Re: [PATCH v7 4/6] drm/xe/uapi: Define drm_xe_vm_get_faults

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




________________________________________
From: Brost, Matthew <matthew.brost@xxxxxxxxx>
Sent: Monday, March 10, 2025 9:50 PM
To: Cavitt, Jonathan
Cc: intel-xe@xxxxxxxxxxxxxxxxxxxxx; Gupta, saurabhg; Zuo, Alex; joonas.lahtinen@xxxxxxxxxxxxxxx; Zhang, Jianxun; Lin, Shuicheng; dri-devel@xxxxxxxxxxxxxxxxxxxxx
Subject: Re: [PATCH v7 4/6] drm/xe/uapi: Define drm_xe_vm_get_faults

On Mon, Mar 10, 2025 at 05:18:32PM +0000, Jonathan Cavitt wrote:
> Add initial declarations for the drm_xe_vm_get_faults ioctl.
>
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@xxxxxxxxx>
> ---
>  include/uapi/drm/xe_drm.h | 49 +++++++++++++++++++++++++++++++++++++++
>  1 file changed, 49 insertions(+)
>
> diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
> index 616916985e3f..90c2fcdbd5c1 100644
> --- a/include/uapi/drm/xe_drm.h
> +++ b/include/uapi/drm/xe_drm.h
> @@ -81,6 +81,7 @@ extern "C" {
>   *  - &DRM_IOCTL_XE_EXEC
>   *  - &DRM_IOCTL_XE_WAIT_USER_FENCE
>   *  - &DRM_IOCTL_XE_OBSERVATION
> + *  - %DRM_IOCTL_XE_VM_GET_FAULTS

This should be a generic "get VM property" IOCTL, not a specific IOCTL
that only retrieves faults. This allows for future expansion of the
uAPI.

>   */
>
>  /*
> @@ -102,6 +103,7 @@ extern "C" {
>  #define DRM_XE_EXEC                  0x09
>  #define DRM_XE_WAIT_USER_FENCE               0x0a
>  #define DRM_XE_OBSERVATION           0x0b
> +#define DRM_XE_VM_GET_FAULTS         0x0c
>
>  /* Must be kept compact -- no holes */
>
> @@ -117,6 +119,7 @@ extern "C" {
>  #define DRM_IOCTL_XE_EXEC                    DRM_IOW(DRM_COMMAND_BASE + DRM_XE_EXEC, struct drm_xe_exec)
>  #define DRM_IOCTL_XE_WAIT_USER_FENCE         DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_WAIT_USER_FENCE, struct drm_xe_wait_user_fence)
>  #define DRM_IOCTL_XE_OBSERVATION             DRM_IOW(DRM_COMMAND_BASE + DRM_XE_OBSERVATION, struct drm_xe_observation_param)
> +#define DRM_IOCTL_XE_VM_GET_FAULTS           DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_VM_GET_FAULTS, struct drm_xe_vm_get_faults)
>
>  /**
>   * DOC: Xe IOCTL Extensions
> @@ -1189,6 +1192,52 @@ struct drm_xe_vm_bind {
>       __u64 reserved[2];
>  };
>
> +struct xe_vm_fault {
> +     /** @address: Address of the fault, if relevant */
> +     __u64 address;
> +#define DRM_XE_FAULT_ADDRESS_TYPE_NONE_EXT           0
> +#define DRM_XE_FAULT_ADDRESS_TYPE_READ_INVALID_EXT   1
> +#define DRM_XE_FAULT_ADDRESS_TYPE_WRITE_INVALID_EXT  2
> +     /** @address_type: , if relevant */
> +     __u32 address_type;
> +     /**
> +      * @address_precision: Precision of faulted address, if relevant.
> +      * Currently only SZ_4K.
The fault address should be no ambiguity, so I think the precision should be 1 for page fault. Please correct me if I am wrong here. 
> +      */
> +     __u32 address_precision;
As I mentioned in offline discussion, I suggest we add engine class and instance into page fault report. I think it is important to know which engine the fault was from.

> +     /** @reserved: MBZ */
> +     __u64 reserved[3];
> +};
> +
> +/**
> + * struct drm_xe_vm_get_faults - Input of &DRM_IOCTL_XE_VM_GET_FAULTS
> + *
> + * The user provides a VM ID, and the ioctl will
> + *
> + */
> +struct drm_xe_vm_get_faults {
> +     /** @extensions: Pointer to the first extension struct, if any */
> +     __u64 extensions;
> +
> +     /** @vm_id: The ID of the VM to query the properties of */
> +     __u32 vm_id;
> +
> +     /** @size: Size to allocate for @ptr */
> +     __u32 size;
> +
> +     /** @fault_count: Number of faults to be returned */
> +     __u32 fault_count;
> +

fault_count is implied by size.

Matt

I think we should remove 'size' and just use fault_count. It is sufficient for user space to calculate the needed size of memory by fault_count  and size of xe_vm_fault. In this way it is also better than just using 'size' that causes everywhere to do a dividing op and check the fraction for getting fault_count. Also, the only place to use 'size' is memory allocation. But I can imagine multiple places in UMD and KMD would need fault_count.

> +     /** @pad: MBZ */
> +     __u32 pad;
> +
> +     /** @reserved: MBZ */
> +     __u64 reserved[2];
> +
> +     /** @faults: Pointer to user-defined array of xe_vm_fault of flexible size */
/** @faults: Pointer to a xe_vm_fault array of fault_count elements provided by user space */
> +     __u64 faults;
> +};
> +
>  /**
>   * struct drm_xe_exec_queue_create - Input of &DRM_IOCTL_XE_EXEC_QUEUE_CREATE
>   *
> --
> 2.43.0
>




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux