Similarly to ETMv3, moving etmv4_drvdata::enable to an atomic
type that gives the 'mode' of a tracer and prevents multiple,
simultanious access by different subsystems.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit c38a9ec2b2)
As with the ETMv3.x driver, calling 'smp_call_function_single()'
twice in a row is highly ineffective. As such moving function
'etm4_os_unlock()' before the default initialisation takes
place, which results in the same outcome.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 66bbbb7754)
Splitting and updating the default initialisation for each etmv4
configuration so that it can be called at the beginning of each
session rather than initialisation time only.
Since the trace ID isn't expected to change with every session,
moving it with the default tracer initialisation.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit fc208abef3)
Similar to what was done on etm3x, splitting driver structure
etmv4_drvdata in two. One half is concerned with the HW
characteristics that are generally static in nature. The other
half deals with user configuration and will change from one
trace session to another.
No gain/loss of functionality is incurred from this patch.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 54ff892b76)
Adding new sysFS management interface to query the configuration
and the traceid registers. Both are required to convey information
to the perf cmd line tools when using ETMv4 tracers as PMU.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 7c38aa4b03)
As with the etm3x driver, sysFS entries are big enough to justify
their own file. As such moving all sysFS related declarations to
a dedicated location.
No gain/loss of functionality is incurred from this patch.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit a77de2637c)
Some STM devices adjust software assigned master numbers depending on
the trace source and its runtime state and whatnot. This patch adds
a sysfs attribute to inform the trace-side software that master numbers
assigned to software sources will not match those in the STP stream,
so that, for example, master/channel allocation policy can be adjusted
accordingly.
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 8e996a2874)
Readability: a postfix increment is used on a pointer which is not
used anywhere afterwards, which may send the reader looking through
the function one extra time. Drop the unnecessary increment.
Reported-by: Alan Cox <alan.cox@intel.com>
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Laurent Fert <laurent.fert@intel.com>
(cherry picked from commit fb0801904b)
Currently, stm_register_device() makes the device visible and then
proceeds to initializing spinlocks and other properties, which leaves
a window when the device can already be opened but is not yet fully
operational.
Fix this by reversing the initialization order.
Reported-by: Alan Cox <alan.cox@intel.com>
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Reviewed-by: Laurent Fert <laurent.fert@intel.com>
(cherry picked from commit 389b6699a2)
Fixes segmentation fault using, for instance:
(gdb) run record -I -e intel_pt/tsc=1,noretcomp=1/u /bin/ls
Starting program: /home/acme/bin/perf record -I -e intel_pt/tsc=1,noretcomp=1/u /bin/ls
Missing separate debuginfos, use: dnf debuginfo-install glibc-2.22-7.fc23.x86_64
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
Program received signal SIGSEGV, Segmentation fault.
0 x00000000004b9ea5 in tracepoint_error (e=0x0, err=13, sys=0x19b1370 "sched", name=0x19a5d00 "sched_switch") at util/parse-events.c:410
(gdb) bt
#0 0x00000000004b9ea5 in tracepoint_error (e=0x0, err=13, sys=0x19b1370 "sched", name=0x19a5d00 "sched_switch") at util/parse-events.c:410
#1 0x00000000004b9fc5 in add_tracepoint (list=0x19a5d20, idx=0x7fffffffb8c0, sys_name=0x19b1370 "sched", evt_name=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:433
#2 0x00000000004ba334 in add_tracepoint_event (list=0x19a5d20, idx=0x7fffffffb8c0, sys_name=0x19b1370 "sched", evt_name=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:498
#3 0x00000000004bb699 in parse_events_add_tracepoint (list=0x19a5d20, idx=0x7fffffffb8c0, sys=0x19b1370 "sched", event=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:936
#4 0x00000000004f6eda in parse_events_parse (_data=0x7fffffffb8b0, scanner=0x19a49d0) at util/parse-events.y:391
#5 0x00000000004bc8e5 in parse_events__scanner (str=0x663ff2 "sched:sched_switch", data=0x7fffffffb8b0, start_token=258) at util/parse-events.c:1361
#6 0x00000000004bca57 in parse_events (evlist=0x19a5220, str=0x663ff2 "sched:sched_switch", err=0x0) at util/parse-events.c:1401
#7 0x0000000000518d5f in perf_evlist__can_select_event (evlist=0x19a3b90, str=0x663ff2 "sched:sched_switch") at util/record.c:253
#8 0x0000000000553c42 in intel_pt_track_switches (evlist=0x19a3b90) at arch/x86/util/intel-pt.c:364
#9 0x00000000005549d1 in intel_pt_recording_options (itr=0x19a2c40, evlist=0x19a3b90, opts=0x8edf68 <record+232>) at arch/x86/util/intel-pt.c:664
#10 0x000000000051e076 in auxtrace_record__options (itr=0x19a2c40, evlist=0x19a3b90, opts=0x8edf68 <record+232>) at util/auxtrace.c:539
#11 0x0000000000433368 in cmd_record (argc=1, argv=0x7fffffffde60, prefix=0x0) at builtin-record.c:1264
#12 0x000000000049bec2 in run_builtin (p=0x8fa2a8 <commands+168>, argc=5, argv=0x7fffffffde60) at perf.c:390
#13 0x000000000049c12a in handle_internal_command (argc=5, argv=0x7fffffffde60) at perf.c:451
#14 0x000000000049c278 in run_argv (argcp=0x7fffffffdcbc, argv=0x7fffffffdcb0) at perf.c:495
#15 0x000000000049c60a in main (argc=5, argv=0x7fffffffde60) at perf.c:618
(gdb)
Intel PT attempts to find the sched:sched_switch tracepoint but that seg
faults if tracefs is not readable, because the error reporting structure
is null, as errors are not reported when automatically adding
tracepoints. Fix by checking before using.
Committer note:
This doesn't take place in a kernel that supports
perf_event_attr.context_switch, that is the default way that will be
used for tracking context switches, only in older kernels, like 4.2, in
a machine with Intel PT (e.g. Broadwell) for non-priviledged users.
Further info from a similar patch by Wang:
The error is in tracepoint_error: it assumes the 'e' parameter is valid.
However, there are many situation a parse_event() can be called without
parse_events_error. See result of
$ grep 'parse_events(.*NULL)' ./tools/perf/ -r'
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Tong Zhang <ztong@vt.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: stable@vger.kernel.org # v4.4+
Fixes: 196581717d ("perf tools: Enhance parsing events tracepoint error output")
Link: http://lkml.kernel.org/r/1453809921-24596-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
(cherry picked from commit ec183d22cc)
In commit 941943cf51 ("drivers/hwtracing:
make coresight-* explicitly non-modular") we removed all uses of
modular functions/macros in favour of their built-in equivlents in
this subsystem.
However that commit and commit 0bcbf2e30f
("coresight: etm-perf: new PMU driver for ETM tracers") were in flight
at the same time, and hence one new non-modular user of module_init
crept back in. Fix it up like we did all the others.
Since module_init translates to device_initcall in the non-modular
case, the init ordering remains unchanged with this commit.
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit ca48fa22c3)
None of the Kconfig currently controlling compilation of any of
the files here are tristate, meaning that none of it currently
is being built as a module by anyone.
We need not be concerned about .remove functions and blocking the
unbind sysfs operations, since that was already done in a recent
commit.
Lets remove any remaining modular references, so that when reading the
drivers there is no doubt they are builtin-only.
All drivers get mostly the same changes, so they are handled in batch.
Changes are (1) convert to builtin_amba_driver, (2) delete module.h
include where unused, and (3) relocate the description into the
comments so we don't need MODULE_DESCRIPTION and associated tags.
The etm3x and etm4x use module_param_named, and have been adjusted
to just include moduleparam.h for that purpose.
In commit f309d44431 ("platform_device:
better support builtin boilerplate avoidance") we introduced the
builtin_driver macro.
Here we use that support and extend it to amba driver registration,
so where a driver is clearly non-modular and builtin-only, we can
update with the simple mapping of
module_amba_driver(...) ---> builtin_amba_driver(...)
Since module_amba_driver() uses the same init level priority as
builtin_amba_driver() the init ordering remains unchanged with
this commit.
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 941943cf51)
TraceID values have to be unique for all tracers and
consistent between drivers and user space. As such
introducing a central function to be used whenever a
traceID value is required.
The patch also account for data traceIDs, which are usually
I(N) + 1.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 17534ceb83)
Perf is a well known and used tool for performance monitoring
and much more. A such it is an ideal candidate for integration
with coresight based HW tracing.
This patch introduces a PMU that represent a coresight tracer to
the Perf core.
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 0bcbf2e30f)
Adding an ETB10 specific AUX area operations to be used
by the perf framework when events are initialised.
Part of this operation involves modeling the mmap'ed area
based on the specific ways a sink buffer gathers information.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 2997aa4063)
Adding an operation mode to the sink->enable() API in order
to prevent simultaneous access from different callers.
TPIU and TMC won't be supplemented with the AUX area
API immediately and as such ignore the new mode.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit e827d4550a)
Moving to use local atomic operations to take advantage of the
lockless implementation, something that will come handy when
the ETB is accessed from the Perf subsystem. Also changing the
name of the variable to something more meaningful.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 27b10da8ff)
That way traces can be enabled and disabled automatically
from the Perf subystem using the PMU abstraction.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 882d5e1124)
There is really no point in having two functions to take care
of doing the initial tracer configuration. As such moving
everything to 'etm_set_default()'.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit c528a25ac7)
Changing default configuration to include the entire address
range rather than just the kernel. That way traces are more
inclusive and it is easier to narrow down if needed.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit e19217299c)
There is no need to use the event enable's "always false" event to
stop trace collection. For that purpose setting the programming bit
(ETMCR:10) is enough.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 47cd066cd0)
Adding a new mode to source API enable() in order to
distinguish where the request comes from. That way it is
possible to perform different operations based on where
the request was issued from.
The ETM4x driver is also modified to keep in sync with the
new interface.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 22fd532eaa)
Splitting "etm_drvdata" in two sections, one for the HW specific
data and another for user configuration.
That way it is easier to manipulate and zero out the configuration
data when more than one concurrent tracing session configuration
is active.
Also taking care of up-lifting all the code affected by this new
arrangement. No loss or gain of functionality (other than what is
mentioned above) is introduced by this patch.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 1925a470ce)
Calling function 'smp_call_function_single()' to unlock a
tracer and calling it again right after to perform the
default initialisation doesn't make sense.
Moving 'etm_os_unlock()' just before making the default
initialisation results in the same outcome while saving
one call to 'smp_call_function_single()'.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit ae69a1da39)
SysFS entries are big enough to justify their own file.
As such moving all sysFS related declarations to a dedicated
location.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit c04148e708)
Moving functions etm_readl/writel to file "coresight-etm.h"
so that the main ETM3x driver can be split in more than one
file.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit c1f8e57c9e)
Moving PM runtime operations in Coresight devices enable() and
disable() API to the framework core when a path is setup. That
way the runtime core doesn't have to be involved everytime a
path is enabled. It also avoids calling runtime PM operations
in IRQ context.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 5da5325fa8)
Add an API allowing external code to quickly get a handle on the
sink within a path. The sink is always last, but adding an API allows
to keep the path's node structure private and remove redundant checks.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit b6404e21f0)
When using the Coresight framework from the sysFS interface a
tracer is always handling a single session and as such, a path
can be associated with a tracer. But when supporting multiple
session per tracer there is no guarantee that sessions will always
have the same path from source to sink.
This patch is removing the automatic association between path and
tracers. The building of a path and enablement of the components
in the path are decoupled, allowing for the association of a path
with a session rather than a tracer.
To keep backward functionality with the current sysFS access methods
a per-cpu place holder is used to keep a handle on the path built when
tracers are enabled. Lastly APIs to build paths and enable tracers are
made public so that other subsystem can interact with the Coresight
framework.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit b3e9440594)
The dma_alloc_coherent return an "void *" not an "void __iomen *".
It uses the wrong parameters when calls dma_free_coherent function.
The sparse tool output logs as the following:
coresight-tmc.c:199:23: expected void *<noident>
coresight-tmc.c:199:23: got void [noderef] <asn:2>*vaddr
coresight-tmc.c:336:30: warning: incorrect type in assignment
(different address spaces)
coresight-tmc.c:336:30: expected char *buf
coresight-tmc.c:336:30: got void [noderef] <asn:2>*
coresight-tmc.c:769:50: warning: incorrect type in argument 4
(different base types)
coresight-tmc.c:769:50: expected unsigned long long
[unsigned] [usertype] dma_handle
coresight-tmc.c:769:50: got restricted gfp_t
Signed-off-by: Eric Long <eric.long@linaro.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 61390593f7)
"DEVICE_ATTR_RO(name)" should be defined as static. And
there is an unnecessary space at the front of the code.
The sparse tool output logs as the following:
coresight-etm4x.c:2224:1: warning: symbol 'dev_attr_trcoslsr' was
not declared. Should it be static?
coresight-etm4x.c:2225:1: warning: symbol 'dev_attr_trcpdcr' was
not declared. Should it be static?
coresight-etm4x.c:2226:1: warning: symbol 'dev_attr_trcpdsr' was
not declared. Should it be static?
And the smatch tool output logs as the following:
of_coresight.c:89 of_coresight_alloc_memory() warn:
inconsistent indenting
Signed-off-by: Eric Long <eric.long@linaro.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit bf16e5b8cd)
Other than plainly parsing the device tree there is no way to
know which CPU a tracer is affined to. As such adding an
interface to lookup the CPU field enclosed in the etm_drvdata
structure that was initialised at boot time.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit 52210c8745)
The coresight drivers have absolutely no control over bind and unbind
operations triggered from sysfs. The operations simply can't be
cancelled or denied event when one or several tracing sessions are
under way. Since the memory associated to individual device is
invariably freed, the end result is a kernel crash when the path from
source to sink is travelled again as demonstrated here[1].
One solution could be to keep track of all the path (i.e tracing
session) that get created and iterate through the elements of those path
looking for the coresight device that is being removed. This proposition
doesn't scale well since there is no upper bound on the amount of
concurrent trace session that can be created.
With the above in mind, this patch prevent devices from being unbounded
from their driver by using the driver->suppress_bind_attr option. That way
trace sessions can be managed without fearing to loose devices.
Since device can't be removed anymore the xyz_remove() functions found in
each driver is also removed.
[1]. http://www.spinics.net/lists/arm-kernel/msg474952.html
Reported-by: Rabin Vincent <rabin@rab.in>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit b15f0fb657)
In function 'coresight_unregister()', all references to the csdev that
is being taken away need to be removed from the topology. Otherwise
building the next coresight path from source to sink may use memory
that has been released.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit ad725aee07)
The reference count taken by function bus_find_device() needs
to be released if a child device is found, something this patch
is adding.
Reported-by: Rabin Vincent <rabin@rab.in>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(cherry picked from commit f2dfab3568)
