commit 726b3d3f141fba6f841d715fc4d8a4a84f02c02a upstream.

When an interrupt or NMI comes in and switches the context, there's a delay
from when the preempt_count() shows the update. As the preempt_count() is
used to detect recursion having each context have its own bit get set when
tracing starts, and if that bit is already set, it is considered a recursion
and the function exits. But if this happens in that section where context
has changed but preempt_count() has not been updated, this will be
incorrectly flagged as a recursion.

To handle this case, create another bit call TRANSITION and test it if the
current context bit is already set. Flag the call as a recursion if the
TRANSITION bit is already set, and if not, set it and continue. The
TRANSITION bit will be cleared normally on the return of the function that
set it, or if the current context bit is clear, set it and clear the
TRANSITION bit to allow for another transition between the current context
and an even higher one.

Cc: stable@vger.kernel.org
Fixes: edc15cafcbfa3 ("tracing: Avoid unnecessary multiple recursion checks")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

commit ee11b93f95eabdf8198edd4668bf9102e7248270 upstream.

The code that checks recursion will work to only do the recursion check once
if there's nested checks. The top one will do the check, the other nested
checks will see recursion was already checked and return zero for its "bit".
On the return side, nothing will be done if the "bit" is zero.

The problem is that zero is returned for the "good" bit when in NMI context.
This will set the bit for NMIs making it look like *all* NMI tracing is
recursing, and prevent tracing of anything in NMI context!

The simple fix is to return "bit + 1" and subtract that bit on the end to
get the real bit.

Cc: stable@vger.kernel.org
Fixes: edc15cafcbfa3 ("tracing: Avoid unnecessary multiple recursion checks")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

I was attempting to use pid filtering with function_graph, but it wasn't
allowing anything to make it through.  Turns out ftrace_trace_task
returns false if ftrace_ignore_pid is not-empty, which isn't correct
anymore.  We're now setting it to FTRACE_PID_IGNORE if we need to ignore
that pid, otherwise it's set to the pid (which is weird considering the
name) or to FTRACE_PID_TRACE.  Fix the check to check for !=
FTRACE_PID_IGNORE.  With this we can now use function_graph with pid
filtering.

Link: https://lkml.kernel.org/r/20200725005048.1790-1-josef@toxicpanda.com

Fixes: 717e3f5ebc82 ("ftrace: Make function trace pid filtering a bit more exact")
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

If a process has the trace_pipe open on a trace_array, the current tracer
for that trace array should not be changed. This was original enforced by a
global lock, but when instances were introduced, it was moved to the
current_trace. But this structure is shared by all instances, and a
trace_pipe is for a single instance. There's no reason that a process that
has trace_pipe open on one instance should prevent another instance from
changing its current tracer. Move the reference counter to the trace_array
instead.

This is marked as "Fixes" but is more of a clean up than a true fix.
Backport if you want, but its not critical.

Fixes: cf6ab6d9143b1 ("tracing: Add ref count to tracer for when they are being read by pipe")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

When using trace-cmd on 5.6-rt for the function graph tracer, the output was
corrupted. It gave output like this:

 funcgraph_entry:       func=0xffffffff depth=38982
 funcgraph_entry:       func=0x1ffffffff depth=16044
 funcgraph_exit:        func=0xffffffff overrun=0x92539aaf00000000 calltime=0x92539c9900000072 rettime=0x100000072 depth=11084
 funcgraph_exit:        func=0xffffffff overrun=0x9253946e00000000 calltime=0x92539e2100000072 rettime=0x72 depth=26033702
 funcgraph_entry:       func=0xffffffff depth=85798
 funcgraph_entry:       func=0x1ffffffff depth=12044

The reason was because the tracefs/events/ftrace/funcgraph_entry/exit format
file was incorrect. The -rt kernel adds more common fields to the trace
events. Namely, common_migrate_disable and common_preempt_lazy_count. Each
is one byte in size. This changes the alignment of the normal payload. Most
events are aligned normally, but the function and function graph events are
defined with a "PACKED" macro, that packs their payload. As the offsets
displayed in the format files are now calculated by an aligned field, the
aligned field for function and function graph events should be 1, not their
normal alignment.

With aligning of the funcgraph_entry event, the format file has:

        field:unsigned short common_type;       offset:0;       size:2; signed:0;
        field:unsigned char common_flags;       offset:2;       size:1; signed:0;
        field:unsigned char common_preempt_count;       offset:3;       size:1; signed:0;
        field:int common_pid;   offset:4;       size:4; signed:1;
        field:unsigned char common_migrate_disable;     offset:8;       size:1; signed:0;
        field:unsigned char common_preempt_lazy_count;  offset:9;       size:1; signed:0;

        field:unsigned long func;       offset:16;      size:8; signed:0;
        field:int depth;        offset:24;      size:4; signed:1;

But the actual alignment is:

	field:unsigned short common_type;	offset:0;	size:2;	signed:0;
	field:unsigned char common_flags;	offset:2;	size:1;	signed:0;
	field:unsigned char common_preempt_count;	offset:3;	size:1;	signed:0;
	field:int common_pid;	offset:4;	size:4;	signed:1;
	field:unsigned char common_migrate_disable;	offset:8;	size:1;	signed:0;
	field:unsigned char common_preempt_lazy_count;	offset:9;	size:1;	signed:0;

	field:unsigned long func;	offset:12;	size:8;	signed:0;
	field:int depth;	offset:20;	size:4;	signed:1;

Link: https://lkml.kernel.org/r/20200609220041.2a3b527f@oasis.local.home

Cc: stable@vger.kernel.org
Fixes: 04ae87a52074e ("ftrace: Rework event_create_dir()")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

Add a new "hist_debug" file for each trace event, which when read will
dump out a bunch of internal details about the hist triggers defined
on that event.

This is normally off but can be enabled by saying 'y' to the new
CONFIG_HIST_TRIGGERS_DEBUG config option.

This is in support of the new Documentation file describing histogram
internals, Documentation/trace/histogram-design.rst, which was
requested by developers trying to understand the internals when
extending or making use of the hist triggers for higher-level tools.

The histogram-design.rst documentation refers to the hist_debug files
and demonstrates their use with output in the test examples.

Link: http://lkml.kernel.org/r/77914c22b0ba493d9783c53bbfbc6087d6a7e1b1.1585941485.git.zanussi@kernel.org

Signed-off-by: Tom Zanussi <zanussi@kernel.org>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

There's currently a way to select a task that should only have its events
traced, but there's no way to select a task not to have itsevents traced.
Add a set_event_notrace_pid file that acts the same as set_event_pid (and is
also affected by event-fork), but the task pids in this file will not be
traced even if they are listed in the set_event_pid file. This makes it easy
for tools like trace-cmd to "hide" itself from beint traced by events when
it is recording other tasks.

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

There's currently a way to select a task that should only be traced by
functions, but there's no way to select a task not to be traced by the
function tracer. Add a set_ftrace_notrace_pid file that acts the same as
set_ftrace_pid (and is also affected by function-fork), but the task pids in
this file will not be traced even if they are listed in the set_ftrace_pid
file. This makes it easy for tools like trace-cmd to "hide" itself from the
function tracer when it is recording other tasks.

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

The set_ftrace_pid file is used to filter function tracing to only trace
tasks that are listed in that file. Instead of testing the pids listed in
that file (it's a bitmask) at each function trace event, the logic is done
via a sched_switch hook. A flag is set when the next task to run is in the
list of pids in the set_ftrace_pid file. But the sched_switch hook is not at
the exact location of when the task switches, and the flag gets set before
the task to be traced actually runs. This leaves a residue of traced
functions that do not belong to the pid that should be filtered on.

By changing the logic slightly, where instead of having  a boolean flag to
test, record the pid that should be traced, with special values for not to
trace and always trace. Then at each function call, a check will be made to
see if the function should be ignored, or if the current pid matches the
function that should be traced, and only trace if it matches (or if it has
the special value to always trace).

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

When opening the "trace" file, it is no longer necessary to disable tracing.

Note, a new option is created called "pause-on-trace", when set, will cause
the trace file to emulate its original behavior.

Link: http://lkml.kernel.org/r/20200317213416.903351225@goodmis.org

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