[PATCH v3 00/11] Raspberry Pi PoE HAT fan support

[PATCH v3 00/11] Raspberry Pi PoE HAT fan support

[Nicolas Saenz Julienne]

The aim of this series is to add support to the fan found on RPi's PoE
HAT. Some commentary on the design can be found below. But the imporant
part to the people CC'd here not involved with PWM is that, in order to
achieve this properly, we also have to fix the firmware interface the
driver uses to communicate with the PWM bus (and many other low level
functions). Specifically, we have to make sure the firmware interface
isn't unbound while consumers are still up. So, patch #1 introduces
reference counting in the firwmware interface driver and patches #2 to
#7 update all firmware users. Patches #8 to #10 introduce the new PWM
driver.

I sent everything as a single series as the final version of the PWM
drivers depends on the firwmare fixes, but I'll be happy to split this
into two separate series if you think it's better.

--- Original cover letter below ---

This series aims at adding support to RPi's official PoE HAT fan[1].

The HW setup is the following:

| Raspberry Pi                               | PoE HAT                    |
 arm core -> Mailbox -> RPi co-processor -> I2C -> Atmel MCU -> PWM -> FAN

The arm cores have only access to the mailbox interface, as i2c0, even if
physically accessible, is to be used solely by the co-processor
(VideoCore 4/6).

This series implements a PWM bus, and has pwm-fan sitting on top of it as per
this discussion: https://lkml.org/lkml/2018/9/2/486. Although this design has a
series of shortcomings:

- It depends on a DT binding: it's not flexible if a new hat shows up with new
  functionality, we're not 100% sure we'll be able to expand it without
  breaking backwards compatibility. But without it we can't make use of DT
  thermal-zones, which IMO is overkill.

- We're using pwm-fan, writing a hwmon driver would, again, give us more
  flexibility, but it's not really needed at the moment.

I personally think that it's not worth the effort, it's unlikely we'll get
things right in advance. And ultimately, if the RPi people come up with
something new, we can always write a new driver/bindings from scratch (as in
not reusing previous code).

That said, I'm more than happy to change things if there is a consensus that
another design will do the trick.

[1] https://www.raspberrypi.org/blog/introducing-power-over-ethernet-poe-hat/

---

Changes since v2:
 - Introduce devm_rpi_firmware_get()
 - Small cleanups in PWM driver

Changes since v1:
 - Address PWM driver changes
 - Fix binding, now with 2 cells
 - Add reference count to rpi_firmware_get()

Nicolas Saenz Julienne (11):
  firmware: raspberrypi: Introduce devm_rpi_firmware_get()
  clk: bcm: rpi: Release firmware handle on unbind
  gpio: raspberrypi-exp: Release firmware handle on unbind
  reset: raspberrypi: Release firmware handle on unbind
  soc: bcm: raspberrypi-power: Release firmware handle on unbind
  staging: vchiq: Release firmware handle on unbind
  input: raspberrypi-ts: Release firmware handle when not needed
  firmware: raspberrypi: Get rid of rpi_firmware_get()
  dt-bindings: pwm: Add binding for RPi firmware PWM bus
  DO NOT MERGE: ARM: dts: Add RPi's official PoE hat support
  pwm: Add Raspberry Pi Firmware based PWM bus

 .../arm/bcm/raspberrypi,bcm2835-firmware.yaml |  20 ++
 arch/arm/boot/dts/bcm2711-rpi-4-b.dts         |  54 +++++
 drivers/clk/bcm/clk-raspberrypi.c             |   2 +-
 drivers/firmware/raspberrypi.c                |  37 ++-
 drivers/gpio/gpio-raspberrypi-exp.c           |   2 +-
 drivers/input/touchscreen/raspberrypi-ts.c    |   2 +-
 drivers/pwm/Kconfig                           |   9 +
 drivers/pwm/Makefile                          |   1 +
 drivers/pwm/pwm-raspberrypi-poe.c             | 216 ++++++++++++++++++
 drivers/reset/reset-raspberrypi.c             |   2 +-
 drivers/soc/bcm/raspberrypi-power.c           |   2 +-
 .../interface/vchiq_arm/vchiq_arm.c           |   2 +-
 .../pwm/raspberrypi,firmware-pwm.h            |  13 ++
 include/soc/bcm2835/raspberrypi-firmware.h    |   6 +-
 14 files changed, 356 insertions(+), 12 deletions(-)
 create mode 100644 drivers/pwm/pwm-raspberrypi-poe.c
 create mode 100644 include/dt-bindings/pwm/raspberrypi,firmware-pwm.h

-- 
2.29.1

[PATCH v3 01/11] firmware: raspberrypi: Introduce devm_rpi_firmware_get()

[Nicolas Saenz Julienne]

When unbinding the firmware device we need to make sure it has no
consumers left. Otherwise we'd leave them with a firmware handle
pointing at freed memory.

Keep a reference count of all consumers and introduce
devm_rpi_firmware_get() which will automatically decrease the reference
count upon unbinding consumer drivers.

Suggested-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Create devm_rpi_firmware_get()

 drivers/firmware/raspberrypi.c             | 46 ++++++++++++++++++++++
 include/soc/bcm2835/raspberrypi-firmware.h |  8 ++++
 2 files changed, 54 insertions(+)

diff --git a/drivers/firmware/raspberrypi.c b/drivers/firmware/raspberrypi.c
index 2371d08bdd17..74bdb3bde9dc 100644
--- a/drivers/firmware/raspberrypi.c
+++ b/drivers/firmware/raspberrypi.c
@@ -11,7 +11,9 @@
 #include <linux/module.h>
 #include <linux/of_platform.h>
 #include <linux/platform_device.h>
+#include <linux/refcount.h>
 #include <linux/slab.h>
+#include <linux/wait.h>
 #include <soc/bcm2835/raspberrypi-firmware.h>
 
 #define MBOX_MSG(chan, data28)		(((data28) & ~0xf) | ((chan) & 0xf))
@@ -27,6 +29,9 @@ struct rpi_firmware {
 	struct mbox_chan *chan; /* The property channel. */
 	struct completion c;
 	u32 enabled;
+
+	refcount_t consumers;
+	wait_queue_head_t wait;
 };
 
 static DEFINE_MUTEX(transaction_lock);
@@ -247,6 +252,8 @@ static int rpi_firmware_probe(struct platform_device *pdev)
 	}
 
 	init_completion(&fw->c);
+	refcount_set(&fw->consumers, 1);
+	init_waitqueue_head(&fw->wait);
 
 	platform_set_drvdata(pdev, fw);
 
@@ -275,11 +282,21 @@ static int rpi_firmware_remove(struct platform_device *pdev)
 	rpi_hwmon = NULL;
 	platform_device_unregister(rpi_clk);
 	rpi_clk = NULL;
+
+	wait_event(fw->wait, refcount_dec_if_one(&fw->consumers));
 	mbox_free_channel(fw->chan);
 
 	return 0;
 }
 
+static void rpi_firmware_put(void *data)
+{
+	struct rpi_firmware *fw = data;
+
+	refcount_dec(&fw->consumers);
+	wake_up(&fw->wait);
+}
+
 /**
  * rpi_firmware_get - Get pointer to rpi_firmware structure.
  * @firmware_node:    Pointer to the firmware Device Tree node.
@@ -297,6 +314,35 @@ struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node)
 }
 EXPORT_SYMBOL_GPL(rpi_firmware_get);
 
+/**
+ * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
+ * @firmware_node:    Pointer to the firmware Device Tree node.
+ *
+ * Returns NULL is the firmware device is not ready.
+ */
+struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
+					   struct device_node *firmware_node)
+{
+	struct platform_device *pdev = of_find_device_by_node(firmware_node);
+	struct rpi_firmware *fw;
+
+	if (!pdev)
+		return NULL;
+
+	fw = platform_get_drvdata(pdev);
+	if (!fw)
+		return NULL;
+
+	if (!refcount_inc_not_zero(&fw->consumers))
+		return NULL;
+
+	if (devm_add_action_or_reset(dev, rpi_firmware_put, fw))
+		return NULL;
+
+	return fw;
+}
+EXPORT_SYMBOL_GPL(devm_rpi_firmware_get);
+
 static const struct of_device_id rpi_firmware_of_match[] = {
 	{ .compatible = "raspberrypi,bcm2835-firmware", },
 	{},
diff --git a/include/soc/bcm2835/raspberrypi-firmware.h b/include/soc/bcm2835/raspberrypi-firmware.h
index cc9cdbc66403..8fe64f53a394 100644
--- a/include/soc/bcm2835/raspberrypi-firmware.h
+++ b/include/soc/bcm2835/raspberrypi-firmware.h
@@ -141,6 +141,8 @@ int rpi_firmware_property(struct rpi_firmware *fw,
 int rpi_firmware_property_list(struct rpi_firmware *fw,
 			       void *data, size_t tag_size);
 struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node);
+struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
+					   struct device_node *firmware_node);
 #else
 static inline int rpi_firmware_property(struct rpi_firmware *fw, u32 tag,
 					void *data, size_t len)
@@ -158,6 +160,12 @@ static inline struct rpi_firmware *rpi_firmware_get(struct device_node *firmware
 {
 	return NULL;
 }
+
+static inline struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
+					struct device_node *firmware_node)
+{
+	return NULL;
+}
 #endif
 
 #endif /* __SOC_RASPBERRY_FIRMWARE_H__ */
-- 
2.29.1

[PATCH v3 02/11] clk: bcm: rpi: Release firmware handle on unbind

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/clk/bcm/clk-raspberrypi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/clk/bcm/clk-raspberrypi.c b/drivers/clk/bcm/clk-raspberrypi.c
index f89b9cfc4309..dd3b71eafabf 100644
--- a/drivers/clk/bcm/clk-raspberrypi.c
+++ b/drivers/clk/bcm/clk-raspberrypi.c
@@ -314,7 +314,7 @@ static int raspberrypi_clk_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	firmware = rpi_firmware_get(firmware_node);
+	firmware = devm_rpi_firmware_get(&pdev->dev, firmware_node);
 	of_node_put(firmware_node);
 	if (!firmware)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 03/11] gpio: raspberrypi-exp: Release firmware handle on unbind

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/gpio/gpio-raspberrypi-exp.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpio/gpio-raspberrypi-exp.c b/drivers/gpio/gpio-raspberrypi-exp.c
index bb100e0124e6..64a552ecc2ad 100644
--- a/drivers/gpio/gpio-raspberrypi-exp.c
+++ b/drivers/gpio/gpio-raspberrypi-exp.c
@@ -208,7 +208,7 @@ static int rpi_exp_gpio_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	fw = rpi_firmware_get(fw_node);
+	fw = devm_rpi_firmware_get(&pdev->dev, fw_node);
 	of_node_put(fw_node);
 	if (!fw)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 04/11] reset: raspberrypi: Release firmware handle on unbind

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/reset/reset-raspberrypi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/reset/reset-raspberrypi.c b/drivers/reset/reset-raspberrypi.c
index 02f59c06f69b..fa23db554bcf 100644
--- a/drivers/reset/reset-raspberrypi.c
+++ b/drivers/reset/reset-raspberrypi.c
@@ -82,7 +82,7 @@ static int rpi_reset_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	fw = rpi_firmware_get(np);
+	fw = devm_rpi_firmware_get(&pdev->dev, np);
 	of_node_put(np);
 	if (!fw)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 05/11] soc: bcm: raspberrypi-power: Release firmware handle on unbind

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/soc/bcm/raspberrypi-power.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/soc/bcm/raspberrypi-power.c b/drivers/soc/bcm/raspberrypi-power.c
index 5d1aacdd84ef..068715d6e66d 100644
--- a/drivers/soc/bcm/raspberrypi-power.c
+++ b/drivers/soc/bcm/raspberrypi-power.c
@@ -177,7 +177,7 @@ static int rpi_power_probe(struct platform_device *pdev)
 		return -ENODEV;
 	}
 
-	rpi_domains->fw = rpi_firmware_get(fw_np);
+	rpi_domains->fw = devm_rpi_firmware_get(&pdev->dev, fw_np);
 	of_node_put(fw_np);
 	if (!rpi_domains->fw)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 06/11] staging: vchiq: Release firmware handle on unbind

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c
index 01125d9f991b..3c25e237e334 100644
--- a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c
+++ b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c
@@ -2723,7 +2723,7 @@ static int vchiq_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	drvdata->fw = rpi_firmware_get(fw_node);
+	drvdata->fw = devm_rpi_firmware_get(&pdev->dev, fw_node);
 	of_node_put(fw_node);
 	if (!drvdata->fw)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 07/11] input: raspberrypi-ts: Release firmware handle when not needed

[Nicolas Saenz Julienne]

Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
interface when unbinding the device.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get(), instead of remove function

 drivers/input/touchscreen/raspberrypi-ts.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/input/touchscreen/raspberrypi-ts.c b/drivers/input/touchscreen/raspberrypi-ts.c
index ef6aaed217cf..efed0efa91d9 100644
--- a/drivers/input/touchscreen/raspberrypi-ts.c
+++ b/drivers/input/touchscreen/raspberrypi-ts.c
@@ -134,7 +134,7 @@ static int rpi_ts_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	fw = rpi_firmware_get(fw_node);
+	fw = devm_rpi_firmware_get(&pdev->dev, fw_node);
 	of_node_put(fw_node);
 	if (!fw)
 		return -EPROBE_DEFER;
-- 
2.29.1

[PATCH v3 08/11] firmware: raspberrypi: Get rid of rpi_firmware_get()

[Nicolas Saenz Julienne]

There a no users left to the function as they all converted to
devm_rpi_firmware_get(). Just get rid of it.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
---
 drivers/firmware/raspberrypi.c             | 17 -----------------
 include/soc/bcm2835/raspberrypi-firmware.h |  6 ------
 2 files changed, 23 deletions(-)

diff --git a/drivers/firmware/raspberrypi.c b/drivers/firmware/raspberrypi.c
index 74bdb3bde9dc..67f7105f3fd2 100644
--- a/drivers/firmware/raspberrypi.c
+++ b/drivers/firmware/raspberrypi.c
@@ -297,23 +297,6 @@ static void rpi_firmware_put(void *data)
 	wake_up(&fw->wait);
 }
 
-/**
- * rpi_firmware_get - Get pointer to rpi_firmware structure.
- * @firmware_node:    Pointer to the firmware Device Tree node.
- *
- * Returns NULL is the firmware device is not ready.
- */
-struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node)
-{
-	struct platform_device *pdev = of_find_device_by_node(firmware_node);
-
-	if (!pdev)
-		return NULL;
-
-	return platform_get_drvdata(pdev);
-}
-EXPORT_SYMBOL_GPL(rpi_firmware_get);
-
 /**
  * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
  * @firmware_node:    Pointer to the firmware Device Tree node.
diff --git a/include/soc/bcm2835/raspberrypi-firmware.h b/include/soc/bcm2835/raspberrypi-firmware.h
index 8fe64f53a394..eaa4e516e4c6 100644
--- a/include/soc/bcm2835/raspberrypi-firmware.h
+++ b/include/soc/bcm2835/raspberrypi-firmware.h
@@ -140,7 +140,6 @@ int rpi_firmware_property(struct rpi_firmware *fw,
 			  u32 tag, void *data, size_t len);
 int rpi_firmware_property_list(struct rpi_firmware *fw,
 			       void *data, size_t tag_size);
-struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node);
 struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
 					   struct device_node *firmware_node);
 #else
@@ -156,11 +155,6 @@ static inline int rpi_firmware_property_list(struct rpi_firmware *fw,
 	return -ENOSYS;
 }
 
-static inline struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node)
-{
-	return NULL;
-}
-
 static inline struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
 					struct device_node *firmware_node)
 {
-- 
2.29.1

[PATCH v3 09/11] dt-bindings: pwm: Add binding for RPi firmware PWM bus

[Nicolas Saenz Julienne]

The PWM bus controlling the fan in RPi's official PoE hat can only be
controlled by the board's co-processor.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Reviewed-by: Rob Herring <robh@kernel.org>

---

Changes since v1:
 - Update bindings to use 2 #pwm-cells

 .../arm/bcm/raspberrypi,bcm2835-firmware.yaml | 20 +++++++++++++++++++
 .../pwm/raspberrypi,firmware-pwm.h            | 13 ++++++++++++
 2 files changed, 33 insertions(+)
 create mode 100644 include/dt-bindings/pwm/raspberrypi,firmware-pwm.h

diff --git a/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.yaml b/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.yaml
index a2c63c8b1d10..8029ce958c48 100644
--- a/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.yaml
+++ b/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.yaml
@@ -64,6 +64,21 @@ properties:
       - compatible
       - "#reset-cells"
 
+  pwm:
+    type: object
+
+    properties:
+      compatible:
+        const: raspberrypi,firmware-pwm
+
+      "#pwm-cells":
+        # See pwm.yaml in this directory for a description of the cells format.
+        const: 2
+
+    required:
+      - compatible
+      - "#pwm-cells"
+
     additionalProperties: false
 
 required:
@@ -87,5 +102,10 @@ examples:
             compatible = "raspberrypi,firmware-reset";
             #reset-cells = <1>;
         };
+
+        pwm: pwm {
+            compatible = "raspberrypi,firmware-pwm";
+            #pwm-cells = <1>;
+        };
     };
 ...
diff --git a/include/dt-bindings/pwm/raspberrypi,firmware-pwm.h b/include/dt-bindings/pwm/raspberrypi,firmware-pwm.h
new file mode 100644
index 000000000000..27c5ce68847b
--- /dev/null
+++ b/include/dt-bindings/pwm/raspberrypi,firmware-pwm.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2020 Nicolas Saenz Julienne
+ * Author: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
+ */
+
+#ifndef _DT_BINDINGS_RASPBERRYPI_FIRMWARE_PWM_H
+#define _DT_BINDINGS_RASPBERRYPI_FIRMWARE_PWM_H
+
+#define RASPBERRYPI_FIRMWARE_PWM_POE		0
+#define RASPBERRYPI_FIRMWARE_PWM_NUM		1
+
+#endif
-- 
2.29.1

[PATCH v3 10/11] DO NOT MERGE: ARM: dts: Add RPi's official PoE hat support

[Nicolas Saenz Julienne]

This is an example on how to enable the fan on top of RPi's official PoE
hat.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v1:
 - Update patch to use 2 pwm cells

 arch/arm/boot/dts/bcm2711-rpi-4-b.dts | 54 +++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)

diff --git a/arch/arm/boot/dts/bcm2711-rpi-4-b.dts b/arch/arm/boot/dts/bcm2711-rpi-4-b.dts
index 09a1182c2936..361db82619a4 100644
--- a/arch/arm/boot/dts/bcm2711-rpi-4-b.dts
+++ b/arch/arm/boot/dts/bcm2711-rpi-4-b.dts
@@ -5,6 +5,7 @@
 #include "bcm283x-rpi-usb-peripheral.dtsi"
 
 #include <dt-bindings/reset/raspberrypi,firmware-reset.h>
+#include <dt-bindings/pwm/raspberrypi,firmware-pwm.h>
 
 / {
 	compatible = "raspberrypi,4-model-b", "brcm,bcm2711";
@@ -68,6 +69,54 @@ sd_vcc_reg: sd_vcc_reg {
 		enable-active-high;
 		gpio = <&expgpio 6 GPIO_ACTIVE_HIGH>;
 	};
+
+	fan: pwm-fan {
+		compatible = "pwm-fan";
+		cooling-levels = <0 50 150 255>;
+		#cooling-cells = <2>;
+		pwms = <&fwpwm RASPBERRYPI_FIRMWARE_PWM_POE 80000>;
+	};
+
+	thermal-zones {
+		cpu_thermal: cpu-thermal {
+			trips {
+				threshold: trip-point@0 {
+					temperature = <45000>;
+					hysteresis = <5000>;
+					type = "active";
+				};
+
+				target: trip-point@1 {
+					temperature = <50000>;
+					hysteresis = <2000>;
+					type = "active";
+				};
+
+				cpu_hot: cpu_hot@0 {
+					temperature = <55000>;
+					hysteresis = <2000>;
+					type = "active";
+				};
+			};
+
+			cooling-maps {
+				map0 {
+					trip = <&threshold>;
+					cooling-device = <&fan 0 1>;
+				};
+
+				map1 {
+					trip = <&target>;
+					cooling-device = <&fan 1 2>;
+				};
+
+				map2 {
+					trip = <&cpu_hot>;
+					cooling-device = <&fan 2 3>;
+				};
+			};
+		};
+	};
 };
 
 &ddc0 {
@@ -103,6 +152,11 @@ reset: reset {
 		compatible = "raspberrypi,firmware-reset";
 		#reset-cells = <1>;
 	};
+
+	fwpwm: pwm {
+		compatible = "raspberrypi,firmware-pwm";
+		#pwm-cells = <2>;
+	};
 };
 
 &gpio {
-- 
2.29.1

[PATCH v3 11/11] pwm: Add Raspberry Pi Firmware based PWM bus

[Nicolas Saenz Julienne]

Adds support to control the PWM bus available in official Raspberry Pi
PoE HAT. Only RPi's co-processor has access to it, so commands have to
be sent through RPi's firmware mailbox interface.

Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

---

Changes since v2:
 - Use devm_rpi_firmware_get()
 - Rename driver
 - Small cleanups as per Andy Shevchenko's comments

Changes since v1:
 - Use default pwm bindings and get rid of xlate() function
 - Correct spelling errors
 - Correct apply() function
 - Round values
 - Fix divisions in arm32 mode
 - Small cleanups

 drivers/pwm/Kconfig               |   9 ++
 drivers/pwm/Makefile              |   1 +
 drivers/pwm/pwm-raspberrypi-poe.c | 216 ++++++++++++++++++++++++++++++
 3 files changed, 226 insertions(+)
 create mode 100644 drivers/pwm/pwm-raspberrypi-poe.c

diff --git a/drivers/pwm/Kconfig b/drivers/pwm/Kconfig
index 63be5362fd3a..ab31615f07fc 100644
--- a/drivers/pwm/Kconfig
+++ b/drivers/pwm/Kconfig
@@ -379,6 +379,15 @@ config PWM_PXA
 	  To compile this driver as a module, choose M here: the module
 	  will be called pwm-pxa.
 
+config PWM_RASPBERRYPI_POE
+	tristate "Raspberry Pi Firwmware PoE Hat PWM support"
+	# Make sure not 'y' when RASPBERRYPI_FIRMWARE is 'm'. This can only
+	# happen when COMPILE_TEST=y, hence the added !RASPBERRYPI_FIRMWARE.
+	depends on RASPBERRYPI_FIRMWARE || (COMPILE_TEST && !RASPBERRYPI_FIRMWARE)
+	help
+	  Enable Raspberry Pi firmware controller PWM bus used to control the
+	  official RPI PoE hat
+
 config PWM_RCAR
 	tristate "Renesas R-Car PWM support"
 	depends on ARCH_RENESAS || COMPILE_TEST
diff --git a/drivers/pwm/Makefile b/drivers/pwm/Makefile
index cbdcd55d69ee..7ecbbbcb323a 100644
--- a/drivers/pwm/Makefile
+++ b/drivers/pwm/Makefile
@@ -35,6 +35,7 @@ obj-$(CONFIG_PWM_MXS)		+= pwm-mxs.o
 obj-$(CONFIG_PWM_OMAP_DMTIMER)	+= pwm-omap-dmtimer.o
 obj-$(CONFIG_PWM_PCA9685)	+= pwm-pca9685.o
 obj-$(CONFIG_PWM_PXA)		+= pwm-pxa.o
+obj-$(CONFIG_PWM_RASPBERRYPI_POE)	+= pwm-raspberrypi-poe.o
 obj-$(CONFIG_PWM_RCAR)		+= pwm-rcar.o
 obj-$(CONFIG_PWM_RENESAS_TPU)	+= pwm-renesas-tpu.o
 obj-$(CONFIG_PWM_ROCKCHIP)	+= pwm-rockchip.o
diff --git a/drivers/pwm/pwm-raspberrypi-poe.c b/drivers/pwm/pwm-raspberrypi-poe.c
new file mode 100644
index 000000000000..91cd826a36f3
--- /dev/null
+++ b/drivers/pwm/pwm-raspberrypi-poe.c
@@ -0,0 +1,216 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2020 Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
+ * For more information on Raspberry Pi's PoE hat see:
+ * https://www.raspberrypi.org/products/poe-hat/
+ *
+ * Limitations:
+ *  - No disable bit, so a disabled PWM is simulated by duty_cycle 0
+ *  - Only normal polarity
+ *  - Fixed 12.5 kHz period
+ *
+ * The current period is completed when HW is reconfigured.
+ */
+
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/pwm.h>
+
+#include <soc/bcm2835/raspberrypi-firmware.h>
+#include <dt-bindings/pwm/raspberrypi,firmware-pwm.h>
+
+#define RPI_PWM_MAX_DUTY		255
+#define RPI_PWM_PERIOD_NS		80000 /* 12.5 kHz */
+
+#define RPI_PWM_CUR_DUTY_REG		0x0
+#define RPI_PWM_DEF_DUTY_REG		0x1
+
+struct raspberrypi_pwm {
+	struct rpi_firmware *firmware;
+	struct pwm_chip chip;
+	unsigned int duty_cycle;
+};
+
+struct raspberrypi_pwm_prop {
+	__le32 reg;
+	__le32 val;
+	__le32 ret;
+} __packed;
+
+static inline struct raspberrypi_pwm *to_raspberrypi_pwm(struct pwm_chip *chip)
+{
+	return container_of(chip, struct raspberrypi_pwm, chip);
+}
+
+static int raspberrypi_pwm_set_property(struct rpi_firmware *firmware,
+					u32 reg, u32 val)
+{
+	struct raspberrypi_pwm_prop msg = {
+		.reg = cpu_to_le32(reg),
+		.val = cpu_to_le32(val),
+	};
+	int ret;
+
+	ret = rpi_firmware_property(firmware, RPI_FIRMWARE_SET_POE_HAT_VAL,
+				    &msg, sizeof(msg));
+	if (ret)
+		return ret;
+	if (msg.ret)
+		return -EIO;
+
+	return 0;
+}
+
+static int raspberrypi_pwm_get_property(struct rpi_firmware *firmware,
+					u32 reg, u32 *val)
+{
+	struct raspberrypi_pwm_prop msg = {
+		.reg = reg
+	};
+	int ret;
+
+	ret = rpi_firmware_property(firmware, RPI_FIRMWARE_GET_POE_HAT_VAL,
+				    &msg, sizeof(msg));
+	if (ret)
+		return ret;
+	if (msg.ret)
+		return -EIO;
+
+	*val = le32_to_cpu(msg.val);
+
+	return 0;
+}
+
+static void raspberrypi_pwm_get_state(struct pwm_chip *chip,
+				      struct pwm_device *pwm,
+				      struct pwm_state *state)
+{
+	struct raspberrypi_pwm *rpipwm = to_raspberrypi_pwm(chip);
+
+	state->period = RPI_PWM_PERIOD_NS;
+	state->duty_cycle = DIV_ROUND_CLOSEST(rpipwm->duty_cycle * RPI_PWM_PERIOD_NS,
+					      RPI_PWM_MAX_DUTY);
+	state->enabled = !!(rpipwm->duty_cycle);
+	state->polarity = PWM_POLARITY_NORMAL;
+}
+
+static int raspberrypi_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
+			         const struct pwm_state *state)
+{
+	struct raspberrypi_pwm *rpipwm = to_raspberrypi_pwm(chip);
+	unsigned int duty_cycle;
+	int ret;
+
+        if (state->period < RPI_PWM_PERIOD_NS ||
+            state->polarity != PWM_POLARITY_NORMAL)
+                return -EINVAL;
+
+        if (!state->enabled)
+                duty_cycle = 0;
+        else if (state->duty_cycle < RPI_PWM_PERIOD_NS)
+                duty_cycle = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * RPI_PWM_MAX_DUTY,
+					           RPI_PWM_PERIOD_NS);
+        else
+                duty_cycle = RPI_PWM_MAX_DUTY;
+
+	if (duty_cycle == rpipwm->duty_cycle)
+		return 0;
+
+	ret = raspberrypi_pwm_set_property(rpipwm->firmware, RPI_PWM_CUR_DUTY_REG,
+					   duty_cycle);
+	if (ret) {
+		dev_err(chip->dev, "Failed to set duty cycle: %d\n", ret);
+		return ret;
+	}
+
+	/*
+	 * This sets the default duty cycle after resetting the board, we
+	 * updated it every time to mimic Raspberry Pi's downstream's driver
+	 * behaviour.
+	 */
+	ret = raspberrypi_pwm_set_property(rpipwm->firmware, RPI_PWM_DEF_DUTY_REG,
+					   duty_cycle);
+	if (ret) {
+		dev_err(chip->dev, "Failed to set default duty cycle: %d\n", ret);
+		return ret;
+	}
+
+        rpipwm->duty_cycle = duty_cycle;
+
+	return 0;
+}
+
+static const struct pwm_ops raspberrypi_pwm_ops = {
+	.get_state = raspberrypi_pwm_get_state,
+	.apply = raspberrypi_pwm_apply,
+	.owner = THIS_MODULE,
+};
+
+static int raspberrypi_pwm_probe(struct platform_device *pdev)
+{
+	struct device_node *firmware_node;
+	struct device *dev = &pdev->dev;
+	struct rpi_firmware *firmware;
+	struct raspberrypi_pwm *rpipwm;
+	int ret;
+
+	firmware_node = of_get_parent(dev->of_node);
+	if (!firmware_node) {
+		dev_err(dev, "Missing firmware node\n");
+		return -ENOENT;
+	}
+
+	firmware = devm_rpi_firmware_get(&pdev->dev, firmware_node);
+	of_node_put(firmware_node);
+	if (!firmware)
+		return -EPROBE_DEFER;
+
+	rpipwm = devm_kzalloc(&pdev->dev, sizeof(*rpipwm), GFP_KERNEL);
+	if (!rpipwm)
+		return -ENOMEM;
+
+	rpipwm->firmware = firmware;
+	rpipwm->chip.dev = dev;
+	rpipwm->chip.ops = &raspberrypi_pwm_ops;
+	rpipwm->chip.base = -1;
+	rpipwm->chip.npwm = RASPBERRYPI_FIRMWARE_PWM_NUM;
+
+	platform_set_drvdata(pdev, rpipwm);
+
+	ret = raspberrypi_pwm_get_property(rpipwm->firmware, RPI_PWM_CUR_DUTY_REG,
+					   &rpipwm->duty_cycle);
+	if (ret) {
+		dev_err(dev, "Failed to get duty cycle: %d\n", ret);
+		return ret;
+	}
+
+	return pwmchip_add(&rpipwm->chip);
+}
+
+static int raspberrypi_pwm_remove(struct platform_device *pdev)
+{
+	struct raspberrypi_pwm *rpipwm = platform_get_drvdata(pdev);
+
+	return pwmchip_remove(&rpipwm->chip);
+}
+
+static const struct of_device_id raspberrypi_pwm_of_match[] = {
+	{ .compatible = "raspberrypi,firmware-pwm", },
+	{ }
+};
+MODULE_DEVICE_TABLE(of, raspberrypi_pwm_of_match);
+
+static struct platform_driver raspberrypi_pwm_driver = {
+	.driver = {
+		.name = "raspberrypi-pwm",
+		.of_match_table = raspberrypi_pwm_of_match,
+	},
+	.probe = raspberrypi_pwm_probe,
+	.remove = raspberrypi_pwm_remove,
+};
+module_platform_driver(raspberrypi_pwm_driver);
+
+MODULE_AUTHOR("Nicolas Saenz Julienne <nsaenzjulienne@suse.de>");
+MODULE_DESCRIPTION("Raspberry Pi Firwmare Based PWM Bus Driver");
+MODULE_LICENSE("GPL v2");
-- 
2.29.1

Re: [PATCH v3 09/11] dt-bindings: pwm: Add binding for RPi firmware PWM bus

[Rob Herring]

On Wed, 04 Nov 2020 11:39:35 +0100, Nicolas Saenz Julienne wrote:
> The PWM bus controlling the fan in RPi's official PoE hat can only be
> controlled by the board's co-processor.
> 
> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
> Reviewed-by: Rob Herring <robh@kernel.org>
> 
> ---
> 
> Changes since v1:
>  - Update bindings to use 2 #pwm-cells
> 
>  .../arm/bcm/raspberrypi,bcm2835-firmware.yaml | 20 +++++++++++++++++++
>  .../pwm/raspberrypi,firmware-pwm.h            | 13 ++++++++++++
>  2 files changed, 33 insertions(+)
>  create mode 100644 include/dt-bindings/pwm/raspberrypi,firmware-pwm.h
> 


My bot found errors running 'make dt_binding_check' on your patch:

yamllint warnings/errors:

dtschema/dtc warnings/errors:
/builds/robherring/linux-dt-review/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.example.dt.yaml: firmware: pwm:#pwm-cells:0:0: 2 was expected
	From schema: /builds/robherring/linux-dt-review/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.yaml


See https://patchwork.ozlabs.org/patch/1393804

The base for the patch is generally the last rc1. Any dependencies
should be noted.

If you already ran 'make dt_binding_check' and didn't see the above
error(s), then make sure 'yamllint' is installed and dt-schema is up to
date:

pip3 install dtschema --upgrade

Please check and re-submit.

Re: [PATCH v3 09/11] dt-bindings: pwm: Add binding for RPi firmware PWM bus

[Nicolas Saenz Julienne]

[-- Attachment #1: Type: text/plain, Size: 1108 bytes --]

On Wed, 2020-11-04 at 13:06 -0600, Rob Herring wrote:
> On Wed, 04 Nov 2020 11:39:35 +0100, Nicolas Saenz Julienne wrote:
> > The PWM bus controlling the fan in RPi's official PoE hat can only be
> > controlled by the board's co-processor.
> > 
> > Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
> > Reviewed-by: Rob Herring <robh@kernel.org>
> > 
> > ---
> > 
> > Changes since v1:
> >  - Update bindings to use 2 #pwm-cells
> > 
> >  .../arm/bcm/raspberrypi,bcm2835-firmware.yaml | 20 +++++++++++++++++++
> >  .../pwm/raspberrypi,firmware-pwm.h            | 13 ++++++++++++
> >  2 files changed, 33 insertions(+)
> >  create mode 100644 include/dt-bindings/pwm/raspberrypi,firmware-pwm.h
> > 
> 
> My bot found errors running 'make dt_binding_check' on your patch:
> 
> yamllint warnings/errors:
> 
> dtschema/dtc warnings/errors:
> /builds/robherring/linux-dt-review/Documentation/devicetree/bindings/arm/bcm/raspberrypi,bcm2835-firmware.example.dt.yaml: firmware: pwm:#pwm-cells:0:0: 2 was expected

Yes I forgot to update the example...

Regards,
Nicolas


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 03/11] gpio: raspberrypi-exp: Release firmware handle on unbind

[Bartosz Golaszewski]

On Wed, Nov 4, 2020 at 11:39 AM Nicolas Saenz Julienne
<nsaenzjulienne@suse.de> wrote:
>
> Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
> interface when unbinding the device.
>
> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
>
> ---
>
> Changes since v2:
>  - Use devm_rpi_firmware_get(), instead of remove function
>
>  drivers/gpio/gpio-raspberrypi-exp.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpio/gpio-raspberrypi-exp.c b/drivers/gpio/gpio-raspberrypi-exp.c
> index bb100e0124e6..64a552ecc2ad 100644
> --- a/drivers/gpio/gpio-raspberrypi-exp.c
> +++ b/drivers/gpio/gpio-raspberrypi-exp.c
> @@ -208,7 +208,7 @@ static int rpi_exp_gpio_probe(struct platform_device *pdev)
>                 return -ENOENT;
>         }
>
> -       fw = rpi_firmware_get(fw_node);
> +       fw = devm_rpi_firmware_get(&pdev->dev, fw_node);
>         of_node_put(fw_node);
>         if (!fw)
>                 return -EPROBE_DEFER;
> --
> 2.29.1
>

Acked-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>

Re: [PATCH v3 01/11] firmware: raspberrypi: Introduce devm_rpi_firmware_get()

[Bartosz Golaszewski]

On Wed, Nov 4, 2020 at 11:39 AM Nicolas Saenz Julienne
<nsaenzjulienne@suse.de> wrote:
>
> When unbinding the firmware device we need to make sure it has no
> consumers left. Otherwise we'd leave them with a firmware handle
> pointing at freed memory.
>
> Keep a reference count of all consumers and introduce
> devm_rpi_firmware_get() which will automatically decrease the reference
> count upon unbinding consumer drivers.
>
> Suggested-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
>
> ---
>
> Changes since v2:
>  - Create devm_rpi_firmware_get()
>
>  drivers/firmware/raspberrypi.c             | 46 ++++++++++++++++++++++
>  include/soc/bcm2835/raspberrypi-firmware.h |  8 ++++
>  2 files changed, 54 insertions(+)
>
> diff --git a/drivers/firmware/raspberrypi.c b/drivers/firmware/raspberrypi.c
> index 2371d08bdd17..74bdb3bde9dc 100644
> --- a/drivers/firmware/raspberrypi.c
> +++ b/drivers/firmware/raspberrypi.c
> @@ -11,7 +11,9 @@
>  #include <linux/module.h>
>  #include <linux/of_platform.h>
>  #include <linux/platform_device.h>
> +#include <linux/refcount.h>
>  #include <linux/slab.h>
> +#include <linux/wait.h>
>  #include <soc/bcm2835/raspberrypi-firmware.h>
>
>  #define MBOX_MSG(chan, data28)         (((data28) & ~0xf) | ((chan) & 0xf))
> @@ -27,6 +29,9 @@ struct rpi_firmware {
>         struct mbox_chan *chan; /* The property channel. */
>         struct completion c;
>         u32 enabled;
> +
> +       refcount_t consumers;
> +       wait_queue_head_t wait;
>  };
>
>  static DEFINE_MUTEX(transaction_lock);
> @@ -247,6 +252,8 @@ static int rpi_firmware_probe(struct platform_device *pdev)
>         }
>
>         init_completion(&fw->c);
> +       refcount_set(&fw->consumers, 1);
> +       init_waitqueue_head(&fw->wait);
>
>         platform_set_drvdata(pdev, fw);
>
> @@ -275,11 +282,21 @@ static int rpi_firmware_remove(struct platform_device *pdev)
>         rpi_hwmon = NULL;
>         platform_device_unregister(rpi_clk);
>         rpi_clk = NULL;
> +
> +       wait_event(fw->wait, refcount_dec_if_one(&fw->consumers));
>         mbox_free_channel(fw->chan);
>
>         return 0;
>  }
>
> +static void rpi_firmware_put(void *data)
> +{
> +       struct rpi_firmware *fw = data;
> +
> +       refcount_dec(&fw->consumers);
> +       wake_up(&fw->wait);
> +}
> +
>  /**
>   * rpi_firmware_get - Get pointer to rpi_firmware structure.
>   * @firmware_node:    Pointer to the firmware Device Tree node.
> @@ -297,6 +314,35 @@ struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node)
>  }
>  EXPORT_SYMBOL_GPL(rpi_firmware_get);
>
> +/**
> + * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
> + * @firmware_node:    Pointer to the firmware Device Tree node.
> + *
> + * Returns NULL is the firmware device is not ready.
> + */
> +struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> +                                          struct device_node *firmware_node)
> +{
> +       struct platform_device *pdev = of_find_device_by_node(firmware_node);
> +       struct rpi_firmware *fw;
> +
> +       if (!pdev)
> +               return NULL;
> +
> +       fw = platform_get_drvdata(pdev);
> +       if (!fw)
> +               return NULL;
> +
> +       if (!refcount_inc_not_zero(&fw->consumers))
> +               return NULL;
> +
> +       if (devm_add_action_or_reset(dev, rpi_firmware_put, fw))
> +               return NULL;
> +
> +       return fw;
> +}
> +EXPORT_SYMBOL_GPL(devm_rpi_firmware_get);

Usually I'd expect the devres variant to simply call
rpi_firmware_get() and then schedule a release callback which would
call whatever function is the release counterpart for it currently.
Devres actions are for drivers which want to schedule some more
unusual tasks at driver detach. Any reason for designing it this way?

Bartosz

> +
>  static const struct of_device_id rpi_firmware_of_match[] = {
>         { .compatible = "raspberrypi,bcm2835-firmware", },
>         {},
> diff --git a/include/soc/bcm2835/raspberrypi-firmware.h b/include/soc/bcm2835/raspberrypi-firmware.h
> index cc9cdbc66403..8fe64f53a394 100644
> --- a/include/soc/bcm2835/raspberrypi-firmware.h
> +++ b/include/soc/bcm2835/raspberrypi-firmware.h
> @@ -141,6 +141,8 @@ int rpi_firmware_property(struct rpi_firmware *fw,
>  int rpi_firmware_property_list(struct rpi_firmware *fw,
>                                void *data, size_t tag_size);
>  struct rpi_firmware *rpi_firmware_get(struct device_node *firmware_node);
> +struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> +                                          struct device_node *firmware_node);
>  #else
>  static inline int rpi_firmware_property(struct rpi_firmware *fw, u32 tag,
>                                         void *data, size_t len)
> @@ -158,6 +160,12 @@ static inline struct rpi_firmware *rpi_firmware_get(struct device_node *firmware
>  {
>         return NULL;
>  }
> +
> +static inline struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> +                                       struct device_node *firmware_node)
> +{
> +       return NULL;
> +}
>  #endif
>
>  #endif /* __SOC_RASPBERRY_FIRMWARE_H__ */
> --
> 2.29.1
>

Re: [PATCH v3 01/11] firmware: raspberrypi: Introduce devm_rpi_firmware_get()

[Nicolas Saenz Julienne]

[-- Attachment #1: Type: text/plain, Size: 1764 bytes --]

Hi Bartosz, thanks for the review.

On Thu, 2020-11-05 at 10:13 +0100, Bartosz Golaszewski wrote:
> > +/**
> > + * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
> > + * @firmware_node:    Pointer to the firmware Device Tree node.
> > + *
> > + * Returns NULL is the firmware device is not ready.
> > + */
> > +struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> > +                                          struct device_node *firmware_node)
> > +{
> > +       struct platform_device *pdev = of_find_device_by_node(firmware_node);
> > +       struct rpi_firmware *fw;
> > +
> > +       if (!pdev)
> > +               return NULL;
> > +
> > +       fw = platform_get_drvdata(pdev);
> > +       if (!fw)
> > +               return NULL;
> > +
> > +       if (!refcount_inc_not_zero(&fw->consumers))
> > +               return NULL;
> > +
> > +       if (devm_add_action_or_reset(dev, rpi_firmware_put, fw))
> > +               return NULL;
> > +
> > +       return fw;
> > +}
> > +EXPORT_SYMBOL_GPL(devm_rpi_firmware_get);
> 
> Usually I'd expect the devres variant to simply call
> rpi_firmware_get() and then schedule a release callback which would
> call whatever function is the release counterpart for it currently.
> Devres actions are for drivers which want to schedule some more
> unusual tasks at driver detach. Any reason for designing it this way?

Yes, see patch #8 where I get rid of rpi_firmware_get() altogether after
converting all users to devres. Since there is no use for the vanilla version
of the function anymore, I figured it'd be better to merge everything into
devm_rpi_firmware_get(). That said it's not something I have strong feelings
about.

Regards,
Nicolas


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 01/11] firmware: raspberrypi: Introduce devm_rpi_firmware_get()

[Bartosz Golaszewski]

On Thu, Nov 5, 2020 at 10:28 AM Nicolas Saenz Julienne
<nsaenzjulienne@suse.de> wrote:
>
> Hi Bartosz, thanks for the review.
>
> On Thu, 2020-11-05 at 10:13 +0100, Bartosz Golaszewski wrote:
> > > +/**
> > > + * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
> > > + * @firmware_node:    Pointer to the firmware Device Tree node.
> > > + *
> > > + * Returns NULL is the firmware device is not ready.
> > > + */
> > > +struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> > > +                                          struct device_node *firmware_node)
> > > +{
> > > +       struct platform_device *pdev = of_find_device_by_node(firmware_node);
> > > +       struct rpi_firmware *fw;
> > > +
> > > +       if (!pdev)
> > > +               return NULL;
> > > +
> > > +       fw = platform_get_drvdata(pdev);
> > > +       if (!fw)
> > > +               return NULL;
> > > +
> > > +       if (!refcount_inc_not_zero(&fw->consumers))
> > > +               return NULL;
> > > +
> > > +       if (devm_add_action_or_reset(dev, rpi_firmware_put, fw))
> > > +               return NULL;
> > > +
> > > +       return fw;
> > > +}
> > > +EXPORT_SYMBOL_GPL(devm_rpi_firmware_get);
> >
> > Usually I'd expect the devres variant to simply call
> > rpi_firmware_get() and then schedule a release callback which would
> > call whatever function is the release counterpart for it currently.
> > Devres actions are for drivers which want to schedule some more
> > unusual tasks at driver detach. Any reason for designing it this way?
>
> Yes, see patch #8 where I get rid of rpi_firmware_get() altogether after
> converting all users to devres. Since there is no use for the vanilla version
> of the function anymore, I figured it'd be better to merge everything into
> devm_rpi_firmware_get(). That said it's not something I have strong feelings
> about.
>

I see. So the previous version didn't really have any reference
counting and it leaked the reference returned by
of_find_device_by_node(), got it. Could you just clarify for me the
logic behind the wait_queue in rpi_firmware_remove()? If the firmware
driver gets detached and remove() stops on the wait_queue - it will be
stuck until the last user releases the firmware. I'm not sure this is
correct. I'd prefer to see a kref with a release callback and remove
would simply decrease the kref and return. Each user would do the same
and then after the last user is detached the firmware would be
destroyed.

Don't we really have some centralized firmware subsystem that would handle this?

Bartosz

Re: [PATCH v3 01/11] firmware: raspberrypi: Introduce devm_rpi_firmware_get()

[Nicolas Saenz Julienne]

[-- Attachment #1: Type: text/plain, Size: 3112 bytes --]

Hi Bartosz, thanks for the feedback.

On Thu, 2020-11-05 at 10:42 +0100, Bartosz Golaszewski wrote:
> On Thu, Nov 5, 2020 at 10:28 AM Nicolas Saenz Julienne
> <nsaenzjulienne@suse.de> wrote:
> > Hi Bartosz, thanks for the review.
> > 
> > On Thu, 2020-11-05 at 10:13 +0100, Bartosz Golaszewski wrote:
> > > > +/**
> > > > + * devm_rpi_firmware_get - Get pointer to rpi_firmware structure.
> > > > + * @firmware_node:    Pointer to the firmware Device Tree node.
> > > > + *
> > > > + * Returns NULL is the firmware device is not ready.
> > > > + */
> > > > +struct rpi_firmware *devm_rpi_firmware_get(struct device *dev,
> > > > +                                          struct device_node *firmware_node)
> > > > +{
> > > > +       struct platform_device *pdev = of_find_device_by_node(firmware_node);
> > > > +       struct rpi_firmware *fw;
> > > > +
> > > > +       if (!pdev)
> > > > +               return NULL;
> > > > +
> > > > +       fw = platform_get_drvdata(pdev);
> > > > +       if (!fw)
> > > > +               return NULL;
> > > > +
> > > > +       if (!refcount_inc_not_zero(&fw->consumers))
> > > > +               return NULL;
> > > > +
> > > > +       if (devm_add_action_or_reset(dev, rpi_firmware_put, fw))
> > > > +               return NULL;
> > > > +
> > > > +       return fw;
> > > > +}
> > > > +EXPORT_SYMBOL_GPL(devm_rpi_firmware_get);
> > > 
> > > Usually I'd expect the devres variant to simply call
> > > rpi_firmware_get() and then schedule a release callback which would
> > > call whatever function is the release counterpart for it currently.
> > > Devres actions are for drivers which want to schedule some more
> > > unusual tasks at driver detach. Any reason for designing it this way?
> > 
> > Yes, see patch #8 where I get rid of rpi_firmware_get() altogether after
> > converting all users to devres. Since there is no use for the vanilla version
> > of the function anymore, I figured it'd be better to merge everything into
> > devm_rpi_firmware_get(). That said it's not something I have strong feelings
> > about.
> > 
> 
> I see. So the previous version didn't really have any reference
> counting and it leaked the reference returned by
> of_find_device_by_node(), got it. Could you just clarify for me the
> logic behind the wait_queue in rpi_firmware_remove()? If the firmware
> driver gets detached and remove() stops on the wait_queue - it will be
> stuck until the last user releases the firmware. I'm not sure this is
> correct.

Yes, that's what I meant to implement.

> I'd prefer to see a kref with a release callback and remove
> would simply decrease the kref and return. Each user would do the same
> and then after the last user is detached the firmware would be
> destroyed.

Sounds good to me. I'll update it.

> Don't we really have some centralized firmware subsystem that would handle this?

Sadly no, this is an RPi specific thing, it doesn't live behind a standard like
other firmware based protocols (for ex. SCMI), and evolves as the needs arise.

Regards,
Nicolas


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 07/11] input: raspberrypi-ts: Release firmware handle when not needed

[Dmitry Torokhov]

Hi Nicolas,

On Wed, Nov 04, 2020 at 11:39:33AM +0100, Nicolas Saenz Julienne wrote:
> Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
> interface when unbinding the device.

Unless I am mistaken this driver does not really need the firmware
structure past rpi_ts_probe(), and will be fine if it disappears earlier
than unbind time.

Thanks.

> 
> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
> 
> ---
> 
> Changes since v2:
>  - Use devm_rpi_firmware_get(), instead of remove function
> 
>  drivers/input/touchscreen/raspberrypi-ts.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/input/touchscreen/raspberrypi-ts.c b/drivers/input/touchscreen/raspberrypi-ts.c
> index ef6aaed217cf..efed0efa91d9 100644
> --- a/drivers/input/touchscreen/raspberrypi-ts.c
> +++ b/drivers/input/touchscreen/raspberrypi-ts.c
> @@ -134,7 +134,7 @@ static int rpi_ts_probe(struct platform_device *pdev)
>  		return -ENOENT;
>  	}
>  
> -	fw = rpi_firmware_get(fw_node);
> +	fw = devm_rpi_firmware_get(&pdev->dev, fw_node);
>  	of_node_put(fw_node);
>  	if (!fw)
>  		return -EPROBE_DEFER;
> -- 
> 2.29.1
> 

-- 
Dmitry

Re: [PATCH v3 07/11] input: raspberrypi-ts: Release firmware handle when not needed

[Nicolas Saenz Julienne]

[-- Attachment #1: Type: text/plain, Size: 506 bytes --]

On Wed, 2020-11-11 at 17:45 -0800, Dmitry Torokhov wrote:
> Hi Nicolas,
> 
> On Wed, Nov 04, 2020 at 11:39:33AM +0100, Nicolas Saenz Julienne wrote:
> > Use devm_rpi_firmware_get() so as to make sure we release RPi's firmware
> > interface when unbinding the device.
> 
> Unless I am mistaken this driver does not really need the firmware
> structure past rpi_ts_probe(), and will be fine if it disappears earlier
> than unbind time.

Yes, I missed that. Will update it.

Regards,
Nicolas


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]