arch/arm64/mm/context.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Based on arch/arm/mm/context.c
   4  *
   5  * Copyright (C) 2002-2003 Deep Blue Solutions Ltd, all rights reserved.
   6  * Copyright (C) 2012 ARM Ltd.
   7  */
   8
   9 #include <linux/bitops.h>
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/mm.h>
  13
  14 #include <asm/cpufeature.h>
  15 #include <asm/mmu_context.h>
  16 #include <asm/smp.h>
  17 #include <asm/tlbflush.h>
  18
  19 static u32 asid_bits;
  20 static DEFINE_RAW_SPINLOCK(cpu_asid_lock);
  21
  22 static atomic64_t asid_generation;
  23 static unsigned long *asid_map;
  24
  25 static DEFINE_PER_CPU(atomic64_t, active_asids);
  26 static DEFINE_PER_CPU(u64, reserved_asids);
  27 static cpumask_t tlb_flush_pending;
  28
  29 #define ASID_MASK               (~GENMASK(asid_bits - 1, 0))
  30 #define ASID_FIRST_VERSION      (1UL << asid_bits)
  31
  32 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
  33 #define NUM_USER_ASIDS          (ASID_FIRST_VERSION >> 1)
  34 #define asid2idx(asid)          (((asid) & ~ASID_MASK) >> 1)
  35 #define idx2asid(idx)           (((idx) << 1) & ~ASID_MASK)
  36 #else
  37 #define NUM_USER_ASIDS          (ASID_FIRST_VERSION)
  38 #define asid2idx(asid)          ((asid) & ~ASID_MASK)
  39 #define idx2asid(idx)           asid2idx(idx)
  40 #endif
  41
  42 /* Get the ASIDBits supported by the current CPU */
  43 static u32 get_cpu_asid_bits(void)
  44 {
  45         u32 asid;
  46         int fld = cpuid_feature_extract_unsigned_field(read_cpuid(ID_AA64MMFR0_EL1),
  47                                                 ID_AA64MMFR0_ASID_SHIFT);
  48
  49         switch (fld) {
  50         default:
  51                 pr_warn("CPU%d: Unknown ASID size (%d); assuming 8-bit\n",
  52                                         smp_processor_id(),  fld);
  53                 /* Fallthrough */
  54         case 0:
  55                 asid = 8;
  56                 break;
  57         case 2:
  58                 asid = 16;
  59         }
  60
  61         return asid;
  62 }
  63
  64 /* Check if the current cpu's ASIDBits is compatible with asid_bits */
  65 void verify_cpu_asid_bits(void)
  66 {
  67         u32 asid = get_cpu_asid_bits();
  68
  69         if (asid < asid_bits) {
  70                 /*
  71                  * We cannot decrease the ASID size at runtime, so panic if we support
  72                  * fewer ASID bits than the boot CPU.
  73                  */
  74                 pr_crit("CPU%d: smaller ASID size(%u) than boot CPU (%u)\n",
  75                                 smp_processor_id(), asid, asid_bits);
  76                 cpu_panic_kernel();
  77         }
  78 }
  79
  80 static void flush_context(void)
  81 {
  82         int i;
  83         u64 asid;
  84
  85         /* Update the list of reserved ASIDs and the ASID bitmap. */
  86         bitmap_clear(asid_map, 0, NUM_USER_ASIDS);
  87
  88         for_each_possible_cpu(i) {
  89                 asid = atomic64_xchg_relaxed(&per_cpu(active_asids, i), 0);
  90                 /*
  91                  * If this CPU has already been through a
  92                  * rollover, but hasn't run another task in
  93                  * the meantime, we must preserve its reserved
  94                  * ASID, as this is the only trace we have of
  95                  * the process it is still running.
  96                  */
  97                 if (asid == 0)
  98                         asid = per_cpu(reserved_asids, i);
  99                 __set_bit(asid2idx(asid), asid_map);
 100                 per_cpu(reserved_asids, i) = asid;
 101         }
 102
 103         /*
 104          * Queue a TLB invalidation for each CPU to perform on next
 105          * context-switch
 106          */
 107         cpumask_setall(&tlb_flush_pending);
 108 }
 109
 110 static bool check_update_reserved_asid(u64 asid, u64 newasid)
 111 {
 112         int cpu;
 113         bool hit = false;
 114
 115         /*
 116          * Iterate over the set of reserved ASIDs looking for a match.
 117          * If we find one, then we can update our mm to use newasid
 118          * (i.e. the same ASID in the current generation) but we can't
 119          * exit the loop early, since we need to ensure that all copies
 120          * of the old ASID are updated to reflect the mm. Failure to do
 121          * so could result in us missing the reserved ASID in a future
 122          * generation.
 123          */
 124         for_each_possible_cpu(cpu) {
 125                 if (per_cpu(reserved_asids, cpu) == asid) {
 126                         hit = true;
 127                         per_cpu(reserved_asids, cpu) = newasid;
 128                 }
 129         }
 130
 131         return hit;
 132 }
 133
 134 static u64 new_context(struct mm_struct *mm)
 135 {
 136         static u32 cur_idx = 1;
 137         u64 asid = atomic64_read(&mm->context.id);
 138         u64 generation = atomic64_read(&asid_generation);
 139
 140         if (asid != 0) {
 141                 u64 newasid = generation | (asid & ~ASID_MASK);
 142
 143                 /*
 144                  * If our current ASID was active during a rollover, we
 145                  * can continue to use it and this was just a false alarm.
 146                  */
 147                 if (check_update_reserved_asid(asid, newasid))
 148                         return newasid;
 149
 150                 /*
 151                  * We had a valid ASID in a previous life, so try to re-use
 152                  * it if possible.
 153                  */
 154                 if (!__test_and_set_bit(asid2idx(asid), asid_map))
 155                         return newasid;
 156         }
 157
 158         /*
 159          * Allocate a free ASID. If we can't find one, take a note of the
 160          * currently active ASIDs and mark the TLBs as requiring flushes.  We
 161          * always count from ASID #2 (index 1), as we use ASID #0 when setting
 162          * a reserved TTBR0 for the init_mm and we allocate ASIDs in even/odd
 163          * pairs.
 164          */
 165         asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, cur_idx);
 166         if (asid != NUM_USER_ASIDS)
 167                 goto set_asid;
 168
 169         /* We're out of ASIDs, so increment the global generation count */
 170         generation = atomic64_add_return_relaxed(ASID_FIRST_VERSION,
 171                                                  &asid_generation);
 172         flush_context();
 173
 174         /* We have more ASIDs than CPUs, so this will always succeed */
 175         asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, 1);
 176
 177 set_asid:
 178         __set_bit(asid, asid_map);
 179         cur_idx = asid;
 180         return idx2asid(asid) | generation;
 181 }
 182
 183 void check_and_switch_context(struct mm_struct *mm, unsigned int cpu)
 184 {
 185         unsigned long flags;
 186         u64 asid, old_active_asid;
 187
 188         if (system_supports_cnp())
 189                 cpu_set_reserved_ttbr0();
 190
 191         asid = atomic64_read(&mm->context.id);
 192
 193         /*
 194          * The memory ordering here is subtle.
 195          * If our active_asids is non-zero and the ASID matches the current
 196          * generation, then we update the active_asids entry with a relaxed
 197          * cmpxchg. Racing with a concurrent rollover means that either:
 198          *
 199          * - We get a zero back from the cmpxchg and end up waiting on the
 200          *   lock. Taking the lock synchronises with the rollover and so
 201          *   we are forced to see the updated generation.
 202          *
 203          * - We get a valid ASID back from the cmpxchg, which means the
 204          *   relaxed xchg in flush_context will treat us as reserved
 205          *   because atomic RmWs are totally ordered for a given location.
 206          */
 207         old_active_asid = atomic64_read(&per_cpu(active_asids, cpu));
 208         if (old_active_asid &&
 209             !((asid ^ atomic64_read(&asid_generation)) >> asid_bits) &&
 210             atomic64_cmpxchg_relaxed(&per_cpu(active_asids, cpu),
 211                                      old_active_asid, asid))
 212                 goto switch_mm_fastpath;
 213
 214         raw_spin_lock_irqsave(&cpu_asid_lock, flags);
 215         /* Check that our ASID belongs to the current generation. */
 216         asid = atomic64_read(&mm->context.id);
 217         if ((asid ^ atomic64_read(&asid_generation)) >> asid_bits) {
 218                 asid = new_context(mm);
 219                 atomic64_set(&mm->context.id, asid);
 220         }
 221
 222         if (cpumask_test_and_clear_cpu(cpu, &tlb_flush_pending))
 223                 local_flush_tlb_all();
 224
 225         atomic64_set(&per_cpu(active_asids, cpu), asid);
 226         raw_spin_unlock_irqrestore(&cpu_asid_lock, flags);
 227
 228 switch_mm_fastpath:
 229
 230         arm64_apply_bp_hardening();
 231
 232         /*
 233          * Defer TTBR0_EL1 setting for user threads to uaccess_enable() when
 234          * emulating PAN.
 235          */
 236         if (!system_uses_ttbr0_pan())
 237                 cpu_switch_mm(mm->pgd, mm);
 238 }
 239
 240 /* Errata workaround post TTBRx_EL1 update. */
 241 asmlinkage void post_ttbr_update_workaround(void)
 242 {
 243         asm(ALTERNATIVE("nop; nop; nop",
 244                         "ic iallu; dsb nsh; isb",
 245                         ARM64_WORKAROUND_CAVIUM_27456,
 246                         CONFIG_CAVIUM_ERRATUM_27456));
 247 }
 248
 249 static int asids_init(void)
 250 {
 251         asid_bits = get_cpu_asid_bits();
 252         /*
 253          * Expect allocation after rollover to fail if we don't have at least
 254          * one more ASID than CPUs. ASID #0 is reserved for init_mm.
 255          */
 256         WARN_ON(NUM_USER_ASIDS - 1 <= num_possible_cpus());
 257         atomic64_set(&asid_generation, ASID_FIRST_VERSION);
 258         asid_map = kcalloc(BITS_TO_LONGS(NUM_USER_ASIDS), sizeof(*asid_map),
 259                            GFP_KERNEL);
 260         if (!asid_map)
 261                 panic("Failed to allocate bitmap for %lu ASIDs\n",
 262                       NUM_USER_ASIDS);
 263
 264         pr_info("ASID allocator initialised with %lu entries\n", NUM_USER_ASIDS);
 265         return 0;
 266 }
 267 early_initcall(asids_init);