Re: [Xen-devel] Performance Monitor and Profiling tools for ARM64

[Julien Grall <julien.grall@xxxxxxx>]

On 4/3/19 7:57 AM, Diego Alejandro Parra Guzman wrote:
>       Hello Julien Grall
Hi Diego,

> thank you for your replied,  you have clarified many of my doubts.
>
> I have been implementing some basic functionality for the PMU :
>
> Here small description:
>
>   * getpmuinfo_attr :                             /* @brief  get PMU
>     info, including number of Counters, and control counters */
>   * enable_all_pmu_counters:              /* @brief  Enable the cycle
>     counter register PMCCNTR_EL0, and any implemented event counter
>     PMEVCNTR<n>_EL0 */
>   * disable_all_pmu_counters /* @brief  Disable the cycle counter
>     register PMCCNTR_EL0, and any implemented event counter
>     PMEVCNTR<n>_EL0. */
>   * reset_all_pmu_counters                  /* @brief  Reset the cycle
>     counter register PMCCNTR_EL0,  and any implemented event counter
>     PMEVCNTR<n>_EL0.*/
>   * enable_user_pmu_access              /* @brief  Enable user-mode
>     access to counters.*/ 
>
> I think these can be used as primitives for XEN:
>
> Now I have some additional questions about XEN:
>
>   * how can I include that primitives as a driver for ARM64 in XEN?
The inclusion in Xen will depend on your use cases. In my previous 
e-mail, I described 2 uses cases:
   1) A guest profiling itself
   2) The hypervisor profiling the system (i.e guests and itself)

Which one are you implementing?

> I've noticed that in linux kernel 4.14 - 4.15  $SOURCE/drivers/perf :
>
>   * there is and implementation for arm_pmu  which is far better than
>     what I've implemented, is it hard to import that driver to XEN ?
arm_pmu is based on the Linux Perf subsystem. So I don't think you can 
re-used the driver directly. Although, you could re-use some parts at 
your convenience.
>   * again how can I include that primitives as a driver for ARM64 in XEN?
The first step is to define your use case. Then we can discuss how we 
can implement PMU support in Xen.
>   * PMU is implemented for each CPU so how to deal with that?
It depends on what you are trying to achieve. If you are exposing the 
PMU to the guest, then it is probably sufficient to store the context in 
the guest vCPU state.
If you need to hold state in Xen as well, then you can use a per-cpu 
storage (see include/xen/percpu.h).
Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel