linux-next: manual merge of the cputime tree

linux-next: manual merge of the cputime tree

[Stephen Rothwell]

[-- Attachment #1: Type: text/plain, Size: 645 bytes --]

Hi Martin,

Today's linux-next merge of the cputime tree got a conflict in
kernel/sched.c between commit 74fcd524e808975dd546dac847119f1995a7c622
("account_steal_time: kill the unneeded account_group_system_time()")
from the tip-core tree and commit
b7f4776b7f575ed8f288c44b64befd241fd44458 ("[PATCH] idle cputime
accounting") from the cputime tree.

The latter removes the call to account_group_system_time() as a side
effect of further changes.  So the fixup is to just take the latter
change.  I can carry the merge fix.

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: linux-next: manual merge of the cputime tree

[Peter Zijlstra]

On Tue, 2008-11-25 at 20:21 +1100, Stephen Rothwell wrote:
> Hi Martin,
> 
> Today's linux-next merge of the cputime tree got a conflict in
> kernel/sched.c between commit 74fcd524e808975dd546dac847119f1995a7c622
> ("account_steal_time: kill the unneeded account_group_system_time()")
> from the tip-core tree and commit
> b7f4776b7f575ed8f288c44b64befd241fd44458 ("[PATCH] idle cputime
> accounting") from the cputime tree.
> 
> The latter removes the call to account_group_system_time() as a side
> effect of further changes.  So the fixup is to just take the latter
> change.  I can carry the merge fix.

Martin,

Why does s390 do its own cpu accounting?

Re: linux-next: manual merge of the cputime tree

[Martin Schwidefsky]

On Tue, 2008-11-25 at 11:06 +0100, Peter Zijlstra wrote:
> On Tue, 2008-11-25 at 20:21 +1100, Stephen Rothwell wrote:
> > Hi Martin,
> > 
> > Today's linux-next merge of the cputime tree got a conflict in
> > kernel/sched.c between commit 74fcd524e808975dd546dac847119f1995a7c622
> > ("account_steal_time: kill the unneeded account_group_system_time()")
> > from the tip-core tree and commit
> > b7f4776b7f575ed8f288c44b64befd241fd44458 ("[PATCH] idle cputime
> > accounting") from the cputime tree.
> > 
> > The latter removes the call to account_group_system_time() as a side
> > effect of further changes.  So the fixup is to just take the latter
> > change.  I can carry the merge fix.
> 
> Why does s390 do its own cpu accounting?

Is that a trick question? I invented the cputime accounting specifically
for s390 because it is a virtual architecture and the standard cpu
accounting numbers are just useless.

-- 
blue skies,
  Martin.

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

Re: linux-next: manual merge of the cputime tree

[Peter Zijlstra]

On Tue, 2008-11-25 at 11:15 +0100, Martin Schwidefsky wrote:
> On Tue, 2008-11-25 at 11:06 +0100, Peter Zijlstra wrote:
> > On Tue, 2008-11-25 at 20:21 +1100, Stephen Rothwell wrote:
> > > Hi Martin,
> > > 
> > > Today's linux-next merge of the cputime tree got a conflict in
> > > kernel/sched.c between commit 74fcd524e808975dd546dac847119f1995a7c622
> > > ("account_steal_time: kill the unneeded account_group_system_time()")
> > > from the tip-core tree and commit
> > > b7f4776b7f575ed8f288c44b64befd241fd44458 ("[PATCH] idle cputime
> > > accounting") from the cputime tree.
> > > 
> > > The latter removes the call to account_group_system_time() as a side
> > > effect of further changes.  So the fixup is to just take the latter
> > > change.  I can carry the merge fix.
> > 
> > Why does s390 do its own cpu accounting?
> 
> Is that a trick question? I invented the cputime accounting specifically
> for s390 because it is a virtual architecture and the standard cpu
> accounting numbers are just useless.

Nah, just general ignorance brought on by the mondays on a tuesday ;-)

Right, makes sense, is it shared with all other virt* folks out there?

Re: linux-next: manual merge of the cputime tree

[Paul Mackerras]

Peter Zijlstra writes:

> Right, makes sense, is it shared with all other virt* folks out there?

PowerPC uses it too.

Paul.

Re: linux-next: manual merge of the cputime tree

[Martin Schwidefsky]

On Tue, 2008-11-25 at 11:28 +0100, Peter Zijlstra wrote:
> On Tue, 2008-11-25 at 11:15 +0100, Martin Schwidefsky wrote:
> > On Tue, 2008-11-25 at 11:06 +0100, Peter Zijlstra wrote:
> > > Why does s390 do its own cpu accounting?
> > 
> > Is that a trick question? I invented the cputime accounting specifically
> > for s390 because it is a virtual architecture and the standard cpu
> > accounting numbers are just useless.
> 
> Nah, just general ignorance brought on by the mondays on a tuesday ;-)

Ah, I usually have this on mondays ..

> Right, makes sense, is it shared with all other virt* folks out there?

Yes, the other virtual architectures have started to use the interface
as well.

-- 
blue skies,
  Martin.

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

linux-next: manual merge of the cputime tree

[Stephen Rothwell]

Hi Martin,

Today's linux-next merge of the cputime tree got a conflict in
kernel/time/tick-sched.c between commit
26105f6432b27212ad5563c4d32826f7b778f0b8 ("tick: cleanup idle checks")
from the timers tree and commit 84dd5d04573c132ff107cf9d5d311d93c6ac6b61
("[PATCH] idle cputime accounting") from the cputime tree.

Just context.  I fixed it up (see below) and can carry it for now.
-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

diff --cc kernel/time/tick-sched.c
index 35a775e,1bcbca6..0000000
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@@ -410,11 -412,13 +410,13 @@@ void tick_nohz_restart_sched_tick(void
  {
  	int cpu = smp_processor_id();
  	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
+ #ifndef CONFIG_VIRT_CPU_ACCOUNTING
  	unsigned long ticks;
+ #endif
 -	ktime_t now;
 +	ktime_t now = ktime_get();
  
  	local_irq_disable();
 -	tick_nohz_stop_idle(cpu);
 +	tick_nohz_stop_idle(ts, now);
  
  	if (!ts->inidle || !ts->tick_stopped) {
  		ts->inidle = 0;

linux-next: manual merge of the cputime tree

[Stephen Rothwell]

Hi Martin,

Today's linux-next merge of the cputime tree got a conflict in
arch/s390/kernel/process.c between commit
2933aa2758dc5e520f83a5f4c3f255aea297bd3f ("[S390] ftrace: disable tracing
on idle psw") from the s390 tree and commit
bbda3e73a200e966955b15a08a8be05a7c1d430e ("[PATCH] improve precision of
idle accounting") from the cputime tree.

I attempted a fix (see below) but please either provide me with a better
patch or fix this conflict in one or the other tree.  You could do this
by merging the s390 tree into the cputime tree (but you would need to be
careful to keep the consistent) and fixing this in the merge commit.

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

diff --cc arch/s390/kernel/process.c
index 04f8c67,9334ca3..0000000
--- a/arch/s390/kernel/process.c
+++ b/arch/s390/kernel/process.c
@@@ -136,12 -133,42 +133,45 @@@ static void default_idle(void
  		return;
  	}
  	trace_hardirqs_on();
 +	/* Don't trace preempt off for idle. */
 +	stop_critical_timings();
+ 	vtime_stop_cpu_timer();
+ 
+ 	/*
+ 	 * The inline assembly is equivalent to
+ 	 * 	idle->idle_enter = get_clock();
+ 	 * 	__load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
+ 	 *			   PSW_MASK_IO | PSW_MASK_EXT);
+ 	 * The difference is that the inline assembly makes sure that
+ 	 * the stck instruction is right before the lpsw instruction.
+ 	 * This is done to increase the precision.
+ 	 */
+ 
  	/* Wait for external, I/O or machine check interrupt. */
- 	__load_psw_mask(psw_kernel_bits | PSW_MASK_WAIT |
- 			PSW_MASK_IO | PSW_MASK_EXT);
+ 	psw.mask = psw_kernel_bits|PSW_MASK_WAIT|PSW_MASK_IO|PSW_MASK_EXT;
+ #ifndef __s390x__
+ 	asm volatile(
+ 		"	basr	%0,0\n"
+ 		"0:	ahi	%0,1f-0b\n"
+ 		"	st	%0,4(%2)\n"
+ 		"	stck	0(%3)\n"
+ 		"	lpsw	0(%2)\n"
+ 		"1:"
+ 		: "=&d" (addr), "=m" (idle->idle_enter)
+ 		: "a" (&psw), "a" (&idle->idle_enter), "m" (psw)
+ 		: "memory", "cc");
+ #else /* __s390x__ */
+ 	asm volatile(
+ 		"	larl	%0,1f\n"
+ 		"	stg	%0,8(%2)\n"
+ 		"	stck	0(%3)\n"
+ 		"	lpswe	0(%2)\n"
+ 		"1:"
+ 		: "=&d" (addr), "=m" (idle->idle_enter)
+ 		: "a" (&psw), "a" (&idle->idle_enter), "m" (psw)
+ 		: "memory", "cc");
+ #endif /* __s390x__ */
 +	start_critical_timings();
  }
  
  void cpu_idle(void)

Re: linux-next: manual merge of the cputime tree

[Martin Schwidefsky]

Hi Stephen,

On Mon, 2008-11-10 at 17:09 +1100, Stephen Rothwell wrote:
> Today's linux-next merge of the cputime tree got a conflict in
> arch/s390/kernel/process.c between commit
> 2933aa2758dc5e520f83a5f4c3f255aea297bd3f ("[S390] ftrace: disable tracing
> on idle psw") from the s390 tree and commit
> bbda3e73a200e966955b15a08a8be05a7c1d430e ("[PATCH] improve precision of
> idle accounting") from the cputime tree.
> 
> I attempted a fix (see below) but please either provide me with a better
> patch or fix this conflict in one or the other tree.  You could do this
> by merging the s390 tree into the cputime tree (but you would need to be
> careful to keep the consistent) and fixing this in the merge commit.

You are pulling the s390 tree first and then the cputime tree, no? Then
I could simply rebase the cputime tree off the s390 tree. You'll find
new features/cputime branches on git390.

-- 
blue skies,
  Martin.

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