[PATCH 0/6] powerpc: SMT priority (PPR) save and restore
On P7 systems, users can define SMT priority levels 2,3 and 4 for
processes so that some can run higher priority than the other ones.
In the current kernel, the default priority is set to 4 which prohibits
processes for using higher priority. Also the kernel boosts the priority to
4 during exceptions without saving the user defined priorities when
the task enters the kernel. So we will be loosing the process PPR value
and can not be restored it back when the task exits the kernel.
This patchset implements saving and restore the user defined PPR value
for all tasks.
With null_syscall testcase (http://ozlabs.org/~anton/junkcode/null_syscall.c),
this feature takes around extra 10 CPU cycles on average for 25 samples.
Haren Myneni (6):
powerpc: Move branch instruction from ACCOUNT_CPU_USER_ENTRY to caller
powerpc: Define CPU_FTR_HAS_PPR
powerpc: Increase exceptions arrays in paca struct to save PPR
powerpc: Define ppr in thread_struct
powerpc: Macros for saving/restore PPR
powerpc: Implement PPR save/restore
arch/powerpc/include/asm/cputable.h | 6 ++-
arch/powerpc/include/asm/exception-64s.h | 45 ++++++++++++++++++++++++++----
arch/powerpc/include/asm/paca.h | 6 ++--
arch/powerpc/include/asm/ppc_asm.h | 27 ++++++++++++++++-
arch/powerpc/include/asm/processor.h | 6 ++++
arch/powerpc/include/asm/reg.h | 1 +
arch/powerpc/kernel/asm-offsets.c | 1 +
arch/powerpc/kernel/entry_64.S | 6 +++-
arch/powerpc/kernel/exceptions-64e.S | 3 +-
arch/powerpc/kernel/exceptions-64s.S | 15 +++++----
arch/powerpc/kernel/process.c | 2 +
11 files changed, 96 insertions(+), 22 deletions(-)
Haren Myneni [off-list ref] wrote:
[PATCH 0/6] powerpc: SMT priority (PPR) save and restore
On P7 systems, users can define SMT priority levels 2,3 and 4 for
processes so that some can run higher priority than the other ones.
In the current kernel, the default priority is set to 4 which prohibits
processes for using higher priority. Also the kernel boosts the priority to
4 during exceptions without saving the user defined priorities when
the task enters the kernel. So we will be loosing the process PPR value
and can not be restored it back when the task exits the kernel.
This patchset implements saving and restore the user defined PPR value
for all tasks.
With null_syscall testcase (http://ozlabs.org/~anton/junkcode/null_syscall.c),
this feature takes around extra 10 CPU cycles on average for 25 samples.
Haren,
This version doesn't do PPR save/restore over syscall, so is this above
statement still true? null_syscall shouldn't be effected at all, right?
Mikey
Haren Myneni (6):
powerpc: Move branch instruction from ACCOUNT_CPU_USER_ENTRY to caller
powerpc: Define CPU_FTR_HAS_PPR
powerpc: Increase exceptions arrays in paca struct to save PPR
powerpc: Define ppr in thread_struct
powerpc: Macros for saving/restore PPR
powerpc: Implement PPR save/restore
arch/powerpc/include/asm/cputable.h | 6 ++-
arch/powerpc/include/asm/exception-64s.h | 45 ++++++++++++++++++++++++++----
arch/powerpc/include/asm/paca.h | 6 ++--
arch/powerpc/include/asm/ppc_asm.h | 27 ++++++++++++++++-
arch/powerpc/include/asm/processor.h | 6 ++++
arch/powerpc/include/asm/reg.h | 1 +
arch/powerpc/kernel/asm-offsets.c | 1 +
arch/powerpc/kernel/entry_64.S | 6 +++-
arch/powerpc/kernel/exceptions-64e.S | 3 +-
arch/powerpc/kernel/exceptions-64s.S | 15 +++++----
arch/powerpc/kernel/process.c | 2 +
11 files changed, 96 insertions(+), 22 deletions(-)
On 10/31/2012 05:44 PM, Michael Neuling wrote:
Haren Myneni [off-list ref] wrote:
quoted
[PATCH 0/6] powerpc: SMT priority (PPR) save and restore
On P7 systems, users can define SMT priority levels 2,3 and 4 for
processes so that some can run higher priority than the other ones.
In the current kernel, the default priority is set to 4 which prohibits
processes for using higher priority. Also the kernel boosts the priority to
4 during exceptions without saving the user defined priorities when
the task enters the kernel. So we will be loosing the process PPR value
and can not be restored it back when the task exits the kernel.
This patchset implements saving and restore the user defined PPR value
for all tasks.
With null_syscall testcase (http://ozlabs.org/~anton/junkcode/null_syscall.c),
this feature takes around extra 10 CPU cycles on average for 25 samples.
Haren,
This version doesn't do PPR save/restore over syscall, so is this above
statement still true? null_syscall shouldn't be effected at all, right?
Mikey
Mikey,
Yes this version does not save/ restore for syscall entries. But we set
default priority 3 for each syscall exit - set HMT_MEDIUM_LOW for
normal return, and set 3 in thread.ppr so that this value will be set
for fast return path.
So I was using null_syscall test case to find out the performance
overhead with this patch-set.
Thanks
Haren
quoted
Haren Myneni (6):
powerpc: Move branch instruction from ACCOUNT_CPU_USER_ENTRY to caller
powerpc: Define CPU_FTR_HAS_PPR
powerpc: Increase exceptions arrays in paca struct to save PPR
powerpc: Define ppr in thread_struct
powerpc: Macros for saving/restore PPR
powerpc: Implement PPR save/restore
arch/powerpc/include/asm/cputable.h | 6 ++-
arch/powerpc/include/asm/exception-64s.h | 45 ++++++++++++++++++++++++++----
arch/powerpc/include/asm/paca.h | 6 ++--
arch/powerpc/include/asm/ppc_asm.h | 27 ++++++++++++++++-
arch/powerpc/include/asm/processor.h | 6 ++++
arch/powerpc/include/asm/reg.h | 1 +
arch/powerpc/kernel/asm-offsets.c | 1 +
arch/powerpc/kernel/entry_64.S | 6 +++-
arch/powerpc/kernel/exceptions-64e.S | 3 +-
arch/powerpc/kernel/exceptions-64s.S | 15 +++++----
arch/powerpc/kernel/process.c | 2 +
11 files changed, 96 insertions(+), 22 deletions(-)
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev