Pull nmi-safe seq_buf printk update from Steven Rostedt:
"This code is a fork from the trace-3.19 pull as it needed the
trace_seq clean ups from that branch.
This code solves the issue of performing stack dumps from NMI context.
The issue is that printk() is not safe from NMI context as if the NMI
were to trigger when a printk() was being performed, the NMI could
deadlock from the printk() internal locks. This has been seen in
practice.
With lots of review from Petr Mladek, this code went through several
iterations, and we feel that it is now at a point of quality to be
accepted into mainline.
Here's what is contained in this patch set:
- Creates a "seq_buf" generic buffer utility that allows a descriptor
to be passed around where functions can write their own "printk()"
formatted strings into it. The generic version was pulled out of
the trace_seq() code that was made specifically for tracing.
- The seq_buf code was change to model the seq_file code. I have a
patch (not included for 3.19) that converts the seq_file.c code
over to use seq_buf.c like the trace_seq.c code does. This was
done to make sure that seq_buf.c is compatible with seq_file.c. I
may try to get that patch in for 3.20.
- The seq_buf.c file was moved to lib/ to remove it from being
dependent on CONFIG_TRACING.
- The printk() was updated to allow for a per_cpu "override" of the
internal calls. That is, instead of writing to the console, a call
to printk() may do something else. This made it easier to allow
the NMI to change what printk() does in order to call dump_stack()
without needing to update that code as well.
- Finally, the dump_stack from all CPUs via NMI code was converted to
use the seq_buf code. The caller to trigger the NMI code would
wait till all the NMIs finished, and then it would print the
seq_buf data to the console safely from a non NMI context
One added bonus is that this code also makes the NMI dump stack work
on PREEMPT_RT kernels. As printk() includes sleeping locks on
PREEMPT_RT, printk() only writes to console if the console does not
use any rt_mutex converted spin locks. Which a lot do"
* tag 'trace-seq-buf-3.19' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
x86/nmi: Fix use of unallocated cpumask_var_t
printk/percpu: Define printk_func when printk is not defined
x86/nmi: Perform a safe NMI stack trace on all CPUs
printk: Add per_cpu printk func to allow printk to be diverted
seq_buf: Move the seq_buf code to lib/
seq-buf: Make seq_buf_bprintf() conditional on CONFIG_BINARY_PRINTF
tracing: Add seq_buf_get_buf() and seq_buf_commit() helper functions
tracing: Have seq_buf use full buffer
seq_buf: Add seq_buf_can_fit() helper function
tracing: Add paranoid size check in trace_printk_seq()
tracing: Use trace_seq_used() and seq_buf_used() instead of len
tracing: Clean up tracing_fill_pipe_page()
seq_buf: Create seq_buf_used() to find out how much was written
tracing: Add a seq_buf_clear() helper and clear len and readpos in init
tracing: Convert seq_buf fields to be like seq_file fields
tracing: Convert seq_buf_path() to be like seq_path()
tracing: Create seq_buf layer in trace_seq
Currently, function graph tracer prints "!" or "+" just before
function execution time to signal a function overhead, depending
on the time. And some tracers tracing latency also print "!" or
"+" just after time to signal overhead, depending on the interval
between events. Even it is usually enough to do that, we sometimes
need to signal for bigger execution time than 100 micro seconds.
For example, I used function graph tracer to detect if there is
any case that exit_mm() takes too much time. I did following steps
in /sys/kernel/debug/tracing. It was easier to detect very large
excution time with patched kernel than with original kernel.
$ echo exit_mm > set_graph_function
$ echo function_graph > current_tracer
$ echo > trace
$ cat trace_pipe > $LOGFILE
... (do something and terminate logging)
$ grep "\\$" $LOGFILE
3) $ 22082032 us | } /* kernel_map_pages */
3) $ 22082040 us | } /* free_pages_prepare */
3) $ 22082113 us | } /* free_hot_cold_page */
3) $ 22083455 us | } /* free_hot_cold_page_list */
3) $ 22083895 us | } /* release_pages */
3) $ 22177873 us | } /* free_pages_and_swap_cache */
3) $ 22178929 us | } /* unmap_single_vma */
3) $ 22198885 us | } /* unmap_vmas */
3) $ 22206949 us | } /* exit_mmap */
3) $ 22207659 us | } /* mmput */
3) $ 22207793 us | } /* exit_mm */
And then, it was easy to find out that a schedule-out occured by
sub_preempt_count() within kernel_map_pages().
To detect very large function exection time caused by either problematic
function implementation or scheduling issues, this patch can be useful.
Link: http://lkml.kernel.org/r/1416789259-24038-1-git-send-email-byungchul.park@lge.com
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
As the seq_buf->len will soon be +1 size when there's an overflow, we
must use trace_seq_used() or seq_buf_used() methods to get the real
length. This will prevent buffer overflow issues if just the len
of the seq_buf descriptor is used to copy memory.
Link: http://lkml.kernel.org/r/20141114121911.09ba3d38@gandalf.local.home
Reported-by: Petr Mladek <pmladek@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Instead of doing individual checks all over the place that makes the code
very messy. Just check trace_seq_has_overflowed() at the end or in
strategic places.
This makes the code much cleaner and also helps with getting closer
to removing the return values of trace_seq_printf() and friends.
Link: http://lkml.kernel.org/r/20141114011410.987913836@goodmis.org
Reviewed-by: Petr Mladek <pmladek@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
On the function_graph tracer, the print_graph_irq() function prints a
trace line with the flag ==========> on an irq handler entry, and the
flag <========== on an irq handler return.
But when the latency-format is enable, it is not printing the
latency-format flags, causing the following error in the trace output:
0) ==========> |
0) d... | smp_apic_timer_interrupt() {
This patch fixes this issue by printing the latency-format flags when
it is enable.
Link: http://lkml.kernel.org/r/7c2e226dac20c940b6242178fab7f0e3c9b5ce58.1415233316.git.bristot@redhat.com
Reviewed-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Using seq_printf to print a simple string or a single character is a
lot more expensive than it needs to be, since seq_puts and seq_putc
exist.
These patches do
seq_printf(m, s) -> seq_puts(m, s)
seq_printf(m, "%s", s) -> seq_puts(m, s)
seq_printf(m, "%c", c) -> seq_putc(m, c)
Subsequent patches will simplify further.
Link: http://lkml.kernel.org/r/1415479332-25944-2-git-send-email-linux@rasmusvillemoes.dk
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently, tracing_thresh works only if we specify it before selecting
function_graph tracer. If we do the opposite, tracing_thresh will change
it's value, but it will not be applied.
To fix it, we add update_thresh callback which is called whenever
tracing_thresh is updated and for function_graph tracer we register
handler which reinitializes tracer depending on tracing_thresh.
Link: http://lkml.kernel.org/p/20140718111727.GA3206@stfomichev-desktop.yandex.net
Signed-off-by: Stanislav Fomichev <stfomichev@yandex-team.ru>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
All archs now use ftrace_graph_is_dead() to stop function graph
tracing. Remove the usage of ftrace_stop() as that is no longer
needed.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ftrace_stop() is going away as it disables parts of function tracing
that affects users that should not be affected. But ftrace_graph_stop()
is built on ftrace_stop(). Here's another example of killing all of
function tracing because something went wrong with function graph
tracing.
Instead of disabling all users of function tracing on function graph
error, disable only function graph tracing.
A new function is created called ftrace_graph_is_dead(). This is called
in strategic paths to prevent function graph from doing more harm and
allowing at least a warning to be printed before the system crashes.
NOTE: ftrace_stop() is still used until all the archs are converted over
to use ftrace_graph_is_dead(). After that, ftrace_stop() will be removed.
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
In the function-graph tracer, add a funcgraph_tail option
to print the function name on all } lines, not just
functions whose first line is no longer in the trace
buffer.
If a function calls other traced functions, its total
time appears on its } line. This change allows grep
to be used to determine the function for which the
line corresponds.
Update Documentation/trace/ftrace.txt to describe
this new option.
Link: http://lkml.kernel.org/p/20140520221041.8359.6782.stgit@beardog.cce.hp.com
Signed-off-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Now that the ring buffer has a built in way to wake up readers
when there's data, using irq_work such that it is safe to do it
in any context. But it was still using the old "poor man's"
wait polling that checks every 1/10 of a second to see if it
should wake up a waiter. This makes the latency for a wake up
excruciatingly long. No need to do that anymore.
Completely remove the different wait_poll types from the tracers
and have them all use the default one now.
Reported-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
As options (flags) may affect instances instead of being global
the set_flag() callbacks need to receive the trace_array descriptor
of the instance they will be modifying.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The duration field of print_graph_duration() can also be used
to do the space filling by passing an enum in it:
DURATION_FILL_FULL
DURATION_FILL_START
DURATION_FILL_END
The problem is that these are enums and defined as negative,
but the duration field is unsigned long long. Most archs are
fine with this but blackfin fails to compile because of it:
kernel/built-in.o: In function `print_graph_duration':
kernel/trace/trace_functions_graph.c:782: undefined reference to `__ucmpdi2'
Overloading a unsigned long long with an signed enum is just
bad in principle. We can accomplish the same thing by using
part of the flags field instead.
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The trace event filters are still tied to event calls rather than
event files, which means you don't get what you'd expect when using
filters in the multibuffer case:
Before:
# echo 'bytes_alloc > 8192' > /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
# cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
bytes_alloc > 8192
# mkdir /sys/kernel/debug/tracing/instances/test1
# echo 'bytes_alloc > 2048' > /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter
# cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
bytes_alloc > 2048
# cat /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter
bytes_alloc > 2048
Setting the filter in tracing/instances/test1/events shouldn't affect
the same event in tracing/events as it does above.
After:
# echo 'bytes_alloc > 8192' > /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
# cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
bytes_alloc > 8192
# mkdir /sys/kernel/debug/tracing/instances/test1
# echo 'bytes_alloc > 2048' > /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter
# cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter
bytes_alloc > 8192
# cat /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter
bytes_alloc > 2048
We'd like to just move the filter directly from ftrace_event_call to
ftrace_event_file, but there are a couple cases that don't yet have
multibuffer support and therefore have to continue using the current
event_call-based filters. For those cases, a new USE_CALL_FILTER bit
is added to the event_call flags, whose main purpose is to keep the
old behavior for those cases until they can be updated with
multibuffer support; at that point, the USE_CALL_FILTER flag (and the
new associated call_filter_check_discard() function) can go away.
The multibuffer support also made filter_current_check_discard()
redundant, so this change removes that function as well and replaces
it with filter_check_discard() (or call_filter_check_discard() as
appropriate).
Link: http://lkml.kernel.org/r/f16e9ce4270c62f46b2e966119225e1c3cca7e60.1382620672.git.tom.zanussi@linux.intel.com
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The selftest for function and function graph tracers are defined as
__init, as they are only executed at boot up. The "tracer" structs
that are associated to those tracers are not setup as __init as they
are used after boot. To stop mismatch warnings, those structures
need to be annotated with __ref_data.
Currently, the tracer structures are defined to __read_mostly, as they
do not really change. But in the future they should be converted to
consts, but that will take a little work because they have a "next"
pointer that gets updated when they are registered. That will have to
wait till the next major release.
Link: http://lkml.kernel.org/r/1373596735.17876.84.camel@gandalf.local.home
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Reported-by: Chen Gang <gang.chen@asianux.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently, the way the latency tracers and snapshot feature works
is to have a separate trace_array called "max_tr" that holds the
snapshot buffer. For latency tracers, this snapshot buffer is used
to swap the running buffer with this buffer to save the current max
latency.
The only items needed for the max_tr is really just a copy of the buffer
itself, the per_cpu data pointers, the time_start timestamp that states
when the max latency was triggered, and the cpu that the max latency
was triggered on. All other fields in trace_array are unused by the
max_tr, making the max_tr mostly bloat.
This change removes the max_tr completely, and adds a new structure
called trace_buffer, that holds the buffer pointer, the per_cpu data
pointers, the time_start timestamp, and the cpu where the latency occurred.
The trace_array, now has two trace_buffers, one for the normal trace and
one for the max trace or snapshot. By doing this, not only do we remove
the bloat from the max_trace but the instances of traces can now use
their own snapshot feature and not have just the top level global_trace have
the snapshot feature and latency tracers for itself.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The global and max-tr currently use static per_cpu arrays for the CPU data
descriptors. But in order to get new allocated trace_arrays, they need to
be allocated per_cpu arrays. Instead of using the static arrays, switch
the global and max-tr to use allocated data.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
While debugging the virtual cputime with the function graph tracer
with a max_depth of 1 (most common use of the max_depth so far),
I found that I was missing kernel execution because of a race condition.
The code for the return side of the function has a slight race:
ftrace_pop_return_trace(&trace, &ret, frame_pointer);
trace.rettime = trace_clock_local();
ftrace_graph_return(&trace);
barrier();
current->curr_ret_stack--;
The ftrace_pop_return_trace() initializes the trace structure for
the callback. The ftrace_graph_return() uses the trace structure
for its own use as that structure is on the stack and is local
to this function. Then the curr_ret_stack is decremented which
is what the trace.depth is set to.
If an interrupt comes in after the ftrace_graph_return() but
before the curr_ret_stack, then the called function will get
a depth of 2. If max_depth is set to 1 this function will be
ignored.
The problem is that the trace has already been called, and the
timestamp for that trace will not reflect the time the function
was about to re-enter userspace. Calls to the interrupt will not
be traced because the max_depth has prevented this.
To solve this issue, the ftrace_graph_return() can safely be
moved after the current->curr_ret_stack has been updated.
This way the timestamp for the return callback will reflect
the actual time.
If an interrupt comes in after the curr_ret_stack update and
ftrace_graph_return(), it will be traced. It may look a little
confusing to see it within the other function, but at least
it will not be lost.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
