Skip to contentSkip to conversation

Redo PolarFire SoC's mailbox/clock devicestrees and related code

PATCH In review Conor Dooley 2 participants 9 messages Nov 10, 2025 – Nov 11, 2025

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v7 cover Redo PolarFire SoC's mailbox/clock devicestrees and related code

From: Conor Dooley <[email protected]>

Hey folks,

Here's another version with the reset stuff moved completely to regmap as you requested Philipp. I think it looks better this way.

Claudiu, I think you're going to "have" to take the reset bits with an ack, because they now depend on changes to the clock driver (and contain some further changes to the regmap config and aux bus portion).

Cheers,
Conor.

v7:
- move reset driver entirely to regmap
- use clear_bits/set_bits instead of update_bits in reset
- drop clock patches that Claudiu applied

v6:
- make reset depend on MFD_SYSCON
- return regmap_update_bits() result directly instead of an additional return 0 in reset
- use regmap_update_bits() instead of regmap_read() -> regmap_write() RMW sequences in clock
- drop clock driver specific lock, since regmap has internal locking, from !msspll clocks
- implement determine_rate instead of round_rate
- drop patch 1 & 2 (soc bits) since I applied them as a pre-req for pinctrl work

v5:
- drop mfd patch applied by Lee
- remove attempt at common regmap divider/gate clocks, and replace it with a return to how the code used to look, before it started using the non-regmap versions of the common divider/gate, with the readl()/writel()s replaced by their regmap equivalents.

v4:
- unify both regmap clk implementations under one option
- change map_offset to a u32, after Gabriel pointed out that u8 was too restrictive.
- remove locking from regmap portion of reset driver, relying on inherent regmap lock

v3 changes:
- drop simple-mfd (for now) from syscon node

v5/6 cover letter:

In v5 the only real change is that I removed the attempt at a common implementation of regmap-based divider/gate clocks. The series hasn't managed to receive feedback on my approach in 2025, despite sending several revisions and bumps, and it is blocking support for both new drivers (gpio interrupt support, pinctrl and hwmon off the top of my head) and a new platform so I have decided to strip out the attempt at making something common in exchange for something that can be merged through the clk-microchip tree without relying on feedback from the clock maintainers.

Currently the driver uses the common gate and divider clocks, but the driver used to use its own custom clock types. Reprising this version of the code allows me to use regmap accessors in the driver without any wider impact, or attempting to create something that works for any other user. It has the advantage that it has already been tested in that prior for, and all that is done to the clock implementations is replacing readl()s and writel()s with their regmap equivalents.

v2 cover letter:

Here's something that I've been mulling over for a while, since I started to understand how devicetree stuff was "meant" to be done. There'd been little reason to actually press forward with it, because it is fairly disruptive. I've finally opted to do it, because a user has come along with a hwmon driver that needs to access the same register region as the mailbox and the author is not keen on using the aux bus, and because I do not want the new pic64gx SoC that's based on PolarFire SoC to use bindings etc that I know to be incorrect.

Given backwards compatibility needs to be maintained, this patch series isn't the prettiest thing I have ever written. The reset driver needs to retain support for the auxiliary bus, which looks a bit mess, but not much can be done there. The mailbox and clock drivers both have to have an "old probe" function to handle the old layout. Thankfully in the clock driver, regmap support can be used to identically handle both old and new devicetree formats - but using a regmap in the mailbox driver was only really possible for the new format, so the code there is unfortunately a bit of an if/else mess that I'm both not proud of, nor really sure is worth "improving".

The series should be pretty splitable per subsystem, only the dts change has some sort of dependency, but I'll not be applying that till everything else is in Linus' tree, so that's not a big deal.

I don't really want this stuff in stable, hence a lack of cc: stable anywhere here, since what's currently in the tree works fine for the currently supported hardware.

AFAIK, the only other project affected here is U-Boot, which I have already modified to support the new format.

I previously submitted this as an RFC, only to Lee and the dt list, in order to get some feedback on the syscon/mfd bindings:
https://lore.kernel.org/all/20240815-shindig-bunny-fd42792d638a@spud/
I'm not really going to bother with a proper changelog, since that was submitted with lots of WIP code to get answers to some questions. The main change was "removing" some of the child nodes of the syscons.

And as a "real" series where discussion lead to me dropping use of the amlogic clk-regmap support:
https://lore.kernel.org/linux-clk/20241002-private-unequal-33cfa6101338@spud/
As a result of that, I've implemented what I think Stephen was asking for - but I'm not at all sure that it is..

CC Conor Dooley <[email protected]>
CC Daire McNamara <[email protected]>
CC Michael Turquette <[email protected]>
CC Stephen Boyd <[email protected]>
CC Rob Herring <[email protected]>
CC Krzysztof Kozlowski <[email protected]>
CC Philipp Zabel <[email protected]>

Conor Dooley (5):
reset: mpfs: add non-auxiliary bus probing
riscv: dts: microchip: fix mailbox description
riscv: dts: microchip: convert clock and reset to use syscon
MAINTAINERS: add new soc drivers to Microchip RISC-V entry
MAINTAINERS: rename Microchip RISC-V entry

 MAINTAINERS                             |  4 +-
 arch/riscv/boot/dts/microchip/mpfs.dtsi | 34 ++++++---
 drivers/clk/microchip/clk-mpfs.c        |  4 +-
 drivers/reset/Kconfig                   |  1 +
 drivers/reset/reset-mpfs.c              | 92 +++++++++++++++----------
 include/soc/microchip/mpfs.h            |  3 +-
 6 files changed, 88 insertions(+), 50 deletions(-)
Signature · 1 line
2.51.0
#1 lore raw

PATCH v7 1/5 reset: mpfs: add non-auxiliary bus probing

From: Conor Dooley <[email protected]>

While the auxiliary bus was a nice bandaid, and meant that re-writing the representation of the clock regions in devicetree was not required, it has run its course. The "mss_top_sysreg" region that contains the clock and reset regions, also contains pinctrl and an interrupt controller, so the time has come rewrite the devicetree and probe the reset controller from an mfd devicetree node, rather than implement those drivers using the auxiliary bus. Wanting to avoid propagating this naive/incorrect description of the hardware to the new pic64gx SoC is a major motivating factor here.

---
v7:
- move entirely to regmap
- use clear/set instead of update

v6:
- depend on MFD_SYSCON
- return regmap_update_bits() result directly instead of an additional return 0

v4:
- Only use driver specific lock for non-regmap writes

v2:
- Implement the request to use regmap_update_bits(). I found that I then hated the read/write helpers since they were just bloat, so I ripped them out. I replaced the regular spin_lock_irqsave() stuff with a guard(spinlock_irqsave), since that's a simpler way of handling the two different paths through such a trivial pair of functions.

---
 drivers/clk/microchip/clk-mpfs.c |  4 +-
 drivers/reset/Kconfig            |  1 +
 drivers/reset/reset-mpfs.c       | 92 +++++++++++++++++++-------------
 include/soc/microchip/mpfs.h     |  3 +-
 4 files changed, 61 insertions(+), 39 deletions(-)
drivers/clk/microchip/clk-mpfs.c +2 -2
index 484893e68b67..ee58304913ef 100644
@@ -38,7 +38,7 @@ static const struct regmap_config mpfs_clk_regmap_config = {
 	.reg_stride = 4,
 	.val_bits = 32,
 	.val_format_endian = REGMAP_ENDIAN_LITTLE,
-	.max_register = REG_SUBBLK_CLOCK_CR,
+	.max_register = REG_SUBBLK_RESET_CR,
 };
 
 /*
@@ -502,7 +502,7 @@ static inline int mpfs_clk_old_format_probe(struct mpfs_clock_data *clk_data,
 	if (IS_ERR(clk_data->regmap))
 		return PTR_ERR(clk_data->regmap);
 
-	return mpfs_reset_controller_register(dev, clk_data->base + REG_SUBBLK_RESET_CR);
+	return mpfs_reset_controller_register(dev, clk_data->regmap);
 }
 
 static int mpfs_clk_probe(struct platform_device *pdev)
drivers/reset/Kconfig +1 -0
index 78b7078478d4..0ec4b7cd08d6 100644
@@ -200,6 +200,7 @@ config RESET_PISTACHIO
 config RESET_POLARFIRE_SOC
 	bool "Microchip PolarFire SoC (MPFS) Reset Driver"
 	depends on MCHP_CLK_MPFS
+	depends on MFD_SYSCON
 	select AUXILIARY_BUS
 	default MCHP_CLK_MPFS
 	help
drivers/reset/reset-mpfs.c +56 -36

168 lines not on this page · raw message

include/soc/microchip/mpfs.h +2 -1
index 0bd67e10b704..ec04c98a8b63 100644
@@ -14,6 +14,7 @@
 
 #include <linux/types.h>
 #include <linux/of_device.h>
+#include <linux/regmap.h>
 
 struct mpfs_sys_controller;
 
@@ -44,7 +45,7 @@ struct mtd_info *mpfs_sys_controller_get_flash(struct mpfs_sys_controller *mpfs_
 
 #if IS_ENABLED(CONFIG_MCHP_CLK_MPFS)
 #if IS_ENABLED(CONFIG_RESET_POLARFIRE_SOC)
-int mpfs_reset_controller_register(struct device *clk_dev, void __iomem *base);
+int mpfs_reset_controller_register(struct device *clk_dev, struct regmap *map);
 #else
 static inline int mpfs_reset_controller_register(struct device *clk_dev, void __iomem *base) { return 0; }
 #endif /* if IS_ENABLED(CONFIG_RESET_POLARFIRE_SOC) */
Signature · 1 line
2.51.0
maintainer Reviewed-by Acked-by on v7 1/5
#2 lore raw

Re: [PATCH v7 1/5] reset: mpfs: add non-auxiliary bus probing

Quoted 78 lines · Conor Dooley

From: Conor Dooley <[email protected]>

While the auxiliary bus was a nice bandaid, and meant that re-writing the representation of the clock regions in devicetree was not required, it has run its course. The "mss_top_sysreg" region that contains the clock and reset regions, also contains pinctrl and an interrupt controller, so the time has come rewrite the devicetree and probe the reset controller from an mfd devicetree node, rather than implement those drivers using the auxiliary bus. Wanting to avoid propagating this naive/incorrect description of the hardware to the new pic64gx SoC is a major motivating factor here.

---
v7:
- move entirely to regmap
- use clear/set instead of update

v6:
- depend on MFD_SYSCON
- return regmap_update_bits() result directly instead of an additional return 0

v4:
- Only use driver specific lock for non-regmap writes

v2:
- Implement the request to use regmap_update_bits(). I found that I then hated the read/write helpers since they were just bloat, so I ripped them out. I replaced the regular spin_lock_irqsave() stuff with a guard(spinlock_irqsave), since that's a simpler way of handling the two different paths through such a trivial pair of functions.

---
 drivers/clk/microchip/clk-mpfs.c |  4 +-
 drivers/reset/Kconfig            |  1 +
 drivers/reset/reset-mpfs.c       | 92 +++++++++++++++++++-------------
 include/soc/microchip/mpfs.h     |  3 +-
 4 files changed, 61 insertions(+), 39 deletions(-)
drivers/clk/microchip/clk-mpfs.c +2 -2
index 484893e68b67..ee58304913ef 100644
@@ -38,7 +38,7 @@ static const struct regmap_config mpfs_clk_regmap_config = {
 	.reg_stride = 4,
 	.val_bits = 32,
 	.val_format_endian = REGMAP_ENDIAN_LITTLE,
-	.max_register = REG_SUBBLK_CLOCK_CR,
+	.max_register = REG_SUBBLK_RESET_CR,
 };
 
 /*
@@ -502,7 +502,7 @@ static inline int mpfs_clk_old_format_probe(struct mpfs_clock_data *clk_data,
 	if (IS_ERR(clk_data->regmap))
 		return PTR_ERR(clk_data->regmap);
 
-	return mpfs_reset_controller_register(dev, clk_data->base + REG_SUBBLK_RESET_CR);
+	return mpfs_reset_controller_register(dev, clk_data->regmap);
 }
 
 static int mpfs_clk_probe(struct platform_device *pdev)
drivers/reset/Kconfig +1 -0
index 78b7078478d4..0ec4b7cd08d6 100644
@@ -200,6 +200,7 @@ config RESET_PISTACHIO
 config RESET_POLARFIRE_SOC
 	bool "Microchip PolarFire SoC (MPFS) Reset Driver"
 	depends on MCHP_CLK_MPFS
+	depends on MFD_SYSCON
 	select AUXILIARY_BUS
 	default MCHP_CLK_MPFS
 	help
drivers/reset/reset-mpfs.c +0 -0
index f6fa10e03ea8..d00212450990 100644

drivers/reset/reset-mpfs.c line 10

Quoted 4 lines
drivers/reset/reset-mpfs.c +1 -0
@@ -7,13 +7,16 @@
  *
  */
 #include <linux/auxiliary_bus.h>
+#include <linux/cleanup.h>

Not used anymore.

[...]

Quoted 21 lines
drivers/reset/reset-mpfs.c +3 -3
@@ -176,12 +196,12 @@ static const struct auxiliary_device_id mpfs_reset_ids[] = {
 };
 MODULE_DEVICE_TABLE(auxiliary, mpfs_reset_ids);
 
-static struct auxiliary_driver mpfs_reset_driver = {
-	.probe		= mpfs_reset_probe,
+static struct auxiliary_driver mpfs_reset_aux_driver = {
+	.probe		= mpfs_reset_adev_probe,
 	.id_table	= mpfs_reset_ids,
 };
 
-module_auxiliary_driver(mpfs_reset_driver);
+module_auxiliary_driver(mpfs_reset_aux_driver);
 
 MODULE_DESCRIPTION("Microchip PolarFire SoC Reset Driver");
 MODULE_AUTHOR("Conor Dooley <[email protected]>");
include/soc/microchip/mpfs.h +0 -0
index 0bd67e10b704..ec04c98a8b63 100644

include/soc/microchip/mpfs.h line 17

Quoted 4 lines
include/soc/microchip/mpfs.h +1 -0
@@ -14,6 +14,7 @@
 
 #include <linux/types.h>
 #include <linux/of_device.h>
+#include <linux/regmap.h>

You don't have to #include <linux/regmap.h> here, a forward declaration

struct regmap;

would suffice.

include/soc/microchip/mpfs.h struct mtd_info *mpfs_sys_controller_get_flash(struct mpfs_sys_controller *mpfs_ · line 51

Quoted 11 lines
include/soc/microchip/mpfs.h +1 -1
 struct mpfs_sys_controller;
 
@@ -44,7 +45,7 @@ struct mtd_info *mpfs_sys_controller_get_flash(struct mpfs_sys_controller *mpfs_
 
 #if IS_ENABLED(CONFIG_MCHP_CLK_MPFS)
 #if IS_ENABLED(CONFIG_RESET_POLARFIRE_SOC)
-int mpfs_reset_controller_register(struct device *clk_dev, void __iomem *base);
+int mpfs_reset_controller_register(struct device *clk_dev, struct regmap *map);
 #else
 static inline int mpfs_reset_controller_register(struct device *clk_dev, void __iomem *base) { return 0; }
 #endif /* if IS_ENABLED(CONFIG_RESET_POLARFIRE_SOC) */

With the superfluous cleanup include fixed.

and

to be merged with the reset of the series.

regards
Philipp

#3 lore raw

Re: [PATCH v7 1/5] reset: mpfs: add non-auxiliary bus probing

Quoted 11 lines · Philipp Zabel

On Mo, 2025-11-10 at 11:23 +0000, Conor Dooley wrote:

With the superfluous cleanup include fixed.

and

to be merged with the reset of the series.

Cool, I have dropped the include and pushed the patch to the clk-microchip branch:
https://git.kernel.org/at91/c/4a75fcd2000e1af452343aac6e34387f8e794f37

I opted to leave the include of regmap.h in mpfs.h unchanged.

Cheers,
Conor.

#4 lore raw

Re: [PATCH v7 1/5] reset: mpfs: add non-auxiliary bus probing

Quoted 9 lines · Conor Dooley
Quoted 8 lines · Philipp Zabel

On Mo, 2025-11-10 at 11:23 +0000, Conor Dooley wrote:

With the superfluous cleanup include fixed.

and

Quoted 7 lines
Quoted 3 lines

to be merged with the reset of the series.

Cool, I have dropped the include and pushed the patch to the clk-microchip branch:
https://git.kernel.org/at91/c/4a75fcd2000e1af452343aac6e34387f8e794f37

And of course I made a mistake hand assembling this, that's the hash from before I removed the header. The actual commit is here:
https://git.kernel.org/pub/scm/linux/kernel/git/at91/linux.git/commit/?h=clk-microchip&id=781f60e45bdfe351aad692ac0fa89e36f8bf4a36

#5 lore raw

PATCH v7 2/5 riscv: dts: microchip: fix mailbox description

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

When the binding for the mailbox on PolarFire SoC was originally written, and later modified, mistakes were made - and the precise nature of the later modification should have been a giveaway, but alas I was naive at the time.

A more correct modelling of the hardware is to use two syscons and have a single reg entry for the mailbox, containing the mailbox region. The two syscons contain the general control/status registers for the mailbox and the interrupt related registers respectively. The reason for two syscons is that the same mailbox is present on the non-SoC version of the FPGA, which has no interrupt controller, and the shared part of the rtl was unchanged between devices.

---
 arch/riscv/boot/dts/microchip/mpfs.dtsi | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
arch/riscv/boot/dts/microchip/mpfs.dtsi +12 -3
index 9883ca3554c5..f9d6bf08e717 100644
@@ -259,6 +259,11 @@ clkcfg: clkcfg@20002000 {
 			#reset-cells = <1>;
 		};
 
+		sysreg_scb: syscon@20003000 {
+			compatible = "microchip,mpfs-sysreg-scb", "syscon";
+			reg = <0x0 0x20003000 0x0 0x1000>;
+		};
+
 		ccc_se: clock-controller@38010000 {
 			compatible = "microchip,mpfs-ccc";
 			reg = <0x0 0x38010000 0x0 0x1000>, <0x0 0x38020000 0x0 0x1000>,
@@ -521,10 +526,14 @@ usb: usb@20201000 {
 			status = "disabled";
 		};
 
-		mbox: mailbox@37020000 {
+		control_scb: syscon@37020000 {
+			compatible = "microchip,mpfs-control-scb", "syscon";
+			reg = <0x0 0x37020000 0x0 0x100>;
+		};
+
+		mbox: mailbox@37020800 {
 			compatible = "microchip,mpfs-mailbox";
-			reg = <0x0 0x37020000 0x0 0x58>, <0x0 0x2000318C 0x0 0x40>,
-			      <0x0 0x37020800 0x0 0x100>;
+			reg = <0x0 0x37020800 0x0 0x1000>;
 			interrupt-parent = <&plic>;
 			interrupts = <96>;
 			#mbox-cells = <1>;
Signature · 1 line
2.51.0
#6 lore raw

PATCH v7 3/5 riscv: dts: microchip: convert clock and reset to use syscon

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

The "subblock" clocks and reset registers on PolarFire SoC are located in the mss-top-sysreg region, alongside pinctrl and interrupt control functionality. Re-write the devicetree to describe the sys explicitly, as its own node, rather than as a region of the clock node. Correspondingly, the phandles to the reset controller must be updated to the new provider. The drivers will continue to support the old way of doing things.

---
 arch/riscv/boot/dts/microchip/mpfs.dtsi | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)
arch/riscv/boot/dts/microchip/mpfs.dtsi +12 -7
index f9d6bf08e717..5c2963e269b8 100644
@@ -251,11 +251,9 @@ pdma: dma-controller@3000000 {
 			#dma-cells = <1>;
 		};
 
-		clkcfg: clkcfg@20002000 {
-			compatible = "microchip,mpfs-clkcfg";
-			reg = <0x0 0x20002000 0x0 0x1000>, <0x0 0x3E001000 0x0 0x1000>;
-			clocks = <&refclk>;
-			#clock-cells = <1>;
+		mss_top_sysreg: syscon@20002000 {
+			compatible = "microchip,mpfs-mss-top-sysreg", "syscon", "simple-mfd";
+			reg = <0x0 0x20002000 0x0 0x1000>;
 			#reset-cells = <1>;
 		};
 
@@ -452,7 +450,7 @@ mac0: ethernet@20110000 {
 			local-mac-address = [00 00 00 00 00 00];
 			clocks = <&clkcfg CLK_MAC0>, <&clkcfg CLK_AHB>;
 			clock-names = "pclk", "hclk";
-			resets = <&clkcfg CLK_MAC0>;
+			resets = <&mss_top_sysreg CLK_MAC0>;
 			status = "disabled";
 		};
 
@@ -466,7 +464,7 @@ mac1: ethernet@20112000 {
 			local-mac-address = [00 00 00 00 00 00];
 			clocks = <&clkcfg CLK_MAC1>, <&clkcfg CLK_AHB>;
 			clock-names = "pclk", "hclk";
-			resets = <&clkcfg CLK_MAC1>;
+			resets = <&mss_top_sysreg CLK_MAC1>;
 			status = "disabled";
 		};
 
@@ -550,5 +548,12 @@ syscontroller_qspi: spi@37020100 {
 			clocks = <&scbclk>;
 			status = "disabled";
 		};
+
+		clkcfg: clkcfg@3e001000 {
+			compatible = "microchip,mpfs-clkcfg";
+			reg = <0x0 0x3e001000 0x0 0x1000>;
+			clocks = <&refclk>;
+			#clock-cells = <1>;
+		};
 	};
 };
Signature · 1 line
2.51.0
#7 lore raw

PATCH v7 4/5 MAINTAINERS: add new soc drivers to Microchip RISC-V entry

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

Add the two new syscon drivers to the RISC-V entry for Microchip platforms.

---
 MAINTAINERS | 2 ++
 1 file changed, 2 insertions(+)
MAINTAINERS +2 -0
index 46126ce2f968..a28740a7d87a 100644
@@ -22105,6 +22105,8 @@ F:	drivers/pci/controller/plda/pcie-microchip-host.c
 F:	drivers/pwm/pwm-microchip-core.c
 F:	drivers/reset/reset-mpfs.c
 F:	drivers/rtc/rtc-mpfs.c
+F:	drivers/soc/microchip/mpfs-control-scb.c
+F:	drivers/soc/microchip/mpfs-mss-top-sysreg.c
 F:	drivers/soc/microchip/mpfs-sys-controller.c
 F:	drivers/spi/spi-microchip-core-qspi.c
 F:	drivers/spi/spi-microchip-core.c
Signature · 1 line
2.51.0
#8 lore raw

PATCH v7 5/5 MAINTAINERS: rename Microchip RISC-V entry

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

There's now non-FPGA RISC-V SoCs from Microchip, so rename the entry to reflect that.

---
 MAINTAINERS | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
MAINTAINERS +1 -1
index a28740a7d87a..24efae3df425 100644
@@ -22079,7 +22079,7 @@ T:	git git://git.kernel.org/pub/scm/linux/kernel/git/iommu/linux.git
 F:	Documentation/devicetree/bindings/iommu/riscv,iommu.yaml
 F:	drivers/iommu/riscv/
 
-RISC-V MICROCHIP FPGA SUPPORT
+RISC-V MICROCHIP SUPPORT
 M:	Conor Dooley <[email protected]>
 M:	Daire McNamara <[email protected]>
 L:	[email protected]
Signature · 1 line
2.51.0