Skip to contentSkip to conversation

remove clocksource inline functions

PATCH john stultz 3 participants 6 messages Aug 13, 2009 – Aug 14, 2009 3 messages not in the archive x86

lore
Share & tools
lore
#2 lore raw
In reply to a message not in the archive
Quoted 25 lines · Martin Schwidefsky

plain text document attachment (clocksource-inline.diff)
From: Martin Schwidefsky <[email protected]>

The three inline functions clocksource_read, clocksource_enable and clocksource_disable are simple wrappers of an indirect call plus the copy from and to the mult_orig value. The functions are exclusively used by the timekeeping code which has intimate knowledge of the clocksource anyway. Therefore remove the inline functions. No functional change.

---
 include/linux/clocksource.h |   58
--------------------------------------------
 kernel/time/timekeeping.c   |   42 +++++++++++++++++++++----------
 2 files changed, 28 insertions(+), 72 deletions(-)

Index: linux-2.6/kernel/time/timekeeping.c
===================================================================

kernel/time/timekeeping.c +0 -0

[snip]

kernel/time/timekeeping.c static int timekeeping_resume(struct sys · line 495

Quoted 7 lines
kernel/time/timekeeping.c +1 -2
@@ -477,8 +492,7 @@ static int timekeeping_resume(struct sys
 	}
 	update_xtime_cache(0);
 	/* re-base the last cycle value */
-	clock->cycle_last = 0;
-	clock->cycle_last = clocksource_read(clock);
+	clock->cycle_last = clock->read(clock);

Minor bug here, the clearing of cycle_last has a side-effect of making sure the TSC doesn't trip over its own cycle_last checking in the read() function. This is part of the uglyness of the TSC pulling this internal timeekping state to avoid minor inconsistencies, but until we find a better way, we have to live with it.

So you'll need to preserve the cycle_last = 0 line.

thanks
-john

#6 lore raw

Re: [patch 11/14] timekeeper read clock helper functions

In reply to a message not in the archive
Quoted 30 lines · Martin Schwidefsky

plain text document attachment (timekeeper-helper.diff)
From: Martin Schwidefsky <[email protected]>

Add timekeeper_read_clock_ntp and timekeeper_read_clock_raw and use them for getnstimeofday, ktime_get, ktime_get_ts and getrawmonotonic.

Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc john stultz <[email protected]>
Cc Daniel Walker <[email protected]>
---
 kernel/time/timekeeping.c |   91 +++++++++++++++++++---------------------------
 1 file changed, 38 insertions(+), 53 deletions(-)

Index: linux-2.6/kernel/time/timekeeping.c
===================================================================

kernel/time/timekeeping.c +7 -0
@@ -95,6 +95,40 @@ static void timekeeper_setup_internals(s
 	timekeeper.mult = clock->mult;
 }
 
+/* Timekeeper helper functions. */
+static inline s64 timekeeping_get_ns(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct clocksource *clock;
+
+	/* read clocksource: */

kernel/time/timekeeping.c static void timekeeper_setup_internals(s · line 114

Quoted 10 lines
kernel/time/timekeeping.c +10 -0
+	clock = timekeeper.clock;
+	cycle_now = clock->read(clock);
+
+	/* calculate the delta since the last update_wall_time: */
+	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+	/* return delta convert to nanoseconds using ntp adjusted mult. */
+	return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
+				  timekeeper.shift);
+}

Again, not a huge issue, but if we kept the read() out of this function and instead passed the cycle_now value in as a argument, we could also use this function in timekeeping_forward_now()

thanks
-john

#8 lore raw

Re: [patch 00/14] clocksource / timekeeping rework V3

In reply to a message not in the archive
Quoted 10 lines · Martin Schwidefsky

Greetings,
version 3 of the clocksource / timekeeping cleanup patches. As far as I'm concerned this is the final version, modulo bugs and review results. All the things I want to have in there are done.

The latest additions are:
1) Change read_persistent_clock to return a struct timespec instead of an unsigned long with the number of seconds.
2) Introduce read_boot_clock to initialize wall_to_monotonic.

Quoted 4 lines

The patch set is based on todays upstream tree plus the patches from the tip tree, if anyone wants to try them you need to pull from the master branch of

    git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-tip

So I got the patch queue up and running and it's functioning correctly on the box I tested it with (although its not a super interesting box:
TSC is unstable and it only has ACPI PM).

Although I've not been able to test changing clocksources, I assume you gave that a whirl? I'll test that as soon as I can get access to a TSC friendly box.

Once the "cycles_last = 0" fix is in to make sure TSC boxes work, I think this set should be ok to go in.

thanks
-john

#3 lore raw
In reply to john stultz
Quoted 20 lines · On Thu, 13 Aug 2009 15:14:34 -0700 john stultz <[email protected]> wrote:
Quoted 4 lines · Martin Schwidefsky

Index: linux-2.6/kernel/time/timekeeping.c
===================================================================

kernel/time/timekeeping.c +0 -0

[snip]

kernel/time/timekeeping.c static int timekeeping_resume(struct sys · line 495

Quoted 7 lines
kernel/time/timekeeping.c +1 -2
@@ -477,8 +492,7 @@ static int timekeeping_resume(struct sys
 	}
 	update_xtime_cache(0);
 	/* re-base the last cycle value */
-	clock->cycle_last = 0;
-	clock->cycle_last = clocksource_read(clock);
+	clock->cycle_last = clock->read(clock);

Minor bug here, the clearing of cycle_last has a side-effect of making sure the TSC doesn't trip over its own cycle_last checking in the read() function. This is part of the uglyness of the TSC pulling this internal timeekping state to avoid minor inconsistencies, but until we find a better way, we have to live with it.

Quoted 1 line

So you'll need to preserve the cycle_last = 0 line.

Whoa, now that is subtle. Good spotting. I would prefer to reset the cycle_last in a resume function though:

arch/x86/kernel/tsc.c +6 -0
Index: linux-2.6/arch/x86/kernel/tsc.c
===================================================================
@@ -744,10 +744,16 @@ static cycle_t __vsyscall_fn vread_tsc(v
 }
 #endif
 
+static void resume_tsc(void)
+{
+       clocksource_tsc.cycle_last = 0;
+}
+
 static struct clocksource clocksource_tsc = {
        .name                   = "tsc",
        .rating                 = 300,
        .read                   = read_tsc,
+       .resume                 = resume_tsc,
        .mask                   = CLOCKSOURCE_MASK(64),
        .shift                  = 22,
        .flags                  = CLOCK_SOURCE_IS_CONTINUOUS |

That puts the subtlety where it belongs.

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.
#4 lore raw
Quoted 29 lines · Martin Schwidefsky
Quoted 1 line · On Thu, 13 Aug 2009 15:14:34 -0700 john stultz <[email protected]> wrote:

So you'll need to preserve the cycle_last = 0 line.

Whoa, now that is subtle. Good spotting. I would prefer to reset the cycle_last in a resume function though:

Index: linux-2.6/arch/x86/kernel/tsc.c
===================================================================

arch/x86/kernel/tsc.c +6 -0
@@ -744,10 +744,16 @@ static cycle_t __vsyscall_fn vread_tsc(v
 }
 #endif
 
+static void resume_tsc(void)
+{
+       clocksource_tsc.cycle_last = 0;
+}
+
 static struct clocksource clocksource_tsc = {
        .name                   = "tsc",
        .rating                 = 300,
        .read                   = read_tsc,
+       .resume                 = resume_tsc,
        .mask                   = CLOCKSOURCE_MASK(64),
        .shift                  = 22,
        .flags                  = CLOCK_SOURCE_IS_CONTINUOUS |
Quoted 1 line

That puts the subtlety where it belongs.

Ack.

tglx

#7 lore raw

Re: [patch 11/14] timekeeper read clock helper functions

In reply to john stultz
Quoted 22 lines · On Thu, 13 Aug 2009 16:30:18 -0700 john stultz <[email protected]> wrote:
Quoted 21 lines · Martin Schwidefsky

Index: linux-2.6/kernel/time/timekeeping.c
===================================================================

kernel/time/timekeeping.c +13 -0
@@ -95,6 +95,40 @@ static void timekeeper_setup_internals(s
 	timekeeper.mult = clock->mult;
 }
 
+/* Timekeeper helper functions. */
+static inline s64 timekeeping_get_ns(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct clocksource *clock;
+
+	/* read clocksource: */
+	clock = timekeeper.clock;
+	cycle_now = clock->read(clock);
+
+	/* calculate the delta since the last update_wall_time: */
+	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
Quoted 8 lines

kernel/time/timekeeping.c static void timekeeper_setup_internals(s · line 114

Quoted 4 lines
kernel/time/timekeeping.c +4 -0
+	/* return delta convert to nanoseconds using ntp adjusted mult. */
+	return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
+				  timekeeper.shift);
+}

Again, not a huge issue, but if we kept the read() out of this function and instead passed the cycle_now value in as a argument, we could also use this function in timekeeping_forward_now()

I actually tested how the code would look like if I do that. Didn't like the result. The thing is that timekeeping_get_ns is a helper that is supposed to reduce the number of lines you need in the caller. If you push clock->read(clock) call back into the caller it doesn't help much ..

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.