perf c2c report: Make calc_width work with struct c2c_hist_entry

We are going to calculate tje column width based on the struct
c2c_hist_entry data, so making calc_width to work with struct
c2c_hist_entry.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Joe Mario <jmario@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180309101442.9224-6-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf c2c record: Record physical addresses in samples

We are going to display NUMA node information in following patches. For
this we need to have physical address data in the sample.

Adding --phys-data as a default option for perf c2c record.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Joe Mario <jmario@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180309101442.9224-5-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Add mem2node object test

Adding mem2node object automated test.

The test prepares few artificial nodes - memory maps and verifies the
mem2node object returns proper node values to given addresses.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180309101442.9224-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add mem2node object

Adding mem2node object to allow the easy lookup of the node for the
physical address.

It has following interface:

  int  mem2node__init(struct mem2node *map, struct perf_env *env);
  void mem2node__exit(struct mem2node *map);
  int  mem2node__node(struct mem2node *map, u64 addr);

The mem2node__toolsinit initialize object from the perf data file
MEM_TOPOLOGY feature data. Following calls to mem2node__node will return
node number for given physical address. The mem2node__exit function
frees the object.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180309101442.9224-3-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf env: Free memory nodes data

Forgot to free env's memory nodes, adding needed code to perf_env__exit.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180309101442.9224-2-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/core: Clear sibling list of detached events

When perf_group_dettach() is called on a group leader, it updates each
sibling's group_leader field to point to that sibling, effectively
upgrading each siblnig to a group leader. After perf_group_detach has
completed, the caller may free the leader event.

We only remove siblings from the group leader's sibling_list when the
leader has a non-empty group_node. This was fine prior to commit:

  8343aae66167df67 ("perf/core: Remove perf_event::group_entry")

... as the sibling's sibling_list would be empty. However, now that we
use the sibling_list field as both the list head and the list entry,
this leaves each sibling with a non-empty sibling list, including the
stale leader event.

If perf_group_detach() is subsequently called on a sibling, it will
appear to be a group leader, and we'll walk the sibling_list,
potentially dereferencing these stale events. In 0day testing, this has
been observed to result in kernel panics.

Let's avoid this by always removing siblings from the sibling list when
we promote them to leaders.

Fixes: 8343aae66167df67 ("perf/core: Remove perf_event::group_entry")
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: vincent.weaver@maine.edu
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: torvalds@linux-foundation.org
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: valery.cherepennikov@intel.com
Cc: linux-tip-commits@vger.kernel.org
Cc: eranian@google.com
Cc: acme@redhat.com
Cc: alexander.shishkin@linux.intel.com
Cc: davidcc@google.com
Cc: kan.liang@intel.com
Cc: Dmitry.Prohorov@intel.com
Cc: Jiri Olsa <jolsa@redhat.com>
Link: https://lkml.kernel.org/r/20180316131741.3svgr64yibc6vsid@lakrids.cambridge.arm.com

perf: Fix sibling iteration

Mark noticed that the change to sibling_list changed some iteration
semantics; because previously we used group_list as list entry,
sibling events would always have an empty sibling_list.

But because we now use sibling_list for both list head and list entry,
siblings will report as having siblings.

Fix this with a custom for_each_sibling_event() iterator.

Fixes: 8343aae66167 ("perf/core: Remove perf_event::group_entry")
Reported-by: Mark Rutland <mark.rutland@arm.com>
Suggested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: vincent.weaver@maine.edu
Cc: alexander.shishkin@linux.intel.com
Cc: torvalds@linux-foundation.org
Cc: alexey.budankov@linux.intel.com
Cc: valery.cherepennikov@intel.com
Cc: eranian@google.com
Cc: acme@redhat.com
Cc: linux-tip-commits@vger.kernel.org
Cc: davidcc@google.com
Cc: kan.liang@intel.com
Cc: Dmitry.Prohorov@intel.com
Cc: jolsa@redhat.com
Link: https://lkml.kernel.org/r/20180315170129.GX4043@hirez.programming.kicks-ass.net

perf/core: Implement fast breakpoint modification via _IOC_MODIFY_ATTRIBUTES

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@gmail.com>
[ Using modify_user_hw_breakpoint_check function. ]
[ Reformated PERF_EVENT_IOC_*, so the values are all in one column. ]
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-8-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tests: Add breakpoint accounting/modify test

Adding test that:

  - detects the number of watch/break-points,
    skip test if any is missing
  - detects PERF_EVENT_IOC_MODIFY_ATTRIBUTES ioctl,
    skip test if it's missing
  - detects if watchpoints and breakpoints share
    same slots
  - create all possible watchpoints on cpu 0
  - change one of it to breakpoint
  - in case wp and bp do not share slots,
    we create another watchpoint to ensure
    the slot accounting is correct

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-9-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Move perf_event_attr::sample_max_stack into perf_copy_attr()

Move the sample_max_stack check and setup into perf_copy_attr(),
so we have all perf_event_attr initial setup in one place
and can easily compare attrs in the new ioctl introduced
in following change.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-7-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

hw_breakpoint: Add perf_event_attr fields check in __modify_user_hw_breakpoint()

And rename it to modify_user_hw_breakpoint_check().

We are about to use modify_user_hw_breakpoint_check() for user space
breakpoints modification, we must be very strict to check only the
fields we can change have changed. As Peter explained:

 "Suppose someone does:

        attr = malloc(sizeof(*attr)); // uninitialized memory
        attr->type = BP;
        attr->bp_addr = new_addr;
        attr->bp_type = bp_type;
        attr->bp_len = bp_len;
        ioctl(fd, PERF_IOC_MOD_ATTR, &attr);

  And feeds absolute shite for the rest of the fields.
  Then we later want to extend IOC_MOD_ATTR to allow changing
  attr::sample_type but we can't, because that would break the
  above application."

I'm making this check optional because we already export
modify_user_hw_breakpoint() and with this check we could
break existing users.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-6-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

hw_breakpoint: Factor out __modify_user_hw_breakpoint() function

Moving out all the functionality without the events
disabling/enabling calls, because we want to call another
disabling/enabling functions in following change.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-5-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

hw_breakpoint: Add modify_bp_slot() function

Add the modify_bp_slot() function to keep slot numbers
correct when changing the breakpoint type.

Using existing __release_bp_slot()/__reserve_bp_slot()
call sequence to update the slot counts.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-4-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

hw_breakpoint: Pass bp_type argument to __reserve_bp_slot|__release_bp_slot()

Passing bp_type argument to __reserve_bp_slot() and __release_bp_slot()
functions, so we can pass another bp_type than the one defined in
bp->attr.bp_type. This will be handy in following change that fixes
breakpoint slot counts during its modification.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-3-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

hw_breakpoint: Pass bp_type directly as find_slot_idx() argument

Pass bp_type directly as a find_slot_idx() argument,
so we don't need to have whole event to get the
breakpoint slot type. It will be used in following
changes.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Milind Chabbi <chabbi.milind@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-2-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Fix installing cgroup events on CPU

There's two problems when installing cgroup events on CPUs: firstly
list_update_cgroup_event() only tries to set cpuctx->cgrp for the
first event, if that mismatches on @cgrp we'll not try again for later
additions.

Secondly, when we install a cgroup event into an active context, only
issue an event reprogram when the event matches the current cgroup
context. This avoids a pointless event reprogramming.

Signed-off-by: leilei.lin <leilei.lin@alibaba-inc.com>
[ Improved the changelog and comments. ]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: brendan.d.gregg@gmail.com
Cc: eranian@gmail.com
Cc: linux-kernel@vger.kernel.org
Cc: yang_oliver@hotmail.com
Link: http://lkml.kernel.org/r/20180306093637.28247-1-linxiulei@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Optimize perf_rotate_context() event scheduling

The event schedule order (as per perf_event_sched_in()) is:

 - cpu  pinned
 - task pinned
 - cpu  flexible
 - task flexible

But perf_rotate_context() will unschedule cpu-flexible even if it
doesn't need a rotation.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Fix tree based event rotation

Similar to how first programming cpu=-1 and then cpu=# is wrong, so is
rotating both. It was especially wrong when we were still programming
the PMU in this same order, because in that scenario we might never
actually end up running cpu=# events at all.

Cure this by using the active_list to pick the rotation event; since
at programming we already select the left-most event.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Simpify perf_event_groups_for_each()

The last argument is, and always must be, the same.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Optimize ctx_sched_out()

When an event group contains more events than can be scheduled on the
hardware, iterating the full event group for ctx_sched_out is a waste
of time.

Keep track of the events that got programmed on the hardware, such
that we can iterate this smaller list in order to schedule them out.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Remove perf_event::group_entry

Now that all the grouping is done with RB trees, we no longer need
group_entry and can replace the whole thing with sibling_list.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Fix event schedule order

Scheduling in events with cpu=-1 before events with cpu=# changes
semantics and is undesirable in that it would priorize these events.

Given that groups->index is across all groups we actually have an
inter-group ordering, meaning we can merge-sort two groups, which is
just what we need to preserve semantics.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Cleanup the rb-tree code

Trivial comment and code fixups..

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/cor: Use RB trees for pinned/flexible groups

Change event groups into RB trees sorted by CPU and then by a 64bit
index, so that multiplexing hrtimer interrupt handler would be able
skipping to the current CPU's list and ignore groups allocated for the
other CPUs.

New API for manipulating event groups in the trees is implemented as well
as adoption on the API in the current implementation.

pinned_group_sched_in() and flexible_group_sched_in() API are
introduced to consolidate code enabling the whole group from pinned
and flexible groups appropriately.

Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitri Prokhorov <Dmitry.Prohorov@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valery Cherepennikov <valery.cherepennikov@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/372f9c8b-0cfe-4240-e44d-83d863d40813@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Fix perf_output_read_group()

Mark reported his arm64 perf fuzzer runs sometimes splat like:

  armv8pmu_read_counter+0x1e8/0x2d8
  armpmu_event_update+0x8c/0x188
  armpmu_read+0xc/0x18
  perf_output_read+0x550/0x11e8
  perf_event_read_event+0x1d0/0x248
  perf_event_exit_task+0x468/0xbb8
  do_exit+0x690/0x1310
  do_group_exit+0xd0/0x2b0
  get_signal+0x2e8/0x17a8
  do_signal+0x144/0x4f8
  do_notify_resume+0x148/0x1e8
  work_pending+0x8/0x14

which asserts that we only call pmu::read() on ACTIVE events.

The above callchain does:

  perf_event_exit_task()
    perf_event_exit_task_context()
      task_ctx_sched_out() // INACTIVE
      perf_event_exit_event()
        perf_event_set_state(EXIT) // EXIT
        sync_child_event()
          perf_event_read_event()
            perf_output_read()
              perf_output_read_group()
                leader->pmu->read()

Which results in doing a pmu::read() on an !ACTIVE event.

I _think_ this is 'new' since we added attr.inherit_stat, which added
the perf_event_read_event() to the exit path, without that
perf_event_read_output() would only trigger from samples and for
@event to trigger a sample, it's leader _must_ be ACTIVE too.

Still, adding this check makes it consistent with the @sub case for
the siblings.

Reported-and-Tested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'perf-core-for-mingo-4.17-20180308' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

- Support to display the IPC/Cycle in 'annotate' TUI, for systems
  where this info can be obtained, like Intel's >= Skylake (Jin Yao)

- Support wildcards on PMU name in dynamic PMU events (Agustin Vega-Frias)

- Display pmu name when printing unmerged events in stat (Agustin Vega-Frias)

- Auto-merge PMU events created by prefix or glob match (Agustin Vega-Frias)

- Fix s390 'call' operations target function annotation (Thomas Richter)

- Handle s390 PC relative load and store instruction in the augmented
  'annotate', code, used so far in the TUI modes of 'perf report' and
  'perf annotate' (Thomas Richter)

- Provide libtraceevent with a kernel symbol resolver, so that
  symbols in tracepoint fields can be resolved when showing them in
  tools such as 'perf report' (Wang YanQing)

- Refactor the cgroups code to look more like other code in tools/perf,
  using cgroup__{put,get} for refcount operations instead of its
  open-coded equivalent, breaking larger functions, etc (Arnaldo Carvalho de Melo)

- Implement support for the -G/--cgroup target in 'perf trace', allowing
  strace like tracing (plus other events, backtraces, etc) for cgroups
  (Arnaldo Carvalho de Melo)

- Update thread shortname in 'perf sched map' when the thread's COMM
  changes (Changbin Du)

- refcount 'struct mem_info', for better sharing it over several
  users, avoid duplicating structs and fixing crashes related to
  use after free (Jiri Olsa)

- Display perf.data version, offsets in 'perf report --header' (Jiri Olsa)

- Record the machine's memory topology information in a perf.data
  feature section, to be used by tools such as 'perf c2c' (Jiri Olsa)

- Fix output of forced groups in the header for 'perf report' --stdio
  and --tui (Jiri Olsa)

- Better support llvm, clang, cxx make tests in the build process (Jiri Olsa)

- Streamline the 'struct perf_mmap' methods, storing some info in the
  struct instead of passing it via various methods, shortening its
  signatures (Kan Liang)

- Update the quipper perf.data parser library site information (Stephane Eranian)

- Correct perf's man pages title markers for asciidoctor (Takashi Iwai)

- Intel PT fixes and refactorings paving the way for implementing
  support for AUX area sampling (Adrian Hunter)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Disable userspace RDPMC usage for large PEBS

Userspace RDPMC cannot possibly work for large PEBS, which was introduced in:

  b8241d20699e ("perf/x86/intel: Implement batched PEBS interrupt handling (large PEBS interrupt threshold)")

When the PEBS interrupt threshold is larger than one, there is no way
to get exact auto-reload times and value for userspace RDPMC.  Disable
the userspace RDPMC usage when large PEBS is enabled.

The only exception is when the PEBS interrupt threshold is 1, in which
case user-space RDPMC works well even with auto-reload events.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Fixes: b8241d20699e ("perf/x86/intel: Implement batched PEBS interrupt handling (large PEBS interrupt threshold)")
Link: http://lkml.kernel.org/r/1518474035-21006-6-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix PMU read for auto-reload

Auto-reload events needs to be specially handled in event count read.

Auto-reload is only available for intel_pmu.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Fixes: b8241d20699e ("perf/x86/intel: Implement batched PEBS interrupt handling (large PEBS interrupt threshold)")
Link: http://lkml.kernel.org/r/1518474035-21006-5-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/ds: Introduce ->read() function for auto-reload events and flush the PEBS buffer there

There is no way to get exact auto-reload times and values which are needed
for event updates unless we flush the PEBS buffer.

Introduce intel_pmu_auto_reload_read() to drain the PEBS buffer for
auto reload event. To prevent races with the hardware, we can only
call drain_pebs() when the PMU is disabled.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Link: http://lkml.kernel.org/r/1518474035-21006-4-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Introduce a ->read() callback in 'struct x86_pmu'

Auto-reload needs to be specially handled when reading event counts.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Link: http://lkml.kernel.org/r/1518474035-21006-3-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix event update for auto-reload

There is a bug when reading event->count with large PEBS enabled.

Here is an example:

  # ./read_count
  0x71f0
  0x122c0
  0x1000000001c54
  0x100000001257d
  0x200000000bdc5

In fixed period mode, the auto-reload mechanism could be enabled for
PEBS events, but the calculation of event->count does not take the
auto-reload values into account.

Anyone who reads event->count will get the wrong result, e.g x86_pmu_read().

This bug was introduced with the auto-reload mechanism enabled since
commit:

  851559e35fd5 ("perf/x86/intel: Use the PEBS auto reload mechanism when possible")

Introduce intel_pmu_save_and_restart_reload() to calculate the
event->count only for auto-reload.

Since the counter increments a negative counter value and overflows on
the sign switch, giving the interval:

        [-period, 0]

the difference between two consequtive reads is:

 A) value2 - value1;
    when no overflows have happened in between,
 B) (0 - value1) + (value2 - (-period));
    when one overflow happened in between,
 C) (0 - value1) + (n - 1) * (period) + (value2 - (-period));
    when @n overflows happened in between.

Here A) is the obvious difference, B) is the extension to the discrete
interval, where the first term is to the top of the interval and the
second term is from the bottom of the next interval and C) the extension
to multiple intervals, where the middle term is the whole intervals
covered.

The equation for all cases is:

    value2 - value1 + n * period

Previously the event->count is updated right before the sample output.
But for case A, there is no PEBS record ready. It needs to be specially
handled.

Remove the auto-reload code from x86_perf_event_set_period() since
we'll not longer call that function in this case.

Based-on-code-from: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Fixes: 851559e35fd5 ("perf/x86/intel: Use the PEBS auto reload mechanism when possible")
Link: http://lkml.kernel.org/r/1518474035-21006-2-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Properly save/restore the PMU state in the NMI handler

The PMU is disabled in intel_pmu_handle_irq(), but cpuc->enabled is not updated
accordingly.

This is fine in current usage because no-one checks it - but fix it
for future code: for example, the drain_pebs() will be modified to
fix an auto-reload bug.

Properly save/restore the old PMU state.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Cc: kernel test robot <fengguang.wu@intel.com>
Link: http://lkml.kernel.org/r/6f44ee84-56f8-79f1-559b-08e371eaeb78@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix large period handling on Broadwell CPUs

Large fixed period values could be truncated on Broadwell, for example:

  perf record -e cycles -c 10000000000

Here the fixed period is 0x2540BE400, but the period which finally applied is
0x540BE400 - which is wrong.

The reason is that x86_pmu::limit_period() uses an u32 parameter, so the
high 32 bits of 'period' get truncated.

This bug was introduced in:

  commit 294fe0f52a44 ("perf/x86/intel: Add INST_RETIRED.ALL workarounds")

It's safe to use u64 instead of u32:

 - Although the 'left' is s64, the value of 'left' must be positive when
   calling limit_period().

 - bdw_limit_period() only modifies the lowest 6 bits, it doesn't touch
   the higher 32 bits.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Fixes: 294fe0f52a44 ("perf/x86/intel: Add INST_RETIRED.ALL workarounds")
Link: http://lkml.kernel.org/r/1519926894-3520-1-git-send-email-kan.liang@linux.intel.com
[ Rewrote unacceptably bad changelog. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Update quipper information

This patch updates the links to the Quipper library.  It is now
available from GitHub and has been updated.

Reported-by: Lakshman Annadorai <lakshmana@google.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1520495985-2147-1-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Handle s390 PC relative load and store instruction.

S390 has several load and store instructions with target operand
addressing relative to the program counter, for example lrl, lgrl, strl,
stgrl.

These instructions are handled similar to x86. Objdump output displays
those instructions as:

   9595c: c4 2d 00 09 9c 54   lgrl   %r7,1c8540 <mp_+0x60>

This output is parsed (like on x86) and perf annotate shows those lines
as:

   lgrl   %r7,mp_+0x60

This patch handles the s390 specific instruction parsing for PC relative
load and store instructions.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180308120913.14802-1-tmricht@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Support to display the IPC/Cycle in TUI mode

Unlike the perf report interactive annotate mode, the perf annotate
doesn't display the IPC/Cycle even if branch info is recorded in perf
data file.

perf record -b ...
perf annotate function

It should show IPC/cycle, but it doesn't.

This patch lets perf annotate support the displaying of IPC/Cycle if
branch info is in perf data.

For example,

  perf annotate compute_flag

  Percent│ IPC Cycle
         │
         │
         │                Disassembly of section .text:
         │
         │                0000000000400640 <compute_flag>:
         │                compute_flag():
         │                volatile int count;
         │                static unsigned int s_randseed;
         │
         │                __attribute__((noinline))
         │                int compute_flag()
         │                {
   22.96 │1.18   584        sub    $0x8,%rsp
         │                        int i;
         │
         │                        i = rand() % 2;
   23.02 │1.18     1      → callq  rand@plt
         │
         │                        return i;
   27.05 │3.37              mov    %eax,%edx
         │                }
         │3.37              add    $0x8,%rsp
         │                {
         │                        int i;
         │
         │                        i = rand() % 2;
         │
         │                        return i;
         │3.37              shr    $0x1f,%edx
         │3.37              add    %edx,%eax
         │3.37              and    $0x1,%eax
         │3.37              sub    %edx,%eax
         │                }
   26.97 │3.37     2      ← retq

Note that, this patch only supports TUI mode. For stdio, now it just keeps
original behavior. Will support it in a follow-up patch.

  $ perf annotate compute_flag --stdio

   Percent |      Source code & Disassembly of div for cycles:ppp (7993 samples)
  ------------------------------------------------------------------------------
           :
           :
           :
           :            Disassembly of section .text:
           :
           :            0000000000400640 <compute_flag>:
           :            compute_flag():
           :            volatile int count;
           :            static unsigned int s_randseed;
           :
           :            __attribute__((noinline))
           :            int compute_flag()
           :            {
      0.29 :   400640:       sub    $0x8,%rsp     # +100.00%
           :                    int i;
           :
           :                    i = rand() % 2;
     42.93 :   400644:       callq  400490 <rand@plt>     # -100.00% (p:100.00%)
           :
           :                    return i;
      0.10 :   400649:       mov    %eax,%edx     # +100.00%
           :            }
      0.94 :   40064b:       add    $0x8,%rsp
           :            {
           :                    int i;
           :
           :                    i = rand() % 2;
           :
           :                    return i;
     27.02 :   40064f:       shr    $0x1f,%edx
      0.15 :   400652:       add    %edx,%eax
      1.24 :   400654:       and    $0x1,%eax
      2.08 :   400657:       sub    %edx,%eax
           :            }
     25.26 :   400659:       retq # -100.00% (p:100.00%)

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/20180223170210.GC7045@tassilo.jf.intel.com
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1519724327-7773-1-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Provide libtraceevent with a kernel symbol resolver

So that beautifiers wanting to resolve kernel function addresses to
names can do its work, and when we use "perf report" for output of "perf
kmem record", we will get kernel symbol output.

This patch affect the output of "perf report" for the record data
generated by "perf kmem record" looks like below:

Before patch:
0.01%  call_site=ffffffff814e5828 ptr=0x99bb000 bytes_req=3616 bytes_alloc=4096 gfp_flags=GFP_ATOMIC
0.01%  call_site=ffffffff81370b87 ptr=0x428a3060 bytes_req=32 bytes_alloc=32 gfp_flags=GFP_KERNEL|GFP_ZERO

After patch:
0.01%  (aa_alloc_task_context+0x27) call_site=ffffffff81370b87 ptr=0x428a3060 bytes_req=32 bytes_alloc=32 gfp_flags=GFP_KERNEL|GFP_ZERO
0.01%  (__tty_buffer_request_room+0x88) call_site=ffffffff814e5828 ptr=0x99bb000 bytes_req=3616 bytes_alloc=4096 gfp_flags=GFP_ATOMIC

Signed-off-by: Wang YanQing <udknight@gmail.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180308032850.GA12383@udknight-ThinkPad-E550
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Force llvm/clang test compile output to .make.output

So we can see the output of feature compile in following files:

  tools/build/feature/test-llvm.make.output
  tools/build/feature/test-llvm-version.make.output
  tools/build/feature/test-clang.make.output

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-20-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Add llvm/clang make targets to FILES

So they can follow the OUTPUT variable setup as the rest of the
features.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-19-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Add llvm/clang/cxx make tests into FEATURE_TESTS_EXTRA

So we can see the status when we build perf, like:

  $ make LIBCLANGLLVM=1 VF=1
  ...                           cxx: [ on  ]
  ...                          llvm: [ on  ]
  ...                  llvm-version: [ on  ]
  ...                         clang: [ on  ]

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-18-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Update tags with .cpp files

We have some .cpp files, make ctags/cscope aware of them.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-17-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add MEM_TOPOLOGY feature to perf data file

Adding MEM_TOPOLOGY feature to perf data file,
that will carry physical memory map and its
node assignments.

The format of data in MEM_TOPOLOGY is as follows:

  0 - version          | for future changes
  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
 16 - count            | number of nodes

 For each node we store map of physical indexes for
 each node:

 32 - node id          | node index
 40 - size             | size of bitmap
 48 - bitmap           | bitmap of memory indexes that belongs to node
                       | /sys/devices/system/node/node<NODE>/memory<INDEX>

The MEM_TOPOLOGY could be displayed with following
report command:

  $ perf report --header-only -I
  ...
  # memory nodes (nr 1, block size 0x8000000):
  #    0 [7G]: 0-23,32-69

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-8-jolsa@kernel.org
[ Rename 'index' to 'idx', as this breaks the build in rhel5, 6 and other systems where this is used by glibc headers ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf c2c: Use mem_info refcnt logic

Switch to refcnt logic instead of duplicating mem_info objects. No
functional change, just saving some memory.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-7-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add refcnt into struct mem_info

It's passed along several hists entries in --hierarchy mode, so it's
better we keep track of it.

The current fail I see is that it gets removed in hierarchy --mem-mode
mode, where it's shared in the different hierarchies, but removed from
the template hist entry, so the report crashes.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-6-jolsa@kernel.org
[ Rename mem_info__aloc() to mem_info__new(), to fix the typo and use the convention for constructors ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Remove progname from struct record

It's no longer used.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-5-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Move machine variable down the function

It's used far more down to be declared on the top of the __cmd_record.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Display perf.data header info

Display more header info from perf.data file, following values:

  $ perf report -i perf.data --header-only
  ...
  # header version : 1
  # data offset    : 424
  # data size      : 3364280
  # feat offset    : 3364704

It's handy for debuging.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-3-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Fix the output for stdio events list

Changing the output header for reporting forced groups via --groups
option on non grouped events, like:

  $ perf record -e 'cycles,instructions'
  $ perf report --stdio --group

Before:

  # Samples: 24  of event 'anon group { cycles:u, instructions:u }'

After:

  # Samples: 24  of events 'cycles:u, instructions:u'

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Fixes: ad52b8cb4886 ("perf report: Add support to display group output for non group events")
Link: http://lkml.kernel.org/r/20180307155020.32613-2-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Fix s390 target function disassembly

'perf annotate' displays function call assembler instructions with a
right arrow. Hitting enter on this line/instruction causes the browser
to disassemble this target function and show it on the screen.  On s390
this results in an error message 'The called function was not found.'

The function call assembly line parsing does not handle the s390 bras
and brasl instructions. Function call__parse expects the target as first
operand:

callq e9140 <__fxstat>

S390 has a register number as first operand:

brasl %r14,41d60 <abort>

Therefore the target addresses on s390 are always zero which is an
invalid address.

Introduce a s390 specific call parsing function which skips the first
operand on s390.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180307134325.96106-1-tmricht@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Adjust overlap-checking to support sampling mode

Adjust overlap-checking to support sampling mode.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-10-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Remove a check for sampling mode

Intel PT code already has some preparation for AUX area sampling mode.

However the implementation has changed from the first proposal and one
of the side-effects is that it will not be impossible to support snapshot
mode and sampling mode at the same time.

Although there are no plans to support it, let validation (not yet
implemented) control whether it is allowed rather than low-level
functions.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-9-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Tidy old_buffer handling in intel_pt_get_trace()

intel_pt_get_trace() fixes overlaps between the current buffer and the
previous buffer ('old_buffer').

However the previous buffer might not have had usable data (no PSB) so
the comparison must be made against the previous buffer that had usable
data.

Tidy that by keeping a pointer for that purpose in struct intel_pt_queue.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-8-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Get rid of intel_pt_use_buffer_pid_tid()

With the new way sampling support will be implemented,
intel_pt_use_buffer_pid_tid() will not be needed. Get rid of it.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-7-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt/bts: In auxtrace_record__init_intel() evlist is never NULL

Tidy auxtrace_record__init_intel() slightly by recognizing that evlist is
never NULL.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix timestamp following overflow

timestamp_insn_cnt is used to estimate the timestamp based on the number of
instructions since the last known timestamp.

If the estimate is not accurate enough decoding might not be correctly
synchronized with side-band events causing more trace errors.

However there are always timestamps following an overflow, so the
estimate is not needed and can indeed result in more errors.

Suppress the estimate by setting timestamp_insn_cnt to zero.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1520431349-30689-5-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix error recovery from missing TIP packet

When a TIP packet is expected but there is a different packet, it is an
error. However the unexpected packet might be something important like a
TSC packet, so after the error, it is necessary to continue from there,
rather than the next packet. That is achieved by setting pkt_step to
zero.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1520431349-30689-4-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix sync_switch

sync_switch is a facility to synchronize decoding more closely with the
point in the kernel when the context actually switched.

The flag when sync_switch is enabled was global to the decoding, whereas
it is really specific to the CPU.

The trace data for different CPUs is put on different queues, so add
sync_switch to the intel_pt_queue structure and use that in preference
to the global setting in the intel_pt structure.

That fixes problems decoding one CPU's trace because sync_switch was
disabled on a different CPU's queue.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1520431349-30689-3-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix overlap detection to identify consecutive buffers correctly

Overlap detection was not not updating the buffer's 'consecutive' flag.
Marking buffers consecutive has the advantage that decoding begins from
the start of the buffer instead of the first PSB. Fix overlap detection
to identify consecutive buffers correctly.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1520431349-30689-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Simplify perf_mmap__read_init()

It isn't necessary to pass the 'start', 'end' and 'overwrite' arguments
to perf_mmap__read_init().  The data is stored in the struct perf_mmap.

Discard the parameters.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-8-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Simplify perf_mmap__read_event()

It isn't necessary to pass the 'overwrite', 'start' and 'end' argument
to perf_mmap__read_event().  Discard them.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-7-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Simplify perf_mmap__consume()

It isn't necessary to pass the 'overwrite' argument to
perf_mmap__consume().  Discard it.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-6-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Use stored 'overwrite' in perf_mmap__consume()

The 'overwrite' is set at allocation. It will not be changed.  Using it
to replace the parameter of perf_mmap__consume().  The parameters will
be discarded later.

No functional change.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-5-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Use the stored data in perf_mmap__read_event()

Using the 'start', 'end' and 'overwrite' which are stored in
struct perf_mmap to replace the parameters of perf_mmap__read_event().
The parameters will be discarded later.

No functional change.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-4-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Use the stored scope data in perf_mmap__push()

Using the 'start' and 'end' which are stored in struct perf_mmap to
replace the temporary 'start' and 'end'.
The temporary variables will be discarded later.

It doesn't need to pass 'overwrite' to perf_mmap__push(). It's stored in
struct perf_mmap.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-3-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Store mmap scope in struct perf_mmap()

There is too much boilerplate in the perf_mmap__read*() interfaces.

The 'start' and 'end' variables should be stored in struct perf_mmap at
initialization. They will be used later.

The old 'startp' and 'endp' pointers are used by perf_mmap__read_event()
now.  They cannot be removed. So the old 'startp/endp' and new
'md->start/md->end' will exist simultaneously now.  The old one will be
removed later.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-2-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Store 'overwrite' in struct perf_mmap

It has been determined that the map is for overwrite mode
(evlist->overwrite_mmap) or non-overwrite mode (evlist->mmap) when
calling perf_evlist__alloc_mmap().

Store the information in struct perf_mmap, which will be used later to
simplify the perf_mmap__read*() interfaces.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Suggested-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1520350567-80082-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmu: Auto-merge PMU events created by prefix or glob match

Auto-merge for these events was disabled when auto-merging of non-alias
events was disabled in commit 63ce844 (perf stat: Only auto-merge events
that are PMU aliases).

Non-merging of legacy events is preserved:

    $ perf stat -ag -e cache-misses,cache-misses sleep 1

     Performance counter stats for 'system wide':

                86,323      cache-misses
                86,323      cache-misses

           1.002623307 seconds time elapsed

But prefix or glob matching auto-merges the events created:

    $ perf stat -a -e l3cache/read-miss/ sleep 1

     Performance counter stats for 'system wide':

                   328      l3cache/read-miss/

           1.002627008 seconds time elapsed

    $ perf stat -a -e l3cache_0_[01]/read-miss/ sleep 1

     Performance counter stats for 'system wide':

                   172      l3cache/read-miss/

           1.002627008 seconds time elapsed

As with events created with aliases, auto-merging can be suppressed with
the --no-merge option:

    $ perf stat -a -e l3cache/read-miss/ --no-merge sleep 1

     Performance counter stats for 'system wide':

                    67      l3cache/read-miss/
                    67      l3cache/read-miss/
                    63      l3cache/read-miss/
                    60      l3cache/read-miss/

           1.002622192 seconds time elapsed

Signed-off-by: Agustin Vega-Frias <agustinv@codeaurora.org>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Timur Tabi <timur@codeaurora.org>
Cc: linux-arm-kernel@lists.infradead.org
Change-Id: I0a47eed54c05e1982ca964d743b37f50f60c508c
Link: http://lkml.kernel.org/r/1520345084-42646-4-git-send-email-agustinv@codeaurora.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmu: Display pmu name when printing unmerged events in stat

To simplify creation of events accross multiple instances of the same
type of PMU stat supports two methods for creating multiple events from
a single event specification:

1. A prefix or glob can be used in the PMU name.
2. Aliases, which are listed immediately after the Kernel PMU events
   by perf list, are used.

When the --no-merge option is passed and these events are displayed
individually the PMU name is lost and it's not possible to see which
count corresponds to which pmu:

    $ perf stat -a -e l3cache/read-miss/ --no-merge ls > /dev/null

     Performance counter stats for 'system wide':

                    67      l3cache/read-miss/
                    67      l3cache/read-miss/
                    63      l3cache/read-miss/
                    60      l3cache/read-miss/

           0.001675706 seconds time elapsed

    $ perf stat -a -e l3cache_read_miss --no-merge ls > /dev/null

     Performance counter stats for 'system wide':

                    12      l3cache_read_miss
                    17      l3cache_read_miss
                    10      l3cache_read_miss
                     8      l3cache_read_miss

           0.001661305 seconds time elapsed

This change adds the original pmu name to the event. For dynamic pmu
events the pmu name is restored in the event name:

    $ perf stat -a -e l3cache/read-miss/ --no-merge ls > /dev/null

     Performance counter stats for 'system wide':

                    63      l3cache_0_3/read-miss/
                    74      l3cache_0_1/read-miss/
                    64      l3cache_0_2/read-miss/
                    74      l3cache_0_0/read-miss/

           0.001675706 seconds time elapsed

For alias events the name is added after the event name:

    $ perf stat -a -e l3cache_read_miss --no-merge ls > /dev/null

     Performance counter stats for 'system wide':

                    10      l3cache_read_miss [l3cache_0_3]
                    12      l3cache_read_miss [l3cache_0_1]
                    10      l3cache_read_miss [l3cache_0_2]
                    17      l3cache_read_miss [l3cache_0_0]

           0.001661305 seconds time elapsed

Signed-off-by: Agustin Vega-Frias <agustinv@codeaurora.org>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Timur Tabi <timur@codeaurora.org>
Cc: linux-arm-kernel@lists.infradead.org
Change-Id: I8056b9eda74bda33e95065056167ad96e97cb1fb
Link: http://lkml.kernel.org/r/1520345084-42646-3-git-send-email-agustinv@codeaurora.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmu: Support wildcards on pmu name in dynamic pmu events

Starting on v4.12 event parsing code for dynamic pmu events already
supports prefix-based matching of multiple pmus when creating dynamic
events. E.g., in a system with the following dynamic pmus:

    mypmu_0
    mypmu_1
    mypmu_2
    mypmu_4

passing mypmu/<config>/ as an event spec will result in the creation of
the event in all of the pmus. This change expands this matching through
the use of fnmatch so glob-like expressions can be used to create events
in multiple pmus. E.g., in the system described above if a user only
wants to create the event in mypmu_0 and mypmu_1, mypmu_[01]/<config>/
can be passed.

Signed-off-by: Agustin Vega-Frias <agustinv@codeaurora.org>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Timur Tabi <timur@codeaurora.org>
Change-Id: Icb25653fc5d5239c20f3bffdfdf4ab4c9c9bb20b
Link: http://lkml.kernel.org/r/1520454947-16977-1-git-send-email-agustinv@codeaurora.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Correct title markers for asciidoctor

I've tested to process the perf man pages with asciidoctor that is
picker than asciidoc, and it revealed minor syntax errors in some
documents.  Namely, the title markers aren't aligned with the previous
line, hence asciidoctor didn't recognize as titles.

This patch corrects these markers to be processed properly.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307105441.28512-1-tiwai@suse.de
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Make auxtrace_queues__add_buffer() return buffer_ptr

In preparation for supporting AUX area sampling buffers,
auxtrace_queues__add_buffer() needs to be more generic. To that end, make
it return buffer_ptr instead of the caller.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520327598-1317-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Rename some buffer-queuing functions

Rename some buffer-queuing functions in preparation for supporting AUX area
sampling buffers.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520327598-1317-5-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Add missing parameters from kernel-doc comments

Add missing parameters from kernel-doc comments.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520327598-1317-4-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Support setting cgroups as targets

One can set a cgroup as a default cgroup to be used by all events or
set cgroups with the 'perf stat' and 'perf record' behaviour, i.e.
'-G A' will be the cgroup for events defined so far in the command line.

Here in my main machine, with a kvm instance running a rhel6 guinea pig
I have:

  # ls -la /sys/fs/cgroup/perf_event/ | grep drw
  drwxr-xr-x. 14 root root 360 Mar  6 12:04 ..
  drwxr-xr-x.  3 root root   0 Mar  6 15:05 machine.slice
  #

So I can go ahead and use that cgroup hierarchy, say lets see what
syscalls are being emitted by threads in that 'machine.slice' hierarchy
that are taking more than 100ms:

  # perf trace --duration 100 -G machine.slice
     0.188 (249.850 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
   250.274 (249.743 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
   500.224 (249.755 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
   750.097 (249.934 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  1000.244 (249.780 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  1250.197 (249.796 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  1500.124 (249.859 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  1750.076 (172.900 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
   902.570 (1021.116 ms): qemu-system-x8/23667 ppoll(ufds: 0x558151e03180, nfds: 74, tsp: 0x7ffc00cd0900, sigsetsize: 8) = 1
  1923.825 (305.133 ms): qemu-system-x8/23667 ppoll(ufds: 0x558151e03180, nfds: 74, tsp: 0x7ffc00cd0900, sigsetsize: 8) = 1
  2000.172 (229.002 ms): CPU 0/KVM/23744 ioctl(fd: 16<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
^C  #

If we look inside that cgroup hierarchy we get:

  # ls -la /sys/fs/cgroup/perf_event/machine.slice/ | grep drw
  drwxr-xr-x. 3 root root 0 Mar  6 15:05 .
  drwxr-xr-x. 2 root root 0 Mar  6 16:16 machine-qemu\x2d2\x2drhel6.sandy.scope
  #

There is just one, but lets say there were more and we would want to see
5 seconds worth of syscall summary for the threads in that cgroup:

  # perf trace --summary -G machine.slice/machine-qemu\\x2d2\\x2drhel6.sandy.scope/ -a sleep 5

   Summary of events:

     qemu-system-x86 (23667), 143858 events, 24.2%

     syscall            calls    total       min       avg       max      stddev
                                 (msec)    (msec)    (msec)    (msec)        (%)
     --------------- -------- --------- --------- --------- ---------     ------
     ppoll              28492  4348.631     0.000     0.153    11.616      1.05%
     futex              19661   140.801     0.001     0.007     2.993      3.20%
     read               18440    68.084     0.001     0.004     1.653      4.33%
     ioctl               5387    24.768     0.002     0.005     0.134      1.62%

     CPU 0/KVM (23744), 449455 events, 75.8%

     syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
     --------------- -------- --------- --------- --------- ---------     ------
     ioctl             148364  3401.812     0.000     0.023    11.801      1.15%
     futex              36131   404.127     0.001     0.011     7.377      2.63%
     writev             29452   339.688     0.003     0.012     1.740      1.36%
     write              11315    45.992     0.001     0.004     0.105      1.10%

  #

See the documentation about how to set more than one cgroup for
different events in the same command line.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-t126jh4occqvu0xdqlcjygex@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Make the cgroup name be const char *

The usual thing is for a constructor to allocate space for its members,
not to require that the caller pass a pre-allocated 'name' and then, at
its destructor, to free something not allocated by it.

Fix it by making cgroup__new() to receive a const char pointer, then
allocate cgroup->name that then can continue to be freed at
cgroup__delete(), balancing the alloc/free operations inside the cgroup
struct methods.

This eases calling evlist__findnew_cgroup() from the custom 'perf trace'
cgroup parser, that will only call parse_cgroups() when the '-G cgroup'
is passed on the command line after '-e event' entries, when it'll
behave just like 'perf stat' and 'perf record', i.e. the previous
parse_cgroup() users that mandate that -G only can come after a -e.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-4leugnuyqi10t98990o3xi1t@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Add evlist__add_default_cgroup()

So that tools like 'perf trace' can allow the user to set a cgroup
to be used for all the evsels still without a crgroup setup by
parse_cgroups(), such as the one to use for the syscalls, vfs_getname
and other events involved in strace like syscall tracing.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-zf9jjsbj661r3lk6qb7g8j70@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Add evlist__findnew_cgroup()

Similar to machine__findnew_thread(), etc, i.e. try to find, get a
refcount if found and return it, otherwise return a new cgroup object.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-im1omevlihhyneiic4nl3g24@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Combine some auxtrace initialization into a single function

In preparation for adding AUX area sampling support, combine some
auxtrace initialization into a single function.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520327598-1317-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sched map: Re-annotate shortname if thread comm changed

This is to show the real name of thread that created via fork-exec.  See
below example for shortname *A0*.

$ sudo ./perf sched map
              *A0   80393.050639 secs A0 => perf:22368
          *.   A0   80393.050748 secs .  => swapper:0
           .  *.    80393.050887 secs
      *B0  .   .    80393.052735 secs B0 => rcu_sched:8
      *.   .   .    80393.052743 secs
       .  *C0  .    80393.056264 secs C0 => kworker/2:1H:287
       .  *A0  .    80393.056270 secs
       .  *D0  .    80393.056769 secs D0 => ksoftirqd/2:22
-      .  *A0  .    80393.056804 secs
+      .  *A0  .    80393.056804 secs A0 => pi:22368
       .  *.   .    80393.056854 secs
      *B0  .   .    80393.060727 secs
      ...

Signed-off-by: Changbin Du <changbin.du@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1520307457-23668-3-git-send-email-changbin.du@intel.com
[ Optimally pack struct thread_runtime when adding the new bool member ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sched: Move thread::shortname to thread_runtime

The thread::shortname only used by sched command, so move it to sched
private structure.

Signed-off-by: Changbin Du <changbin.du@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1520307457-23668-2-git-send-email-changbin.du@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Introduce cgroup__new() out of open coded equivalent

To follow the namespacing convention in tools/perf.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-jaalyl6bkvvji4r5u8wqw4n4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Introduce find_cgroup() method

To break down complexity in add_cgroup().

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-5yqshcf5hm837n7c86u7lhjf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Introduce cgroup__get()

The refcount operation counterpart to cgroup__put(), use it when reusing
a cgroup.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-14ynvrl7y2cz8gyuy5q5v41g@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Rename close_cgroup() to cgroup__put()

It is not really closing the cgroup, but instead dropping a reference
count and if it hits zero, then calling delete, which will, among other
cleanup shores, close the cgroup fd.

So it is really dropping a reference to that cgroup, and the method name
for that is "put", so rename close_cgroup() to cgroup__put() to follow
this naming convention.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-sccxpnd7bgwc1llgokt6fcey@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Introduce cgroup__delete()

Just to make this code look more like other places in tools/perf.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-j3j72vvn2d5j7tenlghdy195@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Rename 'struct cgroup_sel' to 'struct cgroup'

That name isn't used, is shorter, lets switch to it.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-e51yphwgvepd1y4f5fjptmjq@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cgroup: Remove misplaced __maybe_unused

The 'opt' parameter in parse_cgroups() _is_ used. The original patch
used '__used' that was even more confusing :-)

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 023695d96ee0 ("perf tool: Add cgroup support")
Link: https://lkml.kernel.org/n/tip-4jo2puz0empkoou6bbq460tl@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge branch 'perf/urgent' into perf/core, to resolve conflict

Conflicts:
tools/perf/perf.h

Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'perf-urgent-for-mingo-4.16-20180306' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

- Be more robust when drawing arrows in the annotation TUI, avoiding a
  segfault when jump instructions have as a target addresses in functions
  other that the one currently being annotated. The full fix will come in
  the following days, when jumping to other functions will work as call
  instructions (Arnaldo Carvalho de Melo)

- Prevent auxtrace_queues__process_index() from queuing AUX area data for
  decoding when the --no-itrace option has been used (Adrian Hunter)

- Sync copy of kvm UAPI headers and x86's cpufeatures.h (Arnaldo Carvalho de Melo)

- Fix 'perf stat' CSV output format for non-supported counters (Ilya Pronin)

- Fix crash in 'perf record|perf report' pipe mode (Jiri Olsa)

- Fix annoying 'perf top' overwrite fallback message on older kernels (Kan Liang)

- Fix the usage on the 'perf kallsyms' man page (Sangwon Hong)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Fix trigger class trigger_on()

trigger_on() means that the trigger is available but not ready, however
trigger_on() was making it ready. That can segfault if the signal comes
before trigger_ready(). e.g. (USR2 signal delivery not shown)

  $ perf record -e intel_pt//u -S sleep 1
  perf: Segmentation fault
  Obtained 16 stack frames.
  /home/ahunter/bin/perf(sighandler_dump_stack+0x40) [0x4ec550]
  /lib/x86_64-linux-gnu/libc.so.6(+0x36caf) [0x7fa76411acaf]
  /home/ahunter/bin/perf(perf_evsel__disable+0x26) [0x4b9dd6]
  /home/ahunter/bin/perf() [0x43a45b]
  /lib/x86_64-linux-gnu/libc.so.6(+0x36caf) [0x7fa76411acaf]
  /lib/x86_64-linux-gnu/libc.so.6(__xstat64+0x15) [0x7fa7641d2cc5]
  /home/ahunter/bin/perf() [0x4ec6c9]
  /home/ahunter/bin/perf() [0x4ec73b]
  /home/ahunter/bin/perf() [0x4ec73b]
  /home/ahunter/bin/perf() [0x4ec73b]
  /home/ahunter/bin/perf() [0x4eca15]
  /home/ahunter/bin/perf(machine__create_kernel_maps+0x257) [0x4f0b77]
  /home/ahunter/bin/perf(perf_session__new+0xc0) [0x4f86f0]
  /home/ahunter/bin/perf(cmd_record+0x722) [0x43c132]
  /home/ahunter/bin/perf() [0x4a11ae]
  /home/ahunter/bin/perf(main+0x5d4) [0x427fb4]

Note, for testing purposes, this is hard to hit unless you add some sleep()
in builtin-record.c before record__open().

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: stable@vger.kernel.org
Fixes: 3dcc4436fa6f ("perf tools: Introduce trigger class")
Link: http://lkml.kernel.org/r/1519807144-30694-1-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Prevent decoding when --no-itrace

Prevent auxtrace_queues__process_index() from queuing AUX area data for
decoding when the --no-itrace option has been used.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520327598-1317-3-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Fix CVS output format for non-supported counters

When printing stats in CSV mode, 'perf stat' appends extra separators
when a counter is not supported:

<not supported>,,L1-dcache-store-misses,mesos/bd442f34-2b4a-47df-b966-9b281f9f56fc,0,100.00,,,,

Which causes a failure when parsing fields. The numbers of separators
should be the same for each line, no matter if the counter is or not
supported.

Signed-off-by: Ilya Pronin <ipronin@twitter.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/20180306064353.31930-1-xiyou.wangcong@gmail.com
Fixes: 92a61f6412d3 ("perf stat: Implement CSV metrics output")
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-core-for-mingo-4.17-20180305' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

- Be more robust when drawing arrows in the annotation TUI, avoiding a
  segfault when jump instructions have as a target addresses in functions
  other that the one currently being annotated. The full fix will come in
  the following days, when jumping to other functions will work as call
  instructions (Arnaldo Carvalho de Melo)

- Allow asking for the maximum allowed sample rate in 'top' and
  'record', i.e. 'perf record -F max' will read the
  kernel.perf_event_max_sample_rate sysctl and use it (Arnaldo Carvalho de Melo)

- When the user specifies a freq above kernel.perf_event_max_sample_rate,
  Throttle it down to that max freq, and warn the user about it, add as
  well --strict-freq so that the previous behaviour of not starting the
  session when the desired freq can't be used can be selected (Arnaldo Carvalho de Melo)

- Find 'call' instruction target symbol at parsing time, used so far in
  the TUI, part of the infrastructure changes that will end up allowing
  for jumps to navigate to other functions, just like 'call'
  instructions. (Arnaldo Carvalho de Melo)

- Use xyarray dimensions to iterate fds in 'perf stat' (Andi Kleen)

- Ignore threads for which the current user hasn't permissions when
  enabling system-wide --per-thread (Jin Yao)

- Fix some backtrace perf test cases to use 'perf record' + 'perf script'
  instead, till 'perf trace' starts using ordered_events or equivalent
  to avoid symbol resolving artifacts due to reordering of
  PERF_RECORD_MMAP events (Jiri Olsa)

- Fix crash in 'perf record' pipe mode, it needs to allocate the ID
  array even for a single event, unlike non-pipe mode (Jiri Olsa)

- Make annoying fallback message on older kernels with newer 'perf top'
  binaries trying to use overwrite mode and that not being present
  in the older kernels (Kan Liang)

- Switch last users of old APIs to the newer perf_mmap__read_event()
  one, then discard those old mmap read forward APIs (Kan Liang)

- Fix the usage on the 'perf kallsyms' man page (Sangwon Hong)

- Simplify cgroup arguments when tracking multiple events (weiping zhang)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'v4.16-rc4' into perf/core, to pick up fixes

Signed-off-by: Ingo Molnar <mingo@kernel.org>

tools headers: Sync x86's cpufeatures.h

The changes in dd84441a7971 ("x86/speculation: Use IBRS if available
before calling into firmware") don't need any kind of special treatment
in the current tools/perf/ codebase, so just update the copy to get rid
of the perf build warning:

  BUILD:   Doing 'make -j4' parallel build
  Warning: Kernel ABI header at 'tools/arch/x86/include/asm/cpufeatures.h' differs from latest version at 'arch/x86/include/asm/cpufeatures.h'

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: David Woodhouse <dwmw@amazon.co.uk>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-mzmuxocrf96v922xkerey3ns@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools headers: Sync copy of kvm UAPI headers

In 801e459a6f3a ("KVM: x86: Add a framework for supporting MSR-based
features") a new ioctl was introduced, which with this sync of the kvm
UAPI headers, makes 'perf trace' know about it:

  $ cd /tmp/build/perf/trace/beauty/generated/ioctl/
  $ diff -u kvm_ioctl_array.c.old kvm_ioctl_array.c
  --- /tmp/kvm_ioctl_array.c 2018-03-05 11:55:38.409145056 -0300
  +++ /tmp/build/perf/trace/beauty/generated/ioctl/kvm_ioctl_array.c 2018-03-05 11:56:17.456153501 -0300
  @@ -6,6 +6,7 @@
  [0x04] = "GET_VCPU_MMAP_SIZE",
  [0x05] = "GET_SUPPORTED_CPUID",
  [0x09] = "GET_EMULATED_CPUID",
  + [0x0a] = "GET_MSR_FEATURE_INDEX_LIST",
  [0x40] = "SET_MEMORY_REGION",
  [0x41] = "CREATE_VCPU",
  [0x42] = "GET_DIRTY_LOG",

So when using 'perf trace -e ioctl' that will appear along with the
others, like in this excerpt of a system wide session:

  14.556 ( 0.006 ms): CPU 0/KVM/16077 ioctl(fd: 19<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  14.565 ( 0.006 ms): CPU 0/KVM/16077 ioctl(fd: 19<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) = 0
  14.573 (         ): CPU 0/KVM/16077 ioctl(fd: 19<anon_inode:kvm-vcpu:0>, cmd: KVM_RUN) ...
  34.075 ( 0.016 ms): gnome-shell/2192 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_BUSY, arg: 0x7ffe4e73e850) = 0
  40.549 ( 0.012 ms): gnome-shell/2192 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_BUSY, arg: 0x7ffe4e73ece0) = 0
  40.625 ( 0.005 ms): gnome-shell/2192 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_BUSY, arg: 0x7ffe4e73e940) = 0
  40.632 ( 0.003 ms): gnome-shell/2192 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_MADVISE, arg: 0x7ffe4e73e9b0) = 0

This also silences the perf build header copy drift verifier:

  make: Entering directory '/home/acme/git/perf/tools/perf'
    BUILD:   Doing 'make -j4' parallel build
  Warning: Kernel ABI header at 'tools/include/uapi/linux/kvm.h' differs from latest version at 'include/uapi/linux/kvm.h'

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-h31oz5g0mt1dh2s2ajq6o6no@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Fix crash in pipe mode

Currently we can crash perf record when running in pipe mode, like:

  $ perf record ls | perf report
  # To display the perf.data header info, please use --header/--header-only options.
  #
  perf: Segmentation fault
  Error:
  The - file has no samples!

The callstack of the crash is:

    0x0000000000515242 in perf_event__synthesize_event_update_name
  3513            ev = event_update_event__new(len + 1, PERF_EVENT_UPDATE__NAME, evsel->id[0]);
  (gdb) bt
  #0  0x0000000000515242 in perf_event__synthesize_event_update_name
  #1  0x00000000005158a4 in perf_event__synthesize_extra_attr
  #2  0x0000000000443347 in record__synthesize
  #3  0x00000000004438e3 in __cmd_record
  #4  0x000000000044514e in cmd_record
  #5  0x00000000004cbc95 in run_builtin
  #6  0x00000000004cbf02 in handle_internal_command
  #7  0x00000000004cc054 in run_argv
  #8  0x00000000004cc422 in main

The reason of the crash is that the evsel does not have ids array
allocated and the pipe's synthesize code tries to access it.

We don't force evsel ids allocation when we have single event, because
it's not needed. However we need it when we are in pipe mode even for
single event as a key for evsel update event.

Fixing this by forcing evsel ids allocation event for single event, when
we are in pipe mode.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180302161354.30192-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate browser: Be more robust when drawing jump arrows

This first happened with a gcc function, _cpp_lex_token, that has the
usual jumps:

 │1159e6c: ↓ jne    115aa32 <_cpp_lex_token@@Base+0xf92>

I.e. jumps to a label inside that function (_cpp_lex_token), and those
works, but also this kind:

 │1159e8b: ↓ jne    c469be <cpp_named_operator2name@@Base+0xa72>

I.e. jumps to another function, outside _cpp_lex_token, which are not
being correctly handled generating as a side effect references to
ab->offset[] entries that are set to NULL, so to make this code more
robust, check that here.

A proper fix for will be put in place, looking at the function name
right after the '<' token and probably treating this like a 'call'
instruction.

For now just don't draw the arrow.

Reported-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Ingo Molnar <mingo@kernel.org>
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Link: https://lkml.kernel.org/n/tip-5tzvb875ep2sel03aeefgmud@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Fix annoying fallback message on older kernels

On older (e.g. v4.4) kernels, an annoying fallback message can be
observed in 'perf top':

┌─Warning:──────────────────────┐
│fall back to non-overwrite mode│
│                               │
│                               │
│Press any key...               │
└───────────────────────────────┘

The 'perf top' utility has been changed to overwrite mode since commit
ebebbf082357 ("perf top: Switch default mode to overwrite mode").

For older kernels which don't have overwrite mode support, 'perf top'
will fall back to non-overwrite mode and print out the fallback message
using ui__warning(), which needs user's input to close.

The fallback message is not critical for end users. Turning it to debug
message which is printed when running with -vv.

Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Kan Liang <kan.liang@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Fixes: ebebbf082357 ("perf top: Switch default mode to overwrite mode")
Link: http://lkml.kernel.org/r/1519669030-176549-1-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf kallsyms: Fix the usage on the man page

First, all man pages highlight only perf and subcommands except 'perf
kallsyms', which includes the full usage. Fix it for commands to
monopolize underlines.

Second, options can be ommited when executing 'perf kallsyms', so add
square brackets between <option>.

Signed-off-by: Sangwon Hong <qpakzk@gmail.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Taeung Song <treeze.taeung@gmail.com>
Link: http://lkml.kernel.org/r/1518377864-20353-1-git-send-email-qpakzk@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>