1 /*
2  * Copyright © 2014 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Please try to maintain the following order within this file unless it makes
24  * sense to do otherwise. From top to bottom:
25  * 1. typedefs
26  * 2. #defines, and macros
27  * 3. structure definitions
28  * 4. function prototypes
29  *
30  * Within each section, please try to order by generation in ascending order,
31  * from top to bottom (ie. gen6 on the top, gen8 on the bottom).
32  */
33 
34 #ifndef __I915_GEM_GTT_H__
35 #define __I915_GEM_GTT_H__
36 
37 typedef uint32_t gen6_gtt_pte_t;
38 typedef uint64_t gen8_gtt_pte_t;
39 typedef gen8_gtt_pte_t gen8_ppgtt_pde_t;
40 
41 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
42 
43 #define I915_PPGTT_PT_ENTRIES		(PAGE_SIZE / sizeof(gen6_gtt_pte_t))
44 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
45 #define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
46 #define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
47 #define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
48 #define GEN6_PTE_CACHE_LLC		(2 << 1)
49 #define GEN6_PTE_UNCACHED		(1 << 1)
50 #define GEN6_PTE_VALID			(1 << 0)
51 
52 #define GEN6_PPGTT_PD_ENTRIES		512
53 #define GEN6_PD_SIZE			(GEN6_PPGTT_PD_ENTRIES * PAGE_SIZE)
54 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
55 #define GEN6_PDE_VALID			(1 << 0)
56 
57 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
58 
59 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
60 #define BYT_PTE_WRITEABLE		(1 << 1)
61 
62 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
63  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
64  */
65 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
66 					 (((bits) & 0x8) << (11 - 3)))
67 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
68 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
69 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
70 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
71 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
72 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
73 #define HSW_PTE_UNCACHED		(0)
74 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
75 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
76 
77 /* GEN8 legacy style address is defined as a 3 level page table:
78  * 31:30 | 29:21 | 20:12 |  11:0
79  * PDPE  |  PDE  |  PTE  | offset
80  * The difference as compared to normal x86 3 level page table is the PDPEs are
81  * programmed via register.
82  */
83 #define GEN8_PDPE_SHIFT			30
84 #define GEN8_PDPE_MASK			0x3
85 #define GEN8_PDE_SHIFT			21
86 #define GEN8_PDE_MASK			0x1ff
87 #define GEN8_PTE_SHIFT			12
88 #define GEN8_PTE_MASK			0x1ff
89 #define GEN8_LEGACY_PDPS		4
90 #define GEN8_PTES_PER_PAGE		(PAGE_SIZE / sizeof(gen8_gtt_pte_t))
91 #define GEN8_PDES_PER_PAGE		(PAGE_SIZE / sizeof(gen8_ppgtt_pde_t))
92 
93 #define PPAT_UNCACHED_INDEX		(_PAGE_PWT | _PAGE_PCD)
94 #define PPAT_CACHED_PDE_INDEX		0 /* WB LLC */
95 #define PPAT_CACHED_INDEX		_PAGE_PAT /* WB LLCeLLC */
96 #define PPAT_DISPLAY_ELLC_INDEX		_PAGE_PCD /* WT eLLC */
97 
98 #define CHV_PPAT_SNOOP			(1<<6)
99 #define GEN8_PPAT_AGE(x)		(x<<4)
100 #define GEN8_PPAT_LLCeLLC		(3<<2)
101 #define GEN8_PPAT_LLCELLC		(2<<2)
102 #define GEN8_PPAT_LLC			(1<<2)
103 #define GEN8_PPAT_WB			(3<<0)
104 #define GEN8_PPAT_WT			(2<<0)
105 #define GEN8_PPAT_WC			(1<<0)
106 #define GEN8_PPAT_UC			(0<<0)
107 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
108 #define GEN8_PPAT(i, x)			((uint64_t) (x) << ((i) * 8))
109 
110 enum i915_cache_level;
111 /**
112  * A VMA represents a GEM BO that is bound into an address space. Therefore, a
113  * VMA's presence cannot be guaranteed before binding, or after unbinding the
114  * object into/from the address space.
115  *
116  * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
117  * will always be <= an objects lifetime. So object refcounting should cover us.
118  */
119 struct i915_vma {
120 	struct drm_mm_node node;
121 	struct drm_i915_gem_object *obj;
122 	struct i915_address_space *vm;
123 
124 	/** This object's place on the active/inactive lists */
125 	struct list_head mm_list;
126 
127 	struct list_head vma_link; /* Link in the object's VMA list */
128 
129 	/** This vma's place in the batchbuffer or on the eviction list */
130 	struct list_head exec_list;
131 
132 	/**
133 	 * Used for performing relocations during execbuffer insertion.
134 	 */
135 	struct hlist_node exec_node;
136 	unsigned long exec_handle;
137 	struct drm_i915_gem_exec_object2 *exec_entry;
138 
139 	/**
140 	 * How many users have pinned this object in GTT space. The following
141 	 * users can each hold at most one reference: pwrite/pread, pin_ioctl
142 	 * (via user_pin_count), execbuffer (objects are not allowed multiple
143 	 * times for the same batchbuffer), and the framebuffer code. When
144 	 * switching/pageflipping, the framebuffer code has at most two buffers
145 	 * pinned per crtc.
146 	 *
147 	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
148 	 * bits with absolutely no headroom. So use 4 bits. */
149 	unsigned int pin_count:4;
150 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
151 
152 	/** Unmap an object from an address space. This usually consists of
153 	 * setting the valid PTE entries to a reserved scratch page. */
154 	void (*unbind_vma)(struct i915_vma *vma);
155 	/* Map an object into an address space with the given cache flags. */
156 #define GLOBAL_BIND (1<<0)
157 #define PTE_READ_ONLY (1<<1)
158 	void (*bind_vma)(struct i915_vma *vma,
159 			 enum i915_cache_level cache_level,
160 			 u32 flags);
161 };
162 
163 struct i915_address_space {
164 	struct drm_mm mm;
165 	struct drm_device *dev;
166 	struct list_head global_link;
167 	unsigned long start;		/* Start offset always 0 for dri2 */
168 	size_t total;		/* size addr space maps (ex. 2GB for ggtt) */
169 
170 	struct {
171 		dma_addr_t addr;
172 		struct page *page;
173 	} scratch;
174 
175 	/**
176 	 * List of objects currently involved in rendering.
177 	 *
178 	 * Includes buffers having the contents of their GPU caches
179 	 * flushed, not necessarily primitives.  last_rendering_seqno
180 	 * represents when the rendering involved will be completed.
181 	 *
182 	 * A reference is held on the buffer while on this list.
183 	 */
184 	struct list_head active_list;
185 
186 	/**
187 	 * LRU list of objects which are not in the ringbuffer and
188 	 * are ready to unbind, but are still in the GTT.
189 	 *
190 	 * last_rendering_seqno is 0 while an object is in this list.
191 	 *
192 	 * A reference is not held on the buffer while on this list,
193 	 * as merely being GTT-bound shouldn't prevent its being
194 	 * freed, and we'll pull it off the list in the free path.
195 	 */
196 	struct list_head inactive_list;
197 
198 	/* FIXME: Need a more generic return type */
199 	gen6_gtt_pte_t (*pte_encode)(dma_addr_t addr,
200 				     enum i915_cache_level level,
201 				     bool valid, u32 flags); /* Create a valid PTE */
202 	void (*clear_range)(struct i915_address_space *vm,
203 			    uint64_t start,
204 			    uint64_t length,
205 			    bool use_scratch);
206 	void (*insert_entries)(struct i915_address_space *vm,
207 			       struct sg_table *st,
208 			       uint64_t start,
209 			       enum i915_cache_level cache_level, u32 flags);
210 	void (*cleanup)(struct i915_address_space *vm);
211 };
212 
213 /* The Graphics Translation Table is the way in which GEN hardware translates a
214  * Graphics Virtual Address into a Physical Address. In addition to the normal
215  * collateral associated with any va->pa translations GEN hardware also has a
216  * portion of the GTT which can be mapped by the CPU and remain both coherent
217  * and correct (in cases like swizzling). That region is referred to as GMADR in
218  * the spec.
219  */
220 struct i915_gtt {
221 	struct i915_address_space base;
222 	size_t stolen_size;		/* Total size of stolen memory */
223 
224 	unsigned long mappable_end;	/* End offset that we can CPU map */
225 	struct io_mapping *mappable;	/* Mapping to our CPU mappable region */
226 	phys_addr_t mappable_base;	/* PA of our GMADR */
227 
228 	/** "Graphics Stolen Memory" holds the global PTEs */
229 	void __iomem *gsm;
230 
231 	bool do_idle_maps;
232 
233 	int mtrr;
234 
235 	/* global gtt ops */
236 	int (*gtt_probe)(struct drm_device *dev, size_t *gtt_total,
237 			  size_t *stolen, phys_addr_t *mappable_base,
238 			  unsigned long *mappable_end);
239 };
240 
241 struct i915_hw_ppgtt {
242 	struct i915_address_space base;
243 	struct kref ref;
244 	struct drm_mm_node node;
245 	unsigned num_pd_entries;
246 	unsigned num_pd_pages; /* gen8+ */
247 	union {
248 		struct page **pt_pages;
249 		struct page **gen8_pt_pages[GEN8_LEGACY_PDPS];
250 	};
251 	struct page *pd_pages;
252 	union {
253 		uint32_t pd_offset;
254 		dma_addr_t pd_dma_addr[GEN8_LEGACY_PDPS];
255 	};
256 	union {
257 		dma_addr_t *pt_dma_addr;
258 		dma_addr_t *gen8_pt_dma_addr[4];
259 	};
260 
261 	struct intel_context *ctx;
262 
263 	int (*enable)(struct i915_hw_ppgtt *ppgtt);
264 	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
265 			 struct intel_engine_cs *ring,
266 			 bool synchronous);
267 	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
268 };
269 
270 int i915_gem_gtt_init(struct drm_device *dev);
271 void i915_gem_init_global_gtt(struct drm_device *dev);
272 void i915_gem_setup_global_gtt(struct drm_device *dev, unsigned long start,
273 			       unsigned long mappable_end, unsigned long end);
274 
275 int i915_gem_init_ppgtt(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
276 
277 void i915_check_and_clear_faults(struct drm_device *dev);
278 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
279 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
280 
281 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
282 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
283 
284 #endif
285