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.
12  * in lieu of the legacy ACPI processor_idle driver.  The intent is to
23  *	for preventing entry into deep C-states
25  * CPU will flush caches as needed when entering a C-state via MWAIT
26  *	(in contrast to entering ACPI C3, in which case the WBINVD
33  * ACPI has a .suspend hack to turn off deep c-statees during suspend
35  * Have not seen issues with suspend, but may need same workaround here.
39 /* un-comment DEBUG to enable pr_debug() statements */
47 #include <linux/tick.h>
54 #include <asm/intel-family.h>
65 static int max_cstate = CPUIDLE_STATE_MAX - 1;
77 	 * Hardware C-state auto-demotion may not always be optimal.
97  * MWAIT takes an 8-bit "hint" in EAX "suggesting"
98  * the C-state (top nibble) and sub-state (bottom nibble)
107  * intel_idle - Ask the processor to enter the given idle state.
115  * If the local APIC timer is not known to be reliable in the target idle state,
116  * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
126 	struct cpuidle_state *state = &drv->states[index];  in intel_idle()
127 	unsigned long eax = flg2MWAIT(state->flags);  in intel_idle()
136  * intel_idle_s2idle - Ask the processor to enter the given idle state.
144  * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
145  * scheduler tick and suspended scheduler clock on the target CPU.
150 	unsigned long eax = flg2MWAIT(drv->states[index].flags);  in intel_idle_s2idle()
1145 	unsigned long eax = flg2MWAIT(state->flags);  in intel_idle_state_needs_timer_stop()
1151 	 * Switch over to one-shot tick broadcast if the target C-state  in intel_idle_state_needs_timer_stop()
1164 static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1171  * intel_idle_cst_usable - Check if the _CST information can be used.
1173  * Check if all of the C-states listed by _CST in the max_cstate range are
1186 		if (cx->entry_method != ACPI_CSTATE_FFH)  in intel_idle_cst_usable()
1208 		if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))  in intel_idle_acpi_cst_extract()
1239 		if (intel_idle_max_cstate_reached(cstate - 1))  in intel_idle_init_cstates_acpi()
1244 		state = &drv->states[drv->state_count++];  in intel_idle_init_cstates_acpi()
1246 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);  in intel_idle_init_cstates_acpi()
1247 		strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);  in intel_idle_init_cstates_acpi()
1248 		state->exit_latency = cx->latency;  in intel_idle_init_cstates_acpi()
1250 		 * For C1-type C-states use the same number for both the exit  in intel_idle_init_cstates_acpi()
1252 		 * C1 in the majority of the static C-states tables above.  in intel_idle_init_cstates_acpi()
1253 		 * For the other types of C-states, however, set the target  in intel_idle_init_cstates_acpi()
1255 		 * a reasonable balance between energy-efficiency and  in intel_idle_init_cstates_acpi()
1256 		 * performance in the majority of interesting cases.  in intel_idle_init_cstates_acpi()
1258 		state->target_residency = cx->latency;  in intel_idle_init_cstates_acpi()
1259 		if (cx->type > ACPI_STATE_C1)  in intel_idle_init_cstates_acpi()
1260 			state->target_residency *= 3;  in intel_idle_init_cstates_acpi()
1262 		state->flags = MWAIT2flg(cx->address);  in intel_idle_init_cstates_acpi()
1263 		if (cx->type > ACPI_STATE_C2)  in intel_idle_init_cstates_acpi()
1264 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;  in intel_idle_init_cstates_acpi()
1267 			state->flags |= CPUIDLE_FLAG_OFF;  in intel_idle_init_cstates_acpi()
1270 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;  in intel_idle_init_cstates_acpi()
1272 		state->enter = intel_idle;  in intel_idle_init_cstates_acpi()
1273 		state->enter_s2idle = intel_idle_s2idle;  in intel_idle_init_cstates_acpi()
1282 	 * If there are no _CST C-states, do not disable any C-states by  in intel_idle_off_by_default()
1308  * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1310  * Tune IVT multi-socket targets.
1315 	/* IVT uses a different table for 1-2, 3-4, and > 4 sockets */  in ivt_idle_state_table_update()
1337  * irtl_2_usec - IRTL to microseconds conversion.
1358  * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1406  * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1408  * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1420 	/* if PC10 not present in CPUID.MWAIT.EDX */  in sklh_idle_state_table_update()
1426 	/* PC10 is not enabled in PKG C-state limit */  in sklh_idle_state_table_update()
1443 	skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE;	/* C8-SKL */  in sklh_idle_state_table_update()
1444 	skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE;	/* C9-SKL */  in sklh_idle_state_table_update()
1453 	/* Ignore the C-state if there are NO sub-states in CPUID for it. */  in intel_idle_verify_cstate()
1458 		mark_tsc_unstable("TSC halts in idle states deeper than C2");  in intel_idle_verify_cstate()
1502 		drv->states[drv->state_count] = cpuidle_state_table[cstate];  in intel_idle_init_cstates_icpu()
1504 		if ((disabled_states_mask & BIT(drv->state_count)) ||  in intel_idle_init_cstates_icpu()
1505 		    ((icpu->use_acpi || force_use_acpi) &&  in intel_idle_init_cstates_icpu()
1508 			drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;  in intel_idle_init_cstates_icpu()
1510 		if (intel_idle_state_needs_timer_stop(&drv->states[drv->state_count]))  in intel_idle_init_cstates_icpu()
1511 			drv->states[drv->state_count].flags |= CPUIDLE_FLAG_TIMER_STOP;  in intel_idle_init_cstates_icpu()
1513 		drv->state_count++;  in intel_idle_init_cstates_icpu()
1516 	if (icpu->byt_auto_demotion_disable_flag) {  in intel_idle_init_cstates_icpu()
1523  * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1531 		drv->states[0].flags |= CPUIDLE_FLAG_OFF;  in intel_idle_cpuidle_driver_init()
1533 	drv->state_count = 1;  in intel_idle_cpuidle_driver_init()
1560  * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
1563  * Register a cpuidle device object for @cpu and update its MSRs in accordance
1571 	dev->cpu = cpu;  in intel_idle_cpu_init()
1575 		return -EIO;  in intel_idle_cpu_init()
1597 	 * driver in this case  in intel_idle_cpu_online()
1600 	if (!dev->registered)  in intel_idle_cpu_online()
1607  * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
1625 		return -ENODEV;  in intel_idle_init()
1629 		return -EPERM;  in intel_idle_init()
1635 			pr_debug("Please enable MWAIT in BIOS SETUP\n");  in intel_idle_init()
1636 			return -ENODEV;  in intel_idle_init()
1641 			return -ENODEV;  in intel_idle_init()
1645 		return -ENODEV;  in intel_idle_init()
1652 			return -ENODEV;  in intel_idle_init()
1656 	icpu = (const struct idle_cpu *)id->driver_data;  in intel_idle_init()
1658 		cpuidle_state_table = icpu->state_table;  in intel_idle_init()
1659 		auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;  in intel_idle_init()
1660 		disable_promotion_to_c1e = icpu->disable_promotion_to_c1e;  in intel_idle_init()
1661 		if (icpu->use_acpi || force_use_acpi)  in intel_idle_init()
1664 		return -ENODEV;  in intel_idle_init()
1672 		return -ENOMEM;  in intel_idle_init()
1680 		       drv ? drv->name : "none");  in intel_idle_init()
1689 	pr_debug("Local APIC timer is reliable in %s\n",  in intel_idle_init()
1690 		 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");  in intel_idle_init()
1706  * support "intel_idle.max_cstate=..." at boot and also a read-only export of
1707  * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
1712  * The positions of the bits that are set in this number are the indices of the
1714  * corresponding idle state directories in sysfs, "state0", "state1" ...