Skip to contentSkip to conversation

v3 Moorestown clock related patches

PATCH In review Jacob Pan 2 participants 10 messages May 18, 2010 – May 19, 2010 x86

lore
Share & tools
lore

Jacob Pan

author
#0 lore raw

cover v3 Moorestown clock related patches

Hi,
Please review the two patches related Moorestown/Medfield clock selection logic. tglx has suggested a much cleaner logic in
http://lkml.org/lkml/2010/5/17/406
So patch #2 is the implementation of new logic plus small tweaks based on testing results.
I also dropped x86_mask check for Penwell CPU check. It is not needed, CPU family and model should be sufficient/accurate to identify Lincroft vs Penwell.

*** BLURB HERE ***

Jacob Pan (2):
x86/mrst: add cpu type detection
x86/mrst: add more timer config options

 arch/x86/include/asm/apb_timer.h |    1 -
 arch/x86/include/asm/mrst.h      |   20 +++++++
 arch/x86/kernel/apb_timer.c      |   37 +++----------
 arch/x86/kernel/mrst.c           |  109 ++++++++++++++++++++++++++++++--------
 4 files changed, 114 insertions(+), 53 deletions(-)
author
#1 lore raw

PATCH 1/2 v4 x86/mrst: add cpu type detection

Medfield is the follow-up of Moorestown, it is treated under the same HW sub-architecture. However, we do need to know the CPU type in order for some of the driver to act accordingly.
We also have different optimal clock configuration for each CPU type.

---
 arch/x86/include/asm/mrst.h |   19 +++++++++++++++++++
 arch/x86/kernel/mrst.c      |   20 ++++++++++++++++++++
 2 files changed, 39 insertions(+), 0 deletions(-)
arch/x86/include/asm/mrst.h +19 -0
index 451d30e..a25eff3 100644
@@ -11,8 +11,27 @@
 #ifndef _ASM_X86_MRST_H
 #define _ASM_X86_MRST_H
 extern int pci_mrst_init(void);
+extern int mrst_identify_cpu(void);
 int __init sfi_parse_mrtc(struct sfi_table_header *table);
 
+/**
+ * Medfield is the follow-up of Moorestown, it combines two chip solution into
+ * one. Other than that it also added always-on and constant tsc and lapic
+ * timers. Medfield is the platform name, and the chip name is called Penwell
+ * we treat Medfield/Penwell as a variant of Moorestown. Penwell can be
+ * identified via MSRs.
+ */
+enum mrst_cpu_type {
+	MRST_CPU_CHIP_LINCROFT = 1,
+	MRST_CPU_CHIP_PENWELL,
+};
+
+enum mrst_timer_options {
+	MRST_TIMER_DEFAULT,
+	MRST_TIMER_APBT_ONLY,
+	MRST_TIMER_LAPIC_APBT,
+};
+
 #define SFI_MTMR_MAX_NUM 8
 #define SFI_MRTC_MAX	8
 
arch/x86/kernel/mrst.c +20 -0
index 0aad867..0bca837 100644
@@ -216,6 +216,26 @@ static void __init mrst_setup_boot_clock(void)
 		setup_boot_APIC_clock();
 };
 
+int mrst_identify_cpu(void)
+{
+	u32 mrst_cpu_chip;
+
+	if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 0x27)
+		mrst_cpu_chip = MRST_CPU_CHIP_PENWELL;
+	else if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 0x26)
+		mrst_cpu_chip = MRST_CPU_CHIP_LINCROFT;
+	else {
+		pr_err("Unknown Moorestown CPU (%d:%d), default to Lincroft\n",
+			boot_cpu_data.x86, boot_cpu_data.x86_model);
+		mrst_cpu_chip = MRST_CPU_CHIP_LINCROFT;
+	}
+	pr_debug("Moorestown CPU %s identified\n",
+		(mrst_cpu_chip == MRST_CPU_CHIP_LINCROFT) ?
+		"Lincroft" : "Penwell");
+	return mrst_cpu_chip;
+}
+EXPORT_SYMBOL_GPL(mrst_identify_cpu);
+
 /*
  * Moorestown specific x86_init function overrides and early setup
  * calls.
Signature · 1 line
1.6.3.3
#2 lore raw

Re: [PATCH 1/2] v4 x86/mrst: add cpu type detection

Quoted 7 lines · Jacob Pan
arch/x86/kernel/mrst.c +6 -0
+int mrst_identify_cpu(void)
+{
+	u32 mrst_cpu_chip;
+
+	if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 0x27)
+		mrst_cpu_chip = MRST_CPU_CHIP_PENWELL;

arch/x86/kernel/mrst.c static void __init mrst_setup_boot_clock(void) · line 234

Quoted 10 lines
arch/x86/kernel/mrst.c +10 -0
+	else if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 0x26)
+		mrst_cpu_chip = MRST_CPU_CHIP_LINCROFT;
+	else {
+		pr_err("Unknown Moorestown CPU (%d:%d), default to Lincroft\n",
+			boot_cpu_data.x86, boot_cpu_data.x86_model);
+		mrst_cpu_chip = MRST_CPU_CHIP_LINCROFT;
+	}
+	pr_debug("Moorestown CPU %s identified\n",
+		(mrst_cpu_chip == MRST_CPU_CHIP_LINCROFT) ?
+		"Lincroft" : "Penwell");

Do you really want a printk for each call ?

Aside of that wouldn't it make more sense to do the chip identification only once from mrst_init() and cache the result and change mrst_identify_cpu()

int mrst_indetify_cpu(void)

{
	return mrst_cpu_chip;
}

EXPORT_SYMBOL_GPL(mrst_identify_cpu);

I think one of your earlier versions did that, but was not consistent in it's behaviour.

Thanks,

tglx

author
#3 lore raw

Re: [PATCH 1/2] v4 x86/mrst: add cpu type detection

Quoted 5 lines

Aside of that wouldn't it make more sense to do the chip identification only once from mrst_init() and cache the result and change mrst_identify_cpu()

the problem is again the ordering of cpuid, boot_cpu_data is not available by the time we call x86_mrst_early_setup(). I can move it to mrst_time_init() but not very ideal. Unless we go back to do cpuid directly in the v1 patch :).

#4 lore raw

Re: [PATCH 1/2] v4 x86/mrst: add cpu type detection

Quoted 7 lines · jacob pan
Quoted 5 lines

Aside of that wouldn't it make more sense to do the chip identification only once from mrst_init() and cache the result and change mrst_identify_cpu()

the problem is again the ordering of cpuid, boot_cpu_data is not available by the time we call x86_mrst_early_setup(). I can move it to mrst_time_init()

Hmm, right.

Quoted 1 line

but not very ideal. Unless we go back to do cpuid directly in the v1 patch :).

x86_init.oem.arch_setup() is the point where you can do that, that's called after early_cpu_init() which fills boot_cpu_data. It's way before the first user of mrst_chip.

Thanks,

tglx

author
#5 lore raw

PATCH 2/2 v4 x86/mrst: add more timer config options

In reply to Jacob Pan

Always-on local APIC timer (ARAT) has been introduced to Medfield, along with the platform APB timers we have more timer configuration options between Moorestown and Medfield.

This patch adds run-time detection of avaiable timer features so that we can treat Medfield as a variant of Moorestown and set up the optimal timer options for each platform. i.e.

Medfield: per cpu always-on local APIC timer
Moorestown: per cpu APB timer

Manual override is possible via cmdline option x86_mrst_timer.

---
 arch/x86/include/asm/apb_timer.h |    1 -
 arch/x86/include/asm/mrst.h      |    1 +
 arch/x86/kernel/apb_timer.c      |   37 +++------------
 arch/x86/kernel/mrst.c           |   93 +++++++++++++++++++++++++++----------
 4 files changed, 77 insertions(+), 55 deletions(-)
arch/x86/include/asm/apb_timer.h +0 -1
index c74a2ee..a69b1ac 100644
@@ -55,7 +55,6 @@ extern unsigned long apbt_quick_calibrate(void);
 extern int arch_setup_apbt_irqs(int irq, int trigger, int mask, int cpu);
 extern void apbt_setup_secondary_clock(void);
 extern unsigned int boot_cpu_id;
-extern int disable_apbt_percpu;
 
 extern struct sfi_timer_table_entry *sfi_get_mtmr(int hint);
 extern void sfi_free_mtmr(struct sfi_timer_table_entry *mtmr);
arch/x86/include/asm/mrst.h +1 -0
index a25eff3..f683603 100644
@@ -12,6 +12,7 @@
 #define _ASM_X86_MRST_H
 extern int pci_mrst_init(void);
 extern int mrst_identify_cpu(void);
+extern int mrst_timer_options __cpuinitdata;
 int __init sfi_parse_mrtc(struct sfi_table_header *table);
 
 /**
arch/x86/kernel/apb_timer.c +8 -29
index a353475..8dd7780 100644
@@ -43,10 +43,11 @@
 
 #include <asm/fixmap.h>
 #include <asm/apb_timer.h>
+#include <asm/mrst.h>
 
 #define APBT_MASK			CLOCKSOURCE_MASK(32)
 #define APBT_SHIFT			22
-#define APBT_CLOCKEVENT_RATING		150
+#define APBT_CLOCKEVENT_RATING		110
 #define APBT_CLOCKSOURCE_RATING		250
 #define APBT_MIN_DELTA_USEC		200
 
@@ -83,8 +84,6 @@ struct apbt_dev {
 	char name[10];
 };
 
-int disable_apbt_percpu __cpuinitdata;
-
 static DEFINE_PER_CPU(struct apbt_dev, cpu_apbt_dev);
 
 #ifdef CONFIG_SMP
@@ -195,29 +194,6 @@ static struct clock_event_device apbt_clockevent = {
 };
 
 /*
- * if user does not want to use per CPU apb timer, just give it a lower rating
- * than local apic timer and skip the late per cpu timer init.
- */
-static inline int __init setup_x86_mrst_timer(char *arg)
-{
-	if (!arg)
-		return -EINVAL;
-
-	if (strcmp("apbt_only", arg) == 0)
-		disable_apbt_percpu = 0;
-	else if (strcmp("lapic_and_apbt", arg) == 0)
-		disable_apbt_percpu = 1;
-	else {
-		pr_warning("X86 MRST timer option %s not recognised"
-			   " use x86_mrst_timer=apbt_only or lapic_and_apbt\n",
-			   arg);
-		return -EINVAL;
-	}
-	return 0;
-}
-__setup("x86_mrst_timer=", setup_x86_mrst_timer);
-
-/*
  * start count down from 0xffff_ffff. this is done by toggling the enable bit
  * then load initial load count to ~0.
  */
@@ -335,7 +311,7 @@ static int __init apbt_clockevent_register(void)
 	adev->num = smp_processor_id();
 	memcpy(&adev->evt, &apbt_clockevent, sizeof(struct clock_event_device));
 
-	if (disable_apbt_percpu) {
+	if (mrst_timer_options == MRST_TIMER_LAPIC_APBT) {
 		apbt_clockevent.rating = APBT_CLOCKEVENT_RATING - 100;
 		global_clock_event = &adev->evt;
 		printk(KERN_DEBUG "%s clockevent registered as global\n",
@@ -429,7 +405,8 @@ static int apbt_cpuhp_notify(struct notifier_block *n,
 
 static __init int apbt_late_init(void)
 {
-	if (disable_apbt_percpu || !apb_timer_block_enabled)
+	if (mrst_timer_options == MRST_TIMER_LAPIC_APBT ||
+		!apb_timer_block_enabled)
 		return 0;
 	/* This notifier should be called after workqueue is ready */
 	hotcpu_notifier(apbt_cpuhp_notify, -20);
@@ -450,6 +427,8 @@ static void apbt_set_mode(enum clock_event_mode mode,
 	int timer_num;
 	struct apbt_dev *adev = EVT_TO_APBT_DEV(evt);
 
+	BUG_ON(!apbt_virt_address);
+
 	timer_num = adev->num;
 	pr_debug("%s CPU %d timer %d mode=%d\n",
 		 __func__, first_cpu(*evt->cpumask), timer_num, mode);
@@ -676,7 +655,7 @@ void __init apbt_time_init(void)
 	}
 #ifdef CONFIG_SMP
 	/* kernel cmdline disable apb timer, so we will use lapic timers */
-	if (disable_apbt_percpu) {
+	if (mrst_timer_options == MRST_TIMER_LAPIC_APBT) {
 		printk(KERN_INFO "apbt: disabled per cpu timer\n");
 		return;
 	}
arch/x86/kernel/mrst.c +68 -25

129 lines not on this page · raw message

Signature · 1 line
1.6.3.3
#6 lore raw

Re: [PATCH 2/2] v4 x86/mrst: add more timer config options

Quoted 5 lines · Jacob Pan

Always-on local APIC timer (ARAT) has been introduced to Medfield, along with the platform APB timers we have more timer configuration options between Moorestown and Medfield.

+/**

Do not use kerneldoc /** opener from plain comments. Randy complained about that yesterday already

arch/x86/kernel/mrst.c static unsigned long __init mrst_calibrate_tsc(void) · line 211

Quoted 5 lines

void __init mrst_time_init(void)
{

arch/x86/kernel/mrst.c +3 -0
+	switch (mrst_timer_options) {
+	case MRST_TIMER_APBT_ONLY:
+		pr_info("Use APB timer option on Moorestown\n");

Do we really need all this pr_info spew ? /proc/timer_list will tell you if you're interested

arch/x86/kernel/mrst.c static unsigned long __init mrst_calibrate_tsc(void) · line 219

Quoted 8 lines
arch/x86/kernel/mrst.c +8 -0
+		break;
+	case MRST_TIMER_LAPIC_APBT:
+		pr_info("Use LAPIC and APB timer option on Moorestown\n");
+		x86_init.timers.setup_percpu_clockev = setup_boot_APIC_clock;
+		x86_cpuinit.setup_percpu_clockev = setup_secondary_APIC_clock;
+		break;
+	default:
+		pr_info("Use default timer option on Moorestown\n");

And this printk is completely useless. It'd be way more interesting to see which timer gets selected as a default. but see above

arch/x86/kernel/mrst.c static unsigned long __init mrst_calibrate_tsc(void) · line 220

Quoted 1 line
arch/x86/kernel/mrst.c +1 -0
+		if (mrst_identify_cpu() == MRST_CPU_CHIP_PENWELL) {

Why do you want to pin that on PENWELL and not on ARAT ? If you check ARAT, you have coverd all next gen cpus as well.

arch/x86/kernel/mrst.c static unsigned long __init mrst_calibrate_tsc(void) · line 226

Quoted 6 lines
arch/x86/kernel/mrst.c +6 -0
+			x86_init.timers.setup_percpu_clockev =
+				setup_boot_APIC_clock;
+			x86_cpuinit.setup_percpu_clockev =
+				setup_secondary_APIC_clock;
+			return;
+		}

You can avoid that line splits by doing:

      	    	if (!has(ARAT))
			break:

		x86_init.timers. ...
		...
		return;

Way easier to read.

Thanks,

tglx

author
#7 lore raw

Re: [PATCH 2/2] v4 x86/mrst: add more timer config options

Quoted 7 lines
Quoted 3 lines

+/**

Do not use kerneldoc /** opener from plain comments. Randy complained about that yesterday already

I missed randy's comment at the end. sorry. I admit I did not know kernel-doc before that.

		if (mrst_identify_cpu() == MRST_CPU_CHIP_PENWELL) {
Quoted 3 lines

Why do you want to pin that on PENWELL and not on ARAT ? If you check ARAT, you have coverd all next gen cpus as well.

I did try ARAT first but ARAT is not available at this point. the add-on cpu features are set later.

Thanks,

Jacob

#8 lore raw

Re: [PATCH 2/2] v4 x86/mrst: add more timer config options

Quoted 10 lines · jacob pan
Quoted 7 lines
Quoted 3 lines

+/**

Do not use kerneldoc /** opener from plain comments. Randy complained about that yesterday already

I missed randy's comment at the end. sorry. I admit I did not know kernel-doc before that.

Quoted 7 lines
		if (mrst_identify_cpu() == MRST_CPU_CHIP_PENWELL) {
Quoted 3 lines

Why do you want to pin that on PENWELL and not on ARAT ? If you check ARAT, you have coverd all next gen cpus as well.

I did try ARAT first but ARAT is not available at this point. the add-on cpu features are set later.

Hmm, I don't see a reason why we don't to the readout of the extra features right away. I might be missing something really important though :) hpa ?????

Thanks,

tglx

author
#9 lore raw
In reply to Jacob Pan

This is should be the v4 cover letter instead of v3 shown in the subject line. sorry about that.

Jacob Pan Tue, 18 May 2010 05:02:44 -0700

Quoted 22 lines

Hi,
Please review the two patches related Moorestown/Medfield clock selection logic. tglx has suggested a much cleaner logic in
http://lkml.org/lkml/2010/5/17/406
So patch #2 is the implementation of new logic plus small tweaks based on testing results.
I also dropped x86_mask check for Penwell CPU check. It is not needed, CPU family and model should be sufficient/accurate to identify Lincroft vs Penwell.

*** BLURB HERE ***

Jacob Pan (2):
x86/mrst: add cpu type detection
x86/mrst: add more timer config options

arch/x86/include/asm/apb_timer.h |    1 -
arch/x86/include/asm/mrst.h      |   20 +++++++
arch/x86/kernel/apb_timer.c      |   37 +++----------
arch/x86/kernel/mrst.c           |  109 ++++++++++++++++++++++++++++++--------
4 files changed, 114 insertions(+), 53 deletions(-)