Skip to contentSkip to conversation

RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

PATCH New Linus Walleij 4 participants 14 messages Oct 1, 2013 – Oct 28, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

---
 arch/arm/Kconfig            |  3 +-
 arch/arm/mach-gemini/time.c | 97 +++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 91 insertions(+), 9 deletions(-)
arch/arm/Kconfig +2 -1
index 1ad6fb6..1a903e6 100644
@@ -386,8 +386,9 @@ config ARCH_CLPS711X
 config ARCH_GEMINI
 	bool "Cortina Systems Gemini"
 	select ARCH_REQUIRE_GPIOLIB
-	select ARCH_USES_GETTIMEOFFSET
+	select CLKSRC_MMIO
 	select CPU_FA526
+	select GENERIC_CLOCKEVENTS
 	select NEED_MACH_GPIO_H
 	help
 	  Support for the Cortina Systems Gemini family SoCs
arch/arm/mach-gemini/time.c +89 -8

131 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 2 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

I've put a few more people on Cc that I think have access to hardware, according to the openwrt changelog for gemini.

It would be great if someone could test the patch before we apply it.

Arnd

Quoted 7 lines
Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

I've suggested removing the platform before, but got the feedback that it's actively maintained in openwrt. Unfortunately we didn't get much activity to upstream the patches from
http://git.openwrt.org/?p=openwrt.git;a=tree;f=target/linux/gemini/patches
which are apparently required.

Rather than deleting the platform entirely, I'd be more inclined to accept any cleanup patches like yours, testing or not, and let the downstream users deal with fixing it up if it breaks. The ideal case of course would be for someone to actively maintain the platform again and submit any patches that are needed.

Arnd

Quoted 163 lines
---
 arch/arm/Kconfig            |  3 +-
 arch/arm/mach-gemini/time.c | 97 +++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 91 insertions(+), 9 deletions(-)
arch/arm/Kconfig +2 -1
index 1ad6fb6..1a903e6 100644
@@ -386,8 +386,9 @@ config ARCH_CLPS711X
 config ARCH_GEMINI
 	bool "Cortina Systems Gemini"
 	select ARCH_REQUIRE_GPIOLIB
-	select ARCH_USES_GETTIMEOFFSET
+	select CLKSRC_MMIO
 	select CPU_FA526
+	select GENERIC_CLOCKEVENTS
 	select NEED_MACH_GPIO_H
 	help
 	  Support for the Cortina Systems Gemini family SoCs
arch/arm/mach-gemini/time.c +89 -8
index 21dc5a8..0a63c4d 100644
@@ -13,6 +13,8 @@
 #include <mach/hardware.h>
 #include <mach/global_reg.h>
 #include <asm/mach/time.h>
+#include <linux/clockchips.h>
+#include <linux/clocksource.h>
 
 /*
  * Register definitions for the timers
@@ -33,19 +35,89 @@
 #define TIMER_3_CR_CLOCK		(1 << 7)
 #define TIMER_3_CR_INT			(1 << 8)
 
+static unsigned int tick_rate;
+
+static int gemini_timer_set_next_event(unsigned long cycles,
+				       struct clock_event_device *evt)
+{
+	u32 cr;
+
+	cr = readl(TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+
+	/* This may be overdoing it, feel free to test without this */
+	cr &= ~TIMER_2_CR_ENABLE;
+	cr &= ~TIMER_2_CR_INT;
+	writel(cr, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+
+	/* Set next event */
+	writel(cycles, TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER2_BASE)));
+	writel(cycles, TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER2_BASE)));
+	cr |= TIMER_2_CR_ENABLE;
+	cr |= TIMER_2_CR_INT;
+	writel(cr, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+
+	return 0;
+}
+
+static void gemini_timer_set_mode(enum clock_event_mode mode,
+				  struct clock_event_device *evt)
+{
+	u32 period = DIV_ROUND_CLOSEST(tick_rate, HZ);
+	u32 cr;
+
+	switch (mode) {
+        case CLOCK_EVT_MODE_PERIODIC:
+		/* Start the timer */
+		writel(period,
+		       TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER2_BASE)));
+		writel(period,
+		       TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER2_BASE)));
+		cr = readl(TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+		cr |= TIMER_2_CR_ENABLE;
+		cr |= TIMER_2_CR_INT;
+		writel(cr, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+		break;
+	case CLOCK_EVT_MODE_ONESHOT:
+	case CLOCK_EVT_MODE_UNUSED:
+        case CLOCK_EVT_MODE_SHUTDOWN:
+	case CLOCK_EVT_MODE_RESUME:
+		/*
+		 * Disable also for oneshot: the set_next() call will
+		 * arm the timer instead.
+		 */
+		cr = readl(TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+		cr &= ~TIMER_2_CR_ENABLE;
+		cr &= ~TIMER_2_CR_INT;
+		writel(cr, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+		break;
+	default:
+                break;
+	}
+}
+
+/* Use TIMER2 as clock event */
+static struct clock_event_device gemini_clockevent = {
+	.name		= "TIMER2",
+	.rating		= 300, /* Reasonably fast and accurate clock event */
+	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
+	.set_next_event	= gemini_timer_set_next_event,
+	.set_mode	= gemini_timer_set_mode,
+};
+
 /*
  * IRQ handler for the timer
  */
 static irqreturn_t gemini_timer_interrupt(int irq, void *dev_id)
 {
-	timer_tick();
+	struct clock_event_device *evt = &gemini_clockevent;
 
+	evt->event_handler(evt);
 	return IRQ_HANDLED;
 }
 
 static struct irqaction gemini_timer_irq = {
 	.name		= "Gemini Timer Tick",
-	.flags		= IRQF_DISABLED | IRQF_TIMER,
+	.flags		= IRQF_TIMER,
 	.handler	= gemini_timer_interrupt,
 };
 
@@ -54,9 +126,9 @@ static struct irqaction gemini_timer_irq = {
  */
 void __init gemini_timer_init(void)
 {
-	unsigned int tick_rate, reg_v;
+	u32 reg_v;
 
-	reg_v = __raw_readl(IO_ADDRESS(GEMINI_GLOBAL_BASE + GLOBAL_STATUS));
+	reg_v = readl(IO_ADDRESS(GEMINI_GLOBAL_BASE + GLOBAL_STATUS));
 	tick_rate = REG_TO_AHB_SPEED(reg_v) * 1000000;
 
 	printk(KERN_INFO "Bus: %dMHz", tick_rate / 1000000);
@@ -82,8 +154,17 @@ void __init gemini_timer_init(void)
 	 * Make irqs happen for the system timer
 	 */
 	setup_irq(IRQ_TIMER2, &gemini_timer_irq);
-	/* Start the timer */
-	__raw_writel(tick_rate / HZ, TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER2_BASE)));
-	__raw_writel(tick_rate / HZ, TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER2_BASE)));
-	__raw_writel(TIMER_2_CR_ENABLE | TIMER_2_CR_INT, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+
+	/* Enable and use TIMER1 as clock source */
+	writel(0xffffffff, TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER1_BASE)));
+	writel(0xffffffff, TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER1_BASE)));
+	writel(TIMER_1_CR_ENABLE, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+	if (clocksource_mmio_init(TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER1_BASE)),
+				  "TIMER1", tick_rate, 300, 32,
+				  clocksource_mmio_readl_up))
+		pr_err("timer: failed to initialize gemini clock source\n");
+
+	/* Configure and register the clockevent */
+	clockevents_config_and_register(&gemini_clockevent, tick_rate,
+					1, 0xffffffff);
 }
Signature · 1 line
1.8.3.1
#2 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 6 lines · Arnd Bergmann
Quoted 2 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

I've put a few more people on Cc that I think have access to hardware, according to the openwrt changelog for gemini.

Thanks, if the fine OpenWRT folks can suggest a cheapo piece of equipment that is fully supported by OpenWRT and that I can eBay to test my patches on my own I'm all ears.

Quoted 5 lines

I've suggested removing the platform before, but got the feedback that it's actively maintained in openwrt. Unfortunately we didn't get much activity to upstream the patches from
http://git.openwrt.org/?p=openwrt.git;a=tree;f=target/linux/gemini/patches
which are apparently required.

Both with OpenWRT and OpenEmbedded we are seeing this forking instead of working with actively upstreaming and reducing the patch stack. This is not so nice, we need to discuss what can be done to reduce the patch stacks carried by other Open Source projects such as these.

Quoted 5 lines

Rather than deleting the platform entirely, I'd be more inclined to accept any cleanup patches like yours, testing or not, and let the downstream users deal with fixing it up if it breaks. The ideal case of course would be for someone to actively maintain the platform again and submit any patches that are needed.

Yeah :-/

As with any other platform with an active community interested in long-term support it'd be best if someone stepped up to do the necessary conversion to DT and maybe also multiplatform (no idea if that works with the FA526 CPU though) so we can get the maintenance burden down for everone involved.

I suggested project like this for the advanced kernel newbies actually... let's see if it takes hold.

Yours,
Linus Walleij

#3 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

2013/10/1 Linus Walleij <[email protected]>:

Quoted 11 lines · Arnd Bergmann
Quoted 6 lines · Linus Walleij
Quoted 2 lines

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

I've put a few more people on Cc that I think have access to hardware, according to the openwrt changelog for gemini.

Thanks, if the fine OpenWRT folks can suggest a cheapo piece of equipment that is fully supported by OpenWRT and that I can eBay to test my patches on my own I'm all ears.

I can't remember off the top of my head what exact hardware Tomasz Figa handed to me (could probably find it somewhere tomorrow).

Quoted 7 lines
Quoted 6 lines

I've suggested removing the platform before, but got the feedback that it's actively maintained in openwrt. Unfortunately we didn't get much activity to upstream the patches from
http://git.openwrt.org/?p=openwrt.git;a=tree;f=target/linux/gemini/patches
which are apparently required.

Quoted 5 lines

Both with OpenWRT and OpenEmbedded we are seeing this forking instead of working with actively upstreaming and reducing the patch stack. This is not so nice, we need to discuss what can be done to reduce the patch stacks carried by other Open Source projects such as these.

I definitively agree, this has been a bad practice over the years, and even though some of us have tried to resolve that, we are still not quite there yet.

Quoted 9 lines
Quoted 6 lines

Rather than deleting the platform entirely, I'd be more inclined to accept any cleanup patches like yours, testing or not, and let the downstream users deal with fixing it up if it breaks. The ideal case of course would be for someone to actively maintain the platform again and submit any patches that are needed.

Yeah :-/

Quoted 8 lines

As with any other platform with an active community interested in long-term support it'd be best if someone stepped up to do the necessary conversion to DT and maybe also multiplatform (no idea if that works with the FA526 CPU though) so we can get the maintenance burden down for everone involved.

I suggested project like this for the advanced kernel newbies actually... let's see if it takes hold.

One big prerequisite is having actual hardware support, we can probably work out something to give these people access to hardware (either remotely or physically).

Does not anybody actually have a farm of "old" ARM-based platforms besides Olof? That way we could just ship the device to someone kind enough to get some power supply + network + serial + JTAG/system reset?

Signature · 1 line
Florian
#4 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

In reply to Linus Walleij
Quoted 5 lines · Linus Walleij

As with any other platform with an active community interested in long-term support it'd be best if someone stepped up to do the necessary conversion to DT and maybe also multiplatform (no idea if that works with the FA526 CPU though) so we can get the maintenance burden down for everone involved.

Regarding multiplatform, IIRC Jonas Jensen had some issues with ARM9 and FA526 being enabled at the same time (with mach-moxart), which is probably easy enough to fix if necessary. You should also be able to have a combined moxart+gemini kernel once someone looks into multiplatform gemini. The only nonobvious part I can see required for that is NEED_MACH_GPIO_H, and even that should be simpler than the clockevents patch you just sent.

Arnd

#5 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 7 lines · Arnd Bergmann
Quoted 5 lines · Linus Walleij

As with any other platform with an active community interested in long-term support it'd be best if someone stepped up to do the necessary conversion to DT and maybe also multiplatform (no idea if that works with the FA526 CPU though) so we can get the maintenance burden down for everone involved.

Quoted 7 lines

Regarding multiplatform, IIRC Jonas Jensen had some issues with ARM9 and FA526 being enabled at the same time (with mach-moxart), which is probably easy enough to fix if necessary. You should also be able to have a combined moxart+gemini kernel once someone looks into multiplatform gemini. The only nonobvious part I can see required for that is NEED_MACH_GPIO_H, and even that should be simpler than the clockevents patch you just sent.

I sent a patch to delete NEED_MACH_GPIO_H as well and already applied it to the GPIO tree, it was a piece of cake (unless I have now unwittingly broken something).

Yours,
Linus Walleij

#6 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 10 lines · Linus Walleij
Quoted 7 lines · Arnd Bergmann

Regarding multiplatform, IIRC Jonas Jensen had some issues with ARM9 and FA526 being enabled at the same time (with mach-moxart), which is probably easy enough to fix if necessary. You should also be able to have a combined moxart+gemini kernel once someone looks into multiplatform gemini. The only nonobvious part I can see required for that is NEED_MACH_GPIO_H, and even that should be simpler than the clockevents patch you just sent.

Quoted 3 lines

I sent a patch to delete NEED_MACH_GPIO_H as well and already applied it to the GPIO tree, it was a piece of cake (unless I have now unwittingly broken something).

Ah, cool. Upon brief look I could not find any users for the other header files, so I guess we can try to just go ahead and move them all out of include/mach/ to mach-gemini/*.h or to include/debug.

On second look, we need to convert it to MULTI_IRQ_HANDLER first.

Arnd

#7 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

In reply to Linus Walleij
Quoted 10 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

OK nobody is able to test this it seems, but it is a vital refactoring.

Should we just apply this and see what happens?

Or should I just send a deletion patch right away?

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

Yours,
Linus Walleij

#8 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 15 lines · Linus Walleij
Quoted 10 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

OK nobody is able to test this it seems, but it is a vital refactoring.

Quoted 8 lines

Should we just apply this and see what happens?

Or should I just send a deletion patch right away?

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

I suspect the feedback latency will be very long, since they don't track upstream closely (or at least don't on all platforms). So either way, really. We can definitely pick it up and fix up later if needed -- given that you haven't found testers I'm suspecting very few people would likewise be impacted by any kind of breakage introduced.

-Olof

#9 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

In reply to Linus Walleij

Adding openwrt-devel to this thread to get some traction...

Quoted 26 lines · Linus Walleij
Quoted 10 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

OK nobody is able to test this it seems, but it is a vital refactoring.

Should we just apply this and see what happens?

Or should I just send a deletion patch right away?

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

Yours,
Linus Walleij

#10 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Hello Linus,

2013/10/24 Linus Walleij <[email protected]>:

Quoted 1 line

Adding openwrt-devel to this thread to get some traction...

I won't have my gemini device available for the next 6-8 weeks, so just go ahead and merge this, if it breaks, there is still time to figure out how to unbreak later. It cannot possibly be worse since few people have it and actually seem to use it. Does that work for you?

Quoted 33 lines
Quoted 19 lines · Linus Walleij
Quoted 10 lines · Linus Walleij

This converts the gemini machine to use generic clockevents by rewriting the timer driver.

Cc arm at kernel.org
Cc Hans Ulli Kroll <[email protected]>

---
This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

OK nobody is able to test this it seems, but it is a vital refactoring.

Should we just apply this and see what happens?

Or should I just send a deletion patch right away?

Quoted 7 lines

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

Yours,
Linus Walleij

_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel

Signature · 1 line
Florian
#11 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

In reply to Linus Walleij
Quoted 6 lines · Linus Walleij
Quoted 3 lines

This is flagged as Request for Testing (RFT) as I have no Gemini platform. If I do not get help with testing this, the next step will be a patch series to delete the platform.

OK nobody is able to test this it seems, but it is a vital refactoring.

Quoted 8 lines

Should we just apply this and see what happens?

Or should I just send a deletion patch right away?

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

I'd prefer applying it as-is.

Arnd

#12 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 8 lines · Arnd Bergmann
Quoted 4 lines · Linus Walleij

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

What does the ARM SoC people say?

I'd prefer applying it as-is.

OK Olof/Kevin: can either of you strip "RFT" from $SUBJECT and apply this patch to the ARM SoC tree?

Yours,
Linus Walleij

#13 lore raw

PATCH 1/1 RFT: ARM: gemini: convert to GENERIC_CLOCKEVENTS

Quoted 6 lines · Linus Walleij
Quoted 5 lines · Arnd Bergmann
Quoted 3 lines · Linus Walleij

I would lean toward just applying it and see if the OpenWRT people will notice and/or complain.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

What does the ARM SoC people say?

I'd prefer applying it as-is.

OK Olof/Kevin: can either of you strip "RFT" from $SUBJECT and apply this patch to the ARM SoC tree?

Done, applied to next/cleanup.

-Olof