1 // SPDX-License-Identifier: GPL-2.0-only
3  * intel_idle.c - native hardware idle loop for modern Intel processors
5  * Copyright (c) 2013 - 2020, Intel Corporation.
20  * All CPUs have same idle states as boot CPU
23  *	for preventing entry into deep C-states
25  * CPU will flush caches as needed when entering a C-state via MWAIT
33  * ACPI has a .suspend hack to turn off deep c-statees during suspend
39 /* un-comment DEBUG to enable pr_debug() statements */
56 #include <asm/intel-family.h>
58 #include <asm/spec-ctrl.h>
65 	.name = "intel_idle",
69 static int max_cstate = CPUIDLE_STATE_MAX - 1;
89 	 * Hardware C-state auto-demotion may not always be optimal.
104  * Enable interrupts before entering the C-state. On some platforms and for
105  * some C-states, this may measurably decrease interrupt latency.
110  * Enable this state by default even if the ACPI _CST does not list it.
115  * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
121  * Initialize large xstate for the C6-state entrance.
126  * Ignore the sub-state when matching mwait hints between the ACPI _CST and
132  * MWAIT takes an 8-bit "hint" in EAX "suggesting"
133  * the C-state (top nibble) and sub-state (bottom nibble)
136  * We store the hint at the top of our "flags" for each state.
145 	struct cpuidle_state *state = &drv->states[index];  in __intel_idle()  local
146 	unsigned long eax = flg2MWAIT(state->flags);  in __intel_idle()
155  * intel_idle - Ask the processor to enter the given idle state.
158  * @index: Target idle state index.
161  * @dev is idle and it can try to enter the idle state corresponding to @index.
163  * If the local APIC timer is not known to be reliable in the target idle state,
164  * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
206  * intel_idle_s2idle - Ask the processor to enter the given idle state.
209  * @index: Target idle state index.
212  * @dev is idle and it can try to enter the idle state corresponding to @index.
214  * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
221 	struct cpuidle_state *state = &drv->states[index];  in intel_idle_s2idle()  local
222 	unsigned long eax = flg2MWAIT(state->flags);  in intel_idle_s2idle()
224 	if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)  in intel_idle_s2idle()
239 		.name = "C1",
247 		.name = "C1E",
255 		.name = "C3",
263 		.name = "C6",
276 		.name = "C1",
284 		.name = "C1E",
292 		.name = "C3",
300 		.name = "C6",
308 		.name = "C7",
321 		.name = "C1",
329 		.name = "C6N",
337 		.name = "C6S",
345 		.name = "C7",
353 		.name = "C7S",
366 		.name = "C1",
374 		.name = "C6N",
382 		.name = "C6S",
390 		.name = "C7",
398 		.name = "C7S",
411 		.name = "C1",
419 		.name = "C1E",
427 		.name = "C3",
435 		.name = "C6",
443 		.name = "C7",
456 		.name = "C1",
464 		.name = "C1E",
472 		.name = "C3",
480 		.name = "C6",
493 		.name = "C1",
501 		.name = "C1E",
509 		.name = "C3",
517 		.name = "C6",
530 		.name = "C1",
538 		.name = "C1E",
546 		.name = "C3",
554 		.name = "C6",
567 		.name = "C1",
575 		.name = "C1E",
583 		.name = "C3",
591 		.name = "C6",
599 		.name = "C7s",
607 		.name = "C8",
615 		.name = "C9",
623 		.name = "C10",
635 		.name = "C1",
643 		.name = "C1E",
651 		.name = "C3",
659 		.name = "C6",
667 		.name = "C7s",
675 		.name = "C8",
683 		.name = "C9",
691 		.name = "C10",
704 		.name = "C1",
712 		.name = "C1E",
720 		.name = "C3",
728 		.name = "C6",
736 		.name = "C7s",
744 		.name = "C8",
752 		.name = "C9",
760 		.name = "C10",
773 		.name = "C1",
781 		.name = "C1E",
789 		.name = "C6",
802 		.name = "C1",
810 		.name = "C1E",
818 		.name = "C6",
841 		.name = "C1",
849 		.name = "C1E",
857 		.name = "C6",
865 		.name = "C8",
873 		.name = "C10",
886 		.name = "C1",
894 		.name = "C1E",
902 		.name = "C6",
910 		.name = "C8",
918 		.name = "C10",
931 		.name = "C1E",
939 		.name = "C6",
947 		.name = "C10",
960 		.name = "C1",
968 		.name = "C1E",
976 		.name = "C6",
984 		.name = "C8",
992 		.name = "C10",
1005 		.name = "C1",
1013 		.name = "C1E",
1021 		.name = "C6",
1035 		.name = "C1",
1043 		.name = "C1E",
1051 		.name = "C6",
1061 		.name = "C6P",
1076 		.name = "C1",
1084 		.name = "C1E",
1092 		.name = "C6",
1102 		.name = "C6P",
1117 		.name = "C1E",
1125 		.name = "C2",
1133 		.name = "C4",
1141 		.name = "C6",
1153 		.name = "C1",
1161 		.name = "C4",
1169 		.name = "C6",
1177 		.name = "C7",
1185 		.name = "C9",
1197 		.name = "C1",
1205 		.name = "C6",
1217 		.name = "C1",
1225 		.name = "C6",
1238 		.name = "C1",
1246 		.name = "C1E",
1254 		.name = "C6",
1262 		.name = "C7s",
1270 		.name = "C8",
1278 		.name = "C9",
1286 		.name = "C10",
1299 		.name = "C1",
1307 		.name = "C1E",
1315 		.name = "C6",
1332 		.name = "C1",
1340 		.name = "C1E",
1348 		.name = "C6",
1361 		.name = "C1",
1369 		.name = "C1E",
1377 		.name = "C6S",
1390 		.name = "C1",
1398 		.name = "C1E",
1406 		.name = "C6S",
1415 		.name = "C6SP",
1674 static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)  in intel_idle_state_needs_timer_stop()  argument
1676 	unsigned long eax = flg2MWAIT(state->flags);  in intel_idle_state_needs_timer_stop()
1682 	 * Switch over to one-shot tick broadcast if the target C-state  in intel_idle_state_needs_timer_stop()
1693 MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1697 MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1702  * intel_idle_cst_usable - Check if the _CST information can be used.
1704  * Check if all of the C-states listed by _CST in the max_cstate range are
1717 		if (cx->entry_method != ACPI_CSTATE_FFH)  in intel_idle_cst_usable()
1739 		if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))  in intel_idle_acpi_cst_extract()
1768 		struct cpuidle_state *state;  in intel_idle_init_cstates_acpi()  local
1770 		if (intel_idle_max_cstate_reached(cstate - 1))  in intel_idle_init_cstates_acpi()
1775 		state = &drv->states[drv->state_count++];  in intel_idle_init_cstates_acpi()
1777 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);  in intel_idle_init_cstates_acpi()
1778 		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);  in intel_idle_init_cstates_acpi()
1779 		state->exit_latency = cx->latency;  in intel_idle_init_cstates_acpi()
1781 		 * For C1-type C-states use the same number for both the exit  in intel_idle_init_cstates_acpi()
1783 		 * C1 in the majority of the static C-states tables above.  in intel_idle_init_cstates_acpi()
1784 		 * For the other types of C-states, however, set the target  in intel_idle_init_cstates_acpi()
1786 		 * a reasonable balance between energy-efficiency and  in intel_idle_init_cstates_acpi()
1789 		state->target_residency = cx->latency;  in intel_idle_init_cstates_acpi()
1790 		if (cx->type > ACPI_STATE_C1)  in intel_idle_init_cstates_acpi()
1791 			state->target_residency *= 3;  in intel_idle_init_cstates_acpi()
1793 		state->flags = MWAIT2flg(cx->address);  in intel_idle_init_cstates_acpi()
1794 		if (cx->type > ACPI_STATE_C2)  in intel_idle_init_cstates_acpi()
1795 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;  in intel_idle_init_cstates_acpi()
1798 			state->flags |= CPUIDLE_FLAG_OFF;  in intel_idle_init_cstates_acpi()
1800 		if (intel_idle_state_needs_timer_stop(state))  in intel_idle_init_cstates_acpi()
1801 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;  in intel_idle_init_cstates_acpi()
1803 		if (cx->type > ACPI_STATE_C1 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))  in intel_idle_init_cstates_acpi()
1804 			mark_tsc_unstable("TSC halts in idle");  in intel_idle_init_cstates_acpi()
1806 		state->enter = intel_idle;  in intel_idle_init_cstates_acpi()
1807 		state->enter_s2idle = intel_idle_s2idle;  in intel_idle_init_cstates_acpi()
1816 	 * If there are no _CST C-states, do not disable any C-states by  in intel_idle_off_by_default()
1853  * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1855  * Tune IVT multi-socket targets.
1860 	/* IVT uses a different table for 1-2, 3-4, and > 4 sockets */  in ivt_idle_state_table_update()
1882  * irtl_2_usec - IRTL to microseconds conversion.
1903  * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1951  * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1953  * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1971 	/* PC10 is not enabled in PKG C-state limit */  in sklh_idle_state_table_update()
1988 	skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE;	/* C8-SKL */  in sklh_idle_state_table_update()
1989 	skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE;	/* C9-SKL */  in sklh_idle_state_table_update()
1993  * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1994  * idle states table.
2003 	 * 000b: C0/C1 (no package C-state support)  in skx_idle_state_table_update()
2005 	 * 010b: C6 (non-retention)  in skx_idle_state_table_update()
2007 	 * 111b: No Package C state limits.  in skx_idle_state_table_update()
2023  * adl_idle_state_table_update - Adjust AlderLake idle states table.
2042  * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
2049 	 * By default, the C6 state assumes the worst-case scenario of package  in spr_idle_state_table_update()
2069 	/* Ignore the C-state if there are NO sub-states in CPUID for it. */  in intel_idle_verify_cstate()
2074 		mark_tsc_unstable("TSC halts in idle states deeper than C2");  in intel_idle_verify_cstate()
2079 static void state_update_enter_method(struct cpuidle_state *state, int cstate)  in state_update_enter_method()  argument
2081 	if (state->flags & CPUIDLE_FLAG_INIT_XSTATE) {  in state_update_enter_method()
2086 		WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IBRS);  in state_update_enter_method()
2087 		WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);  in state_update_enter_method()
2088 		state->enter = intel_idle_xstate;  in state_update_enter_method()
2093 			((state->flags & CPUIDLE_FLAG_IBRS) || ibrs_off)) {  in state_update_enter_method()
2095 		 * IBRS mitigation requires that C-states are entered  in state_update_enter_method()
2098 		if (ibrs_off && (state->flags & CPUIDLE_FLAG_IRQ_ENABLE))  in state_update_enter_method()
2099 			state->flags &= ~CPUIDLE_FLAG_IRQ_ENABLE;  in state_update_enter_method()
2100 		WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);  in state_update_enter_method()
2101 		state->enter = intel_idle_ibrs;  in state_update_enter_method()
2105 	if (state->flags & CPUIDLE_FLAG_IRQ_ENABLE) {  in state_update_enter_method()
2106 		state->enter = intel_idle_irq;  in state_update_enter_method()
2111 		pr_info("forced intel_idle_irq for state %d\n", cstate);  in state_update_enter_method()
2112 		state->enter = intel_idle_irq;  in state_update_enter_method()
2146 		struct cpuidle_state *state;  in intel_idle_init_cstates_icpu()  local
2156 		/* If marked as unusable, skip this state. */  in intel_idle_init_cstates_icpu()
2158 			pr_debug("state %s is disabled\n",  in intel_idle_init_cstates_icpu()
2159 				 cpuidle_state_table[cstate].name);  in intel_idle_init_cstates_icpu()
2168 		drv->states[drv->state_count] = cpuidle_state_table[cstate];  in intel_idle_init_cstates_icpu()
2169 		state = &drv->states[drv->state_count];  in intel_idle_init_cstates_icpu()
2171 		state_update_enter_method(state, cstate);  in intel_idle_init_cstates_icpu()
2174 		if ((disabled_states_mask & BIT(drv->state_count)) ||  in intel_idle_init_cstates_icpu()
2175 		    ((icpu->use_acpi || force_use_acpi) &&  in intel_idle_init_cstates_icpu()
2176 		     intel_idle_off_by_default(state->flags, mwait_hint) &&  in intel_idle_init_cstates_icpu()
2177 		     !(state->flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))  in intel_idle_init_cstates_icpu()
2178 			state->flags |= CPUIDLE_FLAG_OFF;  in intel_idle_init_cstates_icpu()
2180 		if (intel_idle_state_needs_timer_stop(state))  in intel_idle_init_cstates_icpu()
2181 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;  in intel_idle_init_cstates_icpu()
2183 		drv->state_count++;  in intel_idle_init_cstates_icpu()
2186 	if (icpu->byt_auto_demotion_disable_flag) {  in intel_idle_init_cstates_icpu()
2193  * intel_idle_cpuidle_driver_init - Create the list of available idle states.
2201 		drv->states[0].flags |= CPUIDLE_FLAG_OFF;  in intel_idle_cpuidle_driver_init()
2203 	drv->state_count = 1;  in intel_idle_cpuidle_driver_init()
2205 	if (icpu && icpu->state_table)  in intel_idle_cpuidle_driver_init()
2239  * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
2250 	dev->cpu = cpu;  in intel_idle_cpu_init()
2254 		return -EIO;  in intel_idle_cpu_init()
2281 	if (!dev->registered)  in intel_idle_cpu_online()
2288  * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2304 	/* Do not load intel_idle at all for now if idle= is passed */  in intel_idle_init()
2306 		return -ENODEV;  in intel_idle_init()
2310 		return -EPERM;  in intel_idle_init()
2317 			return -ENODEV;  in intel_idle_init()
2322 			return -ENODEV;  in intel_idle_init()
2330 			return -ENODEV;  in intel_idle_init()
2334 	icpu = (const struct idle_cpu *)id->driver_data;  in intel_idle_init()
2336 		if (icpu->state_table)  in intel_idle_init()
2337 			cpuidle_state_table = icpu->state_table;  in intel_idle_init()
2339 			return -ENODEV;  in intel_idle_init()
2341 		auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;  in intel_idle_init()
2342 		if (icpu->disable_promotion_to_c1e)  in intel_idle_init()
2344 		if (icpu->use_acpi || force_use_acpi)  in intel_idle_init()
2347 		return -ENODEV;  in intel_idle_init()
2355 		return -ENOMEM;  in intel_idle_init()
2363 		       drv ? drv->name : "none");  in intel_idle_init()
2367 	retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",  in intel_idle_init()
2373 		 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");  in intel_idle_init()
2389  * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2390  * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2396  * idle states to be disabled by default (as reflected by the names of the
2397  * corresponding idle state directories in sysfs, "state0", "state1" ...
2398  * "state<i>" ..., where <i> is the index of the given state).
2401 MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2403  * Some platforms come with mutually exclusive C-states, so that if one is
2404  * enabled, the other C-states must not be used. Example: C1 and C1E on
2406  * preferred C-states among the groups of mutually exclusive C-states - the
2407  * selected C-states will be registered, the other C-states from the mutually
2409  * exclusive C-states, this parameter has no effect.
2412 MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");
2414  * Debugging option that forces the driver to enter all C-states with
2415  * interrupts enabled. Does not apply to C-states with
2424 MODULE_PARM_DESC(ibrs_off, "Disable IBRS when idle");