Skip to contentSkip to conversation

ACPI: introduce support for FPDT table

PATCH Superseded Zhang Rui 2 participants 2 messages Oct 31, 2019 – Nov 13, 2019 ACPI

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Zhang Rui

author
#0 lore raw

PATCH 1/1 ACPI: introduce support for FPDT table

ACPI Firmware Performance Data Table (FPDT) provides information about firmware performance during system boot, S3 suspend and S3 resume.

Have the kernel parse the FPDT table, and expose the firmware performance data to userspace as syfs attributes under /sys/firmware/acpi/firmware_performance_data/.

---
 Documentation/ABI/testing/sysfs-firmware-acpi |  43 +++
 drivers/acpi/Kconfig                          |   8 +
 drivers/acpi/Makefile                         |   1 +
 drivers/acpi/acpi_fpdt.c                      | 271 ++++++++++++++++++
 drivers/acpi/internal.h                       |   6 +
 drivers/acpi/sysfs.c                          |   2 +
 6 files changed, 331 insertions(+)
 create mode 100644 drivers/acpi/acpi_fpdt.c
Documentation/ABI/testing/sysfs-firmware-acpi +43 -0
index 613f42a9d5cd..b76f0ef7945a 100644
@@ -1,3 +1,46 @@
+What:		/sys/firmware/acpi/firmware_performance_data/
+Date:		Aug 2019
+Contact:	Zhang Rui <[email protected]>
+Description:
+		ACPI Firmware Performance Data Table (FPDT) provides
+		information for firmware performance data for system boot,
+		S3 suspend and S3 resume. This sysfs entry contains the
+		performance data retrieved from the FPDT.
+
+		boot_performance:
+			firmware_start_ns: Timer value logged at the beginning
+				of firmware image execution. In nanoseconds.
+			bootloader_load_ns: Timer value logged just prior to
+				loading the OS boot loader into memory.
+				In nanoseconds.
+			bootloader_launch_ns: Timer value logged just prior to
+				launching the currently loaded OS boot loader
+				image. In nanoseconds.
+			exitbootservice_start_ns: Timer value logged at the
+				point when the OS loader calls the
+				ExitBootServices function for UEFI compatible
+				firmware. In nanoseconds.
+			exitbootservice_end_ns: Timer value logged at the point
+				just prior to the OS loader gaining control
+				back from the ExitBootServices function for
+				UEFI compatible firmware. In nanoseconds.
+		suspend_performance:
+			suspend_start_ns: Timer value recorded at the previous
+				OS write to SLP_TYP upon entry to S3. In
+				nanoseconds.
+			suspend_end_ns: Timer value recorded at the previous
+				firmware write to SLP_TYP used to trigger
+				hardware entry to S3. In nanoseconds.
+		resume_performance:
+			resume_count: A count of the number of S3 resume cycles
+				since the last full boot sequence.
+			resume_avg_ns: Average timer value of all resume cycles
+				logged since the last full boot sequence,
+				including the most recent resume. In nanoseconds.
+			resume_prev_ns: Timer recorded at the end of the previous
+				platform runtime firmware S3 resume, just prior to
+				handoff to the OS waking vector. In nanoseconds.
+
 What:		/sys/firmware/acpi/bgrt/
 Date:		January 2012
 Contact:	Matthew Garrett <[email protected]>
drivers/acpi/Kconfig +8 -0
index ebe1e9e5fd81..8a7da89eeba8 100644
@@ -88,6 +88,14 @@ config ACPI_SPCR_TABLE
 	  This table provides information about the configuration of the
 	  earlycon console.
 
+config ACPI_FPDT
+	bool "ACPI Firmware Performance Data Table(FPDT)"
+	default y if X86
+	help
+	  Enable support for the Firmware Performance Data Table (FPDT).
+	  This table provides information to describe the platform
+	  initialization performance records.
+
 config ACPI_LPIT
 	bool
 	depends on X86_64
drivers/acpi/Makefile +1 -0
index 5d361e4e3405..3cabb112aec4 100644
@@ -58,6 +58,7 @@ acpi-$(CONFIG_DEBUG_FS)		+= debugfs.o
 acpi-$(CONFIG_ACPI_NUMA)	+= numa.o
 acpi-$(CONFIG_ACPI_PROCFS_POWER) += cm_sbs.o
 acpi-y				+= acpi_lpat.o
+acpi-$(CONFIG_ACPI_FPDT)	+= acpi_fpdt.o
 acpi-$(CONFIG_ACPI_LPIT)	+= acpi_lpit.o
 acpi-$(CONFIG_ACPI_GENERIC_GSI) += irq.o
 acpi-$(CONFIG_ACPI_WATCHDOG)	+= acpi_watchdog.o
drivers/acpi/acpi_fpdt.c new file +271 -0

271 lines not on this page · raw message

drivers/acpi/internal.h +6 -0
index afe6636f9ad3..995dd3462c89 100644
@@ -254,6 +254,12 @@ void acpi_watchdog_init(void);
 static inline void acpi_watchdog_init(void) {}
 #endif
 
+#ifdef CONFIG_ACPI_FPDT
+void acpi_init_fpdt(void);
+#else
+static inline void acpi_init_fpdt(void) { }
+#endif
+
 #ifdef CONFIG_ACPI_LPIT
 void acpi_init_lpit(void);
 #else
drivers/acpi/sysfs.c +2 -0
index 75948a3f1a20..1c28cf341f27 100644
@@ -1035,6 +1035,8 @@ int __init acpi_sysfs_init(void)
 {
 	int result;
 
+	acpi_init_fpdt();
+
 	result = acpi_tables_sysfs_init();
 	if (result)
 		return result;
Signature · 1 line
2.17.1
#1 lore raw
Quoted 75 lines · CET Zhang Rui

ACPI Firmware Performance Data Table (FPDT) provides information about firmware performance during system boot, S3 suspend and S3 resume.

Have the kernel parse the FPDT table, and expose the firmware performance data to userspace as syfs attributes under /sys/firmware/acpi/firmware_performance_data/.

---
 Documentation/ABI/testing/sysfs-firmware-acpi |  43 +++
 drivers/acpi/Kconfig                          |   8 +
 drivers/acpi/Makefile                         |   1 +
 drivers/acpi/acpi_fpdt.c                      | 271 ++++++++++++++++++
 drivers/acpi/internal.h                       |   6 +
 drivers/acpi/sysfs.c                          |   2 +
 6 files changed, 331 insertions(+)
 create mode 100644 drivers/acpi/acpi_fpdt.c
Documentation/ABI/testing/sysfs-firmware-acpi +43 -0
index 613f42a9d5cd..b76f0ef7945a 100644
@@ -1,3 +1,46 @@
+What:		/sys/firmware/acpi/firmware_performance_data/
+Date:		Aug 2019
+Contact:	Zhang Rui <[email protected]>
+Description:
+		ACPI Firmware Performance Data Table (FPDT) provides
+		information for firmware performance data for system boot,
+		S3 suspend and S3 resume. This sysfs entry contains the
+		performance data retrieved from the FPDT.
+
+		boot_performance:
+			firmware_start_ns: Timer value logged at the beginning
+				of firmware image execution. In nanoseconds.
+			bootloader_load_ns: Timer value logged just prior to
+				loading the OS boot loader into memory.
+				In nanoseconds.
+			bootloader_launch_ns: Timer value logged just prior to
+				launching the currently loaded OS boot loader
+				image. In nanoseconds.
+			exitbootservice_start_ns: Timer value logged at the
+				point when the OS loader calls the
+				ExitBootServices function for UEFI compatible
+				firmware. In nanoseconds.
+			exitbootservice_end_ns: Timer value logged at the point
+				just prior to the OS loader gaining control
+				back from the ExitBootServices function for
+				UEFI compatible firmware. In nanoseconds.
+		suspend_performance:
+			suspend_start_ns: Timer value recorded at the previous
+				OS write to SLP_TYP upon entry to S3. In
+				nanoseconds.
+			suspend_end_ns: Timer value recorded at the previous
+				firmware write to SLP_TYP used to trigger
+				hardware entry to S3. In nanoseconds.
+		resume_performance:
+			resume_count: A count of the number of S3 resume cycles
+				since the last full boot sequence.
+			resume_avg_ns: Average timer value of all resume cycles
+				logged since the last full boot sequence,
+				including the most recent resume. In nanoseconds.
+			resume_prev_ns: Timer recorded at the end of the previous
+				platform runtime firmware S3 resume, just prior to
+				handoff to the OS waking vector. In nanoseconds.
+
 What:		/sys/firmware/acpi/bgrt/
 Date:		January 2012
 Contact:	Matthew Garrett <[email protected]>
drivers/acpi/Kconfig +0 -0
index ebe1e9e5fd81..8a7da89eeba8 100644

drivers/acpi/Kconfig config ACPI_SPCR_TABLE · line 93

Quoted 6 lines
drivers/acpi/Kconfig +3 -0
@@ -88,6 +88,14 @@ config ACPI_SPCR_TABLE
 	  This table provides information about the configuration of the
 	  earlycon console.
 
+config ACPI_FPDT
+	bool "ACPI Firmware Performance Data Table(FPDT)"
+	default y if X86

Just X86 or X86_64?

drivers/acpi/Kconfig config ACPI_SPCR_TABLE · line 97

Quoted 4 lines
drivers/acpi/Kconfig +4 -0
+	help
+	  Enable support for the Firmware Performance Data Table (FPDT).
+	  This table provides information to describe the platform
+	  initialization performance records.

Why not "This table provides information on the timing of the system boot, S3 suspend and S3 resume firmware code paths."?

Quoted 22 lines
drivers/acpi/Kconfig +1 -0
+
 config ACPI_LPIT
 	bool
 	depends on X86_64
drivers/acpi/Makefile +1 -0
index 5d361e4e3405..3cabb112aec4 100644
@@ -58,6 +58,7 @@ acpi-$(CONFIG_DEBUG_FS)		+= debugfs.o
 acpi-$(CONFIG_ACPI_NUMA)	+= numa.o
 acpi-$(CONFIG_ACPI_PROCFS_POWER) += cm_sbs.o
 acpi-y				+= acpi_lpat.o
+acpi-$(CONFIG_ACPI_FPDT)	+= acpi_fpdt.o
 acpi-$(CONFIG_ACPI_LPIT)	+= acpi_lpit.o
 acpi-$(CONFIG_ACPI_GENERIC_GSI) += irq.o
 acpi-$(CONFIG_ACPI_WATCHDOG)	+= acpi_watchdog.o
drivers/acpi/acpi_fpdt.c new file +0 -0
new file mode 100644
index 000000000000..393d9451e1af

drivers/acpi/acpi_fpdt.c line 4

Quoted 4 lines
drivers/acpi/acpi_fpdt.c new file +4 -0
@@ -0,0 +1,271 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/*
+ * FPDT support for exporting boot and performance data

"boot and suspend/resume" ?

Quoted 58 lines
drivers/acpi/acpi_fpdt.c new file +58 -0
+ *
+ * Copyright (C) 2019 Intel Corporation. All rights reserved.
+ */
+
+#define pr_fmt(fmt) "ACPI FPDT: " fmt
+
+#include <linux/acpi.h>
+
+/*
+ * FPDT contains ACPI table header and a number of fpdt_subtable_entries.
+ * Each fpdt_subtable_entry points to a subtable: FBPT or S3PT.
+ * Each FPDT subtable (FBPT/S3PT) is composed of a fpdt_subtable_header
+ * and a number of fpdt performance records.
+ * Each FPDT performance record is composed of a fpdt_record_header and
+ * performance data fields, for boot or suspend or resume phase.
+ */
+enum fpdt_subtable_type {
+	SUBTABLE_FBPT,
+	SUBTABLE_S3PT,
+};
+
+struct fpdt_subtable_entry {
+	u16 type;		/* refer to enum fpdt_subtable_type */
+	u8 length;
+	u8 revision;
+	u32 reserved;
+	u64 address;		/* physical address of the S3PT/FBPT table */
+};
+
+struct fpdt_subtable_header {
+	u32 signature;
+	u32 length;
+};
+
+enum fpdt_record_type {
+	RECORD_S3_RESUME,
+	RECORD_S3_SUSPEND,
+	RECORD_BOOT,
+};
+
+struct fpdt_record_header {
+	u16 type;		/* refer to enum fpdt_record_type */
+	u8 length;
+	u8 revision;
+};
+
+struct resume_performance_record {
+	struct fpdt_record_header header;
+	u32 resume_count;
+	u64 resume_prev;
+	u64 resume_avg;
+};
+
+struct boot_performance_record {
+	struct fpdt_record_header header;
+	u32 reserved;
+	u64 firmware_start;
+	u64 bootloader_load;

drivers/acpi/acpi_fpdt.c line 72

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+	u64 bootloader_launch;
+	u64 exitbootservice_start;
+	u64 exitbootservice_end;
+};
+
+/*
+ * All the timer value recorded takes 8 bytes.
+ * Unlike the boot and resume performance records, in which the timer fields
+ * are all 8 bytes aligned, struct suspend_performance_record needs to disable
+ * padding.

This isn't really clear.

I would say something like "Because the header is 4 bytes long, the timer values in a suspend performance records may not be aligned to 8 bytes boundaries, so padding should be avoided in this case".

But overall, I think that adding __attribute__((packed)) to all of the record structures would not be excessive and then you may not need this comment at all.

Quoted 40 lines
drivers/acpi/acpi_fpdt.c new file +40 -0
+ */
+struct suspend_performance_record {
+	struct fpdt_record_header header;
+	u64 suspend_start;
+	u64 suspend_end;
+} __attribute__((packed));
+
+
+static struct resume_performance_record *record_resume;
+static struct suspend_performance_record *record_suspend;
+static struct boot_performance_record *record_boot;
+
+#define FPDT_ATTR(phase, name)	\
+static ssize_t name##_show(struct kobject *kobj,	\
+		 struct kobj_attribute *attr, char *buf)	\
+{	\
+	return sprintf(buf, "%llu\n", record_##phase->name);	\
+}	\
+static struct kobj_attribute name##_attr =	\
+__ATTR(name##_ns, 0444, name##_show, NULL)
+
+FPDT_ATTR(resume, resume_prev);
+FPDT_ATTR(resume, resume_avg);
+FPDT_ATTR(suspend, suspend_start);
+FPDT_ATTR(suspend, suspend_end);
+FPDT_ATTR(boot, firmware_start);
+FPDT_ATTR(boot, bootloader_load);
+FPDT_ATTR(boot, bootloader_launch);
+FPDT_ATTR(boot, exitbootservice_start);
+FPDT_ATTR(boot, exitbootservice_end);
+
+static ssize_t resume_count_show(struct kobject *kobj,
+				 struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%u\n", record_resume->resume_count);
+}
+
+static struct kobj_attribute resume_count_attr =
+__ATTR_RO(resume_count);
+

drivers/acpi/acpi_fpdt.c line 122

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+static struct attribute *resume_attrs[] = {
+	&resume_count_attr.attr,
+	&resume_prev_attr.attr,
+	&resume_avg_attr.attr,
+	NULL
+};
+
+static const struct attribute_group resume_attr_group = {
+	.attrs = resume_attrs,
+	.name = "resume_performance",

Why mot just "resume"?

drivers/acpi/acpi_fpdt.c line 133

Quoted 11 lines
drivers/acpi/acpi_fpdt.c new file +11 -0
+};
+
+static struct attribute *suspend_attrs[] = {
+	&suspend_start_attr.attr,
+	&suspend_end_attr.attr,
+	NULL
+};
+
+static const struct attribute_group suspend_attr_group = {
+	.attrs = suspend_attrs,
+	.name = "suspend_performance",

And why not just "suspend"?

Quoted 4 lines
drivers/acpi/acpi_fpdt.c new file +4 -0
+};
+
+static struct attribute *boot_attrs[] = {
+	&firmware_start_attr.attr,

drivers/acpi/acpi_fpdt.c line 147

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+	&bootloader_load_attr.attr,
+	&bootloader_launch_attr.attr,
+	&exitbootservice_start_attr.attr,
+	&exitbootservice_end_attr.attr,
+	NULL
+};
+
+static const struct attribute_group boot_attr_group = {
+	.attrs = boot_attrs,
+	.name = "boot_performance",

And why not "boot"?

Quoted 5 lines
drivers/acpi/acpi_fpdt.c new file +5 -0
+};
+
+static struct kobject *fpdt_kobj;
+
+static int fpdt_process_subtable(u64 address, u32 subtable_type)

drivers/acpi/acpi_fpdt.c line 162

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+{
+	struct fpdt_subtable_header *subtable_header;
+	struct fpdt_record_header *record_header;
+	char *signature = (subtable_type == SUBTABLE_FBPT ? "FBPT" : "S3PT");
+	u32 length, offset;
+	int result;
+
+	subtable_header = acpi_os_map_memory(address, sizeof(*subtable_header));
+	if (!subtable_header) {
+		pr_err("failed to map subtable\n");

Is this message really needed here?

drivers/acpi/acpi_fpdt.c line 167

Quoted 5 lines
drivers/acpi/acpi_fpdt.c new file +5 -0
+		return -ENOMEM;
+	}
+
+	if (strncmp((char *)&subtable_header->signature, signature, 4)) {
+		pr_err(FW_BUG "subtable signature and type mismatch!\n");

Using pr_err() to complain about firmware bugs seems a bit excessive IMO.

Maybe use pr_info()?

Quoted 75 lines
drivers/acpi/acpi_fpdt.c new file +75 -0
+		return -EINVAL;
+	}
+
+	length = subtable_header->length;
+	acpi_os_unmap_memory(subtable_header, sizeof(*subtable_header));
+
+	subtable_header = acpi_os_map_memory(address, length);
+	if (!subtable_header) {
+		pr_err("failed to map subtable\n");
+		return -ENOMEM;
+	}
+
+	offset = sizeof(*subtable_header);
+	while (offset < length) {
+		record_header = (void *)subtable_header + offset;
+		offset += record_header->length;
+
+		switch (record_header->type) {
+		case RECORD_S3_RESUME:
+			if (subtable_type != SUBTABLE_S3PT) {
+				pr_err(FW_BUG "Invalid record %d for subtable %s\n",
+				     record_header->type, signature);
+				return -EINVAL;
+			}
+			if (record_resume) {
+				pr_err("Duplicate resume performance record found.\n");
+				continue;
+			}
+			record_resume = (struct resume_performance_record *)record_header;
+			result = sysfs_create_group(fpdt_kobj, &resume_attr_group);
+			if (result)
+				return result;
+			break;
+		case RECORD_S3_SUSPEND:
+			if (subtable_type != SUBTABLE_S3PT) {
+				pr_err(FW_BUG "Invalid %d for subtable %s\n",
+				     record_header->type, signature);
+				continue;
+			}
+			if (record_suspend) {
+				pr_err("Duplicate suspend performance record found.\n");
+				continue;
+			}
+			record_suspend = (struct suspend_performance_record *)record_header;
+			result = sysfs_create_group(fpdt_kobj, &suspend_attr_group);
+			if (result)
+				return result;
+			break;
+		case RECORD_BOOT:
+			if (subtable_type != SUBTABLE_FBPT) {
+				pr_err(FW_BUG "Invalid %d for subtable %s\n",
+				     record_header->type, signature);
+				return -EINVAL;
+			}
+			if (record_boot) {
+				pr_err("Duplicate boot performance record found.\n");
+				continue;
+			}
+			record_boot = (struct boot_performance_record *)record_header;
+			result = sysfs_create_group(fpdt_kobj, &boot_attr_group);
+			if (result)
+				return result;
+			break;
+
+		default:
+			pr_err(FW_BUG "Invalid record %d found.\n", record_header->type);
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+void __init acpi_init_fpdt(void)
+{
+	acpi_status status;

drivers/acpi/acpi_fpdt.c line 252

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+	struct acpi_table_header *header;
+	struct fpdt_subtable_entry *subtable;
+	u32 offset = sizeof(*header);
+
+	status = acpi_get_table(ACPI_SIG_FPDT, 0, &header);
+
+	if (ACPI_FAILURE(status))
+		return;
+
+	fpdt_kobj = kobject_create_and_add("firmware_performance_data", acpi_kobj);

Why not to call it "fpdt" (in analogy with "bgrt")?

Quoted 4 lines
drivers/acpi/acpi_fpdt.c new file +4 -0
+	if (!fpdt_kobj)
+		return;
+
+	while (offset < header->length) {

drivers/acpi/acpi_fpdt.c line 266

Quoted 10 lines
drivers/acpi/acpi_fpdt.c new file +10 -0
+		subtable = (void *)header + offset;
+		switch (subtable->type) {
+		case SUBTABLE_FBPT:
+		case SUBTABLE_S3PT:
+			fpdt_process_subtable(subtable->address,
+					      subtable->type);
+			break;
+		default:
+			pr_err("Invalid subtable type %d found.\n",
+			       subtable->type);

That would be FW_BUG, wouldn't it?

Quoted 27 lines
drivers/acpi/acpi_fpdt.c new file +5 -0
+			return;
+		}
+		offset += sizeof(*subtable);
+	}
+}
drivers/acpi/internal.h +6 -0
index afe6636f9ad3..995dd3462c89 100644
@@ -254,6 +254,12 @@ void acpi_watchdog_init(void);
 static inline void acpi_watchdog_init(void) {}
 #endif
 
+#ifdef CONFIG_ACPI_FPDT
+void acpi_init_fpdt(void);
+#else
+static inline void acpi_init_fpdt(void) { }
+#endif
+
 #ifdef CONFIG_ACPI_LPIT
 void acpi_init_lpit(void);
 #else
drivers/acpi/sysfs.c +0 -0
index 75948a3f1a20..1c28cf341f27 100644

drivers/acpi/sysfs.c int __init acpi_sysfs_init(void) · line 1038

Quoted 4 lines
drivers/acpi/sysfs.c +1 -0
@@ -1035,6 +1035,8 @@ int __init acpi_sysfs_init(void)
 {
 	int result;
 
+	acpi_init_fpdt();

Why here? Why not later in particular? Or use fs_initcall() or similar (in which case it wouldn't be necessary to export the _init routine)?

Quoted 5 lines
drivers/acpi/sysfs.c +1 -0
+
 	result = acpi_tables_sysfs_init();
 	if (result)
 		return result;