arch/arm64/mm/context.c

   1 /*
   2  * Based on arch/arm/mm/context.c
   3  *
   4  * Copyright (C) 2002-2003 Deep Blue Solutions Ltd, all rights reserved.
   5  * Copyright (C) 2012 ARM Ltd.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include <linux/bitops.h>
  21 #include <linux/sched.h>
  22 #include <linux/slab.h>
  23 #include <linux/mm.h>
  24
  25 #include <asm/cpufeature.h>
  26 #include <asm/mmu_context.h>
  27 #include <asm/smp.h>
  28 #include <asm/tlbflush.h>
  29
  30 static u32 asid_bits;
  31 static DEFINE_RAW_SPINLOCK(cpu_asid_lock);
  32
  33 static atomic64_t asid_generation;
  34 static unsigned long *asid_map;
  35
  36 static DEFINE_PER_CPU(atomic64_t, active_asids);
  37 static DEFINE_PER_CPU(u64, reserved_asids);
  38 static cpumask_t tlb_flush_pending;
  39
  40 #define ASID_MASK               (~GENMASK(asid_bits - 1, 0))
  41 #define ASID_FIRST_VERSION      (1UL << asid_bits)
  42 #define NUM_USER_ASIDS          ASID_FIRST_VERSION
  43
  44 /* Get the ASIDBits supported by the current CPU */
  45 static u32 get_cpu_asid_bits(void)
  46 {
  47         u32 asid;
  48         int fld = cpuid_feature_extract_unsigned_field(read_cpuid(ID_AA64MMFR0_EL1),
  49                                                 ID_AA64MMFR0_ASID_SHIFT);
  50
  51         switch (fld) {
  52         default:
  53                 pr_warn("CPU%d: Unknown ASID size (%d); assuming 8-bit\n",
  54                                         smp_processor_id(),  fld);
  55                 /* Fallthrough */
  56         case 0:
  57                 asid = 8;
  58                 break;
  59         case 2:
  60                 asid = 16;
  61         }
  62
  63         return asid;
  64 }
  65
  66 /* Check if the current cpu's ASIDBits is compatible with asid_bits */
  67 void verify_cpu_asid_bits(void)
  68 {
  69         u32 asid = get_cpu_asid_bits();
  70
  71         if (asid < asid_bits) {
  72                 /*
  73                  * We cannot decrease the ASID size at runtime, so panic if we support
  74                  * fewer ASID bits than the boot CPU.
  75                  */
  76                 pr_crit("CPU%d: smaller ASID size(%u) than boot CPU (%u)\n",
  77                                 smp_processor_id(), asid, asid_bits);
  78                 cpu_panic_kernel();
  79         }
  80 }
  81
  82 static void set_reserved_asid_bits(void)
  83 {
  84         if (IS_ENABLED(CONFIG_QCOM_FALKOR_ERRATUM_1003) &&
  85             cpus_have_const_cap(ARM64_WORKAROUND_QCOM_FALKOR_E1003))
  86                 __set_bit(FALKOR_RESERVED_ASID, asid_map);
  87 }
  88
  89 static void flush_context(unsigned int cpu)
  90 {
  91         int i;
  92         u64 asid;
  93
  94         /* Update the list of reserved ASIDs and the ASID bitmap. */
  95         bitmap_clear(asid_map, 0, NUM_USER_ASIDS);
  96
  97         set_reserved_asid_bits();
  98
  99         for_each_possible_cpu(i) {
 100                 asid = atomic64_xchg_relaxed(&per_cpu(active_asids, i), 0);
 101                 /*
 102                  * If this CPU has already been through a
 103                  * rollover, but hasn't run another task in
 104                  * the meantime, we must preserve its reserved
 105                  * ASID, as this is the only trace we have of
 106                  * the process it is still running.
 107                  */
 108                 if (asid == 0)
 109                         asid = per_cpu(reserved_asids, i);
 110                 __set_bit(asid & ~ASID_MASK, asid_map);
 111                 per_cpu(reserved_asids, i) = asid;
 112         }
 113
 114         /*
 115          * Queue a TLB invalidation for each CPU to perform on next
 116          * context-switch
 117          */
 118         cpumask_setall(&tlb_flush_pending);
 119 }
 120
 121 static bool check_update_reserved_asid(u64 asid, u64 newasid)
 122 {
 123         int cpu;
 124         bool hit = false;
 125
 126         /*
 127          * Iterate over the set of reserved ASIDs looking for a match.
 128          * If we find one, then we can update our mm to use newasid
 129          * (i.e. the same ASID in the current generation) but we can't
 130          * exit the loop early, since we need to ensure that all copies
 131          * of the old ASID are updated to reflect the mm. Failure to do
 132          * so could result in us missing the reserved ASID in a future
 133          * generation.
 134          */
 135         for_each_possible_cpu(cpu) {
 136                 if (per_cpu(reserved_asids, cpu) == asid) {
 137                         hit = true;
 138                         per_cpu(reserved_asids, cpu) = newasid;
 139                 }
 140         }
 141
 142         return hit;
 143 }
 144
 145 static u64 new_context(struct mm_struct *mm, unsigned int cpu)
 146 {
 147         static u32 cur_idx = 1;
 148         u64 asid = atomic64_read(&mm->context.id);
 149         u64 generation = atomic64_read(&asid_generation);
 150
 151         if (asid != 0) {
 152                 u64 newasid = generation | (asid & ~ASID_MASK);
 153
 154                 /*
 155                  * If our current ASID was active during a rollover, we
 156                  * can continue to use it and this was just a false alarm.
 157                  */
 158                 if (check_update_reserved_asid(asid, newasid))
 159                         return newasid;
 160
 161                 /*
 162                  * We had a valid ASID in a previous life, so try to re-use
 163                  * it if possible.
 164                  */
 165                 asid &= ~ASID_MASK;
 166                 if (!__test_and_set_bit(asid, asid_map))
 167                         return newasid;
 168         }
 169
 170         /*
 171          * Allocate a free ASID. If we can't find one, take a note of the
 172          * currently active ASIDs and mark the TLBs as requiring flushes.
 173          * We always count from ASID #1, as we use ASID #0 when setting a
 174          * reserved TTBR0 for the init_mm.
 175          */
 176         asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, cur_idx);
 177         if (asid != NUM_USER_ASIDS)
 178                 goto set_asid;
 179
 180         /* We're out of ASIDs, so increment the global generation count */
 181         generation = atomic64_add_return_relaxed(ASID_FIRST_VERSION,
 182                                                  &asid_generation);
 183         flush_context(cpu);
 184
 185         /* We have more ASIDs than CPUs, so this will always succeed */
 186         asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, 1);
 187
 188 set_asid:
 189         __set_bit(asid, asid_map);
 190         cur_idx = asid;
 191         return asid | generation;
 192 }
 193
 194 void check_and_switch_context(struct mm_struct *mm, unsigned int cpu)
 195 {
 196         unsigned long flags;
 197         u64 asid;
 198
 199         asid = atomic64_read(&mm->context.id);
 200
 201         /*
 202          * The memory ordering here is subtle.
 203          * If our ASID matches the current generation, then we update
 204          * our active_asids entry with a relaxed xchg. Racing with a
 205          * concurrent rollover means that either:
 206          *
 207          * - We get a zero back from the xchg and end up waiting on the
 208          *   lock. Taking the lock synchronises with the rollover and so
 209          *   we are forced to see the updated generation.
 210          *
 211          * - We get a valid ASID back from the xchg, which means the
 212          *   relaxed xchg in flush_context will treat us as reserved
 213          *   because atomic RmWs are totally ordered for a given location.
 214          */
 215         if (!((asid ^ atomic64_read(&asid_generation)) >> asid_bits)
 216             && atomic64_xchg_relaxed(&per_cpu(active_asids, cpu), asid))
 217                 goto switch_mm_fastpath;
 218
 219         raw_spin_lock_irqsave(&cpu_asid_lock, flags);
 220         /* Check that our ASID belongs to the current generation. */
 221         asid = atomic64_read(&mm->context.id);
 222         if ((asid ^ atomic64_read(&asid_generation)) >> asid_bits) {
 223                 asid = new_context(mm, cpu);
 224                 atomic64_set(&mm->context.id, asid);
 225         }
 226
 227         if (cpumask_test_and_clear_cpu(cpu, &tlb_flush_pending))
 228                 local_flush_tlb_all();
 229
 230         atomic64_set(&per_cpu(active_asids, cpu), asid);
 231         raw_spin_unlock_irqrestore(&cpu_asid_lock, flags);
 232
 233 switch_mm_fastpath:
 234         /*
 235          * Defer TTBR0_EL1 setting for user threads to uaccess_enable() when
 236          * emulating PAN.
 237          */
 238         if (!system_uses_ttbr0_pan())
 239                 cpu_switch_mm(mm->pgd, mm);
 240 }
 241
 242 static int asids_init(void)
 243 {
 244         asid_bits = get_cpu_asid_bits();
 245         /*
 246          * Expect allocation after rollover to fail if we don't have at least
 247          * one more ASID than CPUs. ASID #0 is reserved for init_mm.
 248          */
 249         WARN_ON(NUM_USER_ASIDS - 1 <= num_possible_cpus());
 250         atomic64_set(&asid_generation, ASID_FIRST_VERSION);
 251         asid_map = kzalloc(BITS_TO_LONGS(NUM_USER_ASIDS) * sizeof(*asid_map),
 252                            GFP_KERNEL);
 253         if (!asid_map)
 254                 panic("Failed to allocate bitmap for %lu ASIDs\n",
 255                       NUM_USER_ASIDS);
 256
 257         set_reserved_asid_bits();
 258
 259         pr_info("ASID allocator initialised with %lu entries\n", NUM_USER_ASIDS);
 260         return 0;
 261 }
 262 early_initcall(asids_init);