Skip to contentSkip to conversation

cpuidle: intel_idle: Use ACPI _CTS to get idle states information

RFC PATCH New Rafael J. Wysocki 1 participant 10 messages Dec 10, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v2 cover cpuidle: intel_idle: Use ACPI _CTS to get idle states information

Hi All,

The purpose of this set of patches is to allow the intel_idle driver to use C-states information from ACPI _CST on systems where the processor is not recognized by it.

The first five patches are preparatory (please look into the changelogs for details) and are not expected to make any functional difference.

The patch [6/9] adds ACPI _CST support to intel_idle so that _CST is used when the driver does not have a dedicated list of C-states for the given processor.

Patch [7/9] is the same as https://patchwork.kernel.org/patch/11256815/.

Patch [8/9] changes intel_idle to also use ACPI _CST if there is a tables of C-states for the given processor in the driver, in which case it will use the _CST information to decide which C-state to enable by default.

The last patch adds a module parameter called "no_acpi" that can be used to prevent intel_idle from using ACPI _CST via the kernel command line.

This has been lightly tested on a Dell XPS13 9360 (with an additional patch to drop the Kaby Lake entries from intel_idle_ids[] which normally would prevent intel_idle from registering itself as the cpuidle driver). The difference between using the idle states list from _CST and the built-in one generally is that in the latter case the processor spends more time in package C-state when the system is idle.

If there are any concerns about this series, please let me know.

Thanks,
Rafael

#1 lore raw

RFC PATCH v2 1/9 ACPI: processor: Export function to claim _CST control

From: Rafael J. Wysocki <[email protected]>

The intel_idle driver will be modified to use ACPI _CST subsequently and it will need to notify the platform firmware of that if acpi_gbl_FADT.cst_control is set, so add a routine for this purpose, acpi_processor_claim_cst_control(), to acpi_processor.c (so that it is always present which is required by intel_idle) and export it to allow the ACPI processor driver (which is modular) to call it.

No intentional functional impact.

---

No changes from the previous version.

---
 drivers/acpi/acpi_processor.c |   25 +++++++++++++++++++++++++
 drivers/acpi/processor_idle.c |   12 ++++--------
 include/linux/acpi.h          |    6 ++++++
 3 files changed, 35 insertions(+), 8 deletions(-)
drivers/acpi/acpi_processor.c +25 -0
Index: linux-pm/drivers/acpi/acpi_processor.c
===================================================================
@@ -705,3 +705,28 @@ void __init acpi_processor_init(void)
 	acpi_scan_add_handler_with_hotplug(&processor_handler, "processor");
 	acpi_scan_add_handler(&processor_container_handler);
 }
+
+#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
+/**
+ * acpi_processor_claim_cst_control - Request _CST control from the platform.
+ */
+bool acpi_processor_claim_cst_control(void)
+{
+	static bool cst_control_claimed;
+	acpi_status status;
+
+	if (!acpi_gbl_FADT.cst_control || cst_control_claimed)
+		return true;
+
+	status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
+				    acpi_gbl_FADT.cst_control, 8);
+	if (ACPI_FAILURE(status)) {
+		pr_warn("ACPI: Failed to claim processor _CST control\n");
+		return false;
+	}
+
+	cst_control_claimed = true;
+	return true;
+}
+EXPORT_SYMBOL_GPL(acpi_processor_claim_cst_control);
+#endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
drivers/acpi/processor_idle.c +4 -8
Index: linux-pm/drivers/acpi/processor_idle.c
===================================================================
@@ -909,7 +909,6 @@ static int acpi_processor_setup_cstates(
 
 static inline void acpi_processor_cstate_first_run_checks(void)
 {
-	acpi_status status;
 	static int first_run;
 
 	if (first_run)
@@ -921,13 +920,10 @@ static inline void acpi_processor_cstate
 			  max_cstate);
 	first_run++;
 
-	if (acpi_gbl_FADT.cst_control && !nocst) {
-		status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
-					    acpi_gbl_FADT.cst_control, 8);
-		if (ACPI_FAILURE(status))
-			ACPI_EXCEPTION((AE_INFO, status,
-					"Notifying BIOS of _CST ability failed"));
-	}
+	if (nocst)
+		return;
+
+	acpi_processor_claim_cst_control();
 }
 #else
 
include/linux/acpi.h +6 -0
Index: linux-pm/include/linux/acpi.h
===================================================================
@@ -279,6 +279,12 @@ static inline bool invalid_phys_cpuid(ph
 
 /* Validate the processor object's proc_id */
 bool acpi_duplicate_processor_id(int proc_id);
+/* Processor _CTS control */
+#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
+bool acpi_processor_claim_cst_control(void);
+#else
+static inline bool acpi_processor_claim_cst_control(void) { return false; }
+#endif
 
 #ifdef CONFIG_ACPI_HOTPLUG_CPU
 /* Arch dependent functions for cpu hotplug support */
#2 lore raw

RFC PATCH v2 2/9 ACPI: processor: Introduce acpi_processor_evaluate_cst()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

In order to separate the ACPI _CST evaluation from checks specific to the ACPI processor driver, move the majority of the acpi_processor_get_power_info_cst() function body to a new function, acpi_processor_evaluate_cst(), that will extract the C-states information from _CST output, and redefine acpi_processor_get_power_info_cst() as a wrapper around it.

No intentional functional impact.

---

No changes from the previous version.

---
 drivers/acpi/processor_idle.c |   52 +++++++++++++++++++++++++-----------------
 1 file changed, 32 insertions(+), 20 deletions(-)
drivers/acpi/processor_idle.c +32 -20
Index: linux-pm/drivers/acpi/processor_idle.c
===================================================================
@@ -297,21 +297,17 @@ static int acpi_processor_get_power_info
 	return 0;
 }
 
-static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
+static int acpi_processor_evaluate_cst(acpi_handle handle, u32 cpu,
+				       struct acpi_processor_power *info)
 {
+	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *cst;
 	acpi_status status;
 	u64 count;
-	int current_count;
+	int current_count = 0;
 	int i, ret = 0;
-	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
-	union acpi_object *cst;
-
-	if (nocst)
-		return -ENODEV;
-
-	current_count = 0;
 
-	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
+	status = acpi_evaluate_object(handle, "_CST", NULL, &buffer);
 	if (ACPI_FAILURE(status)) {
 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 		return -ENODEV;
@@ -335,9 +331,6 @@ static int acpi_processor_get_power_info
 		goto end;
 	}
 
-	/* Tell driver that at least _CST is supported. */
-	pr->flags.has_cst = 1;
-
 	for (i = 1; i <= count; i++) {
 		union acpi_object *element;
 		union acpi_object *obj;
@@ -383,7 +376,7 @@ static int acpi_processor_get_power_info
 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 			if (acpi_processor_ffh_cstate_probe
-					(pr->id, &cx, reg) == 0) {
+					(cpu, &cx, reg) == 0) {
 				cx.entry_method = ACPI_CSTATE_FFH;
 			} else if (cx.type == ACPI_STATE_C1) {
 				/*
@@ -432,7 +425,7 @@ static int acpi_processor_get_power_info
 			continue;
 
 		current_count++;
-		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
+		memcpy(&info->states[current_count], &cx, sizeof(cx));
 
 		/*
 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
@@ -446,12 +439,9 @@ static int acpi_processor_get_power_info
 		}
 	}
 
-	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
-			  current_count));
+	acpi_handle_info(handle, "Found %d idle states\n", current_count);
 
-	/* Validate number of power states discovered */
-	if (current_count < 2)
-		ret = -EFAULT;
+	info->count = current_count;
 
       end:
 	kfree(buffer.pointer);
@@ -459,6 +449,28 @@ static int acpi_processor_get_power_info
 	return ret;
 }
 
+static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
+{
+	int ret;
+
+	if (nocst)
+		return -ENODEV;
+
+	ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
+	if (ret)
+		return ret;
+
+	/*
+	 * It is expected that there will be at least 2 states, C1 and
+	 * something else (C2 or C3), so fail if that is not the case.
+	 */
+	if (pr->power.count < 2)
+		return -EFAULT;
+
+	pr->flags.has_cst = 1;
+	return 0;
+}
+
 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 					   struct acpi_processor_cx *cx)
 {
#3 lore raw

RFC PATCH v2 3/9 ACPI: processor: Clean up acpi_processor_evaluate_cst()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Clean up acpi_processor_evaluate_cst() in multiple ways:

* Rename current_count to last_index which matches the purpose of the variable better.

* Consistently use acpi_handle_*() for printing messages and make the messages cleaner.

* Drop redundant parens and braces.

* Rewrite and clarify comments.

* Rearrange checks and drop the redundant ones.

No intentional functional impact.

---

No changes from the previous version.

---
 drivers/acpi/processor_idle.c |  114 +++++++++++++++++++-----------------------
 1 file changed, 52 insertions(+), 62 deletions(-)
drivers/acpi/processor_idle.c +52 -62
Index: linux-pm/drivers/acpi/processor_idle.c
===================================================================
@@ -304,29 +304,29 @@ static int acpi_processor_evaluate_cst(a
 	union acpi_object *cst;
 	acpi_status status;
 	u64 count;
-	int current_count = 0;
+	int last_index = 0;
 	int i, ret = 0;
 
 	status = acpi_evaluate_object(handle, "_CST", NULL, &buffer);
 	if (ACPI_FAILURE(status)) {
-		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
+		acpi_handle_debug(handle, "No _CST\n");
 		return -ENODEV;
 	}
 
 	cst = buffer.pointer;
 
-	/* There must be at least 2 elements */
-	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
-		pr_err("not enough elements in _CST\n");
+	/* There must be at least 2 elements. */
+	if (!cst || cst->type != ACPI_TYPE_PACKAGE || cst->package.count < 2) {
+		acpi_handle_warn(handle, "Invalid _CST output\n");
 		ret = -EFAULT;
 		goto end;
 	}
 
 	count = cst->package.elements[0].integer.value;
 
-	/* Validate number of power states. */
+	/* Validate the number of C-states. */
 	if (count < 1 || count != cst->package.count - 1) {
-		pr_err("count given by _CST is not valid\n");
+		acpi_handle_warn(handle, "Inconsistent _CST data\n");
 		ret = -EFAULT;
 		goto end;
 	}
@@ -337,111 +337,101 @@ static int acpi_processor_evaluate_cst(a
 		struct acpi_power_register *reg;
 		struct acpi_processor_cx cx;
 
+		/*
+		 * If there is not enough space for all C-states, skip the
+		 * excess ones and log a warning.
+		 */
+		if (last_index >= ACPI_PROCESSOR_MAX_POWER - 1) {
+			acpi_handle_warn(handle,
+					 "No room for more idle states (limit: %d)\n",
+					 ACPI_PROCESSOR_MAX_POWER - 1);
+			break;
+		}
+
 		memset(&cx, 0, sizeof(cx));
 
-		element = &(cst->package.elements[i]);
+		element = &cst->package.elements[i];
 		if (element->type != ACPI_TYPE_PACKAGE)
 			continue;
 
 		if (element->package.count != 4)
 			continue;
 
-		obj = &(element->package.elements[0]);
+		obj = &element->package.elements[0];
 
 		if (obj->type != ACPI_TYPE_BUFFER)
 			continue;
 
 		reg = (struct acpi_power_register *)obj->buffer.pointer;
 
-		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
-		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
-			continue;
-
-		/* There should be an easy way to extract an integer... */
-		obj = &(element->package.elements[1]);
+		obj = &element->package.elements[1];
 		if (obj->type != ACPI_TYPE_INTEGER)
 			continue;
 
 		cx.type = obj->integer.value;
 		/*
-		 * Some buggy BIOSes won't list C1 in _CST -
-		 * Let acpi_processor_get_power_info_default() handle them later
+		 * There are known cases in which the _CST output does not
+		 * contain C1, so if the type of the first state found is not
+		 * C1, leave an empty slot for C1 to be filled in later.
 		 */
 		if (i == 1 && cx.type != ACPI_STATE_C1)
-			current_count++;
+			last_index = 1;
 
 		cx.address = reg->address;
-		cx.index = current_count + 1;
+		cx.index = last_index + 1;
 
-		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
-			if (acpi_processor_ffh_cstate_probe
-					(cpu, &cx, reg) == 0) {
-				cx.entry_method = ACPI_CSTATE_FFH;
+			if (!acpi_processor_ffh_cstate_probe(cpu, &cx, reg)) {
+				/*
+				 * In the majority of cases _CST describes C1 as
+				 * a FIXED_HARDWARE C-state, but if the command
+				 * line forbids using MWAIT, use CSTATE_HALT for
+				 * C1 regardless.
+				 */
+				if (cx.type == ACPI_STATE_C1 &&
+				    boot_option_idle_override == IDLE_NOMWAIT) {
+					cx.entry_method = ACPI_CSTATE_HALT;
+					snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
+				} else {
+					cx.entry_method = ACPI_CSTATE_FFH;
+				}
 			} else if (cx.type == ACPI_STATE_C1) {
 				/*
-				 * C1 is a special case where FIXED_HARDWARE
-				 * can be handled in non-MWAIT way as well.
-				 * In that case, save this _CST entry info.
-				 * Otherwise, ignore this info and continue.
+				 * In the special case of C1, FIXED_HARDWARE can
+				 * be handled by executing the HLT instruction.
 				 */
 				cx.entry_method = ACPI_CSTATE_HALT;
 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 			} else {
 				continue;
 			}
-			if (cx.type == ACPI_STATE_C1 &&
-			    (boot_option_idle_override == IDLE_NOMWAIT)) {
-				/*
-				 * In most cases the C1 space_id obtained from
-				 * _CST object is FIXED_HARDWARE access mode.
-				 * But when the option of idle=halt is added,
-				 * the entry_method type should be changed from
-				 * CSTATE_FFH to CSTATE_HALT.
-				 * When the option of idle=nomwait is added,
-				 * the C1 entry_method type should be
-				 * CSTATE_HALT.
-				 */
-				cx.entry_method = ACPI_CSTATE_HALT;
-				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
-			}
-		} else {
+		} else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
+			cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 				 cx.address);
+		} else {
+			continue;
 		}
 
-		if (cx.type == ACPI_STATE_C1) {
+		if (cx.type == ACPI_STATE_C1)
 			cx.valid = 1;
-		}
 
-		obj = &(element->package.elements[2]);
+		obj = &element->package.elements[2];
 		if (obj->type != ACPI_TYPE_INTEGER)
 			continue;
 
 		cx.latency = obj->integer.value;
 
-		obj = &(element->package.elements[3]);
+		obj = &element->package.elements[3];
 		if (obj->type != ACPI_TYPE_INTEGER)
 			continue;
 
-		current_count++;
-		memcpy(&info->states[current_count], &cx, sizeof(cx));
-
-		/*
-		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
-		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
-		 */
-		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
-			pr_warn("Limiting number of power states to max (%d)\n",
-				ACPI_PROCESSOR_MAX_POWER);
-			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
-			break;
-		}
+		memcpy(&info->states[++last_index], &cx, sizeof(cx));
 	}
 
-	acpi_handle_info(handle, "Found %d idle states\n", current_count);
+	acpi_handle_info(handle, "Found %d idle states\n", last_index);
 
-	info->count = current_count;
+	info->count = last_index;
 
       end:
 	kfree(buffer.pointer);
#4 lore raw

RFC PATCH v2 4/9 ACPI: processor: Export acpi_processor_evaluate_cst()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

The intel_idle driver will be modified to use ACPI _CST subsequently and it will need to call acpi_processor_evaluate_cst(), so move that function to acpi_processor.c so that it is always present (which is required by intel_idle) and export it to modules to allow the ACPI processor driver (which is modular) to call it.

No intentional functional impact.

---

No changes from the previous version.

---
 drivers/acpi/acpi_processor.c |  154 ++++++++++++++++++++++++++++++++++++++++++
 drivers/acpi/processor_idle.c |  139 -------------------------------------
 include/linux/acpi.h          |    9 ++
 3 files changed, 163 insertions(+), 139 deletions(-)
drivers/acpi/acpi_processor.c +157 -0
Index: linux-pm/drivers/acpi/acpi_processor.c
===================================================================
@@ -729,4 +729,161 @@ bool acpi_processor_claim_cst_control(vo
 	return true;
 }
 EXPORT_SYMBOL_GPL(acpi_processor_claim_cst_control);
+
+/**
+ * acpi_processor_evaluate_cst - Evaluate the processor _CST control method.
+ * @handle: ACPI handle of the processor object containing the _CST.
+ * @cpu: The numeric ID of the target CPU.
+ * @info: Object write the C-states information into.
+ *
+ * Extract the C-state information for the given CPU from the output of the _CST
+ * control method under the corresponding ACPI processor object (or processor
+ * device object) and populate @info with it.
+ *
+ * If any ACPI_ADR_SPACE_FIXED_HARDWARE C-states are found, invoke
+ * acpi_processor_ffh_cstate_probe() to verify them and update the
+ * cpu_cstate_entry data for @cpu.
+ */
+int acpi_processor_evaluate_cst(acpi_handle handle, u32 cpu,
+				struct acpi_processor_power *info)
+{
+	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *cst;
+	acpi_status status;
+	u64 count;
+	int last_index = 0;
+	int i, ret = 0;
+
+	status = acpi_evaluate_object(handle, "_CST", NULL, &buffer);
+	if (ACPI_FAILURE(status)) {
+		acpi_handle_debug(handle, "No _CST\n");
+		return -ENODEV;
+	}
+
+	cst = buffer.pointer;
+
+	/* There must be at least 2 elements. */
+	if (!cst || cst->type != ACPI_TYPE_PACKAGE || cst->package.count < 2) {
+		acpi_handle_warn(handle, "Invalid _CST output\n");
+		ret = -EFAULT;
+		goto end;
+	}
+
+	count = cst->package.elements[0].integer.value;
+
+	/* Validate the number of C-states. */
+	if (count < 1 || count != cst->package.count - 1) {
+		acpi_handle_warn(handle, "Inconsistent _CST data\n");
+		ret = -EFAULT;
+		goto end;
+	}
+
+	for (i = 1; i <= count; i++) {
+		union acpi_object *element;
+		union acpi_object *obj;
+		struct acpi_power_register *reg;
+		struct acpi_processor_cx cx;
+
+		/*
+		 * If there is not enough space for all C-states, skip the
+		 * excess ones and log a warning.
+		 */
+		if (last_index >= ACPI_PROCESSOR_MAX_POWER - 1) {
+			acpi_handle_warn(handle,
+					 "No room for more idle states (limit: %d)\n",
+					 ACPI_PROCESSOR_MAX_POWER - 1);
+			break;
+		}
+
+		memset(&cx, 0, sizeof(cx));
+
+		element = &cst->package.elements[i];
+		if (element->type != ACPI_TYPE_PACKAGE)
+			continue;
+
+		if (element->package.count != 4)
+			continue;
+
+		obj = &element->package.elements[0];
+
+		if (obj->type != ACPI_TYPE_BUFFER)
+			continue;
+
+		reg = (struct acpi_power_register *)obj->buffer.pointer;
+
+		obj = &element->package.elements[1];
+		if (obj->type != ACPI_TYPE_INTEGER)
+			continue;
+
+		cx.type = obj->integer.value;
+		/*
+		 * There are known cases in which the _CST output does not
+		 * contain C1, so if the type of the first state found is not
+		 * C1, leave an empty slot for C1 to be filled in later.
+		 */
+		if (i == 1 && cx.type != ACPI_STATE_C1)
+			last_index = 1;
+
+		cx.address = reg->address;
+		cx.index = last_index + 1;
+
+		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
+			if (!acpi_processor_ffh_cstate_probe(cpu, &cx, reg)) {
+				/*
+				 * In the majority of cases _CST describes C1 as
+				 * a FIXED_HARDWARE C-state, but if the command
+				 * line forbids using MWAIT, use CSTATE_HALT for
+				 * C1 regardless.
+				 */
+				if (cx.type == ACPI_STATE_C1 &&
+				    boot_option_idle_override == IDLE_NOMWAIT) {
+					cx.entry_method = ACPI_CSTATE_HALT;
+					snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
+				} else {
+					cx.entry_method = ACPI_CSTATE_FFH;
+				}
+			} else if (cx.type == ACPI_STATE_C1) {
+				/*
+				 * In the special case of C1, FIXED_HARDWARE can
+				 * be handled by executing the HLT instruction.
+				 */
+				cx.entry_method = ACPI_CSTATE_HALT;
+				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
+			} else {
+				continue;
+			}
+		} else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
+			cx.entry_method = ACPI_CSTATE_SYSTEMIO;
+			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
+				 cx.address);
+		} else {
+			continue;
+		}
+
+		if (cx.type == ACPI_STATE_C1)
+			cx.valid = 1;
+
+		obj = &element->package.elements[2];
+		if (obj->type != ACPI_TYPE_INTEGER)
+			continue;
+
+		cx.latency = obj->integer.value;
+
+		obj = &element->package.elements[3];
+		if (obj->type != ACPI_TYPE_INTEGER)
+			continue;
+
+		memcpy(&info->states[++last_index], &cx, sizeof(cx));
+	}
+
+	acpi_handle_info(handle, "Found %d idle states\n", last_index);
+
+	info->count = last_index;
+
+      end:
+	kfree(buffer.pointer);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(acpi_processor_evaluate_cst);
 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
include/linux/acpi.h +9 -0
Index: linux-pm/include/linux/acpi.h
===================================================================
@@ -280,10 +280,19 @@ static inline bool invalid_phys_cpuid(ph
 /* Validate the processor object's proc_id */
 bool acpi_duplicate_processor_id(int proc_id);
 /* Processor _CTS control */
+struct acpi_processor_power;
+
 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
 bool acpi_processor_claim_cst_control(void);
+int acpi_processor_evaluate_cst(acpi_handle handle, u32 cpu,
+				struct acpi_processor_power *info);
 #else
 static inline bool acpi_processor_claim_cst_control(void) { return false; }
+static inline int acpi_processor_evaluate_cst(acpi_handle handle, u32 cpu,
+					      struct acpi_processor_power *info)
+{
+	return -ENODEV;
+}
 #endif
 
 #ifdef CONFIG_ACPI_HOTPLUG_CPU
drivers/acpi/processor_idle.c +0 -142

148 lines not on this page · raw message

#5 lore raw

RFC PATCH v2 5/9 intel_idle: Refactor intel_idle_cpuidle_driver_init()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move the C-state verification and checks from intel_idle_cpuidle_driver_init() to a separate function, intel_idle_verify_cstate(), and make the former call it after checking the CPUIDLE_FLAG_UNUSABLE state flag.

Also combine the drv->states[] updates with the incrementation of drv->state_count.

No intentional functional impact.

---

No changes from the previous version.

---
 drivers/idle/intel_idle.c |   49 +++++++++++++++++++++++-----------------------
 1 file changed, 25 insertions(+), 24 deletions(-)
drivers/idle/intel_idle.c +25 -24
Index: linux-pm/drivers/idle/intel_idle.c
===================================================================
@@ -944,6 +944,22 @@ static void intel_idle_s2idle(struct cpu
 	mwait_idle_with_hints(eax, ecx);
 }
 
+static bool intel_idle_verify_cstate(unsigned int mwait_hint)
+{
+	unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
+	unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
+					MWAIT_SUBSTATE_MASK;
+
+	/* Ignore the C-state if there are NO sub-states in CPUID for it. */
+	if (num_substates == 0)
+		return false;
+
+	if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
+		mark_tsc_unstable("TSC halts in idle states deeper than C2");
+
+	return true;
+}
+
 static void __setup_broadcast_timer(bool on)
 {
 	if (on)
@@ -1332,10 +1348,10 @@ static void __init intel_idle_cpuidle_dr
 	drv->state_count = 1;
 
 	for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
-		int num_substates, mwait_hint, mwait_cstate;
+		unsigned int mwait_hint;
 
-		if ((cpuidle_state_table[cstate].enter == NULL) &&
-		    (cpuidle_state_table[cstate].enter_s2idle == NULL))
+		if (!cpuidle_state_table[cstate].enter &&
+		    !cpuidle_state_table[cstate].enter_s2idle)
 			break;
 
 		if (cstate + 1 > max_cstate) {
@@ -1343,34 +1359,19 @@ static void __init intel_idle_cpuidle_dr
 			break;
 		}
 
-		mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
-		mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
-
-		/* number of sub-states for this state in CPUID.MWAIT */
-		num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
-					& MWAIT_SUBSTATE_MASK;
-
-		/* if NO sub-states for this state in CPUID, skip it */
-		if (num_substates == 0)
-			continue;
-
-		/* if state marked as disabled, skip it */
+		/* If marked as unusable, skip this state. */
 		if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
 			pr_debug("state %s is disabled\n",
 				 cpuidle_state_table[cstate].name);
 			continue;
 		}
 
+		mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
+		if (!intel_idle_verify_cstate(mwait_hint))
+			continue;
 
-		if (((mwait_cstate + 1) > 2) &&
-			!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
-			mark_tsc_unstable("TSC halts in idle"
-					" states deeper than C2");
-
-		drv->states[drv->state_count] =	/* structure copy */
-			cpuidle_state_table[cstate];
-
-		drv->state_count += 1;
+		/* Structure copy. */
+		drv->states[drv->state_count++] = cpuidle_state_table[cstate];
 	}
 
 	if (icpu->byt_auto_demotion_disable_flag) {
#6 lore raw

RFC PATCH v2 6/9 intel_idle: Use ACPI _CST for processor models without C-state tables

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Modify the intel_idle driver to get the C-states information from ACPI _CST if the processor model is not recognized by it.

The processor is still required to support MWAIT and the information from ACPI _CST will only be used if all of the C-states listed by _CST are of the ACPI_CSTATE_FFH type (which means that they are expected to be entered via MWAIT).

Moreover, the driver assumes that the _CST information is the same for all CPUs in the system, so it is sufficient to evaluate _CST for one of them and extract the common list of C-states from there. Also _CST is evaluated once at the system initialization time and the driver does not respond to _CST change notifications (that can be changed in the future).

---

Changes from the previous version:

- Append "_ACPI" to the names of the C-states coming from _CST.
- For ACPI C-states types other than C1 set target_residency as 4 times the exit latency (which is closer to the numbers used by intel_idle for "known" processors).

---
 drivers/idle/intel_idle.c |  181 ++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 153 insertions(+), 28 deletions(-)
drivers/idle/intel_idle.c +153 -28
Index: linux-pm/drivers/idle/intel_idle.c
===================================================================
@@ -41,6 +41,7 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
+#include <linux/acpi.h>
 #include <linux/kernel.h>
 #include <linux/cpuidle.h>
 #include <linux/tick.h>
@@ -1111,6 +1112,120 @@ static const struct x86_cpu_id intel_idl
 	{}
 };
 
+#define INTEL_CPU_FAM6_MWAIT \
+	{ X86_VENDOR_INTEL, 6, X86_MODEL_ANY, X86_FEATURE_MWAIT, 0 }
+
+static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
+	INTEL_CPU_FAM6_MWAIT,
+	{}
+};
+
+static bool intel_idle_max_cstate_reached(int cstate)
+{
+	if (cstate + 1 > max_cstate) {
+		pr_info("max_cstate %d reached\n", max_cstate);
+		return true;
+	}
+	return false;
+}
+
+#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
+#include <acpi/processor.h>
+
+static struct acpi_processor_power acpi_state_table;
+
+/**
+ * intel_idle_cst_usable - Check if the _CST information can be used.
+ *
+ * Check if all of the C-states listed by _CST in the max_cstate range are
+ * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
+ */
+static bool intel_idle_cst_usable(void)
+{
+	int cstate, limit;
+
+	limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
+		      acpi_state_table.count);
+
+	for (cstate = 1; cstate < limit; cstate++) {
+		struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
+
+		if (cx->entry_method != ACPI_CSTATE_FFH)
+			return false;
+	}
+
+	return true;
+}
+
+static bool intel_idle_acpi_cst_extract(void)
+{
+	unsigned int cpu;
+
+	for_each_possible_cpu(cpu) {
+		struct acpi_processor *pr = per_cpu(processors, cpu);
+
+		if (!pr)
+			continue;
+
+		if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
+			continue;
+
+		acpi_state_table.count++;
+
+		if (!intel_idle_cst_usable())
+			continue;
+
+		if (!acpi_processor_claim_cst_control()) {
+			acpi_state_table.count = 0;
+			return false;
+		}
+
+		return true;
+	}
+
+	pr_debug("ACPI _CST not found or not usable\n");
+	return false;
+}
+
+static void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
+{
+	int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
+
+	/*
+	 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
+	 * the interesting states are ACPI_CSTATE_FFH.
+	 */
+	for (cstate = 1; cstate < limit; cstate++) {
+		struct acpi_processor_cx *cx;
+		struct cpuidle_state *state;
+
+		if (intel_idle_max_cstate_reached(cstate))
+			break;
+
+		cx = &acpi_state_table.states[cstate];
+
+		state = &drv->states[drv->state_count++];
+
+		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
+		strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
+		state->exit_latency = cx->latency;
+		state->target_residency = cx->latency;
+		if (cx->type > ACPI_STATE_C1)
+			state->target_residency *= 4;
+
+		state->flags = MWAIT2flg(cx->address);
+		if (cx->type > ACPI_STATE_C2)
+			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
+
+		state->enter = intel_idle;
+		state->enter_s2idle = intel_idle_s2idle;
+	}
+}
+#else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
+static inline bool intel_idle_acpi_cst_extract(void) { return false; }
+static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
+#endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
+
 /*
  * intel_idle_probe()
  */
@@ -1125,17 +1240,15 @@ static int __init intel_idle_probe(void)
 	}
 
 	id = x86_match_cpu(intel_idle_ids);
-	if (!id) {
-		if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
-		    boot_cpu_data.x86 == 6)
-			pr_debug("does not run on family %d model %d\n",
-				 boot_cpu_data.x86, boot_cpu_data.x86_model);
-		return -ENODEV;
-	}
-
-	if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
-		pr_debug("Please enable MWAIT in BIOS SETUP\n");
-		return -ENODEV;
+	if (id) {
+		if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
+			pr_debug("Please enable MWAIT in BIOS SETUP\n");
+			return -ENODEV;
+		}
+	} else {
+		id = x86_match_cpu(intel_mwait_ids);
+		if (!id)
+			return -ENODEV;
 	}
 
 	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
@@ -1151,7 +1264,10 @@ static int __init intel_idle_probe(void)
 	pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
 
 	icpu = (const struct idle_cpu *)id->driver_data;
-	cpuidle_state_table = icpu->state_table;
+	if (icpu)
+		cpuidle_state_table = icpu->state_table;
+	else if (!intel_idle_acpi_cst_extract())
+		return -ENODEV;
 
 	pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
 		 boot_cpu_data.x86_model);
@@ -1333,31 +1449,19 @@ static void intel_idle_state_table_updat
 	}
 }
 
-/*
- * intel_idle_cpuidle_driver_init()
- * allocate, initialize cpuidle_states
- */
-static void __init intel_idle_cpuidle_driver_init(void)
+static void intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
 {
 	int cstate;
-	struct cpuidle_driver *drv = &intel_idle_driver;
-
-	intel_idle_state_table_update();
-
-	cpuidle_poll_state_init(drv);
-	drv->state_count = 1;
 
 	for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
 		unsigned int mwait_hint;
 
-		if (!cpuidle_state_table[cstate].enter &&
-		    !cpuidle_state_table[cstate].enter_s2idle)
+		if (intel_idle_max_cstate_reached(cstate))
 			break;
 
-		if (cstate + 1 > max_cstate) {
-			pr_info("max_cstate %d reached\n", max_cstate);
+		if (!cpuidle_state_table[cstate].enter &&
+		    !cpuidle_state_table[cstate].enter_s2idle)
 			break;
-		}
 
 		/* If marked as unusable, skip this state. */
 		if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
@@ -1380,6 +1484,24 @@ static void __init intel_idle_cpuidle_dr
 	}
 }
 
+/*
+ * intel_idle_cpuidle_driver_init()
+ * allocate, initialize cpuidle_states
+ */
+static void __init intel_idle_cpuidle_driver_init(void)
+{
+	struct cpuidle_driver *drv = &intel_idle_driver;
+
+	intel_idle_state_table_update();
+
+	cpuidle_poll_state_init(drv);
+	drv->state_count = 1;
+
+	if (icpu)
+		intel_idle_init_cstates_icpu(drv);
+	else
+		intel_idle_init_cstates_acpi(drv);
+}
 
 /*
  * intel_idle_cpu_init()
@@ -1398,6 +1520,9 @@ static int intel_idle_cpu_init(unsigned
 		return -EIO;
 	}
 
+	if (!icpu)
+		return 0;
+
 	if (icpu->auto_demotion_disable_flags)
 		auto_demotion_disable();
 
#7 lore raw

RFC PATCH v2 7/9 cpuidle: Allow idle states to be disabled by default

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

In certain situations it may be useful to prevent some idle states from being used by default while allowing user space to enable them later on.

For this purpose, introduce a new state flag, CPUIDLE_FLAG_OFF, to mark idle states that should be disabled by default, make the core set CPUIDLE_STATE_DISABLED_BY_USER for those states at the initialization time and add a new state attribute in sysfs, "initial_status", to inform user space of the initial status of the given idle state ("disabled" if CPUIDLE_FLAG_OFF is set for it, "enabled" otherwise).

---

No changes from the previous version.

---
 Documentation/ABI/testing/sysfs-devices-system-cpu |    6 ++++++
 Documentation/admin-guide/pm/cpuidle.rst           |    3 +++
 drivers/cpuidle/cpuidle.c                          |    6 +++++-
 drivers/cpuidle/sysfs.c                            |   10 ++++++++++
 include/linux/cpuidle.h                            |    1 +
 5 files changed, 25 insertions(+), 1 deletion(-)
drivers/cpuidle/sysfs.c +10 -0
Index: linux-pm/drivers/cpuidle/sysfs.c
===================================================================
@@ -327,6 +327,14 @@ static ssize_t store_state_disable(struc
 	return size;
 }
 
+static ssize_t show_state_initial_status(struct cpuidle_state *state,
+					  struct cpuidle_state_usage *state_usage,
+					  char *buf)
+{
+	return sprintf(buf, "%s\n",
+		       state->flags & CPUIDLE_FLAG_OFF ? "disabled" : "enabled");
+}
+
 define_one_state_ro(name, show_state_name);
 define_one_state_ro(desc, show_state_desc);
 define_one_state_ro(latency, show_state_exit_latency);
@@ -337,6 +345,7 @@ define_one_state_ro(time, show_state_tim
 define_one_state_rw(disable, show_state_disable, store_state_disable);
 define_one_state_ro(above, show_state_above);
 define_one_state_ro(below, show_state_below);
+define_one_state_ro(initial_status, show_state_initial_status);
 
 static struct attribute *cpuidle_state_default_attrs[] = {
 	&attr_name.attr,
@@ -349,6 +358,7 @@ static struct attribute *cpuidle_state_d
 	&attr_disable.attr,
 	&attr_above.attr,
 	&attr_below.attr,
+	&attr_initial_status.attr,
 	NULL
 };
 
include/linux/cpuidle.h +1 -0
Index: linux-pm/include/linux/cpuidle.h
===================================================================
@@ -77,6 +77,7 @@ struct cpuidle_state {
 #define CPUIDLE_FLAG_COUPLED	BIT(1) /* state applies to multiple cpus */
 #define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */
 #define CPUIDLE_FLAG_UNUSABLE	BIT(3) /* avoid using this state */
+#define CPUIDLE_FLAG_OFF	BIT(4) /* disable this state by default */
 
 struct cpuidle_device_kobj;
 struct cpuidle_state_kobj;
drivers/cpuidle/cpuidle.c +5 -1
Index: linux-pm/drivers/cpuidle/cpuidle.c
===================================================================
@@ -571,10 +571,14 @@ static int __cpuidle_register_device(str
 	if (!try_module_get(drv->owner))
 		return -EINVAL;
 
-	for (i = 0; i < drv->state_count; i++)
+	for (i = 0; i < drv->state_count; i++) {
 		if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
 			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
 
+		if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
+			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
+	}
+
 	per_cpu(cpuidle_devices, dev->cpu) = dev;
 	list_add(&dev->device_list, &cpuidle_detected_devices);
 
Documentation/ABI/testing/sysfs-devices-system-cpu +6 -0
Index: linux-pm/Documentation/ABI/testing/sysfs-devices-system-cpu
===================================================================
@@ -196,6 +196,12 @@ Description:
 		does not reflect it. Likewise, if one enables a deep state but a
 		lighter state still is disabled, then this has no effect.
 
+What:		/sys/devices/system/cpu/cpuX/cpuidle/stateN/initial_status
+Date:		November 2019
+KernelVersion:	v5.6
+Contact:	Linux power management list <[email protected]>
+Description:
+		(RO) The initial status of this state, "enabled" or "disabled".
 
 What:		/sys/devices/system/cpu/cpuX/cpuidle/stateN/residency
 Date:		March 2014
Documentation/admin-guide/pm/cpuidle.rst +3 -0
Index: linux-pm/Documentation/admin-guide/pm/cpuidle.rst
===================================================================
@@ -506,6 +506,9 @@ object corresponding to it, as follows:
 ``disable``
 	Whether or not this idle state is disabled.
 
+``initial_status``
+	The initial status of this state, "enabled" or "disabled".
+
 ``latency``
 	Exit latency of the idle state in microseconds.
 
#8 lore raw

RFC PATCH v2 8/9 intel_idle: Use ACPI _CST for selected known processors

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Update the intel_idle driver to get the C-states information from ACPI _CST in some cases in which the processor is known to the driver, as long as that information is available and the new use_acpi flag is set in the profile of the processor in question.

In the cases when there is a specific table of C-states for the given processor in the driver, that table is used as the primary source of information on the available C-states, but if ACPI _CST is present, the C-states that are not listed by it will not be enabled by default (they still can be enabled later by user space via sysfs, though).

---
 drivers/idle/intel_idle.c |   45 ++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 42 insertions(+), 3 deletions(-)
drivers/idle/intel_idle.c +42 -3
Index: linux-pm/drivers/idle/intel_idle.c
===================================================================
@@ -80,6 +80,7 @@ struct idle_cpu {
 	unsigned long auto_demotion_disable_flags;
 	bool byt_auto_demotion_disable_flag;
 	bool disable_promotion_to_c1e;
+	bool use_acpi;
 };
 
 static const struct idle_cpu *icpu;
@@ -91,6 +92,11 @@ static void intel_idle_s2idle(struct cpu
 static struct cpuidle_state *cpuidle_state_table;
 
 /*
+ * Enable this state by default even if the ACPI _CST does not list it.
+ */
+#define CPUIDLE_FLAG_IGNORE_ACPI	BIT(15)
+
+/*
  * Set this flag for states where the HW flushes the TLB for us
  * and so we don't need cross-calls to keep it consistent.
  * If this flag is set, SW flushes the TLB, so even if the
@@ -1221,9 +1227,33 @@ static void intel_idle_init_cstates_acpi
 		state->enter_s2idle = intel_idle_s2idle;
 	}
 }
+
+static bool intel_idle_off_by_default(u32 mwait_hint)
+{
+	int cstate, limit;
+
+	/*
+	 * If there are no _CST C-states, do not disable any C-states by
+	 * default.
+	 */
+	if (!acpi_state_table.count)
+		return false;
+
+	limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
+	/*
+	 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
+	 * the interesting states are ACPI_CSTATE_FFH.
+	 */
+	for (cstate = 1; cstate < limit; cstate++) {
+		if (acpi_state_table.states[cstate].address == mwait_hint)
+			return false;
+	}
+	return true;
+}
 #else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
 static inline bool intel_idle_acpi_cst_extract(void) { return false; }
 static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
+static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
 #endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
 
 /*
@@ -1264,10 +1294,13 @@ static int __init intel_idle_probe(void)
 	pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
 
 	icpu = (const struct idle_cpu *)id->driver_data;
-	if (icpu)
+	if (icpu) {
 		cpuidle_state_table = icpu->state_table;
-	else if (!intel_idle_acpi_cst_extract())
+		if (icpu->use_acpi)
+			intel_idle_acpi_cst_extract();
+	} else if (!intel_idle_acpi_cst_extract()) {
 		return -ENODEV;
+	}
 
 	pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
 		 boot_cpu_data.x86_model);
@@ -1475,7 +1508,13 @@ static void intel_idle_init_cstates_icpu
 			continue;
 
 		/* Structure copy. */
-		drv->states[drv->state_count++] = cpuidle_state_table[cstate];
+		drv->states[drv->state_count] = cpuidle_state_table[cstate];
+
+		if (icpu->use_acpi && intel_idle_off_by_default(mwait_hint) &&
+		    !(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IGNORE_ACPI))
+			drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;
+
+		drv->state_count++;
 	}
 
 	if (icpu->byt_auto_demotion_disable_flag) {
#9 lore raw

RFC PATCH v2 9/9 intel_idle: Add module parameter to prevent ACPI _CST from being used

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new module parameter called "no_acpi" to the intel_idle driver to allow the driver to be prevented from using ACPI _CST via kernel command line.

---
 drivers/idle/intel_idle.c |    9 +++++++++
 1 file changed, 9 insertions(+)
drivers/idle/intel_idle.c +9 -0
Index: linux-pm/drivers/idle/intel_idle.c
===================================================================
@@ -1138,6 +1138,10 @@ static bool intel_idle_max_cstate_reache
 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
 #include <acpi/processor.h>
 
+static bool no_acpi __read_mostly;
+module_param(no_acpi, bool, 0444);
+MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
+
 static struct acpi_processor_power acpi_state_table;
 
 /**
@@ -1167,6 +1171,11 @@ static bool intel_idle_acpi_cst_extract(
 {
 	unsigned int cpu;
 
+	if (no_acpi) {
+		pr_debug("Not allowed to use ACPI _CST\n");
+		return false;
+	}
+
 	for_each_possible_cpu(cpu) {
 		struct acpi_processor *pr = per_cpu(processors, cpu);