Re: Inconsistent load average on tickless kernels
From: Peter Zijlstra
Date: Mon Mar 05 2012 - 18:32:51 EST
On Tue, 2012-03-06 at 00:25 +0100, Peter Zijlstra wrote:
> I tried writing hpet64 support so we could idle that long, killed all
> kinds of stupid kernel threads (watchdogs mostly) that keep waking up
> and got a brick..
Just in case someone wants to have a go at fixing this mess.. :-)
I _think_ the below was the latest, but it was 2am on friday night or
something, so recollections are somewhat hazy.
---
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index ad0de0c..fd2aab0 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -17,6 +17,7 @@
#include <asm/time.h>
#define HPET_MASK CLOCKSOURCE_MASK(32)
+#define HPET64_MASK CLOCKSOURCE_MASK(64)
/* FSEC = 10^-15
NSEC = 10^-9 */
@@ -43,6 +44,12 @@ static unsigned long hpet_num_timers;
#endif
static void __iomem *hpet_virt_address;
+#ifdef X86_64
+static int hpet64;
+#else
+ #define hpet64 (0)
+#endif
+
struct hpet_dev {
struct clock_event_device evt;
unsigned int num;
@@ -67,6 +74,26 @@ static inline void hpet_writel(unsigned int d, unsigned int a)
writel(d, hpet_virt_address + a);
}
+inline u64 hpet_read(unsigned int a)
+{
+ u64 ret;
+
+ if (hpet64)
+ ret = readq(hpet_virt_address + a);
+ else
+ ret = readl(hpet_virt_address + a);
+
+ return ret;
+}
+
+static inline void hpet_write(u64 d, unsigned int a)
+{
+ if (hpet64)
+ writeq(d, hpet_virt_address + a);
+ else
+ writel(d, hpet_virt_address + a);
+}
+
#ifdef CONFIG_X86_64
#include <asm/pgtable.h>
#endif
@@ -91,6 +118,10 @@ static inline void hpet_clear_mapping(void)
static int boot_hpet_disable;
int hpet_force_user;
static int hpet_verbose;
+#ifdef X86_64
+static int hpet_force_64;
+static int hpet_force_32;
+#endif
static int __init hpet_setup(char *str)
{
@@ -101,6 +132,12 @@ static int __init hpet_setup(char *str)
hpet_force_user = 1;
if (!strncmp("verbose", str, 7))
hpet_verbose = 1;
+#ifdef X86_64
+ if(!strncmp("force64", str, 7))
+ hpet_force_64 = 1;
+ if(!strncmp("force32", str, 7))
+ hpet_force_32 = 1;
+#endif
}
return 1;
}
@@ -249,8 +286,11 @@ static void hpet_stop_counter(void)
static void hpet_reset_counter(void)
{
- hpet_writel(0, HPET_COUNTER);
- hpet_writel(0, HPET_COUNTER + 4);
+ if (!hpet64) {
+ hpet_writel(0, HPET_COUNTER);
+ hpet_writel(0, HPET_COUNTER + 4);
+ } else
+ hpet_write(0, HPET_COUNTER);
}
static void hpet_start_counter(void)
@@ -298,7 +338,8 @@ static void hpet_legacy_clockevent_register(void)
*/
hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
clockevents_config_and_register(&hpet_clockevent, hpet_freq,
- HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
+ HPET_MIN_PROG_DELTA,
+ hpet64 ? 0x7FFFFFFFFFFFFFFF : 0x7FFFFFFF);
global_clock_event = &hpet_clockevent;
printk(KERN_DEBUG "hpet clockevent registered\n");
}
@@ -308,23 +349,25 @@ static int hpet_setup_msi_irq(unsigned int irq);
static void hpet_set_mode(enum clock_event_mode mode,
struct clock_event_device *evt, int timer)
{
- unsigned int cfg, cmp, now;
- uint64_t delta;
+ uint64_t delta, cmp, now;
+ unsigned int cfg;
switch (mode) {
case CLOCK_EVT_MODE_PERIODIC:
hpet_stop_counter();
delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
delta >>= evt->shift;
- now = hpet_readl(HPET_COUNTER);
+ now = hpet_read(HPET_COUNTER);
cmp = now + (unsigned int) delta;
cfg = hpet_readl(HPET_Tn_CFG(timer));
/* Make sure we use edge triggered interrupts */
cfg &= ~HPET_TN_LEVEL;
cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
- HPET_TN_SETVAL | HPET_TN_32BIT;
+ HPET_TN_SETVAL;
+ if (!hpet64)
+ cfg |= HPET_TN_32BIT;
hpet_writel(cfg, HPET_Tn_CFG(timer));
- hpet_writel(cmp, HPET_Tn_CMP(timer));
+ hpet_write(cmp, HPET_Tn_CMP(timer));
udelay(1);
/*
* HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
@@ -333,7 +376,7 @@ static void hpet_set_mode(enum clock_event_mode mode,
* (See AMD-8111 HyperTransport I/O Hub Data Sheet,
* Publication # 24674)
*/
- hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
+ hpet_write(delta, HPET_Tn_CMP(timer));
hpet_start_counter();
hpet_print_config();
break;
@@ -341,7 +384,9 @@ static void hpet_set_mode(enum clock_event_mode mode,
case CLOCK_EVT_MODE_ONESHOT:
cfg = hpet_readl(HPET_Tn_CFG(timer));
cfg &= ~HPET_TN_PERIODIC;
- cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
+ cfg |= HPET_TN_ENABLE;
+ if (!hpet64)
+ cfg |= HPET_TN_32BIT;
hpet_writel(cfg, HPET_Tn_CFG(timer));
break;
@@ -370,12 +415,12 @@ static void hpet_set_mode(enum clock_event_mode mode,
static int hpet_next_event(unsigned long delta,
struct clock_event_device *evt, int timer)
{
- u32 cnt;
- s32 res;
+ u64 cnt;
+ s64 res;
- cnt = hpet_readl(HPET_COUNTER);
- cnt += (u32) delta;
- hpet_writel(cnt, HPET_Tn_CMP(timer));
+ cnt = hpet_read(HPET_COUNTER);
+ cnt += delta;
+ hpet_write(cnt, HPET_Tn_CMP(timer));
/*
* HPETs are a complete disaster. The compare register is
@@ -399,7 +444,7 @@ static int hpet_next_event(unsigned long delta,
* the event. The minimum programming delta for the generic
* clockevents code is set to 1.5 * HPET_MIN_CYCLES.
*/
- res = (s32)(cnt - hpet_readl(HPET_COUNTER));
+ res = (s64)(cnt - hpet_read(HPET_COUNTER));
return res < HPET_MIN_CYCLES ? -ETIME : 0;
}
@@ -739,7 +784,7 @@ static int hpet_cpuhp_notify(struct notifier_block *n,
*/
static cycle_t read_hpet(struct clocksource *cs)
{
- return (cycle_t)hpet_readl(HPET_COUNTER);
+ return (cycle_t)hpet_read(HPET_COUNTER);
}
static struct clocksource clocksource_hpet = {
@@ -763,7 +808,7 @@ static int hpet_clocksource_register(void)
hpet_restart_counter();
/* Verify whether hpet counter works */
- t1 = hpet_readl(HPET_COUNTER);
+ t1 = hpet_read(HPET_COUNTER);
rdtscll(start);
/*
@@ -777,7 +822,7 @@ static int hpet_clocksource_register(void)
rdtscll(now);
} while ((now - start) < 200000UL);
- if (t1 == hpet_readl(HPET_COUNTER)) {
+ if (t1 == hpet_read(HPET_COUNTER)) {
printk(KERN_WARNING
"HPET counter not counting. HPET disabled\n");
return -ENODEV;
@@ -847,6 +892,13 @@ int __init hpet_enable(void)
id = hpet_readl(HPET_ID);
hpet_print_config();
+#ifdef X86_64
+ if (((id & HPET_ID_64BIT) || hpet_force_64) && !hpet_force_32) {
+ hpet64 = 1;
+ clocksource_hpet.mask = HPET64_MASK;
+ }
+#endif
+
#ifdef CONFIG_HPET_EMULATE_RTC
/*
* The legacy routing mode needs at least two channels, tick timer
@@ -962,9 +1014,9 @@ static unsigned long hpet_rtc_flags;
static int hpet_prev_update_sec;
static struct rtc_time hpet_alarm_time;
static unsigned long hpet_pie_count;
-static u32 hpet_t1_cmp;
-static u32 hpet_default_delta;
-static u32 hpet_pie_delta;
+static u64 hpet_t1_cmp;
+static u64 hpet_default_delta;
+static u64 hpet_pie_delta;
static unsigned long hpet_pie_limit;
static rtc_irq_handler irq_handler;
@@ -972,9 +1024,9 @@ static rtc_irq_handler irq_handler;
/*
* Check that the hpet counter c1 is ahead of the c2
*/
-static inline int hpet_cnt_ahead(u32 c1, u32 c2)
+static inline int hpet_cnt_ahead(u64 c1, u64 c2)
{
- return (s32)(c2 - c1) < 0;
+ return (s64)(c2 - c1) < 0;
}
/*
@@ -1015,7 +1067,8 @@ EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
*/
int hpet_rtc_timer_init(void)
{
- unsigned int cfg, cnt, delta;
+ unsigned int cfg;
+ u64 cnt, delta;
unsigned long flags;
if (!is_hpet_enabled())
@@ -1036,13 +1089,15 @@ int hpet_rtc_timer_init(void)
local_irq_save(flags);
- cnt = delta + hpet_readl(HPET_COUNTER);
- hpet_writel(cnt, HPET_T1_CMP);
+ cnt = delta + hpet_read(HPET_COUNTER);
+ hpet_write(cnt, HPET_T1_CMP);
hpet_t1_cmp = cnt;
cfg = hpet_readl(HPET_T1_CFG);
cfg &= ~HPET_TN_PERIODIC;
- cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
+ cfg |= HPET_TN_ENABLE;
+ if (!hpet64)
+ cfg |= HPET_TN_32BIT;
hpet_writel(cfg, HPET_T1_CFG);
local_irq_restore(flags);
@@ -1155,9 +1210,9 @@ static void hpet_rtc_timer_reinit(void)
*/
do {
hpet_t1_cmp += delta;
- hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
+ hpet_write(hpet_t1_cmp, HPET_T1_CMP);
lost_ints++;
- } while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
+ } while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_read(HPET_COUNTER)));
if (lost_ints) {
if (hpet_rtc_flags & RTC_PIE)
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/