Re: [PATCH] ACPI: introduce support for FPDT table

["Rafael J. Wysocki" <rjw@rjwysocki.net>]

On Thursday, October 31, 2019 2:13:14 PM CET Zhang Rui wrote:
> 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/.
> 
> Tested-by: Todd Brandt <todd.e.brandt@linux.intel.com>
> Signed-off-by: Zhang Rui <rui.zhang@intel.com>
> ---
>  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
> 
> diff --git a/Documentation/ABI/testing/sysfs-firmware-acpi b/Documentation/ABI/testing/sysfs-firmware-acpi
> index 613f42a9d5cd..b76f0ef7945a 100644
> --- a/Documentation/ABI/testing/sysfs-firmware-acpi
> +++ b/Documentation/ABI/testing/sysfs-firmware-acpi
> @@ -1,3 +1,46 @@
> +What:		/sys/firmware/acpi/firmware_performance_data/
> +Date:		Aug 2019
> +Contact:	Zhang Rui <rui.zhang@intel.com>
> +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 <mjg@redhat.com>
> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
> index ebe1e9e5fd81..8a7da89eeba8 100644
> --- a/drivers/acpi/Kconfig
> +++ b/drivers/acpi/Kconfig
> @@ -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?

> +	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."?

> +
>  config ACPI_LPIT
>  	bool
>  	depends on X86_64
> diff --git a/drivers/acpi/Makefile b/drivers/acpi/Makefile
> index 5d361e4e3405..3cabb112aec4 100644
> --- a/drivers/acpi/Makefile
> +++ b/drivers/acpi/Makefile
> @@ -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
> diff --git a/drivers/acpi/acpi_fpdt.c b/drivers/acpi/acpi_fpdt.c
> new file mode 100644
> index 000000000000..393d9451e1af
> --- /dev/null
> +++ b/drivers/acpi/acpi_fpdt.c
> @@ -0,0 +1,271 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +
> +/*
> + * FPDT support for exporting boot and performance data

"boot and suspend/resume" ?

> + *
> + * 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;
> +	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.

> + */
> +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);
> +
> +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"?

> +};
> +
> +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"?

> +};
> +
> +static struct attribute *boot_attrs[] = {
> +	&firmware_start_attr.attr,
> +	&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"?

> +};
> +
> +static struct kobject *fpdt_kobj;
> +
> +static int fpdt_process_subtable(u64 address, u32 subtable_type)
> +{
> +	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?

> +		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()?

> +		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;
> +	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")?

> +	if (!fpdt_kobj)
> +		return;
> +
> +	while (offset < header->length) {
> +		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?

> +			return;
> +		}
> +		offset += sizeof(*subtable);
> +	}
> +}
> diff --git a/drivers/acpi/internal.h b/drivers/acpi/internal.h
> index afe6636f9ad3..995dd3462c89 100644
> --- a/drivers/acpi/internal.h
> +++ b/drivers/acpi/internal.h
> @@ -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
> diff --git a/drivers/acpi/sysfs.c b/drivers/acpi/sysfs.c
> index 75948a3f1a20..1c28cf341f27 100644
> --- a/drivers/acpi/sysfs.c
> +++ b/drivers/acpi/sysfs.c
> @@ -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)?

> +
>  	result = acpi_tables_sysfs_init();
>  	if (result)
>  		return result;
>