diff options
author | Arnd Bergmann <arnd@arndb.de> | 2018-03-07 16:45:07 +0100 |
---|---|---|
committer | Arnd Bergmann <arnd@arndb.de> | 2018-03-07 16:45:07 +0100 |
commit | f46f11dc1e86270935041fbc3920ba71a050a5fd (patch) | |
tree | e049f00eb27f573e2f3d90d379554c869f6469c3 /drivers/cpufreq | |
parent | Merge tag 'scpi-updates-4.17' of ssh://gitolite.kernel.org/pub/scm/linux/kern... (diff) | |
parent | cpufreq: scmi: add support for fast frequency switching (diff) | |
download | linux-f46f11dc1e86270935041fbc3920ba71a050a5fd.tar.xz linux-f46f11dc1e86270935041fbc3920ba71a050a5fd.zip |
Merge tag 'scmi-updates-4.17' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux into next/drivers
Pull "ARM SCMI support for v4.17" from Sudeep Holla:
ARM System Control and Management Interface(SCMI)[1] is more flexible and
easily extensible than any of the existing interfaces.
Few existing as well as future ARM platforms provide micro-controllers
to abstract various power and other system management tasks which have
similar interfaces, both in terms of the functions that are provided by
them, and in terms of how requests are communicated to them.
There are quite a few protocols like ARM SCPI, TI SCI, QCOM RPM, Nvidia Tegra
BPMP, and so on already. This specification is to standardize and avoid any
further fragmentation in the design of such interface by various vendors.
The current SCMI driver implementation is very basic and initial support.
It lacks support for notifications, asynchronous/delayed response, perf/power
statistics region and sensor register region.
Mailbox is the only form of transport supported currently in the driver.
SCMI supports interrupt based mailbox communication, where, on completion
of the processing of a message, the caller receives an interrupt as well as
polling for completion.
SCMI is designed to minimize the dependency on the mailbox/transport
hardware. So in terms of SCMI, each channel in the mailbox includes
memory area, doorbell and completion interrupt.
However the doorbell and completion interrupt is highly mailbox dependent
which was bit of controversial as part of SCMI/mailbox discussions.
Arnd and me discussed about the few aspects of SCMI and the mailbox framework:
1. Use of mailbox framework for doorbell type mailbox controller:
- Such hardware may not require any data to be sent to signal the remote
about the presence of a message. The channel will have in-built
information on how to trigger the signal to the remote.
There are few mailbox controller drivers which are purely doorbell based.
e.g.QCOM IPC, STM, Tegra, ACPI PCC,..etc
2. Supporting other mailbox controller:
- SCMI just needs a mechanism to signal the remote firmware. Such
controller may need fixed message to be sent to trigger a doorbell.
In such case we may need to get that data from DT and pass the same
to the controller. It's not covered in the current DT binding, but
can be extended as optional property in future.
However handling notifications may be interesting on such mailbox, but
again there is no way to interpret what the data field(remote message)
means, it could be a bit mask or a number or don't-care.
Arnd mentioned that he doesn't like the way the mailbox binding deals
with doorbell-type hardware, but we do have quite a few precedent drivers
already and changing the binding to add a data field would not make it any
better, but could cause other problems. So he is happy with the status quo
of SCMI implementation.
[1] http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.den0056a/index.html
* tag 'scmi-updates-4.17' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux:
cpufreq: scmi: add support for fast frequency switching
cpufreq: add support for CPU DVFS based on SCMI message protocol
hwmon: add support for sensors exported via ARM SCMI
hwmon: (core) Add hwmon_max to hwmon_sensor_types enumeration
clk: add support for clocks provided by SCMI
firmware: arm_scmi: add device power domain support using genpd
firmware: arm_scmi: add per-protocol channels support using idr objects
firmware: arm_scmi: refactor in preparation to support per-protocol channels
firmware: arm_scmi: add option for polling based performance domain operations
firmware: arm_scmi: add support for polling based SCMI transfers
firmware: arm_scmi: probe and initialise all the supported protocols
firmware: arm_scmi: add initial support for sensor protocol
firmware: arm_scmi: add initial support for power protocol
firmware: arm_scmi: add initial support for clock protocol
firmware: arm_scmi: add initial support for performance protocol
firmware: arm_scmi: add scmi protocol bus to enumerate protocol devices
firmware: arm_scmi: add common infrastructure and support for base protocol
firmware: arm_scmi: add basic driver infrastructure for SCMI
dt-bindings: arm: add support for ARM System Control and Management Interface(SCMI) protocol
dt-bindings: mailbox: add support for mailbox client shared memory
Diffstat (limited to 'drivers/cpufreq')
-rw-r--r-- | drivers/cpufreq/Kconfig.arm | 11 | ||||
-rw-r--r-- | drivers/cpufreq/Makefile | 1 | ||||
-rw-r--r-- | drivers/cpufreq/scmi-cpufreq.c | 264 |
3 files changed, 276 insertions, 0 deletions
diff --git a/drivers/cpufreq/Kconfig.arm b/drivers/cpufreq/Kconfig.arm index fb586e09682d..9bbb5b39d18a 100644 --- a/drivers/cpufreq/Kconfig.arm +++ b/drivers/cpufreq/Kconfig.arm @@ -238,6 +238,17 @@ config ARM_SA1100_CPUFREQ config ARM_SA1110_CPUFREQ bool +config ARM_SCMI_CPUFREQ + tristate "SCMI based CPUfreq driver" + depends on ARM_SCMI_PROTOCOL || COMPILE_TEST + select PM_OPP + help + This adds the CPUfreq driver support for ARM platforms using SCMI + protocol for CPU power management. + + This driver uses SCMI Message Protocol driver to interact with the + firmware providing the CPU DVFS functionality. + config ARM_SPEAR_CPUFREQ bool "SPEAr CPUFreq support" depends on PLAT_SPEAR diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile index c60c1e141d9d..4987227b67df 100644 --- a/drivers/cpufreq/Makefile +++ b/drivers/cpufreq/Makefile @@ -75,6 +75,7 @@ obj-$(CONFIG_ARM_S3C24XX_CPUFREQ_DEBUGFS) += s3c24xx-cpufreq-debugfs.o obj-$(CONFIG_ARM_S5PV210_CPUFREQ) += s5pv210-cpufreq.o obj-$(CONFIG_ARM_SA1100_CPUFREQ) += sa1100-cpufreq.o obj-$(CONFIG_ARM_SA1110_CPUFREQ) += sa1110-cpufreq.o +obj-$(CONFIG_ARM_SCMI_CPUFREQ) += scmi-cpufreq.o obj-$(CONFIG_ARM_SCPI_CPUFREQ) += scpi-cpufreq.o obj-$(CONFIG_ARM_SPEAR_CPUFREQ) += spear-cpufreq.o obj-$(CONFIG_ARM_STI_CPUFREQ) += sti-cpufreq.o diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c new file mode 100644 index 000000000000..959a1dbe3835 --- /dev/null +++ b/drivers/cpufreq/scmi-cpufreq.c @@ -0,0 +1,264 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * System Control and Power Interface (SCMI) based CPUFreq Interface driver + * + * Copyright (C) 2018 ARM Ltd. + * Sudeep Holla <sudeep.holla@arm.com> + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include <linux/cpu.h> +#include <linux/cpufreq.h> +#include <linux/cpumask.h> +#include <linux/cpu_cooling.h> +#include <linux/export.h> +#include <linux/module.h> +#include <linux/pm_opp.h> +#include <linux/slab.h> +#include <linux/scmi_protocol.h> +#include <linux/types.h> + +struct scmi_data { + int domain_id; + struct device *cpu_dev; + struct thermal_cooling_device *cdev; +}; + +static const struct scmi_handle *handle; + +static unsigned int scmi_cpufreq_get_rate(unsigned int cpu) +{ + struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu); + struct scmi_perf_ops *perf_ops = handle->perf_ops; + struct scmi_data *priv = policy->driver_data; + unsigned long rate; + int ret; + + ret = perf_ops->freq_get(handle, priv->domain_id, &rate, false); + if (ret) + return 0; + return rate / 1000; +} + +/* + * perf_ops->freq_set is not a synchronous, the actual OPP change will + * happen asynchronously and can get notified if the events are + * subscribed for by the SCMI firmware + */ +static int +scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index) +{ + int ret; + struct scmi_data *priv = policy->driver_data; + struct scmi_perf_ops *perf_ops = handle->perf_ops; + u64 freq = policy->freq_table[index].frequency * 1000; + + ret = perf_ops->freq_set(handle, priv->domain_id, freq, false); + if (!ret) + arch_set_freq_scale(policy->related_cpus, freq, + policy->cpuinfo.max_freq); + return ret; +} + +static unsigned int scmi_cpufreq_fast_switch(struct cpufreq_policy *policy, + unsigned int target_freq) +{ + struct scmi_data *priv = policy->driver_data; + struct scmi_perf_ops *perf_ops = handle->perf_ops; + + if (!perf_ops->freq_set(handle, priv->domain_id, + target_freq * 1000, true)) { + arch_set_freq_scale(policy->related_cpus, target_freq, + policy->cpuinfo.max_freq); + return target_freq; + } + + return 0; +} + +static int +scmi_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask) +{ + int cpu, domain, tdomain; + struct device *tcpu_dev; + + domain = handle->perf_ops->device_domain_id(cpu_dev); + if (domain < 0) + return domain; + + for_each_possible_cpu(cpu) { + if (cpu == cpu_dev->id) + continue; + + tcpu_dev = get_cpu_device(cpu); + if (!tcpu_dev) + continue; + + tdomain = handle->perf_ops->device_domain_id(tcpu_dev); + if (tdomain == domain) + cpumask_set_cpu(cpu, cpumask); + } + + return 0; +} + +static int scmi_cpufreq_init(struct cpufreq_policy *policy) +{ + int ret; + unsigned int latency; + struct device *cpu_dev; + struct scmi_data *priv; + struct cpufreq_frequency_table *freq_table; + + cpu_dev = get_cpu_device(policy->cpu); + if (!cpu_dev) { + pr_err("failed to get cpu%d device\n", policy->cpu); + return -ENODEV; + } + + ret = handle->perf_ops->add_opps_to_device(handle, cpu_dev); + if (ret) { + dev_warn(cpu_dev, "failed to add opps to the device\n"); + return ret; + } + + ret = scmi_get_sharing_cpus(cpu_dev, policy->cpus); + if (ret) { + dev_warn(cpu_dev, "failed to get sharing cpumask\n"); + return ret; + } + + ret = dev_pm_opp_set_sharing_cpus(cpu_dev, policy->cpus); + if (ret) { + dev_err(cpu_dev, "%s: failed to mark OPPs as shared: %d\n", + __func__, ret); + return ret; + } + + ret = dev_pm_opp_get_opp_count(cpu_dev); + if (ret <= 0) { + dev_dbg(cpu_dev, "OPP table is not ready, deferring probe\n"); + ret = -EPROBE_DEFER; + goto out_free_opp; + } + + priv = kzalloc(sizeof(*priv), GFP_KERNEL); + if (!priv) { + ret = -ENOMEM; + goto out_free_opp; + } + + ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table); + if (ret) { + dev_err(cpu_dev, "failed to init cpufreq table: %d\n", ret); + goto out_free_priv; + } + + priv->cpu_dev = cpu_dev; + priv->domain_id = handle->perf_ops->device_domain_id(cpu_dev); + + policy->driver_data = priv; + + ret = cpufreq_table_validate_and_show(policy, freq_table); + if (ret) { + dev_err(cpu_dev, "%s: invalid frequency table: %d\n", __func__, + ret); + goto out_free_cpufreq_table; + } + + /* SCMI allows DVFS request for any domain from any CPU */ + policy->dvfs_possible_from_any_cpu = true; + + latency = handle->perf_ops->get_transition_latency(handle, cpu_dev); + if (!latency) + latency = CPUFREQ_ETERNAL; + + policy->cpuinfo.transition_latency = latency; + + policy->fast_switch_possible = true; + return 0; + +out_free_cpufreq_table: + dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table); +out_free_priv: + kfree(priv); +out_free_opp: + dev_pm_opp_cpumask_remove_table(policy->cpus); + + return ret; +} + +static int scmi_cpufreq_exit(struct cpufreq_policy *policy) +{ + struct scmi_data *priv = policy->driver_data; + + cpufreq_cooling_unregister(priv->cdev); + dev_pm_opp_free_cpufreq_table(priv->cpu_dev, &policy->freq_table); + kfree(priv); + dev_pm_opp_cpumask_remove_table(policy->related_cpus); + + return 0; +} + +static void scmi_cpufreq_ready(struct cpufreq_policy *policy) +{ + struct scmi_data *priv = policy->driver_data; + + priv->cdev = of_cpufreq_cooling_register(policy); +} + +static struct cpufreq_driver scmi_cpufreq_driver = { + .name = "scmi", + .flags = CPUFREQ_STICKY | CPUFREQ_HAVE_GOVERNOR_PER_POLICY | + CPUFREQ_NEED_INITIAL_FREQ_CHECK, + .verify = cpufreq_generic_frequency_table_verify, + .attr = cpufreq_generic_attr, + .target_index = scmi_cpufreq_set_target, + .fast_switch = scmi_cpufreq_fast_switch, + .get = scmi_cpufreq_get_rate, + .init = scmi_cpufreq_init, + .exit = scmi_cpufreq_exit, + .ready = scmi_cpufreq_ready, +}; + +static int scmi_cpufreq_probe(struct scmi_device *sdev) +{ + int ret; + + handle = sdev->handle; + + if (!handle || !handle->perf_ops) + return -ENODEV; + + ret = cpufreq_register_driver(&scmi_cpufreq_driver); + if (ret) { + dev_err(&sdev->dev, "%s: registering cpufreq failed, err: %d\n", + __func__, ret); + } + + return ret; +} + +static void scmi_cpufreq_remove(struct scmi_device *sdev) +{ + cpufreq_unregister_driver(&scmi_cpufreq_driver); +} + +static const struct scmi_device_id scmi_id_table[] = { + { SCMI_PROTOCOL_PERF }, + { }, +}; +MODULE_DEVICE_TABLE(scmi, scmi_id_table); + +static struct scmi_driver scmi_cpufreq_drv = { + .name = "scmi-cpufreq", + .probe = scmi_cpufreq_probe, + .remove = scmi_cpufreq_remove, + .id_table = scmi_id_table, +}; +module_scmi_driver(scmi_cpufreq_drv); + +MODULE_AUTHOR("Sudeep Holla <sudeep.holla@arm.com>"); +MODULE_DESCRIPTION("ARM SCMI CPUFreq interface driver"); +MODULE_LICENSE("GPL v2"); |