From: Maynard J. <may...@us...> - 2013-07-24 20:00:10
|
We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. Thanks. -Maynard Johnson ----------------------------------------------------------------- Release Notes =============== OProfile provides powerful and flexible profiling for Linux. And now with the announcement of the 0.9.9 release, OProfile can also be used for simple event counting with the new 'ocount' tool. Read more at http://oprofile.sf.net. OProfile 0.9.9 has been released. OProfile is still in alpha, but has been proven stable for many users. Incompatibilities with previous release --------------------------------------- - None New features ------------ - The new 'ocount' program is introduced, for collecting raw event counts on a per-application, per-process, per-cpu, or system-wide basis. Unlike the profiling tools, post-processing of the data collected is not necessary -- the data is displayed in the output of ocount. A common use case for event counting tools is for computing the CPI (cycles per instruction) for an application. High CPI implies possible stalls, and many architectures provide events that give detailed information about the different types of stalls. This new feature requires a kernel version of 2.6.31 or greater. - New Processor Support * Intel Haswell * IBM zEnterprise EC12 (zEC12) * AMD Generic Performance Events * IBM Power ISA 2.07 Architected Events * IBM POWER8 Bug fixes --------- Filed bug reports: ------------------------------------------------------------------------- | BUG ID | Summary |-----------|------------------------------------------------------------ | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ | 219 | operf problems with PATH env variable | 220 | opjitconv fails with Floating point exception | 221 | opreport to find nonexistent symbol cause segfault | 222 | JIT profiling with oprofile fails when PID is > 32768 | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump | 225 | operf should record throttling/multiplexing | | NOTE: Multiplexing detection had to be backed out because | | the kernel provides no guaranteed method for detecting this. | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix | 233 | Default named unit mask with no extra parameter is broken | 234 | operf/opcontrol accept some invalid unit mask values | 235 | Unit mask bitmasks containing non-unique values should fail | 237 | operf problems with --pid and --system-wide options | 240 | parsed_filename.jit_dumpfile_exists represent wrong value | 241 | Flesh out user manual doc on oparchive/opimport commands ------------------------------------------------------------------------- Bug fixes and improvements without a filed report (e.g., posted to the list): --------------- - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails - operf: handle spaces in arguments correctly - Add support for ARC architecture to operf - Fix configure to handle architectures that do not implement perf_event_open - operf: Fix ppc64-specific code that handles _GRP## suffix in event names - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 - Fix operf default unit mask handling - Fix various compile issues with recent gcc - Fix opreport header info on unit mask when operf is run without a UM specified - operf: Fix 'Permission denied' error on early perf_events kernels - operf does not run opjitconv if --pid or --system-wide used - operf does not properly collect samples for children of already-running app - Update configure.ac to work with automake 1.13 - Change configure to distinguish between ppc32 and ppc64 architectures - Make convertPerfData procedure more robust - Fix operf seg fault due to event array size being too small; - Fix Ivybridge EP support - Performance improvement for operf's perf_event-to-oprofile format conversion - Fix broken --with-kernel configure option - oprofile pp tools should print messages about operf's lost samples - Fix opjitconv error message for bfd_set_arch_mach failure - Fix holes in operf system-wide profiling of forked processes - Fix Coverity issues identified against oprofile 0.9.8 release - Fix support for Intel Netburst (e.g., Pentium P4) with operf - Fix for oprof_start when daemonrc file does not exist - Print debug message when module summary count differs from total symbols counts - oprof_start: Don't show irrelevant bits in unit mask sub-window - Fix size problem of oprof_start GUI (overlapping sub-windows) - Various man page updates to match actual oprofile tools options Known problems and limitations ------------------------- - AMD Instruction Based Sampling (IBS) is not currently supported with the new operf program. Use the "legacy" opcontrol commands for IBS profiling. - When using operf to profile multiple events, the absolute number of events recorded will usually be substantially fewer than expected. This is due to a bug in the Linux kernel's Performance Events Subsystem that was fixed sometime between Linux kernel version 3.1 and 3.5. - If NMI watchdog is not disabled (x86), opcontrol may fail to allocate the hardware performance counter(s) it needs (bug #1468916). - Many Alpha ev67 events do not work (bug #931875). |
From: Maynard J. <may...@us...> - 2013-07-25 15:43:20
|
On 07/24/2013 02:59 PM, Maynard Johnson wrote: > We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: > https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ > > Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. A user running on IBM POWER7 reported this problem to me: ocount --system-wide -i 5:5 --events PM_GRP_CMPL Unable to convert all oprofile event values to perf_event values This is a bug in ocount (operf works OK for this event). I'll fix this in RC2. -Maynard > > Thanks. > -Maynard Johnson > > ----------------------------------------------------------------- > > Release Notes > =============== > OProfile provides powerful and flexible profiling for Linux. And now > with the announcement of the 0.9.9 release, OProfile can also > be used for simple event counting with the new 'ocount' tool. > > Read more at http://oprofile.sf.net. > > OProfile 0.9.9 has been released. OProfile is still in alpha, > but has been proven stable for many users. > > > Incompatibilities with previous release > --------------------------------------- > > - None > > > New features > ------------ > > - The new 'ocount' program is introduced, for collecting raw event > counts on a per-application, per-process, per-cpu, or system-wide basis. > Unlike the profiling tools, post-processing of the data collected is not > necessary -- the data is displayed in the output of ocount. A common use case > for event counting tools is for computing the CPI (cycles per instruction) for > an application. High CPI implies possible stalls, and many architectures > provide events that give detailed information about the different types of > stalls. This new feature requires a kernel version of 2.6.31 or greater. > > - New Processor Support > * Intel Haswell > * IBM zEnterprise EC12 (zEC12) > * AMD Generic Performance Events > * IBM Power ISA 2.07 Architected Events > * IBM POWER8 > > > Bug fixes > --------- > > Filed bug reports: > ------------------------------------------------------------------------- > | BUG ID | Summary > |-----------|------------------------------------------------------------ > | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ > | 219 | operf problems with PATH env variable > | 220 | opjitconv fails with Floating point exception > | 221 | opreport to find nonexistent symbol cause segfault > | 222 | JIT profiling with oprofile fails when PID is > 32768 > | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump > | 225 | operf should record throttling/multiplexing > | | NOTE: Multiplexing detection had to be backed out because > | | the kernel provides no guaranteed method for detecting this. > | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared > | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks > | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix > | 233 | Default named unit mask with no extra parameter is broken > | 234 | operf/opcontrol accept some invalid unit mask values > | 235 | Unit mask bitmasks containing non-unique values should fail > | 237 | operf problems with --pid and --system-wide options > | 240 | parsed_filename.jit_dumpfile_exists represent wrong value > | 241 | Flesh out user manual doc on oparchive/opimport commands > ------------------------------------------------------------------------- > > Bug fixes and improvements without a filed report (e.g., posted to the list): > --------------- > - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails > - operf: handle spaces in arguments correctly > - Add support for ARC architecture to operf > - Fix configure to handle architectures that do not implement perf_event_open > - operf: Fix ppc64-specific code that handles _GRP## suffix in event names > - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 > - Fix operf default unit mask handling > - Fix various compile issues with recent gcc > - Fix opreport header info on unit mask when operf is run without a UM specified > - operf: Fix 'Permission denied' error on early perf_events kernels > - operf does not run opjitconv if --pid or --system-wide used > - operf does not properly collect samples for children of already-running app > - Update configure.ac to work with automake 1.13 > - Change configure to distinguish between ppc32 and ppc64 architectures > - Make convertPerfData procedure more robust > - Fix operf seg fault due to event array size being too small; > - Fix Ivybridge EP support > - Performance improvement for operf's perf_event-to-oprofile format conversion > - Fix broken --with-kernel configure option > - oprofile pp tools should print messages about operf's lost samples > - Fix opjitconv error message for bfd_set_arch_mach failure > - Fix holes in operf system-wide profiling of forked processes > - Fix Coverity issues identified against oprofile 0.9.8 release > - Fix support for Intel Netburst (e.g., Pentium P4) with operf > - Fix for oprof_start when daemonrc file does not exist > - Print debug message when module summary count differs from total symbols counts > - oprof_start: Don't show irrelevant bits in unit mask sub-window > - Fix size problem of oprof_start GUI (overlapping sub-windows) > - Various man page updates to match actual oprofile tools options > > > > Known problems and limitations > ------------------------- > - AMD Instruction Based Sampling (IBS) is not currently supported with the new > operf program. Use the "legacy" opcontrol commands for IBS profiling. > > - When using operf to profile multiple events, the absolute number of events > recorded will usually be substantially fewer than expected. This is due > to a bug in the Linux kernel's Performance Events Subsystem that was fixed > sometime between Linux kernel version 3.1 and 3.5. > > - If NMI watchdog is not disabled (x86), opcontrol may fail to > allocate the hardware performance counter(s) it needs (bug #1468916). > > - Many Alpha ev67 events do not work (bug #931875). > > > ------------------------------------------------------------------------------ > See everything from the browser to the database with AppDynamics > Get end-to-end visibility with application monitoring from AppDynamics > Isolate bottlenecks and diagnose root cause in seconds. > Start your free trial of AppDynamics Pro today! > http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk > _______________________________________________ > oprofile-list mailing list > opr...@li... > https://lists.sourceforge.net/lists/listinfo/oprofile-list > |
From: William C. <wc...@re...> - 2013-07-25 17:05:51
|
On 07/24/2013 03:59 PM, Maynard Johnson wrote: > We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: > https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ > > Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. > > Thanks. > -Maynard Johnson Hi Maynard, I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. -Will > > ----------------------------------------------------------------- > > Release Notes > =============== > OProfile provides powerful and flexible profiling for Linux. And now > with the announcement of the 0.9.9 release, OProfile can also > be used for simple event counting with the new 'ocount' tool. > > Read more at http://oprofile.sf.net. > > OProfile 0.9.9 has been released. OProfile is still in alpha, > but has been proven stable for many users. > > > Incompatibilities with previous release > --------------------------------------- > > - None > > > New features > ------------ > > - The new 'ocount' program is introduced, for collecting raw event > counts on a per-application, per-process, per-cpu, or system-wide basis. > Unlike the profiling tools, post-processing of the data collected is not > necessary -- the data is displayed in the output of ocount. A common use case > for event counting tools is for computing the CPI (cycles per instruction) for > an application. High CPI implies possible stalls, and many architectures > provide events that give detailed information about the different types of > stalls. This new feature requires a kernel version of 2.6.31 or greater. > > - New Processor Support > * Intel Haswell > * IBM zEnterprise EC12 (zEC12) > * AMD Generic Performance Events > * IBM Power ISA 2.07 Architected Events > * IBM POWER8 > > > Bug fixes > --------- > > Filed bug reports: > ------------------------------------------------------------------------- > | BUG ID | Summary > |-----------|------------------------------------------------------------ > | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ > | 219 | operf problems with PATH env variable > | 220 | opjitconv fails with Floating point exception > | 221 | opreport to find nonexistent symbol cause segfault > | 222 | JIT profiling with oprofile fails when PID is > 32768 > | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump > | 225 | operf should record throttling/multiplexing > | | NOTE: Multiplexing detection had to be backed out because > | | the kernel provides no guaranteed method for detecting this. > | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared > | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks > | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix > | 233 | Default named unit mask with no extra parameter is broken > | 234 | operf/opcontrol accept some invalid unit mask values > | 235 | Unit mask bitmasks containing non-unique values should fail > | 237 | operf problems with --pid and --system-wide options > | 240 | parsed_filename.jit_dumpfile_exists represent wrong value > | 241 | Flesh out user manual doc on oparchive/opimport commands > ------------------------------------------------------------------------- > > Bug fixes and improvements without a filed report (e.g., posted to the list): > --------------- > - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails > - operf: handle spaces in arguments correctly > - Add support for ARC architecture to operf > - Fix configure to handle architectures that do not implement perf_event_open > - operf: Fix ppc64-specific code that handles _GRP## suffix in event names > - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 > - Fix operf default unit mask handling > - Fix various compile issues with recent gcc > - Fix opreport header info on unit mask when operf is run without a UM specified > - operf: Fix 'Permission denied' error on early perf_events kernels > - operf does not run opjitconv if --pid or --system-wide used > - operf does not properly collect samples for children of already-running app > - Update configure.ac to work with automake 1.13 > - Change configure to distinguish between ppc32 and ppc64 architectures > - Make convertPerfData procedure more robust > - Fix operf seg fault due to event array size being too small; > - Fix Ivybridge EP support > - Performance improvement for operf's perf_event-to-oprofile format conversion > - Fix broken --with-kernel configure option > - oprofile pp tools should print messages about operf's lost samples > - Fix opjitconv error message for bfd_set_arch_mach failure > - Fix holes in operf system-wide profiling of forked processes > - Fix Coverity issues identified against oprofile 0.9.8 release > - Fix support for Intel Netburst (e.g., Pentium P4) with operf > - Fix for oprof_start when daemonrc file does not exist > - Print debug message when module summary count differs from total symbols counts > - oprof_start: Don't show irrelevant bits in unit mask sub-window > - Fix size problem of oprof_start GUI (overlapping sub-windows) > - Various man page updates to match actual oprofile tools options > > > > Known problems and limitations > ------------------------- > - AMD Instruction Based Sampling (IBS) is not currently supported with the new > operf program. Use the "legacy" opcontrol commands for IBS profiling. > > - When using operf to profile multiple events, the absolute number of events > recorded will usually be substantially fewer than expected. This is due > to a bug in the Linux kernel's Performance Events Subsystem that was fixed > sometime between Linux kernel version 3.1 and 3.5. > > - If NMI watchdog is not disabled (x86), opcontrol may fail to > allocate the hardware performance counter(s) it needs (bug #1468916). > > - Many Alpha ev67 events do not work (bug #931875). > > > ------------------------------------------------------------------------------ > See everything from the browser to the database with AppDynamics > Get end-to-end visibility with application monitoring from AppDynamics > Isolate bottlenecks and diagnose root cause in seconds. > Start your free trial of AppDynamics Pro today! > http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk > _______________________________________________ > oprofile-list mailing list > opr...@li... > https://lists.sourceforge.net/lists/listinfo/oprofile-list > |
From: Maynard J. <may...@us...> - 2013-07-25 20:23:38
|
On 07/25/2013 12:05 PM, William Cohen wrote: > On 07/24/2013 03:59 PM, Maynard Johnson wrote: >> We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: >> https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ >> >> Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. >> >> Thanks. >> -Maynard Johnson > > Hi Maynard, > > I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: > > x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 > arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 > > I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. Will, thanks a bunch for trying out 0.9.9-rc1! > > Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. I presume Andi can lay hands on one, so I added him to the conversation. *Andi*, Will has made updates to the oprofile-tests to support Haswell. Can you please do the following steps on a Haswell box: 1. 'git clone git://git.code.sf.net/p/oprofile/oprofile-tests' 2. 'cd oprofile-tests/testsuite' 3. Switch to root user profile. 4. 'runtest --tool oprofile' It should end with output similar to the following: ------------------------------------------- Running ./oprofile-opcontrol/oprofile-opcontrol-run.exp ... Running ./oprofile-operf/oprofile-operf-run.exp ... Running ./oprofile-single_process/oprofile-single_process-run.exp ... === oprofile Summary === # of expected passes 89 kernel version: 2.6.32-358.11.1.el6.x86_64 module path: /lib/modules/2.6.32-358.11.1.el6.x86_64/kernel/arch/x86/oprofile/oprofile.ko userspace version: oprofile 0.9.9-rc1 compiled on Jul 25 2013 10:40:26 ------------------------------------------- The number of expected passes varies between architectures (and I think it may even vary between processor models). The important thing is that you should *not* see any indication of failures. Thanks! -Maynard > > > -Will >> >> ----------------------------------------------------------------- >> >> Release Notes >> =============== >> OProfile provides powerful and flexible profiling for Linux. And now >> with the announcement of the 0.9.9 release, OProfile can also >> be used for simple event counting with the new 'ocount' tool. >> >> Read more at http://oprofile.sf.net. >> >> OProfile 0.9.9 has been released. OProfile is still in alpha, >> but has been proven stable for many users. >> >> >> Incompatibilities with previous release >> --------------------------------------- >> >> - None >> >> >> New features >> ------------ >> >> - The new 'ocount' program is introduced, for collecting raw event >> counts on a per-application, per-process, per-cpu, or system-wide basis. >> Unlike the profiling tools, post-processing of the data collected is not >> necessary -- the data is displayed in the output of ocount. A common use case >> for event counting tools is for computing the CPI (cycles per instruction) for >> an application. High CPI implies possible stalls, and many architectures >> provide events that give detailed information about the different types of >> stalls. This new feature requires a kernel version of 2.6.31 or greater. >> >> - New Processor Support >> * Intel Haswell >> * IBM zEnterprise EC12 (zEC12) >> * AMD Generic Performance Events >> * IBM Power ISA 2.07 Architected Events >> * IBM POWER8 >> >> >> Bug fixes >> --------- >> >> Filed bug reports: >> ------------------------------------------------------------------------- >> | BUG ID | Summary >> |-----------|------------------------------------------------------------ >> | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ >> | 219 | operf problems with PATH env variable >> | 220 | opjitconv fails with Floating point exception >> | 221 | opreport to find nonexistent symbol cause segfault >> | 222 | JIT profiling with oprofile fails when PID is > 32768 >> | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump >> | 225 | operf should record throttling/multiplexing >> | | NOTE: Multiplexing detection had to be backed out because >> | | the kernel provides no guaranteed method for detecting this. >> | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared >> | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks >> | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix >> | 233 | Default named unit mask with no extra parameter is broken >> | 234 | operf/opcontrol accept some invalid unit mask values >> | 235 | Unit mask bitmasks containing non-unique values should fail >> | 237 | operf problems with --pid and --system-wide options >> | 240 | parsed_filename.jit_dumpfile_exists represent wrong value >> | 241 | Flesh out user manual doc on oparchive/opimport commands >> ------------------------------------------------------------------------- >> >> Bug fixes and improvements without a filed report (e.g., posted to the list): >> --------------- >> - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails >> - operf: handle spaces in arguments correctly >> - Add support for ARC architecture to operf >> - Fix configure to handle architectures that do not implement perf_event_open >> - operf: Fix ppc64-specific code that handles _GRP## suffix in event names >> - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 >> - Fix operf default unit mask handling >> - Fix various compile issues with recent gcc >> - Fix opreport header info on unit mask when operf is run without a UM specified >> - operf: Fix 'Permission denied' error on early perf_events kernels >> - operf does not run opjitconv if --pid or --system-wide used >> - operf does not properly collect samples for children of already-running app >> - Update configure.ac to work with automake 1.13 >> - Change configure to distinguish between ppc32 and ppc64 architectures >> - Make convertPerfData procedure more robust >> - Fix operf seg fault due to event array size being too small; >> - Fix Ivybridge EP support >> - Performance improvement for operf's perf_event-to-oprofile format conversion >> - Fix broken --with-kernel configure option >> - oprofile pp tools should print messages about operf's lost samples >> - Fix opjitconv error message for bfd_set_arch_mach failure >> - Fix holes in operf system-wide profiling of forked processes >> - Fix Coverity issues identified against oprofile 0.9.8 release >> - Fix support for Intel Netburst (e.g., Pentium P4) with operf >> - Fix for oprof_start when daemonrc file does not exist >> - Print debug message when module summary count differs from total symbols counts >> - oprof_start: Don't show irrelevant bits in unit mask sub-window >> - Fix size problem of oprof_start GUI (overlapping sub-windows) >> - Various man page updates to match actual oprofile tools options >> >> >> >> Known problems and limitations >> ------------------------- >> - AMD Instruction Based Sampling (IBS) is not currently supported with the new >> operf program. Use the "legacy" opcontrol commands for IBS profiling. >> >> - When using operf to profile multiple events, the absolute number of events >> recorded will usually be substantially fewer than expected. This is due >> to a bug in the Linux kernel's Performance Events Subsystem that was fixed >> sometime between Linux kernel version 3.1 and 3.5. >> >> - If NMI watchdog is not disabled (x86), opcontrol may fail to >> allocate the hardware performance counter(s) it needs (bug #1468916). >> >> - Many Alpha ev67 events do not work (bug #931875). >> >> >> ------------------------------------------------------------------------------ >> See everything from the browser to the database with AppDynamics >> Get end-to-end visibility with application monitoring from AppDynamics >> Isolate bottlenecks and diagnose root cause in seconds. >> Start your free trial of AppDynamics Pro today! >> http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk >> _______________________________________________ >> oprofile-list mailing list >> opr...@li... >> https://lists.sourceforge.net/lists/listinfo/oprofile-list >> > |
From: Andi K. <an...@fi...> - 2013-07-25 20:39:14
|
On Thu, Jul 25, 2013 at 03:13:11PM -0500, Maynard Johnson wrote: > On 07/25/2013 12:05 PM, William Cohen wrote: > > On 07/24/2013 03:59 PM, Maynard Johnson wrote: > >> We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: > >> https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ > >> > >> Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. > >> > >> Thanks. > >> -Maynard Johnson > > > > Hi Maynard, > > > > I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: > > > > x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 > > arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 > > > > I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. > Will, thanks a bunch for trying out 0.9.9-rc1! > > > > Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. > I presume Andi can lay hands on one, so I added him to the conversation. > > *Andi*, Will has made updates to the oprofile-tests to support Haswell. Can you please do the following steps on a Haswell box: I get lots of failures, but not sure what the cause is. This is with a fairly old distribution (so may be something else). Will investigate later. # runtest --tool oprofile WARNING: Couldn't find the global config file. WARNING: Couldn't find tool init file Test Run By ak on Thu Jul 25 07:18:35 2013 Native configuration is x86_64-unknown-linux-gnu === oprofile tests === Schedule of variations: unix Running target unix Using /usr/share/dejagnu/baseboards/unix.exp as board description file for target. Using /usr/share/dejagnu/config/unix.exp as generic interface file for target. Using ./config/unix.exp as tool-and-target-specific interface file. Running ./oprofile-opcontrol/oprofile-opcontrol-run.exp ... FAIL: opcontrol --init FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles op_daemon_check_samples_exist, op_daemon not running as expected FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles op_daemon_check_samples_exist, op_daemon not running as expected FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles FAIL: opcontrol --setup --no-vmlinux --event=uops_issued:2000000:stall_cycles --event=uops_retired:2000000:stall_cycles op_daemon_check_samples_exist, op_daemon not running as expected Running ./oprofile-operf/oprofile-operf-run.exp ... ERROR: tcl error sourcing ./oprofile-operf/oprofile-operf-run.exp. ERROR: kill: usage: kill [-s sigspec | -n signum | -sigspec] pid | jobspec ... or kill -l [sigspec] while executing "exec /tmp/operf_stop.zzzz" (procedure "do_test" line 78) invoked from within "do_test $spec $symbol_check" (procedure "oprofile-operf_run_tests" line 15) invoked from within > > 1. 'git clone git://git.code.sf.net/p/oprofile/oprofile-tests' > 2. 'cd oprofile-tests/testsuite' > 3. Switch to root user profile. > 4. 'runtest --tool oprofile' > > It should end with output similar to the following: > > ------------------------------------------- > Running ./oprofile-opcontrol/oprofile-opcontrol-run.exp ... > Running ./oprofile-operf/oprofile-operf-run.exp ... > Running ./oprofile-single_process/oprofile-single_process-run.exp ... > > === oprofile Summary === > > # of expected passes 89 > > kernel version: 2.6.32-358.11.1.el6.x86_64 > module path: /lib/modules/2.6.32-358.11.1.el6.x86_64/kernel/arch/x86/oprofile/oprofile.ko > userspace version: oprofile 0.9.9-rc1 compiled on Jul 25 2013 10:40:26 > > ------------------------------------------- > > > The number of expected passes varies between architectures (and I think it may even vary between processor models). The important thing is that you should *not* see any indication of failures. > > > Thanks! > > -Maynard > > > > > > -Will > >> > >> ----------------------------------------------------------------- > >> > >> Release Notes > >> =============== > >> OProfile provides powerful and flexible profiling for Linux. And now > >> with the announcement of the 0.9.9 release, OProfile can also > >> be used for simple event counting with the new 'ocount' tool. > >> > >> Read more at http://oprofile.sf.net. > >> > >> OProfile 0.9.9 has been released. OProfile is still in alpha, > >> but has been proven stable for many users. > >> > >> > >> Incompatibilities with previous release > >> --------------------------------------- > >> > >> - None > >> > >> > >> New features > >> ------------ > >> > >> - The new 'ocount' program is introduced, for collecting raw event > >> counts on a per-application, per-process, per-cpu, or system-wide basis. > >> Unlike the profiling tools, post-processing of the data collected is not > >> necessary -- the data is displayed in the output of ocount. A common use case > >> for event counting tools is for computing the CPI (cycles per instruction) for > >> an application. High CPI implies possible stalls, and many architectures > >> provide events that give detailed information about the different types of > >> stalls. This new feature requires a kernel version of 2.6.31 or greater. > >> > >> - New Processor Support > >> * Intel Haswell > >> * IBM zEnterprise EC12 (zEC12) > >> * AMD Generic Performance Events > >> * IBM Power ISA 2.07 Architected Events > >> * IBM POWER8 > >> > >> > >> Bug fixes > >> --------- > >> > >> Filed bug reports: > >> ------------------------------------------------------------------------- > >> | BUG ID | Summary > >> |-----------|------------------------------------------------------------ > >> | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ > >> | 219 | operf problems with PATH env variable > >> | 220 | opjitconv fails with Floating point exception > >> | 221 | opreport to find nonexistent symbol cause segfault > >> | 222 | JIT profiling with oprofile fails when PID is > 32768 > >> | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump > >> | 225 | operf should record throttling/multiplexing > >> | | NOTE: Multiplexing detection had to be backed out because > >> | | the kernel provides no guaranteed method for detecting this. > >> | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared > >> | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks > >> | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix > >> | 233 | Default named unit mask with no extra parameter is broken > >> | 234 | operf/opcontrol accept some invalid unit mask values > >> | 235 | Unit mask bitmasks containing non-unique values should fail > >> | 237 | operf problems with --pid and --system-wide options > >> | 240 | parsed_filename.jit_dumpfile_exists represent wrong value > >> | 241 | Flesh out user manual doc on oparchive/opimport commands > >> ------------------------------------------------------------------------- > >> > >> Bug fixes and improvements without a filed report (e.g., posted to the list): > >> --------------- > >> - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails > >> - operf: handle spaces in arguments correctly > >> - Add support for ARC architecture to operf > >> - Fix configure to handle architectures that do not implement perf_event_open > >> - operf: Fix ppc64-specific code that handles _GRP## suffix in event names > >> - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 > >> - Fix operf default unit mask handling > >> - Fix various compile issues with recent gcc > >> - Fix opreport header info on unit mask when operf is run without a UM specified > >> - operf: Fix 'Permission denied' error on early perf_events kernels > >> - operf does not run opjitconv if --pid or --system-wide used > >> - operf does not properly collect samples for children of already-running app > >> - Update configure.ac to work with automake 1.13 > >> - Change configure to distinguish between ppc32 and ppc64 architectures > >> - Make convertPerfData procedure more robust > >> - Fix operf seg fault due to event array size being too small; > >> - Fix Ivybridge EP support > >> - Performance improvement for operf's perf_event-to-oprofile format conversion > >> - Fix broken --with-kernel configure option > >> - oprofile pp tools should print messages about operf's lost samples > >> - Fix opjitconv error message for bfd_set_arch_mach failure > >> - Fix holes in operf system-wide profiling of forked processes > >> - Fix Coverity issues identified against oprofile 0.9.8 release > >> - Fix support for Intel Netburst (e.g., Pentium P4) with operf > >> - Fix for oprof_start when daemonrc file does not exist > >> - Print debug message when module summary count differs from total symbols counts > >> - oprof_start: Don't show irrelevant bits in unit mask sub-window > >> - Fix size problem of oprof_start GUI (overlapping sub-windows) > >> - Various man page updates to match actual oprofile tools options > >> > >> > >> > >> Known problems and limitations > >> ------------------------- > >> - AMD Instruction Based Sampling (IBS) is not currently supported with the new > >> operf program. Use the "legacy" opcontrol commands for IBS profiling. > >> > >> - When using operf to profile multiple events, the absolute number of events > >> recorded will usually be substantially fewer than expected. This is due > >> to a bug in the Linux kernel's Performance Events Subsystem that was fixed > >> sometime between Linux kernel version 3.1 and 3.5. > >> > >> - If NMI watchdog is not disabled (x86), opcontrol may fail to > >> allocate the hardware performance counter(s) it needs (bug #1468916). > >> > >> - Many Alpha ev67 events do not work (bug #931875). > >> > >> > >> ------------------------------------------------------------------------------ > >> See everything from the browser to the database with AppDynamics > >> Get end-to-end visibility with application monitoring from AppDynamics > >> Isolate bottlenecks and diagnose root cause in seconds. > >> Start your free trial of AppDynamics Pro today! > >> http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk > >> _______________________________________________ > >> oprofile-list mailing list > >> opr...@li... > >> https://lists.sourceforge.net/lists/listinfo/oprofile-list > >> > > > -- ak...@li... -- Speaking for myself only. |
From: William C. <wc...@re...> - 2013-07-25 21:29:53
|
On 07/25/2013 04:39 PM, Andi Kleen wrote: > On Thu, Jul 25, 2013 at 03:13:11PM -0500, Maynard Johnson wrote: >> On 07/25/2013 12:05 PM, William Cohen wrote: >>> On 07/24/2013 03:59 PM, Maynard Johnson wrote: >>>> We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: >>>> https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ >>>> >>>> Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. >>>> >>>> Thanks. >>>> -Maynard Johnson >>> >>> Hi Maynard, >>> >>> I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: >>> >>> x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 >>> arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 >>> >>> I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. >> Will, thanks a bunch for trying out 0.9.9-rc1! >>> >>> Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. >> I presume Andi can lay hands on one, so I added him to the conversation. >> >> *Andi*, Will has made updates to the oprofile-tests to support Haswell. Can you please do the following steps on a Haswell box: > > I get lots of failures, but not sure what the cause is. This is with a > fairly old distribution (so may be something else). Will investigate later. This is maybe due to the newer kernel having the nmi watchdog enabled. Could you try doing the following as root then rerunning the tests: opcontrol --deinit echo 0 > /proc/sys/kernel/nmi_watchdog -Will |
From: Andi K. <an...@fi...> - 2013-07-25 21:31:55
Attachments:
fix-hsw
|
Ok so I investigated the failures now. The oprofile --init failure was spurious. On rerun it doesn't happen. The other failures are all fixed by the attached patch for the test suite. The newer files use prime number minimums, so it had to be adjusted. So I think it's good to go. -Andi |
From: William C. <wc...@re...> - 2013-07-25 21:42:15
|
On 07/25/2013 05:31 PM, Andi Kleen wrote: > > Ok so I investigated the failures now. > > The oprofile --init failure was spurious. On rerun it doesn't happen. > > The other failures are all fixed by the attached patch for the test > suite. The newer files use prime number minimums, so it had to be > adjusted. > > So I think it's good to go. > > -Andi > Pushed the patch into the testsuite. -Will |
From: William C. <wc...@re...> - 2013-07-25 21:53:48
|
On 07/25/2013 01:05 PM, William Cohen wrote: > On 07/24/2013 03:59 PM, Maynard Johnson wrote: >> We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: >> https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ >> >> Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. >> >> Thanks. >> -Maynard Johnson > > Hi Maynard, > > I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: > > x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 > arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 > > I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. > > Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. > > > -Will The tests seemed to run fine on Intel westmere, ivybridge, and amd family_10h. When running the tests on a ARM v7 cortex a9 I found that the INST_RETIRED doesn't work Looking through the ARM documentation I found that particular event is disabled for the ARM v7 cortex a9. The linux kernel uses another event to approximate INST_RETIRED. For the ARM cortex a15 I am currently stuck using the prebuilt chromebook kernel which doesn't have oprofile enabled, so I can't run the opcontrol portion of the tests. However, the oprofile-operf tests seemed to run fine. -Will |
From: William C. <wc...@re...> - 2013-07-29 13:12:54
|
On 07/25/2013 01:05 PM, William Cohen wrote: > On 07/24/2013 03:59 PM, Maynard Johnson wrote: >> We are pleased to announce OProfile 0.9.9 Release Candidate 1. You can download this release at: >> https://sourceforge.net/projects/oprofile/files/oprofile/oprofile-0.9.9-rc1/ >> >> Please download and test this release candidate, and send your feedback by replying to this message. Please include your hardware platform and Linux distribution information in your reply. >> >> Thanks. >> -Maynard Johnson > > Hi Maynard, > > I have built a couple scratch fedora 19 rpms to test the oprofile-0.9.9-rc1: > > x86: http://koji.fedoraproject.org/koji/taskinfo?taskID=5653386 > arm: http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=2017231 > > I have been trying the rpms out on a number of different machines (amd opteron, intel westmere, intel ivybridge, arm cortex a15, and arm cortex a9. I have found that some updates are needed for the testsuite code to handle newer processors such as the intel ivybridge and ARM cortex A-5, A-7, and A-15 processors. > > Has someone tested the Intel haswell support? There doesn't seem to be any free/available Intel Haswell machines at Red Hat. > > > -Will OProfile 0.9.9-rc1 appeared to work in the machines I tried Intel ivy bridge, Intel Westemere, arm cortex a9. Anything else needed for the release of oprofile? For the ARM cortex a9 the INST_RETIRED event (included from armv7-common) doesn't work, but that is not a regression. -the other Will >> >> ----------------------------------------------------------------- >> >> Release Notes >> =============== >> OProfile provides powerful and flexible profiling for Linux. And now >> with the announcement of the 0.9.9 release, OProfile can also >> be used for simple event counting with the new 'ocount' tool. >> >> Read more at http://oprofile.sf.net. >> >> OProfile 0.9.9 has been released. OProfile is still in alpha, >> but has been proven stable for many users. >> >> >> Incompatibilities with previous release >> --------------------------------------- >> >> - None >> >> >> New features >> ------------ >> >> - The new 'ocount' program is introduced, for collecting raw event >> counts on a per-application, per-process, per-cpu, or system-wide basis. >> Unlike the profiling tools, post-processing of the data collected is not >> necessary -- the data is displayed in the output of ocount. A common use case >> for event counting tools is for computing the CPI (cycles per instruction) for >> an application. High CPI implies possible stalls, and many architectures >> provide events that give detailed information about the different types of >> stalls. This new feature requires a kernel version of 2.6.31 or greater. >> >> - New Processor Support >> * Intel Haswell >> * IBM zEnterprise EC12 (zEC12) >> * AMD Generic Performance Events >> * IBM Power ISA 2.07 Architected Events >> * IBM POWER8 >> >> >> Bug fixes >> --------- >> >> Filed bug reports: >> ------------------------------------------------------------------------- >> | BUG ID | Summary >> |-----------|------------------------------------------------------------ >> | 193 | oprofile leaves a world-writable /var/lib/oprofile/jitdump/ >> | 219 | operf problems with PATH env variable >> | 220 | opjitconv fails with Floating point exception >> | 221 | opreport to find nonexistent symbol cause segfault >> | 222 | JIT profiling with oprofile fails when PID is > 32768 >> | 223 | jvmti agent reports missing /var/lib/oprofile/jitdump >> | 225 | operf should record throttling/multiplexing >> | | NOTE: Multiplexing detection had to be backed out because >> | | the kernel provides no guaranteed method for detecting this. >> | 226 | operf build fails: PERF_RECORD_MISC_GUEST_* not declared >> | 228 | On Fedora18, oprofile's CRC method to find debuginfo breaks >> | 229 | operf/ppc64: Allow use of event names without _GRP<n> suffix >> | 233 | Default named unit mask with no extra parameter is broken >> | 234 | operf/opcontrol accept some invalid unit mask values >> | 235 | Unit mask bitmasks containing non-unique values should fail >> | 237 | operf problems with --pid and --system-wide options >> | 240 | parsed_filename.jit_dumpfile_exists represent wrong value >> | 241 | Flesh out user manual doc on oparchive/opimport commands >> ------------------------------------------------------------------------- >> >> Bug fixes and improvements without a filed report (e.g., posted to the list): >> --------------- >> - OProfile build fails for 32-bit ppc; operf_utils.cpp compile fails >> - operf: handle spaces in arguments correctly >> - Add support for ARC architecture to operf >> - Fix configure to handle architectures that do not implement perf_event_open >> - operf: Fix ppc64-specific code that handles _GRP## suffix in event names >> - Change configure to look for libpfm4 first (for ppc64); fall back to libpfm3 >> - Fix operf default unit mask handling >> - Fix various compile issues with recent gcc >> - Fix opreport header info on unit mask when operf is run without a UM specified >> - operf: Fix 'Permission denied' error on early perf_events kernels >> - operf does not run opjitconv if --pid or --system-wide used >> - operf does not properly collect samples for children of already-running app >> - Update configure.ac to work with automake 1.13 >> - Change configure to distinguish between ppc32 and ppc64 architectures >> - Make convertPerfData procedure more robust >> - Fix operf seg fault due to event array size being too small; >> - Fix Ivybridge EP support >> - Performance improvement for operf's perf_event-to-oprofile format conversion >> - Fix broken --with-kernel configure option >> - oprofile pp tools should print messages about operf's lost samples >> - Fix opjitconv error message for bfd_set_arch_mach failure >> - Fix holes in operf system-wide profiling of forked processes >> - Fix Coverity issues identified against oprofile 0.9.8 release >> - Fix support for Intel Netburst (e.g., Pentium P4) with operf >> - Fix for oprof_start when daemonrc file does not exist >> - Print debug message when module summary count differs from total symbols counts >> - oprof_start: Don't show irrelevant bits in unit mask sub-window >> - Fix size problem of oprof_start GUI (overlapping sub-windows) >> - Various man page updates to match actual oprofile tools options >> >> >> >> Known problems and limitations >> ------------------------- >> - AMD Instruction Based Sampling (IBS) is not currently supported with the new >> operf program. Use the "legacy" opcontrol commands for IBS profiling. >> >> - When using operf to profile multiple events, the absolute number of events >> recorded will usually be substantially fewer than expected. This is due >> to a bug in the Linux kernel's Performance Events Subsystem that was fixed >> sometime between Linux kernel version 3.1 and 3.5. >> >> - If NMI watchdog is not disabled (x86), opcontrol may fail to >> allocate the hardware performance counter(s) it needs (bug #1468916). >> >> - Many Alpha ev67 events do not work (bug #931875). >> >> >> ------------------------------------------------------------------------------ >> See everything from the browser to the database with AppDynamics >> Get end-to-end visibility with application monitoring from AppDynamics >> Isolate bottlenecks and diagnose root cause in seconds. >> Start your free trial of AppDynamics Pro today! >> http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk >> _______________________________________________ >> oprofile-list mailing list >> opr...@li... >> https://lists.sourceforge.net/lists/listinfo/oprofile-list >> > > > ------------------------------------------------------------------------------ > See everything from the browser to the database with AppDynamics > Get end-to-end visibility with application monitoring from AppDynamics > Isolate bottlenecks and diagnose root cause in seconds. > Start your free trial of AppDynamics Pro today! > http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk > _______________________________________________ > oprofile-list mailing list > opr...@li... > https://lists.sourceforge.net/lists/listinfo/oprofile-list > |