JAZZ: disable PIT; cleanup R4030 clockevent

PATCH Applied Thomas Bogendoerfer 2 participants 8 messages Nov 1, 2007 – Nov 4, 2007 MIPS

lore

Share & tools

lore

Thomas Bogendoerfer

author maintainer

2007-11-01 12:52 UTC #0 lore raw

PATCH 1/1 JAZZ: disable PIT; cleanup R4030 clockevent

PIT doesn't work, disable it completly
make r4030 clockevent code look like other mips clockevent code

Signed-off-by Thomas Bogendoerfer <[email protected]>

---

arch/mips/Kconfig +0 -1

index 97da953..c4f7e60 100644
@@ -119,7 +119,6 @@ config MACH_JAZZ
 	select CEVT_R4K
 	select GENERIC_ISA_DMA
 	select IRQ_CPU
-	select I8253
 	select I8259
 	select ISA
 	select PCSPEAKER

arch/mips/jazz/irq.c +12 -9

index 1dac9e1..a629719 100644
@@ -118,16 +118,16 @@ static void r4030_set_mode(enum clock_event_mode mode,
 struct clock_event_device r4030_clockevent = {
 	.name		= "r4030",
 	.features	= CLOCK_EVT_FEAT_PERIODIC,
-	.rating		= 100,
+	.rating		= 300,
 	.irq		= JAZZ_TIMER_IRQ,
-	.cpumask	= CPU_MASK_CPU0,
 	.set_mode	= r4030_set_mode,
 };
 
 static irqreturn_t r4030_timer_interrupt(int irq, void *dev_id)
 {
-	r4030_clockevent.event_handler(&r4030_clockevent);
+	struct clock_event_device *cd = dev_id;
 
+	cd->event_handler(cd);
 	return IRQ_HANDLED;
 }

@@ -135,15 +135,22 @@ static struct irqaction r4030_timer_irqaction = {
 	.handler	= r4030_timer_interrupt,
 	.flags		= IRQF_DISABLED,
 	.mask		= CPU_MASK_CPU0,
-	.name		= "timer",
+	.name		= "R4030 timer",
 };
 
 void __init plat_time_init(void)
 {
-	struct irqaction *irq = &r4030_timer_irqaction;
+	struct clock_event_device *cd = &r4030_clockevent;
+	struct irqaction *action = &r4030_timer_irqaction;
+	unsigned int cpu = smp_processor_id();
 
 	BUG_ON(HZ != 100);
 
+	cd->cpumask             = cpumask_of_cpu(cpu);
+	clockevents_register_device(cd);
+	action->dev_id = cd;
+	setup_irq(JAZZ_TIMER_IRQ, action);
+
 	/*
 	 * Set clock to 100Hz.
 	 *

@@ -151,8 +158,4 @@ void __init plat_time_init(void)
 	 * a programmable 4-bit divider.  This makes it fairly inflexible.
 	 */
 	r4030_write_reg32(JAZZ_TIMER_INTERVAL, 9);
-	setup_irq(JAZZ_TIMER_IRQ, irq);
-
-	clockevents_register_device(&r4030_clockevent);
-	setup_pit_timer();
 }

Signature · 2 lines

Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]

Ralf Baechle

2007-11-01 15:08 UTC #1 lore raw

Quoted 1 line · Thomas Bogendoerfer

PIT doesn't work, disable it completly

I think this is the explanation:

include/asm-mips/mach-jazz/timex.h:#define CLOCK_TICK_RATE 100

while the PIT code actually expects 1193182.

Turns out that due to a recent Qemu bug which made the probe for the cp0 compare interrupt fail the Malta code did fall back from the compare timer to the i8253 PIT for the clockevent device. Works perfectly well.

Ralf

Ralf Baechle

2007-11-01 16:02 UTC #2 lore raw

Quoted 6 lines · Ralf Baechle

Quoted 1 line · Thomas Bogendoerfer
PIT doesn't work, disable it completly
I think this is the explanation:

Quoted 7 lines

include/asm-mips/mach-jazz/timex.h:#define CLOCK_TICK_RATE 100
while the PIT code actually expects 1193182.
Turns out that due to a recent Qemu bug which made the probe for the cp0 compare interrupt fail the Malta code did fall back from the compare timer to the i8253 PIT for the clockevent device. Works perfectly well.

So I just fixed the MIPS part of the CLOCK_TICK_RATE mess which is really all over the kernel. I hope this should bring the i2853 to life for you. Could you test this? It's pretty much the only hope for Jazz to go tickless so we should try to get it to work. Unfortunately there is a locking issue lurking there as well so I need some feedback if you can get it to work.

Ralf

Thomas Bogendoerfer

author maintainer

2007-11-02 10:20 UTC #3 lore raw

Quoted 1 line · Ralf Baechle

all over the kernel. I hope this should bring the i2853 to life for you.

it does now, even pit clockevent works now (if you apply the patch below).

Thomas.

Fix ISA irq acknowledge
make r4030 clockevent code look like other mips clockevent code

Signed-off-by Thomas Bogendoerfer <[email protected]>

---

arch/mips/jazz/irq.c +16 ~~-11~~

index ae25b48..d7f8a78 100644
@@ -97,9 +97,10 @@ asmlinkage void plat_irq_dispatch(void)
 	if (pending & IE_IRQ4) {
 		r4030_read_reg32(JAZZ_TIMER_REGISTER);
 		do_IRQ(JAZZ_TIMER_IRQ);
-	} else if (pending & IE_IRQ2)
-		do_IRQ(r4030_read_reg32(JAZZ_EISA_IRQ_ACK));
-	else if (pending & IE_IRQ1) {
+	} else if (pending & IE_IRQ2) {
+		irq = *(volatile u8 *)JAZZ_EISA_IRQ_ACK;
+		do_IRQ(irq);
+	} else if (pending & IE_IRQ1) {
 		irq = *(volatile u8 *)JAZZ_IO_IRQ_SOURCE >> 2;
 		if (likely(irq > 0))
 			do_IRQ(irq + JAZZ_IRQ_START - 1);

@@ -117,16 +118,16 @@ static void r4030_set_mode(enum clock_event_mode mode,
 struct clock_event_device r4030_clockevent = {
 	.name		= "r4030",
 	.features	= CLOCK_EVT_FEAT_PERIODIC,
-	.rating		= 100,
+	.rating		= 300,
 	.irq		= JAZZ_TIMER_IRQ,
-	.cpumask	= CPU_MASK_CPU0,
 	.set_mode	= r4030_set_mode,
 };
 
 static irqreturn_t r4030_timer_interrupt(int irq, void *dev_id)
 {
-	r4030_clockevent.event_handler(&r4030_clockevent);
+	struct clock_event_device *cd = dev_id;
 
+	cd->event_handler(cd);
 	return IRQ_HANDLED;
 }

@@ -134,15 +135,22 @@ static struct irqaction r4030_timer_irqaction = {
 	.handler	= r4030_timer_interrupt,
 	.flags		= IRQF_DISABLED,
 	.mask		= CPU_MASK_CPU0,
-	.name		= "timer",
+	.name		= "R4030 timer",
 };
 
 void __init plat_time_init(void)
 {
-	struct irqaction *irq = &r4030_timer_irqaction;
+	struct clock_event_device *cd = &r4030_clockevent;
+	struct irqaction *action = &r4030_timer_irqaction;
+	unsigned int cpu = smp_processor_id();
 
 	BUG_ON(HZ != 100);
 
+	cd->cpumask             = cpumask_of_cpu(cpu);
+	clockevents_register_device(cd);
+	action->dev_id = cd;
+	setup_irq(JAZZ_TIMER_IRQ, action);
+
 	/*
 	 * Set clock to 100Hz.
 	 *

@@ -150,8 +158,5 @@ void __init plat_time_init(void)
 	 * a programmable 4-bit divider.  This makes it fairly inflexible.
 	 */
 	r4030_write_reg32(JAZZ_TIMER_INTERVAL, 9);
-	setup_irq(JAZZ_TIMER_IRQ, irq);
-
-	clockevents_register_device(&r4030_clockevent);
 	setup_pit_timer();
 }

Signature · 2 lines

Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]

Ralf Baechle

Applied

2007-11-02 12:20 UTC #4 lore raw

Quoted 5 lines · Thomas Bogendoerfer

Quoted 1 line · Ralf Baechle
all over the kernel. I hope this should bring the i2853 to life for you.
it does now, even pit clockevent works now (if you apply the patch below).

Applied, thanks,

One thing I'm still wondering about, does the kernel actually go tickless for you?

Ralf

Thomas Bogendoerfer

author maintainer

2007-11-02 22:12 UTC #5 lore raw

Quoted 2 lines · Ralf Baechle

One thing I'm still wondering about, does the kernel actually go tickless for you?

a kernel with CONFIG_NO_HZ boots and acts normal. But it looks like the PIT is still ticking at the selected 100HZ...

Thomas.

Signature · 2 lines

Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]

Ralf Baechle

2007-11-04 00:33 UTC #6 lore raw

Quoted 6 lines · Thomas Bogendoerfer

Quoted 2 lines · Ralf Baechle
One thing I'm still wondering about, does the kernel actually go tickless for you?
a kernel with CONFIG_NO_HZ boots and acts normal. But it looks like the PIT is still ticking at the selected 100HZ...

I think this happens because the R4030 clockevent device has a rather high rating of 300 while the PIT because of an omission or intension doesn't set it's rating at all so has rating of 0. So Linux prefers the R4030. You can see which clockevent device is actually getting used by Linux in /proc/timer_list.

Ralf

Thomas Bogendoerfer

author maintainer

2007-11-04 10:06 UTC #7 lore raw

Quoted 10 lines · Ralf Baechle

Quoted 6 lines · Thomas Bogendoerfer
Quoted 2 lines · Ralf Baechle
One thing I'm still wondering about, does the kernel actually go tickless for you?
a kernel with CONFIG_NO_HZ boots and acts normal. But it looks like the PIT is still ticking at the selected 100HZ...
I think this happens because the R4030 clockevent device has a rather

r4030 is of course disabled in my test.

Thomas.

Signature · 2 lines

Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]