[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20171129083853.28022-8-jolsa@kernel.org>
Date: Wed, 29 Nov 2017 09:38:52 +0100
From: Jiri Olsa <jolsa@...nel.org>
To: Ingo Molnar <mingo@...nel.org>,
Peter Zijlstra <a.p.zijlstra@...llo.nl>,
Arnaldo Carvalho de Melo <acme@...nel.org>
Cc: lkml <linux-kernel@...r.kernel.org>,
Namhyung Kim <namhyung@...nel.org>,
David Ahern <dsahern@...il.com>,
Andi Kleen <ak@...ux.intel.com>,
Milind Chabbi <chabbi.milind@...il.com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Michael Ellerman <mpe@...erman.id.au>,
Hari Bathini <hbathini@...ux.vnet.ibm.com>,
Jin Yao <yao.jin@...ux.intel.com>,
Kan Liang <kan.liang@...el.com>,
Sukadev Bhattiprolu <sukadev@...ux.vnet.ibm.com>,
Oleg Nesterov <onestero@...hat.com>,
Will Deacon <will.deacon@....com>
Subject: [PATCH 7/8] perf/core: fast breakpoint modification via _IOC_MODIFY_ATTRIBUTES.
From: Milind Chabbi <chabbi.milind@...il.com>
Problem and motivation: Once a breakpoint perf event (PERF_TYPE_BREAKPOINT)
is created, there is no flexibility to change the breakpoint type
(bp_type), breakpoint address (bp_addr), or breakpoint length (bp_len). The
only option is to close the perf event and configure a new breakpoint
event. This inflexibility has a significant performance overhead. For
example, sampling-based, lightweight performance profilers (and also
concurrency bug detection tools), monitor different addresses for a short
duration using PERF_TYPE_BREAKPOINT and change the address (bp_addr) to
another address or change the kind of breakpoint (bp_type) from "write" to
a "read" or vice-versa or change the length (bp_len) of the address being
monitored. The cost of these modifications is prohibitive since it involves
unmapping the circular buffer associated with the perf event, closing the
perf event, opening another perf event and mmaping another circular buffer.
Solution: The new ioctl flag for perf events,
PERF_EVENT_IOC_MODIFY_ATTRIBUTES, introduced in this patch takes a pointer
to a struct perf_event_attr as an argument to update an old breakpoint
event with new address, type, and size. This facility allows retaining a
previous mmaped perf events ring buffer and avoids having to close and
reopen another perf event.
This patch supports only changing PERF_TYPE_BREAKPOINT event type; future
implementations can extend this feature. The patch replicates some of its
functionality of modify_user_hw_breakpoint() in
kernel/events/hw_breakpoint.c. modify_user_hw_breakpoint cannot be called
directly since perf_event_ctx_lock() is already held in _perf_ioctl().
Evidence: Experiments show that the baseline (not able to modify an already
created breakpoint) costs an order of magnitude (~10x) more than the
suggested optimization (having the ability to dynamically modifying a
configured breakpoint via ioctl). When the breakpoints typically do not
trap, the speedup due to the suggested optimization is ~10x; even when the
breakpoints always trap, the speedup is ~4x due to the suggested
optimization.
Testing: tests posted at
https://github.com/linux-contrib/perf_event_modify_bp demonstrate the
performance significance of this patch. Tests also check the functional
correctness of the patch.
Signed-off-by: Milind Chabbi <chabbi.milind@...il.com>
[ Using modify_user_hw_breakpoint_check function. ]
Signed-off-by: Jiri Olsa <jolsa@...nel.org>
---
include/linux/hw_breakpoint.h | 7 ++++++
include/uapi/linux/perf_event.h | 2 ++
kernel/events/core.c | 47 +++++++++++++++++++++++++++++++++++
kernel/events/hw_breakpoint.c | 2 +-
tools/include/uapi/linux/perf_event.h | 2 ++
5 files changed, 59 insertions(+), 1 deletion(-)
diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h
index cf045885a499..abeba094f080 100644
--- a/include/linux/hw_breakpoint.h
+++ b/include/linux/hw_breakpoint.h
@@ -53,6 +53,9 @@ register_user_hw_breakpoint(struct perf_event_attr *attr,
/* FIXME: only change from the attr, and don't unregister */
extern int
modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr);
+extern int
+modify_user_hw_breakpoint_check(struct perf_event *bp, struct perf_event_attr *attr,
+ bool check);
/*
* Kernel breakpoints are not associated with any particular thread.
@@ -97,6 +100,10 @@ register_user_hw_breakpoint(struct perf_event_attr *attr,
static inline int
modify_user_hw_breakpoint(struct perf_event *bp,
struct perf_event_attr *attr) { return -ENOSYS; }
+static inline int
+__modify_user_hw_breakpoint(struct perf_event *bp,
+ struct perf_event_attr *attr) { return -ENOSYS; }
+
static inline struct perf_event *
register_wide_hw_breakpoint_cpu(struct perf_event_attr *attr,
perf_overflow_handler_t triggered,
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 362493a2f950..cd430821f022 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -433,6 +433,8 @@ struct perf_event_attr {
#define PERF_EVENT_IOC_ID _IOR('$', 7, __u64 *)
#define PERF_EVENT_IOC_SET_BPF _IOW('$', 8, __u32)
#define PERF_EVENT_IOC_PAUSE_OUTPUT _IOW('$', 9, __u32)
+#define PERF_EVENT_IOC_MODIFY_ATTRIBUTES \
+ _IOW('$', 10, struct perf_event_attr *)
enum perf_event_ioc_flags {
PERF_IOC_FLAG_GROUP = 1U << 0,
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 028adb24bf7a..d6e5e2eb472b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2644,6 +2644,41 @@ int perf_event_refresh(struct perf_event *event, int refresh)
}
EXPORT_SYMBOL_GPL(perf_event_refresh);
+static int perf_event_modify_breakpoint(struct perf_event *bp,
+ struct perf_event_attr *attr)
+{
+ int err;
+
+ _perf_event_disable(bp);
+
+ err = modify_user_hw_breakpoint_check(bp, attr, true);
+ if (err) {
+ if (!bp->attr.disabled)
+ _perf_event_enable(bp);
+
+ return err;
+ }
+
+ if (!attr->disabled)
+ _perf_event_enable(bp);
+ return 0;
+}
+
+static int perf_event_modify_attr(struct perf_event *event,
+ struct perf_event_attr *attr)
+{
+ if (event->attr.type != attr->type)
+ return -EINVAL;
+
+ switch (event->attr.type) {
+ case PERF_TYPE_BREAKPOINT:
+ return perf_event_modify_breakpoint(event, attr);
+ default:
+ /* Place holder for future additions. */
+ return -EOPNOTSUPP;
+ }
+}
+
static void ctx_sched_out(struct perf_event_context *ctx,
struct perf_cpu_context *cpuctx,
enum event_type_t event_type)
@@ -4655,6 +4690,8 @@ static int perf_event_set_output(struct perf_event *event,
struct perf_event *output_event);
static int perf_event_set_filter(struct perf_event *event, void __user *arg);
static int perf_event_set_bpf_prog(struct perf_event *event, u32 prog_fd);
+static int perf_copy_attr(struct perf_event_attr __user *uattr,
+ struct perf_event_attr *attr);
static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned long arg)
{
@@ -4724,6 +4761,16 @@ static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned lon
rcu_read_unlock();
return 0;
}
+ case PERF_EVENT_IOC_MODIFY_ATTRIBUTES: {
+ struct perf_event_attr new_attr;
+ int err = perf_copy_attr((struct perf_event_attr __user *)arg,
+ &new_attr);
+
+ if (err)
+ return err;
+
+ return perf_event_modify_attr(event, &new_attr);
+ }
default:
return -ENOTTY;
}
diff --git a/kernel/events/hw_breakpoint.c b/kernel/events/hw_breakpoint.c
index 0c82663395f7..6253d5519cd8 100644
--- a/kernel/events/hw_breakpoint.c
+++ b/kernel/events/hw_breakpoint.c
@@ -456,7 +456,7 @@ register_user_hw_breakpoint(struct perf_event_attr *attr,
}
EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
-static int
+int
modify_user_hw_breakpoint_check(struct perf_event *bp, struct perf_event_attr *attr,
bool check)
{
diff --git a/tools/include/uapi/linux/perf_event.h b/tools/include/uapi/linux/perf_event.h
index 362493a2f950..4602c260f802 100644
--- a/tools/include/uapi/linux/perf_event.h
+++ b/tools/include/uapi/linux/perf_event.h
@@ -433,6 +433,8 @@ struct perf_event_attr {
#define PERF_EVENT_IOC_ID _IOR('$', 7, __u64 *)
#define PERF_EVENT_IOC_SET_BPF _IOW('$', 8, __u32)
#define PERF_EVENT_IOC_PAUSE_OUTPUT _IOW('$', 9, __u32)
+#define PERF_EVENT_IOC_MODIFY_ATTRIBUTES \
+ _IOW('$', 10, struct perf_event_attr *)
enum perf_event_ioc_flags {
PERF_IOC_FLAG_GROUP = 1U << 0,
--
2.13.6
Powered by blists - more mailing lists