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