The contents of this directory allow users to specify PMU events in their
CPUs by their symbolic names rather than raw event codes (see example below).
The main program in this directory, is the 'jevents', which is built and
executed _BEFORE_ the perf binary itself is built.
The 'jevents' program tries to locate and process JSON files in the directory
tree tools/perf/pmu-events/arch/foo.
- Regular files with '.json' extension in the name are assumed to be
JSON files, each of which describes a set of PMU events.
- Regular files with basename starting with 'mapfile.csv' are assumed
to be a CSV file that maps a specific CPU to its set of PMU events.
(see below for mapfile format)
- Directories are traversed, but all other files are ignored.
The PMU events supported by a CPU model are expected to grouped into topics
such as Pipelining, Cache, Memory, Floating-point etc. All events for a topic
should be placed in a separate JSON file - where the file name identifies
the topic. Eg: "Floating-point.json".
All the topic JSON files for a CPU model/family should be in a separate
sub directory. Thus for the Silvermont X86 CPU:
$ ls tools/perf/pmu-events/arch/x86/Silvermont_core
Cache.json Memory.json Virtual-Memory.json
Frontend.json Pipeline.json
Using the JSON files and the mapfile, 'jevents' generates the C source file,
'pmu-events.c', which encodes the two sets of tables:
- Set of 'PMU events tables' for all known CPUs in the architecture,
(one table like the following, per JSON file; table name 'pme_power8'
is derived from JSON file name, 'power8.json').
struct pmu_event pme_power8[] = {
...
{
.name = "pm_1plus_ppc_cmpl",
.event = "event=0x100f2",
.desc = "1 or more ppc insts finished,",
},
...
}
- A 'mapping table' that maps each CPU of the architecture, to its
'PMU events table'
struct pmu_events_map pmu_events_map[] = {
{
.cpuid = "004b0000",
.version = "1",
.type = "core",
.table = pme_power8
},
...
};
After the 'pmu-events.c' is generated, it is compiled and the resulting
'pmu-events.o' is added to 'libperf.a' which is then used to build perf.
NOTES:
1. Several CPUs can support same set of events and hence use a common
JSON file. Hence several entries in the pmu_events_map[] could map
to a single 'PMU events table'.
2. The 'pmu-events.h' has an extern declaration for the mapping table
and the generated 'pmu-events.c' defines this table.
3. _All_ known CPU tables for architecture are included in the perf
binary.
At run time, perf determines the actual CPU it is running on, finds the
matching events table and builds aliases for those events. This allows
users to specify events by their name:
$ perf stat -e pm_1plus_ppc_cmpl sleep 1
where 'pm_1plus_ppc_cmpl' is a Power8 PMU event.
In case of errors when processing files in the tools/perf/pmu-events/arch
directory, 'jevents' tries to create an empty mapping file to allow the perf
build to succeed even if the PMU event aliases cannot be used.
However some errors in processing may cause the perf build to fail.
Mapfile format
===============
The mapfile enables multiple CPU models to share a single set of PMU events.
It is required even if such mapping is 1:1.
The mapfile.csv format is expected to be:
Header line
CPUID,Version,Dir/path/name,Type
where:
Comma:
is the required field delimiter (i.e other fields cannot
have commas within them).
Comments:
Lines in which the first character is either '\n' or '#'
are ignored.
Header line
The header line is the first line in the file, which is
always _IGNORED_. It can empty.
CPUID:
CPUID is an arch-specific char string, that can be used
to identify CPU (and associate it with a set of PMU events
it supports). Multiple CPUIDS can point to the same
File/path/name.json.
Example:
CPUID == 'GenuineIntel-6-2E' (on x86).
CPUID == '004b0100' (PVR value in Powerpc)
Version:
is the Version of the mapfile.
Dir/path/name:
is the pathname to the directory containing the CPU's JSON
files, relative to the directory containing the mapfile.csv
Type:
indicates whether the events or "core" or "uncore" events.
Eg:
$ grep Silvermont tools/perf/pmu-events/arch/x86/mapfile.csv
GenuineIntel-6-37,V13,Silvermont_core,core
GenuineIntel-6-4D,V13,Silvermont_core,core
GenuineIntel-6-4C,V13,Silvermont_core,core
i.e the three CPU models use the JSON files (i.e PMU events) listed
in the directory 'tools/perf/pmu-events/arch/x86/Silvermont_core'.
ype='submit' value='reload'/>
percpu-refcount: fix reference leak during percpu-atomic transition
percpu_ref_tryget() and percpu_ref_tryget_live() should return
"true" IFF they acquire a reference. But the return value from
atomic_long_inc_not_zero() is a long and may have high bits set,
e.g. PERCPU_COUNT_BIAS, and the return value of the tryget routines
is bool so the reference may actually be acquired but the routines
return "false" which results in a reference leak since the caller
assumes it does not need to do a corresponding percpu_ref_put().
This was seen when performing CPU hotplug during I/O, as hangs in
blk_mq_freeze_queue_wait where percpu_ref_kill (blk_mq_freeze_queue_start)
raced with percpu_ref_tryget (blk_mq_timeout_work).
Sample stack trace:
__switch_to+0x2c0/0x450
__schedule+0x2f8/0x970
schedule+0x48/0xc0
blk_mq_freeze_queue_wait+0x94/0x120
blk_mq_queue_reinit_work+0xb8/0x180
blk_mq_queue_reinit_prepare+0x84/0xa0
cpuhp_invoke_callback+0x17c/0x600
cpuhp_up_callbacks+0x58/0x150
_cpu_up+0xf0/0x1c0
do_cpu_up+0x120/0x150
cpu_subsys_online+0x64/0xe0
device_online+0xb4/0x120
online_store+0xb4/0xc0
dev_attr_store+0x68/0xa0
sysfs_kf_write+0x80/0xb0
kernfs_fop_write+0x17c/0x250
__vfs_write+0x6c/0x1e0
vfs_write+0xd0/0x270
SyS_write+0x6c/0x110
system_call+0x38/0xe0
Examination of the queue showed a single reference (no PERCPU_COUNT_BIAS,
and __PERCPU_REF_DEAD, __PERCPU_REF_ATOMIC set) and no requests.
However, conditions at the time of the race are count of PERCPU_COUNT_BIAS + 0
and __PERCPU_REF_DEAD and __PERCPU_REF_ATOMIC set.
The fix is to make the tryget routines use an actual boolean internally instead
of the atomic long result truncated to a int.
Fixes: e625305b3907 percpu-refcount: make percpu_ref based on longs instead of ints
Link: https://bugzilla.kernel.org/show_bug.cgi?id=190751
Signed-off-by: Douglas Miller <dougmill@linux.vnet.ibm.com>
Reviewed-by: Jens Axboe <axboe@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: e625305b3907 ("percpu-refcount: make percpu_ref based on longs instead of ints")
Cc: stable@vger.kernel.org # v3.18+