tracing: Introduce TRACE_FIELD_ZERO() macro

Use TRACE_FIELD_ZERO(type, item) instead of TRACE_FIELD_ZERO_CHAR(item).
This also includes a typo fix of TRACE_ZERO_CHAR() macro.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203501.31965.30172.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Ftrace dynamic ftrace_event_call support

Add dynamic ftrace_event_call support to ftrace. Trace engines can add
new ftrace_event_call to ftrace on the fly. Each operator function of
the call takes an ftrace_event_call data structure as an argument,
because these functions may be shared among several ftrace_event_calls.

Changes from v13:
 - Define remove_subsystem_dir() always (revirt a2ca5e03), because
   trace_remove_event_call() uses it.
 - Modify syscall tracer because of ftrace_event_call change.

[fweisbec@gmail.com: Fixed conflict against latest tracing/core]

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203453.31965.71901.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

x86: Add pt_regs register and stack access APIs

Add following APIs for accessing registers and stack entries from
pt_regs.
These APIs are required by kprobes-based event tracer on ftrace.
Some other debugging tools might be able to use it too.

- regs_query_register_offset(const char *name)
   Query the offset of "name" register.

- regs_query_register_name(unsigned int offset)
   Query the name of register by its offset.

- regs_get_register(struct pt_regs *regs, unsigned int offset)
   Get the value of a register by its offset.

- regs_within_kernel_stack(struct pt_regs *regs, unsigned long addr)
   Check the address is in the kernel stack.

- regs_get_kernel_stack_nth(struct pt_regs *reg, unsigned int nth)
   Get Nth entry of the kernel stack. (N >= 0)

- regs_get_argument_nth(struct pt_regs *reg, unsigned int nth)
   Get Nth argument at function call. (N >= 0)

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: linux-arch@vger.kernel.org
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203444.31965.26374.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

kprobes: Cleanup fix_riprel() using insn decoder on x86

Cleanup fix_riprel() in arch/x86/kernel/kprobes.c by using the new x86
instruction decoder instead of using comparisons with raw ad hoc numeric
opcodes.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203436.31965.34374.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

kprobes: Checks probe address is instruction boudary on x86

Ensure safeness of inserting kprobes by checking whether the specified
address is at the first byte of an instruction on x86.
This is done by decoding probed function from its head to the probe
point.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203428.31965.21939.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

x86: X86 instruction decoder build-time selftest

Add a user-space selftest of x86 instruction decoder at kernel build
time.
When CONFIG_X86_DECODER_SELFTEST=y, Kbuild builds a test harness of x86
instruction decoder and performs it after building vmlinux.
The test compares the results of objdump and x86 instruction decoder
code and check there are no differences.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203421.31965.29006.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

x86: Instruction decoder API

Add x86 instruction decoder to arch-specific libraries. This decoder
can decode x86 instructions used in kernel into prefix, opcode, modrm,
sib, displacement and immediates. This can also show the length of
instructions.

This version introduces instruction attributes for decoding
instructions.
The instruction attribute tables are generated from the opcode map file
(x86-opcode-map.txt) by the generator script(gen-insn-attr-x86.awk).

Currently, the opcode maps are based on opcode maps in Intel(R) 64 and
IA-32 Architectures Software Developers Manual Vol.2: Appendix.A,
and consist of below two types of opcode tables.

1-byte/2-bytes/3-bytes opcodes, which has 256 elements, are
written as below;

 Table: table-name
 Referrer: escaped-name
 opcode: mnemonic|GrpXXX [operand1[,operand2...]] [(extra1)[,(extra2)...] [| 2nd-mnemonic ...]
  (or)
 opcode: escape # escaped-name
 EndTable

Group opcodes, which has 8 elements, are written as below;

 GrpTable: GrpXXX
 reg:  mnemonic [operand1[,operand2...]] [(extra1)[,(extra2)...] [| 2nd-mnemonic ...]
 EndTable

These opcode maps include a few SSE and FP opcodes (for setup), because
those opcodes are used in the kernel.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jason Baron <jbaron@redhat.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Przemysław Pawełczyk <przemyslaw@pawelczyk.it>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
LKML-Reference: <20090813203413.31965.49709.stgit@localhost.localdomain>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

Merge branch 'tracing/core' of git://git./linux/kernel/git/frederic/random-tracing into tracing/core

Conflicts:
include/linux/tracepoint.h

Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing: add comments to explain TRACE_EVENT out of protection

The commit:
  commit 5ac35daa9343936038a3c9c4f4d6d3fe6a2a7bd8
  Author: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
  tracing/events: fix the include file dependencies

Moved the TRACE_EVENT out of the ifdef protection of tracepoints.h
but uses the define of TRACE_EVENT itself as protection. This patch
adds comments to explain why.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing/events: fix the include file dependencies

The TRACE_EVENT depends on the include/linux/tracepoint.h first
and include/trace/ftrace.h later, if we include the ftrace.h early,
a building error will occur.

Both define TRACE_EVENT in trace_a.h and trace_b.h, if we include
those in .c file, like this:

#define CREATE_TRACE_POINTS
include <trace/events/trace_a.h>
include <trace/events/trace_b.h>

The above will not work, because the TRACE_EVENT was re-defined by
the previous .h file.

Reported-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
LKML-Reference: <4A937F5E.3020802@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

ftrace: Move setting of clock-source out of options

There are many clock sources for the tracing system but we can only
enable/disable one at a time with the trace/options file.
We can move the setting of clock-source out of options and add a separate
file for it:
 # cat trace_clock
 [local] global
 # echo global > trace_clock
 # cat trace_clock
 local [global]

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
LKML-Reference: <4A939D08.6050604@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing/filters: Support filtering for char * strings

Usually, char * entries are dangerous in traces because the string
can be released whereas a pointer to it can still wait to be read from
the ring buffer.

But sometimes we can assume it's safe, like in case of RO data
(eg: __file__ or __line__, used in bkl trace event). If these RO data
are in a module and so is the call to the trace event, then it's safe,
because the ring buffer will be flushed once this module get unloaded.

To allow char * to be treated as a string:

TRACE_EVENT(...,

TP_STRUCT__entry(
__field_ext(const char *, name, FILTER_PTR_STRING)
...
)

...
);

The filtering will not dereference "char *" unless the developer
explicitly sets FILTER_PTR_STR in __field_ext.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A7B9287.90205@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing/filters: Add __field_ext() to TRACE_EVENT

Add __field_ext(), so a field can be assigned to a specific
filter_type, which matches a corresponding filter function.

For example, a later patch will allow this:
__field_ext(const char *, str, FILTER_PTR_STR);

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A7B9272.6050709@cn.fujitsu.com>

[
  Fixed a -1 to FILTER_OTHER
  Forward ported to latest kernel.
]

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing/filters: Add filter_type to struct ftrace_event_field

The type of a field is stored as a string in @type, and here
we add @filter_type which is an enum value.

This prepares for later patches, so we can specifically assign
different @filter_type for the same @type.

For example normally a "char *" field is treated as a ptr,
but we may want it to be treated as a string when doing filting.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A7B925E.9030605@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing: Add vim script to enable folding for function_graph traces

function_graph traces look like nested function calls, complete with
braces denoting the start and end of functions.  function-graph-fold.vim
teaches vim how to fold these functions, to make it more convenient to
browse them.

To use, :source function-graph-fold.vim while viewing a function_graph
trace, or use "view -S function-graph-fold.vim some-trace" to load it
from the command-line together with a trace.  You can then use the usual
vim fold commands, such as "za", to open and close nested functions.
While closed, a fold will show the total time taken for a call, as would
normally appear on the line with the closing brace.  Folded functions
will not include finish_task_switch(), so folding should remain
relatively sane even through a context switch.

Note that this will almost certainly only work well with a single-CPU
trace (e.g. trace-cmd report --cpu 1).  It also takes some time to run
(a few seconds for a large trace on my laptop).  Nevertheless, I found
it very handy to get an overview of a trace and then drill down on
problematic calls.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
LKML-Reference: <20090806145701.GB7661@feather>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing/sched: show CPU task wakes up on in trace event

While debugging the scheduler push / pull algorithm, I found
it very annoying that the sched wake up events did not show
the CPU that the task was waking on. In order to analyze the
scheduler, I needed that information.

This patch adds recording of the CPU that a task is waking up
on.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing: Create generic syscall TRACE_EVENTs

This converts the syscall_enter/exit tracepoints into TRACE_EVENTs, so
you can have generic ftrace events that capture all system calls with
arguments and return values.  These generic events are also renamed to
sys_enter/exit, so they're more closely aligned to the specific
sys_enter_foo events.

Signed-off-by: Josh Stone <jistone@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
LKML-Reference: <1251150194-1713-5-git-send-email-jistone@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Move tracepoint callbacks from declaration to definition

It's not strictly correct for the tracepoint reg/unreg callbacks to
occur when a client is hooking up, because the actual tracepoint may not
be present yet.  This happens to be fine for syscall, since that's in
the core kernel, but it would cause problems for tracepoints defined in
a module that hasn't been loaded yet.  It also means the reg/unreg has
to be EXPORTed for any modules to use the tracepoint (as in SystemTap).

This patch removes DECLARE_TRACE_WITH_CALLBACK, and instead introduces
DEFINE_TRACE_FN which stores the callbacks in struct tracepoint.  The
callbacks are used now when the active state of the tracepoint changes
in set_tracepoint & disable_tracepoint.

This also introduces TRACE_EVENT_FN, so ftrace events can also provide
registration callbacks if needed.

Signed-off-by: Josh Stone <jistone@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
LKML-Reference: <1251150194-1713-4-git-send-email-jistone@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Make syscall tracepoints conditional

The syscall enter/exit tracepoints are only supported on archs that
HAVE_SYSCALL_TRACEPOINTS, so the declarations should be #ifdef'ed.
Also, the definition of syscall_regfunc and syscall_unregfunc should
depend on this same config, rather than the ftrace-specific one.

Signed-off-by: Josh Stone <jistone@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <1251150194-1713-3-git-send-email-jistone@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Rename FTRACE_SYSCALLS for tracepoints

s/HAVE_FTRACE_SYSCALLS/HAVE_SYSCALL_TRACEPOINTS/g
s/TIF_SYSCALL_FTRACE/TIF_SYSCALL_TRACEPOINT/g

The syscall enter/exit tracing is no longer specific to just ftrace, so
they now have names that reflect their tie to tracepoints instead.

Signed-off-by: Josh Stone <jistone@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
LKML-Reference: <1251150194-1713-2-git-send-email-jistone@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Eliminate code duplication in kernel/tracepoint.c

Signed-off-by: Anirban Sinha <asinha@zeugmasystems.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: "Oleg Nesterov" <oleg@tv-sign.ru>
LKML-Reference: <DDFD17CC94A9BD49A82147DDF7D545C501EA9047@exchange.ZeugmaSystems.local>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/syscalls: Fix the output of syscalls with no arguments

Before:

  # echo 1 > events/syscalls/sys_enter_sync/enable
  # cat events/syscalls/sys_enter_sync/format
  ...
        field:int nr;   offset:12;      size:4;

  print fmt: "# sync
  # cat trace
  ...
            sync-8950  [000]  2366.087670: sys_sync(

After:

  # echo 1 > events/syscalls/sys_enter_sync/enable
  # cat events/syscalls/sys_enter_sync/format
  ...
        field:int nr;   offset:12;      size:4;

  print fmt: ""
  # sync
  # cat trace
            sync-2134  [001]   136.780735: sys_sync()

Reported-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
LKML-Reference: <4A8D05AF.20103@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/syscalls: Add filtering support

Add filtering support for syscall events:

 # echo 'mode == 0666' > events/syscalls/sys_enter_open
 # echo 'ret == 0' > events/syscalls/sys_exit_open
 # echo 1 > events/syscalls/sys_enter_open
 # echo 1 > events/syscalls/sys_exit_open
 # cat trace
 ...
   modprobe-3084 [001] 117.463140: sys_open(filename: 917d3e8, flags: 0, mode: 1b6)
   modprobe-3084 [001] 117.463176: sys_open -> 0x0
       less-3086 [001] 117.510455: sys_open(filename: 9c6bdb8, flags: 8000, mode: 1b6)
   sendmail-2574 [001] 122.145840: sys_open(filename: b807a365, flags: 0, mode: 1b6)
 ...

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A8BAFCB.1040006@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/events: Add trace_define_common_fields()

Extract duplicate code. Also prepare for the later patch.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A8BAFB8.1010304@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/events: Add ftrace_event_call param to define_fields()

This parameter is needed by syscall events to add define_fields()
handler.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A8BAF90.6060801@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/syscalls: Add fields format for exit events

Add "format" file for syscall exit events:

 # cat events/syscalls/sys_exit_open/format
 name: sys_exit_open
 ID: 344
 format:
         field:unsigned short common_type;       offset:0;       size:2;
         field:unsigned char common_flags;       offset:2;       size:1;
         field:unsigned char common_preempt_count;       offset:3;       size:1;
         field:int common_pid;   offset:4;       size:4;
         field:int common_tgid;  offset:8;       size:4;

         field:int nr;   offset:12;      size:4;
         field:unsigned long ret;        offset:16;      size:4;

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A8BAF61.3060307@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/syscalls: Fix fields format for enter events

The "format" file of a trace event is originally for parsers to
parse ftrace binary output.

But the "format" file of a syscall event can only be used by
perfcounter, because it describes the format of struct
syscall_enter_record not struct syscall_trace_enter.

To fix this, we remove struct syscall_enter_record, and then
struct syscall_trace_enter will be used by both perf profile
and ftrace.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A8BAF39.1030404@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[S390] ftrace: update system call tracer support

Commit fb34a08c3 ("tracing: Add trace events for each syscall
entry/exit") changed the lowlevel API to ftrace syscall tracing
but did not update s390 which started making use of it recently.

This broke the s390 build, as reported by Paul Mundt.

Update the callbacks with the syscall number and the syscall
return code values. This allows per syscall tracepoints,
syscall argument enumeration /debug/tracing/events/syscalls/
and perfcounters support and integration on s390 too.

Reported-by: Paul Mundt <lethal@linux-sh.org>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <tip-fb34a08c3469b2be9eae626ccb96476b4687b810@git.kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

ftrace: Simplify seqfile code

Use seq_release_private().

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4A891AAB.8090701@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

trace_stack: Simplify seqfile code

Extract duplicate code in t_start() and t_next().

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A891A91.4030602@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

trace_stat: Fix missing entry in stat file

One entry is missing in the output of a stat file.

The cause is, when stat_seq_start() is called the 2nd time, we
should start from the (pos-1)th elem in the rbtree but not pos,
because pos == 0 is the header.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <4A891A65.70009@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/syscalls: Fix to print parameter types

When syscall tracing was implemented as a tracer,
"syscall_arg_type" trace option could be set to enable the
display of syscall parameter types.

Now this option is gone since it's no longer a tracer, but the
code is still there but dead.

So we remove dead code and re-enable the printing of paramete
types via the verbose option:

  # echo verbose > trace_options
  # echo syscalls > set_event
  # cat trace
...
        bash-3331  [000]    95.348937: sys_fcntl64 -> 0x1
        bash-3331  [000]    95.348942: sys_close(unsigned int fd: a)
...

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jason Baron <jbaron@redhat.com>
LKML-Reference: <4A891AF6.5050102@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing/events: Add module tracepoints

Add trace points to trace module_load, module_free, module_get,
module_put and module_request, and use trace_event facility to
get the trace output.

Here's the sample output:

     TASK-PID    CPU#    TIMESTAMP  FUNCTION
        | |       |          |         |
    <...>-42    [000]     1.758380: module_request: fb0 wait=1 call_site=fb_open
    ...
    <...>-60    [000]     3.269403: module_load: scsi_wait_scan
    <...>-60    [000]     3.269432: module_put: scsi_wait_scan call_site=sys_init_module refcnt=0
    <...>-61    [001]     3.273168: module_free: scsi_wait_scan
    ...
    <...>-1021  [000]    13.836081: module_load: sunrpc
    <...>-1021  [000]    13.840589: module_put: sunrpc call_site=sys_init_module refcnt=-1
    <...>-1027  [000]    13.848098: module_get: sunrpc call_site=try_module_get refcnt=0
    <...>-1027  [000]    13.848308: module_get: sunrpc call_site=get_filesystem refcnt=1
    <...>-1027  [000]    13.848692: module_put: sunrpc call_site=put_filesystem refcnt=0
    ...
 modprobe-2587  [001]  1088.437213: module_load: trace_events_sample F
 modprobe-2587  [001]  1088.437786: module_put: trace_events_sample call_site=sys_init_module refcnt=0

Note:

- the taints flag can be 'F', 'C' and/or 'P' if mod->taints != 0

- the module refcnt is percpu, so it can be negative in a
  specific cpu

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
LKML-Reference: <4A891B3C.5030608@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing: Fix syscall tracing on !HAVE_FTRACE_SYSCALLS architectures

The new syscall_regfunc()/unregfunc() functions rely on
the existence of TIF_SYSCALL_FTRACE - but that TIF flag
is only offered by HAVE_FTRACE_SYSCALLS.

Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tracing: Support for syscall events raw records in perfcounters

This bring the support for raw syscall events in perfcounters.
The arguments or exit value are saved as a raw sample using
the PERF_SAMPLE_RAW attribute in a perf counter.

Example (for now you must explicitly set the PERF_SAMPLE_RAW flag
in perf record):

perf record -e syscalls:sys_enter_open -f -F 1 -a
perf report -D

0x2cbb8 [0x50]: event: 9
.
. ... raw event: size 80 bytes
.  0000:  09 00 00 00 02 00 50 00 20 e9 39 ab 0a 7f 00 00  ......P. .9....
.  0010:  bc 14 00 00 bc 14 00 00 01 00 00 00 00 00 00 00  ...............
.  0020:  2c 00 00 00 15 01 01 00 bc 14 00 00 bc 14 00 00  ,..............
                  ^  ^  ^  ^  ^  ^  ^  ..........................
                  Event Size  struct trace_entry

.  0030:  00 00 00 00 46 98 43 02 00 00 00 00 80 08 00 00  ....F.C........
                  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^
                  ptr to file name        open flags

.  0040:  00 00 00 00 02 00 00 00 00 00 00 00 00 00 00 00  ...............
                  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^  ^
.         open mode               padding

0x2cbb8 [0x50]: PERF_EVENT_SAMPLE (IP, 2): 5308: 0x7f0aab39e920 period: 1

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>

tracing: Add fields format definition for syscall events

Define the format of the syscall trace fields to parse the binary
values from a raw trace using the syscall events "format" file.

This is defined dynamically using the syscalls metadata.
It prepares the export of syscall event raw records to perf
counters.

Example:

$ cat /debug/tracing/events/syscalls/sys_enter_sched_getparam/format
name: sys_enter_sched_getparam
ID: 39
format:
field:unsigned short common_type; offset:0; size:2;
field:unsigned char common_flags; offset:2; size:1;
field:unsigned char common_preempt_count; offset:3; size:1;
field:int common_pid; offset:4; size:4;
field:int common_tgid; offset:8; size:4;

field:pid_t pid; offset:12; size:8;
field:struct sched_param * param; offset:20; size:8;

print fmt: "pid: 0x%08lx, param: 0x%08lx", ((unsigned long)(REC->pid)), ((unsigned long)(REC->param))

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>

tracing: Add ftrace event call parameter to its field descriptor handler

Add the struct ftrace_event_call as a parameter of its show_format()
callback. This way we can use it from the syscall trace events to
retrieve the syscall name from the ftrace event call parameter and
describe its fields using the syscalls metadata.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>

tracing: Convert x86_64 mmap and uname to use DEFINE_SYSCALL

A number of syscalls are not using 'DEFINE_SYSCALL'. I'm not sure why.
Convert x86_64 uname and mmap to use DEFINE_SYSCALL.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add more namespace area to 'perf list' output

The new syscall tracepoints names can be too long for the 'perf list'
output.
Add a few more characters.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add perf counter support for syscalls tracing

The perf counter support is automated for usual trace events. But we
have to define specific callbacks for this to handle syscalls trace
events

Make 'perf stat -e syscalls:sys_enter_blah' work with syscall style
tracepoints.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add individual syscalls tracepoint id support

The current state of syscalls tracepoints generates only one event id
for every syscall events.

This patch associates an id with each syscall trace event, so that we
can identify each syscall trace event using the 'perf' tool.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add trace events for each syscall entry/exit

Layer Frederic's syscall tracer on tracepoints. We create trace events
via hooking into the SYSCALL_DEFINE macros. This allows us to
individually toggle syscall entry and exit points on/off.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add ftrace_event_call void * 'data' field

add an optional void * pointer to 'ftrace_event_call' that is
passed in for regfunc and unregfunc.

This prepares for syscall tracepoints creation by passing the name of
the syscall we want to trace and then retrieve its number through our
arch syscall table.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Raw_init() bailout in trace event register fail case

Allow the return value of raw_init() trace event callback to bail us out
of creating a trace event file, in case we fail to register our
event.

Also, we plan to return -ENOSYS for syscall events that don't match any
syscalls listed in our arch tracing syscall table, we don't want to warn
in that case, we just want this event to be invisible in debugfs and
ignored.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Update FTRACE_SYSCALL_MAX

update FTRACE_SYSCALL_MAX to the current number of syscalls

FTRACE_SYSCALL_MAX is a temporary solution to get the number of
syscalls supported by the arch until we find a more dynamic way
to get this number.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add syscall tracepoints

add two tracepoints in syscall exit and entry path, conditioned on
TIF_SYSCALL_FTRACE. Supports the syscall trace event code.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Add DECLARE_TRACE_WITH_CALLBACK() macro

Introduce a new 'DECLARE_TRACE_WITH_CALLBACK()' macro, so that
tracepoints can associate an external register/unregister function.

This prepares for the syscalls tracer conversion to trace events. We
will need to perform arch level operations once a syscall event is
turned on/off, such as TIF flags setting, hence the need of such
specific callbacks.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Call arch_init_ftrace_syscalls at boot

Call arch_init_ftrace_syscalls at boot, so we can determine early the
set of syscalls for the syscall trace events.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Map syscall name to number

Add a new function to support translating a syscall name to number at
runtime.
This allows the syscall event tracer to map syscall names to number.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Jiaying Zhang <jiayingz@google.com>
Cc: Martin Bligh <mbligh@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

tracing: Rename set_tracer_flags()'s local variable trace_flags

set_tracer_flags() have a local variable named trace_flags which has
the same name than a global one in the same scope.
This leads to confusion, using tracer_flags should be better by its
meaning.

Changelog:
v1->v2: Simplified another patch in this patchset, no change in this
        patch.

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>

Merge branch 'linus' into tracing/core

Conflicts:
kernel/trace/trace_events_filter.c

We use the tracing/core version.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

pty: fix data loss when stopped (^S/^Q)

Commit d945cb9cc ("pty: Rework the pty layer to use the normal buffering
logic") dropped the test for 'tty->stopped' in pty_write_room(), which
then causes the n_tty line discipline thing to not throttle the data
properly when the tty is stopped.

So instead of pausing the write due to the tty being stopped, the ldisc
layer would go ahead and push it down to the pty.  The pty write()
routine would then refuse to take the data (because it _did_ check
'stopped'), and the data wouldn't actually be written.

This whole stopped test should eventually be moved into the tty ldisc
layer rather than have low-level tty drivers care about these things,
but right now the fix is to just re-instate the missing pty 'stopped'
handling.

Reported-and-tested-by: Artur Skawina <art.08.09@gmail.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'perfcounters-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'perfcounters-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (27 commits)
  perf_counter: Zero dead bytes from ftrace raw samples size alignment
  perf_counter: Subtract the buffer size field from the event record size
  perf_counter: Require CAP_SYS_ADMIN for raw tracepoint data
  perf_counter: Correct PERF_SAMPLE_RAW output
  perf tools: callchain: Fix bad rounding of minimum rate
  perf_counter tools: Fix libbfd detection for systems with libz dependency
  perf: "Longum est iter per praecepta, breve et efficax per exempla"
  perf_counter: Fix a race on perf_counter_ctx
  perf_counter: Fix tracepoint sampling to be part of generic sampling
  perf_counter: Work around gcc warning by initializing tracepoint record unconditionally
  perf tools: callchain: Fix sum of percentages to be 100% by displaying amount of ignored chains in fractal mode
  perf tools: callchain: Fix 'perf report' display to be callchain by default
  perf tools: callchain: Fix spurious 'perf report' warnings: ignore empty callchains
  perf record: Fix the -A UI for empty or non-existent perf.data
  perf util: Fix do_read() to fail on EOF instead of busy-looping
  perf list: Fix the output to not include tracepoints without an id
  perf_counter/powerpc: Fix oops on cpus without perf_counter hardware support
  perf stat: Fix tool option consistency: rename -S/--scale to -c/--scale
  perf report: Add debug help for the finding of symbol bugs - show the symtab origin (DSO, build-id, kernel, etc)
  perf report: Fix per task mult-counter stat reporting
  ...

Merge branch 'irq-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'irq-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86/irq: Fix move_irq_desc() for nodes without ram

Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86: Fix serialization in pit_expect_msb()

Merge branch 'for-linus' of git://git./linux/kernel/git/jbarnes/pci-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci-2.6:
  PCI hotplug: SGI hotplug: do not use hotplug_slot_attr
  PCI hotplug: SGI hotplug: fix build failure

x86: Fix serialization in pit_expect_msb()

Wei Chong Tan reported a fast-PIT-calibration corner-case:

| pit_expect_msb() is vulnerable to SMI disturbance corner case
| in some platforms which causes /proc/cpuinfo to show wrong
| CPU MHz value when quick_pit_calibrate() jumps to success
| section.

I think that the real issue isn't even an SMI - but the fact
that in the very last iteration of the loop, there's no
serializing instruction _after_ the last 'rdtsc'. So even in
the absense of SMI's, we do have a situation where the cycle
counter was read without proper serialization.

The last check should be done outside the outer loop, since
_inside_ the outer loop, we'll be testing that the PIT has
the right MSB value has the right value in the next iteration.

So only the _last_ iteration is special, because that's the one
that will not check the PIT MSB value any more, and because the
final 'get_cycles()' isn't serialized.

In other words:

 - I'd like to move the PIT MSB check to after the last
   iteration, rather than in every iteration

 - I think we should comment on the fact that it's also a
   serializing instruction and so 'fences in' the TSC read.

Here's a suggested replacement.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Reported-by: "Tan, Wei Chong" <wei.chong.tan@intel.com>
Tested-by: "Tan, Wei Chong" <wei.chong.tan@intel.com>
LKML-Reference: <B28277FD4E0F9247A3D55704C440A140D5D683F3@pgsmsx504.gar.corp.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/security-testing-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/security-testing-2.6:
  mm_for_maps: take ->cred_guard_mutex to fix the race with exec
  mm_for_maps: shift down_read(mmap_sem) to the caller
  mm_for_maps: simplify, use ptrace_may_access()

Merge branch 'merge' of git://git./linux/kernel/git/benh/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
  powerpc/dma: pci_set_dma_mask() shouldn't fail if mask fits in RAM

MN10300: includecheck fix: mn10300, pci.h

Fix the following 'make includecheck' warning:

  arch/mn10300/include/asm/pci.h: linux/mm.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mempool.c: clean up type-casting

clean up type-casting twice.  "size_t" is typedef as "unsigned long" in
64-bit system, and "unsigned int" in 32-bit system, and the intermediate
cast to 'long' is pointless.

Signed-off-by: Figo.zhang <figo1802@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

documentation: register ioctl entry of nilfs2

This will register the ioctl range used by nilfs2 file system to the
table listed in Documentation/ioctl/ioctl-number.txt.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

perf_counter: Zero dead bytes from ftrace raw samples size alignment

After aligning the ftrace raw samples, there are dead bytes storing
random data from the stack. We don't want to leak these to userspace,
then zero these out.

Before:

0x2de88 [0x50]: event: 9
.
. ... raw event: size 80 bytes
.  0000:  09 00 00 00 01 00 50 00 d0 c7 00 81 ff ff ff ff  ......P........
.  0010:  68 01 00 00 68 01 00 00 2c 00 00 00 00 00 00 00  h...h...,......
.  0020:  2c 00 00 00 2b 00 01 02 68 01 00 00 68 01 00 00  ,...+...h...h..
.  0030:  6b 6f 6e 64 65 6d 61 6e 64 2f 30 00 00 00 00 00  kondemand/0....
.  0040:  68 01 00 00 40 7f 46 81 ff ff ff ff 00 10 1b 7f  h...@.F........
                                                      ^  ^  ^  ^
                                                         Leak

After:

0x2d318 [0x50]: event: 9
.
. ... raw event: size 80 bytes
.  0000:  09 00 00 00 01 00 50 00 d0 c7 00 81 ff ff ff ff  ......P........
.  0010:  68 01 00 00 68 01 00 00 68 14 00 00 00 00 00 00  h...h...h......
.  0020:  2c 00 00 00 2b 00 01 02 68 01 00 00 68 01 00 00  ,...+...h...h..
.  0030:  6b 6f 6e 64 65 6d 61 6e 64 2f 30 00 00 00 00 00  kondemand/0....
.  0040:  68 01 00 00 a0 80 46 81 ff ff ff ff 00 00 00 00  h.....F........
                                                      ^  ^  ^  ^
 Fixed

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1249915116-5210-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>

perf_counter: Subtract the buffer size field from the event record size

We compute the perf raw sample size by aligning the raw ftrace
event size plus the buffer size field itself. We do that
instead of aligning only the perf raw sample size, so that we
might economize some in some cases.

But this buffer size field is not stored in the perf raw
sample, we must then substract its size from the buffer once we
computed the alignment unless we may get a useless u32 field in
the buffer.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <20090810141129.GA5124@nowhere>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

mm_for_maps: take ->cred_guard_mutex to fix the race with exec

The problem is minor, but without ->cred_guard_mutex held we can race
with exec() and get the new ->mm but check old creds.

Now we do not need to re-check task->mm after ptrace_may_access(), it
can't be changed to the new mm under us.

Strictly speaking, this also fixes another very minor problem. Unless
security check fails or the task exits mm_for_maps() should never
return NULL, the caller should get either old or new ->mm.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>

mm_for_maps: shift down_read(mmap_sem) to the caller

mm_for_maps() takes ->mmap_sem after security checks, this looks
strange and obfuscates the locking rules. Move this lock to its
single caller, m_start().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>

mm_for_maps: simplify, use ptrace_may_access()

It would be nice to kill __ptrace_may_access(). It requires task_lock(),
but this lock is only needed to read mm->flags in the middle.

Convert mm_for_maps() to use ptrace_may_access(), this also simplifies
the code a little bit.

Also, we do not need to take ->mmap_sem in advance. In fact I think
mm_for_maps() should not play with ->mmap_sem at all, the caller should
take this lock.

With or without this patch, without ->cred_guard_mutex held we can race
with exec() and get the new ->mm but check old creds.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>

perf_counter: Require CAP_SYS_ADMIN for raw tracepoint data

Raw tracepoint data contains various kernel internals and
data from other users, so restrict this to CAP_SYS_ADMIN.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1249896452.17467.75.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter: Correct PERF_SAMPLE_RAW output

PERF_SAMPLE_* output switches should unconditionally output the
correct format, as they are the only way to unambiguously parse
the PERF_EVENT_SAMPLE data.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1249896447.17467.74.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

powerpc/dma: pci_set_dma_mask() shouldn't fail if mask fits in RAM

On an iMac G5, the b43 driver is failing to initialise because trying to
set the dma mask to 30-bit fails. Even though there's only 512MiB of RAM
in the machine anyway:
https://bugzilla.redhat.com/show_bug.cgi?id=514787

We should probably let it succeed if the available RAM in the system
doesn't exceed the requested limit.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

Merge branch 'for-linus' of git://git.infradead.org/ubi-2.6

* 'for-linus' of git://git.infradead.org/ubi-2.6:
  UBI: compatible fallback in absense of sequence numbers
  UBI: fix double free on error path

Merge branch 'kvm-updates/2.6.31' of git://git./virt/kvm/kvm

* 'kvm-updates/2.6.31' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: Avoid redelivery of edge interrupt before next edge
  KVM: MMU: limit rmap chain length
  KVM: ia64: fix build failures due to ia64/unsigned long mismatches
  KVM: Make KVM_HPAGES_PER_HPAGE unsigned long to avoid build error on powerpc
  KVM: fix ack not being delivered when msi present
  KVM: s390: fix wait_queue handling
  KVM: VMX: Fix locking imbalance on emulation failure
  KVM: VMX: Fix locking order in handle_invalid_guest_state
  KVM: MMU: handle n_free_mmu_pages > n_alloc_mmu_pages in kvm_mmu_change_mmu_pages
  KVM: SVM: force new asid on vcpu migration
  KVM: x86: verify MTRR/PAT validity
  KVM: PIT: fix kpit_elapsed division by zero
  KVM: Fix KVM_GET_MSR_INDEX_LIST

Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/i915: silence vblank warnings
  drm: silence pointless vblank warning.
  drm: When adding probed modes, preserve duplicate mode types

Merge branch 'timers-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'timers-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  posix_cpu_timers_exit_group(): Do not use thread_group_cputimer()

Merge branch 'tracing-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  perf_counter: Fix/complete ftrace event records sampling
  perf_counter, ftrace: Fix perf_counter integration
  tracing/filters: Always free pred on filter_add_subsystem_pred() failure
  tracing/filters: Don't use pred on alloc failure
  ring-buffer: Fix memleak in ring_buffer_free()
  tracing: Fix recordmcount.pl to handle sections with only weak functions
  ring-buffer: Fix advance of reader in rb_buffer_peek()
  tracing: do not use functions starting with .L in recordmcount.pl
  ring-buffer: do not disable ring buffer on oops_in_progress
  ring-buffer: fix check of try_to_discard result

Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86: fix buffer overflow in efi_init()
  x86: Add quirk to make Apple MacBookPro5,1 use reboot=pci
  x86: Fix MSI-X initialization by using online_mask for x2apic target_cpus
  x86: Fix VMI && stack protector

Merge branch 'core-fixes-for-linus-2' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  lockdep: Fix typos in documentation
  lockdep: Fix file mode of lock_stat
  rtmutex: Avoid deadlock in rt_mutex_start_proxy_lock()

perf tools: callchain: Fix bad rounding of minimum rate

Sometimes we get callchain branches that have a rate under the
limit given by the user.

Say you launched:

 perf record -f -g -a ./hackbench 10
 perf report -g fractal,10.0

And you got:

2.33%       hackbench  [kernel]                  [k] _spin_lock_irqsave
                |
                |--78.57%-- remove_wait_queue
                |          poll_freewait
                |          do_sys_poll
                |          sys_poll
                |          sysenter_dispatch
                |          0xf7ffa430
                |          0x1ffadea3c
                |
                |--7.14%-- __up_read
                |          up_read
                |          do_page_fault
                |          page_fault
                |          0xf7ffa430
                |          0xa0df710000000a
                ...

It is abnormal to get a 7.14% branch whereas we passed a 10%
filter.

The problem is that we round down the minimum threshold. This
happens mostly when we have very low number of events. If the
total amount of your branch is 4 and you have a subranch of 3
events, filtering to 90% will be computed like follows:

  limit = 4 * 0.9;

The result is about 3.6, but the cast to integer will round
down to 3. It means that our filter is actually of 75%

We must then explicitly round up the minimum threshold.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: acme@redhat.com
Cc: peterz@infradead.org
Cc: efault@gmx.de
LKML-Reference: <20090809024235.GA10146@nowhere>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter tools: Fix libbfd detection for systems with libz dependency

Due to a libz dependency in some distro's binutils package,
C++ demangle support isn't compiled in despite the necessary
libraries being available.

Fix this by adding a -lz link test to the dependency detection
rules.

Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1249733655.6929.5.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf: "Longum est iter per praecepta, breve et efficax per exempla"

A few examples of how 'perf' can be used, from an e-mail by
Ingo Molnar http://lkml.org/lkml/2009/8/4/346.

Signed-off-by: Carlos R. Mafra <crmafra2@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Valdis.Kletnieks@vt.edu
LKML-Reference: <20090805185334.GA4535@Pilar.aei.mpg.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter: Fix a race on perf_counter_ctx

While extending perfcounters with BTS hw-tracing, Markus
Metzger managed to trigger this warning:

   [  995.557128] WARNING: at kernel/perf_counter.c:1191 __perf_counter_task_sched_out+0x48/0x6b()

triggers because commit
9f498cc5be7e013d8d6e4c616980ed0ffc8680d2 (perf_counter: Full
task tracing) removed clearing of tsk->perf_counter_ctxp out
from under ctx->lock which introduced a race (against
perf_lock_task_context).

Move it back and deal with the exit notification by explicitly
passing along the former task context.

Reported-by: Markus T Metzger <markus.t.metzger@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1249667341.17467.5.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter: Fix tracepoint sampling to be part of generic sampling

Based on Peter's comments, make tracepoint sampling generic
just like all the other sampling bits are. This is a rename
with no code changes:

- PERF_SAMPLE_TP_RECORD to PERF_SAMPLE_RAW
- struct perf_tracepoint_record to perf_raw_record

We want the system in place that transport tracepoints raw
samples events into the perf ring buffer to be generalized and
usable by any type of counter.

Reported-by; Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1249698400-5441-4-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter: Work around gcc warning by initializing tracepoint record unconditionally

Despite that the tracepoint record is always present when the
PERF_SAMPLE_TP_RECORD flag is set, gcc raises a warning,
thinking it might not be initialized:

  kernel/perf_counter.c: In function ‘perf_counter_output’:
  kernel/perf_counter.c:2650: warning: ‘tp’ may be used uninitialized in this function

Then, initialize it to NULL and always check if it's not NULL
before dereference it.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1249698400-5441-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: callchain: Fix sum of percentages to be 100% by displaying amount of ignored chains in fractal mode

When we filter the callchains below a given percentage, we
ignore them and the end result only shows entries that have an
upper percentage than the filter threshold.

It seems to users then that we have an imbalance in the
percentage, as if the sum inside a profiled branch doesn't
reach 100%.

Since in the past there have been real perf report bugs that
showed the same sypmtom, it would be nice to assure the user
that the data is perfect and trustable and it all sums up to
100.00%.

So fix this by displaying the remaining hits that have been
filtered but without more detail than their amount in each
branches. Example while filtering below 50%:

7.73%  [k] delay_tsc
                |
                |--98.22%-- __const_udelay
                |          |
                |          |--86.37%-- ath5k_hw_register_timeout
                |          |          ath5k_hw_noise_floor_calibration
                |          |          ath5k_hw_reset
                |          |          ath5k_reset
                |          |          ath5k_config
                |          |          ieee80211_hw_config
                |          |          |
                |          |          |--88.53%-- ieee80211_scan_work
                |          |          |          worker_thread
                |          |          |          kthread
                |          |          |          child_rip
                |          |           --11.47%-- [...]
                |           --13.63%-- [...]
                 --1.78%-- [...]

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1249690585-9145-4-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: callchain: Fix 'perf report' display to be callchain by default

If we recorded with -g option to record the callchain, right now
we require a -g option to perf report as well - and people reported
this as unnecessary complication: the user already specified -g
once, no need to require it a second time.

So if the recording includes call-chains, display the callchain by
default from perf report.

( The user can override this default using "-g none" option from
  perf report. )

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1249690585-9145-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: callchain: Fix spurious 'perf report' warnings: ignore empty callchains

When the callchain tree comes to insert an empty backtrace, it
raises a spurious warning about the fact we are inserting an
empty. This is spurious because the radix tree assumes it did
something wrong to reach this state. But it didn't, we just met
an empty callchain that has to be ignored.

This happens occasionally with certain types of call-chain
recordings. If it happens it's a big nuisance as perf report
output starts with thousands of warning lines.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1249690585-9145-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf record: Fix the -A UI for empty or non-existent perf.data

1. Ignore the -A argument if there is no perf.data file
2. Treat an empty file like a non existent file.

Else, perf will try to read the perf.data header, and fail with
an error.

Treating an empty file like a non-existent file makes sense,
since an interupted (as in SIGKILLed) perf could leave such
files around, and you don't want to annoy the user with errors
for files with no data in it.

Signed-off-by: Pierre Habouzit <pierre.habouzit@intersec.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf util: Fix do_read() to fail on EOF instead of busy-looping

While toying with perf, I've noticed that perf record can
easily enter a busy loop when doing something as silly as:

    $ perf record -A ls

Yeah, do_read here really wants to read a known size, not being
able to should die(), not busy-loop ;)

That was the cause for the bug.

Signed-off-by: Pierre Habouzit <pierre.habouzit@intersec.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf list: Fix the output to not include tracepoints without an id

Stop perf list from displaying tracepoints without an id file,
those are special tracepoints that are not interfaced to
perfcounters so listing them is erroneous and passing them as
events will produce no output.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Jason Baron <jbaron@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter/powerpc: Fix oops on cpus without perf_counter hardware support

If we have the powerpc perf_counter backend compiled in, but
the cpu we are running on is one where we don't support the
PMU, we currently oops in hw_perf_group_sched_in if we try to
use any counters, because ppmu is NULL in that case, and we
unconditionally dereference ppmu.

This fixes the problem by adding a check if ppmu is NULL at the
beginning of hw_perf_group_sched_in, and also at the beginning
of the other functions that get called from the perf_counter
core, i.e. hw_perf_disable, hw_perf_enable, and
hw_perf_counter_setup.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: benh@kernel.crashing.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf stat: Fix tool option consistency: rename -S/--scale to -c/--scale

We want to use a coherent flag for -S/--stat across all tools,
so free up -S in perf stat.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus@samba.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf report: Add debug help for the finding of symbol bugs - show the symtab origin (DSO, build-id, kernel, etc)

Used with perf report --verbose:

[acme@doppio linux-2.6-tip]$ perf report -v | head -16
     5.17%  firefox  /usr/lib64/xulrunner-1.9.1/libxul.so   0x00000000005d8eee f [.] imgContainer::DrawFrameTo(gfxIImageFrame*, gfxIImageFrame*, nsRect&)
     2.56%  firefox  /lib64/libpthread-2.10.1.so            0x0000000000008e02 d [.] __pthread_mutex_lock_internal
     1.94%  firefox  /usr/lib64/xulrunner-1.9.1/libxul.so   0x0000000000d0af8f f [.] SearchTable
     1.75%  firefox  [kernel]                               0xffffffffff60013b k [.] vread_hpet
     1.63%  firefox  /lib64/libpthread-2.10.1.so            0x000000000000a404 d [.] __pthread_mutex_unlock
     1.47%  firefox  /usr/lib64/xulrunner-1.9.1/libmozjs.so 0x00000000000482ea f [.] js_Interpret
     1.42%  firefox  /usr/lib64/xulrunner-1.9.1/libmozjs.so 0x000000000003eda3 f [.] JS_CallTracer
     1.24%  firefox  [kernel]                               0xffffffff8102ca4a k [k] read_hpet
     1.16%  firefox  [kernel]                               0xffffffff810f3dd4 k [k] fget_light
     1.11%  firefox  /usr/lib64/xulrunner-1.9.1/libmozjs.so 0x00000000000567ff f [.] js_TraceObject
     0.98%  firefox  /usr/lib64/firefox-3.5.2/firefox       0x000000000000dd23 b [.] arena_ralloc
[acme@doppio linux-2.6-tip]$

The new field is just after the symbol address. To help in
figuring out symbol resolution bugs.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf report: Fix per task mult-counter stat reporting

Brice Goglin reported:

> I can easily sort them by thread id, but I don't know how to match
> my 4 events with each group of 4 lines.

Also report the counter id and the time running/enabled
stats (in case the counter got time-shared).

Reported-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Fix multi-counter stat bug caused by incorrect reading of perf.data file header

Brice Goglin reported that only the first result from a
multi-counter perf record --stat run is accurate, the
rest looks bogus.

A silly mistake made us re-read the first attribute for
every recorded attribute.

Reported-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Cc: paulus@samba.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf tools: Fix call-chain cumul hit based sub-total (fractal mode)

The callchain fractal mode builds each new total hits in a new
branch of profiling by using the parent's hits of the current
branch plus the hits of the children.

This is wrong, the total hits of a branch should be made of the
sum of every children hits, we must ignore the parent hits in
this scope.

This patch also fixes another mistake with the hit counting.

Now the rates are correct.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf top: Update man page

perf_counter tools: update perf top manual page to reflect
current implementation.

Signed-off-by: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf top: Improve interactive key handling

Pressing any key which is not currently mapped to
functionality, based on startup command line options, displays
currently mapped keys, and prompts for input.

Pressing any unmapped key at the prompt returns the user to
display mode with variables unchanged.  eg, pressing ? <SPACE>
<ESC> etc displays currently available keys, the value of the
variable associated with that key, and prompts.

Pressing same again aborts input.

Signed-off-by: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter: Fix software counters for fast moving event sources

Reimplement the software counters to deal with fast moving
event sources (such as tracepoints). This means being able
to generate multiple overflows from a single 'event' as well
as support throttling.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter tools: Allow perf top top users to switch between weighted and individual counter display

Add [w]eighted hotkey.  Pressing [w] toggles between displaying
weighted total of all counters, and the counter selected via
[E]vent select key.

------------------------------------------------------------------------------
   PerfTop:   90395 irqs/sec  kernel:16.1% [cache-misses/cache-references/instructions],  (all, 4 CPUs)
------------------------------------------------------------------------------

  weight     samples    pcnt         RIP          kernel function
  ______     _______   _____   ________________   _______________

1275408.6      10881 -  5.3% - ffffffff81146f70 : copy_page_c
 553683.4      43569 - 21.3% - ffffffff81146f20 : clear_page_c
  74075.0       6768 -  3.3% - ffffffff81147190 : copy_user_generic_string
  40602.9       7538 -  3.7% - ffffffff81284ba2 : _spin_lock
  26882.1        965 -  0.5% - ffffffff8109d280 : file_ra_state_init

[w]

------------------------------------------------------------------------------
   PerfTop:   91221 irqs/sec  kernel:14.5% [10000Hz cache-misses],  (all, 4 CPUs)
------------------------------------------------------------------------------

  weight     samples    pcnt         RIP          kernel function
  ______     _______   _____   ________________   _______________

            47320.00 - 22.3% - ffffffff81146f20 : clear_page_c
            14261.00 -  6.7% - ffffffff810992f5 : __rmqueue
            11046.00 -  5.2% - ffffffff81146f70 : copy_page_c
             7842.00 -  3.7% - ffffffff81284ba2 : _spin_lock
             7234.00 -  3.4% - ffffffff810aa1d6 : unmap_vmas

Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

perf_counter tools: Fix/resurrect perf top annotation in a simple interactive form

perf top used to have annotation support, but it has bitrotted and
removed.

This patch restores that: it allows the user to select any symbol
in kernel space for source level annotation on the fly, switch
between event counters and alter display variables. When symbol
details are being displayed, stopping annotation reverts to normal.

known keys:
        [d]     select display delay.
        [e]     select display entries (lines).
        [E]     select annotation event counter.
        [f]     select normal display count filter.
        [F]     select annotation display count filter (percentage).
        [qQ]    quit.
        [s]     select annotation symbol and start annotation.
        [S]     stop annotation, revert to normal display.
        [z]     toggle event count zeroing.

Sample:
------------------------------------------------------------------------------
   PerfTop:   16719 irqs/sec  kernel:78.7% [cache-misses/cache-references/instructions/cycles],  (all, 4 CPUs)
------------------------------------------------------------------------------

Showing cache-misses for e1000_clean_rx_irq
  Events  Pcnt (>=3%)
       0  0.0%                  /* adjust length to remove Ethernet CRC */
       0  0.0%                  if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
       0  0.0%                          length -= 4;
     436  5.0%      f039:       41 f6 84 24 5c 29 00    testb  $0x1,0x295c(%r12)
       0  0.0%      f089:       8b 4d 84                mov    -0x7c(%rbp),%ecx
       0  0.0%      f08c:       48 83 ef 02             sub    $0x2,%rdi
       0  0.0%      f090:       48 83 ee 02             sub    $0x2,%rsi
     811  9.3%      f094:       f3 a4                   rep movsb %ds:(%rsi),%es:(%rdi)
       0  0.0%
       0  0.0%          while (rx_desc->status & E1000_RXD_STAT_DD) {
       0  0.0%      f114:       41 f6 47 0c 01          testb  $0x1,0xc(%r15)
    7226 82.6%      f119:       0f 85 24 fe ff ff       jne    ef43 <e1000_clean_rx_irq+0x84>

Available events:
        0 cache-misses
        1 cache-references
        2 instructions
        3 cycles
Enter details event counter: 2
------------------------------------------------------------------------------
   PerfTop:   15035 irqs/sec  kernel:79.0% [cache-misses/cache-references/instructions/cycles],  (all, 4 CPUs)
------------------------------------------------------------------------------

Showing instructions for e1000_clean_rx_irq
  Events  Pcnt (>=3%)
       0  0.0%                                 int *work_done, int work_to_do)
       0  0.0%  {
     175  0.9%      eebf:       55                      push   %rbp
    1898  9.8%      eec0:       48 89 e5                mov    %rsp,%rbp
       0  0.0%
       0  0.0%          i = rx_ring->next_to_clean;
     140  0.7%      ef0a:       0f b7 41 1a             movzwl 0x1a(%rcx),%eax
     670  3.4%      ef0e:       89 45 ac                mov    %eax,-0x54(%rbp)
       0  0.0%  {
       0  0.0%          memcpy(skb->data + offset, from, len);
      91  0.5%      f07b:       49 8b b6 e8 00 00 00    mov    0xe8(%r14),%rsi
    1153  5.9%      f082:       48 8b b8 e8 00 00 00    mov    0xe8(%rax),%rdi
      42  0.2%      f089:       8b 4d 84                mov    -0x7c(%rbp),%ecx
      14  0.1%      f08c:       48 83 ef 02             sub    $0x2,%rdi
       0  0.0%      f090:       48 83 ee 02             sub    $0x2,%rsi
    1618  8.3%      f094:       f3 a4                   rep movsb %ds:(%rsi),%es:(%rdi)
       0  0.0%
       0  0.0%                  /* return some buffers to hardware, one at a time is too slow */
       0  0.0%                  if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
     867  4.5%      f0e7:       83 7d b0 0f             cmpl   $0xf,-0x50(%rbp)
       0  0.0%
       0  0.0%          while (rx_desc->status & E1000_RXD_STAT_DD) {
      37  0.2%      f114:       41 f6 47 0c 01          testb  $0x1,0xc(%r15)
    4047 20.8%      f119:       0f 85 24 fe ff ff       jne    ef43 <e1000_clean_rx_irq+0x84>

Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>