How to measure VM events using perf-event

How to measure VM events using perf-event

[Jueyuan Zhu]

Hello,

I am trying to use the perf-event to measure the events for one specified VM in KVM platform. Since each VM is just a process in the host OS, so I thought we could just set the process id of this VM in the perf-event_open API to count the events. The following is my code (based on the example from perf_event_open linux manual). 

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>

int main(int argc, char **argv)
{
    struct perf_event_attr pe;
    long long count;
    int fd;
    int pid = VM_id; // the process id of the VM.

    memset(&pe, 0, sizeof(struct perf_event_attr));
    pe.size = sizeof(struct perf_event_attr);
    pe.type = PERF_TYPE_HARDWARE;
    pe.config = PERF_COUNT_HW_INSTRUCTIONS;
    pe.disabled = 1;
    pe.exclude_kernel = 0;
    pe.exclude_hv = 0;
    pe.exclude_host = 0;
    pe.exclude_guest = 0;

    fd = syscall(__NR_perf_event_open, &pe, pid, -1, -1, 0);
    if (fd == -1) {
       fprintf(stderr, "Error opening leader %llx\n", pe.config);
       exit(EXIT_FAILURE);
    }

    ioctl(fd, PERF_EVENT_IOC_RESET, 0);
    ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);

    sleep(1);

    ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
    read(fd, &count, sizeof(long long));

    printf("# of instructions:  %lld\n", count);

    close(fd);
}

However, this program’s output keeps constant whenever this VM is idle, or running different benchmarks. So it seems the event counting is incorrect. Can anyone tell me how to set the flags or perf_event_attr to correctly count the VM events using perf_event_open? 

Thank you very much in advance!!

Best regards,

Jueyuan--
To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: How to measure VM events using perf-event

[Christopher Covington]

Hi Jueyuan,

On 04/09/2015 11:46 PM, Jueyuan Zhu wrote:
> Hello,
> 
> I am trying to use the perf-event to measure the events for one specified
> VM in KVM platform. Since each VM is just a process in the host OS, so I thought
> we could just set the process id of this VM in the perf-event_open API to
> count the events. The following is my code (based on the example from
> perf_event_open linux manual). 

> However, this program’s output keeps constant whenever this VM is idle, or
> running different benchmarks. So it seems the event counting is incorrect. Can
> anyone tell me how to set the flags or perf_event_attr to correctly count the
> VM events using perf_event_open? 

You may want to check that the QEMU or kvmtool you're using correctly
virtualizes or emulates the PMU for your architecture. Information about this
might be in the documentation or release notes or if not you could try asking
on the QEMU or kvmtool mailing list tool.

Chris

-- 
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: How to measure VM events using perf-event

[Jueyuan Zhu]

Hi Chris,

Thanks for your suggestions. Since I am using the perf_event_open in the host OS, not in the guest OS. So does it need the support of PMU virtualization? I used the perf command below to measure the VM, and it can give correct results. So I am wondering how to use perf_event_open to get the same results as the perf user command?

#perf stat -e instructions -p VM_id sleep 1

Thank you!

-Jueyuan

On Apr 10, 2015, at 11:39 AM, Christopher Covington <cov@codeaurora.org> wrote:

> Hi Jueyuan,
> 
> On 04/09/2015 11:46 PM, Jueyuan Zhu wrote:
>> Hello,
>> 
>> I am trying to use the perf-event to measure the events for one specified
>> VM in KVM platform. Since each VM is just a process in the host OS, so I thought
>> we could just set the process id of this VM in the perf-event_open API to
>> count the events. The following is my code (based on the example from
>> perf_event_open linux manual). 
> 
>> However, this program’s output keeps constant whenever this VM is idle, or
>> running different benchmarks. So it seems the event counting is incorrect. Can
>> anyone tell me how to set the flags or perf_event_attr to correctly count the
>> VM events using perf_event_open? 
> 
> You may want to check that the QEMU or kvmtool you're using correctly
> virtualizes or emulates the PMU for your architecture. Information about this
> might be in the documentation or release notes or if not you could try asking
> on the QEMU or kvmtool mailing list tool.
> 
> Chris
> 
> -- 
> Qualcomm Innovation Center, Inc.
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
> a Linux Foundation Collaborative Project

Re: How to measure VM events using perf-event

[Christopher Covington]

Hi Jueyuan,

On 04/10/2015 12:17 PM, Jueyuan Zhu wrote:
> Hi Chris,
> 
> Thanks for your suggestions. Since I am using the perf_event_open in the
> host OS, not in the guest OS. So does it need the support of PMU
> virtualization? I used the perf command below to measure the VM, and it can
> give correct results. So I am wondering how to use perf_event_open to get the
> same results as the perf user command?
> 
> #perf stat -e instructions -p VM_id sleep 1

Sorry, I overlooked that. I like to set pinned and inherit, but nothing jumps
out to me as wrong. If you'd like to look at more examples of using
perf_event_open, the test suite has the largest collection I know of.

https://github.com/deater/perf_event_tests

Chris

> On Apr 10, 2015, at 11:39 AM, Christopher Covington <cov@codeaurora.org> wrote:
> 
>> Hi Jueyuan,
>>
>> On 04/09/2015 11:46 PM, Jueyuan Zhu wrote:
>>> Hello,
>>>
>>> I am trying to use the perf-event to measure the events for one specified
>>> VM in KVM platform. Since each VM is just a process in the host OS, so I thought
>>> we could just set the process id of this VM in the perf-event_open API to
>>> count the events. The following is my code (based on the example from
>>> perf_event_open linux manual). 
>>
>>> However, this program’s output keeps constant whenever this VM is idle, or
>>> running different benchmarks. So it seems the event counting is incorrect. Can
>>> anyone tell me how to set the flags or perf_event_attr to correctly count the
>>> VM events using perf_event_open? 
>>
>> You may want to check that the QEMU or kvmtool you're using correctly
>> virtualizes or emulates the PMU for your architecture. Information about this
>> might be in the documentation or release notes or if not you could try asking
>> on the QEMU or kvmtool mailing list tool.
>>
>> Chris
>>
>> -- 
>> Qualcomm Innovation Center, Inc.
>> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
>> a Linux Foundation Collaborative Project
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


-- 
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: How to measure VM events using perf-event

[Tianwei Zhang]

In the KVM-based machine, each VM is a process, with several threads running concurrently. You can use ps -eLf to display all the VM threads. When measuring the performance of the VM, you should also specify the thread ids in the perf_event_open function. 

Tianwei

On Apr 10, 2015, at 12:17 PM, Jueyuan Zhu <jueyuan.zhu@gmail.com> wrote:

> Hi Chris,
> 
> Thanks for your suggestions. Since I am using the perf_event_open in the host OS, not in the guest OS. So does it need the support of PMU virtualization? I used the perf command below to measure the VM, and it can give correct results. So I am wondering how to use perf_event_open to get the same results as the perf user command?
> 
> #perf stat -e instructions -p VM_id sleep 1
> 
> Thank you!
> 
> -Jueyuan
> 
> On Apr 10, 2015, at 11:39 AM, Christopher Covington <cov@codeaurora.org> wrote:
> 
>> Hi Jueyuan,
>> 
>> On 04/09/2015 11:46 PM, Jueyuan Zhu wrote:
>>> Hello,
>>> 
>>> I am trying to use the perf-event to measure the events for one specified
>>> VM in KVM platform. Since each VM is just a process in the host OS, so I thought
>>> we could just set the process id of this VM in the perf-event_open API to
>>> count the events. The following is my code (based on the example from
>>> perf_event_open linux manual). 
>> 
>>> However, this program’s output keeps constant whenever this VM is idle, or
>>> running different benchmarks. So it seems the event counting is incorrect. Can
>>> anyone tell me how to set the flags or perf_event_attr to correctly count the
>>> VM events using perf_event_open? 
>> 
>> You may want to check that the QEMU or kvmtool you're using correctly
>> virtualizes or emulates the PMU for your architecture. Information about this
>> might be in the documentation or release notes or if not you could try asking
>> on the QEMU or kvmtool mailing list tool.
>> 
>> Chris
>> 
>> -- 
>> Qualcomm Innovation Center, Inc.
>> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
>> a Linux Foundation Collaborative Project
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: How to measure VM events using perf-event

[Arnaldo Carvalho de Melo]

Em Fri, Apr 10, 2015 at 12:17:08PM -0400, Jueyuan Zhu escreveu:
> Hi Chris,
> 
> Thanks for your suggestions. Since I am using the perf_event_open in the host OS, not in the guest OS. So does it need the support of PMU virtualization? I used the perf command below to measure the VM, and it can give correct results. So I am wondering how to use perf_event_open to get the same results as the perf user command?
> 
> #perf stat -e instructions -p VM_id sleep 1
 
Try adding -vv to the above perf stat command, it will show you how it
is setting up perf_event_attr, as well as the other arguments to
sys_perf_event_open, for example:

[acme@zoo linux]$ perf stat -vv -e instructions usleep 1
------------------------------------------------------------
perf_event_attr:
  size                             112
  config                           1
  read_format                      TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
  disabled                         1
  inherit                          1
  enable_on_exec                   1
  exclude_guest                    1
------------------------------------------------------------
sys_perf_event_open: pid 3530  cpu -1  group_fd -1  flags 0x8
instructions: 632294 767399 767399

 Performance counter stats for 'usleep 1':

           632.294      instructions                                                

       0,001521851 seconds time elapsed

[acme@zoo linux]$

- Arnaldo