[Intel-gfx] [RFC 01/10] drm/i915/vm_bind: Introduce VM_BIND ioctl

Hellstrom, Thomas thomas.hellstrom at intel.com
Tue Jul 5 09:59:24 UTC 2022


Hi,


On Fri, 2022-07-01 at 15:50 -0700, Niranjana Vishwanathapura wrote:
> Add VM_BIND and VM_UNBIND ioctls to bind/unbind a section of an
> object at the specified GPU virtual addresses.
> 
> Add I915_PARAM_VM_BIND_VERSION to indicate version of VM_BIND feature
> supported and I915_VM_CREATE_FLAGS_USE_VM_BIND for UMDs to select the
> vm_bind mode of binding.
> 
> Signed-off-by: Niranjana Vishwanathapura
> <niranjana.vishwanathapura at intel.com>

Some comments on patch ordering. In order to ease reviews and to not
introduce unwanted surprises, could we

1) Add patches that introduce needed internal functionality /
refactoring / helpers.
2) Add patches that add enable intended user-space functionality, any
yet unsupported functionality disabled.
3) Add patches that introduce additional internal functionality /
refactoring / helpers.
4) Add patches that enable that additional functionality.

Fixes that are known at series submission time squashed before a
feature is enabled.


> ---
>  drivers/gpu/drm/i915/gem/i915_gem_context.c |  20 +-
>  drivers/gpu/drm/i915/gem/i915_gem_context.h |  15 ++
>  drivers/gpu/drm/i915/gt/intel_gtt.h         |   6 +
>  drivers/gpu/drm/i915/i915_driver.c          |  30 +++
>  drivers/gpu/drm/i915/i915_getparam.c        |   3 +
>  include/uapi/drm/i915_drm.h                 | 192
> +++++++++++++++++++-
>  6 files changed, 248 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> index dabdfe09f5e5..e3f5fbf2ac05 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> @@ -81,7 +81,6 @@
>  
>  #include "pxp/intel_pxp.h"
>  
> -#include "i915_file_private.h"
>  #include "i915_gem_context.h"
>  #include "i915_trace.h"
>  #include "i915_user_extensions.h"
> @@ -346,20 +345,6 @@ static int proto_context_register(struct
> drm_i915_file_private *fpriv,
>         return ret;
>  }
>  
> -static struct i915_address_space *
> -i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
> -{
> -       struct i915_address_space *vm;
> -
> -       xa_lock(&file_priv->vm_xa);
> -       vm = xa_load(&file_priv->vm_xa, id);
> -       if (vm)
> -               kref_get(&vm->ref);
> -       xa_unlock(&file_priv->vm_xa);
> -
> -       return vm;
> -}
> -
>  static int set_proto_ctx_vm(struct drm_i915_file_private *fpriv,
>                             struct i915_gem_proto_context *pc,
>                             const struct drm_i915_gem_context_param
> *args)
> @@ -1799,7 +1784,7 @@ int i915_gem_vm_create_ioctl(struct drm_device
> *dev, void *data,
>         if (!HAS_FULL_PPGTT(i915))
>                 return -ENODEV;
>  
> -       if (args->flags)
> +       if (args->flags & I915_VM_CREATE_FLAGS_UNKNOWN)
>                 return -EINVAL;
>  
>         ppgtt = i915_ppgtt_create(to_gt(i915), 0);
> @@ -1819,6 +1804,9 @@ int i915_gem_vm_create_ioctl(struct drm_device
> *dev, void *data,
>         if (err)
>                 goto err_put;
>  
> +       if (args->flags & I915_VM_CREATE_FLAGS_USE_VM_BIND)
> +               ppgtt->vm.vm_bind_mode = true;
> +
>         GEM_BUG_ON(id == 0); /* reserved for invalid/unassigned ppgtt
> */
>         args->vm_id = id;
>         return 0;
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.h
> b/drivers/gpu/drm/i915/gem/i915_gem_context.h
> index e5b0f66ea1fe..723bf446c934 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_context.h
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.h
> @@ -12,6 +12,7 @@
>  #include "gt/intel_context.h"
>  
>  #include "i915_drv.h"
> +#include "i915_file_private.h"
>  #include "i915_gem.h"
>  #include "i915_scheduler.h"
>  #include "intel_device_info.h"
> @@ -139,6 +140,20 @@ int i915_gem_context_setparam_ioctl(struct
> drm_device *dev, void *data,
>  int i915_gem_context_reset_stats_ioctl(struct drm_device *dev, void
> *data,
>                                        struct drm_file *file);
>  
> +static inline struct i915_address_space *
> +i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
> +{
> +       struct i915_address_space *vm;
> +
> +       xa_lock(&file_priv->vm_xa);
> +       vm = xa_load(&file_priv->vm_xa, id);
> +       if (vm)
> +               kref_get(&vm->ref);
> +       xa_unlock(&file_priv->vm_xa);
> +
> +       return vm;
> +}

Does this really need to be inlined?

> +
>  struct i915_gem_context *
>  i915_gem_context_lookup(struct drm_i915_file_private *file_priv, u32
> id);
>  
> diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h
> b/drivers/gpu/drm/i915/gt/intel_gtt.h
> index e639434e97fd..c812aa9708ae 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gtt.h
> +++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
> @@ -271,6 +271,12 @@ struct i915_address_space {
>         /* Skip pte rewrite on unbind for suspend. Protected by
> @mutex */
>         bool skip_pte_rewrite:1;
>  
> +       /**
> +        * true: allow only vm_bind method of binding.
> +        * false: allow only legacy execbuff method of binding.
> +        */

Use proper kerneldoc. (Same holds for structure documentation across
the series).
Also please follow internal locking guidelines on documentation of
members that need protection with locks.

> +       bool vm_bind_mode:1;
> +
>         u8 top;
>         u8 pd_shift;
>         u8 scratch_order;
> diff --git a/drivers/gpu/drm/i915/i915_driver.c
> b/drivers/gpu/drm/i915/i915_driver.c
> index deb8a8b76965..ccf990dfd99b 100644
> --- a/drivers/gpu/drm/i915/i915_driver.c
> +++ b/drivers/gpu/drm/i915/i915_driver.c
> @@ -1778,6 +1778,34 @@ i915_gem_reject_pin_ioctl(struct drm_device
> *dev, void *data,
>         return -ENODEV;
>  }
>  
> +static int i915_gem_vm_bind_ioctl(struct drm_device *dev, void
> *data,
> +                                 struct drm_file *file)
> +{
> +       struct drm_i915_gem_vm_bind *args = data;
> +       struct i915_address_space *vm;
> +
> +       vm = i915_gem_vm_lookup(file->driver_priv, args->vm_id);
> +       if (unlikely(!vm))
> +               return -ENOENT;
> +
> +       i915_vm_put(vm);
> +       return -EINVAL;
> +}
> +
> +static int i915_gem_vm_unbind_ioctl(struct drm_device *dev, void
> *data,
> +                                   struct drm_file *file)
> +{
> +       struct drm_i915_gem_vm_unbind *args = data;
> +       struct i915_address_space *vm;
> +
> +       vm = i915_gem_vm_lookup(file->driver_priv, args->vm_id);
> +       if (unlikely(!vm))
> +               return -ENOENT;
> +
> +       i915_vm_put(vm);
> +       return -EINVAL;
> +}
> +

Move these functions to the file of the actual implementation?

>  static const struct drm_ioctl_desc i915_ioctls[] = {
>         DRM_IOCTL_DEF_DRV(I915_INIT, drm_noop,
> DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
>         DRM_IOCTL_DEF_DRV(I915_FLUSH, drm_noop, DRM_AUTH),
> @@ -1838,6 +1866,8 @@ static const struct drm_ioctl_desc
> i915_ioctls[] = {
>         DRM_IOCTL_DEF_DRV(I915_QUERY, i915_query_ioctl,
> DRM_RENDER_ALLOW),
>         DRM_IOCTL_DEF_DRV(I915_GEM_VM_CREATE,
> i915_gem_vm_create_ioctl, DRM_RENDER_ALLOW),
>         DRM_IOCTL_DEF_DRV(I915_GEM_VM_DESTROY,
> i915_gem_vm_destroy_ioctl, DRM_RENDER_ALLOW),
> +       DRM_IOCTL_DEF_DRV(I915_GEM_VM_BIND, i915_gem_vm_bind_ioctl,
> DRM_RENDER_ALLOW),
> +       DRM_IOCTL_DEF_DRV(I915_GEM_VM_UNBIND,
> i915_gem_vm_unbind_ioctl, DRM_RENDER_ALLOW),
>  };
>  
>  /*
> diff --git a/drivers/gpu/drm/i915/i915_getparam.c
> b/drivers/gpu/drm/i915/i915_getparam.c
> index 6fd15b39570c..c1d53febc5de 100644
> --- a/drivers/gpu/drm/i915/i915_getparam.c
> +++ b/drivers/gpu/drm/i915/i915_getparam.c
> @@ -175,6 +175,9 @@ int i915_getparam_ioctl(struct drm_device *dev,
> void *data,
>         case I915_PARAM_PERF_REVISION:
>                 value = i915_perf_ioctl_version();
>                 break;
> +       case I915_PARAM_VM_BIND_VERSION:
> +               value = GRAPHICS_VER(i915) >= 12 ? 1 : 0;
> +               break;
>         default:
>                 DRM_DEBUG("Unknown parameter %d\n", param->param);
>                 return -EINVAL;
> diff --git a/include/uapi/drm/i915_drm.h
> b/include/uapi/drm/i915_drm.h
> index 3e78a00220ea..26cca49717f8 100644
> --- a/include/uapi/drm/i915_drm.h
> +++ b/include/uapi/drm/i915_drm.h
> @@ -470,6 +470,8 @@ typedef struct _drm_i915_sarea {
>  #define DRM_I915_GEM_VM_CREATE         0x3a
>  #define DRM_I915_GEM_VM_DESTROY                0x3b
>  #define DRM_I915_GEM_CREATE_EXT                0x3c
> +#define DRM_I915_GEM_VM_BIND           0x3d
> +#define DRM_I915_GEM_VM_UNBIND         0x3e
>  /* Must be kept compact -- no holes */
>  
>  #define DRM_IOCTL_I915_INIT            DRM_IOW( DRM_COMMAND_BASE +
> DRM_I915_INIT, drm_i915_init_t)
> @@ -534,6 +536,8 @@ typedef struct _drm_i915_sarea {
>  #define
> DRM_IOCTL_I915_QUERY                   DRM_IOWR(DRM_COMMAND_BASE +
> DRM_I915_QUERY, struct drm_i915_query)
>  #define DRM_IOCTL_I915_GEM_VM_CREATE   DRM_IOWR(DRM_COMMAND_BASE +
> DRM_I915_GEM_VM_CREATE, struct drm_i915_gem_vm_control)
>  #define DRM_IOCTL_I915_GEM_VM_DESTROY  DRM_IOW (DRM_COMMAND_BASE +
> DRM_I915_GEM_VM_DESTROY, struct drm_i915_gem_vm_control)
> +#define DRM_IOCTL_I915_GEM_VM_BIND     DRM_IOWR(DRM_COMMAND_BASE +
> DRM_I915_GEM_VM_BIND, struct drm_i915_gem_vm_bind)
> +#define DRM_IOCTL_I915_GEM_VM_UNBIND   DRM_IOWR(DRM_COMMAND_BASE +
> DRM_I915_GEM_VM_UNBIND, struct drm_i915_gem_vm_unbind)
>  
>  /* Allow drivers to submit batchbuffers directly to hardware,
> relying
>   * on the security mechanisms provided by hardware.
> @@ -749,6 +753,25 @@ typedef struct drm_i915_irq_wait {
>  /* Query if the kernel supports the I915_USERPTR_PROBE flag. */
>  #define I915_PARAM_HAS_USERPTR_PROBE 56
>  
> +/*
> + * VM_BIND feature version supported.
> + *
> + * The following versions of VM_BIND have been defined:
> + *
> + * 0: No VM_BIND support.
> + *
> + * 1: In VM_UNBIND calls, the UMD must specify the exact mappings
> created
> + *    previously with VM_BIND, the ioctl will not support unbinding
> multiple
> + *    mappings or splitting them. Similarly, VM_BIND calls will not
> replace
> + *    any existing mappings.
> + *
> + * 2: The restrictions on unbinding partial or multiple mappings is
> + *    lifted, Similarly, binding will replace any mappings in the
> given range.
> + *
> + * See struct drm_i915_gem_vm_bind and struct
> drm_i915_gem_vm_unbind.
> + */
> +#define I915_PARAM_VM_BIND_VERSION     57

Perhaps clarify that new versions are always backwards compatible?

> +
>  /* Must be kept compact -- no holes and well documented */
>  
>  typedef struct drm_i915_getparam {
> @@ -1441,6 +1464,41 @@ struct drm_i915_gem_execbuffer2 {
>  #define i915_execbuffer2_get_context_id(eb2) \
>         ((eb2).rsvd1 & I915_EXEC_CONTEXT_ID_MASK)
>  
> +/**
> + * struct drm_i915_gem_timeline_fence - An input or output timeline
> fence.
> + *
> + * The operation will wait for input fence to signal.
> + *
> + * The returned output fence will be signaled after the completion
> of the
> + * operation.
> + */
> +struct drm_i915_gem_timeline_fence {
> +       /** @handle: User's handle for a drm_syncobj to wait on or
> signal. */
> +       __u32 handle;
> +
> +       /**
> +        * @flags: Supported flags are:
> +        *
> +        * I915_TIMELINE_FENCE_WAIT:
> +        * Wait for the input fence before the operation.
> +        *
> +        * I915_TIMELINE_FENCE_SIGNAL:
> +        * Return operation completion fence as output.
> +        */
> +       __u32 flags;
> +#define I915_TIMELINE_FENCE_WAIT            (1 << 0)
> +#define I915_TIMELINE_FENCE_SIGNAL          (1 << 1)
> +#define __I915_TIMELINE_FENCE_UNKNOWN_FLAGS (-
> (I915_TIMELINE_FENCE_SIGNAL << 1))
> +
> +       /**
> +        * @value: A point in the timeline.
> +        * Value must be 0 for a binary drm_syncobj. A Value of 0 for
> a
> +        * timeline drm_syncobj is invalid as it turns a drm_syncobj
> into a
> +        * binary one.
> +        */
> +       __u64 value;
> +};
> +
>  struct drm_i915_gem_pin {
>         /** Handle of the buffer to be pinned. */
>         __u32 handle;
> @@ -2397,8 +2455,6 @@ struct drm_i915_gem_context_destroy {
>   * The id of new VM (bound to the fd) for use with
> I915_CONTEXT_PARAM_VM is
>   * returned in the outparam @id.
>   *
> - * No flags are defined, with all bits reserved and must be zero.
> - *
>   * An extension chain maybe provided, starting with @extensions, and
> terminated
>   * by the @next_extension being 0. Currently, no extensions are
> defined.
>   *
> @@ -2410,6 +2466,10 @@ struct drm_i915_gem_context_destroy {
>   */
>  struct drm_i915_gem_vm_control {
>         __u64 extensions;
> +
> +#define I915_VM_CREATE_FLAGS_USE_VM_BIND       (1u << 0)
> +#define I915_VM_CREATE_FLAGS_UNKNOWN \
> +       (-(I915_VM_CREATE_FLAGS_USE_VM_BIND << 1))
>         __u32 flags;
>         __u32 vm_id;
>  };
> @@ -3602,6 +3662,134 @@ struct
> drm_i915_gem_create_ext_protected_content {
>  /* ID of the protected content session managed by i915 when PXP is
> active */
>  #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf
>  
> +/**
> + * struct drm_i915_gem_vm_bind - VA to object mapping to bind.
> + *
> + * This structure is passed to VM_BIND ioctl and specifies the
> mapping of GPU
> + * virtual address (VA) range to the section of an object that
> should be bound
> + * in the device page table of the specified address space (VM).
> + * The VA range specified must be unique (ie., not currently bound)
> and can
> + * be mapped to whole object or a section of the object (partial
> binding).
> + * Multiple VA mappings can be created to the same section of the
> object
> + * (aliasing).
> + *
> + * The @start, @offset and @length must be 4K page aligned. However
> the DG2
> + * and XEHPSDV has 64K page size for device local memory and has
> compact page
> + * table. On those platforms, for binding device local-memory
> objects, the
> + * @start, @offset and @length must be 64K aligned. Also, UMDs
> should not mix
> + * the local memory 64K page and the system memory 4K page bindings
> in the same
> + * 2M range.
> + *
> + * Error code -EINVAL will be returned if @start, @offset and
> @length are not
> + * properly aligned. In version 1 (See I915_PARAM_VM_BIND_VERSION),
> error code
> + * -ENOSPC will be returned if the VA range specified can't be
> reserved.
> + *
> + * VM_BIND/UNBIND ioctl calls executed on different CPU threads
> concurrently
> + * are not ordered. Furthermore, parts of the VM_BIND operation can
> be done
> + * asynchronously, if valid @fence is specified.
> + */
> +struct drm_i915_gem_vm_bind {
> +       /** @vm_id: VM (address space) id to bind */
> +       __u32 vm_id;
> +
> +       /** @handle: Object handle */
> +       __u32 handle;
> +
> +       /** @start: Virtual Address start to bind */
> +       __u64 start;
> +
> +       /** @offset: Offset in object to bind */
> +       __u64 offset;
> +
> +       /** @length: Length of mapping to bind */
> +       __u64 length;
> +
> +       /**
> +        * @flags: Currently reserved, MBZ.
> +        *
> +        * Note that @fence carries its own flags.
> +        */
> +       __u64 flags;
> +
> +       /**
> +        * @fence: Timeline fence for bind completion signaling.
> +        *
> +        * Timeline fence is of format struct
> drm_i915_gem_timeline_fence.
> +        *
> +        * It is an out fence, hence using I915_TIMELINE_FENCE_WAIT
> flag
> +        * is invalid, and an error will be returned.
> +        *
> +        * If I915_TIMELINE_FENCE_SIGNAL flag is not set, then out
> fence
> +        * is not requested and binding is completed synchronously.
> +        */
> +       struct drm_i915_gem_timeline_fence fence;
> +
> +       /**
> +        * @extensions: Zero-terminated chain of extensions.
> +        *
> +        * For future extensions. See struct i915_user_extension.
> +        */
> +       __u64 extensions;
> +};
> +
> +/**
> + * struct drm_i915_gem_vm_unbind - VA to object mapping to unbind.
> + *
> + * This structure is passed to VM_UNBIND ioctl and specifies the GPU
> virtual
> + * address (VA) range that should be unbound from the device page
> table of the
> + * specified address space (VM). VM_UNBIND will force unbind the
> specified
> + * range from device page table without waiting for any GPU job to
> complete.
> + * It is UMDs responsibility to ensure the mapping is no longer in
> use before
> + * calling VM_UNBIND.
> + *
> + * If the specified mapping is not found, the ioctl will simply
> return without
> + * any error.
> + *
> + * VM_BIND/UNBIND ioctl calls executed on different CPU threads
> concurrently
> + * are not ordered. Furthermore, parts of the VM_UNBIND operation
> can be done
> + * asynchronously, if valid @fence is specified.
> + */
> +struct drm_i915_gem_vm_unbind {
> +       /** @vm_id: VM (address space) id to bind */
> +       __u32 vm_id;
> +
> +       /** @rsvd: Reserved, MBZ */
> +       __u32 rsvd;
> +
> +       /** @start: Virtual Address start to unbind */
> +       __u64 start;
> +
> +       /** @length: Length of mapping to unbind */
> +       __u64 length;
> +
> +       /**
> +        * @flags: Currently reserved, MBZ.
> +        *
> +        * Note that @fence carries its own flags.
> +        */
> +       __u64 flags;
> +
> +       /**
> +        * @fence: Timeline fence for unbind completion signaling.
> +        *
> +        * Timeline fence is of format struct
> drm_i915_gem_timeline_fence.
> +        *
> +        * It is an out fence, hence using I915_TIMELINE_FENCE_WAIT
> flag
> +        * is invalid, and an error will be returned.
> +        *
> +        * If I915_TIMELINE_FENCE_SIGNAL flag is not set, then out
> fence
> +        * is not requested and unbinding is completed synchronously.
> +        */
> +       struct drm_i915_gem_timeline_fence fence;
> +
> +       /**
> +        * @extensions: Zero-terminated chain of extensions.
> +        *
> +        * For future extensions. See struct i915_user_extension.
> +        */
> +       __u64 extensions;
> +};
> +
>  #if defined(__cplusplus)
>  }
>  #endif



More information about the Intel-gfx mailing list