Re: Clarification on support of arm32 with kernel4.1 for oprofile 1.2.0

On 02/15/2018 11:07 AM, Michael Petlan wrote:
> On Thu, 15 Feb 2018, Vinoth Natarajan wrote:
>> Hello Michael,
>> Thanks so much for the help.
>>
>> Cross compiled the hello.c with crosstools-arm-gcc-5.3-linux-4.1-glibc-2.22
>> and tested the executable on the target(arm32)
>>
>> Received the following error
>> " ./hello
>> Error opening leader 1"
>>
> 
> This points to the kernel perf subsystem. It cannot even open a simple event.
> Not an OProfile's fault.
> 
>>
>> Following are the kernel configs related to PERF in target
>>
>>  cat /tmp/test |grep PERF
>>
>> CONFIG_HAVE_PERF_EVENTS=y
>> CONFIG_PERF_USE_VMALLOC=y
>> CONFIG_PERF_EVENTS=y
>> # CONFIG_DEBUG_PERF_USE_VMALLOC is not set
>> CONFIG_HAVE_PERF_REGS=y
>> CONFIG_HAVE_PERF_USER_STACK_DUMP=y
>> # CONFIG_PCIEASPM_PERFORMANCE is not set
>> # CONFIG_HW_PERF_EVENTS is not set
> 
> I believe you need CONFIG_HW_PERF_EVENTS=y on ARM.
> 
>> # CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set
>> # CONFIG_CPU_FREQ_GOV_PERFORMANCE is not set
>> # CONFIG_CLS_U32_PERF is not set
>>
>> Could you please share your feedback on this.
> 
> You may also try to grep for perf in logs:
> 
> # dmesg | grep -i -e perf -e pmu
> [    1.375699] hw perfevents: enabled with armv8_pmuv3_0 PMU driver, 5 counters available
> [    1.403155] kvm [1]: GICV region size/alignment is unsafe, using trapping (reduced performance)
> [    2.036285] xgene-pmu APMC0D5C:00: X-Gene PMU version 2
> 
> ^^^ This is a log from the moment when my kernel identified and initialized
> the PMU. You should see something like that too. Or better, after recompiling
> your kernel with CONFIG_HW_PERF_EVENTS=y and booting, you should see something
> like this (PMU-related logs). If there is some error in PMU init, you will see
> it in the logs.
> 
> I am afraid that without CONFIG_HW_PERF_EVENTS=y you cannot use the events.
> 
> However, I don't guarantee that with CONFIG_HW_PERF_EVENTS=y it works. There
> can be some firmware issues, ACPI issues, etc. I roughly remember some issue
> when you'd need to boot with acpi=off to access the events, etc.
> 
> Cheers,
> Michael

Hi,

Another way that perf may fail to work is due to the device tree for the platform not describing the performance monitoring hardware:

https://bugzilla.redhat.com/show_bug.cgi?id=741325

There also may be some ARM processors that don't have performance monitoring hardware or that kernel perf support doesn't recognize.

What 32-bit ARM platform is the target?

-Will


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
oprofile-list mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/oprofile-list