Skip to contentSkip to conversation

rtc: class: support hctosys from modular RTC drivers

PATCH In review Steve Muckle 2 participants 7 messages Nov 6, 2019 – Mar 23, 2020 1 message not in the archive Real time clock (RTC)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 rtc: class: support hctosys from modular RTC drivers

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

---
 drivers/rtc/Kconfig   | 18 ++++++-----
 drivers/rtc/Makefile  |  1 -
 drivers/rtc/class.c   | 63 +++++++++++++++++++++++++++++++++++++++
 drivers/rtc/hctosys.c | 69 -------------------------------------------
 4 files changed, 73 insertions(+), 78 deletions(-)
 delete mode 100644 drivers/rtc/hctosys.c
drivers/rtc/Kconfig +10 -8
index 1adf9f815652..f663d77deb41 100644
@@ -35,14 +35,16 @@ config RTC_HCTOSYS_DEVICE
 	depends on RTC_HCTOSYS
 	default "rtc0"
 	help
-	  The RTC device that will be used to (re)initialize the system
-	  clock, usually rtc0. Initialization is done when the system
-	  starts up, and when it resumes from a low power state. This
-	  device should record time in UTC, since the kernel won't do
-	  timezone correction.
-
-	  The driver for this RTC device must be loaded before late_initcall
-	  functions run, so it must usually be statically linked.
+	  The RTC device that will be used to (re)initialize the system clock,
+	  usually rtc0. Initialization is done when the driver for the RTC
+	  device registers, and when it resumes from a low power state. This
+	  device should record time in UTC, since the kernel won't do timezone
+	  correction.
+
+	  During startup it is useful to initialize the system clock from the
+	  RTC as early as possible. For this reason the driver for this RTC
+	  device should be statically linked, or alternately, the kernel module
+	  for the RTC device driver loaded immediately.
 
 	  This clock should be battery-backed, so that it reads the correct
 	  time when the system boots from a power-off state. Otherwise, your
drivers/rtc/Makefile +0 -1
index 4ac8f19fb631..33f932d08262 100644
@@ -6,7 +6,6 @@
 ccflags-$(CONFIG_RTC_DEBUG)	:= -DDEBUG
 
 obj-$(CONFIG_RTC_LIB)		+= lib.o
-obj-$(CONFIG_RTC_HCTOSYS)	+= hctosys.o
 obj-$(CONFIG_RTC_SYSTOHC)	+= systohc.o
 obj-$(CONFIG_RTC_CLASS)		+= rtc-core.o
 obj-$(CONFIG_RTC_MC146818_LIB)	+= rtc-mc146818-lib.o
drivers/rtc/class.c +63 -0

75 lines not on this page · raw message

drivers/rtc/hctosys.c deleted +0 -69

69 lines not on this page · raw message

Signature · 1 line
2.24.0.rc1.363.gb1bccd3e3d-goog
#1 lore raw

Hi,

Quoted 7 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Can't you move away from HCTOSYS and do the correct thing in userspace instead of the crap hctosys is doing?

Quoted 209 lines
---
 drivers/rtc/Kconfig   | 18 ++++++-----
 drivers/rtc/Makefile  |  1 -
 drivers/rtc/class.c   | 63 +++++++++++++++++++++++++++++++++++++++
 drivers/rtc/hctosys.c | 69 -------------------------------------------
 4 files changed, 73 insertions(+), 78 deletions(-)
 delete mode 100644 drivers/rtc/hctosys.c
drivers/rtc/Kconfig +10 -8
index 1adf9f815652..f663d77deb41 100644
@@ -35,14 +35,16 @@ config RTC_HCTOSYS_DEVICE
 	depends on RTC_HCTOSYS
 	default "rtc0"
 	help
-	  The RTC device that will be used to (re)initialize the system
-	  clock, usually rtc0. Initialization is done when the system
-	  starts up, and when it resumes from a low power state. This
-	  device should record time in UTC, since the kernel won't do
-	  timezone correction.
-
-	  The driver for this RTC device must be loaded before late_initcall
-	  functions run, so it must usually be statically linked.
+	  The RTC device that will be used to (re)initialize the system clock,
+	  usually rtc0. Initialization is done when the driver for the RTC
+	  device registers, and when it resumes from a low power state. This
+	  device should record time in UTC, since the kernel won't do timezone
+	  correction.
+
+	  During startup it is useful to initialize the system clock from the
+	  RTC as early as possible. For this reason the driver for this RTC
+	  device should be statically linked, or alternately, the kernel module
+	  for the RTC device driver loaded immediately.
 
 	  This clock should be battery-backed, so that it reads the correct
 	  time when the system boots from a power-off state. Otherwise, your
drivers/rtc/Makefile +0 -1
index 4ac8f19fb631..33f932d08262 100644
@@ -6,7 +6,6 @@
 ccflags-$(CONFIG_RTC_DEBUG)	:= -DDEBUG
 
 obj-$(CONFIG_RTC_LIB)		+= lib.o
-obj-$(CONFIG_RTC_HCTOSYS)	+= hctosys.o
 obj-$(CONFIG_RTC_SYSTOHC)	+= systohc.o
 obj-$(CONFIG_RTC_CLASS)		+= rtc-core.o
 obj-$(CONFIG_RTC_MC146818_LIB)	+= rtc-mc146818-lib.o
drivers/rtc/class.c +63 -0
index 9458e6d6686a..b1ed5f3be223 100644
@@ -34,6 +34,64 @@ static void rtc_device_release(struct device *dev)
 #ifdef CONFIG_RTC_HCTOSYS_DEVICE
 /* Result of the last RTC to system clock attempt. */
 int rtc_hctosys_ret = -ENODEV;
+
+/* IMPORTANT: the RTC only stores whole seconds. It is arbitrary
+ * whether it stores the most close value or the value with partial
+ * seconds truncated. However, it is important that we use it to store
+ * the truncated value. This is because otherwise it is necessary,
+ * in an rtc sync function, to read both xtime.tv_sec and
+ * xtime.tv_nsec. On some processors (i.e. ARM), an atomic read
+ * of >32bits is not possible. So storing the most close value would
+ * slow down the sync API. So here we have the truncated value and
+ * the best guess is to add 0.5s.
+ */
+
+static int rtc_hctosys(void)
+{
+	int err = -ENODEV;
+	struct rtc_time tm;
+	struct timespec64 tv64 = {
+		.tv_nsec = NSEC_PER_SEC >> 1,
+	};
+	struct rtc_device *rtc = rtc_class_open(CONFIG_RTC_HCTOSYS_DEVICE);
+
+	if (!rtc) {
+		pr_info("unable to open rtc device (%s)\n",
+			CONFIG_RTC_HCTOSYS_DEVICE);
+		goto err_open;
+	}
+
+	err = rtc_read_time(rtc, &tm);
+	if (err) {
+		dev_err(rtc->dev.parent,
+			"hctosys: unable to read the hardware clock\n");
+		goto err_read;
+	}
+
+	tv64.tv_sec = rtc_tm_to_time64(&tm);
+
+#if BITS_PER_LONG == 32
+	if (tv64.tv_sec > INT_MAX) {
+		err = -ERANGE;
+		goto err_read;
+	}
+#endif
+
+	err = do_settimeofday64(&tv64);
+
+	dev_info(rtc->dev.parent, "setting system clock to %ptR UTC (%lld)\n",
+		 &tm, (long long)tv64.tv_sec);
+
+err_read:
+	rtc_class_close(rtc);
+
+err_open:
+	rtc_hctosys_ret = err;
+
+	return err;
+}
+
+
 #endif
 
 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
@@ -375,6 +433,11 @@ int __rtc_register_device(struct module *owner, struct rtc_device *rtc)
 	dev_info(rtc->dev.parent, "registered as %s\n",
 		 dev_name(&rtc->dev));
 
+#ifdef CONFIG_RTC_HCTOSYS_DEVICE
+	if (!strcmp(dev_name(&rtc->dev), CONFIG_RTC_HCTOSYS_DEVICE))
+		rtc_hctosys();
+#endif
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(__rtc_register_device);
drivers/rtc/hctosys.c deleted +0 -69
deleted file mode 100644
index a74d0d890600..000000000000
@@ -1,69 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-/*
- * RTC subsystem, initialize system time on startup
- *
- * Copyright (C) 2005 Tower Technologies
- * Author: Alessandro Zummo <[email protected]>
- */
-
-#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
-
-#include <linux/rtc.h>
-
-/* IMPORTANT: the RTC only stores whole seconds. It is arbitrary
- * whether it stores the most close value or the value with partial
- * seconds truncated. However, it is important that we use it to store
- * the truncated value. This is because otherwise it is necessary,
- * in an rtc sync function, to read both xtime.tv_sec and
- * xtime.tv_nsec. On some processors (i.e. ARM), an atomic read
- * of >32bits is not possible. So storing the most close value would
- * slow down the sync API. So here we have the truncated value and
- * the best guess is to add 0.5s.
- */
-
-static int __init rtc_hctosys(void)
-{
-	int err = -ENODEV;
-	struct rtc_time tm;
-	struct timespec64 tv64 = {
-		.tv_nsec = NSEC_PER_SEC >> 1,
-	};
-	struct rtc_device *rtc = rtc_class_open(CONFIG_RTC_HCTOSYS_DEVICE);
-
-	if (!rtc) {
-		pr_info("unable to open rtc device (%s)\n",
-			CONFIG_RTC_HCTOSYS_DEVICE);
-		goto err_open;
-	}
-
-	err = rtc_read_time(rtc, &tm);
-	if (err) {
-		dev_err(rtc->dev.parent,
-			"hctosys: unable to read the hardware clock\n");
-		goto err_read;
-	}
-
-	tv64.tv_sec = rtc_tm_to_time64(&tm);
-
-#if BITS_PER_LONG == 32
-	if (tv64.tv_sec > INT_MAX) {
-		err = -ERANGE;
-		goto err_read;
-	}
-#endif
-
-	err = do_settimeofday64(&tv64);
-
-	dev_info(rtc->dev.parent, "setting system clock to %ptR UTC (%lld)\n",
-		 &tm, (long long)tv64.tv_sec);
-
-err_read:
-	rtc_class_close(rtc);
-
-err_open:
-	rtc_hctosys_ret = err;
-
-	return err;
-}
-
-late_initcall(rtc_hctosys);
Signature · 1 line
2.24.0.rc1.363.gb1bccd3e3d-goog
Signature · 3 lines
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#2 lore raw
Quoted 11 lines · Alexandre Belloni
Quoted 7 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Just for my own edification, why is that? Is rtc0 normally useless on PC for some reason?

On the platforms I'm working with I believe it can be assured that rtc0 will be the correct rtc. That doesn't help typical distributions though.

What about a kernel parameter to optionally override the rtc hctosys device at runtime?

Quoted 2 lines

Can't you move away from HCTOSYS and do the correct thing in userspace instead of the crap hctosys is doing?

Yes, I just figured it's a small change, and if hctosys can be made to work might as well use that.

#3 lore raw
Quoted 10 lines · Steve Muckle
Quoted 9 lines · Alexandre Belloni
Quoted 7 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

Quoted 6 lines
Quoted 2 lines

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Just for my own edification, why is that? Is rtc0 normally useless on PC for some reason?

On PC, rtc0 is probably fine which is not the case for other architectures where rtc0 is the SoC RTC and is often not battery backed.

Quoted 6 lines

On the platforms I'm working with I believe it can be assured that rtc0 will be the correct rtc. That doesn't help typical distributions though.

What about a kernel parameter to optionally override the rtc hctosys device at runtime?

What about keeping that in userspace instead which is way easier than messing with kernel parameters?

Quoted 5 lines
Quoted 2 lines

Can't you move away from HCTOSYS and do the correct thing in userspace instead of the crap hctosys is doing?

Yes, I just figured it's a small change, and if hctosys can be made to work might as well use that.

The fact is that hctosys is more related to time keeping than it is to the RTC subsytem. It also does a very poor job setting the system time because adding 0.5s is not the smartest thing to do. The rtc granularity is indeed 1 second but is can be very precisely set.

Signature · 3 lines
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#4 lore raw

(accidentally sent as HTML, resending in text)

Quoted 24 lines · Alexandre Belloni
Quoted 10 lines · Steve Muckle
Quoted 9 lines · Alexandre Belloni
Quoted 7 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

Quoted 6 lines
Quoted 2 lines

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Just for my own edification, why is that? Is rtc0 normally useless on PC for some reason?

On PC, rtc0 is probably fine which is not the case for other architectures where rtc0 is the SoC RTC and is often not battery backed.

Quoted 3 lines

On the platforms I'm working with I believe it can be assured that rtc0 will be the correct rtc. That doesn't help typical distributions though.

Quoted 6 lines
Quoted 3 lines

What about a kernel parameter to optionally override the rtc hctosys device at runtime?

What about keeping that in userspace instead which is way easier than messing with kernel parameters?

This should ideally happen before file systems are mounted so I don't see many alternatives for communicating which RTC should be used. Android uses the kernel command line for userspace parameters as well and that's an option but that defeats part of the value of doing it in userspace IMO. There's also device tree but I'm not sure this belongs there.

Hctosys is also saving and restoring the system time on suspend/resume. It seems more efficient to me to do this (which happens very frequently on an Android device) in the kernel as opposed to in userspace.

If I set the initial system time from the rtc in userspace but continue to rely on the hctosys suspend/resume code, as it stands there will be a window after the rtc driver is loaded but before the system time is set where if suspend is entered, the correct time in the rtc will be lost.

Quoted 10 lines
Quoted 5 lines
Quoted 2 lines

Can't you move away from HCTOSYS and do the correct thing in userspace instead of the crap hctosys is doing?

Yes, I just figured it's a small change, and if hctosys can be made to work might as well use that.

The fact is that hctosys is more related to time keeping than it is to the RTC subsytem. It also does a very poor job setting the system time because adding 0.5s is not the smartest thing to do. The rtc granularity is indeed 1 second but is can be very precisely set.

No argument with that, but millions of devices successfully rely on it today. AFAICT this simple patch doesn't make anything worse. Together with a change to support a kernel parameter for runtime rtc selection, it should allow RTC drivers to be modularized on many systems. Can it be adopted as a stopgap measure?

#5 lore raw
Quoted 58 lines · Steve Muckle
Quoted 24 lines · Alexandre Belloni
Quoted 10 lines · Steve Muckle
Quoted 9 lines · Alexandre Belloni
Quoted 7 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically compiled or not) by checking at the end of RTC device registration whether the time should be synced.

Quoted 6 lines
Quoted 2 lines

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Just for my own edification, why is that? Is rtc0 normally useless on PC for some reason?

On PC, rtc0 is probably fine which is not the case for other architectures where rtc0 is the SoC RTC and is often not battery backed.

Quoted 3 lines

On the platforms I'm working with I believe it can be assured that rtc0 will be the correct rtc. That doesn't help typical distributions though.

Quoted 6 lines
Quoted 3 lines

What about a kernel parameter to optionally override the rtc hctosys device at runtime?

What about keeping that in userspace instead which is way easier than messing with kernel parameters?

This should ideally happen before file systems are mounted so I don't see many alternatives for communicating which RTC should be used. Android uses the kernel command line for userspace parameters as well and that's an option but that defeats part of the value of doing it in userspace IMO. There's also device tree but I'm not sure this belongs there.

Hctosys is also saving and restoring the system time on suspend/resume. It seems more efficient to me to do this (which happens very frequently on an Android device) in the kernel as opposed to in userspace.

If I set the initial system time from the rtc in userspace but continue to rely on the hctosys suspend/resume code, as it stands there will be a window after the rtc driver is loaded but before the system time is set where if suspend is entered, the correct time in the rtc will be lost.

Quoted 10 lines
Quoted 5 lines
Quoted 2 lines

Can't you move away from HCTOSYS and do the correct thing in userspace instead of the crap hctosys is doing?

Yes, I just figured it's a small change, and if hctosys can be made to work might as well use that.

The fact is that hctosys is more related to time keeping than it is to the RTC subsytem. It also does a very poor job setting the system time because adding 0.5s is not the smartest thing to do. The rtc granularity is indeed 1 second but is can be very precisely set.

Quoted 5 lines

No argument with that, but millions of devices successfully rely on it today. AFAICT this simple patch doesn't make anything worse. Together with a change to support a kernel parameter for runtime rtc selection, it should allow RTC drivers to be modularized on many systems. Can it be adopted as a stopgap measure?

Hi Alexandre, ping...

A message that is not in the archive (sent off-list or not yet received): CAL21Ktd3JmRbkwPCCb77knXg4AWi0vWdU147sVaDaoWeEMauDQ@mail.gmail.com

#7 lore raw
Quoted 67 lines · Steve Muckle
Quoted 27 lines · Alexandre Belloni
Quoted 6 lines · Steve Muckle
Quoted 5 lines · Alexandre Belloni
Quoted 4 lines · Steve Muckle

Due to distribution constraints it may not be possible to statically compile the required RTC driver into the kernel.

Expand RTC_HCTOSYS support to cover all RTC devices (statically

compiled

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

or not) by checking at the end of RTC device registration whether the time should be synced.

This does not really help distributions because most of them will still have "rtc0" hardcoded and rtc0 is often the rtc that shouldn't be used.

Just for my own edification, why is that? Is rtc0 normally useless on PC

for

Quoted 2 lines

some reason?

On PC, rtc0 is probably fine which is not the case for other architectures where rtc0 is the SoC RTC and is often not battery backed.

Quoted 1 line

On the platforms I'm working with I believe it can be assured that rtc0

will

Quoted 2 lines

be the correct rtc. That doesn't help typical distributions though.

Quoted 8 lines
Quoted 1 line

What about a kernel parameter to optionally override the rtc hctosys

device

Quoted 2 lines

at runtime?

What about keeping that in userspace instead which is way easier than messing with kernel parameters?

This should ideally happen before file systems are mounted so I don't see many alternatives for communicating which RTC should be used. Android uses the kernel command line for userspace parameters as well and that's an option but that defeats part of the value of doing it in userspace IMO. There's also device tree but I'm not sure this belongs there.

Hctosys is also saving and restoring the system time on suspend/resume. It seems more efficient to me to do this (which happens very frequently on an Android device) in the kernel as opposed to in userspace.

If I set the initial system time from the rtc in userspace but continue to rely on the hctosys suspend/resume code, as it stands there will be a window after the rtc driver is loaded but before the system time is set where if suspend is entered, the correct time in the rtc will be lost.

Quoted 12 lines
Quoted 4 lines

Can't you move away from HCTOSYS and do the correct thing in userspace

Quoted 1 line

instead of the crap hctosys is doing?

Yes, I just figured it's a small change, and if hctosys can be made to

work

Quoted 1 line

might as well use that.

The fact is that hctosys is more related to time keeping than it is to the RTC subsytem. It also does a very poor job setting the system time because adding 0.5s is not the smartest thing to do. The rtc granularity is indeed 1 second but is can be very precisely set.

Quoted 5 lines

No argument with that, but millions of devices successfully rely on it today. AFAICT this simple patch doesn't make anything worse. Together with a change to support a kernel parameter for runtime rtc selection, it should allow RTC drivers to be modularized on many systems. Can it be adopted as a stopgap measure?

I've applied this patch for 5.7 after removing the unnecessary reformatting of the kconfig help.