Add a new RTC driver for recent mvebu SoCs
Subsystems (4)
v2 cover Add a new RTC driver for recent mvebu SoCs
Hi,
The Marvell Armada 38x SoCs contains an RTC which differs from the RTC used in the other mvebu SoCs until now. This second version of the patch set adds support for this new IP and enable it in the Device Tree of the Armada 38x SoC.
Thanks,
Gr??gory
Changelog
v1 -> v2:
- Used reg-names to identify the registers in the device tree
- Changed the space into tab in the MAINTAINERS file
- Emphasized that the 5s wait was only needed between two consecutive writes
- Reduced the wait in the set_time function, 100ms were enough
- If no interrupt was available, then disable the alarm related functions
- Fixed the critical sections
- Updated the mvebu_v7_defconfig by enabling the new RTC support
Gregory CLEMENT (5):
rtc: armada38x: Add the device tree binding documentation drivers/rtc/rtc-armada38x: Add a new RTC driver for recent mvebu SoCs
MAINTAINERS: Add the RTC driver for the Armada38x
ARM: mvebu: add Device Tree description of RTC on Armada 38x
ARM: mvebu: enable Armada 38x RTC driver in mvebu_v7_defconfig
.../devicetree/bindings/rtc/armada-380-rtc.txt | 22 ++ MAINTAINERS | 1 + arch/arm/boot/dts/armada-38x.dtsi | 7 + arch/arm/configs/mvebu_v7_defconfig | 1 + drivers/rtc/Kconfig | 10 + drivers/rtc/Makefile | 1 + drivers/rtc/rtc-armada38x.c | 319 +++++++++++++++++++++ 7 files changed, 361 insertions(+) create mode 100644 Documentation/devicetree/bindings/rtc/armada-380-rtc.txt create mode 100644 drivers/rtc/rtc-armada38x.c
Signature · 1 line
1.9.1
PATCH v2 1/5 rtc: armada38x: Add the device tree binding documentation
The Armada 38x SoCs come with a new RTC which differs from the one used in the other mvebu SoCs until now. This patch describes the binding of this RTC.
--- .../devicetree/bindings/rtc/armada-380-rtc.txt | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) create mode 100644 Documentation/devicetree/bindings/rtc/armada-380-rtc.txt
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt new file +22 -0
@@ -0,0 +1,22 @@ +* Real Time Clock of te Armada 38x SoCs + +RTC controller for the Armada 38x SoCs + +Required properties: +- compatible : Should be "marvell,armada-380-rtc" +- reg: physical base address of the controller and length of memory + mapped region, associated to the reg-name "rtc". The other entry is + related to the interrupt control from the SoC, associated to the + reg-name "soc-interrupt". +- reg-names: names of the mapped memory regions listed in regs + property in the same order: "rtc" and "soc-int". +- interrupts: IRQ line for the RTC. + +Example: + +rtc at a3800 { + compatible = "marvell,armada-380-rtc"; + reg = <0xa3800 0x20>, <0x184a0 0x0c>; + reg-names = "rtc", "soc-int"; + interrupts = <GIC_SPI 21 IRQ_TYPE_LEVEL_HIGH>; +};
Signature · 1 line
1.9.1
PATCH v2 1/5 rtc: armada38x: Add the device tree binding documentation
Dear Gregory CLEMENT,
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt line 10
Quoted 6 lines · Gregory CLEMENT
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt new file +6
-0+Required properties: +- compatible : Should be "marvell,armada-380-rtc" +- reg: physical base address of the controller and length of memory + mapped region, associated to the reg-name "rtc". The other entry is + related to the interrupt control from the SoC, associated to the + reg-name "soc-interrupt".
soc-interrupt or...
Quoted 2 lines
+- reg-names: names of the mapped memory regions listed in regs + property in the same order: "rtc" and "soc-int".
soc-int ?
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt line 19
Quoted 3 lines
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt new file +3
-0+rtc at a3800 { + compatible = "marvell,armada-380-rtc"; + reg = <0xa3800 0x20>, <0x184a0 0x0c>;
Any reason to use <0x184A0 0xC> instead of <0x184A8 0x4> ? According to the datasheet, there is only this 184A8 register for RTC stuff.
Thanks,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons Embedded Linux, Kernel and Android engineering http://free-electrons.com
PATCH v2 1/5 rtc: armada38x: Add the device tree binding documentation
Hi Thomas,
Quoted 4 lines · Thomas Petazzoni
Dear Gregory CLEMENT,
On Wed, 14 Jan 2015 18:39:11 +0100, Gregory CLEMENT wrote:
Quoted 8 lines
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt line 10
Quoted 6 lines
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt new file +6
-0+Required properties: +- compatible : Should be "marvell,armada-380-rtc" +- reg: physical base address of the controller and length of memory + mapped region, associated to the reg-name "rtc". The other entry is + related to the interrupt control from the SoC, associated to the + reg-name "soc-interrupt".soc-interrupt or...
I changed all the name from soc-interrupt to soc-int except here. I will fix it.
Quoted 4 lines
Quoted 3 lines
+- reg-names: names of the mapped memory regions listed in regs + property in the same order: "rtc" and "soc-int".
Quoted 8 lines
soc-int ?
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt line 19
Quoted 3 lines
Documentation/devicetree/bindings/rtc/armada-380-rtc.txt new file +3
-0+rtc at a3800 { + compatible = "marvell,armada-380-rtc"; + reg = <0xa3800 0x20>, <0x184a0 0x0c>;Any reason to use <
0x184A00xC> instead of <0x184A80x4> ? According to the datasheet, there is only this 184A8 register for RTC stuff.
Yes but according to the code I saw there were other registers related to the RTC from 0x184A0. Even if we don't use them now I prefer having an accurate mapping from the beginning for avoiding using negative offset as we needed to do in the past.
Thanks,
Gregory
Signature · 4 lines
Gregory Clement, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH v2 1/5 rtc: armada38x: Add the device tree binding documentation
Dear Gregory CLEMENT,
Quoted 6 lines · Gregory CLEMENT
Quoted 2 lines
Any reason to use <
0x184A00xC> instead of <0x184A80x4> ? According to the datasheet, there is only this 184A8 register for RTC stuff.Yes but according to the code I saw there were other registers related to the RTC from
0x184A0. Even if we don't use them now I prefer having an accurate mapping from the beginning for avoiding using negative offset as we needed to do in the past.
So I presume you mean that there are undocumented registers, since I didn't see them in the datasheet. Is this correct?
Thanks,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons Embedded Linux, Kernel and Android engineering http://free-electrons.com
PATCH v2 1/5 rtc: armada38x: Add the device tree binding documentation
Hi Thomas,
Quoted 7 lines · Thomas Petazzoni
Dear Gregory CLEMENT,
Quoted 3 lines · Gregory CLEMENT
Quoted 2 lines
Any reason to use <
0x184A00xC> instead of <0x184A80x4> ? According to the datasheet, there is only this 184A8 register for RTC stuff.
Quoted 6 lines
Quoted 3 lines
Yes but according to the code I saw there were other registers related to the RTC from
0x184A0. Even if we don't use them now I prefer having an accurate mapping from the beginning for avoiding using negative offset as we needed to do in the past.So I presume you mean that there are undocumented registers, since I didn't see them in the datasheet. Is this correct?
Yes indeed there are not documented in the current version of our datasheet.
Thanks,
Gregory
Signature · 4 lines
Gregory Clement, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH v2 2/5 drivers/rtc/rtc-armada38x: Add a new RTC driver for recent mvebu SoCs
In reply to Gregory CLEMENTThe new mvebu SoCs come with a new RTC driver. This patch adds the support for this new IP which is currently found in the Armada 38x SoCs.
This RTC provides two alarms, but only the first one is used in the driver. The RTC also allows using periodic interrupts.
--- drivers/rtc/Kconfig | 10 ++ drivers/rtc/Makefile | 1 + drivers/rtc/rtc-armada38x.c | 319 ++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 330 insertions(+) create mode 100644 drivers/rtc/rtc-armada38x.c
drivers/rtc/Kconfig +10 -0
@@ -1269,6 +1269,16 @@ config RTC_DRV_MV This driver can also be built as a module. If so, the module will be called rtc-mv. +config RTC_DRV_ARMADA38X + tristate "Armada 38x Marvell SoC RTC" + depends on ARCH_MVEBU + help + If you say yes here you will get support for the in-chip RTC + that can be found in the Armada 38x Marvell's SoC device + + This driver can also be built as a module. If so, the module + will be called armada38x-rtc. + config RTC_DRV_PS3 tristate "PS3 RTC" depends on PPC_PS3
drivers/rtc/Makefile +1 -0
@@ -24,6 +24,7 @@ obj-$(CONFIG_RTC_DRV_88PM860X) += rtc-88pm860x.o obj-$(CONFIG_RTC_DRV_88PM80X) += rtc-88pm80x.o obj-$(CONFIG_RTC_DRV_AB3100) += rtc-ab3100.o obj-$(CONFIG_RTC_DRV_AB8500) += rtc-ab8500.o +obj-$(CONFIG_RTC_DRV_ARMADA38X) += rtc-armada38x.o obj-$(CONFIG_RTC_DRV_AS3722) += rtc-as3722.o obj-$(CONFIG_RTC_DRV_AT32AP700X)+= rtc-at32ap700x.o obj-$(CONFIG_RTC_DRV_AT91RM9200)+= rtc-at91rm9200.o
drivers/rtc/rtc-armada38x.c new file +319 -0
319 lines not on this page · raw message
Signature · 1 line
1.9.1
PATCH v2 2/5 drivers/rtc/rtc-armada38x: Add a new RTC driver for recent mvebu SoCs
Hi Gregory,
Quoted 98 lines · Gregory CLEMENT
The new mvebu SoCs come with a new RTC driver. This patch adds the support for this new IP which is currently found in the Armada 38x SoCs.
This RTC provides two alarms, but only the first one is used in the driver. The RTC also allows using periodic interrupts.
Signed-off-by Gregory CLEMENT <[email protected]>--- drivers/rtc/Kconfig | 10 ++ drivers/rtc/Makefile | 1 + drivers/rtc/rtc-armada38x.c | 319 ++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 330 insertions(+) create mode 100644 drivers/rtc/rtc-armada38x.cdrivers/rtc/Kconfig +10
-0@@ -1269,6 +1269,16 @@ config RTC_DRV_MV This driver can also be built as a module. If so, the module will be called rtc-mv. +config RTC_DRV_ARMADA38X + tristate "Armada 38x Marvell SoC RTC" + depends on ARCH_MVEBU + help + If you say yes here you will get support for the in-chip RTC + that can be found in the Armada 38x Marvell's SoC device + + This driver can also be built as a module. If so, the module + will be called armada38x-rtc. + config RTC_DRV_PS3 tristate "PS3 RTC" depends on PPC_PS3drivers/rtc/Makefile +1
-0@@ -24,6 +24,7 @@ obj-$(CONFIG_RTC_DRV_88PM860X) += rtc-88pm860x.o obj-$(CONFIG_RTC_DRV_88PM80X) += rtc-88pm80x.o obj-$(CONFIG_RTC_DRV_AB3100) += rtc-ab3100.o obj-$(CONFIG_RTC_DRV_AB8500) += rtc-ab8500.o +obj-$(CONFIG_RTC_DRV_ARMADA38X) += rtc-armada38x.o obj-$(CONFIG_RTC_DRV_AS3722) += rtc-as3722.o obj-$(CONFIG_RTC_DRV_AT32AP700X)+= rtc-at32ap700x.o obj-$(CONFIG_RTC_DRV_AT91RM9200)+= rtc-at91rm9200.odrivers/rtc/rtc-armada38x.c new file +44
-0@@ -0,0 +1,319 @@ +/* +* RTC driver for the Armada 38x Marvell SoCs +* +* Copyright (C) 2015 Marvell +* +* Gregory Clement <[email protected]> +* +* This program is free software; you can redistribute it and/or +* modify it under the terms of the GNU General Public License as +* published by the Free Software Foundation; either version 2 of the +* License, or (at your option) any later version. +* +*/ + +#include <linux/delay.h> +#include <linux/io.h> +#include <linux/module.h> +#include <linux/of.h> +#include <linux/platform_device.h> +#include <linux/rtc.h> + +#define RTC_STATUS 0x0 +#define RTC_STATUS_ALARM1 BIT(0) +#define RTC_STATUS_ALARM2 BIT(1) +#define RTC_IRQ1_CONF 0x4 +#define RTC_IRQ1_AL_EN BIT(0) +#define RTC_IRQ1_FREQ_EN BIT(1) +#define RTC_IRQ1_FREQ_1HZ BIT(2) +#define RTC_TIME 0xC +#define RTC_ALARM1 0x10 + +#define SOC_RTC_INTERRUPT 0x8 +#define SOC_RTC_ALARM1 BIT(0) +#define SOC_RTC_ALARM2 BIT(1) +#define SOC_RTC_ALARM1_MASK BIT(2) +#define SOC_RTC_ALARM2_MASK BIT(3) + +struct armada38x_rtc { +struct rtc_device *rtc_dev; +void __iomem *regs; +void __iomem *regs_soc; +spinlock_t lock; +int irq; +};
drivers/rtc/rtc-armada38x.c line 54
Quoted 10 lines
drivers/rtc/rtc-armada38x.c new file +10
-0+ +/* +* According to the datasheet, we need to wait for 5us only between +* two consecutive writes +*/ +static void rtc_delayed_write (u32 val,struct armada38x_rtc *rtc,int offset) +{ +writel (val, rtc->regs + offset); +udelay (5 ); +}
The thing I do not get is how you can guarantee that an undelayed writel() in a function will not be followed less than 5?s later by another writel() in another function. For instance:
1) a call to set_alarm() followed by a call to alarm_irq_enable().
2) some writel() or even rtc_udelayed_write() w/ sth asynchronous occuring like your interrupt handler just after that.
I guess it may be possible to make assumptions on the fact that the amount of code before a writel() or rtc_udelayed_write() may prevent such scenario but it's difficult to tell, all the more w/ a spinlock before the writel() in irq_enable().
Considering the description of the bug, why not doing the following:
1) implement rtc_delayed_write() a bit differently:
static inline void rtc_delayed_write(u32 val, struct armada38x_rtc *rtc, int offset)
{
udelay(5);
writel(val, rtc->regs + offset);
} 2) remove all writel() in the driver and use rtc_delayed_write() everywhere.
All writes being under the protection of your spinlock, this will guarantee the 5?s delay in all cases.
Quoted 41 lines
drivers/rtc/rtc-armada38x.c new file +41
-0+static int armada38x_rtc_read_time (struct device *dev,struct rtc_time *tm) +{ +struct armada38x_rtc *rtc =dev_get_drvdata (dev); +unsigned long time, time_check, flags; + +spin_lock_irqsave (&rtc->lock, flags); + + time =readl (rtc->regs +RTC_TIME ); +/* +* WA for failing time set attempts. As stated in HW ERRATA if +* more than one second between two time reads is detected +* then read once again. +*/ + time_check =readl (rtc->regs +RTC_TIME ); +if ((time_check - time) >1 ) + time_check =readl (rtc->regs +RTC_TIME ); + +spin_unlock_irqrestore (&rtc->lock, flags); + +rtc_time_to_tm (time_check, tm); + +return 0 ; +} + +static int armada38x_rtc_set_time (struct device *dev,struct rtc_time *tm) +{ +struct armada38x_rtc *rtc =dev_get_drvdata (dev); +int ret =0 ; +unsigned long time, flags; + + ret =rtc_tm_to_time (tm, &time); + +if (ret) + goto out; +/* +* Setting the RTC time not always succeeds. According to the +* errata we need to first write on the status register and +* then wait for 100ms before writing to the time register to be +* sure that the data will be taken into account. +*/ +spin_lock_irqsave (&rtc->lock, flags);
drivers/rtc/rtc-armada38x.c line 106
Quoted 10 lines
drivers/rtc/rtc-armada38x.c new file +10
-0+ +writel (0 , rtc->regs +RTC_STATUS ); + +spin_unlock_irqrestore (&rtc->lock, flags); + +msleep (100 ); + +spin_lock_irqsave (&rtc->lock, flags); + +writel (time, rtc->regs +RTC_TIME );
probably not critical (it's rtc clock and not system clock) but you still introduce a 100ms shift when setting time here.
As for the two writel() not being done w/o releasing the spinlock, it looks a bit weird but it seems it's ok for other functions manipulating RTC_STATUS reg.
Nonetheless, in the reverse direction, if a writel() occurs somewhere (e.g. in the alarm irq handler) during your msleep(), you may do your final writel() w/o having enforced the expected 100ms delay.
Quoted 2 lines
[SNIP]
PATCH v2 2/5 drivers/rtc/rtc-armada38x: Add a new RTC driver for recent mvebu SoCs
Hi Arnaud,
[...]
Quoted 33 lines
drivers/rtc/rtc-armada38x.c line 54
Quoted 9 lines
drivers/rtc/rtc-armada38x.c new file +9
-0+/* +* According to the datasheet, we need to wait for 5us only between +* two consecutive writes +*/ +static void rtc_delayed_write (u32 val,struct armada38x_rtc *rtc,int offset) +{ +writel (val, rtc->regs + offset); +udelay (5 ); +}The thing I do not get is how you can guarantee that an undelayed
writel()in a function will not be followed less than 5?s later by anotherwritel()in another function. For instance:1) a call to
set_alarm()followed by a call toalarm_irq_enable().
2) somewritel()or evenrtc_udelayed_write()w/ sth asynchronous occuring like your interrupt handler just after that.I guess it may be possible to make assumptions on the fact that the amount of code before a
writel()orrtc_udelayed_write()may prevent such scenario but it's difficult to tell, all the more w/ a spinlock before thewritel()inirq_enable().Considering the description of the bug, why not doing the following:
1) implement
rtc_delayed_write()a bit differently:static inline void rtc_delayed_write(u32 val, struct armada38x_rtc *rtc, int offset) { udelay(5); writel(val, rtc->regs + offset); }
Quoted 5 lines
2) remove all
writel()in the driver and usertc_delayed_write()everywhere.All writes being under the protection of your spinlock, this will guarantee the 5?s delay in all cases.
You're right. I tried avoiding loosing microseconds here and there, but there is too many case where it can fail. So let's us it everywhere.
[...]
Quoted 22 lines
Quoted 22 lines
drivers/rtc/rtc-armada38x.c new file +22
-0+static int armada38x_rtc_set_time (struct device *dev,struct rtc_time *tm) +{ +struct armada38x_rtc *rtc =dev_get_drvdata (dev); +int ret =0 ; +unsigned long time, flags; + + ret =rtc_tm_to_time (tm, &time); + +if (ret) + goto out; +/* +* Setting the RTC time not always succeeds. According to the +* errata we need to first write on the status register and +* then wait for 100ms before writing to the time register to be +* sure that the data will be taken into account. +*/ +spin_lock_irqsave (&rtc->lock, flags); + +writel (0 , rtc->regs +RTC_STATUS ); + +spin_unlock_irqrestore (&rtc->lock, flags); +
Quoted 8 lines
drivers/rtc/rtc-armada38x.c line 106
Quoted 5 lines
drivers/rtc/rtc-armada38x.c new file +5
-0+msleep (100 ); + +spin_lock_irqsave (&rtc->lock, flags); + +writel (time, rtc->regs +RTC_TIME );probably not critical (it's rtc clock and not system clock) but you still introduce a 100ms shift when setting time here.
I am aware of this shift but the granularity is the second, so we can't do better, we have to deal with the limitation of the hardware. :(
Quoted 8 lines
As for the two
writel()not being done w/o releasing the spinlock, it looks a bit weird but it seems it's ok for other functions manipulatingRTC_STATUSreg.Nonetheless, in the reverse direction, if a
writel()occurs somewhere (e.g. in the alarm irq handler) during yourmsleep(), you may do your finalwritel()w/o having enforced the expected 100ms delay.
Actually I don't know if it is problematic if a writel occur in an other register during the 100ms. I still wait for an answer for it.
Thanks,
Gregory
Signature · 4 lines
Gregory Clement, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH v2 3/5 MAINTAINERS: Add the RTC driver for the Armada38x
In reply to Gregory CLEMENTPut it in the mvebu entry.
--- MAINTAINERS | 1 + 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
@@ -1157,6 +1157,7 @@ M: Sebastian Hesselbarth <[email protected]> L: linux-arm-kernel at lists.infradead.org (moderated for non-subscribers) S: Maintained F: arch/arm/mach-mvebu/ +F: drivers/rtc/armada38x-rtc ARM/Marvell Berlin SoC support M: Sebastian Hesselbarth <[email protected]>
Signature · 1 line
1.9.1
PATCH v2 4/5 ARM: mvebu: add Device Tree description of RTC on Armada 38x
In reply to Gregory CLEMENTThe Marvell Armada 38x SoCs contains an RTC which differs from the RTC used in the other mvebu SoCs until now. This commit adds the Device Tree description of this interface at the SoC level.
--- arch/arm/boot/dts/armada-38x.dtsi | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm/boot/dts/armada-38x.dtsi +7 -0
@@ -466,6 +466,13 @@ clocks = <&gateclk 4>; }; + rtc at a3800 { + compatible = "marvell,armada-380-rtc"; + reg = <0xa3800 0x20>, <0x184a0 0x0c>; + reg-names = "rtc", "soc-int"; + interrupts = <GIC_SPI 21 IRQ_TYPE_LEVEL_HIGH>; + }; + sata at a8000 { compatible = "marvell,armada-380-ahci"; reg = <0xa8000 0x2000>;
Signature · 1 line
1.9.1
PATCH v2 5/5 ARM: mvebu: enable Armada 38x RTC driver in mvebu_v7_defconfig
In reply to Gregory CLEMENTNow that the Armada 38x RTC driver has been pushed, let's enable it in mvebu_v7_defconfig.
--- arch/arm/configs/mvebu_v7_defconfig | 1 + 1 file changed, 1 insertion(+)
arch/arm/configs/mvebu_v7_defconfig +1 -0
@@ -112,6 +112,7 @@ CONFIG_LEDS_TRIGGER_HEARTBEAT=y CONFIG_RTC_CLASS=y CONFIG_RTC_DRV_S35390A=y CONFIG_RTC_DRV_MV=y +CONFIG_RTC_DRV_ARMADA38X=y CONFIG_DMADEVICES=y CONFIG_MV_XOR=y # CONFIG_IOMMU_SUPPORT is not set
Signature · 1 line
1.9.1