[linux-2.6-block.git] / drivers / gpu / drm / i915 / gem / i915_gem_object_types.h

/*
 * SPDX-License-Identifier: MIT
 *
 * Copyright © 2016 Intel Corporation
 */

#ifndef __I915_GEM_OBJECT_TYPES_H__
#define __I915_GEM_OBJECT_TYPES_H__

#include <drm/drm_gem.h>
#include <uapi/drm/i915_drm.h>

#include "i915_active.h"
#include "i915_selftest.h"

struct drm_i915_gem_object;
struct intel_fronbuffer;

/*
 * struct i915_lut_handle tracks the fast lookups from handle to vma used
 * for execbuf. Although we use a radixtree for that mapping, in order to
 * remove them as the object or context is closed, we need a secondary list
 * and a translation entry (i915_lut_handle).
 */
struct i915_lut_handle {
	struct list_head obj_link;
	struct i915_gem_context *ctx;
	u32 handle;
};

struct drm_i915_gem_object_ops {
	unsigned int flags;
#define I915_GEM_OBJECT_HAS_STRUCT_PAGE	BIT(0)
#define I915_GEM_OBJECT_HAS_IOMEM	BIT(1)
#define I915_GEM_OBJECT_IS_SHRINKABLE	BIT(2)
#define I915_GEM_OBJECT_IS_PROXY	BIT(3)
#define I915_GEM_OBJECT_NO_MMAP		BIT(4)
#define I915_GEM_OBJECT_ASYNC_CANCEL	BIT(5)

	/* Interface between the GEM object and its backing storage.
	 * get_pages() is called once prior to the use of the associated set
	 * of pages before to binding them into the GTT, and put_pages() is
	 * called after we no longer need them. As we expect there to be
	 * associated cost with migrating pages between the backing storage
	 * and making them available for the GPU (e.g. clflush), we may hold
	 * onto the pages after they are no longer referenced by the GPU
	 * in case they may be used again shortly (for example migrating the
	 * pages to a different memory domain within the GTT). put_pages()
	 * will therefore most likely be called when the object itself is
	 * being released or under memory pressure (where we attempt to
	 * reap pages for the shrinker).
	 */
	int (*get_pages)(struct drm_i915_gem_object *obj);
	void (*put_pages)(struct drm_i915_gem_object *obj,
			  struct sg_table *pages);
	void (*truncate)(struct drm_i915_gem_object *obj);
	void (*writeback)(struct drm_i915_gem_object *obj);

	int (*pread)(struct drm_i915_gem_object *obj,
		     const struct drm_i915_gem_pread *arg);
	int (*pwrite)(struct drm_i915_gem_object *obj,
		      const struct drm_i915_gem_pwrite *arg);

	int (*dmabuf_export)(struct drm_i915_gem_object *obj);
	void (*release)(struct drm_i915_gem_object *obj);

	const char *name; /* friendly name for debug, e.g. lockdep classes */
};

enum i915_map_type {
	I915_MAP_WB = 0,
	I915_MAP_WC,
#define I915_MAP_OVERRIDE BIT(31)
	I915_MAP_FORCE_WB = I915_MAP_WB | I915_MAP_OVERRIDE,
	I915_MAP_FORCE_WC = I915_MAP_WC | I915_MAP_OVERRIDE,
};

enum i915_mmap_type {
	I915_MMAP_TYPE_GTT = 0,
	I915_MMAP_TYPE_WC,
	I915_MMAP_TYPE_WB,
	I915_MMAP_TYPE_UC,
};

struct i915_mmap_offset {
	struct drm_vma_offset_node vma_node;
	struct drm_i915_gem_object *obj;
	enum i915_mmap_type mmap_type;

	struct rb_node offset;
};

struct i915_gem_object_page_iter {
	struct scatterlist *sg_pos;
	unsigned int sg_idx; /* in pages, but 32bit eek! */

	struct radix_tree_root radix;
	struct mutex lock; /* protects this cache */
};

struct drm_i915_gem_object {
	struct drm_gem_object base;

	const struct drm_i915_gem_object_ops *ops;

	struct {
		/**
		 * @vma.lock: protect the list/tree of vmas
		 */
		spinlock_t lock;

		/**
		 * @vma.list: List of VMAs backed by this object
		 *
		 * The VMA on this list are ordered by type, all GGTT vma are
		 * placed at the head and all ppGTT vma are placed at the tail.
		 * The different types of GGTT vma are unordered between
		 * themselves, use the @vma.tree (which has a defined order
		 * between all VMA) to quickly find an exact match.
		 */
		struct list_head list;

		/**
		 * @vma.tree: Ordered tree of VMAs backed by this object
		 *
		 * All VMA created for this object are placed in the @vma.tree
		 * for fast retrieval via a binary search in
		 * i915_vma_instance(). They are also added to @vma.list for
		 * easy iteration.
		 */
		struct rb_root tree;
	} vma;

	/**
	 * @lut_list: List of vma lookup entries in use for this object.
	 *
	 * If this object is closed, we need to remove all of its VMA from
	 * the fast lookup index in associated contexts; @lut_list provides
	 * this translation from object to context->handles_vma.
	 */
	struct list_head lut_list;
	spinlock_t lut_lock; /* guards lut_list */

	/**
	 * @obj_link: Link into @i915_gem_ww_ctx.obj_list
	 *
	 * When we lock this object through i915_gem_object_lock() with a
	 * context, we add it to the list to ensure we can unlock everything
	 * when i915_gem_ww_ctx_backoff() or i915_gem_ww_ctx_fini() are called.
	 */
	struct list_head obj_link;

	union {
		struct rcu_head rcu;
		struct llist_node freed;
	};

	/**
	 * Whether the object is currently in the GGTT mmap.
	 */
	unsigned int userfault_count;
	struct list_head userfault_link;

	struct {
		spinlock_t lock; /* Protects access to mmo offsets */
		struct rb_root offsets;
	} mmo;

	I915_SELFTEST_DECLARE(struct list_head st_link);

	unsigned long flags;
#define I915_BO_ALLOC_CONTIGUOUS BIT(0)
#define I915_BO_ALLOC_VOLATILE   BIT(1)
#define I915_BO_ALLOC_FLAGS (I915_BO_ALLOC_CONTIGUOUS | I915_BO_ALLOC_VOLATILE)
#define I915_BO_READONLY         BIT(2)
#define I915_TILING_QUIRK_BIT    3 /* unknown swizzling; do not release! */

	/*
	 * Is the object to be mapped as read-only to the GPU
	 * Only honoured if hardware has relevant pte bit
	 */
	unsigned int cache_level:3;
	unsigned int cache_coherent:2;
#define I915_BO_CACHE_COHERENT_FOR_READ BIT(0)
#define I915_BO_CACHE_COHERENT_FOR_WRITE BIT(1)
	unsigned int cache_dirty:1;

	/**
	 * @read_domains: Read memory domains.
	 *
	 * These monitor which caches contain read/write data related to the
	 * object. When transitioning from one set of domains to another,
	 * the driver is called to ensure that caches are suitably flushed and
	 * invalidated.
	 */
	u16 read_domains;

	/**
	 * @write_domain: Corresponding unique write memory domain.
	 */
	u16 write_domain;

	struct intel_frontbuffer __rcu *frontbuffer;

	/** Current tiling stride for the object, if it's tiled. */
	unsigned int tiling_and_stride;
#define FENCE_MINIMUM_STRIDE 128 /* See i915_tiling_ok() */
#define TILING_MASK (FENCE_MINIMUM_STRIDE - 1)
#define STRIDE_MASK (~TILING_MASK)

	struct {
		/*
		 * Protects the pages and their use. Do not use directly, but
		 * instead go through the pin/unpin interfaces.
		 */
		struct mutex lock;
		atomic_t pages_pin_count;
		atomic_t shrink_pin;

		/**
		 * Memory region for this object.
		 */
		struct intel_memory_region *region;
		/**
		 * List of memory region blocks allocated for this object.
		 */
		struct list_head blocks;
		/**
		 * Element within memory_region->objects or region->purgeable
		 * if the object is marked as DONTNEED. Access is protected by
		 * region->obj_lock.
		 */
		struct list_head region_link;

		struct sg_table *pages;
		void *mapping;

		struct i915_page_sizes {
			/**
			 * The sg mask of the pages sg_table. i.e the mask of
			 * of the lengths for each sg entry.
			 */
			unsigned int phys;

			/**
			 * The gtt page sizes we are allowed to use given the
			 * sg mask and the supported page sizes. This will
			 * express the smallest unit we can use for the whole
			 * object, as well as the larger sizes we may be able
			 * to use opportunistically.
			 */
			unsigned int sg;

			/**
			 * The actual gtt page size usage. Since we can have
			 * multiple vma associated with this object we need to
			 * prevent any trampling of state, hence a copy of this
			 * struct also lives in each vma, therefore the gtt
			 * value here should only be read/write through the vma.
			 */
			unsigned int gtt;
		} page_sizes;

		I915_SELFTEST_DECLARE(unsigned int page_mask);

		struct i915_gem_object_page_iter get_page;
		struct i915_gem_object_page_iter get_dma_page;

		/**
		 * Element within i915->mm.unbound_list or i915->mm.bound_list,
		 * locked by i915->mm.obj_lock.
		 */
		struct list_head link;

		/**
		 * Advice: are the backing pages purgeable?
		 */
		unsigned int madv:2;

		/**
		 * This is set if the object has been written to since the
		 * pages were last acquired.
		 */
		bool dirty:1;
	} mm;

	/** Record of address bit 17 of each page at last unbind. */
	unsigned long *bit_17;

	union {
		struct i915_gem_userptr {
			uintptr_t ptr;

			struct i915_mm_struct *mm;
			struct i915_mmu_object *mmu_object;
			struct work_struct *work;
		} userptr;

		struct drm_mm_node *stolen;

		unsigned long scratch;
		u64 encode;

		void *gvt_info;
	};
};

static inline struct drm_i915_gem_object *
to_intel_bo(struct drm_gem_object *gem)
{
	/* Assert that to_intel_bo(NULL) == NULL */
	BUILD_BUG_ON(offsetof(struct drm_i915_gem_object, base));

	return container_of(gem, struct drm_i915_gem_object, base);
}

#endif
Commit	Line	Data
5e5d2e20 CW	1	/*
	2	* SPDX-License-Identifier: MIT
	3	*
	4	* Copyright © 2016 Intel Corporation
	5	*/
	6
	7	#ifndef __I915_GEM_OBJECT_TYPES_H__
	8	#define __I915_GEM_OBJECT_TYPES_H__
	9
5e5d2e20	10	#include <drm/drm_gem.h>
b1e3177b	11	#include <uapi/drm/i915_drm.h>
5e5d2e20 CW	12
	13	#include "i915_active.h"
	14	#include "i915_selftest.h"
	15
	16	struct drm_i915_gem_object;
8e7cb179	17	struct intel_fronbuffer;
5e5d2e20 CW	18
	19	/*
	20	* struct i915_lut_handle tracks the fast lookups from handle to vma used
	21	* for execbuf. Although we use a radixtree for that mapping, in order to
	22	* remove them as the object or context is closed, we need a secondary list
	23	* and a translation entry (i915_lut_handle).
	24	*/
	25	struct i915_lut_handle {
	26	struct list_head obj_link;
5e5d2e20 CW	27	struct i915_gem_context *ctx;
	28	u32 handle;
	29	};
	30
	31	struct drm_i915_gem_object_ops {
	32	unsigned int flags;
	33	#define I915_GEM_OBJECT_HAS_STRUCT_PAGE BIT(0)
01377a0d AJ	34	#define I915_GEM_OBJECT_HAS_IOMEM BIT(1)
	35	#define I915_GEM_OBJECT_IS_SHRINKABLE BIT(2)
	36	#define I915_GEM_OBJECT_IS_PROXY BIT(3)
f6c26b55	37	#define I915_GEM_OBJECT_NO_MMAP BIT(4)
01377a0d	38	#define I915_GEM_OBJECT_ASYNC_CANCEL BIT(5)
5e5d2e20 CW	39
	40	/* Interface between the GEM object and its backing storage.
	41	* get_pages() is called once prior to the use of the associated set
	42	* of pages before to binding them into the GTT, and put_pages() is
	43	* called after we no longer need them. As we expect there to be
	44	* associated cost with migrating pages between the backing storage
	45	* and making them available for the GPU (e.g. clflush), we may hold
	46	* onto the pages after they are no longer referenced by the GPU
	47	* in case they may be used again shortly (for example migrating the
	48	* pages to a different memory domain within the GTT). put_pages()
	49	* will therefore most likely be called when the object itself is
	50	* being released or under memory pressure (where we attempt to
	51	* reap pages for the shrinker).
	52	*/
	53	int (get_pages)(struct drm_i915_gem_object obj);
	54	void (put_pages)(struct drm_i915_gem_object obj,
	55	struct sg_table *pages);
f033428d CW	56	void (truncate)(struct drm_i915_gem_object obj);
f033428d CW	57	void (writeback)(struct drm_i915_gem_object obj);
5e5d2e20	58
0049b688 MA	59	int (pread)(struct drm_i915_gem_object obj,
0049b688 MA	60	const struct drm_i915_gem_pread *arg);
5e5d2e20 CW	61	int (pwrite)(struct drm_i915_gem_object obj,
	62	const struct drm_i915_gem_pwrite *arg);
	63
	64	int (dmabuf_export)(struct drm_i915_gem_object obj);
	65	void (release)(struct drm_i915_gem_object obj);
7d192daa CW	66
7d192daa CW	67	const char name; / friendly name for debug, e.g. lockdep classes */
5e5d2e20 CW	68	};
5e5d2e20 CW	69
e2f4367a MA	70	enum i915_map_type {
	71	I915_MAP_WB = 0,
	72	I915_MAP_WC,
	73	#define I915_MAP_OVERRIDE BIT(31)
	74	I915_MAP_FORCE_WB = I915_MAP_WB \| I915_MAP_OVERRIDE,
	75	I915_MAP_FORCE_WC = I915_MAP_WC \| I915_MAP_OVERRIDE,
	76	};
	77
cc662126 AJ	78	enum i915_mmap_type {
	79	I915_MMAP_TYPE_GTT = 0,
	80	I915_MMAP_TYPE_WC,
	81	I915_MMAP_TYPE_WB,
	82	I915_MMAP_TYPE_UC,
	83	};
	84
	85	struct i915_mmap_offset {
cc662126 AJ	86	struct drm_vma_offset_node vma_node;
cc662126 AJ	87	struct drm_i915_gem_object *obj;
cc662126 AJ	88	enum i915_mmap_type mmap_type;
cc662126 AJ	89
78655598	90	struct rb_node offset;
cc662126 AJ	91	};
cc662126 AJ	92
934941ed TU	93	struct i915_gem_object_page_iter {
	94	struct scatterlist *sg_pos;
	95	unsigned int sg_idx; /* in pages, but 32bit eek! */
	96
	97	struct radix_tree_root radix;
	98	struct mutex lock; /* protects this cache */
	99	};
	100
5e5d2e20 CW	101	struct drm_i915_gem_object {
	102	struct drm_gem_object base;
	103
	104	const struct drm_i915_gem_object_ops *ops;
	105
	106	struct {
	107	/**
	108	* @vma.lock: protect the list/tree of vmas
	109	*/
	110	spinlock_t lock;
	111
	112	/**
	113	* @vma.list: List of VMAs backed by this object
	114	*
	115	* The VMA on this list are ordered by type, all GGTT vma are
	116	* placed at the head and all ppGTT vma are placed at the tail.
	117	* The different types of GGTT vma are unordered between
	118	* themselves, use the @vma.tree (which has a defined order
	119	* between all VMA) to quickly find an exact match.
	120	*/
	121	struct list_head list;
	122
	123	/**
	124	* @vma.tree: Ordered tree of VMAs backed by this object
	125	*
	126	* All VMA created for this object are placed in the @vma.tree
	127	* for fast retrieval via a binary search in
	128	* i915_vma_instance(). They are also added to @vma.list for
	129	* easy iteration.
	130	*/
	131	struct rb_root tree;
	132	} vma;
	133
	134	/**
	135	* @lut_list: List of vma lookup entries in use for this object.
	136	*
	137	* If this object is closed, we need to remove all of its VMA from
	138	* the fast lookup index in associated contexts; @lut_list provides
	139	* this translation from object to context->handles_vma.
	140	*/
	141	struct list_head lut_list;
096a42dd	142	spinlock_t lut_lock; /* guards lut_list */
5e5d2e20	143
80f0b679 ML	144	/**
	145	* @obj_link: Link into @i915_gem_ww_ctx.obj_list
	146	*
	147	* When we lock this object through i915_gem_object_lock() with a
	148	* context, we add it to the list to ensure we can unlock everything
	149	* when i915_gem_ww_ctx_backoff() or i915_gem_ww_ctx_fini() are called.
	150	*/
	151	struct list_head obj_link;
	152
5e5d2e20 CW	153	union {
	154	struct rcu_head rcu;
	155	struct llist_node freed;
	156	};
	157
	158	/**
	159	* Whether the object is currently in the GGTT mmap.
	160	*/
	161	unsigned int userfault_count;
	162	struct list_head userfault_link;
	163
cc662126 AJ	164	struct {
cc662126 AJ	165	spinlock_t lock; /* Protects access to mmo offsets */
78655598	166	struct rb_root offsets;
cc662126 AJ	167	} mmo;
cc662126 AJ	168
5e5d2e20 CW	169	I915_SELFTEST_DECLARE(struct list_head st_link);
5e5d2e20 CW	170
2f0b97ca MA	171	unsigned long flags;
2f0b97ca MA	172	#define I915_BO_ALLOC_CONTIGUOUS BIT(0)
7c98501a MA	173	#define I915_BO_ALLOC_VOLATILE BIT(1)
7c98501a MA	174	#define I915_BO_ALLOC_FLAGS (I915_BO_ALLOC_CONTIGUOUS \| I915_BO_ALLOC_VOLATILE)
cc662126	175	#define I915_BO_READONLY BIT(2)
0175969e	176	#define I915_TILING_QUIRK_BIT 3 /* unknown swizzling; do not release! */
2f0b97ca	177
5e5d2e20 CW	178	/*
	179	* Is the object to be mapped as read-only to the GPU
	180	* Only honoured if hardware has relevant pte bit
	181	*/
	182	unsigned int cache_level:3;
	183	unsigned int cache_coherent:2;
	184	#define I915_BO_CACHE_COHERENT_FOR_READ BIT(0)
	185	#define I915_BO_CACHE_COHERENT_FOR_WRITE BIT(1)
	186	unsigned int cache_dirty:1;
	187
	188	/**
	189	* @read_domains: Read memory domains.
	190	*
	191	* These monitor which caches contain read/write data related to the
	192	* object. When transitioning from one set of domains to another,
	193	* the driver is called to ensure that caches are suitably flushed and
	194	* invalidated.
	195	*/
	196	u16 read_domains;
	197
	198	/**
	199	* @write_domain: Corresponding unique write memory domain.
	200	*/
	201	u16 write_domain;
	202
da42104f	203	struct intel_frontbuffer __rcu *frontbuffer;
5e5d2e20 CW	204
	205	/** Current tiling stride for the object, if it's tiled. */
	206	unsigned int tiling_and_stride;
	207	#define FENCE_MINIMUM_STRIDE 128 /* See i915_tiling_ok() */
	208	#define TILING_MASK (FENCE_MINIMUM_STRIDE - 1)
	209	#define STRIDE_MASK (~TILING_MASK)
	210
5e5d2e20	211	struct {
f86dbacb DV	212	/*
	213	* Protects the pages and their use. Do not use directly, but
	214	* instead go through the pin/unpin interfaces.
	215	*/
	216	struct mutex lock;
5e5d2e20	217	atomic_t pages_pin_count;
99013b10	218	atomic_t shrink_pin;
5e5d2e20	219
232a6eba MA	220	/**
	221	* Memory region for this object.
	222	*/
	223	struct intel_memory_region *region;
	224	/**
	225	* List of memory region blocks allocated for this object.
	226	*/
	227	struct list_head blocks;
7c98501a MA	228	/**
	229	* Element within memory_region->objects or region->purgeable
	230	* if the object is marked as DONTNEED. Access is protected by
	231	* region->obj_lock.
	232	*/
	233	struct list_head region_link;
232a6eba	234
5e5d2e20 CW	235	struct sg_table *pages;
	236	void *mapping;
	237
5e5d2e20 CW	238	struct i915_page_sizes {
	239	/**
	240	* The sg mask of the pages sg_table. i.e the mask of
	241	* of the lengths for each sg entry.
	242	*/
	243	unsigned int phys;
	244
	245	/**
	246	* The gtt page sizes we are allowed to use given the
	247	* sg mask and the supported page sizes. This will
	248	* express the smallest unit we can use for the whole
	249	* object, as well as the larger sizes we may be able
	250	* to use opportunistically.
	251	*/
	252	unsigned int sg;
	253
	254	/**
	255	* The actual gtt page size usage. Since we can have
	256	* multiple vma associated with this object we need to
	257	* prevent any trampling of state, hence a copy of this
	258	* struct also lives in each vma, therefore the gtt
	259	* value here should only be read/write through the vma.
	260	*/
	261	unsigned int gtt;
	262	} page_sizes;
	263
	264	I915_SELFTEST_DECLARE(unsigned int page_mask);
	265
934941ed TU	266	struct i915_gem_object_page_iter get_page;
934941ed TU	267	struct i915_gem_object_page_iter get_dma_page;
5e5d2e20 CW	268
	269	/**
	270	* Element within i915->mm.unbound_list or i915->mm.bound_list,
	271	* locked by i915->mm.obj_lock.
	272	*/
	273	struct list_head link;
	274
	275	/**
	276	* Advice: are the backing pages purgeable?
	277	*/
	278	unsigned int madv:2;
	279
	280	/**
	281	* This is set if the object has been written to since the
	282	* pages were last acquired.
	283	*/
	284	bool dirty:1;
5e5d2e20 CW	285	} mm;
5e5d2e20 CW	286
5e5d2e20 CW	287	/** Record of address bit 17 of each page at last unbind. */
	288	unsigned long *bit_17;
	289
	290	union {
	291	struct i915_gem_userptr {
	292	uintptr_t ptr;
	293
	294	struct i915_mm_struct *mm;
	295	struct i915_mmu_object *mmu_object;
	296	struct work_struct *work;
	297	} userptr;
	298
41a9c75d CW	299	struct drm_mm_node *stolen;
41a9c75d CW	300
5e5d2e20	301	unsigned long scratch;
89351925	302	u64 encode;
5e5d2e20 CW	303
	304	void *gvt_info;
	305	};
5e5d2e20 CW	306	};
	307
	308	static inline struct drm_i915_gem_object *
	309	to_intel_bo(struct drm_gem_object *gem)
	310	{
	311	/* Assert that to_intel_bo(NULL) == NULL */
	312	BUILD_BUG_ON(offsetof(struct drm_i915_gem_object, base));
	313
	314	return container_of(gem, struct drm_i915_gem_object, base);
	315	}
	316
	317	#endif