| Message ID | 1543230756-15319-1-git-send-email-andrew.murray@arm.com |
|---|---|
| Headers | show |
| Series | perf/core: Generalise event exclusion checking | expand |
Hi Andrew, On 26/11/2018 11:12, Andrew Murray wrote: > Many PMU drivers do not have the capability to exclude counting events > that occur in specific contexts such as idle, kernel, guest, etc. These > drivers indicate this by returning an error in their event_init upon > testing the events attribute flags. This approach is error prone and > often inconsistent. > > Let's instead allow PMU drivers to advertise their ability to exclude > based on context via a new capability: PERF_PMU_CAP_EXCLUDE. This > allows the perf core to reject requests for exclusion events where > there is no support in the PMU. > > Signed-off-by: Andrew Murray <andrew.murray@arm.com> > --- > include/linux/perf_event.h | 1 + > kernel/events/core.c | 9 +++++++++ > 2 files changed, 10 insertions(+) > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h > index b2e806f..69b3d65 100644 > --- a/include/linux/perf_event.h > +++ b/include/linux/perf_event.h > @@ -244,6 +244,7 @@ struct perf_event; > #define PERF_PMU_CAP_EXCLUSIVE 0x10 > #define PERF_PMU_CAP_ITRACE 0x20 > #define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40 > +#define PERF_PMU_CAP_EXCLUDE 0x80 > > /** > * struct pmu - generic performance monitoring unit > diff --git a/kernel/events/core.c b/kernel/events/core.c > index 5a97f34..9afb33c 100644 > --- a/kernel/events/core.c > +++ b/kernel/events/core.c > @@ -9743,6 +9743,15 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event) > if (ctx) > perf_event_ctx_unlock(event->group_leader, ctx); > > + if (!ret) { > + if (!(pmu->capabilities & PERF_PMU_CAP_EXCLUDE) && > + event_has_any_exclude_flag(event)) { Technically this is a bisection-breaker, since no driver has this capability yet - ideally, this patch should come after all the ones introducing it to the relevant drivers (with the removal of the now-redundant code from the other drivers at the end). Alternatively, since we already have several other negative capabilities, unless there's a strong feeling against adding any more then it might work out simpler to flip it to PERF_PMU_CAP_NO_EXCLUDE, such that we only need to introduce the core check then directly replace the open-coded event checks with the capability in the appropriate drivers, and need not touch the exclusion-supporting ones at all. Robin. > + if (event->destroy) > + event->destroy(event); > + ret = -EINVAL; > + } > + } > + > if (ret) > module_put(pmu->module); > >
On Mon, Nov 26, 2018 at 02:10:24PM +0000, Robin Murphy wrote: > Hi Andrew, > > On 26/11/2018 11:12, Andrew Murray wrote: > > Many PMU drivers do not have the capability to exclude counting events > > that occur in specific contexts such as idle, kernel, guest, etc. These > > drivers indicate this by returning an error in their event_init upon > > testing the events attribute flags. This approach is error prone and > > often inconsistent. > > > > Let's instead allow PMU drivers to advertise their ability to exclude > > based on context via a new capability: PERF_PMU_CAP_EXCLUDE. This > > allows the perf core to reject requests for exclusion events where > > there is no support in the PMU. > > > > Signed-off-by: Andrew Murray <andrew.murray@arm.com> > > --- > > include/linux/perf_event.h | 1 + > > kernel/events/core.c | 9 +++++++++ > > 2 files changed, 10 insertions(+) > > > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h > > index b2e806f..69b3d65 100644 > > --- a/include/linux/perf_event.h > > +++ b/include/linux/perf_event.h > > @@ -244,6 +244,7 @@ struct perf_event; > > #define PERF_PMU_CAP_EXCLUSIVE 0x10 > > #define PERF_PMU_CAP_ITRACE 0x20 > > #define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40 > > +#define PERF_PMU_CAP_EXCLUDE 0x80 > > /** > > * struct pmu - generic performance monitoring unit > > diff --git a/kernel/events/core.c b/kernel/events/core.c > > index 5a97f34..9afb33c 100644 > > --- a/kernel/events/core.c > > +++ b/kernel/events/core.c > > @@ -9743,6 +9743,15 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event) > > if (ctx) > > perf_event_ctx_unlock(event->group_leader, ctx); > > + if (!ret) { > > + if (!(pmu->capabilities & PERF_PMU_CAP_EXCLUDE) && > > + event_has_any_exclude_flag(event)) { > > Technically this is a bisection-breaker, since no driver has this capability > yet - ideally, this patch should come after all the ones introducing it to > the relevant drivers (with the removal of the now-redundant code from the > other drivers at the end). Indeed. Thought it is possible to first introduce the capability, update the relevant drivers to advertise it, then add the change to core.c and finally remove the unnecessary error checks as a result of using the new capability. This approach could be bisection-proof. > > Alternatively, since we already have several other negative capabilities, > unless there's a strong feeling against adding any more then it might work > out simpler to flip it to PERF_PMU_CAP_NO_EXCLUDE, such that we only need to > introduce the core check then directly replace the open-coded event checks > with the capability in the appropriate drivers, and need not touch the > exclusion-supporting ones at all. This would certaintly be less risky and invasive (e.g. compare the number of files touched between this v2 and the previous v1). I'm happy with either approach. Thanks, Andrew Murray > > Robin. > > > + if (event->destroy) > > + event->destroy(event); > > + ret = -EINVAL; > > + } > > + } > > + > > if (ret) > > module_put(pmu->module); > >
From: Andrew Murray <andrew.murray@arm.com> Date: Mon, 26 Nov 2018 11:12:32 +0000 > The SPARC PMU has the capability to exclude events based on context > - let's advertise that we support the PERF_PMU_CAP_EXCLUDE > capability to ensure that perf doesn't prevent us from handling > events where any exclusion flags are set. > > Signed-off-by: Andrew Murray <andrew.murray@arm.com> Acked-by: David S. Miller <davem@davemloft.net>
Hi Andrew, On 26/11/2018 11:12, Andrew Murray wrote: > Update design.txt to reflect the presence of the exclude_host > and exclude_guest perf flags. > > Signed-off-by: Andrew Murray <andrew.murray@arm.com> Thanks a lot for adding this ! > --- > tools/perf/design.txt | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/tools/perf/design.txt b/tools/perf/design.txt > index a28dca2..5b2b23b 100644 > --- a/tools/perf/design.txt > +++ b/tools/perf/design.txt > @@ -222,6 +222,10 @@ The 'exclude_user', 'exclude_kernel' and 'exclude_hv' bits provide a > way to request that counting of events be restricted to times when the > CPU is in user, kernel and/or hypervisor mode. > > +Furthermore the 'exclude_host' and 'exclude_guest' bits provide a way > +to request counting of events restricted to guest and host contexts when > +using KVM virtualisation. minor nit: could we generalise this to : "using Linux as the hypervisor". Otherwise, looks good to me. Cheers Suzuki
On Mon, Nov 26, 2018 at 11:12:31AM +0000, Andrew Murray wrote: > The s390 cpum_cf and cpum_sf PMUs have the capability to exclude > events based on context. Let's advertise that we support the > PERF_PMU_CAP_EXCLUDE capability to ensure that perf doesn't > prevent us from handling events where any exclusion flags are set. > > Signed-off-by: Andrew Murray <andrew.murray@arm.com> > --- > arch/s390/kernel/perf_cpum_cf.c | 1 + > arch/s390/kernel/perf_cpum_sf.c | 2 ++ > 2 files changed, 3 insertions(+) Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Many PMU drivers do not have the capability to exclude counting events that occur in specific contexts such as idle, kernel, guest, etc. These drivers indicate this by returning an error in their event_init upon testing the events attribute flags. However this approach requires that each time a new event modifier is added to perf, all the perf drivers need to be modified to indicate that they don't support the attribute. This results in additional boiler-plate code common to many drivers that needs to be maintained. Furthermore the drivers are not consistent with regards to the error value they return when reporting unsupported attributes. This patchset allow PMU drivers to advertise their ability to exclude based on context via a new capability: PERF_PMU_CAP_EXCLUDE. This allows the perf core to reject requests for exclusion events where there is no support in the PMU. This is a functional change, in particular: - Some drivers will now additionally (but correctly) report unsupported exclusion flags. It's typical for existing userspace tools such as perf to handle such errors by retrying the system call without the unsupported flags. - Drivers that do not support any exclusion that previously reported -EPERM or -EOPNOTSUPP will now report -EINVAL - this is consistent with the majority and results in userspace perf retrying without exclusion. All drivers touched by this patchset have been compile tested. Changes from v1: - Changed approach from explicitly rejecting events in unsupporting PMU drivers to explicitly advertising a capability in PMU drivers that do support exclusion events - Added additional information to tools/perf/design.txt - Rename event_has_exclude_flags to event_has_any_exclude_flag and update commit log to reflect it's a function Andrew Murray (20): perf/doc: update design.txt for exclude_{host|guest} flags perf/core: add function to test for event exclusion flags perf/core: add PERF_PMU_CAP_EXCLUDE for exclusion capable PMUs perf/hw_breakpoint: perf/core: advertise PMU exclusion capability alpha: perf/core: remove unnecessary checks for exclusion arc: perf/core: advertise PMU exclusion capability arm: perf: conditionally advertise PMU exclusion capability arm: perf/core: remove unnecessary checks for exclusion drivers/perf: perf/core: remove unnecessary checks for exclusion drivers/perf: perf/core: remove unnecessary checks for exclusion drivers/perf: perf/core: advertise PMU exclusion capability mips: perf/core: advertise PMU exclusion capability powerpc: perf/core: advertise PMU exclusion capability powerpc: perf/core: remove unnecessary checks for exclusion s390: perf/events: advertise PMU exclusion capability sparc: perf/core: advertise PMU exclusion capability x86: perf/core: remove unnecessary checks for exclusion x86: perf/core remove unnecessary checks for exclusion x86: perf/core: advertise PMU exclusion capability perf/core: remove unused perf_flags arch/alpha/kernel/perf_event.c | 6 ------ arch/arc/kernel/perf_event.c | 1 + arch/arm/mach-imx/mmdc.c | 8 +------- arch/arm/mm/cache-l2x0-pmu.c | 8 -------- arch/mips/kernel/perf_event_mipsxx.c | 1 + arch/powerpc/perf/core-book3s.c | 1 + arch/powerpc/perf/core-fsl-emb.c | 1 + arch/powerpc/perf/hv-24x7.c | 9 --------- arch/powerpc/perf/hv-gpci.c | 9 --------- arch/powerpc/perf/imc-pmu.c | 18 ------------------ arch/s390/kernel/perf_cpum_cf.c | 1 + arch/s390/kernel/perf_cpum_sf.c | 2 ++ arch/sparc/kernel/perf_event.c | 1 + arch/x86/events/amd/ibs.c | 12 ------------ arch/x86/events/amd/iommu.c | 5 ----- arch/x86/events/amd/power.c | 9 +-------- arch/x86/events/amd/uncore.c | 5 ----- arch/x86/events/core.c | 2 ++ arch/x86/events/intel/bts.c | 2 +- arch/x86/events/intel/cstate.c | 8 +------- arch/x86/events/intel/pt.c | 4 +++- arch/x86/events/intel/rapl.c | 8 +------- arch/x86/events/intel/uncore.c | 8 -------- arch/x86/events/intel/uncore_snb.c | 8 +------- arch/x86/events/msr.c | 8 +------- drivers/perf/arm-cci.c | 9 --------- drivers/perf/arm-ccn.c | 5 +---- drivers/perf/arm_dsu_pmu.c | 8 +------- drivers/perf/arm_pmu.c | 15 +++++---------- drivers/perf/arm_spe_pmu.c | 3 ++- drivers/perf/hisilicon/hisi_uncore_pmu.c | 9 --------- drivers/perf/qcom_l2_pmu.c | 8 -------- drivers/perf/qcom_l3_pmu.c | 7 ------- drivers/perf/xgene_pmu.c | 5 ----- include/linux/perf_event.h | 10 ++++++++++ include/uapi/linux/perf_event.h | 2 -- kernel/events/core.c | 9 +++++++++ kernel/events/hw_breakpoint.c | 2 ++ tools/include/uapi/linux/perf_event.h | 2 -- tools/perf/design.txt | 4 ++++ 40 files changed, 54 insertions(+), 189 deletions(-)