perfcounters: provide expansion room in the ABI
RFC PATCH 1/1 perfcounters: provide expansion room in the ABI
Impact: ABI change
This expands several fields in the perf_counter_hw_event struct and adds a "flags" argument to the perf_counter_open system call, in order that features can be added in future without ABI changes.
In particular the record_type field is expanded to 64 bits, and the space for flag bits has been expanded from 32 to 64 bits.
This also adds some new fields:
* read_format (64 bits) is intended to provide a way to specify what userspace wants to get back when it does a read() on a simple (non-interrupting) counter;
* exclude_idle (1 bit) provides a way for userspace to ask that events that occur when the cpu is idle be excluded;
* extra_config_len will provide a way for userspace to supply an arbitrary amount of extra machine-specific PMU configuration data immediately following the perf_counter_hw_event struct, to allow sophisticated users to program things such as instruction matching CAMs and address range registers;
* __reserved_3 and __reserved_4 provide space for future expansion.
---
This should provide plenty of room for future expansion, in fact maybe it's a little over the top.
Ingo, if this looks OK, you can pull it from the rfc branch of my perfcounters.git tree at:
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/perfcounters.git rfc
include/linux/perf_counter.h | 12 +++++++++--- include/linux/syscalls.h | 2 +- kernel/perf_counter.c | 10 +++++++--- 3 files changed, 17 insertions(+), 7 deletions(-)
include/linux/perf_counter.h +9 -3
@@ -69,9 +69,10 @@ struct perf_counter_hw_event {__s64 type;__u64 irq_period; -__u32 record_type; +__u64 record_type; +__u64 read_format; - __u32 disabled : 1, /* off by default */ + __u64 disabled : 1, /* off by default */ nmi : 1, /* NMI sampling */ raw : 1, /* raw event type */ inherit : 1, /* children inherit it */
@@ -80,10 +81,15 @@ struct perf_counter_hw_event { exclude_user : 1, /* don't count user */ exclude_kernel : 1, /* ditto kernel */ exclude_hv : 1, /* ditto hypervisor */ + exclude_idle : 1, /* don't count when idle */ - __reserved_1 : 23; + __reserved_1 : 55; + +__u32 extra_config_len; +__u32 __reserved_4;__u64 __reserved_2; +__u64 __reserved_3; };/*
include/linux/syscalls.h +1 -1
@@ -698,5 +698,5 @@ int kernel_execve(const char *filename, char *const argv[], char *const envp[]);asmlinkage long sys_perf_counter_open (const struct perf_counter_hw_event __user *hw_event_uptr, -pid_t pid,int cpu,int group_fd); +pid_t pid,int cpu,int group_fd,unsigned long flags);#endif
kernel/perf_counter.c +7 -3
@@ -1683,16 +1683,16 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, }/** -* sys_perf_task_open - open a performance counter, associate it to a task/cpu +* sys_perf_counter_open - open a performance counter, associate it to a task/cpu * * @hw_event_uptr: event type attributes for monitoring/sampling * @pid: target pid * @cpu: target cpu * @group_fd: group leader counter fd */ -SYSCALL_DEFINE4 (perf_counter_open, +SYSCALL_DEFINE5 (perf_counter_open, const struct perf_counter_hw_event __user *, hw_event_uptr, - pid_t, pid, int, cpu, int, group_fd) + pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) {struct perf_counter *counter, *group_leader;struct perf_counter_hw_event hw_event;
@@ -1703,6 +1703,10 @@ SYSCALL_DEFINE4(perf_counter_open,int fput_needed2 =0 ;int ret; +/* for future expandability... */ +if (flags) +return -EINVAL ; +if (copy_from_user (&hw_event, hw_event_uptr,sizeof (hw_event)) !=0 )return -EFAULT ;
Signature · 1 line
1.5.6.3
Quoted 26 lines · Paul Mackerras
Impact: ABI change
This expands several fields in the
perf_counter_hw_eventstruct and adds a "flags" argument to theperf_counter_opensystem call, in order that features can be added in future without ABI changes.In particular the
record_typefield is expanded to 64 bits, and the space for flag bits has been expanded from 32 to 64 bits.This also adds some new fields:
*
read_format(64 bits) is intended to provide a way to specify what userspace wants to get back when it does aread()on a simple (non-interrupting) counter;*
exclude_idle(1 bit) provides a way for userspace to ask that events that occur when the cpu is idle be excluded;*
extra_config_lenwill provide a way for userspace to supply an arbitrary amount of extra machine-specific PMU configuration data immediately following theperf_counter_hw_eventstruct, to allow sophisticated users to program things such as instruction matching CAMs and address range registers;
Quoted 6 lines
*
__reserved_3and__reserved_4provide space for future expansion.Signed-off-by Paul Mackerras <[email protected]>---
This should provide plenty of room for future expansion, in fact maybe it's a little over the top.
Yeah - but i think it does not hurt to be defensive when it comes to ABI extensibility.
Your changes look very good to me.
Quoted 9 lines
Ingo, if this looks OK, you can pull it from the rfc branch of my perfcounters.git tree at:
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/perfcounters.gitrfcinclude/linux/perf_counter.h | 12 +++++++++--- include/linux/syscalls.h | 2 +- kernel/perf_counter.c | 10 +++++++--- 3 files changed, 17 insertions(+), 7 deletions(-)
Pulled into tip:perfcounters/core, thanks Paul!
Ingo
Quoted 8 lines · Ingo Molnar
Quoted 6 lines
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/perfcounters.gitrfcinclude/linux/perf_counter.h | 12 +++++++++--- include/linux/syscalls.h | 2 +- kernel/perf_counter.c | 10 +++++++--- 3 files changed, 17 insertions(+), 7 deletions(-)Pulled into tip:perfcounters/core, thanks Paul!
Actually, I goofed slightly. In this bit of the patch:
@@ -80,10 +81,15 @@ struct perf_counter_hw_event {
exclude_user : 1, /* don't count user */ exclude_kernel : 1, /* ditto kernel */ exclude_hv : 1, /* ditto hypervisor */ + exclude_idle : 1, /* don't count when idle */
- __reserved_1 : 23;
+ __reserved_1 : 55;
+
the 55 should have been 54 if we want those bitfields to occupy 64 bits. With 55 there, __reserved_1 will go in the next 64-bit quadword (x86-speak, or doubleword in powerpc-speak).
Do you think that's worth changing? If so I'll do a commit changing the 55 to 54.
Paul.
Quoted 28 lines · Paul Mackerras
Quoted 8 lines · Ingo Molnar
Quoted 6 lines
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/perfcounters.gitrfcinclude/linux/perf_counter.h | 12 +++++++++--- include/linux/syscalls.h | 2 +- kernel/perf_counter.c | 10 +++++++--- 3 files changed, 17 insertions(+), 7 deletions(-)Pulled into tip:perfcounters/core, thanks Paul!
Actually, I goofed slightly. In this bit of the patch:
include/linux/perf_counter.h +3
-1@@ -80,10 +81,15 @@ struct perf_counter_hw_event { exclude_user : 1, /* don't count user */ exclude_kernel : 1, /* ditto kernel */ exclude_hv : 1, /* ditto hypervisor */ + exclude_idle : 1, /* don't count when idle */ - __reserved_1 : 23; + __reserved_1 : 55; +the 55 should have been 54 if we want those bitfields to occupy 64 bits. With 55 there,
__reserved_1will go in the next 64-bit quadword (x86-speak, or doubleword in powerpc-speak).
Quoted 2 lines
Do you think that's worth changing? If so I'll do a commit changing the 55 to 54.
How about extending it to 118 instead? The creation of a counter is a slow path in any case.
Ingo