* [PATCH] fix stolen time for SMT without LPAR
@ 2007-05-24 13:33 Michael Neuling
2007-05-25 5:08 ` Michael Neuling
0 siblings, 1 reply; 4+ messages in thread
From: Michael Neuling @ 2007-05-24 13:33 UTC (permalink / raw)
To: paulus; +Cc: linuxppc-dev, anton
For POWERPC, stolen time accounts for cycles lost to the hypervisor or
PURR cycles attributed to the other SMT thread. Hence, when a PURR is
available, we should still calculate stolen time, irrespective of being
virtualised.
Signed-off-by: Michael Neuling <mikey@neuling.org>
---
Kudos to sfr for realizing the problem here.
arch/powerpc/kernel/time.c | 6 ++++--
include/asm-powerpc/time.h | 2 +-
2 files changed, 5 insertions(+), 3 deletions(-)
Index: linux-2.6-ozlabs/arch/powerpc/kernel/time.c
===================================================================
--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/time.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/time.c
@@ -244,7 +244,8 @@ void snapshot_timebases(void)
{
int cpu;
- if (!cpu_has_feature(CPU_FTR_PURR))
+ if (!cpu_has_feature(CPU_FTR_PURR) &&
+ !firmware_has_feature(FW_FEATURE_SPLPAR))
return;
for_each_possible_cpu(cpu)
spin_lock_init(&per_cpu(cpu_purr_data, cpu).lock);
@@ -257,7 +258,8 @@ void calculate_steal_time(void)
s64 stolen;
struct cpu_purr_data *pme;
- if (!cpu_has_feature(CPU_FTR_PURR))
+ if (!cpu_has_feature(CPU_FTR_PURR) &&
+ !firmware_has_feature(FW_FEATURE_SPLPAR))
return;
pme = &per_cpu(cpu_purr_data, smp_processor_id());
if (!pme->initialized)
Index: linux-2.6-ozlabs/include/asm-powerpc/time.h
===================================================================
--- linux-2.6-ozlabs.orig/include/asm-powerpc/time.h
+++ linux-2.6-ozlabs/include/asm-powerpc/time.h
@@ -232,7 +232,7 @@ extern void account_process_vtime(struct
#define account_process_vtime(tsk) do { } while (0)
#endif
-#if defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)
+#if defined(CONFIG_VIRT_CPU_ACCOUNTING)
extern void calculate_steal_time(void);
extern void snapshot_timebases(void);
#else
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] fix stolen time for SMT without LPAR
2007-05-24 13:33 [PATCH] fix stolen time for SMT without LPAR Michael Neuling
@ 2007-05-25 5:08 ` Michael Neuling
2007-06-07 1:33 ` Paul Mackerras
0 siblings, 1 reply; 4+ messages in thread
From: Michael Neuling @ 2007-05-25 5:08 UTC (permalink / raw)
To: paulus; +Cc: linuxppc-dev
For POWERPC, stolen time accounts for cycles lost to the hypervisor or
PURR cycles attributed to the other SMT thread. Hence, when a PURR is
available, we should still calculate stolen time, irrespective of being
virtualised.
Signed-off-by: Michael Neuling <mikey@neuling.org>
---
Updated to fix link error when CONFIG_PPC_SPLPAR is off.
arch/powerpc/kernel/time.c | 8 +++++---
include/asm-powerpc/time.h | 2 +-
2 files changed, 6 insertions(+), 4 deletions(-)
Index: linux-2.6-ozlabs/arch/powerpc/kernel/time.c
===================================================================
--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/time.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/time.c
@@ -214,7 +214,6 @@ static void account_process_time(struct
run_posix_cpu_timers(current);
}
-#ifdef CONFIG_PPC_SPLPAR
/*
* Stuff for accounting stolen time.
*/
@@ -244,7 +243,8 @@ void snapshot_timebases(void)
{
int cpu;
- if (!cpu_has_feature(CPU_FTR_PURR))
+ if (!cpu_has_feature(CPU_FTR_PURR) &&
+ !firmware_has_feature(FW_FEATURE_SPLPAR))
return;
for_each_possible_cpu(cpu)
spin_lock_init(&per_cpu(cpu_purr_data, cpu).lock);
@@ -257,7 +257,8 @@ void calculate_steal_time(void)
s64 stolen;
struct cpu_purr_data *pme;
- if (!cpu_has_feature(CPU_FTR_PURR))
+ if (!cpu_has_feature(CPU_FTR_PURR) &&
+ !firmware_has_feature(FW_FEATURE_SPLPAR))
return;
pme = &per_cpu(cpu_purr_data, smp_processor_id());
if (!pme->initialized)
@@ -273,6 +274,7 @@ void calculate_steal_time(void)
spin_unlock(&pme->lock);
}
+#ifdef CONFIG_PPC_SPLPAR
/*
* Must be called before the cpu is added to the online map when
* a cpu is being brought up at runtime.
Index: linux-2.6-ozlabs/include/asm-powerpc/time.h
===================================================================
--- linux-2.6-ozlabs.orig/include/asm-powerpc/time.h
+++ linux-2.6-ozlabs/include/asm-powerpc/time.h
@@ -232,7 +232,7 @@ extern void account_process_vtime(struct
#define account_process_vtime(tsk) do { } while (0)
#endif
-#if defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)
+#if defined(CONFIG_VIRT_CPU_ACCOUNTING)
extern void calculate_steal_time(void);
extern void snapshot_timebases(void);
#else
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] fix stolen time for SMT without LPAR
2007-05-25 5:08 ` Michael Neuling
@ 2007-06-07 1:33 ` Paul Mackerras
2007-06-08 3:18 ` Michael Neuling
0 siblings, 1 reply; 4+ messages in thread
From: Paul Mackerras @ 2007-06-07 1:33 UTC (permalink / raw)
To: Michael Neuling; +Cc: linuxppc-dev
Michael Neuling writes:
> - if (!cpu_has_feature(CPU_FTR_PURR))
> + if (!cpu_has_feature(CPU_FTR_PURR) &&
> + !firmware_has_feature(FW_FEATURE_SPLPAR))
> return;
> for_each_possible_cpu(cpu)
> spin_lock_init(&per_cpu(cpu_purr_data, cpu).lock);
This means that on a machine with SPLPAR but not a PURR, e.g. a JS21,
we'll try to read the PURR register and do the stolen time calculation
with whatever value was in the register before the mfspr from the
PURR. If we're going to make this change then we also need to add
code to do the H_PURR hcall instead of reading the PURR directly on
such machines.
Paul.
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH] fix stolen time for SMT without LPAR
2007-06-07 1:33 ` Paul Mackerras
@ 2007-06-08 3:18 ` Michael Neuling
0 siblings, 0 replies; 4+ messages in thread
From: Michael Neuling @ 2007-06-08 3:18 UTC (permalink / raw)
To: Paul Mackerras; +Cc: linuxppc-dev
For POWERPC, stolen time accounts for cycles lost to the hypervisor or
PURR cycles attributed to the other SMT thread. Hence, when a PURR is
available, we should still calculate stolen time, irrespective of being
virtualised.
Signed-off-by: Michael Neuling <mikey@neuling.org>
---
> This means that on a machine with SPLPAR but not a PURR, e.g. a JS21,
> we'll try to read the PURR register and do the stolen time calculation
> with whatever value was in the register before the mfspr from the
> PURR. If we're going to make this change then we also need to add
> code to do the H_PURR hcall instead of reading the PURR directly on
> such machines.
This version bails on reading the PURR if it doesn't exist, so it
doesn't make JS21 any worse. We still need to fix JS21, but our initial
patches using H_PURR showed some lmbench crappy-ness.
Tested on JS21 and BML. Like you said offline, this can be 2.6.23.
arch/powerpc/kernel/time.c | 2 +-
include/asm-powerpc/time.h | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
Index: linux-2.6-ozlabs/arch/powerpc/kernel/time.c
===================================================================
--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/time.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/time.c
@@ -214,7 +214,6 @@ static void account_process_time(struct
run_posix_cpu_timers(current);
}
-#ifdef CONFIG_PPC_SPLPAR
/*
* Stuff for accounting stolen time.
*/
@@ -273,6 +272,7 @@ void calculate_steal_time(void)
spin_unlock(&pme->lock);
}
+#ifdef CONFIG_PPC_SPLPAR
/*
* Must be called before the cpu is added to the online map when
* a cpu is being brought up at runtime.
Index: linux-2.6-ozlabs/include/asm-powerpc/time.h
===================================================================
--- linux-2.6-ozlabs.orig/include/asm-powerpc/time.h
+++ linux-2.6-ozlabs/include/asm-powerpc/time.h
@@ -232,7 +232,7 @@ extern void account_process_vtime(struct
#define account_process_vtime(tsk) do { } while (0)
#endif
-#if defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)
+#if defined(CONFIG_VIRT_CPU_ACCOUNTING)
extern void calculate_steal_time(void);
extern void snapshot_timebases(void);
#else
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2007-06-08 3:18 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-05-24 13:33 [PATCH] fix stolen time for SMT without LPAR Michael Neuling
2007-05-25 5:08 ` Michael Neuling
2007-06-07 1:33 ` Paul Mackerras
2007-06-08 3:18 ` Michael Neuling
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).