Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 1 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 2 | |
Suresh Siddha | 61c4628 | 2008-03-10 15:28:04 -0700 | [diff] [blame] | 3 | #include <linux/errno.h> |
| 4 | #include <linux/kernel.h> |
| 5 | #include <linux/mm.h> |
| 6 | #include <linux/smp.h> |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 7 | #include <linux/prctl.h> |
Suresh Siddha | 61c4628 | 2008-03-10 15:28:04 -0700 | [diff] [blame] | 8 | #include <linux/slab.h> |
| 9 | #include <linux/sched.h> |
Paul Gortmaker | 186f436 | 2016-07-13 20:18:56 -0400 | [diff] [blame] | 10 | #include <linux/init.h> |
| 11 | #include <linux/export.h> |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 12 | #include <linux/pm.h> |
Thomas Gleixner | 162a688 | 2015-04-03 02:01:28 +0200 | [diff] [blame] | 13 | #include <linux/tick.h> |
Amerigo Wang | 9d62dcd | 2009-05-11 22:05:28 -0400 | [diff] [blame] | 14 | #include <linux/random.h> |
Avi Kivity | 7c68af6 | 2009-09-19 09:40:22 +0300 | [diff] [blame] | 15 | #include <linux/user-return-notifier.h> |
Andy Isaacson | 814e2c8 | 2009-12-08 00:29:42 -0800 | [diff] [blame] | 16 | #include <linux/dmi.h> |
| 17 | #include <linux/utsname.h> |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 18 | #include <linux/stackprotector.h> |
| 19 | #include <linux/tick.h> |
| 20 | #include <linux/cpuidle.h> |
Arjan van de Ven | 6161352 | 2009-09-17 16:11:28 +0200 | [diff] [blame] | 21 | #include <trace/events/power.h> |
Frederic Weisbecker | 24f1e32c | 2009-09-09 19:22:48 +0200 | [diff] [blame] | 22 | #include <linux/hw_breakpoint.h> |
Borislav Petkov | 93789b3 | 2011-01-20 15:42:52 +0100 | [diff] [blame] | 23 | #include <asm/cpu.h> |
Ivan Vecera | d3ec5ca | 2008-11-11 14:33:44 +0100 | [diff] [blame] | 24 | #include <asm/apic.h> |
Jaswinder Singh Rajput | 2c1b284 | 2009-04-11 00:03:10 +0530 | [diff] [blame] | 25 | #include <asm/syscalls.h> |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 26 | #include <asm/idle.h> |
| 27 | #include <asm/uaccess.h> |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 28 | #include <asm/mwait.h> |
Ingo Molnar | 78f7f1e | 2015-04-24 02:54:44 +0200 | [diff] [blame] | 29 | #include <asm/fpu/internal.h> |
K.Prasad | 66cb591 | 2009-06-01 23:44:55 +0530 | [diff] [blame] | 30 | #include <asm/debugreg.h> |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 31 | #include <asm/nmi.h> |
Andy Lutomirski | 375074c | 2014-10-24 15:58:07 -0700 | [diff] [blame] | 32 | #include <asm/tlbflush.h> |
Ashok Raj | 8838eb6 | 2015-08-12 18:29:40 +0200 | [diff] [blame] | 33 | #include <asm/mce.h> |
Brian Gerst | 9fda6a0 | 2015-07-29 01:41:16 -0400 | [diff] [blame] | 34 | #include <asm/vm86.h> |
Brian Gerst | 7b32aea | 2016-08-13 12:38:18 -0400 | [diff] [blame] | 35 | #include <asm/switch_to.h> |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 36 | #include <asm/spec-ctrl.h> |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 37 | |
Thomas Gleixner | 4504689 | 2012-05-03 09:03:01 +0000 | [diff] [blame] | 38 | /* |
| 39 | * per-CPU TSS segments. Threads are completely 'soft' on Linux, |
| 40 | * no more per-task TSS's. The TSS size is kept cacheline-aligned |
| 41 | * so they are allowed to end up in the .data..cacheline_aligned |
| 42 | * section. Since TSS's are completely CPU-local, we want them |
| 43 | * on exact cacheline boundaries, to eliminate cacheline ping-pong. |
| 44 | */ |
Richard Fellner | 13be448 | 2017-05-04 14:26:50 +0200 | [diff] [blame] | 45 | __visible DEFINE_PER_CPU_SHARED_ALIGNED_USER_MAPPED(struct tss_struct, cpu_tss) = { |
Andy Lutomirski | d0a0de2 | 2015-03-05 19:19:06 -0800 | [diff] [blame] | 46 | .x86_tss = { |
Andy Lutomirski | d9e05cc | 2015-03-10 11:05:59 -0700 | [diff] [blame] | 47 | .sp0 = TOP_OF_INIT_STACK, |
Andy Lutomirski | d0a0de2 | 2015-03-05 19:19:06 -0800 | [diff] [blame] | 48 | #ifdef CONFIG_X86_32 |
| 49 | .ss0 = __KERNEL_DS, |
| 50 | .ss1 = __KERNEL_CS, |
| 51 | .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, |
| 52 | #endif |
| 53 | }, |
| 54 | #ifdef CONFIG_X86_32 |
| 55 | /* |
| 56 | * Note that the .io_bitmap member must be extra-big. This is because |
| 57 | * the CPU will access an additional byte beyond the end of the IO |
| 58 | * permission bitmap. The extra byte must be all 1 bits, and must |
| 59 | * be within the limit. |
| 60 | */ |
| 61 | .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, |
| 62 | #endif |
Andy Lutomirski | 2a41aa4 | 2016-03-09 19:00:33 -0800 | [diff] [blame] | 63 | #ifdef CONFIG_X86_32 |
| 64 | .SYSENTER_stack_canary = STACK_END_MAGIC, |
| 65 | #endif |
Andy Lutomirski | d0a0de2 | 2015-03-05 19:19:06 -0800 | [diff] [blame] | 66 | }; |
Marc Dionne | de71ad2 | 2015-05-04 15:16:44 -0300 | [diff] [blame] | 67 | EXPORT_PER_CPU_SYMBOL(cpu_tss); |
Thomas Gleixner | 4504689 | 2012-05-03 09:03:01 +0000 | [diff] [blame] | 68 | |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 69 | #ifdef CONFIG_X86_64 |
| 70 | static DEFINE_PER_CPU(unsigned char, is_idle); |
| 71 | static ATOMIC_NOTIFIER_HEAD(idle_notifier); |
| 72 | |
| 73 | void idle_notifier_register(struct notifier_block *n) |
| 74 | { |
| 75 | atomic_notifier_chain_register(&idle_notifier, n); |
| 76 | } |
| 77 | EXPORT_SYMBOL_GPL(idle_notifier_register); |
| 78 | |
| 79 | void idle_notifier_unregister(struct notifier_block *n) |
| 80 | { |
| 81 | atomic_notifier_chain_unregister(&idle_notifier, n); |
| 82 | } |
| 83 | EXPORT_SYMBOL_GPL(idle_notifier_unregister); |
| 84 | #endif |
Zhao Yakui | c1e3b37 | 2008-06-24 17:58:53 +0800 | [diff] [blame] | 85 | |
Suresh Siddha | 55ccf3f | 2012-05-16 15:03:51 -0700 | [diff] [blame] | 86 | /* |
| 87 | * this gets called so that we can store lazy state into memory and copy the |
| 88 | * current task into the new thread. |
| 89 | */ |
Suresh Siddha | 61c4628 | 2008-03-10 15:28:04 -0700 | [diff] [blame] | 90 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) |
| 91 | { |
Ingo Molnar | 5aaeb5c | 2015-07-17 12:28:12 +0200 | [diff] [blame] | 92 | memcpy(dst, src, arch_task_struct_size); |
Andy Lutomirski | 2459ee8 | 2015-10-30 22:42:46 -0700 | [diff] [blame] | 93 | #ifdef CONFIG_VM86 |
| 94 | dst->thread.vm86 = NULL; |
| 95 | #endif |
Oleg Nesterov | f185350 | 2014-09-02 19:57:23 +0200 | [diff] [blame] | 96 | |
Ingo Molnar | c69e098 | 2015-04-24 02:07:15 +0200 | [diff] [blame] | 97 | return fpu__copy(&dst->thread.fpu, &src->thread.fpu); |
Suresh Siddha | 61c4628 | 2008-03-10 15:28:04 -0700 | [diff] [blame] | 98 | } |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 99 | |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 100 | /* |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 101 | * Free current thread data structures etc.. |
| 102 | */ |
Jiri Slaby | e646469 | 2016-05-20 17:00:20 -0700 | [diff] [blame] | 103 | void exit_thread(struct task_struct *tsk) |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 104 | { |
Jiri Slaby | e646469 | 2016-05-20 17:00:20 -0700 | [diff] [blame] | 105 | struct thread_struct *t = &tsk->thread; |
Thomas Gleixner | 250981e | 2009-03-16 13:07:21 +0100 | [diff] [blame] | 106 | unsigned long *bp = t->io_bitmap_ptr; |
Ingo Molnar | ca6787b | 2015-04-23 12:33:50 +0200 | [diff] [blame] | 107 | struct fpu *fpu = &t->fpu; |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 108 | |
Thomas Gleixner | 250981e | 2009-03-16 13:07:21 +0100 | [diff] [blame] | 109 | if (bp) { |
Andy Lutomirski | 24933b8 | 2015-03-05 19:19:05 -0800 | [diff] [blame] | 110 | struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu()); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 111 | |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 112 | t->io_bitmap_ptr = NULL; |
| 113 | clear_thread_flag(TIF_IO_BITMAP); |
| 114 | /* |
| 115 | * Careful, clear this in the TSS too: |
| 116 | */ |
| 117 | memset(tss->io_bitmap, 0xff, t->io_bitmap_max); |
| 118 | t->io_bitmap_max = 0; |
| 119 | put_cpu(); |
Thomas Gleixner | 250981e | 2009-03-16 13:07:21 +0100 | [diff] [blame] | 120 | kfree(bp); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 121 | } |
Suresh Siddha | 1dcc8d7 | 2012-05-16 15:03:54 -0700 | [diff] [blame] | 122 | |
Brian Gerst | 9fda6a0 | 2015-07-29 01:41:16 -0400 | [diff] [blame] | 123 | free_vm86(t); |
| 124 | |
Ingo Molnar | 5033861 | 2015-04-29 19:04:31 +0200 | [diff] [blame] | 125 | fpu__drop(fpu); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 126 | } |
| 127 | |
| 128 | void flush_thread(void) |
| 129 | { |
| 130 | struct task_struct *tsk = current; |
| 131 | |
Frederic Weisbecker | 24f1e32c | 2009-09-09 19:22:48 +0200 | [diff] [blame] | 132 | flush_ptrace_hw_breakpoint(tsk); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 133 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); |
Oleg Nesterov | 110d7f7 | 2015-01-19 19:52:12 +0100 | [diff] [blame] | 134 | |
Ingo Molnar | 04c8e01 | 2015-04-29 20:35:33 +0200 | [diff] [blame] | 135 | fpu__clear(&tsk->thread.fpu); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 136 | } |
| 137 | |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 138 | void disable_TSC(void) |
| 139 | { |
| 140 | preempt_disable(); |
| 141 | if (!test_and_set_thread_flag(TIF_NOTSC)) |
| 142 | /* |
| 143 | * Must flip the CPU state synchronously with |
| 144 | * TIF_NOTSC in the current running context. |
| 145 | */ |
Thomas Gleixner | 5ed7788 | 2017-02-14 00:11:04 -0800 | [diff] [blame] | 146 | cr4_set_bits(X86_CR4_TSD); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 147 | preempt_enable(); |
| 148 | } |
| 149 | |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 150 | static void enable_TSC(void) |
| 151 | { |
| 152 | preempt_disable(); |
| 153 | if (test_and_clear_thread_flag(TIF_NOTSC)) |
| 154 | /* |
| 155 | * Must flip the CPU state synchronously with |
| 156 | * TIF_NOTSC in the current running context. |
| 157 | */ |
Thomas Gleixner | 5ed7788 | 2017-02-14 00:11:04 -0800 | [diff] [blame] | 158 | cr4_clear_bits(X86_CR4_TSD); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 159 | preempt_enable(); |
| 160 | } |
| 161 | |
| 162 | int get_tsc_mode(unsigned long adr) |
| 163 | { |
| 164 | unsigned int val; |
| 165 | |
| 166 | if (test_thread_flag(TIF_NOTSC)) |
| 167 | val = PR_TSC_SIGSEGV; |
| 168 | else |
| 169 | val = PR_TSC_ENABLE; |
| 170 | |
| 171 | return put_user(val, (unsigned int __user *)adr); |
| 172 | } |
| 173 | |
| 174 | int set_tsc_mode(unsigned int val) |
| 175 | { |
| 176 | if (val == PR_TSC_SIGSEGV) |
| 177 | disable_TSC(); |
| 178 | else if (val == PR_TSC_ENABLE) |
| 179 | enable_TSC(); |
| 180 | else |
| 181 | return -EINVAL; |
| 182 | |
| 183 | return 0; |
| 184 | } |
| 185 | |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 186 | static inline void switch_to_bitmap(struct tss_struct *tss, |
| 187 | struct thread_struct *prev, |
| 188 | struct thread_struct *next, |
| 189 | unsigned long tifp, unsigned long tifn) |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 190 | { |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 191 | if (tifn & _TIF_IO_BITMAP) { |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 192 | /* |
| 193 | * Copy the relevant range of the IO bitmap. |
| 194 | * Normally this is 128 bytes or less: |
| 195 | */ |
| 196 | memcpy(tss->io_bitmap, next->io_bitmap_ptr, |
| 197 | max(prev->io_bitmap_max, next->io_bitmap_max)); |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 198 | } else if (tifp & _TIF_IO_BITMAP) { |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 199 | /* |
| 200 | * Clear any possible leftover bits: |
| 201 | */ |
| 202 | memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); |
| 203 | } |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 204 | } |
| 205 | |
Thomas Gleixner | d0cb78f | 2018-05-09 21:53:09 +0200 | [diff] [blame] | 206 | #ifdef CONFIG_SMP |
| 207 | |
| 208 | struct ssb_state { |
| 209 | struct ssb_state *shared_state; |
| 210 | raw_spinlock_t lock; |
| 211 | unsigned int disable_state; |
| 212 | unsigned long local_state; |
| 213 | }; |
| 214 | |
| 215 | #define LSTATE_SSB 0 |
| 216 | |
| 217 | static DEFINE_PER_CPU(struct ssb_state, ssb_state); |
| 218 | |
| 219 | void speculative_store_bypass_ht_init(void) |
| 220 | { |
| 221 | struct ssb_state *st = this_cpu_ptr(&ssb_state); |
| 222 | unsigned int this_cpu = smp_processor_id(); |
| 223 | unsigned int cpu; |
| 224 | |
| 225 | st->local_state = 0; |
| 226 | |
| 227 | /* |
| 228 | * Shared state setup happens once on the first bringup |
| 229 | * of the CPU. It's not destroyed on CPU hotunplug. |
| 230 | */ |
| 231 | if (st->shared_state) |
| 232 | return; |
| 233 | |
| 234 | raw_spin_lock_init(&st->lock); |
| 235 | |
| 236 | /* |
| 237 | * Go over HT siblings and check whether one of them has set up the |
| 238 | * shared state pointer already. |
| 239 | */ |
| 240 | for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) { |
| 241 | if (cpu == this_cpu) |
| 242 | continue; |
| 243 | |
| 244 | if (!per_cpu(ssb_state, cpu).shared_state) |
| 245 | continue; |
| 246 | |
| 247 | /* Link it to the state of the sibling: */ |
| 248 | st->shared_state = per_cpu(ssb_state, cpu).shared_state; |
| 249 | return; |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * First HT sibling to come up on the core. Link shared state of |
| 254 | * the first HT sibling to itself. The siblings on the same core |
| 255 | * which come up later will see the shared state pointer and link |
| 256 | * themself to the state of this CPU. |
| 257 | */ |
| 258 | st->shared_state = st; |
| 259 | } |
| 260 | |
| 261 | /* |
| 262 | * Logic is: First HT sibling enables SSBD for both siblings in the core |
| 263 | * and last sibling to disable it, disables it for the whole core. This how |
| 264 | * MSR_SPEC_CTRL works in "hardware": |
| 265 | * |
| 266 | * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL |
| 267 | */ |
| 268 | static __always_inline void amd_set_core_ssb_state(unsigned long tifn) |
| 269 | { |
| 270 | struct ssb_state *st = this_cpu_ptr(&ssb_state); |
| 271 | u64 msr = x86_amd_ls_cfg_base; |
| 272 | |
| 273 | if (!static_cpu_has(X86_FEATURE_ZEN)) { |
| 274 | msr |= ssbd_tif_to_amd_ls_cfg(tifn); |
| 275 | wrmsrl(MSR_AMD64_LS_CFG, msr); |
| 276 | return; |
| 277 | } |
| 278 | |
| 279 | if (tifn & _TIF_SSBD) { |
| 280 | /* |
| 281 | * Since this can race with prctl(), block reentry on the |
| 282 | * same CPU. |
| 283 | */ |
| 284 | if (__test_and_set_bit(LSTATE_SSB, &st->local_state)) |
| 285 | return; |
| 286 | |
| 287 | msr |= x86_amd_ls_cfg_ssbd_mask; |
| 288 | |
| 289 | raw_spin_lock(&st->shared_state->lock); |
| 290 | /* First sibling enables SSBD: */ |
| 291 | if (!st->shared_state->disable_state) |
| 292 | wrmsrl(MSR_AMD64_LS_CFG, msr); |
| 293 | st->shared_state->disable_state++; |
| 294 | raw_spin_unlock(&st->shared_state->lock); |
| 295 | } else { |
| 296 | if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state)) |
| 297 | return; |
| 298 | |
| 299 | raw_spin_lock(&st->shared_state->lock); |
| 300 | st->shared_state->disable_state--; |
| 301 | if (!st->shared_state->disable_state) |
| 302 | wrmsrl(MSR_AMD64_LS_CFG, msr); |
| 303 | raw_spin_unlock(&st->shared_state->lock); |
| 304 | } |
| 305 | } |
| 306 | #else |
| 307 | static __always_inline void amd_set_core_ssb_state(unsigned long tifn) |
| 308 | { |
| 309 | u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn); |
| 310 | |
| 311 | wrmsrl(MSR_AMD64_LS_CFG, msr); |
| 312 | } |
| 313 | #endif |
| 314 | |
Tom Lendacky | 7c0b2dc | 2018-05-17 17:09:18 +0200 | [diff] [blame^] | 315 | static __always_inline void amd_set_ssb_virt_state(unsigned long tifn) |
| 316 | { |
| 317 | /* |
| 318 | * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL, |
| 319 | * so ssbd_tif_to_spec_ctrl() just works. |
| 320 | */ |
| 321 | wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn)); |
| 322 | } |
| 323 | |
Thomas Gleixner | d0cb78f | 2018-05-09 21:53:09 +0200 | [diff] [blame] | 324 | static __always_inline void intel_set_ssb_state(unsigned long tifn) |
| 325 | { |
| 326 | u64 msr = x86_spec_ctrl_base | ssbd_tif_to_spec_ctrl(tifn); |
| 327 | |
| 328 | wrmsrl(MSR_IA32_SPEC_CTRL, msr); |
| 329 | } |
| 330 | |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 331 | static __always_inline void __speculative_store_bypass_update(unsigned long tifn) |
| 332 | { |
Tom Lendacky | 7c0b2dc | 2018-05-17 17:09:18 +0200 | [diff] [blame^] | 333 | if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) |
| 334 | amd_set_ssb_virt_state(tifn); |
| 335 | else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) |
Thomas Gleixner | d0cb78f | 2018-05-09 21:53:09 +0200 | [diff] [blame] | 336 | amd_set_core_ssb_state(tifn); |
| 337 | else |
| 338 | intel_set_ssb_state(tifn); |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 339 | } |
| 340 | |
| 341 | void speculative_store_bypass_update(void) |
| 342 | { |
Thomas Gleixner | d0cb78f | 2018-05-09 21:53:09 +0200 | [diff] [blame] | 343 | preempt_disable(); |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 344 | __speculative_store_bypass_update(current_thread_info()->flags); |
Thomas Gleixner | d0cb78f | 2018-05-09 21:53:09 +0200 | [diff] [blame] | 345 | preempt_enable(); |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 346 | } |
| 347 | |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 348 | void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, |
| 349 | struct tss_struct *tss) |
| 350 | { |
| 351 | struct thread_struct *prev, *next; |
| 352 | unsigned long tifp, tifn; |
| 353 | |
| 354 | prev = &prev_p->thread; |
| 355 | next = &next_p->thread; |
| 356 | |
| 357 | tifn = READ_ONCE(task_thread_info(next_p)->flags); |
| 358 | tifp = READ_ONCE(task_thread_info(prev_p)->flags); |
| 359 | switch_to_bitmap(tss, prev, next, tifp, tifn); |
| 360 | |
Avi Kivity | 7c68af6 | 2009-09-19 09:40:22 +0300 | [diff] [blame] | 361 | propagate_user_return_notify(prev_p, next_p); |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 362 | |
Kyle Huey | 439f2ef8 | 2017-02-14 00:11:03 -0800 | [diff] [blame] | 363 | if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) && |
| 364 | arch_has_block_step()) { |
| 365 | unsigned long debugctl, msk; |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 366 | |
Kyle Huey | 439f2ef8 | 2017-02-14 00:11:03 -0800 | [diff] [blame] | 367 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 368 | debugctl &= ~DEBUGCTLMSR_BTF; |
Kyle Huey | 439f2ef8 | 2017-02-14 00:11:03 -0800 | [diff] [blame] | 369 | msk = tifn & _TIF_BLOCKSTEP; |
| 370 | debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT; |
| 371 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); |
Kyle Huey | fd01e82 | 2017-02-14 00:11:02 -0800 | [diff] [blame] | 372 | } |
| 373 | |
Thomas Gleixner | 5ed7788 | 2017-02-14 00:11:04 -0800 | [diff] [blame] | 374 | if ((tifp ^ tifn) & _TIF_NOTSC) |
| 375 | cr4_toggle_bits(X86_CR4_TSD); |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 376 | |
Konrad Rzeszutek Wilk | bf3da84 | 2018-05-09 21:41:38 +0200 | [diff] [blame] | 377 | if ((tifp ^ tifn) & _TIF_SSBD) |
Thomas Gleixner | 89c6e9b | 2018-04-29 15:21:42 +0200 | [diff] [blame] | 378 | __speculative_store_bypass_update(tifn); |
Jeremy Fitzhardinge | 389d1fb | 2009-02-27 13:25:28 -0800 | [diff] [blame] | 379 | } |
| 380 | |
Brian Gerst | df59e7b | 2009-12-09 12:34:44 -0500 | [diff] [blame] | 381 | /* |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 382 | * Idle related variables and functions |
| 383 | */ |
Thomas Renninger | d189604 | 2010-11-03 17:06:14 +0100 | [diff] [blame] | 384 | unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 385 | EXPORT_SYMBOL(boot_option_idle_override); |
| 386 | |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 387 | static void (*x86_idle)(void); |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 388 | |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 389 | #ifndef CONFIG_SMP |
| 390 | static inline void play_dead(void) |
| 391 | { |
| 392 | BUG(); |
| 393 | } |
| 394 | #endif |
| 395 | |
| 396 | #ifdef CONFIG_X86_64 |
| 397 | void enter_idle(void) |
| 398 | { |
Alex Shi | c6ae41e | 2012-05-11 15:35:27 +0800 | [diff] [blame] | 399 | this_cpu_write(is_idle, 1); |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 400 | atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL); |
| 401 | } |
| 402 | |
| 403 | static void __exit_idle(void) |
| 404 | { |
| 405 | if (x86_test_and_clear_bit_percpu(0, is_idle) == 0) |
| 406 | return; |
| 407 | atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL); |
| 408 | } |
| 409 | |
| 410 | /* Called from interrupts to signify idle end */ |
| 411 | void exit_idle(void) |
| 412 | { |
| 413 | /* idle loop has pid 0 */ |
| 414 | if (current->pid) |
| 415 | return; |
| 416 | __exit_idle(); |
| 417 | } |
| 418 | #endif |
| 419 | |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 420 | void arch_cpu_idle_enter(void) |
| 421 | { |
| 422 | local_touch_nmi(); |
| 423 | enter_idle(); |
| 424 | } |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 425 | |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 426 | void arch_cpu_idle_exit(void) |
| 427 | { |
| 428 | __exit_idle(); |
| 429 | } |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 430 | |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 431 | void arch_cpu_idle_dead(void) |
| 432 | { |
| 433 | play_dead(); |
Richard Weinberger | 90e2401 | 2012-03-25 23:00:04 +0200 | [diff] [blame] | 434 | } |
| 435 | |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 436 | /* |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 437 | * Called from the generic idle code. |
| 438 | */ |
| 439 | void arch_cpu_idle(void) |
| 440 | { |
Nicolas Pitre | 16f8b05 | 2014-01-29 12:45:12 -0500 | [diff] [blame] | 441 | x86_idle(); |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 442 | } |
| 443 | |
| 444 | /* |
| 445 | * We use this if we don't have any better idle routine.. |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 446 | */ |
Chris Metcalf | 6727ad9 | 2016-10-07 17:02:55 -0700 | [diff] [blame] | 447 | void __cpuidle default_idle(void) |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 448 | { |
Daniel Lezcano | 4d0e42c | 2012-10-25 18:13:11 +0200 | [diff] [blame] | 449 | trace_cpu_idle_rcuidle(1, smp_processor_id()); |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 450 | safe_halt(); |
Daniel Lezcano | 4d0e42c | 2012-10-25 18:13:11 +0200 | [diff] [blame] | 451 | trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 452 | } |
Andy Whitcroft | 60b8b1d | 2011-06-14 12:45:10 -0700 | [diff] [blame] | 453 | #ifdef CONFIG_APM_MODULE |
Thomas Gleixner | 00dba56 | 2008-06-09 18:35:28 +0200 | [diff] [blame] | 454 | EXPORT_SYMBOL(default_idle); |
| 455 | #endif |
| 456 | |
Len Brown | 6a377dd | 2013-02-09 23:08:07 -0500 | [diff] [blame] | 457 | #ifdef CONFIG_XEN |
| 458 | bool xen_set_default_idle(void) |
Konrad Rzeszutek Wilk | e5fd47b | 2011-11-21 18:02:02 -0500 | [diff] [blame] | 459 | { |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 460 | bool ret = !!x86_idle; |
Konrad Rzeszutek Wilk | e5fd47b | 2011-11-21 18:02:02 -0500 | [diff] [blame] | 461 | |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 462 | x86_idle = default_idle; |
Konrad Rzeszutek Wilk | e5fd47b | 2011-11-21 18:02:02 -0500 | [diff] [blame] | 463 | |
| 464 | return ret; |
| 465 | } |
Len Brown | 6a377dd | 2013-02-09 23:08:07 -0500 | [diff] [blame] | 466 | #endif |
Ivan Vecera | d3ec5ca | 2008-11-11 14:33:44 +0100 | [diff] [blame] | 467 | void stop_this_cpu(void *dummy) |
| 468 | { |
| 469 | local_irq_disable(); |
| 470 | /* |
| 471 | * Remove this CPU: |
| 472 | */ |
Rusty Russell | 4f06289 | 2009-03-13 14:49:54 +1030 | [diff] [blame] | 473 | set_cpu_online(smp_processor_id(), false); |
Ivan Vecera | d3ec5ca | 2008-11-11 14:33:44 +0100 | [diff] [blame] | 474 | disable_local_APIC(); |
Ashok Raj | 8838eb6 | 2015-08-12 18:29:40 +0200 | [diff] [blame] | 475 | mcheck_cpu_clear(this_cpu_ptr(&cpu_info)); |
Ivan Vecera | d3ec5ca | 2008-11-11 14:33:44 +0100 | [diff] [blame] | 476 | |
Len Brown | 27be457 | 2013-02-10 02:28:46 -0500 | [diff] [blame] | 477 | for (;;) |
| 478 | halt(); |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 479 | } |
| 480 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 481 | bool amd_e400_c1e_detected; |
| 482 | EXPORT_SYMBOL(amd_e400_c1e_detected); |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 483 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 484 | static cpumask_var_t amd_e400_c1e_mask; |
Thomas Gleixner | 4faac97 | 2008-09-22 18:54:29 +0200 | [diff] [blame] | 485 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 486 | void amd_e400_remove_cpu(int cpu) |
Thomas Gleixner | 4faac97 | 2008-09-22 18:54:29 +0200 | [diff] [blame] | 487 | { |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 488 | if (amd_e400_c1e_mask != NULL) |
| 489 | cpumask_clear_cpu(cpu, amd_e400_c1e_mask); |
Thomas Gleixner | 4faac97 | 2008-09-22 18:54:29 +0200 | [diff] [blame] | 490 | } |
| 491 | |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 492 | /* |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 493 | * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 494 | * pending message MSR. If we detect C1E, then we handle it the same |
| 495 | * way as C3 power states (local apic timer and TSC stop) |
| 496 | */ |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 497 | static void amd_e400_idle(void) |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 498 | { |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 499 | if (!amd_e400_c1e_detected) { |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 500 | u32 lo, hi; |
| 501 | |
| 502 | rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); |
Michal Schmidt | e8c534e | 2010-07-27 18:53:35 +0200 | [diff] [blame] | 503 | |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 504 | if (lo & K8_INTP_C1E_ACTIVE_MASK) { |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 505 | amd_e400_c1e_detected = true; |
Venki Pallipadi | 40fb171 | 2008-11-17 16:11:37 -0800 | [diff] [blame] | 506 | if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) |
Andreas Herrmann | 09bfeea | 2008-09-18 21:12:10 +0200 | [diff] [blame] | 507 | mark_tsc_unstable("TSC halt in AMD C1E"); |
Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 508 | pr_info("System has AMD C1E enabled\n"); |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 509 | } |
| 510 | } |
| 511 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 512 | if (amd_e400_c1e_detected) { |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 513 | int cpu = smp_processor_id(); |
| 514 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 515 | if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) { |
| 516 | cpumask_set_cpu(cpu, amd_e400_c1e_mask); |
Thomas Gleixner | 162a688 | 2015-04-03 02:01:28 +0200 | [diff] [blame] | 517 | /* Force broadcast so ACPI can not interfere. */ |
| 518 | tick_broadcast_force(); |
Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 519 | pr_info("Switch to broadcast mode on CPU%d\n", cpu); |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 520 | } |
Thomas Gleixner | 435c350 | 2015-04-03 02:05:53 +0200 | [diff] [blame] | 521 | tick_broadcast_enter(); |
Thomas Gleixner | 0beefa2 | 2008-06-17 09:12:03 +0200 | [diff] [blame] | 522 | |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 523 | default_idle(); |
Thomas Gleixner | 0beefa2 | 2008-06-17 09:12:03 +0200 | [diff] [blame] | 524 | |
| 525 | /* |
| 526 | * The switch back from broadcast mode needs to be |
| 527 | * called with interrupts disabled. |
| 528 | */ |
Peter Zijlstra | ea81174 | 2013-09-11 12:43:13 +0200 | [diff] [blame] | 529 | local_irq_disable(); |
Thomas Gleixner | 435c350 | 2015-04-03 02:05:53 +0200 | [diff] [blame] | 530 | tick_broadcast_exit(); |
Peter Zijlstra | ea81174 | 2013-09-11 12:43:13 +0200 | [diff] [blame] | 531 | local_irq_enable(); |
Thomas Gleixner | aa276e1 | 2008-06-09 19:15:00 +0200 | [diff] [blame] | 532 | } else |
| 533 | default_idle(); |
| 534 | } |
| 535 | |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 536 | /* |
| 537 | * Intel Core2 and older machines prefer MWAIT over HALT for C1. |
| 538 | * We can't rely on cpuidle installing MWAIT, because it will not load |
| 539 | * on systems that support only C1 -- so the boot default must be MWAIT. |
| 540 | * |
| 541 | * Some AMD machines are the opposite, they depend on using HALT. |
| 542 | * |
| 543 | * So for default C1, which is used during boot until cpuidle loads, |
| 544 | * use MWAIT-C1 on Intel HW that has it, else use HALT. |
| 545 | */ |
| 546 | static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c) |
| 547 | { |
| 548 | if (c->x86_vendor != X86_VENDOR_INTEL) |
| 549 | return 0; |
| 550 | |
Peter Zijlstra | 08e237f | 2016-07-18 11:41:10 -0700 | [diff] [blame] | 551 | if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR)) |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 552 | return 0; |
| 553 | |
| 554 | return 1; |
| 555 | } |
| 556 | |
| 557 | /* |
Huang Rui | 0fb0328 | 2015-05-26 10:28:09 +0200 | [diff] [blame] | 558 | * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT |
| 559 | * with interrupts enabled and no flags, which is backwards compatible with the |
| 560 | * original MWAIT implementation. |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 561 | */ |
Chris Metcalf | 6727ad9 | 2016-10-07 17:02:55 -0700 | [diff] [blame] | 562 | static __cpuidle void mwait_idle(void) |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 563 | { |
Mike Galbraith | f8e617f | 2014-01-18 17:14:44 +0100 | [diff] [blame] | 564 | if (!current_set_polling_and_test()) { |
Jisheng Zhang | e43d018 | 2015-08-20 12:54:39 +0800 | [diff] [blame] | 565 | trace_cpu_idle_rcuidle(1, smp_processor_id()); |
Mike Galbraith | f8e617f | 2014-01-18 17:14:44 +0100 | [diff] [blame] | 566 | if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) { |
Michael S. Tsirkin | ca59809 | 2016-01-28 19:02:51 +0200 | [diff] [blame] | 567 | mb(); /* quirk */ |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 568 | clflush((void *)¤t_thread_info()->flags); |
Michael S. Tsirkin | ca59809 | 2016-01-28 19:02:51 +0200 | [diff] [blame] | 569 | mb(); /* quirk */ |
Mike Galbraith | f8e617f | 2014-01-18 17:14:44 +0100 | [diff] [blame] | 570 | } |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 571 | |
| 572 | __monitor((void *)¤t_thread_info()->flags, 0, 0); |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 573 | if (!need_resched()) |
| 574 | __sti_mwait(0, 0); |
| 575 | else |
| 576 | local_irq_enable(); |
Jisheng Zhang | e43d018 | 2015-08-20 12:54:39 +0800 | [diff] [blame] | 577 | trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); |
Mike Galbraith | f8e617f | 2014-01-18 17:14:44 +0100 | [diff] [blame] | 578 | } else { |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 579 | local_irq_enable(); |
Mike Galbraith | f8e617f | 2014-01-18 17:14:44 +0100 | [diff] [blame] | 580 | } |
| 581 | __current_clr_polling(); |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 582 | } |
| 583 | |
Paul Gortmaker | 148f9bb | 2013-06-18 18:23:59 -0400 | [diff] [blame] | 584 | void select_idle_routine(const struct cpuinfo_x86 *c) |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 585 | { |
Ingo Molnar | 3e5095d | 2009-01-27 17:07:08 +0100 | [diff] [blame] | 586 | #ifdef CONFIG_SMP |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 587 | if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) |
Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 588 | pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n"); |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 589 | #endif |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 590 | if (x86_idle || boot_option_idle_override == IDLE_POLL) |
Thomas Gleixner | 6ddd2a2 | 2008-06-09 16:59:53 +0200 | [diff] [blame] | 591 | return; |
| 592 | |
Thomas Gleixner | bd7e769 | 2016-12-09 19:29:09 +0100 | [diff] [blame] | 593 | if (boot_cpu_has_bug(X86_BUG_AMD_E400)) { |
Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 594 | pr_info("using AMD E400 aware idle routine\n"); |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 595 | x86_idle = amd_e400_idle; |
Len Brown | b253149 | 2014-01-15 00:37:34 -0500 | [diff] [blame] | 596 | } else if (prefer_mwait_c1_over_halt(c)) { |
| 597 | pr_info("using mwait in idle threads\n"); |
| 598 | x86_idle = mwait_idle; |
Thomas Gleixner | 6ddd2a2 | 2008-06-09 16:59:53 +0200 | [diff] [blame] | 599 | } else |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 600 | x86_idle = default_idle; |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 601 | } |
| 602 | |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 603 | void __init init_amd_e400_c1e_mask(void) |
Rusty Russell | 30e1e6d | 2009-03-17 14:50:34 +1030 | [diff] [blame] | 604 | { |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 605 | /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */ |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 606 | if (x86_idle == amd_e400_idle) |
Len Brown | 02c68a0 | 2011-04-01 16:59:53 -0400 | [diff] [blame] | 607 | zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL); |
Rusty Russell | 30e1e6d | 2009-03-17 14:50:34 +1030 | [diff] [blame] | 608 | } |
| 609 | |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 610 | static int __init idle_setup(char *str) |
| 611 | { |
Cyrill Gorcunov | ab6bc3e | 2008-07-05 15:53:36 +0400 | [diff] [blame] | 612 | if (!str) |
| 613 | return -EINVAL; |
| 614 | |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 615 | if (!strcmp(str, "poll")) { |
Joe Perches | c767a54 | 2012-05-21 19:50:07 -0700 | [diff] [blame] | 616 | pr_info("using polling idle threads\n"); |
Thomas Renninger | d189604 | 2010-11-03 17:06:14 +0100 | [diff] [blame] | 617 | boot_option_idle_override = IDLE_POLL; |
Thomas Gleixner | 7d1a941 | 2013-03-21 22:50:03 +0100 | [diff] [blame] | 618 | cpu_idle_poll_ctrl(true); |
Thomas Renninger | d189604 | 2010-11-03 17:06:14 +0100 | [diff] [blame] | 619 | } else if (!strcmp(str, "halt")) { |
Zhao Yakui | c1e3b37 | 2008-06-24 17:58:53 +0800 | [diff] [blame] | 620 | /* |
| 621 | * When the boot option of idle=halt is added, halt is |
| 622 | * forced to be used for CPU idle. In such case CPU C2/C3 |
| 623 | * won't be used again. |
| 624 | * To continue to load the CPU idle driver, don't touch |
| 625 | * the boot_option_idle_override. |
| 626 | */ |
Len Brown | a476bda | 2013-02-09 21:45:03 -0500 | [diff] [blame] | 627 | x86_idle = default_idle; |
Thomas Renninger | d189604 | 2010-11-03 17:06:14 +0100 | [diff] [blame] | 628 | boot_option_idle_override = IDLE_HALT; |
Zhao Yakui | da5e09a | 2008-06-24 18:01:09 +0800 | [diff] [blame] | 629 | } else if (!strcmp(str, "nomwait")) { |
| 630 | /* |
| 631 | * If the boot option of "idle=nomwait" is added, |
| 632 | * it means that mwait will be disabled for CPU C2/C3 |
| 633 | * states. In such case it won't touch the variable |
| 634 | * of boot_option_idle_override. |
| 635 | */ |
Thomas Renninger | d189604 | 2010-11-03 17:06:14 +0100 | [diff] [blame] | 636 | boot_option_idle_override = IDLE_NOMWAIT; |
Zhao Yakui | c1e3b37 | 2008-06-24 17:58:53 +0800 | [diff] [blame] | 637 | } else |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 638 | return -1; |
| 639 | |
Peter Zijlstra | 7f424a8 | 2008-04-25 17:39:01 +0200 | [diff] [blame] | 640 | return 0; |
| 641 | } |
| 642 | early_param("idle", idle_setup); |
| 643 | |
Amerigo Wang | 9d62dcd | 2009-05-11 22:05:28 -0400 | [diff] [blame] | 644 | unsigned long arch_align_stack(unsigned long sp) |
| 645 | { |
| 646 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
| 647 | sp -= get_random_int() % 8192; |
| 648 | return sp & ~0xf; |
| 649 | } |
| 650 | |
| 651 | unsigned long arch_randomize_brk(struct mm_struct *mm) |
| 652 | { |
Jason Cooper | 9c6f090 | 2016-10-11 13:53:56 -0700 | [diff] [blame] | 653 | return randomize_page(mm->brk, 0x02000000); |
Amerigo Wang | 9d62dcd | 2009-05-11 22:05:28 -0400 | [diff] [blame] | 654 | } |
| 655 | |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 656 | /* |
Brian Gerst | ffcb043 | 2016-08-13 12:38:21 -0400 | [diff] [blame] | 657 | * Return saved PC of a blocked thread. |
| 658 | * What is this good for? it will be always the scheduler or ret_from_fork. |
| 659 | */ |
| 660 | unsigned long thread_saved_pc(struct task_struct *tsk) |
| 661 | { |
| 662 | struct inactive_task_frame *frame = |
| 663 | (struct inactive_task_frame *) READ_ONCE(tsk->thread.sp); |
| 664 | return READ_ONCE_NOCHECK(frame->ret_addr); |
| 665 | } |
| 666 | |
| 667 | /* |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 668 | * Called from fs/proc with a reference on @p to find the function |
| 669 | * which called into schedule(). This needs to be done carefully |
| 670 | * because the task might wake up and we might look at a stack |
| 671 | * changing under us. |
| 672 | */ |
| 673 | unsigned long get_wchan(struct task_struct *p) |
| 674 | { |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 675 | unsigned long start, bottom, top, sp, fp, ip, ret = 0; |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 676 | int count = 0; |
| 677 | |
| 678 | if (!p || p == current || p->state == TASK_RUNNING) |
| 679 | return 0; |
| 680 | |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 681 | if (!try_get_task_stack(p)) |
| 682 | return 0; |
| 683 | |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 684 | start = (unsigned long)task_stack_page(p); |
| 685 | if (!start) |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 686 | goto out; |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 687 | |
| 688 | /* |
| 689 | * Layout of the stack page: |
| 690 | * |
| 691 | * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long) |
| 692 | * PADDING |
| 693 | * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING |
| 694 | * stack |
Andy Lutomirski | 15f4eae | 2016-09-13 14:29:25 -0700 | [diff] [blame] | 695 | * ----------- bottom = start |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 696 | * |
| 697 | * The tasks stack pointer points at the location where the |
| 698 | * framepointer is stored. The data on the stack is: |
| 699 | * ... IP FP ... IP FP |
| 700 | * |
| 701 | * We need to read FP and IP, so we need to adjust the upper |
| 702 | * bound by another unsigned long. |
| 703 | */ |
| 704 | top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; |
| 705 | top -= 2 * sizeof(unsigned long); |
Andy Lutomirski | 15f4eae | 2016-09-13 14:29:25 -0700 | [diff] [blame] | 706 | bottom = start; |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 707 | |
| 708 | sp = READ_ONCE(p->thread.sp); |
| 709 | if (sp < bottom || sp > top) |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 710 | goto out; |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 711 | |
Brian Gerst | 7b32aea | 2016-08-13 12:38:18 -0400 | [diff] [blame] | 712 | fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp); |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 713 | do { |
| 714 | if (fp < bottom || fp > top) |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 715 | goto out; |
Andrey Ryabinin | f7d27c3 | 2015-10-19 11:37:18 +0300 | [diff] [blame] | 716 | ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long))); |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 717 | if (!in_sched_functions(ip)) { |
| 718 | ret = ip; |
| 719 | goto out; |
| 720 | } |
Andrey Ryabinin | f7d27c3 | 2015-10-19 11:37:18 +0300 | [diff] [blame] | 721 | fp = READ_ONCE_NOCHECK(*(unsigned long *)fp); |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 722 | } while (count++ < 16 && p->state != TASK_RUNNING); |
Andy Lutomirski | 74327a3 | 2016-09-15 22:45:46 -0700 | [diff] [blame] | 723 | |
| 724 | out: |
| 725 | put_task_stack(p); |
| 726 | return ret; |
Thomas Gleixner | 7ba7805 | 2015-09-30 08:38:23 +0000 | [diff] [blame] | 727 | } |