
  88 }
  89 static inline void destroy_context_ldt(struct mm_struct *mm) { }
  90 static inline void ldt_arch_exit_mmap(struct mm_struct *mm) { }
  91 #endif
  92
  93 #ifdef CONFIG_MODIFY_LDT_SYSCALL
  94 extern void load_mm_ldt(struct mm_struct *mm);
  95 extern void switch_ldt(struct mm_struct *prev, struct mm_struct *next);
  96 #else
  97 static inline void load_mm_ldt(struct mm_struct *mm)
  98 {
  99         clear_LDT();
 100 }
 101 static inline void switch_ldt(struct mm_struct *prev, struct mm_struct *next)
 102 {
 103         DEBUG_LOCKS_WARN_ON(preemptible());
 104 }
 105 #endif
 106
 107 extern void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk);
 108
 109 /*
 110  * Init a new mm.  Used on mm copies, like at fork()
 111  * and on mm's that are brand-new, like at execve().
 112  */
 113 static inline int init_new_context(struct task_struct *tsk,
 114                                    struct mm_struct *mm)
 115 {
 116         mutex_init(&mm->context.lock);
 117
 118         mm->context.ctx_id = atomic64_inc_return(&last_mm_ctx_id);
 119         atomic64_set(&mm->context.tlb_gen, 0);
 120
 121 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 122         if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
 123                 /* pkey 0 is the default and allocated implicitly */
 124                 mm->context.pkey_allocation_map = 0x1;
 125                 /* -1 means unallocated or invalid */
 126                 mm->context.execute_only_pkey = -1;
 127         }
 128 #endif
 129         init_new_context_ldt(mm);
 130         return 0;
 131 }
 132 static inline void destroy_context(struct mm_struct *mm)
 133 {
 134         destroy_context_ldt(mm);
 135 }
 136
 137 extern void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 138                       struct task_struct *tsk);
 139
 140 extern void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
 141                                struct task_struct *tsk);
 142 #define switch_mm_irqs_off switch_mm_irqs_off
 143
 144 #define activate_mm(prev, next)                 \
 145 do {                                            \
 146         paravirt_activate_mm((prev), (next));   \
 147         switch_mm((prev), (next), NULL);        \
 148 } while (0);
 149
 150 #ifdef CONFIG_X86_32
 151 #define deactivate_mm(tsk, mm)                  \
 152 do {                                            \
 153         lazy_load_gs(0);                        \
 154 } while (0)
 155 #else
 156 #define deactivate_mm(tsk, mm)                  \
 157 do {                                            \
 158         load_gs_index(0);                       \
 159         loadsegment(fs, 0);                     \
 160 } while (0)
 161 #endif
 162
 163 static inline void arch_dup_pkeys(struct mm_struct *oldmm,
 164                                   struct mm_struct *mm)
 165 {
 166 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 167         if (!cpu_feature_enabled(X86_FEATURE_OSPKE))
 168                 return;
 169
 170         /* Duplicate the oldmm pkey state in mm: */
 171         mm->context.pkey_allocation_map = oldmm->context.pkey_allocation_map;
 172         mm->context.execute_only_pkey   = oldmm->context.execute_only_pkey;
 173 #endif
 174 }
 175
 176 static inline int arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
 177 {
 178         arch_dup_pkeys(oldmm, mm);
 179         paravirt_arch_dup_mmap(oldmm, mm);
 180         return ldt_dup_context(oldmm, mm);
 181 }
 182
 183 static inline void arch_exit_mmap(struct mm_struct *mm)
 184 {
 185         paravirt_arch_exit_mmap(mm);
 186         ldt_arch_exit_mmap(mm);
 187 }
 188
 189 #ifdef CONFIG_X86_64
 190 static inline bool is_64bit_mm(struct mm_struct *mm)
 191 {
 192         return  !IS_ENABLED(CONFIG_IA32_EMULATION) ||
 193                 !(mm->context.ia32_compat == TIF_IA32);
 194 }
 195 #else
 196 static inline bool is_64bit_mm(struct mm_struct *mm)
 197 {
 198         return false;
 199 }
 200 #endif
 201
 202 static inline void arch_unmap(struct mm_struct *mm, unsigned long start,
 203                               unsigned long end)
 204 {
 205 }
 206
 207 /*
 208  * We only want to enforce protection keys on the current process
 209  * because we effectively have no access to PKRU for other
 210  * processes or any way to tell *which * PKRU in a threaded
 211  * process we could use.
 212  *
 213  * So do not enforce things if the VMA is not from the current
 214  * mm, or if we are in a kernel thread.
 215  */
 216 static inline bool arch_vma_access_permitted(struct vm_area_struct *vma,
 217                 bool write, bool execute, bool foreign)
 218 {
 219         /* pkeys never affect instruction fetches */
 220         if (execute)
 221                 return true;
 222         /* allow access if the VMA is not one from this process */
 223         if (foreign || vma_is_foreign(vma))
 224                 return true;
 225         return __pkru_allows_pkey(vma_pkey(vma), write);
 226 }
 227
 228 /*
 229  * This can be used from process context to figure out what the value of
 230  * CR3 is without needing to do a (slow) __read_cr3().
 231  *
 232  * It's intended to be used for code like KVM that sneakily changes CR3
 233  * and needs to restore it.  It needs to be used very carefully.
 234  */
 235 static inline unsigned long __get_current_cr3_fast(void)
 236 {
 237         unsigned long cr3 = build_cr3(this_cpu_read(cpu_tlbstate.loaded_mm)->pgd,
 238                 this_cpu_read(cpu_tlbstate.loaded_mm_asid));
 239
 240         /* For now, be very restrictive about when this can be called. */
 241         VM_WARN_ON(in_nmi() || preemptible());
 242
 243         VM_BUG_ON(cr3 != __read_cr3());
 244         return cr3;
 245 }
 246
 247 typedef struct {
 248         struct mm_struct *mm;
 249 } temp_mm_state_t;
 250
 251 /*
 252  * Using a temporary mm allows to set temporary mappings that are not accessible
 253  * by other CPUs. Such mappings are needed to perform sensitive memory writes
 254  * that override the kernel memory protections (e.g., W^X), without exposing the
 255  * temporary page-table mappings that are required for these write operations to
 256  * other CPUs. Using a temporary mm also allows to avoid TLB shootdowns when the
 257  * mapping is torn down.
 258  *
 259  * Context: The temporary mm needs to be used exclusively by a single core. To
 260  *          harden security IRQs must be disabled while the temporary mm is
 261  *          loaded, thereby preventing interrupt handler bugs from overriding
 262  *          the kernel memory protection.
 263  */
 264 static inline temp_mm_state_t use_temporary_mm(struct mm_struct *mm)
 265 {
 266         temp_mm_state_t temp_state;
 267
 268         lockdep_assert_irqs_disabled();
 269         temp_state.mm = this_cpu_read(cpu_tlbstate.loaded_mm);
 270         switch_mm_irqs_off(NULL, mm, current);
 271
 272         /*
 273          * If breakpoints are enabled, disable them while the temporary mm is
 274          * used. Userspace might set up watchpoints on addresses that are used
 275          * in the temporary mm, which would lead to wrong signals being sent or
 276          * crashes.
 277          *
 278          * Note that breakpoints are not disabled selectively, which also causes
 279          * kernel breakpoints (e.g., perf's) to be disabled. This might be
 280          * undesirable, but still seems reasonable as the code that runs in the
 281          * temporary mm should be short.
 282          */
 283         if (hw_breakpoint_active())
 284                 hw_breakpoint_disable();
 285
 286         return temp_state;
 287 }
 288
 289 static inline void unuse_temporary_mm(temp_mm_state_t prev_state)
 290 {
 291         lockdep_assert_irqs_disabled();
 292         switch_mm_irqs_off(NULL, prev_state.mm, current);
 293
 294         /*
 295          * Restore the breakpoints if they were disabled before the temporary mm
 296          * was loaded.
 297          */
 298         if (hw_breakpoint_active())
 299                 hw_breakpoint_restore();
 300 }
 301
 302 #endif /* _ASM_X86_MMU_CONTEXT_H */