[PATCH 01/24] drm/amdkfd/kfd_ioctl: add pc sampling support

[Alex Deucher]

On Fri, Nov 3, 2023 at 9:22 AM James Zhu <James.Zhu at amd.com> wrote:
>
> From: David Yat Sin <david.yatsin at amd.com>
>
> Add pc sampling support in kfd_ioctl.
>
> Co-developed-by: James Zhu <James.Zhu at amd.com>
> Signed-off-by: James Zhu <James.Zhu at amd.com>
> Signed-off-by: David Yat Sin <david.yatsin at amd.com>

For any new IOCTL interfaces, please provide a link to the user mode
code branch which uses it in the patch description.

Thanks,

Alex

> ---
>  include/uapi/linux/kfd_ioctl.h | 57 +++++++++++++++++++++++++++++++++-
>  1 file changed, 56 insertions(+), 1 deletion(-)
>
> diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
> index f0ed68974c54..5202e29c9560 100644
> --- a/include/uapi/linux/kfd_ioctl.h
> +++ b/include/uapi/linux/kfd_ioctl.h
> @@ -1446,6 +1446,58 @@ struct kfd_ioctl_dbg_trap_args {
>         };
>  };
>
> +/**
> + * kfd_ioctl_pc_sample_op - PC Sampling ioctl operations
> + *
> + * @KFD_IOCTL_PCS_OP_QUERY_CAPABILITIES: Query device PC Sampling capabilities
> + * @KFD_IOCTL_PCS_OP_CREATE:             Register this process with a per-device PC sampler instance
> + * @KFD_IOCTL_PCS_OP_DESTROY:            Unregister from a previously registered PC sampler instance
> + * @KFD_IOCTL_PCS_OP_START:              Process begins taking samples from a previously registered PC sampler instance
> + * @KFD_IOCTL_PCS_OP_STOP:               Process stops taking samples from a previously registered PC sampler instance
> + */
> +enum kfd_ioctl_pc_sample_op {
> +       KFD_IOCTL_PCS_OP_QUERY_CAPABILITIES,
> +       KFD_IOCTL_PCS_OP_CREATE,
> +       KFD_IOCTL_PCS_OP_DESTROY,
> +       KFD_IOCTL_PCS_OP_START,
> +       KFD_IOCTL_PCS_OP_STOP,
> +};
> +
> +/* Values have to be a power of 2*/
> +#define KFD_IOCTL_PCS_FLAG_POWER_OF_2 0x00000001
> +
> +enum kfd_ioctl_pc_sample_method {
> +       KFD_IOCTL_PCS_METHOD_HOSTTRAP = 1,
> +       KFD_IOCTL_PCS_METHOD_STOCHASTIC,
> +};
> +
> +enum kfd_ioctl_pc_sample_type {
> +       KFD_IOCTL_PCS_TYPE_TIME_US,
> +       KFD_IOCTL_PCS_TYPE_CLOCK_CYCLES,
> +       KFD_IOCTL_PCS_TYPE_INSTRUCTIONS
> +};
> +
> +struct kfd_pc_sample_info {
> +       __u64 value;         /* [IN] if PCS_TYPE_INTERVAL_US: sample interval in us
> +                             * if PCS_TYPE_CLOCK_CYCLES: sample interval in graphics core clk cycles
> +                             * if PCS_TYPE_INSTRUCTIONS: sample interval in instructions issued by
> +                             * graphics compute units
> +                             */
> +       __u64 value_min;     /* [OUT] */
> +       __u64 value_max;     /* [OUT] */
> +       __u64 flags;         /* [OUT] indicate potential restrictions e.g FLAG_POWER_OF_2 */
> +       __u32 method;        /* [IN/OUT] kfd_ioctl_pc_sample_method */
> +       __u32 type;          /* [IN/OUT] kfd_ioctl_pc_sample_type */
> +};
> +
> +struct kfd_ioctl_pc_sample_args {
> +       __u64 sample_info_ptr;   /* array of kfd_pc_sample_info */
> +       __u32 num_sample_info;
> +       __u32 op;                /* kfd_ioctl_pc_sample_op */
> +       __u32 gpu_id;
> +       __u32 trace_id;
> +};
> +
>  #define AMDKFD_IOCTL_BASE 'K'
>  #define AMDKFD_IO(nr)                  _IO(AMDKFD_IOCTL_BASE, nr)
>  #define AMDKFD_IOR(nr, type)           _IOR(AMDKFD_IOCTL_BASE, nr, type)
> @@ -1566,7 +1618,10 @@ struct kfd_ioctl_dbg_trap_args {
>  #define AMDKFD_IOC_DBG_TRAP                    \
>                 AMDKFD_IOWR(0x26, struct kfd_ioctl_dbg_trap_args)
>
> +#define AMDKFD_IOC_PC_SAMPLE           \
> +               AMDKFD_IOWR(0x27, struct kfd_ioctl_pc_sample_args)
> +
>  #define AMDKFD_COMMAND_START           0x01
> -#define AMDKFD_COMMAND_END             0x27
> +#define AMDKFD_COMMAND_END             0x28
>
>  #endif
> --
> 2.25.1
>