Adds a call back interface for register/rating change events. This is also used
later in this series to signal other interesting events.
Signed-Off-By: Daniel Walker <[email protected]>
---
include/linux/clocksource.h | 37 +++++++++++++++++++++++++++++++++++++
include/linux/timekeeping.h | 3 +++
kernel/time/clocksource.c | 10 ++++++++++
3 files changed, 50 insertions(+)
Index: linux-2.6.19/include/linux/clocksource.h
===================================================================
--- linux-2.6.19.orig/include/linux/clocksource.h
+++ linux-2.6.19/include/linux/clocksource.h
@@ -13,6 +13,7 @@
#include <linux/time.h>
#include <linux/list.h>
#include <linux/sysdev.h>
+#include <linux/notifier.h>
#include <asm/div64.h>
#include <asm/io.h>
@@ -23,6 +24,42 @@ typedef u64 cycle_t;
/* XXX - Would like a better way for initializing curr_clocksource */
extern struct clocksource clocksource_jiffies;
+/*
+ * Allows inlined calling for notifier routines.
+ */
+extern struct atomic_notifier_head clocksource_list_notifier;
+
+/*
+ * Block notifier flags.
+ */
+#define CLOCKSOURCE_NOTIFY_REGISTER 1
+#define CLOCKSOURCE_NOTIFY_RATING 2
+#define CLOCKSOURCE_NOTIFY_FREQ 4
+
+/**
+ * clocksource_notifier_register - Registers a list change notifier
+ * @nb: pointer to a notifier block
+ *
+ * Returns zero always.
+ */
+static inline int clocksource_notifier_register(struct notifier_block *nb)
+{
+ return atomic_notifier_chain_register(&clocksource_list_notifier, nb);
+}
+
+/**
+ * clocksource_freq_change - Allows notification of dynamic frequency changes.
+ *
+ * Signals that a clocksource is dynamically changing it's frequency.
+ * This could happen if a clocksource becomes more/less stable.
+ */
+static inline void clocksource_freq_change(struct clocksource *c)
+{
+ atomic_notifier_call_chain(&clocksource_list_notifier,
+ CLOCKSOURCE_NOTIFY_FREQ, c);
+}
+
+
/**
* struct clocksource - hardware abstraction for a free running counter
* Provides mostly state-free accessors to the underlying hardware.
Index: linux-2.6.19/include/linux/timekeeping.h
===================================================================
--- linux-2.6.19.orig/include/linux/timekeeping.h
+++ linux-2.6.19/include/linux/timekeeping.h
@@ -8,6 +8,9 @@ static inline int change_clocksource(voi
{
return 0;
}
+
+static inline void change_clocksource(void) { }
+
#endif /* !CONFIG_GENERIC_TIME */
#endif /* _LINUX_TIMEKEEPING_H */
Index: linux-2.6.19/kernel/time/clocksource.c
===================================================================
--- linux-2.6.19.orig/kernel/time/clocksource.c
+++ linux-2.6.19/kernel/time/clocksource.c
@@ -49,6 +49,8 @@ static DEFINE_SPINLOCK(clocksource_lock)
static char override_name[32];
static int finished_booting;
+ATOMIC_NOTIFIER_HEAD(clocksource_list_notifier);
+
/* clocksource_done_booting - Called near the end of bootup
*
* Hack to avoid lots of clocksource churn at boot time
@@ -193,6 +195,10 @@ int clocksource_register(struct clocksou
*/
next_clocksource = select_clocksource();
spin_unlock_irqrestore(&clocksource_lock, flags);
+
+ atomic_notifier_call_chain(&clocksource_list_notifier,
+ CLOCKSOURCE_NOTIFY_REGISTER, c);
+
return ret;
}
EXPORT_SYMBOL(clocksource_register);
@@ -218,6 +224,10 @@ void clocksource_rating_change(struct cl
next_clocksource = select_clocksource();
spin_unlock_irqrestore(&clocksource_lock, flags);
+
+ atomic_notifier_call_chain(&clocksource_list_notifier,
+ CLOCKSOURCE_NOTIFY_RATING, c);
+
}
EXPORT_SYMBOL(clocksource_rating_change);
--
* Daniel Walker <[email protected]> wrote:
> Adds a call back interface for register/rating change events. This is
> also used later in this series to signal other interesting events.
This patch adds:
> +#define CLOCKSOURCE_NOTIFY_REGISTER 1
> +#define CLOCKSOURCE_NOTIFY_RATING 2
> +#define CLOCKSOURCE_NOTIFY_FREQ 4
and a later patch adds:
> +#define CLOCKSOURCE_NOTIFY_UNSTABLE 8
while i tentatively agree with the introduction of clocksource driver
notifications in general (given any actual real users of that
infrastructure), i can see three problems with your notifier changes:
firstly, with the full changes applied, CLOCKSOURCE_NOTIFY_RATING is
totally unused (and not even driven by the core clocksource code).
secondly, CLOCKSOURCE_NOTIFY_REGISTER is only used by the core
clocksource code and no register-notifier interface is exposed: this
makes the whole thing an expensive NOP.
thirdly, CLOCKSOURCE_NOTIFY_FREQ is totally unused as well.
(there are tons of examples in the kernel of how to do driver
notification APIs properly.)
Ingo
On Wed, 2007-01-31 at 11:17 +0100, Ingo Molnar wrote:
> * Daniel Walker <[email protected]> wrote:
>
> > Adds a call back interface for register/rating change events. This is
> > also used later in this series to signal other interesting events.
>
> This patch adds:
>
> > +#define CLOCKSOURCE_NOTIFY_REGISTER 1
> > +#define CLOCKSOURCE_NOTIFY_RATING 2
> > +#define CLOCKSOURCE_NOTIFY_FREQ 4
>
> and a later patch adds:
>
> > +#define CLOCKSOURCE_NOTIFY_UNSTABLE 8
>
> while i tentatively agree with the introduction of clocksource driver
> notifications in general (given any actual real users of that
> infrastructure), i can see three problems with your notifier changes:
>
> firstly, with the full changes applied, CLOCKSOURCE_NOTIFY_RATING is
> totally unused (and not even driven by the core clocksource code).
True, ratings are constant so the notification never happens (and never
needs to)
> secondly, CLOCKSOURCE_NOTIFY_REGISTER is only used by the core
> clocksource code and no register-notifier interface is exposed: this
> makes the whole thing an expensive NOP.
I don't know what you mean when you say "no register-notifier interface
is exposed" .. The timekeeping code doesn't use this notification , but
it really should. It's not that big of an issue since we don't have many
clocks in modules..
> thirdly, CLOCKSOURCE_NOTIFY_FREQ is totally unused as well.
Indeed . Consider it dropped.
> (there are tons of examples in the kernel of how to do driver
> notification APIs properly.)
Are you referring to the comments above , or something else your seeing
in the code?
Daniel
* Daniel Walker <[email protected]> wrote:
> > secondly, CLOCKSOURCE_NOTIFY_REGISTER is only used by the core
> > clocksource code and no register-notifier interface is exposed: this
> > makes the whole thing an expensive NOP.
>
> I don't know what you mean when you say "no register-notifier
> interface is exposed" .. The timekeeping code doesn't use this
> notification , but it really should. [...]
what i say is that there's no public interface exposed of this notifier
(and even if there was, there isnt any user of it) and there's only a
single notifier used internally: this means that this thing is an
expensive function call.
Ingo