Introduce RPM Master stats
v5 cover Introduce RPM Master stats
v4 -> v5:
- pick up tags
- uintXY_t -> uXY
- driver.name = "rpm_master_stats" -> "qcom_rpm_master_stats"
- use dev_err_probe for debugfs file creation fail
- use PTR_ERR(dent) intead of -EINVAL for debugfs file creation fail
v4: https://lore.kernel.org/r/[email protected]
v3 -> v4:
- Use "data" instead of "d" for naming local struct master_stats_data *
- Add "qcom_" prefix in the debugfs path
- Add "Qualcomm" prefix in MODULE_NAME
- Make the struct __packed__ to avoid padding surprises (even though it looks like there shouldn't be any, by eye)
- use IS_ERR for checking debugfs_create_file return value instead of nullchecking, that apparently changed..
- Use -ENODEV instead of -EINVAL when a MSG RAM handle is not found
- Add missing of_node_put() after of_parse_phandle()
- Vastly improve the documentation bit
v3: https://lore.kernel.org/r/[email protected]
v2 -> v3:
- rename rpm-master-stats.yaml to qcom,rpm-master-stats.yaml
v2: https://lore.kernel.org/r/[email protected]
v1 -> v2:
- Drop the - in /properties/compatible to make our entry be of the correct type [1/2]
- Change %s to %d for printing out the iterator [2/2]
v1: https://lore.kernel.org/r/[email protected]
The RPM MSG ram includes per-subsystem low-power mode entry/exit/ residence/etc. statistics which are very useful for trying to debug what I'd call "SoC insomnia", or IOW the plaftorm refusing to drop the voltage rails to a minimum and gate the non-critical clocks.
This series adds a very short and simple driver to query that data and expose it through debugfs.
The base used for writing this driver is:
https://github.com/sonyxperiadev/kernel/blob/aosp/LA.UM.9.14.r1/drivers/soc/qcom/rpm_master_stat.c
---
Konrad Dybcio (2):
dt-bindings: soc: qcom: Add RPM Master stats
soc: qcom: Introduce RPM master stats driver
.../bindings/soc/qcom/qcom,rpm-master-stats.yaml | 69 +++++++++ drivers/soc/qcom/Kconfig | 11 ++ drivers/soc/qcom/Makefile | 1 + drivers/soc/qcom/rpm_master_stats.c | 163 +++++++++++++++++++++ 4 files changed, 244 insertions(+)
---
base-commit: 67d5d9f013d6c3829383c08162939cabff14fccc
change-id: 20230405-topic-master_stats-ba201a9af93d
Best regards,
Signature · 1 line
Konrad Dybcio <[email protected]>
PATCH v5 1/2 dt-bindings: soc: qcom: Add RPM Master stats
The RPM MSG RAM contains per-RPM-master (e.g. APPS, ADSP etc.) sleep statistics. They let one assess which core is actively preventing the system from entering a true low-power mode.
--- .../bindings/soc/qcom/qcom,rpm-master-stats.yaml | 69 ++++++++++++++++++++++ 1 file changed, 69 insertions(+)
Documentation/devicetree/bindings/soc/qcom/qcom,rpm-master-stats.yaml new file +69 -0
@@ -0,0 +1,69 @@ +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) +%YAML 1.2 +--- +$id: http://devicetree.org/schemas/soc/qcom/qcom,rpm-master-stats.yaml# +$schema: http://devicetree.org/meta-schemas/core.yaml# + +title: Qualcomm Technologies, Inc. (QTI) RPM Master Stats + +maintainers: + - Konrad Dybcio <[email protected]> + +description: | + The Qualcomm RPM (Resource Power Manager) architecture includes a concept + of "RPM Masters". They can be thought of as "the local gang leaders", usually + spanning a single subsystem (e.g. APSS, ADSP, CDSP). All of the RPM decisons + (particularly around entering hardware-driven low power modes: XO shutdown + and total system-wide power collapse) are first made at Master-level, and + only then aggregated for the entire system. + + The Master Stats provide a few useful bits that can be used to assess whether + our device has entered the desired low-power mode, how long it took to do so, + the duration of that residence, how long it took to come back online, + how many times a given sleep state was entered and which cores are actively + voting for staying awake. + + This scheme has been used on various SoCs in the 2013-2023 era, with some + newer or higher-end designs providing this information through an SMEM query. + +properties: + compatible: + const: qcom,rpm-master-stats + + qcom,rpm-msg-ram: + $ref: /schemas/types.yaml#/definitions/phandle-array + description: Phandle to an RPM MSG RAM slice containing the master stats + minItems: 1 + maxItems: 5 + + qcom,master-names: + $ref: /schemas/types.yaml#/definitions/string-array + description: + The name of the RPM Master which owns the MSG RAM slice where this + instance of Master Stats resides + minItems: 1 + maxItems: 5 + +required: + - compatible + - qcom,rpm-msg-ram + - qcom,master-names + +additionalProperties: false + +examples: + - | + stats { + compatible = "qcom,rpm-master-stats"; + qcom,rpm-msg-ram = <&apss_master_stats>, + <&mpss_master_stats>, + <&adsp_master_stats>, + <&cdsp_master_stats>, + <&tz_master_stats>; + qcom,master-names = "APSS", + "MPSS", + "ADSP", + "CDSP", + "TZ"; + }; +...
Signature · 1 line
2.40.0
PATCH v5 2/2 soc: qcom: Introduce RPM master stats driver
In reply to Konrad DybcioIntroduce a driver to query and expose detailed, per-subsystem (as opposed to the existing qcom_stats driver which exposes SoC-wide data) about low power mode states of a given RPM master. That includes the APSS (ARM), MPSS (modem) and other remote cores, depending on the platform configuration.
This is a vastly cleaned up and restructured version of a similar driver found in msm-5.4.
--- drivers/soc/qcom/Kconfig | 11 +++ drivers/soc/qcom/Makefile | 1 + drivers/soc/qcom/rpm_master_stats.c | 163 ++++++++++++++++++++++++++++++++++++ 3 files changed, 175 insertions(+)
drivers/soc/qcom/Kconfig +11 -0
@@ -135,6 +135,17 @@ config QCOM_RMTFS_MEM Say y here if you intend to boot the modem remoteproc. +config QCOM_RPM_MASTER_STATS + tristate "Qualcomm RPM Master stats" + depends on ARCH_QCOM || COMPILE_TEST + help + The RPM Master sleep stats driver provides detailed per-subsystem + sleep/wake data, read from the RPM message RAM. It can be used to + assess whether all the low-power modes available are entered as + expected or to check which part of the SoC prevents it from sleeping. + + Say y here if you intend to debug or monitor platform sleep. + config QCOM_RPMH tristate "Qualcomm RPM-Hardened (RPMH) Communication" depends on ARCH_QCOM || COMPILE_TEST
drivers/soc/qcom/Makefile +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_QCOM_QMI_HELPERS) += qmi_helpers.o qmi_helpers-y += qmi_encdec.o qmi_interface.o obj-$(CONFIG_QCOM_RAMP_CTRL) += ramp_controller.o obj-$(CONFIG_QCOM_RMTFS_MEM) += rmtfs_mem.o +obj-$(CONFIG_QCOM_RPM_MASTER_STATS) += rpm_master_stats.o obj-$(CONFIG_QCOM_RPMH) += qcom_rpmh.o qcom_rpmh-y += rpmh-rsc.o qcom_rpmh-y += rpmh.o
drivers/soc/qcom/rpm_master_stats.c new file +163 -0
163 lines not on this page · raw message
Signature · 1 line
2.40.0
Re: [PATCH v5 2/2] soc: qcom: Introduce RPM master stats driver
Quoted 10 lines · Konrad Dybcio
Introduce a driver to query and expose detailed, per-subsystem (as opposed to the existing
qcom_statsdriver which exposes SoC-wide data) about low power mode states of a given RPM master. That includes the APSS (ARM), MPSS (modem) and other remote cores, depending on the platform configuration.This is a vastly cleaned up and restructured version of a similar driver found in msm-5.4.
Signed-off-by Konrad Dybcio <[email protected]>
- Mani
Quoted 213 lines
--- drivers/soc/qcom/Kconfig | 11 +++ drivers/soc/qcom/Makefile | 1 + drivers/soc/qcom/rpm_master_stats.c | 163 ++++++++++++++++++++++++++++++++++++ 3 files changed, 175 insertions(+)drivers/soc/qcom/Kconfig +11
-0@@ -135,6 +135,17 @@ config QCOM_RMTFS_MEM Say y here if you intend to boot the modem remoteproc. +config QCOM_RPM_MASTER_STATS + tristate "Qualcomm RPM Master stats" + depends on ARCH_QCOM || COMPILE_TEST + help + The RPM Master sleep stats driver provides detailed per-subsystem + sleep/wake data, read from the RPM message RAM. It can be used to + assess whether all the low-power modes available are entered as + expected or to check which part of the SoC prevents it from sleeping. + + Say y here if you intend to debug or monitor platform sleep. + config QCOM_RPMH tristate "Qualcomm RPM-Hardened (RPMH) Communication" depends on ARCH_QCOM || COMPILE_TESTdrivers/soc/qcom/Makefile +1
-0@@ -14,6 +14,7 @@ obj-$(CONFIG_QCOM_QMI_HELPERS) += qmi_helpers.o qmi_helpers-y += qmi_encdec.o qmi_interface.o obj-$(CONFIG_QCOM_RAMP_CTRL) += ramp_controller.o obj-$(CONFIG_QCOM_RMTFS_MEM) += rmtfs_mem.o +obj-$(CONFIG_QCOM_RPM_MASTER_STATS) += rpm_master_stats.o obj-$(CONFIG_QCOM_RPMH) += qcom_rpmh.o qcom_rpmh-y += rpmh-rsc.o qcom_rpmh-y += rpmh.odrivers/soc/qcom/rpm_master_stats.c new file +163
-0@@ -0,0 +1,163 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (c) 2012-2020, The Linux Foundation. All rights reserved. +* Copyright (c) 2023, Linaro Limited +* +* This driver supports what is known as "Master Stats v2" in Qualcomm +* downstream kernel terms, which seems to be the only version which has +* ever shipped, all the way from 2013 to 2023. +*/ + +#include <linux/debugfs.h> +#include <linux/io.h> +#include <linux/module.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/platform_device.h> + +struct master_stats_data { +void __iomem *base; +const char *label; +}; + +struct rpm_master_stats { +u32 active_cores; +u32 num_shutdowns; +u64 shutdown_req; +u64 wakeup_idx; +u64 bringup_req; +u64 bringup_ack; +u32 wakeup_reason;/* 0 = "rude wakeup", 1 = scheduled wakeup */ +u32 last_sleep_trans_dur; +u32 last_wake_trans_dur; + +/* Per-subsystem (*not necessarily* SoC-wide) XO shutdown stats */ +u32 xo_count; +u64 xo_last_enter; +u64 last_exit; +u64 xo_total_dur; +} __packed; + +static int master_stats_show (struct seq_file *s,void *unused) +{ +struct master_stats_data *data = s->private; +struct rpm_master_stats stat; + +memcpy_fromio (&stat, data->base,sizeof (stat)); + +seq_printf (s,"%s:\n" , data->label); + +seq_printf (s,"\tLast shutdown @ %llu\n" , stat.shutdown_req); +seq_printf (s,"\tLast bringup req @ %llu\n" , stat.bringup_req); +seq_printf (s,"\tLast bringup ack @ %llu\n" , stat.bringup_ack); +seq_printf (s,"\tLast wakeup idx: %llu\n" , stat.wakeup_idx); +seq_printf (s,"\tLast XO shutdown enter @ %llu\n" , stat.xo_last_enter); +seq_printf (s,"\tLast XO shutdown exit @ %llu\n" , stat.last_exit); +seq_printf (s,"\tXO total duration: %llu\n" , stat.xo_total_dur); +seq_printf (s,"\tLast sleep transition duration: %u\n" , stat.last_sleep_trans_dur); +seq_printf (s,"\tLast wake transition duration: %u\n" , stat.last_wake_trans_dur); +seq_printf (s,"\tXO shutdown count: %u\n" , stat.xo_count); +seq_printf (s,"\tWakeup reason: 0x%x\n" , stat.wakeup_reason); +seq_printf (s,"\tShutdown count: %u\n" , stat.num_shutdowns); +seq_printf (s,"\tActive cores bitmask: 0x%x\n" , stat.active_cores); + +return 0 ; +} +DEFINE_SHOW_ATTRIBUTE (master_stats); + +static int master_stats_probe (struct platform_device *pdev) +{ +struct device *dev = &pdev->dev; +struct master_stats_data *data; +struct device_node *msgram_np; +struct dentry *dent, *root; +struct resource res; +int count, i, ret; + + count =of_property_count_strings (dev->of_node,"qcom,master-names" ); +if (count <0 ) +return count; + + data =devm_kzalloc (dev, count *sizeof (*data),GFP_KERNEL ); +if (!data) +return -ENOMEM ; + + root =debugfs_create_dir ("qcom_rpm_master_stats" ,NULL ); +platform_set_drvdata (pdev, root); + +for (i =0 ; i < count; i++) { + msgram_np =of_parse_phandle (dev->of_node,"qcom,rpm-msg-ram" , i); +if (!msgram_np) { +debugfs_remove_recursive (root); +return dev_err_probe (dev, -ENODEV , +"Couldn't parse MSG RAM phandle idx %d" , i); + } + +/* +* Purposefully skip devm_platform helpers as we're using a +* shared resource. +*/ + ret =of_address_to_resource (msgram_np,0 , &res); +of_node_put (msgram_np); +if (ret <0 ) { +debugfs_remove_recursive (root); +return ret; + } + + data[i].base =devm_ioremap (dev, res.start,resource_size (&res)); +if (IS_ERR (data[i].base)) { +debugfs_remove_recursive (root); +return dev_err_probe (dev, -EINVAL , +"Could not map the MSG RAM slice idx %d!\n" , i); + } + + ret =of_property_read_string_index (dev->of_node,"qcom,master-names" , i, + &data[i].label); +if (ret <0 ) { +debugfs_remove_recursive (root); +return dev_err_probe (dev, ret, +"Could not read name idx %d!\n" , i); + } + +/* +* Generally it's not advised to fail on debugfs errors, but this +* driver's only job is exposing data therein. +*/ + dent =debugfs_create_file (data[i].label,0444 , root, + &data[i], &master_stats_fops); +if (IS_ERR (dent)) { +debugfs_remove_recursive (root); +return dev_err_probe (dev,PTR_ERR (dent), +"Failed to create debugfs file %s!\n" , data[i].label); + } + } + +device_set_pm_not_required (dev); + +return 0 ; +} + +static void master_stats_remove (struct platform_device *pdev) +{ +struct dentry *root =platform_get_drvdata (pdev); + +debugfs_remove_recursive (root); +} + +static const struct of_device_id rpm_master_table[] = { + { .compatible ="qcom,rpm-master-stats" }, + { }, +}; + +static struct platform_driver master_stats_driver = { + .probe = master_stats_probe, + .remove_new = master_stats_remove, + .driver = { + .name ="qcom_rpm_master_stats" , + .of_match_table = rpm_master_table, + }, +}; +module_platform_driver (master_stats_driver); + +MODULE_DESCRIPTION ("Qualcomm RPM Master Statistics driver" ); +MODULE_LICENSE ("GPL" );Signature · 2 lines
-- 2.40.0
Signature · 1 line
மணிவண்ணன் சதாசிவம்