The virPerfGetEvent method pointlessly checks for a NULL
parameter and the range of an enum value. The whole point
of using an enum is that we can avoid such checks. Just
replace calls to virPerfGetEvent, with perf->events[type]
array access.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The virPerfGetEventAttr method contains a totally pointless
loop. Remove it, verify the array size statically, and then
just use an array index to access the perf event.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

This patch adds support and documentation
for the emulation_faults perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the alignment_faults perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the page_faults_maj perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the page_faults_min perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the cpu_migrations perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the context_switches perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the page_faults perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation for
the task_clock perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation for
the cpu_clock perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

Currently, on every --enable perf_event command,
a new event->fd is created and counting of perf
event counter starts from zero and previous
event->fd is lost. This patch prevents this
behaviour.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This reverts commit ae16c95f.

This patch adds support and documentation for
a generalized hardware cache event called cache_l1d
perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

The PERF_COUNT_HW_REF_CPU_CYCLES constant is not available
on all Linux distros libvirt targets, so its use must be
made conditional. Other constant have existed long enough
that we can assume they exist, as we don't support very
old distros like RHEL-5 any more.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

This patch adds support and documentation for
the ref_cpu_cycles perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation for
the stalled_cycles_backend perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the stalled_cycles_frontend perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the bus_cycles perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

This patch adds support and documentation
for the branch_misses perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

With current perf framework, this patch adds support and documentation
for the branch_instructions perf event.
Signed-off-by: NNitesh Konkar <nitkon12@linux.vnet.ibm.com>

If the event is already disabled, then don't bother with setting it
disabled again.  Causes unnecessary error on systems that don't support
the feature anyway.

Make it clearer that the perf event is based/for the host cpu and
use the virPerfEventTypeToString to convert the type to a string

Commit id 'b00d7f29' shifted the opening of the /sys/devices/intel_cqm/type
file from event enable to perf event initialization. If the file did not
exist, then an error would be written to the domain log:

2016-09-06 20:51:21.677+0000: 7310: error : virFileReadAll:1360 : Failed to open file '/sys/devices/intel_cqm/type': No such file or directory

Since the error is now handled in virPerfEventEnable by checking if the
event_attr->attrType == 0 for CMT, MBML, and MBMT events - we can just
use the Quiet API in order to not log the error we're going to throw away.

Additionally, rather than using virReportSystemError, use virReportError
and VIR_ERR_ARGUMENT_UNSUPPORTED in order to signify that support isn't there
for that type of perf event - adjust the error message as well.

With current perf framework, this patch adds support and documentation
for more perf events, including cache misses, cache references, cpu cycles,
and instructions.
Signed-off-by: NQiaowei Ren <qiaowei.ren@intel.com>

Introduce a static attr table and refactor virPerfEventEnable() for
general purpose usage.

This patch creates a static table/matrix that converts the VIR_PERF_EVENT_*
events into their respective "attr.type" and "attr.config" so that
virPerfEventEnable doesn't have the switch the calling function passes
by value the 'type'.
Signed-off-by: NQiaowei Ren <qiaowei.ren@intel.com>

Move them to the bottom under the #ifdef code.

So imagine the following. You connect read only to a daemon and
try to fetch stats for a shut off domain, e.g.:

  virsh -r domstats $dom

but all of a sudden, virsh instead of printing the stats throws
the following error at you:

  error: Disconnected from qemu:///system due to I/O error
  error: End of file while reading data: Input/output error

The daemon crashed. This is its backtrace:

#0  0x00007fa43e3751a8 in virPerfEventIsEnabled (perf=0x0, type=VIR_PERF_EVENT_MBMT) at util/virperf.c:241
#1  0x00007fa424a9f042 in qemuDomainGetStatsPerf (driver=0x7fa3f4022a30, dom=0x7fa3f40e24c0, record=0x7fa41c000e20, maxparams=0x7fa4360b38d0, privflags=1) at qemu/qemu_driver.c:19110
#2  0x00007fa424a9f2e7 in qemuDomainGetStats (conn=0x7fa41c001b20, dom=0x7fa3f40e24c0, stats=127, record=0x7fa4360b3970, flags=1) at qemu/qemu_driver.c:19213
#3  0x00007fa424a9f672 in qemuConnectGetAllDomainStats (conn=0x7fa41c001b20, doms=0x7fa41c0017f0, ndoms=1, stats=127, retStats=0x7fa4360b3a50, flags=0) at qemu/qemu_driver.c:19303
#4  0x00007fa43e4e15f6 in virDomainListGetStats (doms=0x7fa41c0017f0, stats=0, retStats=0x7fa4360b3a50, flags=0) at libvirt-domain.c:11615

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7f28d1a38700 (LWP 16154)]
0x00007f28da4fa1a8 in virPerfEventIsEnabled (perf=0x0, type=VIR_PERF_EVENT_MBMT) at util/virperf.c:241
241         return event->enabled;

Problem is, shut off domains don't have priv->perf allocated.
Therefore if in frame #1 qemuDomainGetStatsPerf() tries to check
if perf events are enabled, NULL is passed to
virPerfEventIsEnabled() which due to some incredible
implementation dereference it. Fix this by checking whether
passed object is not NULL.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

This function is not used anywhere. Moreover, the code that would
use lives in virperf.c and therefore has access to the FD anyway.
Well, for instance virPerfReadEvent is doing just that.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

Some Intel processor families (e.g. the Intel Xeon processor E5 v3
family) introduced some RDT (Resource Director Technology) features
to monitor or control shared resource. Among these features, MBM
(Memory Bandwidth Monitoring), which is build on the CMT (Cache
Monitoring Technology) infrastructure, provides OS/VMM a way to
monitor bandwidth from one level of cache to another.

With current perf framework, this patch adds support to perf event
for MBM.
Signed-off-by: NQiaowei Ren <qiaowei.ren@intel.com>

The label is used only for the error path, thus rename cleanup.

Do I really need to explain why?
Well, if read() is interrupted int the middle of reading, we will
never read the rest (even though it's highly unlikely as we are
reading just 8 bytes).
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

 * Sync stubbed functions with prototypes
 * Add missing ATTRIBUTE_UNUSED where needed

Pushing under the build breaker rule.

This patch implement a set of interfaces for perf event. Based on
these interfaces, we can implement internal driver API for perf,
and get the results of perf conuter you care about.
Signed-off-by: NQiaowei Ren <qiaowei.ren@intel.com>
Message-id: 1459171833-26416-4-git-send-email-qiaowei.ren@intel.com