We hit ftrace_bug report when booting Android on a 64bit ATOM SOC chip.
Basically, there is a race between insmod and ftrace_run_update_code.
After load_module=>ftrace_module_init, another thread jumps in to call
ftrace_run_update_code=>ftrace_arch_code_modify_prepare
=>set_all_modules_text_rw, to change all modules
as RW. Since the new module is at MODULE_STATE_UNFORMED, the text attribute
is not changed. Then, the 2nd thread goes ahead to change codes.
However, load_module continues to call complete_formation=>set_section_ro_nx,
then 2nd thread would fail when probing the module's TEXT.
The patch fixes it by using notifier to delay the enabling of ftrace
records to the time when module is at state MODULE_STATE_COMING.
Link: http://lkml.kernel.org/r/567CE628.3000609@intel.com
Signed-off-by: Qiu Peiyang <peiyangx.qiu@intel.com>
Signed-off-by: Zhang Yanmin <yanmin.zhang@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Qiu Peiyang pointed out that there's a race when enabling function tracing
and loading a module. In order to make the modifications of converting nops
in the prologue of functions into callbacks, the text needs to be converted
from read-only to read-write. When enabling function tracing, the text
permission is updated, the functions are modified, and then they are put
back.
When loading a module, the updates to convert function calls to mcount is
done before the module text is set to read-only. But after it is done, the
module text is visible by the function tracer. Thus we have the following
race:
CPU 0 CPU 1
----- -----
start function tracing
set text to read-write
load_module
add functions to ftrace
set module text read-only
update all functions to callbacks
modify module functions too
< Can't it's read-only >
When this happens, ftrace detects the issue and disables itself till the
next reboot.
To fix this, a new DISABLED flag is added for ftrace records, which all
module functions get when they are added. Then later, after the module code
is all set, the records will have the DISABLED flag cleared, and they will
be enabled if any callback wants all functions to be traced.
Note, this doesn't add the delay to later. It simply changes the
ftrace_module_init() to do both the setting of DISABLED records, and then
immediately calls the enable code. This helps with testing this new code as
it has the same behavior as previously. Another change will come after this
to have the ftrace_module_enable() called after the text is set to
read-only.
Cc: Qiu Peiyang <peiyangx.qiu@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The start and end variables were only used when ftrace_module_init() was
split up into multiple functions. No need to keep them around after the
merger.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Jiri Olsa noted that the change to replace the control_ops did not update
the trampoline for when running perf on a single CPU and with CONFIG_PREEMPT
disabled (where dynamic ops, like perf, can use trampolines directly). The
result was that perf function could be called when RCU is not watching as
well as not handle the ftrace_local_disable().
Modify the ftrace_ops_get_func() to also check the RCU and PER_CPU ops flags
and use the recursive function if they are set. The recursive function is
modified to check those flags and execute the appropriate checks if they are
set.
Link: http://lkml.kernel.org/r/20151201134213.GA14155@krava.brq.redhat.com
Reported-by: Jiri Olsa <jolsa@redhat.com>
Patch-fixed-up-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently perf has its own list function within the ftrace infrastructure
that seems to be used only to allow for it to have per-cpu disabling as well
as a check to make sure that it's not called while RCU is not watching. It
uses something called the "control_ops" which is used to iterate over ops
under it with the control_list_func().
The problem is that this control_ops and control_list_func unnecessarily
complicates the code. By replacing FTRACE_OPS_FL_CONTROL with two new flags
(FTRACE_OPS_FL_RCU and FTRACE_OPS_FL_PER_CPU) we can remove all the code
that is special with the control ops and add the needed checks within the
generic ftrace_list_func().
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When showing all tramps registered to a ftrace record in the file
enabled_functions, it exits the loop with ops == NULL. But then it is
suppose to show the function on the ops->trampoline and
add_trampoline_func() is called with the given ops. But because ops is now
NULL (to exit the loop), it always shows the static trampoline instead of
the one that is really registered to the record.
The call to add_trampoline_func() that shows the trampoline for the given
ops needs to be called at every iteration.
Fixes: 39daa7b9e8 "ftrace: Show all tramps registered to a record on ftrace_bug()"
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When an anomaly is detected in the function call modification code,
ftrace_bug() is called to disable function tracing as well as give any
information that may help debug the problem. Currently, only the first found
trampoline that is attached to the failed record is reported. Instead, show
all trampolines that are hooked to it.
Also, not only show the ops pointer but also report the function it calls.
While at it, add this info to the enabled_functions debug file too.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When an anomaly is found while modifying function code, ftrace_bug() is
called which disables the function tracing infrastructure and reports
information about what failed. If the code that is to be replaced does not
match what is expected, then actual code is shown. Currently there is no
arch generic way to show what was expected.
Add a new variable pointer calld ftrace_expected that the arch code can set
to point to what it expected so that ftrace_bug() can report the actual text
as well as the text that was expected to be there.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The ftrace function hook utility has several internal checks to make sure
that whatever it modifies is exactly what it expects to be modifying. This
is essential as modifying running code can be extremely dangerous to the
system.
When an anomaly is detected, ftrace_bug() is called which sends a splat to
the console and disables function tracing. There's some extra information
that is printed to help diagnose the issue.
One thing that is missing though is output of what ftrace was doing at the
time of the crash. Was it updating a call site or perhaps converting a call
site to a nop? A new global enum variable is created to state what ftrace
was doing at the time of the anomaly, and this is reported in ftrace_bug().
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
stable tags to them. I searched through my INBOX just as the merge window
opened and found lots of patches to pull. I ran them through all my tests
and they were in linux-next for a few days.
Features added this release:
----------------------------
o Module globbing. You can now filter function tracing to several
modules. # echo '*:mod:*snd*' > set_ftrace_filter (Dmitry Safonov)
o Tracer specific options are now visible even when the tracer is not
active. It was rather annoying that you can only see and modify tracer
options after enabling the tracer. Now they are in the options/ directory
even when the tracer is not active. Although they are still only visible
when the tracer is active in the trace_options file.
o Trace options are now per instance (although some of the tracer specific
options are global)
o New tracefs file: set_event_pid. If any pid is added to this file, then
all events in the instance will filter out events that are not part of
this pid. sched_switch and sched_wakeup events handle next and the wakee
pids.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJWPLQ5AAoJEKKk/i67LK/8CTYIAI1u8DE5QCzv3J0p54jVpNVR
J5FqEU3eXIzd6FS4JXD4nxCeMpUZAy21YnhlZpsnrbJJM5bc9bUsBCwiKKM+MuSZ
ztmy2sgYKkO0h/KUdhNgYJrzis3/Ojquyx9iAqK5ST/Fr+nKYx81akFKjNK53iur
RJRut45sSa8rv11LaL8sgJ6hAWQTc+YkybUdZ5xaMdJmZ6A61T7Y6VzTjbUexuvL
hntCfTjYLtVd8dbfknAnf3B7n/VOO3IFF85wr7ciYR5oEVfPrF8tHmJBlhHExPpX
kaXAiDDRY/UTg/5DQqnp4zmxJoR5BQ2l4pT5PwiLcnwhcphIDNYS8EYUmOYAWjU=
=TjOE
-----END PGP SIGNATURE-----
Merge tag 'trace-v4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracking updates from Steven Rostedt:
"Most of the changes are clean ups and small fixes. Some of them have
stable tags to them. I searched through my INBOX just as the merge
window opened and found lots of patches to pull. I ran them through
all my tests and they were in linux-next for a few days.
Features added this release:
----------------------------
- Module globbing. You can now filter function tracing to several
modules. # echo '*:mod:*snd*' > set_ftrace_filter (Dmitry Safonov)
- Tracer specific options are now visible even when the tracer is not
active. It was rather annoying that you can only see and modify
tracer options after enabling the tracer. Now they are in the
options/ directory even when the tracer is not active. Although
they are still only visible when the tracer is active in the
trace_options file.
- Trace options are now per instance (although some of the tracer
specific options are global)
- New tracefs file: set_event_pid. If any pid is added to this file,
then all events in the instance will filter out events that are not
part of this pid. sched_switch and sched_wakeup events handle next
and the wakee pids"
* tag 'trace-v4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace: (68 commits)
tracefs: Fix refcount imbalance in start_creating()
tracing: Put back comma for empty fields in boot string parsing
tracing: Apply tracer specific options from kernel command line.
tracing: Add some documentation about set_event_pid
ring_buffer: Remove unneeded smp_wmb() before wakeup of reader benchmark
tracing: Allow dumping traces without tracking trace started cpus
ring_buffer: Fix more races when terminating the producer in the benchmark
ring_buffer: Do no not complete benchmark reader too early
tracing: Remove redundant TP_ARGS redefining
tracing: Rename max_stack_lock to stack_trace_max_lock
tracing: Allow arch-specific stack tracer
recordmcount: arm64: Replace the ignored mcount call into nop
recordmcount: Fix endianness handling bug for nop_mcount
tracepoints: Fix documentation of RCU lockdep checks
tracing: ftrace_event_is_function() can return boolean
tracing: is_legal_op() can return boolean
ring-buffer: rb_event_is_commit() can return boolean
ring-buffer: rb_per_cpu_empty() can return boolean
ring_buffer: ring_buffer_empty{cpu}() can return boolean
ring-buffer: rb_is_reader_page() can return boolean
...
Extend module command for function filter selection with globbing.
It uses the same globbing as function filter.
sh# echo '*alloc*:mod:*' > set_ftrace_filter
Will trace any function with the letters 'alloc' in the name in any
module but not in kernel.
sh# echo '!*alloc*:mod:ipv6' >> set_ftrace_filter
Will prevent from tracing functions with 'alloc' in the name from module
ipv6 (do not forget to append to set_ftrace_filter file).
sh# echo '*alloc*:mod:!ipv6' > set_ftrace_filter
Will trace functions with 'alloc' in the name from kernel and any
module except ipv6.
sh# echo '*alloc*:mod:!*' > set_ftrace_filter
Will trace any function with the letters 'alloc' in the name only from
kernel, but not from any module.
sh# echo '*:mod:!*' > set_ftrace_filter
or
sh# echo ':mod:!' > set_ftrace_filter
Will trace every function in the kernel, but will not trace functions
from any module.
sh# echo '*:mod:*' > set_ftrace_filter
or
sh# echo ':mod:' > set_ftrace_filter
As the opposite will trace all functions from all modules, but not from
kernel.
sh# echo '*:mod:*snd*' > set_ftrace_filter
Will trace your sound drivers only (if any).
Link: http://lkml.kernel.org/r/1443545176-3215-4-git-send-email-0x7f454c46@gmail.com
Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
[ Made format changes ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ftrace_match parameters are very related and I reduce the number of local
variables & parameters with it.
This is also preparation for module globbing as it would introduce more
realated variables & parameters.
Link: http://lkml.kernel.org/r/1443545176-3215-3-git-send-email-0x7f454c46@gmail.com
Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
[ Made some formatting changes ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
"Not" is too abstract variable name - changed to clear_filter.
Removed ftrace_match_module_records function: comparison with !* or *
not does the general code in filter_parse_regex() as it works without
mod command for
sh# echo '!*' > /sys/kernel/debug/tracing/set_ftrace_filter
Link: http://lkml.kernel.org/r/1443545176-3215-2-git-send-email-0x7f454c46@gmail.com
Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
By now there isn't any subcommand for mod.
Before:
sh$ echo '*:mod:ipv6:a' > set_ftrace_filter
sh$ echo '*:mod:ipv6' > set_ftrace_filter
had the same results, but now first will result in:
sh$ echo '*:mod:ipv6:a' > set_ftrace_filter
-bash: echo: write error: Invalid argument
Also, I clarified ftrace_mod_callback code a little.
Link: http://lkml.kernel.org/r/1443545176-3215-1-git-send-email-0x7f454c46@gmail.com
Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
[ converted 'if (ret == 0)' to 'if (!ret)' ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
__trace_sched_switch_state() is the last remaining PREEMPT_ACTIVE
user, move trace_sched_switch() from prepare_task_switch() to
__schedule() and propagate the @preempt argument.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
To cover the common case of sorting an array of pointers, Daniel
Wagner recently modified the library sort() to use a specific swap
function for size==8, in addition to the size==4 case which was
already handled. Since sizeof(long) is either 4 or 8,
ftrace_swap_ips() is redundant and we can just let sort() pick an
appropriate and fast swap callback.
Link: http://lkml.kernel.org/r/1441834023-13130-1-git-send-email-linux@rasmusvillemoes.dk
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The sleep-time and graph-time options are only for the function graph tracer
and are not used by anything else. As tracer options are now visible when
the tracer is not activated, its better to move the function graph specific
tracer options into the function graph tracer.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The changes with more meat are:
o Allowing the trace event filters to filter on CPU number and process ids
o Two new markers for trace output latency were added
(10 and 100 msec latencies)
o Have tracing_thresh filter function profiling time
I also worked on modifying the ring buffer code for some future
work, and moved the adding of the timestamp around. One of my changes
caused a regression, and since other changes were built on top of it
and already tested, I had to operate a revert of that change. Instead
of rebasing, this change set has the code that caused a regression
as well as the code to revert that change without touching the other
changes that were made on top of it.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJV6aZEAAoJEEjnJuOKh9ldrR4H/A1RcQf1prLLoUibPP4w3lat
dmQcdpS1NY+cqyiKuKPAOkFDGQL7qWzRqZ8whcPSJIsHq57ufqNSLf+0bbQYPzg9
g3CgGL7OApmGi5ulj0sNxhadvc9TFm/SAN0nVJlNuUWdm8e1UWHLsrJZaMfopu2r
RDEtkOhg619mhDL4rktNdS6rk0B92Fhu2o2PwLZPVlUl1NNEt4WJU+ejitXUVO1A
Nb70/rTGGJKtyHbW+74on4LnEN5Uu0Viu6rMwGfYyIgRmC2otdBDvE4xfKMiTUKr
SzBjzrhIoMIRn4Vl0vElfulkpYaw7pcC2BdpZ4d9VpIOiLSlZs0x/TgCtpFEv5M=
=baZ3
-----END PGP SIGNATURE-----
Merge tag 'trace-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracing update from Steven Rostedt:
"Mostly this is just clean ups and micro optimizations.
The changes with more meat are:
- Allowing the trace event filters to filter on CPU number and
process ids
- Two new markers for trace output latency were added (10 and 100
msec latencies)
- Have tracing_thresh filter function profiling time
I also worked on modifying the ring buffer code for some future work,
and moved the adding of the timestamp around. One of my changes
caused a regression, and since other changes were built on top of it
and already tested, I had to operate a revert of that change. Instead
of rebasing, this change set has the code that caused a regression as
well as the code to revert that change without touching the other
changes that were made on top of it"
* tag 'trace-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
ring-buffer: Revert "ring-buffer: Get timestamp after event is allocated"
tracing: Don't make assumptions about length of string on task rename
tracing: Allow triggers to filter for CPU ids and process names
ftrace: Format MCOUNT_ADDR address as type unsigned long
tracing: Introduce two additional marks for delay
ftrace: Fix function_graph duration spacing with 7-digits
ftrace: add tracing_thresh to function profile
tracing: Clean up stack tracing and fix fentry updates
ring-buffer: Reorganize function locations
ring-buffer: Make sure event has enough room for extend and padding
ring-buffer: Get timestamp after event is allocated
ring-buffer: Move the adding of the extended timestamp out of line
ring-buffer: Add event descriptor to simplify passing data
ftrace: correct the counter increment for trace_buffer data
tracing: Fix for non-continuous cpu ids
tracing: Prefer kcalloc over kzalloc with multiply
Commit 4104d326b6 ("ftrace: Remove global function list and call function
directly") simplified the ftrace code by removing the global_ops list with a
new design. But this cleanup also broke the filtering of PIDs that are added
to the set_ftrace_pid file.
Add back the proper hooks to have pid filtering working once again.
Cc: stable@vger.kernel.org # 3.16+
Reported-by: Matt Fleming <matt@console-pimps.org>
Reported-by: Richard Weinberger <richard.weinberger@gmail.com>
Tested-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
This patch extends tracing_thresh functionality to function profile tracer.
If tracing_thresh is set, print those entries only,
whose average is > tracing thresh.
Link: http://lkml.kernel.org/r/1434972488-8571-1-git-send-email-umesh.t@samsung.com
Signed-off-by: Umesh Tiwari <umesh.t@samsung.com>
[ Removed unnecessary 'moved' comment ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
of the TRACE_DEFINE_ENUM() macro that can be used by tracepoints.
Tracepoints have helper functions for the TP_printk() called
__print_symbolic() and __print_flags() that lets a numeric number be
displayed as a a human comprehensible text. What is placed in the
TP_printk() is also shown in the tracepoint format file such that
user space tools like perf and trace-cmd can parse the binary data
and express the values too. Unfortunately, the way the TRACE_EVENT()
macro works, anything placed in the TP_printk() will be shown pretty
much exactly as is. The problem arises when enums are used. That's
because unlike macros, enums will not be changed into their values
by the C pre-processor. Thus, the enum string is exported to the
format file, and this makes it useless for user space tools.
The TRACE_DEFINE_ENUM() solves this by converting the enum strings
in the TP_printk() format into their number, and that is what is
shown to user space. For example, the tracepoint tlb_flush currently
has this in its format file:
__print_symbolic(REC->reason,
{ TLB_FLUSH_ON_TASK_SWITCH, "flush on task switch" },
{ TLB_REMOTE_SHOOTDOWN, "remote shootdown" },
{ TLB_LOCAL_SHOOTDOWN, "local shootdown" },
{ TLB_LOCAL_MM_SHOOTDOWN, "local mm shootdown" })
After adding:
TRACE_DEFINE_ENUM(TLB_FLUSH_ON_TASK_SWITCH);
TRACE_DEFINE_ENUM(TLB_REMOTE_SHOOTDOWN);
TRACE_DEFINE_ENUM(TLB_LOCAL_SHOOTDOWN);
TRACE_DEFINE_ENUM(TLB_LOCAL_MM_SHOOTDOWN);
Its format file will contain this:
__print_symbolic(REC->reason,
{ 0, "flush on task switch" },
{ 1, "remote shootdown" },
{ 2, "local shootdown" },
{ 3, "local mm shootdown" })
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJVLBTuAAoJEEjnJuOKh9ldjHMIALdRS755TXCZGOf0r7O2akOR
wMPeum7C+ae1mH+jCsJKUC0/jUfQKaMt/UxoHlipDgcGg8kD2jtGnGCw4Xlwvdsr
y4rFmcTRSl1mo0zDSsg6ujoupHlVYN0+JPjrd7S3cv/llJoY49zcanNLF7S2XLeM
dZCtWRLWYpBiWO68ai6AqJTnE/eGFIqBI048qb5Eg8dbK243SSeSIf9Ywhb+VsA+
aq6F7cWI/H6j4tbeza8tAN19dcwenDro5EfCDY8ARQHJu1f6Y3+DLf2imjkd6Aiu
JVAoGIjHIpI+djwCZC1u4gi4urjfOqYartrM3Q54tb3YWYqHeNqP2ASI2a4EpYk=
=Ixwt
-----END PGP SIGNATURE-----
Merge tag 'trace-v4.1' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracing updates from Steven Rostedt:
"Some clean ups and small fixes, but the biggest change is the addition
of the TRACE_DEFINE_ENUM() macro that can be used by tracepoints.
Tracepoints have helper functions for the TP_printk() called
__print_symbolic() and __print_flags() that lets a numeric number be
displayed as a a human comprehensible text. What is placed in the
TP_printk() is also shown in the tracepoint format file such that user
space tools like perf and trace-cmd can parse the binary data and
express the values too. Unfortunately, the way the TRACE_EVENT()
macro works, anything placed in the TP_printk() will be shown pretty
much exactly as is. The problem arises when enums are used. That's
because unlike macros, enums will not be changed into their values by
the C pre-processor. Thus, the enum string is exported to the format
file, and this makes it useless for user space tools.
The TRACE_DEFINE_ENUM() solves this by converting the enum strings in
the TP_printk() format into their number, and that is what is shown to
user space. For example, the tracepoint tlb_flush currently has this
in its format file:
__print_symbolic(REC->reason,
{ TLB_FLUSH_ON_TASK_SWITCH, "flush on task switch" },
{ TLB_REMOTE_SHOOTDOWN, "remote shootdown" },
{ TLB_LOCAL_SHOOTDOWN, "local shootdown" },
{ TLB_LOCAL_MM_SHOOTDOWN, "local mm shootdown" })
After adding:
TRACE_DEFINE_ENUM(TLB_FLUSH_ON_TASK_SWITCH);
TRACE_DEFINE_ENUM(TLB_REMOTE_SHOOTDOWN);
TRACE_DEFINE_ENUM(TLB_LOCAL_SHOOTDOWN);
TRACE_DEFINE_ENUM(TLB_LOCAL_MM_SHOOTDOWN);
Its format file will contain this:
__print_symbolic(REC->reason,
{ 0, "flush on task switch" },
{ 1, "remote shootdown" },
{ 2, "local shootdown" },
{ 3, "local mm shootdown" })"
* tag 'trace-v4.1' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace: (27 commits)
tracing: Add enum_map file to show enums that have been mapped
writeback: Export enums used by tracepoint to user space
v4l: Export enums used by tracepoints to user space
SUNRPC: Export enums in tracepoints to user space
mm: tracing: Export enums in tracepoints to user space
irq/tracing: Export enums in tracepoints to user space
f2fs: Export the enums in the tracepoints to userspace
net/9p/tracing: Export enums in tracepoints to userspace
x86/tlb/trace: Export enums in used by tlb_flush tracepoint
tracing/samples: Update the trace-event-sample.h with TRACE_DEFINE_ENUM()
tracing: Allow for modules to convert their enums to values
tracing: Add TRACE_DEFINE_ENUM() macro to map enums to their values
tracing: Update trace-event-sample with TRACE_SYSTEM_VAR documentation
tracing: Give system name a pointer
brcmsmac: Move each system tracepoints to their own header
iwlwifi: Move each system tracepoints to their own header
mac80211: Move message tracepoints to their own header
tracing: Add TRACE_SYSTEM_VAR to xhci-hcd
tracing: Add TRACE_SYSTEM_VAR to kvm-s390
tracing: Add TRACE_SYSTEM_VAR to intel-sst
...
more than one release, as I had it ready for the 4.0 merge window, but
a last minute thing that needed to go into Linux first had to be done.
That was that perf hard coded the file system number when reading
/sys/kernel/debugfs/tracing directory making sure that the path had
the debugfs mount # before it would parse the tracing file. This broke
other use cases of perf, and the check is removed.
Now when mounting /sys/kernel/debug, tracefs is automatically mounted
in /sys/kernel/debug/tracing such that old tools will still see that
path as expected. But now system admins can mount tracefs directly
and not need to mount debugfs, which can expose security issues.
A new directory is created when tracefs is configured such that
system admins can now mount it separately (/sys/kernel/tracing).
This branch is based off of Al Viro's vfs debugfs_automount branch
at commit 163f9eb95a
debugfs: Provide a file creation function that also takes an initial size
to get the debugfs_create_automount() operation.
I just noticed that Al rebased the pull to add his Signed-off-by to
that commit, and the commit is now e59b4e9187.
I did a git diff of those two and see they are the same. Only the
latter has Al's SOB.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJVLA6YAAoJEEjnJuOKh9ldv6AH/1JUINDQwV+M0VTwzbLogloo
Sco0byLhskmx5KLVD7Vs8BJAGrgHTdit32kzBGmLGJvVCKBa+c8lwmRw6rnXg3uX
K4kGp7BIyn1/geXoGpCmDKaLGXhDcw49hRzejKDg/OqFtxKTsSeQtG8fo29ps9Do
0VaF6UDp8gYplC2N2BfpB59LVndrITQ3mSsBBeFPvS7IxFJXAhDBOq2yi0aI6HyJ
ICo2L/bA9HLxMuceWrXbsun+RP68+AQlnFfAtok7AcuBzUYPCKY0shT2VMOUtpTt
1dGxMxq6q1ACfmY7gbp47WMX9aKjWcSEr0V+IYx/xex6Maf0Xsujsy99bKYUWvs=
=OcgU
-----END PGP SIGNATURE-----
Merge tag 'trace-4.1-tracefs' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracefs from Steven Rostedt:
"This adds the new tracefs file system.
This has been in linux-next for more than one release, as I had it
ready for the 4.0 merge window, but a last minute thing that needed to
go into Linux first had to be done. That was that perf hard coded the
file system number when reading /sys/kernel/debugfs/tracing directory
making sure that the path had the debugfs mount # before it would
parse the tracing file. This broke other use cases of perf, and the
check is removed.
Now when mounting /sys/kernel/debug, tracefs is automatically mounted
in /sys/kernel/debug/tracing such that old tools will still see that
path as expected. But now system admins can mount tracefs directly
and not need to mount debugfs, which can expose security issues. A
new directory is created when tracefs is configured such that system
admins can now mount it separately (/sys/kernel/tracing)"
* tag 'trace-4.1-tracefs' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
tracing: Have mkdir and rmdir be part of tracefs
tracefs: Add directory /sys/kernel/tracing
tracing: Automatically mount tracefs on debugfs/tracing
tracing: Convert the tracing facility over to use tracefs
tracefs: Add new tracefs file system
tracing: Create cmdline tracer options on tracing fs init
tracing: Only create tracer options files if directory exists
debugfs: Provide a file creation function that also takes an initial size
Dynamically allocated trampolines call ftrace_ops_get_func to get the
function which they should call. For dynamic fops (FTRACE_OPS_FL_DYNAMIC
flag is set) ftrace_ops_list_func is always returned. This is reasonable
for static trampolines but goes against the main advantage of dynamic
ones, that is avoidance of going through the list of all registered
callbacks for functions that are only being traced by a single callback.
We can fix it by returning ops->func (or recursion safe version) from
ftrace_ops_get_func whenever it is possible for dynamic trampolines.
Note that dynamic trampolines are not allowed for dynamic fops if
CONFIG_PREEMPT=y.
Link: http://lkml.kernel.org/r/alpine.LNX.2.00.1501291023000.25445@pobox.suse.cz
Link: http://lkml.kernel.org/r/1424357773-13536-1-git-send-email-mbenes@suse.cz
Reported-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Some archs (specifically PowerPC), are sensitive with the ordering of
the enabling of the calls to function tracing and setting of the
function to use to be traced.
That is, update_ftrace_function() sets what function the ftrace_caller
trampoline should call. Some archs require this to be set before
calling ftrace_run_update_code().
Another bug was discovered, that ftrace_startup_sysctl() called
ftrace_run_update_code() directly. If the function the ftrace_caller
trampoline changes, then it will not be updated. Instead a call
to ftrace_startup_enable() should be called because it tests to see
if the callback changed since the code was disabled, and will
tell the arch to update appropriately. Most archs do not need this
notification, but PowerPC does.
The problem could be seen by the following commands:
# echo 0 > /proc/sys/kernel/ftrace_enabled
# echo function > /sys/kernel/debug/tracing/current_tracer
# echo 1 > /proc/sys/kernel/ftrace_enabled
# cat /sys/kernel/debug/tracing/trace
The trace will show that function tracing was not active.
Cc: stable@vger.kernel.org # 2.6.27+
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When ftrace is enabled globally through the proc interface, we must check if
ftrace_graph_active is set. If it is set, then we should also pass the
FTRACE_START_FUNC_RET command to ftrace_run_update_code(). Similarly, when
ftrace is disabled globally through the proc interface, we must check if
ftrace_graph_active is set. If it is set, then we should also pass the
FTRACE_STOP_FUNC_RET command to ftrace_run_update_code().
Consider the following situation.
# echo 0 > /proc/sys/kernel/ftrace_enabled
After this ftrace_enabled = 0.
# echo function_graph > /sys/kernel/debug/tracing/current_tracer
Since ftrace_enabled = 0, ftrace_enable_ftrace_graph_caller() is never
called.
# echo 1 > /proc/sys/kernel/ftrace_enabled
Now ftrace_enabled will be set to true, but still
ftrace_enable_ftrace_graph_caller() will not be called, which is not
desired.
Further if we execute the following after this:
# echo nop > /sys/kernel/debug/tracing/current_tracer
Now since ftrace_enabled is set it will call
ftrace_disable_ftrace_graph_caller(), which causes a kernel warning on
the ARM platform.
On the ARM platform, when ftrace_enable_ftrace_graph_caller() is called,
it checks whether the old instruction is a nop or not. If it's not a nop,
then it returns an error. If it is a nop then it replaces instruction at
that address with a branch to ftrace_graph_caller.
ftrace_disable_ftrace_graph_caller() behaves just the opposite. Therefore,
if generic ftrace code ever calls either ftrace_enable_ftrace_graph_caller()
or ftrace_disable_ftrace_graph_caller() consecutively two times in a row,
then it will return an error, which will cause the generic ftrace code to
raise a warning.
Note, x86 does not have an issue with this because the architecture
specific code for ftrace_enable_ftrace_graph_caller() and
ftrace_disable_ftrace_graph_caller() does not check the previous state,
and calling either of these functions twice in a row has no ill effect.
Link: http://lkml.kernel.org/r/e4fbe64cdac0dd0e86a3bf914b0f83c0b419f146.1425666454.git.panand@redhat.com
Cc: stable@vger.kernel.org # 2.6.31+
Signed-off-by: Pratyush Anand <panand@redhat.com>
[
removed extra if (ftrace_start_up) and defined ftrace_graph_active as 0
if CONFIG_FUNCTION_GRAPH_TRACER is not set.
]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When /proc/sys/kernel/ftrace_enabled is set to zero, all function
tracing is disabled. But the records that represent the functions
still hold information about the ftrace_ops that are hooked to them.
ftrace_ops may request "REGS" (have a full set of pt_regs passed to
the callback), or "TRAMP" (the ops has its own trampoline to use).
When the record is updated to represent the state of the ops hooked
to it, it sets "REGS_EN" and/or "TRAMP_EN" to state that the callback
points to the correct trampoline (REGS has its own trampoline).
When ftrace_enabled is set to zero, all ftrace locations are a nop,
so they do not point to any trampoline. But the _EN flags are still
set. This can cause the accounting to go wrong when ftrace_enabled
is cleared and an ops that has a trampoline is registered or unregistered.
For example, the following will cause ftrace to crash:
# echo function_graph > /sys/kernel/debug/tracing/current_tracer
# echo 0 > /proc/sys/kernel/ftrace_enabled
# echo nop > /sys/kernel/debug/tracing/current_tracer
# echo 1 > /proc/sys/kernel/ftrace_enabled
# echo function_graph > /sys/kernel/debug/tracing/current_tracer
As function_graph uses a trampoline, when ftrace_enabled is set to zero
the updates to the record are not done. When enabling function_graph
again, the record will still have the TRAMP_EN flag set, and it will
look for an op that has a trampoline other than the function_graph
ops, and fail to find one.
Cc: stable@vger.kernel.org # 3.17+
Reported-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
o Several clean ups to the code
One such clean up was to convert to 64 bit time keeping, in the
ring buffer benchmark code.
o Adding of __print_array() helper macro for TRACE_EVENT()
o Updating the sample/trace_events/ to add samples of different ways to
make trace events. Lots of features have been added since the sample
code was made, and these features are mostly unknown. Developers
have been making their own hacks to do things that are already available.
o Performance improvements. Most notably, I found a performance bug where
a waiter that is waiting for a full page from the ring buffer will
see that a full page is not available, and go to sleep. The sched
event caused by it going to sleep would cause it to wake up again.
It would see that there was still not a full page, and go back to sleep
again, and that would wake it up again, until finally it would see a
full page. This change has been marked for stable.
Other improvements include removing global locks from fast paths.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJU3M+GAAoJEEjnJuOKh9ldpWQIAJTUzeVXlU0cf3bVn768VW7e
XS41WHF34l1tNevmKTh6fCPiw8+U0UMGLQt5WKtyaaARsZn2MlefLVuvHPKFlK2w
+qcI4OEVHH97Qgf9HWJSsYgnZaOnOE+TENqnokEgXMimRMuVcd/S4QaGxwJVDcjm
iBF5j2TaG4aGbx4a3J7KueoZ3K+39r3ut15hIGi/IZBZldQ1pt26ytafD/KA3CU3
BLRM2HLttAMsV1ds0EDLgZjSGICVetFcdOmI5Gwj7Qr3KrOTRPYJMNc8NdDL7Js9
v8VhujhFGvcCrhO/IKpVvd9yluz3RCF+Z7ihc+D/+1B3Nsm0PTwN3Fl5J+f89AA=
=u2Mm
-----END PGP SIGNATURE-----
Merge tag 'trace-v3.20' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracing updates from Steven Rostedt:
"The updates included in this pull request for ftrace are:
o Several clean ups to the code
One such clean up was to convert to 64 bit time keeping, in the
ring buffer benchmark code.
o Adding of __print_array() helper macro for TRACE_EVENT()
o Updating the sample/trace_events/ to add samples of different ways
to make trace events. Lots of features have been added since the
sample code was made, and these features are mostly unknown.
Developers have been making their own hacks to do things that are
already available.
o Performance improvements. Most notably, I found a performance bug
where a waiter that is waiting for a full page from the ring buffer
will see that a full page is not available, and go to sleep. The
sched event caused by it going to sleep would cause it to wake up
again. It would see that there was still not a full page, and go
back to sleep again, and that would wake it up again, until finally
it would see a full page. This change has been marked for stable.
Other improvements include removing global locks from fast paths"
* tag 'trace-v3.20' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
ring-buffer: Do not wake up a splice waiter when page is not full
tracing: Fix unmapping loop in tracing_mark_write
tracing: Add samples of DECLARE_EVENT_CLASS() and DEFINE_EVENT()
tracing: Add TRACE_EVENT_FN example
tracing: Add TRACE_EVENT_CONDITION sample
tracing: Update the TRACE_EVENT fields available in the sample code
tracing: Separate out initializing top level dir from instances
tracing: Make tracing_init_dentry_tr() static
trace: Use 64-bit timekeeping
tracing: Add array printing helper
tracing: Remove newline from trace_printk warning banner
tracing: Use IS_ERR() check for return value of tracing_init_dentry()
tracing: Remove unneeded includes of debugfs.h and fs.h
tracing: Remove taking of trace_types_lock in pipe files
tracing: Add ref count to tracer for when they are being read by pipe
debugfs was fine for the tracing facility as a quick way to get
an interface. Now that tracing has matured, it should separate itself
from debugfs such that it can be mounted separately without needing
to mount all of debugfs with it. That is, users resist using tracing
because it requires mounting debugfs. Having tracing have its own file
system lets users get the features of tracing without needing to bring
in the rest of the kernel's debug infrastructure.
Another reason for tracefs is that debubfs does not support mkdir.
Currently, to create instances, one does a mkdir in the tracing/instance
directory. This is implemented via a hack that forces debugfs to do
something it is not intended on doing. By converting over to tracefs, this
hack can be removed and mkdir can be properly implemented. This patch does
not address this yet, but it lays the ground work for that to be done.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Pull in Al Viro's changes to debugfs that implement the new primitive:
debugfs_create_automount(), that creates a directory in debugfs that will
safely mount another file system automatically when debugfs is mounted.
This will let tracefs automount itself on top of debugfs/tracing directory.
tracing_init_dentry() will soon return NULL as a valid pointer for the
top level tracing directroy. NULL can not be used as an error value.
Instead, switch to ERR_PTR() and check the return status with
IS_ERR().
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Using just the filter for checking for trampolines or regs is not enough
when updating the code against the records that represent all functions.
Both the filter hash and the notrace hash need to be checked.
To trigger this bug (using trace-cmd and perf):
# perf probe -a do_fork
# trace-cmd start -B foo -e probe
# trace-cmd record -p function_graph -n do_fork sleep 1
The trace-cmd record at the end clears the filter before it disables
function_graph tracing and then that causes the accounting of the
ftrace function records to become incorrect and causes ftrace to bug.
Link: http://lkml.kernel.org/r/20150114154329.358378039@goodmis.org
Cc: stable@vger.kernel.org
[ still need to switch old_hash_ops to old_ops_hash ]
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
As the set_ftrace_filter affects both the function tracer as well as the
function graph tracer, the ops that represent each have a shared
ftrace_ops_hash structure. This allows both to be updated when the filter
files are updated.
But if function graph is enabled and the global_ops (function tracing) ops
is not, then it is possible that the filter could be changed without the
update happening for the function graph ops. This will cause the changes
to not take place and may even cause a ftrace_bug to occur as it could mess
with the trampoline accounting.
The solution is to check if the ops uses the shared global_ops filter and
if the ops itself is not enabled, to check if there's another ops that is
enabled and also shares the global_ops filter. In that case, the
modification still needs to be executed.
Link: http://lkml.kernel.org/r/20150114154329.055980438@goodmis.org
Cc: stable@vger.kernel.org # 3.17+
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Introduce FTRACE_OPS_FL_IPMODIFY to avoid conflict among
ftrace users who may modify regs->ip to change the execution
path. If two or more users modify the regs->ip on the same
function entry, one of them will be broken. So they must add
IPMODIFY flag and make sure that ftrace_set_filter_ip() succeeds.
Note that ftrace doesn't allow ftrace_ops which has IPMODIFY
flag to have notrace hash, and the ftrace_ops must have a
filter hash (so that the ftrace_ops can hook only specific
entries), because it strongly depends on the address and
must be allowed for only few selected functions.
Link: http://lkml.kernel.org/r/20141121102516.11844.27829.stgit@localhost.localdomain
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Seth Jennings <sjenning@redhat.com>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Vojtech Pavlik <vojtech@suse.cz>
Cc: Miroslav Benes <mbenes@suse.cz>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
[ fixed up some of the comments ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Fix up a few typos in comments and convert an int into a bool in
update_traceon_count().
Link: http://lkml.kernel.org/r/546DD445.5080108@hitachi.com
Suggested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Stack traces that happen from function tracing check if the address
on the stack is a __kernel_text_address(). That is, is the address
kernel code. This calls core_kernel_text() which returns true
if the address is part of the builtin kernel code. It also calls
is_module_text_address() which returns true if the address belongs
to module code.
But what is missing is ftrace dynamically allocated trampolines.
These trampolines are allocated for individual ftrace_ops that
call the ftrace_ops callback functions directly. But if they do a
stack trace, the code checking the stack wont detect them as they
are neither core kernel code nor module address space.
Adding another field to ftrace_ops that also stores the size of
the trampoline assigned to it we can create a new function called
is_ftrace_trampoline() that returns true if the address is a
dynamically allocate ftrace trampoline. Note, it ignores trampolines
that are not dynamically allocated as they will return true with
the core_kernel_text() function.
Link: http://lkml.kernel.org/r/20141119034829.497125839@goodmis.org
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
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>
With the new logic, if only a single user of ftrace function hooks is
used, it will get its own trampoline assigned to it.
The problem is that the control_ops is an indirect ops that perf ops
uses. What that means is that when perf registers its ops with
register_ftrace_function(), it has the CONTROL flag set and gets added
to the control list instead of the global ftrace list. The control_ops
gets added to that instead and the mcount trampoline calls the control_ops
function. The control_ops function will iterate the control list and
call the ops functions that are attached to it.
But currently the trampoline is added to the perf ops and not the
control ops, and when ftrace tries to find a trampoline hook for it,
it fails to find one and gives the following splat:
------------[ cut here ]------------
WARNING: CPU: 0 PID: 10133 at kernel/trace/ftrace.c:2033 ftrace_get_addr_new+0x6f/0xc0()
Modules linked in: [...]
CPU: 0 PID: 10133 Comm: perf Tainted: P 3.18.0-rc1-test+ #388
Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v02.05 05/07/2012
00000000000007f1 ffff8800c2643bc8 ffffffff814fca6e ffff88011ea0ed01
0000000000000000 ffff8800c2643c08 ffffffff81041ffd 0000000000000000
ffffffff810c388c ffffffff81a5a350 ffff880119b00000 ffffffff810001c8
Call Trace:
[<ffffffff814fca6e>] dump_stack+0x46/0x58
[<ffffffff81041ffd>] warn_slowpath_common+0x81/0x9b
[<ffffffff810c388c>] ? ftrace_get_addr_new+0x6f/0xc0
[<ffffffff810001c8>] ? 0xffffffff810001c8
[<ffffffff81042031>] warn_slowpath_null+0x1a/0x1c
[<ffffffff810c388c>] ftrace_get_addr_new+0x6f/0xc0
[<ffffffff8102e938>] ftrace_replace_code+0xd6/0x334
[<ffffffff810c4116>] ftrace_modify_all_code+0x41/0xc5
[<ffffffff8102eba6>] arch_ftrace_update_code+0x10/0x19
[<ffffffff810c293c>] ftrace_run_update_code+0x21/0x42
[<ffffffff810c298f>] ftrace_startup_enable+0x32/0x34
[<ffffffff810c3049>] ftrace_startup+0x14e/0x15a
[<ffffffff810c307c>] register_ftrace_function+0x27/0x40
[<ffffffff810dc118>] perf_ftrace_event_register+0x3e/0xee
[<ffffffff810dbfbe>] perf_trace_init+0x29d/0x2a9
[<ffffffff810eb422>] perf_tp_event_init+0x27/0x3a
[<ffffffff810f18bc>] perf_init_event+0x9e/0xed
[<ffffffff810f1ba4>] perf_event_alloc+0x299/0x330
[<ffffffff810f236b>] SYSC_perf_event_open+0x3ee/0x816
[<ffffffff8115a066>] ? mntput+0x2d/0x2f
[<ffffffff81142b00>] ? __fput+0xa7/0x1b2
[<ffffffff81091300>] ? do_gettimeofday+0x22/0x3a
[<ffffffff810f279c>] SyS_perf_event_open+0x9/0xb
[<ffffffff81502a92>] system_call_fastpath+0x12/0x17
---[ end trace 81a53565150e4982 ]---
Bad trampoline accounting at: ffffffff810001c8 (run_init_process+0x0/0x2d) (10000001)
Update the control_ops trampoline instead of the perf ops one.
Reported-by: lkp@01.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
With the introduction of the dynamic trampolines, it is useful that if
things go wrong that ftrace_bug() produces more information about what
the current state is. This can help debug issues that may arise.
Ftrace has lots of checks to make sure that the state of the system it
touchs is exactly what it expects it to be. When it detects an abnormality
it calls ftrace_bug() and disables itself to prevent any further damage.
It is crucial that ftrace_bug() produces sufficient information that
can be used to debug the situation.
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Borislav Petkov <bp@suse.de>
Tested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When the static ftrace_ops (like function tracer) enables tracing, and it
is the only callback that is referencing a function, a trampoline is
dynamically allocated to the function that calls the callback directly
instead of calling a loop function that iterates over all the registered
ftrace ops (if more than one ops is registered).
But when it comes to dynamically allocated ftrace_ops, where they may be
freed, on a CONFIG_PREEMPT kernel there's no way to know when it is safe
to free the trampoline. If a task was preempted while executing on the
trampoline, there's currently no way to know when it will be off that
trampoline.
But this is not true when it comes to !CONFIG_PREEMPT. The current method
of calling schedule_on_each_cpu() will force tasks off the trampoline,
becaues they can not schedule while on it (kernel preemption is not
configured). That means it is safe to free a dynamically allocated
ftrace ops trampoline when CONFIG_PREEMPT is not configured.
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Borislav Petkov <bp@suse.de>
Tested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The file /sys/kernel/debug/tracing/eneabled_functions is used to debug
ftrace function hooks. Add to the output what function is being called
by the trampoline if the arch supports it.
Add support for this feature in x86_64.
Cc: H. Peter Anvin <hpa@linux.intel.com>
Tested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The current method of handling multiple function callbacks is to register
a list function callback that calls all the other callbacks based on
their hash tables and compare it to the function that the callback was
called on. But this is very inefficient.
For example, if you are tracing all functions in the kernel and then
add a kprobe to a function such that the kprobe uses ftrace, the
mcount trampoline will switch from calling the function trace callback
to calling the list callback that will iterate over all registered
ftrace_ops (in this case, the function tracer and the kprobes callback).
That means for every function being traced it checks the hash of the
ftrace_ops for function tracing and kprobes, even though the kprobes
is only set at a single function. The kprobes ftrace_ops is checked
for every function being traced!
Instead of calling the list function for functions that are only being
traced by a single callback, we can call a dynamically allocated
trampoline that calls the callback directly. The function graph tracer
already uses a direct call trampoline when it is being traced by itself
but it is not dynamically allocated. It's trampoline is static in the
kernel core. The infrastructure that called the function graph trampoline
can also be used to call a dynamically allocated one.
For now, only ftrace_ops that are not dynamically allocated can have
a trampoline. That is, users such as function tracer or stack tracer.
kprobes and perf allocate their ftrace_ops, and until there's a safe
way to free the trampoline, it can not be used. The dynamically allocated
ftrace_ops may, although, use the trampoline if the kernel is not
compiled with CONFIG_PREEMPT. But that will come later.
Tested-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When modifying code, ftrace has several checks to make sure things
are being done correctly. One of them is to make sure any code it
modifies is exactly what it expects it to be before it modifies it.
In order to do so with the new trampoline logic, it must be able
to find out what trampoline a function is hooked to in order to
see if the code that hooks to it is what's expected.
The logic to find the trampoline from a record (accounting descriptor
for a function that is hooked) needs to only look at the "old_hash"
of an ops that is being modified. The old_hash is the list of function
an ops is hooked to before its update. Since a record would only be
pointing to an ops that is being modified if it was already hooked
before.
Currently, it can pick a modified ops based on its new functions it
will be hooked to, and this picks the wrong trampoline and causes
the check to fail, disabling ftrace.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ftrace: squash into ordering of ops for modification
The code that checks for trampolines when modifying function hooks
tests against a modified ops "old_hash". But the ops old_hash pointer
is not being updated before the changes are made, making it possible
to not find the right hash to the callback and possibly causing
ftrace to break in accounting and disable itself.
Have the ops set its old_hash before the modifying takes place.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When the last ftrace_ops is unregistered, all the function records should
have a zeroed flags value. Make sure that is the case when the last ftrace_ops
is unregistered.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Allowing function callbacks to declare their own trampolines requires
that each ftrace_ops that has a trampoline must have some sort of
accounting that keeps track of which ops has a trampoline attached
to a record.
The easy way to solve this was to add a "tramp_hash" that created a
hash entry for every function that a ops uses with a trampoline.
But since we can have literally tens of thousands of functions being
traced, that means we need tens of thousands of descriptors to map
the ops to the function in the hash. This is quite expensive and
can cause enabling and disabling the function graph tracer to take
some time to start and stop. It can take up to several seconds to
disable or enable all functions in the function graph tracer for this
reason.
The better approach albeit more complex, is to keep track of how ops
are being enabled and disabled, and use that along with the counting
of the number of ops attached to records, to determive what ops has
a trampoline attached to a record at enabling and disabling of
tracing.
To do this, the tramp_hash has been replaced with an old_filter_hash
and old_notrace_hash, which get the copy of the ops filter_hash and
notrace_hash respectively. The old hashes is kept until the ops has
been modified or removed and the old hashes are used with the logic
of the accounting to determine the ops that have the trampoline of
a record. The reason this has less of a footprint is due to the trick
that an "empty" hash in the filter_hash means "all functions" and
an empty hash in the notrace hash means "no functions" in the hash.
This is much more efficienct, doesn't have the delay, and takes up
much less memory, as we do not need to map all the functions but
just figure out which functions are mapped at the time it is
enabled or disabled.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add three new flags for ftrace_ops:
FTRACE_OPS_FL_ADDING
FTRACE_OPS_FL_REMOVING
FTRACE_OPS_FL_MODIFYING
These will be set for the ftrace_ops when they are first added
to the function tracing, being removed from function tracing
or just having their functions changed from function tracing,
respectively.
This will be needed to remove the tramp_hash, which can grow quite
big. The tramp_hash is used to note what functions a ftrace_ops
is using a trampoline for. Denoting which ftrace_ops is being
modified, will allow us to use the ftrace_ops hashes themselves,
which are much smaller as they have a global flag to denote if
a ftrace_ops is tracing all functions, as well as a notrace hash
if the ftrace_ops is tracing all but a few. The tramp_hash just
creates a hash item for every function, which can go into the 10s
of thousands if all functions are using the ftrace_ops trampoline.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When dumping the enabled_functions, use the first op that is
found with a trampoline to the record, as there should only be
one, as only one ops can be registered to a function that has
a trampoline.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ftrace_hash_move() currently frees the old hash that is passed to it
after replacing the pointer with the new hash. Instead of having the
function do that chore, have the caller perform the free.
This lets the ftrace_hash_move() be used a bit more freely, which
is needed for changing the way the trampoline logic is done.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Qiu Peiyang
Steven Rostedt (Red Hat)
Abel Vesa
Li Bin
Linus Torvalds
Dmitry Safonov
Peter Zijlstra
Rasmus Villemoes
Umesh Tiwari
Pratyush Anand
Masami Hiramatsu