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_pte_t;
40 typedef uint64_t gen8_pte_t;
41 typedef uint64_t gen8_pde_t;
42 
43 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
44 
45 
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 I915_PTES(pte_len)		(PAGE_SIZE / (pte_len))
55 #define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
56 #define I915_PDES			512
57 #define I915_PDE_MASK			(I915_PDES - 1)
58 #define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
59 
60 #define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
61 #define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
62 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
63 #define GEN6_PDE_SHIFT			22
64 #define GEN6_PDE_VALID			(1 << 0)
65 
66 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
67 
68 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
69 #define BYT_PTE_WRITEABLE		(1 << 1)
70 
71 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
72  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
73  */
74 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
75 					 (((bits) & 0x8) << (11 - 3)))
76 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
77 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
78 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
79 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
80 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
81 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
82 #define HSW_PTE_UNCACHED		(0)
83 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
84 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
85 
86 /* GEN8 legacy style address is defined as a 3 level page table:
87  * 31:30 | 29:21 | 20:12 |  11:0
88  * PDPE  |  PDE  |  PTE  | offset
89  * The difference as compared to normal x86 3 level page table is the PDPEs are
90  * programmed via register.
91  */
92 #define GEN8_PDPE_SHIFT			30
93 #define GEN8_PDPE_MASK			0x3
94 #define GEN8_PDE_SHIFT			21
95 #define GEN8_PDE_MASK			0x1ff
96 #define GEN8_PTE_SHIFT			12
97 #define GEN8_PTE_MASK			0x1ff
98 #define GEN8_LEGACY_PDPES		4
99 #define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
100 
101 #define PPAT_UNCACHED_INDEX		(_PAGE_PWT | _PAGE_PCD)
102 #define PPAT_CACHED_PDE_INDEX		0 /* WB LLC */
103 #define PPAT_CACHED_INDEX		_PAGE_PAT /* WB LLCeLLC */
104 #define PPAT_DISPLAY_ELLC_INDEX		_PAGE_PCD /* WT eLLC */
105 
106 #define CHV_PPAT_SNOOP			(1<<6)
107 #define GEN8_PPAT_AGE(x)		(x<<4)
108 #define GEN8_PPAT_LLCeLLC		(3<<2)
109 #define GEN8_PPAT_LLCELLC		(2<<2)
110 #define GEN8_PPAT_LLC			(1<<2)
111 #define GEN8_PPAT_WB			(3<<0)
112 #define GEN8_PPAT_WT			(2<<0)
113 #define GEN8_PPAT_WC			(1<<0)
114 #define GEN8_PPAT_UC			(0<<0)
115 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
116 #define GEN8_PPAT(i, x)			((uint64_t) (x) << ((i) * 8))
117 
118 enum i915_ggtt_view_type {
119 	I915_GGTT_VIEW_NORMAL = 0,
120 	I915_GGTT_VIEW_ROTATED
121 };
122 
123 struct intel_rotation_info {
124 	unsigned int height;
125 	unsigned int pitch;
126 	uint32_t pixel_format;
127 	uint64_t fb_modifier;
128 };
129 
130 struct i915_ggtt_view {
131 	enum i915_ggtt_view_type type;
132 
133 	struct sg_table *pages;
134 
135 	union {
136 		struct intel_rotation_info rotation_info;
137 	};
138 };
139 
140 extern const struct i915_ggtt_view i915_ggtt_view_normal;
141 extern const struct i915_ggtt_view i915_ggtt_view_rotated;
142 
143 enum i915_cache_level;
144 
145 /**
146  * A VMA represents a GEM BO that is bound into an address space. Therefore, a
147  * VMA's presence cannot be guaranteed before binding, or after unbinding the
148  * object into/from the address space.
149  *
150  * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
151  * will always be <= an objects lifetime. So object refcounting should cover us.
152  */
153 struct i915_vma {
154 	struct drm_mm_node node;
155 	struct drm_i915_gem_object *obj;
156 	struct i915_address_space *vm;
157 
158 	/** Flags and address space this VMA is bound to */
159 #define GLOBAL_BIND	(1<<0)
160 #define LOCAL_BIND	(1<<1)
161 #define PTE_READ_ONLY	(1<<2)
162 	unsigned int bound : 4;
163 
164 	/**
165 	 * Support different GGTT views into the same object.
166 	 * This means there can be multiple VMA mappings per object and per VM.
167 	 * i915_ggtt_view_type is used to distinguish between those entries.
168 	 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
169 	 * assumed in GEM functions which take no ggtt view parameter.
170 	 */
171 	struct i915_ggtt_view ggtt_view;
172 
173 	/** This object's place on the active/inactive lists */
174 	struct list_head mm_list;
175 
176 	struct list_head vma_link; /* Link in the object's VMA list */
177 
178 	/** This vma's place in the batchbuffer or on the eviction list */
179 	struct list_head exec_list;
180 
181 	/**
182 	 * Used for performing relocations during execbuffer insertion.
183 	 */
184 	struct hlist_node exec_node;
185 	unsigned long exec_handle;
186 	struct drm_i915_gem_exec_object2 *exec_entry;
187 
188 	/**
189 	 * How many users have pinned this object in GTT space. The following
190 	 * users can each hold at most one reference: pwrite/pread, execbuffer
191 	 * (objects are not allowed multiple times for the same batchbuffer),
192 	 * and the framebuffer code. When switching/pageflipping, the
193 	 * framebuffer code has at most two buffers pinned per crtc.
194 	 *
195 	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
196 	 * bits with absolutely no headroom. So use 4 bits. */
197 	unsigned int pin_count:4;
198 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
199 
200 	/** Unmap an object from an address space. This usually consists of
201 	 * setting the valid PTE entries to a reserved scratch page. */
202 	void (*unbind_vma)(struct i915_vma *vma);
203 	/* Map an object into an address space with the given cache flags. */
204 	void (*bind_vma)(struct i915_vma *vma,
205 			 enum i915_cache_level cache_level,
206 			 u32 flags);
207 };
208 
209 struct i915_page_table_entry {
210 	struct page *page;
211 	dma_addr_t daddr;
212 
213 	unsigned long *used_ptes;
214 };
215 
216 struct i915_page_directory_entry {
217 	struct page *page; /* NULL for GEN6-GEN7 */
218 	union {
219 		uint32_t pd_offset;
220 		dma_addr_t daddr;
221 	};
222 
223 	struct i915_page_table_entry *page_table[I915_PDES]; /* PDEs */
224 };
225 
226 struct i915_page_directory_pointer_entry {
227 	/* struct page *page; */
228 	struct i915_page_directory_entry *page_directory[GEN8_LEGACY_PDPES];
229 };
230 
231 struct i915_address_space {
232 	struct drm_mm mm;
233 	struct drm_device *dev;
234 	struct list_head global_link;
235 	unsigned long start;		/* Start offset always 0 for dri2 */
236 	size_t total;		/* size addr space maps (ex. 2GB for ggtt) */
237 
238 	struct {
239 		dma_addr_t addr;
240 		struct page *page;
241 	} scratch;
242 
243 	/**
244 	 * List of objects currently involved in rendering.
245 	 *
246 	 * Includes buffers having the contents of their GPU caches
247 	 * flushed, not necessarily primitives. last_read_req
248 	 * represents when the rendering involved will be completed.
249 	 *
250 	 * A reference is held on the buffer while on this list.
251 	 */
252 	struct list_head active_list;
253 
254 	/**
255 	 * LRU list of objects which are not in the ringbuffer and
256 	 * are ready to unbind, but are still in the GTT.
257 	 *
258 	 * last_read_req is NULL while an object is in this list.
259 	 *
260 	 * A reference is not held on the buffer while on this list,
261 	 * as merely being GTT-bound shouldn't prevent its being
262 	 * freed, and we'll pull it off the list in the free path.
263 	 */
264 	struct list_head inactive_list;
265 
266 	/* FIXME: Need a more generic return type */
267 	gen6_pte_t (*pte_encode)(dma_addr_t addr,
268 				 enum i915_cache_level level,
269 				 bool valid, u32 flags); /* Create a valid PTE */
270 	int (*allocate_va_range)(struct i915_address_space *vm,
271 				 uint64_t start,
272 				 uint64_t length);
273 	void (*clear_range)(struct i915_address_space *vm,
274 			    uint64_t start,
275 			    uint64_t length,
276 			    bool use_scratch);
277 	void (*insert_entries)(struct i915_address_space *vm,
278 			       struct sg_table *st,
279 			       uint64_t start,
280 			       enum i915_cache_level cache_level, u32 flags);
281 	void (*cleanup)(struct i915_address_space *vm);
282 };
283 
284 /* The Graphics Translation Table is the way in which GEN hardware translates a
285  * Graphics Virtual Address into a Physical Address. In addition to the normal
286  * collateral associated with any va->pa translations GEN hardware also has a
287  * portion of the GTT which can be mapped by the CPU and remain both coherent
288  * and correct (in cases like swizzling). That region is referred to as GMADR in
289  * the spec.
290  */
291 struct i915_gtt {
292 	struct i915_address_space base;
293 	size_t stolen_size;		/* Total size of stolen memory */
294 
295 	unsigned long mappable_end;	/* End offset that we can CPU map */
296 	struct io_mapping *mappable;	/* Mapping to our CPU mappable region */
297 	phys_addr_t mappable_base;	/* PA of our GMADR */
298 
299 	/** "Graphics Stolen Memory" holds the global PTEs */
300 	void __iomem *gsm;
301 
302 	bool do_idle_maps;
303 
304 	int mtrr;
305 
306 	/* global gtt ops */
307 	int (*gtt_probe)(struct drm_device *dev, size_t *gtt_total,
308 			  size_t *stolen, phys_addr_t *mappable_base,
309 			  unsigned long *mappable_end);
310 };
311 
312 struct i915_hw_ppgtt {
313 	struct i915_address_space base;
314 	struct kref ref;
315 	struct drm_mm_node node;
316 	unsigned long pd_dirty_rings;
317 	unsigned num_pd_entries;
318 	unsigned num_pd_pages; /* gen8+ */
319 	union {
320 		struct i915_page_directory_pointer_entry pdp;
321 		struct i915_page_directory_entry pd;
322 	};
323 
324 	struct i915_page_table_entry *scratch_pt;
325 
326 	struct drm_i915_file_private *file_priv;
327 
328 	gen6_pte_t __iomem *pd_addr;
329 
330 	int (*enable)(struct i915_hw_ppgtt *ppgtt);
331 	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
332 			 struct intel_engine_cs *ring);
333 	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
334 };
335 
336 /* For each pde iterates over every pde between from start until start + length.
337  * If start, and start+length are not perfectly divisible, the macro will round
338  * down, and up as needed. The macro modifies pde, start, and length. Dev is
339  * only used to differentiate shift values. Temp is temp.  On gen6/7, start = 0,
340  * and length = 2G effectively iterates over every PDE in the system.
341  *
342  * XXX: temp is not actually needed, but it saves doing the ALIGN operation.
343  */
344 #define gen6_for_each_pde(pt, pd, start, length, temp, iter) \
345 	for (iter = gen6_pde_index(start); \
346 	     pt = (pd)->page_table[iter], length > 0 && iter < I915_PDES; \
347 	     iter++, \
348 	     temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT) - start, \
349 	     temp = min_t(unsigned, temp, length), \
350 	     start += temp, length -= temp)
351 
352 static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
353 {
354 	const uint32_t mask = NUM_PTE(pde_shift) - 1;
355 
356 	return (address >> PAGE_SHIFT) & mask;
357 }
358 
359 /* Helper to counts the number of PTEs within the given length. This count
360  * does not cross a page table boundary, so the max value would be
361  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
362 */
363 static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
364 				      uint32_t pde_shift)
365 {
366 	const uint64_t mask = ~((1 << pde_shift) - 1);
367 	uint64_t end;
368 
369 	WARN_ON(length == 0);
370 	WARN_ON(offset_in_page(addr|length));
371 
372 	end = addr + length;
373 
374 	if ((addr & mask) != (end & mask))
375 		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
376 
377 	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
378 }
379 
380 static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
381 {
382 	return (addr >> shift) & I915_PDE_MASK;
383 }
384 
385 static inline uint32_t gen6_pte_index(uint32_t addr)
386 {
387 	return i915_pte_index(addr, GEN6_PDE_SHIFT);
388 }
389 
390 static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
391 {
392 	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
393 }
394 
395 static inline uint32_t gen6_pde_index(uint32_t addr)
396 {
397 	return i915_pde_index(addr, GEN6_PDE_SHIFT);
398 }
399 
400 int i915_gem_gtt_init(struct drm_device *dev);
401 void i915_gem_init_global_gtt(struct drm_device *dev);
402 void i915_global_gtt_cleanup(struct drm_device *dev);
403 
404 
405 int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
406 int i915_ppgtt_init_hw(struct drm_device *dev);
407 void i915_ppgtt_release(struct kref *kref);
408 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
409 					struct drm_i915_file_private *fpriv);
410 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
411 {
412 	if (ppgtt)
413 		kref_get(&ppgtt->ref);
414 }
415 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
416 {
417 	if (ppgtt)
418 		kref_put(&ppgtt->ref, i915_ppgtt_release);
419 }
420 
421 void i915_check_and_clear_faults(struct drm_device *dev);
422 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
423 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
424 
425 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
426 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
427 
428 static inline bool
429 i915_ggtt_view_equal(const struct i915_ggtt_view *a,
430                      const struct i915_ggtt_view *b)
431 {
432 	if (WARN_ON(!a || !b))
433 		return false;
434 
435 	return a->type == b->type;
436 }
437 
438 #endif
439