1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/arch/arm/mm/context.c 4 * 5 * Copyright (C) 2002-2003 Deep Blue Solutions Ltd, all rights reserved. 6 * Copyright (C) 2012 ARM Limited 7 * 8 * Author: Will Deacon <will.deacon@arm.com> 9 */ 10 #include <linux/init.h> 11 #include <linux/sched.h> 12 #include <linux/mm.h> 13 #include <linux/smp.h> 14 #include <linux/percpu.h> 15 16 #include <asm/mmu_context.h> 17 #include <asm/smp_plat.h> 18 #include <asm/thread_notify.h> 19 #include <asm/tlbflush.h> 20 #include <asm/proc-fns.h> 21 22 /* 23 * On ARMv6, we have the following structure in the Context ID: 24 * 25 * 31 7 0 26 * +-------------------------+-----------+ 27 * | process ID | ASID | 28 * +-------------------------+-----------+ 29 * | context ID | 30 * +-------------------------------------+ 31 * 32 * The ASID is used to tag entries in the CPU caches and TLBs. 33 * The context ID is used by debuggers and trace logic, and 34 * should be unique within all running processes. 35 * 36 * In big endian operation, the two 32 bit words are swapped if accessed 37 * by non-64-bit operations. 38 */ 39 #define ASID_FIRST_VERSION (1ULL << ASID_BITS) 40 #define NUM_USER_ASIDS ASID_FIRST_VERSION 41 42 static DEFINE_RAW_SPINLOCK(cpu_asid_lock); 43 static atomic64_t asid_generation = ATOMIC64_INIT(ASID_FIRST_VERSION); 44 static DECLARE_BITMAP(asid_map, NUM_USER_ASIDS); 45 46 static DEFINE_PER_CPU(atomic64_t, active_asids); 47 static DEFINE_PER_CPU(u64, reserved_asids); 48 static cpumask_t tlb_flush_pending; 49 50 #ifdef CONFIG_ARM_ERRATA_798181 51 void a15_erratum_get_cpumask(int this_cpu, struct mm_struct *mm, 52 cpumask_t *mask) 53 { 54 int cpu; 55 unsigned long flags; 56 u64 context_id, asid; 57 58 raw_spin_lock_irqsave(&cpu_asid_lock, flags); 59 context_id = mm->context.id.counter; 60 for_each_online_cpu(cpu) { 61 if (cpu == this_cpu) 62 continue; 63 /* 64 * We only need to send an IPI if the other CPUs are 65 * running the same ASID as the one being invalidated. 66 */ 67 asid = per_cpu(active_asids, cpu).counter; 68 if (asid == 0) 69 asid = per_cpu(reserved_asids, cpu); 70 if (context_id == asid) 71 cpumask_set_cpu(cpu, mask); 72 } 73 raw_spin_unlock_irqrestore(&cpu_asid_lock, flags); 74 } 75 #endif 76 77 #ifdef CONFIG_ARM_LPAE 78 /* 79 * With LPAE, the ASID and page tables are updated atomicly, so there is 80 * no need for a reserved set of tables (the active ASID tracking prevents 81 * any issues across a rollover). 82 */ 83 #define cpu_set_reserved_ttbr0() 84 #else 85 static void cpu_set_reserved_ttbr0(void) 86 { 87 u32 ttb; 88 /* 89 * Copy TTBR1 into TTBR0. 90 * This points at swapper_pg_dir, which contains only global 91 * entries so any speculative walks are perfectly safe. 92 */ 93 asm volatile( 94 " mrc p15, 0, %0, c2, c0, 1 @ read TTBR1\n" 95 " mcr p15, 0, %0, c2, c0, 0 @ set TTBR0\n" 96 : "=r" (ttb)); 97 isb(); 98 } 99 #endif 100 101 #ifdef CONFIG_PID_IN_CONTEXTIDR 102 static int contextidr_notifier(struct notifier_block *unused, unsigned long cmd, 103 void *t) 104 { 105 u32 contextidr; 106 pid_t pid; 107 struct thread_info *thread = t; 108 109 if (cmd != THREAD_NOTIFY_SWITCH) 110 return NOTIFY_DONE; 111 112 pid = task_pid_nr(thread->task) << ASID_BITS; 113 asm volatile( 114 " mrc p15, 0, %0, c13, c0, 1\n" 115 " and %0, %0, %2\n" 116 " orr %0, %0, %1\n" 117 " mcr p15, 0, %0, c13, c0, 1\n" 118 : "=r" (contextidr), "+r" (pid) 119 : "I" (~ASID_MASK)); 120 isb(); 121 122 return NOTIFY_OK; 123 } 124 125 static struct notifier_block contextidr_notifier_block = { 126 .notifier_call = contextidr_notifier, 127 }; 128 129 static int __init contextidr_notifier_init(void) 130 { 131 return thread_register_notifier(&contextidr_notifier_block); 132 } 133 arch_initcall(contextidr_notifier_init); 134 #endif 135 136 static void flush_context(unsigned int cpu) 137 { 138 int i; 139 u64 asid; 140 141 /* Update the list of reserved ASIDs and the ASID bitmap. */ 142 bitmap_clear(asid_map, 0, NUM_USER_ASIDS); 143 for_each_possible_cpu(i) { 144 asid = atomic64_xchg(&per_cpu(active_asids, i), 0); 145 /* 146 * If this CPU has already been through a 147 * rollover, but hasn't run another task in 148 * the meantime, we must preserve its reserved 149 * ASID, as this is the only trace we have of 150 * the process it is still running. 151 */ 152 if (asid == 0) 153 asid = per_cpu(reserved_asids, i); 154 __set_bit(asid & ~ASID_MASK, asid_map); 155 per_cpu(reserved_asids, i) = asid; 156 } 157 158 /* Queue a TLB invalidate and flush the I-cache if necessary. */ 159 cpumask_setall(&tlb_flush_pending); 160 161 if (icache_is_vivt_asid_tagged()) 162 __flush_icache_all(); 163 } 164 165 static bool check_update_reserved_asid(u64 asid, u64 newasid) 166 { 167 int cpu; 168 bool hit = false; 169 170 /* 171 * Iterate over the set of reserved ASIDs looking for a match. 172 * If we find one, then we can update our mm to use newasid 173 * (i.e. the same ASID in the current generation) but we can't 174 * exit the loop early, since we need to ensure that all copies 175 * of the old ASID are updated to reflect the mm. Failure to do 176 * so could result in us missing the reserved ASID in a future 177 * generation. 178 */ 179 for_each_possible_cpu(cpu) { 180 if (per_cpu(reserved_asids, cpu) == asid) { 181 hit = true; 182 per_cpu(reserved_asids, cpu) = newasid; 183 } 184 } 185 186 return hit; 187 } 188 189 static u64 new_context(struct mm_struct *mm, unsigned int cpu) 190 { 191 static u32 cur_idx = 1; 192 u64 asid = atomic64_read(&mm->context.id); 193 u64 generation = atomic64_read(&asid_generation); 194 195 if (asid != 0) { 196 u64 newasid = generation | (asid & ~ASID_MASK); 197 198 /* 199 * If our current ASID was active during a rollover, we 200 * can continue to use it and this was just a false alarm. 201 */ 202 if (check_update_reserved_asid(asid, newasid)) 203 return newasid; 204 205 /* 206 * We had a valid ASID in a previous life, so try to re-use 207 * it if possible., 208 */ 209 asid &= ~ASID_MASK; 210 if (!__test_and_set_bit(asid, asid_map)) 211 return newasid; 212 } 213 214 /* 215 * Allocate a free ASID. If we can't find one, take a note of the 216 * currently active ASIDs and mark the TLBs as requiring flushes. 217 * We always count from ASID #1, as we reserve ASID #0 to switch 218 * via TTBR0 and to avoid speculative page table walks from hitting 219 * in any partial walk caches, which could be populated from 220 * overlapping level-1 descriptors used to map both the module 221 * area and the userspace stack. 222 */ 223 asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, cur_idx); 224 if (asid == NUM_USER_ASIDS) { 225 generation = atomic64_add_return(ASID_FIRST_VERSION, 226 &asid_generation); 227 flush_context(cpu); 228 asid = find_next_zero_bit(asid_map, NUM_USER_ASIDS, 1); 229 } 230 231 __set_bit(asid, asid_map); 232 cur_idx = asid; 233 cpumask_clear(mm_cpumask(mm)); 234 return asid | generation; 235 } 236 237 void check_and_switch_context(struct mm_struct *mm, struct task_struct *tsk) 238 { 239 unsigned long flags; 240 unsigned int cpu = smp_processor_id(); 241 u64 asid; 242 243 if (unlikely(mm->context.vmalloc_seq != init_mm.context.vmalloc_seq)) 244 __check_vmalloc_seq(mm); 245 246 /* 247 * We cannot update the pgd and the ASID atomicly with classic 248 * MMU, so switch exclusively to global mappings to avoid 249 * speculative page table walking with the wrong TTBR. 250 */ 251 cpu_set_reserved_ttbr0(); 252 253 asid = atomic64_read(&mm->context.id); 254 if (!((asid ^ atomic64_read(&asid_generation)) >> ASID_BITS) 255 && atomic64_xchg(&per_cpu(active_asids, cpu), asid)) 256 goto switch_mm_fastpath; 257 258 raw_spin_lock_irqsave(&cpu_asid_lock, flags); 259 /* Check that our ASID belongs to the current generation. */ 260 asid = atomic64_read(&mm->context.id); 261 if ((asid ^ atomic64_read(&asid_generation)) >> ASID_BITS) { 262 asid = new_context(mm, cpu); 263 atomic64_set(&mm->context.id, asid); 264 } 265 266 if (cpumask_test_and_clear_cpu(cpu, &tlb_flush_pending)) { 267 local_flush_bp_all(); 268 local_flush_tlb_all(); 269 } 270 271 atomic64_set(&per_cpu(active_asids, cpu), asid); 272 cpumask_set_cpu(cpu, mm_cpumask(mm)); 273 raw_spin_unlock_irqrestore(&cpu_asid_lock, flags); 274 275 switch_mm_fastpath: 276 cpu_switch_mm(mm->pgd, mm); 277 } 278