drm/i915: Rework ppgtt init to no require an aliasing ppgtt
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / drivers / gpu / drm / i915 / i915_gem_gtt.h
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 struct drm_i915_file_private;
38
39 typedef uint32_t gen6_gtt_pte_t;
40 typedef uint64_t gen8_gtt_pte_t;
41 typedef gen8_gtt_pte_t gen8_ppgtt_pde_t;
42
43 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
44
45 #define I915_PPGTT_PT_ENTRIES (PAGE_SIZE / sizeof(gen6_gtt_pte_t))
46 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
47 #define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
48 #define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
49 #define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
50 #define GEN6_PTE_CACHE_LLC (2 << 1)
51 #define GEN6_PTE_UNCACHED (1 << 1)
52 #define GEN6_PTE_VALID (1 << 0)
53
54 #define GEN6_PPGTT_PD_ENTRIES 512
55 #define GEN6_PD_SIZE (GEN6_PPGTT_PD_ENTRIES * PAGE_SIZE)
56 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
57 #define GEN6_PDE_VALID (1 << 0)
58
59 #define GEN7_PTE_CACHE_L3_LLC (3 << 1)
60
61 #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
62 #define BYT_PTE_WRITEABLE (1 << 1)
63
64 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
65 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
66 */
67 #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
68 (((bits) & 0x8) << (11 - 3)))
69 #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
70 #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
71 #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
72 #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
73 #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
74 #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
75 #define HSW_PTE_UNCACHED (0)
76 #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
77 #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
78
79 /* GEN8 legacy style address is defined as a 3 level page table:
80 * 31:30 | 29:21 | 20:12 | 11:0
81 * PDPE | PDE | PTE | offset
82 * The difference as compared to normal x86 3 level page table is the PDPEs are
83 * programmed via register.
84 */
85 #define GEN8_PDPE_SHIFT 30
86 #define GEN8_PDPE_MASK 0x3
87 #define GEN8_PDE_SHIFT 21
88 #define GEN8_PDE_MASK 0x1ff
89 #define GEN8_PTE_SHIFT 12
90 #define GEN8_PTE_MASK 0x1ff
91 #define GEN8_LEGACY_PDPS 4
92 #define GEN8_PTES_PER_PAGE (PAGE_SIZE / sizeof(gen8_gtt_pte_t))
93 #define GEN8_PDES_PER_PAGE (PAGE_SIZE / sizeof(gen8_ppgtt_pde_t))
94
95 #define PPAT_UNCACHED_INDEX (_PAGE_PWT | _PAGE_PCD)
96 #define PPAT_CACHED_PDE_INDEX 0 /* WB LLC */
97 #define PPAT_CACHED_INDEX _PAGE_PAT /* WB LLCeLLC */
98 #define PPAT_DISPLAY_ELLC_INDEX _PAGE_PCD /* WT eLLC */
99
100 #define CHV_PPAT_SNOOP (1<<6)
101 #define GEN8_PPAT_AGE(x) (x<<4)
102 #define GEN8_PPAT_LLCeLLC (3<<2)
103 #define GEN8_PPAT_LLCELLC (2<<2)
104 #define GEN8_PPAT_LLC (1<<2)
105 #define GEN8_PPAT_WB (3<<0)
106 #define GEN8_PPAT_WT (2<<0)
107 #define GEN8_PPAT_WC (1<<0)
108 #define GEN8_PPAT_UC (0<<0)
109 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
110 #define GEN8_PPAT(i, x) ((uint64_t) (x) << ((i) * 8))
111
112 enum i915_cache_level;
113 /**
114 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
115 * VMA's presence cannot be guaranteed before binding, or after unbinding the
116 * object into/from the address space.
117 *
118 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
119 * will always be <= an objects lifetime. So object refcounting should cover us.
120 */
121 struct i915_vma {
122 struct drm_mm_node node;
123 struct drm_i915_gem_object *obj;
124 struct i915_address_space *vm;
125
126 /** This object's place on the active/inactive lists */
127 struct list_head mm_list;
128
129 struct list_head vma_link; /* Link in the object's VMA list */
130
131 /** This vma's place in the batchbuffer or on the eviction list */
132 struct list_head exec_list;
133
134 /**
135 * Used for performing relocations during execbuffer insertion.
136 */
137 struct hlist_node exec_node;
138 unsigned long exec_handle;
139 struct drm_i915_gem_exec_object2 *exec_entry;
140
141 /**
142 * How many users have pinned this object in GTT space. The following
143 * users can each hold at most one reference: pwrite/pread, pin_ioctl
144 * (via user_pin_count), execbuffer (objects are not allowed multiple
145 * times for the same batchbuffer), and the framebuffer code. When
146 * switching/pageflipping, the framebuffer code has at most two buffers
147 * pinned per crtc.
148 *
149 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
150 * bits with absolutely no headroom. So use 4 bits. */
151 unsigned int pin_count:4;
152 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
153
154 /** Unmap an object from an address space. This usually consists of
155 * setting the valid PTE entries to a reserved scratch page. */
156 void (*unbind_vma)(struct i915_vma *vma);
157 /* Map an object into an address space with the given cache flags. */
158 #define GLOBAL_BIND (1<<0)
159 #define PTE_READ_ONLY (1<<1)
160 void (*bind_vma)(struct i915_vma *vma,
161 enum i915_cache_level cache_level,
162 u32 flags);
163 };
164
165 struct i915_address_space {
166 struct drm_mm mm;
167 struct drm_device *dev;
168 struct list_head global_link;
169 unsigned long start; /* Start offset always 0 for dri2 */
170 size_t total; /* size addr space maps (ex. 2GB for ggtt) */
171
172 struct {
173 dma_addr_t addr;
174 struct page *page;
175 } scratch;
176
177 /**
178 * List of objects currently involved in rendering.
179 *
180 * Includes buffers having the contents of their GPU caches
181 * flushed, not necessarily primitives. last_rendering_seqno
182 * represents when the rendering involved will be completed.
183 *
184 * A reference is held on the buffer while on this list.
185 */
186 struct list_head active_list;
187
188 /**
189 * LRU list of objects which are not in the ringbuffer and
190 * are ready to unbind, but are still in the GTT.
191 *
192 * last_rendering_seqno is 0 while an object is in this list.
193 *
194 * A reference is not held on the buffer while on this list,
195 * as merely being GTT-bound shouldn't prevent its being
196 * freed, and we'll pull it off the list in the free path.
197 */
198 struct list_head inactive_list;
199
200 /* FIXME: Need a more generic return type */
201 gen6_gtt_pte_t (*pte_encode)(dma_addr_t addr,
202 enum i915_cache_level level,
203 bool valid, u32 flags); /* Create a valid PTE */
204 void (*clear_range)(struct i915_address_space *vm,
205 uint64_t start,
206 uint64_t length,
207 bool use_scratch);
208 void (*insert_entries)(struct i915_address_space *vm,
209 struct sg_table *st,
210 uint64_t start,
211 enum i915_cache_level cache_level, u32 flags);
212 void (*cleanup)(struct i915_address_space *vm);
213 };
214
215 /* The Graphics Translation Table is the way in which GEN hardware translates a
216 * Graphics Virtual Address into a Physical Address. In addition to the normal
217 * collateral associated with any va->pa translations GEN hardware also has a
218 * portion of the GTT which can be mapped by the CPU and remain both coherent
219 * and correct (in cases like swizzling). That region is referred to as GMADR in
220 * the spec.
221 */
222 struct i915_gtt {
223 struct i915_address_space base;
224 size_t stolen_size; /* Total size of stolen memory */
225
226 unsigned long mappable_end; /* End offset that we can CPU map */
227 struct io_mapping *mappable; /* Mapping to our CPU mappable region */
228 phys_addr_t mappable_base; /* PA of our GMADR */
229
230 /** "Graphics Stolen Memory" holds the global PTEs */
231 void __iomem *gsm;
232
233 bool do_idle_maps;
234
235 int mtrr;
236
237 /* global gtt ops */
238 int (*gtt_probe)(struct drm_device *dev, size_t *gtt_total,
239 size_t *stolen, phys_addr_t *mappable_base,
240 unsigned long *mappable_end);
241 };
242
243 struct i915_hw_ppgtt {
244 struct i915_address_space base;
245 struct kref ref;
246 struct drm_mm_node node;
247 unsigned num_pd_entries;
248 unsigned num_pd_pages; /* gen8+ */
249 union {
250 struct page **pt_pages;
251 struct page **gen8_pt_pages[GEN8_LEGACY_PDPS];
252 };
253 struct page *pd_pages;
254 union {
255 uint32_t pd_offset;
256 dma_addr_t pd_dma_addr[GEN8_LEGACY_PDPS];
257 };
258 union {
259 dma_addr_t *pt_dma_addr;
260 dma_addr_t *gen8_pt_dma_addr[4];
261 };
262
263 struct drm_i915_file_private *file_priv;
264
265 int (*enable)(struct i915_hw_ppgtt *ppgtt);
266 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
267 struct intel_engine_cs *ring,
268 bool synchronous);
269 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
270 };
271
272 int i915_gem_gtt_init(struct drm_device *dev);
273 void i915_gem_init_global_gtt(struct drm_device *dev);
274 int i915_gem_setup_global_gtt(struct drm_device *dev, unsigned long start,
275 unsigned long mappable_end, unsigned long end);
276
277
278 int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
279 int i915_ppgtt_init_hw(struct drm_device *dev);
280 void i915_ppgtt_release(struct kref *kref);
281 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
282 struct drm_i915_file_private *fpriv);
283 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
284 {
285 if (ppgtt)
286 kref_get(&ppgtt->ref);
287 }
288 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
289 {
290 if (ppgtt)
291 kref_put(&ppgtt->ref, i915_ppgtt_release);
292 }
293
294 void i915_check_and_clear_faults(struct drm_device *dev);
295 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
296 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
297
298 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
299 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
300
301 #endif