Re: [PATCH v1 1/8] thermal: Add Remote Proc cooling driver
From: Gaurav Kohli
Date: Mon Feb 09 2026 - 00:29:04 EST
On 2/2/2026 4:29 PM, Lukasz Luba wrote:
Hi Gaurav,
On 12/23/25 12:32, Gaurav Kohli wrote:
Add a new generic driver for thermal cooling devices that control
remote processors (modem, DSP, etc.) through various communication
channels.
This driver provides an abstraction layer between the thermal
subsystem and vendor-specific remote processor communication
mechanisms.
Is this the patch about proposing the new cooling
device type at last LPC2025 conference (what we've discussed with Amit)?
thanks Lukasz for review, yes this is the same.
sorry for late reply, was on leave last week.
There was some feedback asking you to add a bit more description
into this patch header, please do that (with some background as well).
Sure, will update.
Suggested-by: Amit Kucheria <amit.kucheria@xxxxxxxxxxxxxxxx>
Signed-off-by: Gaurav Kohli <gaurav.kohli@xxxxxxxxxxxxxxxx>
---
MAINTAINERS | 8 ++
drivers/thermal/Kconfig | 11 ++
drivers/thermal/Makefile | 2 +
drivers/thermal/remoteproc_cooling.c | 154 +++++++++++++++++++++++++++
include/linux/remoteproc_cooling.h | 52 +++++++++
5 files changed, 227 insertions(+)
create mode 100644 drivers/thermal/remoteproc_cooling.c
create mode 100644 include/linux/remoteproc_cooling.h
diff --git a/MAINTAINERS b/MAINTAINERS
index 679e5f11e672..c1ba87315cdf 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -25935,6 +25935,14 @@ F: drivers/thermal/cpufreq_cooling.c
F: drivers/thermal/cpuidle_cooling.c
F: include/linux/cpu_cooling.h
+THERMAL/REMOTEPROC_COOLING
+M: Gaurav Kohli <gaurav.kohli@xxxxxxxxxxxxxxxx>
+L: linux-pm@xxxxxxxxxxxxxxx
+S: Supported
+F: drivers/thermal/remoteproc_cooling.c
+F: include/linux/remoteproc_cooling.h
+
+
THERMAL/POWER_ALLOCATOR
M: Lukasz Luba <lukasz.luba@xxxxxxx>
L: linux-pm@xxxxxxxxxxxxxxx
diff --git a/drivers/thermal/Kconfig b/drivers/thermal/Kconfig
index b10080d61860..31e92be34387 100644
--- a/drivers/thermal/Kconfig
+++ b/drivers/thermal/Kconfig
@@ -229,6 +229,17 @@ config PCIE_THERMAL
If you want this support, you should say Y here.
+
+config REMOTEPROC_THERMAL
+ bool "Remote processor cooling support"
+ help
+ This implements a generic cooling mechanism for remote processors
+ (modem, DSP, etc.) that allows vendor-specific implementations to
+ register thermal cooling devices and provide callbacks for thermal
+ mitigation.
+
+ If you want this support, you should say Y here.
+
config THERMAL_EMULATION
bool "Thermal emulation mode support"
help
diff --git a/drivers/thermal/Makefile b/drivers/thermal/Makefile
index bb21e7ea7fc6..ae747dde54fe 100644
--- a/drivers/thermal/Makefile
+++ b/drivers/thermal/Makefile
@@ -34,6 +34,8 @@ thermal_sys-$(CONFIG_DEVFREQ_THERMAL) += devfreq_cooling.o
thermal_sys-$(CONFIG_PCIE_THERMAL) += pcie_cooling.o
+thermal_sys-$(CONFIG_REMOTEPROC_THERMAL) += remoteproc_cooling.o
+
obj-$(CONFIG_K3_THERMAL) += k3_bandgap.o k3_j72xx_bandgap.o
# platform thermal drivers
obj-y += broadcom/
diff --git a/drivers/thermal/remoteproc_cooling.c b/drivers/thermal/ remoteproc_cooling.c
new file mode 100644
index 000000000000..a1f948cbde0f
--- /dev/null
+++ b/drivers/thermal/remoteproc_cooling.c
@@ -0,0 +1,154 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Remote Processor Cooling Device
+ *
+ * Copyright (c) 2025, Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/of.h>
+#include <linux/slab.h>
+#include <linux/thermal.h>
+
+#define REMOTEPROC_PREFIX "rproc_"
+
+struct remoteproc_cooling_ops {
+ int (*get_max_level)(void *devdata, unsigned long *level);
+ int (*get_cur_level)(void *devdata, unsigned long *level);
+ int (*set_cur_level)(void *devdata, unsigned long level);
+};
1. There is no comment for struct and the functions like you did below.
2. Why you need those 3 callbacks?
It looks like they are simple wrappers on stuff in
'struct thermal_cooling_device_ops'.
Please try to get rid of them and re-use the existing fwk callbacks.
thanks for this suggestion, i will use thermal_cooling_device_ops directly.
+
+/**
+ * struct remoteproc_cdev - Remote processor cooling device
+ * @cdev: Thermal cooling device handle
+ * @ops: Vendor-specific operation callbacks
+ * @devdata: Private data for vendor implementation
+ * @np: Device tree node associated with this cooling device
+ * @lock: Mutex to protect cooling device operations
+ */
+struct remoteproc_cdev {
Please use the full naming:
remoteproc_cooling_device
+ struct thermal_cooling_device *cdev;
You don't need to keep it here. AFAICS it's only
used in the 'unregister' function. Please check my
comment here and then remove this pointer.
(It creates uneseccery linkage between those devices).
+ const struct remoteproc_cooling_ops *ops;
So here it can be simply:
struct thermal_cooling_device_ops cooling_ops;
yes, i will use this as part of remoteproc_cooling_device struct.
+ void *devdata;
+ struct device_node *np;
This 'np' is also not used, remove it please.
+ struct mutex lock;
+};
+
+
+/* Thermal cooling device callbacks */
+
+static int remoteproc_get_max_state(struct thermal_cooling_device *cdev,
+ unsigned long *state)
+{
+ struct remoteproc_cdev *rproc_cdev = cdev->devdata;
+ int ret;
+
+ if (!rproc_cdev || !rproc_cdev->ops)
+ return -EINVAL;
This mustn't be changed in runtime accidenly. We don't guard in
cpufreq-/devfreq- cooling these callbacks that way. Please drop them.
Sure, let me rewrite this and update in next version.
+
+ mutex_lock(&rproc_cdev->lock);
+ ret = rproc_cdev->ops->get_max_level(rproc_cdev->devdata, state);
+ mutex_unlock(&rproc_cdev->lock);
+
+ return ret;
+}
+
+static int remoteproc_get_cur_state(struct thermal_cooling_device *cdev,
+ unsigned long *state)
+{
+ struct remoteproc_cdev *rproc_cdev = cdev->devdata;
+ int ret;
+
+ if (!rproc_cdev || !rproc_cdev->ops)
+ return -EINVAL;
as above
+
+ mutex_lock(&rproc_cdev->lock);
+ ret = rproc_cdev->ops->get_cur_level(rproc_cdev->devdata, state);
+ mutex_unlock(&rproc_cdev->lock);
+
+ return ret;
+}
+
+static int remoteproc_set_cur_state(struct thermal_cooling_device *cdev,
+ unsigned long state)
+{
+ struct remoteproc_cdev *rproc_cdev = cdev->devdata;
+ int ret;
+
+ if (!rproc_cdev || !rproc_cdev->ops)
+ return -EINVAL;
as above
will fix in all callbacks.
+
+ mutex_lock(&rproc_cdev->lock);
+ ret = rproc_cdev->ops->set_cur_level(rproc_cdev->devdata, state);
+ mutex_unlock(&rproc_cdev->lock);
+
+ return ret;
+}
+
+static const struct thermal_cooling_device_ops remoteproc_cooling_ops = {
+ .get_max_state = remoteproc_get_max_state,
+ .get_cur_state = remoteproc_get_cur_state,
+ .set_cur_state = remoteproc_set_cur_state,
+};
+
+struct remoteproc_cdev *
+remoteproc_cooling_register(struct device_node *np,
+ const char *name, const struct remoteproc_cooling_ops *ops,
+ void *devdata)
Since this is based on device_node, please align to the naming
convention from cpufreq-/devfreq-cooling and use prefix 'of_'
of_remoteproc_cooling_register()
+{
+ struct remoteproc_cdev *rproc_cdev;
+ struct thermal_cooling_device *cdev;
+ int ret;
+
+ if (!name || !ops) {
IMO you should check the '!np' here, not the lines below.
We can simply bail out very early.
thanks will put explicit check for np, but please let me know for non np, do we have to add support for non np also.
so they can directly register with thermal_cooling_device_register.
+ return ERR_PTR(-EINVAL);
+ }
+
+ rproc_cdev = kzalloc(sizeof(*rproc_cdev), GFP_KERNEL);
+ if (!rproc_cdev)
+ return ERR_PTR(-ENOMEM);
+
+ rproc_cdev->ops = ops;
+ rproc_cdev->devdata = devdata;
+ rproc_cdev->np = np;
+ mutex_init(&rproc_cdev->lock);
+
+ char *rproc_name __free(kfree) =
+ kasprintf(GFP_KERNEL, REMOTEPROC_PREFIX "%s", name);
+ /* Register with thermal framework */
+ if (np) {
Too late to test 'np'.
yes, will put a check in the start of function.
+ cdev = thermal_of_cooling_device_register(np, rproc_name, rproc_cdev,
+ &remoteproc_cooling_ops);
+ }
+
+ if (IS_ERR(cdev)) {
This check should be just below the line when you get the 'cdev'
+ ret = PTR_ERR(cdev);
+ goto free_rproc_cdev;
+ }
+
+ rproc_cdev->cdev = cdev;
+
+ return rproc_cdev;
+
+free_rproc_cdev:
+ kfree(rproc_cdev);
+ return ERR_PTR(ret);
+}
+EXPORT_SYMBOL_GPL(remoteproc_cooling_register);
+
+void remoteproc_cooling_unregister(struct remoteproc_cdev *rproc_cdev)
Change the API to be alined with cpufreq-cooling and devfreq-cooling
types of devices, so:
void remoteproc_cooling_unregister(struct thermal_cooling_device *cdev)
You still should be able to get the rptoc_cdev like:
rproc_cdev = cdev->devdata;
and free it.
thanks, will change something like below
+ rproc_cdev = cdev->devdata;
+ thermal_cooling_device_unregister(cdev);
+{
+ if (!rproc_cdev)
+ return;
+
+ thermal_cooling_device_unregister(rproc_cdev->cdev);
+ mutex_destroy(&rproc_cdev->lock);
+ kfree(rproc_cdev);
+}
+EXPORT_SYMBOL_GPL(remoteproc_cooling_unregister);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Remote Processor Cooling Device");
diff --git a/include/linux/remoteproc_cooling.h b/include/linux/ remoteproc_cooling.h
new file mode 100644
index 000000000000..ef94019d220d
--- /dev/null
+++ b/include/linux/remoteproc_cooling.h
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Remote Processor Cooling Device
+ *
+ * Copyright (c) 2025, Qualcomm Innovation Center
+ */
+
+#ifndef __REMOTEPROC_COOLING_H__
+#define __REMOTEPROC_COOLING_H__
+
+#include <linux/thermal.h>
+
+struct device;
+struct device_node;
+
+struct remoteproc_cooling_ops {
+ int (*get_max_level)(void *devdata, unsigned long *level);
+ int (*get_cur_level)(void *devdata, unsigned long *level);
+ int (*set_cur_level)(void *devdata, unsigned long level);
+};
That duplicate w/ .c file content.
We don't need this in the header, please follow the cpufreq-/devfreq-
design.
Yes, with new approach of using thermal_cooling_device_ops directly can
save this.
+
+struct remoteproc_cdev;
+
+#ifdef CONFIG_REMOTEPROC_THERMAL
+
+struct remoteproc_cdev *
+remoteproc_cooling_register(struct device_node *np,
+ const char *name,
+ const struct remoteproc_cooling_ops *ops,
+ void *devdata);
+
+void remoteproc_cooling_unregister(struct remoteproc_cdev *rproc_cdev);
+
+#else /* !CONFIG_REMOTEPROC_THERMAL */
+
+static inline struct remoteproc_cdev *
+remoteproc_cooling_register(struct device_node *np,
+ const char *name,
+ const struct remoteproc_cooling_ops *ops,
+ void *devdata)
+{
+ return ERR_PTR(-EINVAL);
+}
Function naming convention here as well
thanks a lot, let me rewrite as per suggestion and update in newer version.
Regards,
Lukasz